Structural and functional characterization of the transforming growth factor-beta -induced Smad3/c-Jun transcriptional cooperativity.

Smads are intracellular proteins that act as central effectors for transforming growth factor-beta (TGF-beta) and related proteins from the activated receptor into the nucleus, where they regulate ligand-induced gene expression. AP-1 binding sites have been functionally linked to the transcriptional activation of various genes in response to TGF-beta. Accordingly, we have previously shown that the heteromeric complex of Smad3 and Smad4 synergizes with c-Jun/c-Fos at the AP-1 binding site of the collagenase I promoter to induce transcriptional activation in response to TGF-beta. Using the collagenase I promoter as model system, we have now investigated the role of the c-Jun and Smad3 interactions with the promoter DNA and have further characterized the physical basis of the c-Jun/Smad3 interaction in the transcriptional response. Mutational analyses of the c-Jun protein and the AP-1 binding site in the promoter revealed that the interaction of c-Jun with DNA is necessary for transcriptional activation by TGF-beta and Smad3. Similar analyses of Smad3 and the Smad binding sites revealed that binding of Smad3 to DNA is also required, but that its DNA sequence-specific recognition is not essential. We also found that the basic leucine zipper domain of c-Jun and a short sequence close to the N terminus of Smad3 mediate their physical interaction, and that these regions are critical for their DNA-binding function. Our studies provide a basis for understanding the functional cooperativity of Smads with the diversity of transcription factors, which underlies the Smad-induced transcriptional activation in response to TGF-beta and related factors.

The transforming growth factor-␤ (TGF-␤) 1 superfamily constitutes a large group of secreted polypeptide signaling mole-cules, which play pivotal roles in a broad array of cellular processes, including cell proliferation and differentiation, apoptosis, deposition of extracellular matrix, and cell adhesion (1)(2)(3). This family of factors also regulates cell fate decisions and pattern formation during development in species from nematodes to vertebrates (4 -6). Many of the effects induced by TGF-␤ and related factors result from their ability to regulate transcription of specific sets of genes. The characterization of the molecular mechanisms of transcriptional regulation by these factors will provide insights into their role in normal physiological and pathological states.
TGF-␤ signaling is initiated by a cell surface heteromeric complex of type I and type II receptors, which are both transmembrane serine/theronine kinases (7)(8)(9). Upon ligand binding, the activated type II receptors phosphorylate and activate the type I receptor kinases, which then phosphorylate intracellular signaling mediators, including members of the Smad family of proteins. Smads have been identified as central mediators of the transcriptional effects of the TGF-␤ superfamily (9 -15). In vertebrates, nine Smad proteins have been identified thus far. Bone morphogeneic protein receptors phosphorylate and thereby activate Smads 1, 5, and 8, whereas activin/TGF-␤ receptors activate Smads 2 and 3. These pathway-specific Smad proteins then form heteromeric complexes with Smad4 in the cytoplasm, and subsequently translocate into the nucleus, where they induce or repress transcription of specific genes.
The mechanisms by which Smads regulate transcription are a subject of rapidly progressing research. Recent studies have revealed that three important functional characteristics are at the basis of the ability of Smads to activate transcription (11,14). First, Smads have an intrinsic DNA binding activity, which is mediated by a sequence in the conserved N-terminal domain, i.e. the N-or MH1-domain. Mad, the Drosophila homolog of Smads 1, 5, and 8, was first shown to be able to bind a GC-rich DNA sequence in the Dpp-responsive vestigial promoter (16), and, similarly, Smad4 interacts with GC-rich sequences in the Xenopus goosecoid promoter (17). In addition, Smads 3 and 4 bind to CAGA-or GTCT-like sequences in several other promoters (18 -23), and GTCTAGAC has been proposed as the optimal Smad binding element (24). Mutations of the "CAGA box" in the PAI-1 promoter (19), JunB promoter (20), and c-Jun promoter (22) severely diminish the TGF-␤ responsiveness. Second, Smads regulate transcription through functional and/or physical interactions with various transcription factors. For example, the Smad2/4 complex interacts with FAST-1 or -2 to activate transcription of the activin-responsive Xenopus Mix.2 or mouse goosecoid genes (17,(25)(26)(27). In addition, the Smad3/4 complex interacts with c-Jun or c-Jun/c-Fos at the TGF-␤-responsive collagenase I promoter (23) and the synthetic 3TP-promoter (18). Similarly, Smad3 interacts with the vitamin D receptor to mediate the effect of TGF-␤ on vitamin D 3 -induced transcription (28). Smad3 and Smad4 have also been shown to cooperate and interact with the basic helixloop-helix protein TFE3 to activate transcription from PAI-1 promoter (29,30). Third, the ability of Smads to activate transcription depends on their ability to interact with the general transcription machinery. Thus, the C-terminal sequence of the receptor-activated Smads interacts directly with CBP or p300, two closely related coactivators, which mediate the transcriptional activity of a variety of transcription factors (31)(32)(33)(34)(35). These biochemical and molecular studies have established a general model of transcriptional regulation by Smads, whereby the Smads activate transcription through physical interactions and functional cooperativity with selected transcription factors and through their ability to bind DNA sequences in the promoter. The relative contributions of the Smad binding sites in the transcriptional response seem to vary depending on the promoter (14,36,37). Besides the cooperating transcription factors mentioned above, Smads are likely to engage in interactions with various other transcription factors as well, and the promoter sequence requirements may accordingly substantially differ from each other. Given this information, it is important to characterize the roles of protein-protein interaction and protein-DNA interaction in transducing the transcriptional responses to TGF-␤ or to Smads.
