A Set of Hox Proteins Interact with the Maf Oncoprotein to Inhibit Its DNA Binding, Transactivation, and Transforming Activities*

Maf oncoprotein is a basic-leucine zipper (bZip) type of transcriptional activator. Since many transcription factors are known to form functional complexes, we searched for proteins that interact with the DNA-binding domain of Maf using the phage display method and identified two homeodomain-containing proteins, Hoxd12 and MHox/Prx1/Phox1/Pmx1. Studies with mutants of Hox and Maf proteins showed that they associate through their DNA-binding domains; the homeodomain of Hox and the bZip domain of Maf, respectively. Reflecting the high similarity of the bZip domain, all other Maf family members tested (c-/v-Maf, MafB, MafK, MafF, and MafG) also associated with the Hox proteins. Pax6, whose homeodomain is relatively similar to MHox, also could interact with Maf. However, two other bZip oncoproteins, Fos and Jun, failed to associate with the Hox proteins, while a distantly related Hox family member, Meis1, could not interact with Maf. Through interactions with the bZip domain, the Hox proteins inhibited the DNA binding activity of Maf, whereas the binding of Hox proteins to their recognition sequences was not abrogated by Maf. We further showed that coexpression of the Hox proteins repressed transcriptional activation and transforming activity of Maf. These results suggested that the interaction of a set of Hox proteins with Maf family members may interfere not only with their oncogenicity but also with their physiological roles.

v-maf oncogene, originally identified in the genome of acute transforming avian retrovirus AS42, induces musculoaponeurotic fibrosarcoma in chickens and causes transformation of chicken embryo fibroblast (CEF) 1 cells (1,2). Its product, v-Maf, contains a basic-leucine zipper (bZip) structure at its carboxyl terminus, forms a homodimer, and recognizes rela-tively long palindromic sequences named Maf-recognition elements (MARE: TGCTGACTCAGCA and TGCTGACGTCAG-CA) (3)(4)(5). MARE sequences include the phorbol 12-Otetradecanoate-13-acetate-responsive element (TGACTCA) and cyclic AMP-responsive element (TGACGTCA), which are recognized by various homo-and heterodimers of AP-1 (Jun/ Fos) and the ATF/CREB family of bZip proteins. Maf is a transcriptional transactivator, and the DNA binding and transactivation activities are necessary for its transforming ability (4,6). Structural analysis of v-maf and its cellular counterpart c-maf revealed that the v-Maf protein had no structural change compared with c-Maf except that it was fused to the viral Gag protein at the initiator methionine residue of c-Maf (2). In accordance with this finding, expression of c-maf under control of a retroviral long terminal repeat causes efficient transformation of CEF cells (5). Recently, c-maf overexpression was found in a fraction of human multiple myelomas that resulted from chromosomal translocation to the immunoglobulin heavy or light chain locus (7). Up-regulation of c-maf was also found in melanoma cells due to the insertion of the melanoma-associated retrovirus (MelARV) (8).
To date, several maf-related genes have been isolated from various vertebrates, including human, mouse, rat, chicken, quail, frog, and zebrafish. They are divided into two subfamilies according to the structures of the encoded proteins. The first group, which consists of c-Maf, MafB, Nrl, and MafA/L-Maf, has an amino-terminal transactivation domain in addition to the COOH-terminal bZip domain and are called large Maf proteins. The second group, MafK, MafF and MafG, lack the amino-terminal domain and thus are called small Maf proteins.
One of the most important control mechanisms of transcriptional regulation by bZip factors is homo-and heterodimerization through leucine zipper domains. For example, the small Maf proteins can positively regulate a set of erythroid-specific genes as heterodimers with another b-Zip protein, NF-E2 p45, a member of the Cap'n'collar family of transcription factors (9,10). However, the small Maf proteins also can act as repressors for the erythroid-specific genes by forming homodimers or heterodimers with c-Fos (11). Similarly, the small Maf proteins can form dimers with another bZip factor Bach2 and act as a B-cell-specific negative regulator of the immunoglobulin heavy chain 3Ј enhancer (12). As for v-/c-Maf, it has been shown to form heterodimers with both Jun and Fos (4,13). The functional importance of Maf/Jun or Maf/Fos heterodimers is not yet known, but these heterodimers are different in their DNA binding specificity from Maf homodimers or AP-1 complexes and are likely to regulate a distinct set of cellular genes.
Jun and Fos are the most well characterized nuclear oncoproteins, and their interaction with members of other bZip protein families such as ATF/CREB and C/EBP has been examined (14 -16). Furthermore, Jun and Fos are known to form many kinds of functional complexes with different classes of transcription factors. For example, direct interaction of the Jun/Fos heterodimer with the NF-AT transcription factor is necessary to activate expression of T-cell-specific genes (17,18). Interactions of Jun and Fos with the glucocorticoid receptor or some members of the Ets family also have been reported (19 -21). These studies shed light not only on their physiological roles but also on the mechanisms of cell transformation. In the case of c-Maf, interaction with the proto-oncogene product c-Myb was found to regulate transcription of the CD13/APN gene during myeloid cell differentiation (22). Its close relative, MafB, also has been shown to interact with and to repress transcriptional activity of c-Ets-1, which leads to inhibition of differentiation of erythroid cells (23). Collectively, fine tuning of cell type-specific gene expression and lineage-specific cell differentiation seems to be achieved by cooperative and inhibitory interactions of transcription factors. Disregulated expression of a nuclear oncoprotein thus may affect the cell differentiation program, which in turn may lead to cell transformation.
To understand the mechanism of transcriptional regulation and cell transformation by Maf, it is important to clarify the regulatory cross-talk that occurs with other transcription factors. To this end, we utilized the phage display method to screen a cDNA expression library for genes whose products interact with Maf and isolated two genes that encode homeodomain-containing transcription factors, Hoxd12 and MHox/ Prx1/Phox1/Pmx1. These two proteins repressed DNA binding and transforming activities of Maf by interacting with its bZip domain. The possible biological significance of this interaction network also will be discussed.

EXPERIMENTAL PROCEDURES
Phage Display-cDNA was made from poly(A) ϩ RNA isolated from whole chicken at embryonic day 8 and was inserted into the EcoRI and HindIII digested pT7Select1-1b vector (Novagen), which should display products of the inserted cDNA on the surface of T7 phage particles as a fusion product with its capsid protein. Phage were packaged in vitro and used to infect the host bacterial strain, BLT5403 (Novagen). Enrichment of recombinant phage that expressed candidate Maf-binding proteins by biopanning was performed as recommended by the manufacturer with some modifications. Briefly, wells of microtiter plates were coated with bacterially expressed and purified maltose binding protein (MBP) or MBP-Maf fusion protein in Tris-buffered saline (TBS) and blocked with 10% Blockace (Dainihon Seiyaku, Osaka, Japan) in TBS. A 10-l aliquot of fresh phage lysate (1 ϫ 10 8 pfu/l) containing 2.2 ϫ 10 6 independent clones was incubated in a MBP-coated well at room temperature for 30 min to absorb phage particles that have affinity for MBP or blocking reagent. Then the lysate was transferred into a MBP-Maf-coated well, incubated for 30 min, removed, and the well washed five times with 200 l of TBS containing 0.1% of Tween 20. The phages that remained in the well were eluted with 20 l of TBS containing 0.1% Tween 20 and 1% sodium dodecyl sulfate (SDS) and amplified in BLT5403 cells. After seven rounds of this two-step biopanning procedure, the titer of MBP-Maf bound phage was five times higher than background. The eluted phages were combined with host cells to form plaques on agar plates. The cDNA insert of individual phages was amplified by the polymerase chain reaction (PCR) using T7Select UP and DOWN primers (Novagen), subcloned into the EcoRI-HindIII site of pUC19, and subjected to nucleotide sequence analysis.
Coprecipitation Analysis by MBP or GST Fusion Proteins-Construction of the prokaryotic expression vector for the MBP-Maf fusion protein has been described (4,11,24). The procedures for expression of MBP and MBP-Maf proteins in Escherichia coli and for purification by amylose resin chromatograhy (New England Biolabs) also have been described previously (4).
For in vitro translation of human c-Jun and c-Fos, a PmaCI-StyI fragment of c-jun cDNA and a HaeII-BanI fragment of c-fos cDNA, both of which contain entire open reading frames, were blunt-ended by T4 DNA polymerase, ligated with MluI linkers (pGACACGCGTGTC), and inserted into pGEM4-MluI.
These plasmids were linearized with appropriate restriction enzymes and were transcribed and translated in vitro using the TNT Coupled Wheat Germ Extract System (Promega) in the presence of [ 35 S]methionine. 10 l of the programmed extracts were added to 800 l of phosphate-buffered saline containing 1% bovine serum albumin and 0.05% Tween 20 together with 20 l of amylose resin or glutathione-Sepharose 4B immobilized with MBP or GST fusion proteins, respectively. After overnight incubation at 4°C, samples were washed three times with 800 l of phosphate-buffered saline containing 0.05% Tween 20 and analyzed by SDS-polyacrylamide gel electrophoresis (SDS-PAGE).