We have previously shown that Smad3 and Smad4 synergize with c-Jun to mediate TGF-␤-induced immediate early transcriptional activation of a minimal human collagenase I promoter. This functional synergy correlates with the ability of Smad3 to interact directly with c-Jun, while c-Jun and Smad3 both can bind to an AP-1 binding promoter sequence and a partially overlapping Smad binding sequence, respectively (23). This functional cooperativity between Smad3 and c-Jun is of particular interest to understand the ability of TGF-␤ to activate transcription of a variety of genes. Many TGF-␤-inducible genes, such as PAI-1 (38), clusterin (39), type I collagen (40), interleukin 11 (41), retinoid acid receptors (42), and TGF-␤1 itself (43), contain AP-1 binding sites in the regulatory regions of their promoters and, in several cases, these AP-1 binding sites have been functionally linked to the transcriptional activation by TGF-␤ (38,40,41,43). Thus, the physical interaction of Smad3 with AP-1 family members in a nucleoprotein complex at the promoter (23,44) and the functional cooperativity of Smad3 with c-Jun or other AP-1 transcription factors (23,44,45) may represent a mechanism for transcriptional activation of a subset of genes by TGF-␤. Additionally, TGF-␤ enhances the expression of some AP-1 family members (20,22), suggesting that increased expression of these proteins may also account for or contribute to TGF-␤-mediated expression of some genes.
In this report, we further characterized the functional cooperativity of c-Jun and Smad3 in TGF-␤-induced transcriptional activation using the collagenase I promoter as model system. We have now: 1) examined the roles of the c-Jun and Smad DNA binding sites in transactivation of the collagenase I promoter, 2) assessed the relative contributions of DNA binding activities by c-Jun and Smad3 to TGF-␤-induced transcriptional activation and to their functional cooperativity, and 3) characterized the physical interaction between Smad3 and c-Jun, and defined the sequence requirement for their association. We find that the AP-1 binding site and the DNA-binding functions of both c-Jun and Smad3 are essential for promoter activation by TGF-␤. However, in contrast to the AP-1 binding site, the sequence of the Smad-binding site is less critical. We also find that the N-terminal domain of Smad3 and the basic leucine zipper domain of c-Jun mediate their physical interactions and are critical for their DNA-binding activity. Our studies provide a basis for understanding the functional cooperation of Smads with other transcription factors.
c-Jun and derivatives were expressed as N-terminally hemagglutinin-tagged versions. pRSV-c-Jun and the RSV vector have been described (23). The DNA fragments encoding wild-type c-Jun, c-JunV273, or c-Jun⌬RK were generated by PCR using pRSV-c-Jun, pSV-c-JunV273, or pSV-c-Jun⌬RK (46) as templates, and inserted into the EcoRI-HindIII sites of pXF1H, a pRK5 derivative with an hemagglutinin epitope tag inserted into the ClaI-EcoRI sites of pRK5 (32). The DNA fragments encoding c-Jun⌬C19 and c-Jun⌬LZ were generated by PCR and ligated into the PstI-HindIII sites of pXF1H-c-Jun. Details about plasmid constructions will be provided upon request.
Cell Cultures, Transient Transfection, and Luciferase Reporter Assays-Mv1Lu epithelial cells were maintained in Eagle's minimum essential medium supplemented with non-essential amino acids, 10% fetal bovine serum, 100 IU/ml ampicillin, and 100 g/ml streptomycin. F9 and HepG2 cells were maintained in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum, 100 IU/ml ampicillin, and 100 g/ml streptomycin. Transient transfections of Mv1Lu cells were performed in six-well tissue culture plates using the standard DEAE-dextran method (49). Transfections of F9 and HepG2 cells were carried out in six-well plates using LipofectAMINE according to manufacturer's instructions (Life Technologies, Inc.). For each transfection, 0.5 g of the luciferase reporter plasmid and, when indicated, 0.25 g of the Smad3 expression plasmid and 0.15 g of the c-Jun expression plasmid were used. The total amount of transfected DNA was kept constant by adding empty vector, as needed. Cotransfection of 0.15 g of the ␤-galactosidase expression plasmid (pSV-␤-Gal) allowed all transfections to be normalized to the ␤-galactosidase activity. TGF-␤ treatment and luciferase assays were carried out as described (50). All experiments were carried out in duplicate and repeated at least three times.