Gel Mobility Shift Analysis-XhoI restriction sites were introduced at the first leucine residue of the zipper structure of Maf and Jun by the method of Kunkel (26) using oligonucleotide primers (5Ј-CTCCGACTC-gAGGACGTGCCGC-3Ј for v-maf and 5Ј-CACTTTTTCcTCgAgCCTG-GCAATCC-3Ј for v-jun) without changing amino acids, and the resultant plasmids were used to construct a chimera of maf and jun (maf(junzip)). Maf, Jun, Fos, and Hox proteins were transcribed and translated in vitro using the TNT Coupled Wheat Germ Extract System (Promega), and were analyzed by a gel mobility shift assay as described previously (4). Oligonucleotide probes containing MARE (probe 7) or AP-1 (probe 11) also have been described (4). A double-stranded oligonucleotide probe containing the Hoxd12 recognition sequence (27) was made from the oligonucleotides: 5Ј-GATCCTTCACTCCGTTTTACG-ACAGGAGTA-3Ј and 5Ј-GATCTACTCCTGTCGTAAAACGGAGTG-AAG-3Ј.
Luciferase Assay-For eukaryotic expression, BssHII fragments of hoxd12 or Hoxd12⌬HD were inserted into a unique BssHII site of the pEF-BssHII expression vector (24), a derivative of pEF-BOS. The pEF/ v-maf expression plasmid and the luciferase reporter plasmid 3ϫMARE/RBGP-luc, containing three copies of MARE (oligonucleotide 7) and rabbit ␤-globin minimal promoter, have been described (6,24). pEF-Rluc, used to normalize transfection efficiencies, was constructed by inserting a XbaI-NheI fragment containing the open reading frame of the Renilla luciferase gene (pRL-TK, Promega) into the XbaI site of pEF-BOS, followed by deletion of the replication origin of SV40 by HindIII digestion and self-ligation.
3 ϫ 10 5 CEF cells, grown on a 35-mm dish, were transfected with a total of 1 g of plasmid (125 ng of luciferase reporter plasmid, 750 ng of expression plasmids, and 125 ng of pEF-Rluc plasmid) using 1 l of Superfect Transfection Reagent (Qiagen) as recommended by the manufacturer. The cells were harvested at 18 h post-transfection, and luciferase activity was measured using the Dual-Luciferase Reporter Assay System (Promega).
Focus Formation Assay-To construct a NotI cassette plasmid with the internal ribosome entry site (IRES), an EcoRI-NotI-EcoRI adaptor (5Ј-AATTCGGCGGCCGCGGATCCCCTCGAGTTCG-3Ј and 5Ј-pAATT-CGAACTCGAGGGGATCCGCGGCCGCCG-3Ј) and an NcoI-SmaI-Not-I-EcoRI adaptor (5Ј-AATTCGGCGGCCGCGGATCCGACCCGGGAT-C-3Ј and 5Ј-pCATGGATCCCGGGTCGGATCCGCGGCCGCCG-3Ј) were added to the 5Ј and 3Ј ends, respectively, of the EcoRI-NcoI fragment of pCITE-1 (Novagen) containing the IRES of encephalomyocarditis virus and inserted into the EcoRI site of a modified pUC vector. An NcoI restriction site was introduced into the 75th codon of hoxd12 and Hoxd12⌬HD in the pGEM3-MluI plasmid by PCR using the primers, 5Ј-GAGCCATGGGCTCCGTTCCAATC-3Ј and SP6 primer. The PCR fragment was digested with HindIII at the 3Ј polylinker site, treated with T4 DNA polymerase, digested with NcoI, and inserted into the NcoI and SmaI-digested IRES cassette plasmid to make pUC/IRES-hoxd12 and pUC/IRES-Hoxd12⌬HD. The NotI fragments containing IRES-hoxd12 and IRES-Hoxd12⌬HD were then inserted into the unique NotI site of the replication-competent avian retroviral vector, pRV-9. Subsequently, the MluI fragment containing chicken c-maf or chicken v-fos was introduced into a unique MluI site in the plasmid. To construct the chicken v-fos MluI fragment, the initiator methionine codon and an NcoI restriction site were created at the fusion point of gag and fos of the NK24 provirus (28), and the MluI site in the coding region was deleted by site-directed mutagenesis using the oligonucleotides, 5Ј-CCCCAGGAGccATggTCAACTCGCAGG-3Ј and 5Ј-CCTTC-TAtGCaTCGGACTGGGAG-3Ј, respectively. The modified v-fos fragment was introduced into the NcoI-MluI site of the pRAM plasmid (5) by addition of an MluI linker to the ApaI site at the 3Ј-noncoding region followed by digestion with NcoI and MluI.
Preparation and maintenance of CEF cells have been described previously (24). For the focus formation assay, 2.5 g of the recombinant retrovirus vector plasmid DNA was transfected into 1.2 ϫ 10 6 CEF cells grown on a 60-mm dish using 12.5 l of Effectene Transfection Reagent (Qiagen) as recommended by the manufacturer. On the day following transfection, the cells were trypsinized and appropriately diluted with fresh CEF cells to give a total of 1.2 ϫ 10 6 cells/60-mm dish. The cells were then overlaid with medium containing 0.4% agar and tested for focus formation.

RESULTS
Isolation of Hoxd12 and MHox as Maf-associated Proteins-We selected cDNAs encoding proteins that interact with the carboxyl terminus DNA-binding domain of Maf by the phage display method (for datails see "Experimental Procedures"). In brief, a cDNA library constructed from chicken whole embryo RNA was inserted into the T7Select vector, which expresses protein encoded by inserted cDNA on the surface of the phage particle as a fusion with phage capsid protein. Recombinant phage particles that express putative Maf-binding proteins were enriched by multiple rounds of biopanning on microtiter plates coated with purified Maf protein fused to maltose-binding protein (MBP-Maf). After seven rounds of selection, cDNA inserts of individual phages were amplified by PCR, subcloned, and sequenced. As a result, of 16 clones analyzed, 11 clones were derived from at least three independent cDNA molecules of the chicken mhox/prx1/ phox1/pmx1 gene. Four other clones were derived from at least two independent cDNA species of chicken hoxd12.
As shown schematically in Fig. 1A, both MHox and Hoxd12 proteins contain a homeodomain, which is a DNA-binding motif composed of 60 amino acids. Although all cDNA species iso-lated were partial, they all fused to the capsid protein of the T7 phage in frame and retained their homeodomains. Furthermore, they had no sequence similarity with each other except for their homeodomains (Fig. 1B), suggesting that Maf binds to these conserved domains of MHox and Hoxd12. hoxd12 is a member of the Abdominal B (Abd-B) group of hoxD cluster genes and has been shown to be involved in limb pattern formation (29). MHox was originally isolated as a member of the paired class homeobox proteins from mice that are expressed only in cells of mesodermal origin and bind to A/T-rich elements of the muscle creatin kinase enhancer (30). Its chicken and human counterparts, prx1 and phox1, were isolated as homeobox genes predominantly expressed in the developing limb (31) and as an interacting partner of the serum response factor (32), respectively. mhox is also identical to a recently reported gene named pmx1, which was found at a chromosome translocation break region in human acute myelogenous leukemia cells (33). The fact that two homeodomainencoding cDNAs were isolated independently and that homeobox proteins are transcriptional regulators prompted us to further analyze these cDNAs.
The Homeodomain Is Necessary for Hox Association with Maf-To confirm the interaction of these homeobox proteins with Maf, we cloned the entire open reading frames of hoxd12 and mhox cDNA by reverse transcriptase-PCR, subjected them to in vitro transcription and translation in the presence of [ 35 S]methionine, and tested for coprecipitation with MBP or MBP-Maf protein immobilized onto amylose resin. As shown in Fig. 2A, both Hoxd12 and MHox proteins were specifically coprecipitated with MBP-Maf but not with MBP. Deletion of the homeodomain of the Hoxd12 protein (Hoxd12⌬HD) resulted in loss of specific association with Maf, indicating that the homeodomain is necessary for this interaction. We also could detect recombinant GST-Hoxd12 and GST-MHox proteins (see below) immobilized on nitrocellulose membranes by West-Western method using the MBP-Maf fusion protein as a probe (data not shown), indicating direct interaction of Maf and Hox proteins.
Maf Also Interacts with Pax6 -MHox and Hoxd12 are not the closest members among the homeoprotein family (Fig. 1B). We thus suspected that other homeodomain-containing proteins also could associate with Maf. To test this possibility, we chose Pax6, which contains the paired class of homeodomain with high homology to that of Mhox. We also chose Meis1, whose homeodomain is distantly related to both MHox and Hoxd12 (Fig. 1B). Pax6 and Meis1 were translated in vitro and subjected to the coprecipitation assay with the MBP-Maf fusion bZip Domain of Maf Is Necessary for Interaction with Hox Proteins-We next tried to identify domains in Maf required for interaction with the homeobox proteins. Hoxd12 and MHox proteins fused to GST were expressed in E. coli, purified by glutathione beads, and used to test for association with a set of in vitro translated mutant Maf proteins. Schematic structures of the Maf mutants used in this assay are shown in Fig. 3A, together with their DNA binding abilities and transforming activities (4,5). Fig. 3B shows the coprecipitation results for a set of mutant Maf proteins with GST or GST-Hoxd12. Essentially the same results were obtained using GST-MHox (data not shown). The full-length v-Maf protein was efficiently coprecipitated with GST-Hoxd12 but not with GST, confirming the specific association of Maf and Hoxd12 proteins. Mutants with a deletion of the COOH-terminal, but which retain most of the zipper structure (CD1 and CD2), could associate with Hox, but further deletion of three out of six leucine repeats (CD3) resulted in a large reduction in the amount of coprecipitated protein. Deletion of the entire bZip domain (CD4) lead to complete loss of association. Disruption of the ␣-helical structure of the leucine zipper by amino acid substitutions of two leucine residues with prolines (L2PL4P) significantly affected the binding with GST-Hoxd12, suggesting that the intact leucine zipper structure is required for the interaction. A mutant, ND5, with a deletion of the amino-terminal region, contains the same region present in the MBP-Maf fusion protein and thus can associate with GST-Hoxd12. The mutant, ND6, which contains further deletions and cannot bind to DNA, could still efficiently be coprecipitated. On the other hand, deletion of either the basic domain (MD56) or the preceding region (MD45), which is also necessary for DNA binding, significantly affected the interaction. A deletion of only five amino acids in the basic region (MD26.22) effectively abrogated the association. The Q5H mutant, which contains a glutamine to histidine substitution in the hinge region of the basic domain and the leucine zipper that results in enhanced transforming ability by an unknown mechanism (5, 6), was coprecipitated with GST-Hox at a comparable level as the wild type Maf protein. These results together suggested that the domain spanning the basic region and the leucine zipper structure of Maf was necessary for association with Hox proteins.
Maf Family Proteins, but Not Jun and Fos, Interact with Hoxd12 and Mhox-We also tested the interaction between other bZip proteins and Hox proteins with the GST-pull-down assay (Fig. 4). In vitro translated MafB protein, a close relative of v-/c-Maf, was efficiently coprecipitated with the GST-Hoxd12 protein. All three small Maf subfamily proteins, MafK, MafF and MafG, also interacted with Hoxd12. However, c-Jun could only marginally bind to the GST-Hoxd12 protein, and c-Fos could not bind at all. The efficiency of coprecipitation of Jun (3% of input) was similar to those of the Maf mutants CD3, L2PL4P, MD26.22, and MD45 (2-4% of input, see Fig. 3B) and was much lower than those of Maf and the Maf family members (more than 20% of input). Again, essentially the same results were obtained using GST-MHox (data not shown). Thus, the interaction of the Hox proteins seemed to be specific for Maf family proteins.
DNA Binding Activity of Maf Is Inhibited by Interaction with Hox-The fact that Maf and Hox proteins interact with each other through their DNA-binding domains led us to examine the effect of the association on their DNA binding activities. For this purpose, a constant amount of linearized template plasmid containing maf (Nd5) was cotranscribed and translated in vitro with increasing amounts of hox template (1:1 to 1:5) and subjected to an electrophoretic gel mobility shift assay using the MARE probe (Fig. 5). The intensity of the retarded band of the Maf-DNA complex was decreased by the presence of increasing amounts of Hoxd12 and MHox proteins, while the homeodomain deletion mutant (Hoxd12⌬HD) had little effect (Fig. 5A). Coexpression of Pax6 also inhibited the DNA binding activity of Maf (Fig. 5B), whereas Meis1, which cannot associate with Maf, did not affect DNA binding. These results clearly indicated that the association of Hox proteins with Maf specifically inhibited its DNA binding activity.
On the other hand, cotranslation of Hoxd12, MHox, Pax6, or Meis1 had no effect on binding of the Jun homodimer to the AP-1 site (Fig. 6A). DNA binding of the Jun/Fos heterodimer also was not abrogated by Hoxd12 or MHox (Fig. 6B). These results again indicated specific interaction of Hox proteins with Maf.
We next constructed a chimeric Maf molecule whose leucine zipper was substituted by one from Jun. As expected, this molecule, Maf(Jun-zip), could efficiently form a homodimer and bind to MARE (Fig. 6C), because the DNA binding specificity of bZip proteins depends on the basic region. Importantly, the DNA binding activity of Maf(Jun-zip) was abrogated as efficiently as that of the Maf homodimer by Hoxd12. Based on these results, together with the fact that Hox proteins do not interact with Jun, and that the basic and leucine zipper regions of Maf are necessary for interaction with Hox (see Fig. 3), we believe that Hox proteins interact with the basic region of Maf only when it forms a dimer, and the leucine zipper can be substituted by those from other bZip proteins.
We next asked if Hox proteins could interact with heterodimers of Maf and other bZip proteins. For this purpose, we translated constant amounts of Maf and Fos together with increasing amounts of Hoxd12 and subjected the samples to gel mobility shift analysis using an oligonucleotide probe that can be efficiently bound by the Maf homodimer and Maf/Fos heterodimer (4). As shown in Fig. 6D, the intensity of the retarded band of the Maf/Fos heterodimer was decreased in the presence of Hoxd12 similarly to that of the Maf homodimer. One possible explanation for this inhibition of DNA binding is that Hoxd12 formed a ternary complex with Maf and Fos by interacting with the basic region of Maf. Although more careful examination is necessary to determine the mechanism of inhibition, these results suggest that Hox proteins interfere with the DNA binding activities of not only the Maf homodimer but also heterodimers of Maf and other bZip proteins.
DNA Binding of Hox Is Not Abrogated by Maf-Since the DNA-binding domains of Hox proteins can associate with Maf, we tested whether DNA binding of Hox proteins was also inhibited by Maf. Unexpectedly, binding of Hoxd12 protein to its recognition sequence was not affected by Maf (Fig. 7A). Maf also did not have any effect on binding of MHox protein to the muscle creatin kinase enhancer sequence (data not shown). We then examined the relative affinity of Hox proteins to DNA and to Maf. In vitro translated, 35 S-labeled Hoxd12 protein was mixed with increasing concentrations of oligonucleotide containing the Hoxd12-binding site or an unrelated sequence and then incubated with the MBP-Maf fusion protein immobilized on amylose resin. As shown in Fig. 7B, the amount of Hoxd12 protein coprecipitated with MBP-Maf was significantly decreased by addition of the Hoxd12-binding site oligonucleotide but not by the unrelated oligonucleotide, clearly demonstrating that the affinity of Hoxd12 protein to its binding site was higher than its affinity to the Maf protein. It thus seems reasonable that Maf is unable to compete with the Hox recognition sequence for the binding of Hox proteins.
Hoxd12 Inhibits Transactivation and Cell Transformation by Maf-The negative effect of Hox proteins on the DNA binding activity of Maf prompted us to investigate the effect of Hox proteins on the transactivation and cell transforming potential of Maf. The transactivation assay was performed on CEF cells using a luciferase reporter plasmid containing three tandem repeats of MAREs upstream of the TATA sequence of the rabbit ␤-globin promoter (3ϫMARE/RBGP-luc) (6,24). As shown in Fig. 8, cotransfection of the expression plasmid for Maf resulted in significant induction of luciferase activity, and the magnitude of transactivation was reduced by cotransfection of a Hoxd12 expression plasmid in dose-dependent manner. On the   lanes 2-12) was cotranslated in vitro with increasing amounts of template for Hoxd12 (lanes 3-5), Hoxd12⌬HD (lanes 6 -8), and MHox (lanes 10 -12) at a ratio of 1 :1 (lanes 3, 6, and 10), 1:3 (lanes 4, 7, and 11 (lanes 2 and 3) or a mutant Maf construct (Maf(Jun-zip)) whose leucine zipper was substituted by one from Jun (lanes 4 and 5), was cotranslated with (lanes 3 and 5) or without (lanes 2 and 4) Hoxd12, and tested for binding to MARE. D, effect of Hoxd12 on the Maf/Fos heterodimer. Constant amounts of template plasmids for Maf (ND5) and Fos (lanes 2-5) were cotranslated with increasing amounts of Hoxd12 template (lanes 3-5) and were subjected to gel mobility shift analysis using probe 11, which can be bound by both Maf/Fos heterodimer and Maf homodimer (indicated by arrows). The endogenous binding activity is indicated by an asterisk.
other hand, cotransfection of the Hoxd12⌬HD expression plasmid had little effect on the transactivation by Maf. We confirmed the nuclear localization of both Hoxd12 and Hoxd12⌬HD proteins in the transfected cells by producing GST fusion proteins and performing immunofluorescent staining of cells using anti-GST antiserum (data not shown). These results indicated that the Hoxd12 protein inhibits the transactivation activity of Maf by abrogating its DNA binding ability.
To measure the antagonistic effect of Hox proteins on transformation by Maf, we developed a replication competent avian retrovirus vector system (pRV-9), which can express two exogenous genes in a single cell (Fig. 9A). 2 Upon transfection into CEF cells, we expect one gene to be expressed from fully spliced subgenomic RNA and the other to be expressed from any RNA species by utilizing the IRES. We constructed pRV-9/c-maf-IRES-hoxd12 and pRV-9/c-maf-IRES-hoxd12⌬HD plasmids, transfected them into CEF cells, and assayed for focus formation. As shown in Fig. 9B, pRV-9/c-maf efficiently induced foci, but pRV-9/c-maf-IRES-hoxd12 did not. On the other hand, pRV-9/c-maf-IRES-hoxd12⌬HD induced foci with a similar efficiency to that of pRV-9/c-maf. Furthermore, coexpression of Hoxd12 did not interfere with the transforming ability of v-Fos (pRV-9/v-fos and pRV-9/v-fos-IRES-hoxd12). These results in-dicated that the Hoxd12 protein specifically suppressed the transforming ability of Maf.

DISCUSSION
In this study, we isolated two homeodomain-containing proteins, Hoxd12 and MHox, that interact with v-/c-Maf, using the phage display method. The Hox proteins also could associate with the other Maf protein family members, MafB, MafK, MafF, and MafG, but not with Jun and Fos. The Hox proteins negatively regulated the DNA binding, transactivation and cell-transforming abilities of Maf.
To date, several classes of transcriptional regulators are known to interact with Maf and/or Maf family proteins. For instance, Jun, Fos, (4,13) and Bach1 3 have been identified as heterodimeric partners of v-/c-Maf. In addition to these bZip proteins, several members of other transcription factor superfamilies have been reported to interact with Maf proteins. For instance, USF2, a member of the basic helix-loop-helix zipper transcription factor family, has been shown to interact with c-Maf and inhibit its DNA binding activity (34). c-Maf was also shown to form a transcriptionally inert complex with c-Myb in a developmentally regulated manner in cells of the myeloid lineage (22). Similarly, MafB has been reported to repress transcriptional activity of c-Ets-1 through direct interaction, which results in inhibition of erythroid cell differentiation (23). These transcription factors recognize the bZip domain of Maf proteins, which is highly conserved among the family members.
As shown in this study, Maf and Hox proteins can form complexes through interaction between the conserved bZip domain of Maf and the homeodomain of the Hox proteins. To date, more than 150 homeodomain-encoding genes have been identified in the vertebrate genome. Among them, Hoxd12 and MHox, identified in this study, are members of different subclasses and display higher amino acid sequence homology to other Hox proteins than to each other. These facts lead us to investigate whether other Hox proteins interact with Maf. We found that Pax6, whose homeodomain was more similar to MHox than to Hoxd12, could associate with Maf. In contrast, Maf could not bind to Meis1, whose homeodomain was distantly related to both Hoxd12 and MHox. These observations indicated that a set of, but not all, homeodomain-containing proteins could interact with Maf family proteins.
The Hox family transcription factors are known to play pivotal roles in the establishment of cell identity and regional information as well as for cell differentiation and morphogenesis. Maf also act as a differentiation factor in specific cell types, although it was originally identified as an oncogene product. For example, c-Maf is a key regulator for the specific expression of the interleukin-4 gene and differentiation of the Th2 subset of helper T-cells (35). The mafB gene has also been shown to establish specific rhombomeres in the developing hindbrain and is responsible for the kreisler and valentino mutations of mouse and zebrafish, respectively (36,37). As both Maf and Hox are transcriptional regulators, we examined the effect of their physical interaction on their functions and found that Hox proteins inhibited the DNA binding ability of Maf. In contrast, Maf could neither interfere with the DNA binding activities of Hox proteins or form a ternary complex with Hox bound to DNA, indicating that Maf was inert with regard to the biological activities of Hox proteins. However, interaction of Maf and Pax6 is of interest because Pax6 contains another DNA-binding motif paired domain in addition to the homeodomain. It therefore might be possible that Maf interacts with Pax6 at the homeodomain when Pax6 is bound to DNA by its paired domain. It previously has been shown that Pax6 and large Maf family members were expressed in overlapping regions of the eye and that both Pax6-binding sites and MAREs were identified in transcriptional regulatory regions of crystallin genes in a variety of species. Accordingly, large Maf proteins have been shown to activate lens specific expression of crystallin genes through such MAREs (38 -40). Pax6 also has been reported to activate the ␣A-crystallin gene of chicken (41) and mouse (42) and ␦-crystallin in chicken (43), and to activate synergistically with Maf, the -crystallin gene of guinea pig (44). On the other hand, Pax6 negatively regulates ␤-crystallin genes of chicken in lens fiber cells, while Maf activates these genes (38 -40, 45). The fine tuning of the spacio-temporal expression of a set of crystallin genes during lens development may require such synergistic and antagonistic regulation by Maf and Pax6, which might in part be accomplished by direct association of these two transcription factors.
Recently, it has been reported that a large Maf family protein, Nrl, and a paired-type homeodomain protein, Crx, synergistically regulate the photoreceptor cell-specific expression of the rhodopsin gene (46). Nrl is expressed not only in photoreceptor cells but also in other retinal cells (47). In contrast, Crx is exclusively expressed in photoreceptor cells, and synergistic activation of rhodopsin gene transcription is achieved only by the copresence of Nrl, which binds to a MARE-like cis-regulatory element, and Crx, which binds to another cis-element, Ret4, located next to the MARE on its promoter (46,48). Al-though direct physical interaction of Nrl and Crx has not been reported, such functional synergism between Maf family members and Hox proteins may occur depending on the context of the promoter.
As we have reported previously, the v-maf-carrying retrovirus, AS42, induces a specific type of tumor, musculoaponeurotic fibrosarcoma, in chicken (2), whereas v-Fos and v-Jun induce nephroblastoma and osteosarcoma (NK24 virus) (28) and fibrosarcoma (ASV17 virus) (49), respectively. Although the mechanism of tissue specificity in tumor induction is still not known, similar but clearly distinct DNA binding specificities of Maf and Fos/Jun may explain these differences. These oncogene products are likely to activate overlapping but different sets of downstream target genes. In addition, the finding that coexpression of Hoxd12 in CEF cells specifically inhibited the transcriptional activation and transforming ability of Maf, but not of Fos, suggests that the expression levels of a group of tissue-specific, Maf inhibitory factors such as USF2 or Hox proteins also may be important for oncogene-specific tumor induction. These two possibilities are not mutually exclusive and require further study.
Disregulated expression of some homeodomain-containing proteins also has been implicated in malignancies in humans and rodents. For example, Pax6 has been shown to be oncogenic when overexpressed (50), and chromosome translocations of pax3 and pax7 genes, close relatives of pax6, to the fkhr gene have been found in human rhabdomyosarcomas (51,52). It is especially noteworthy that not only MHox/Pmx1 itself but also HoxA9 and HoxD13, whose homeodomains are closely related to that of Hoxd12, have been found to form a fusion product with the nucleoporin, Nup98, by chromosome translocation in human acute myelogenous leukemia cells (33,(53)(54)(55). All these potentially oncogenic Hox proteins might interact with Maf family members, considering the similarity of the amino acids in their homeodomains. If these interactions do occur, together with recent findings that both c-Maf and MafB induce monocytic differentiation (56,57), then overexpression of these Hox proteins in cells of the myeloid lineage should inhibit differentiation-inducing functions of Maf proteins, which may result in inhibition of the differentiation program and may cause myelogenous leukemia. This idea of a common protein target for the action of oncogenic Hox proteins is worthy of further study, because DNA sequences recognized by these Hox proteins are different from each other, and it does not seem plausible that they share common downstream target gene(s) for cell transformation.
Recently, it was reported that MHox and a set of related homeodomain-containing proteins, Chx10, B4, and Pax3, interacted with the pRB family proteins, pRB, p107 and p130, through their homeodomains (58). Thus, Maf may compete with pRB family members for binding to these Hox proteins. Conversely, Hox proteins may compete for binding to Maf with other Maf-interacting molecules, such as USF2, c-Ets1, and c-Myb, because the COOH-terminal bZip domain of Maf is necessary for interaction with these three proteins (22,23,34). Moreover, in this study, we demonstrated that Hox proteins abrogated DNA binding of not only the homodimer of Maf family members but also the heterodimer of Maf and Fos, suggesting the possibility that Hox interferes with heterodimers of Maf and other bZip partners such as Jun and Cap'n'collar family members. Although the functional and physiological relevance of interactions between these transcription factors must be examined more carefully, we propose that these different classes of transcription factors form a regulatory network through protein-protein and protein-DNA interaction, which may allow the fine tuning of the regulation of gene expression. Disorder of the network by dysfunction or abnormal expression of a factor by mutation, viral transduction, or chromosome translocation may lead to a developmental abnormality, genetic disease, or cell transformation.