Generation of GST-fused Smad3 Proteins and in Vitro Protein Binding Assays-Plasmids pGEX-Smad3, pGEX-Smad3N, pGEX-Smad3NL, and pGEX-Smad3LC, which allow expression of GST-fused Smad3 or Smad3 fragments in Escherichia coli, have been described (23). GST-Smad3⌬H2 was made by subcloning a PCR-generated fragment encoding amino acid 48 -425 of Smad3 into the BamHI-SalI sites of pGEX-4T2 (Amersham Pharmacia Biotech). All other GST-Smad3 fusions were generated by subcloning PCR-generated cDNA fragments into the EcoRI/SalI sites of pGEX-5X-1 (Amersham Pharmacia Biotech). Detailed information about plasmid constructions will be provided upon request. The full-length and mutants of Smad3 fused to GST were expressed in E. coli and semi-purified by glutathione-Sepharose 4B adsorption according to manufacturer's recommendations (Amersham Pharmacia Biotech). 35 S-labeled c-Jun and mutants of c-Jun were generated by in vitro transcription and translation from the pXF1H-based expression vectors described above. In vitro protein binding assays to test the ability of these radiolabeled products to interact with Smad3 were performed as described. To avoid nonspecific interaction through DNA, the glutathione-Sepharose beads with adsorbed GST or GST-Smad3 proteins were washed with TE buffer (50 mM Tris-Cl, pH 7.5, 0.1 mM EDTA) containing 0.6 M NaCl to remove bacterial DNA. For each protein binding assay, 3-5 g of GST fusion protein bound to the beads was incubated with 35 S-labeled c-Jun or c-Jun mutant in binding buffer (20 mM Hepes, pH 7.95, 100 mM NaCl, 0.1 mM EDTA, 2.5 mM MgCl 2 , 1 mM dithiothreitol, 0.05% Nonidet P-40, 1% skimmed milk, and protease inhibitors) at 4°C for 1 h, then washed four times in binding buffer without milk. Specifically associated proteins were separated by SDS-polyacrylamide gel electrophoresis and visualized by autoradiography.
Electrophoretic Mobility Gel Shift Assays (EMSA)-Oligonucleotides, corresponding to the human collagenase I promoter, as indicated, were 5Ј-labeled with [␥-32 P]dCTP using the T4 polynucleotide kinase or 3Јlabeled with [␣-32 P]dCTP using Klenow fragment. The overlapping AP-1/Smad binding sequence in the wild-type collagenase I promoter oligonucleotide (Fig. 2) differed from the EMSA probe used in Zhang et al. (23) with one nucleotide in the SBE, i.e. AGAC in this study versus AGCC in Zhang et al. (23). Additionally, the commercially available oligonucleotide (Promega) used in the latter study (23) contained flanking sequences unrelated to the collagenase I promoter sequence. DNA binding assays were carried out as described (24) except that 26 g/ml poly(dI⅐dC) was used as a nonspecific competitor, and the preincubation of DNA and protein was carried out at 4°C. For each assay, 100 -300 ng of Smad3 proteins fused to GST were used. Excess unlabeled competitor oligonucleotide was added to the reaction mixture before preincubation, as required. The sequence of the double-stranded SBE oligonucleotide used as a probe was 5Ј-TAA AGC ATG AGT CTA GAC ACC TCT G-3Ј and its complementary strand. The nonspecific competitor oligonucleotides Oct1 and AP2 were purchased from Promega.
Methylation Interference Assays-To generate the methylation interference probes, the Ϫ73 to ϩ13 region of the collagenase I promoter was excised from D14-Luc or D14 m1m2m3-Luc using EcoRI and BamHI, and subcloned into pBluescript SK (Stratagene), generating pBS-D14 or pBS-D14 m1m2m3. The wild-type and mutant probes were then generated by restriction digestion using SstII and ClaI (to label the upper strand), or KpnI and XbaI (to label the lower strand), and labeled using [␣-32 P]dCTP and Klenow DNA polymerase. 0.5 l of DMS was then added per 100 l of labeling reaction mixture containing 500 ng of DNA. After incubation at room temperature for 5 min, 40 l of 1.5 M sodium acetate, 0.5 M dithiothreitol was added. The probes were then precipitated by addition of ethanol, dissolved in 100 mM Tris, pH 8.0, and used in EMSA as described above. After a short exposure of the gel at 4°C, the unbound free DNA probes and shifted complexes were excised and the DNA was electro-eluted, precipitated, and resuspended in 100 l of 1 M piperidine. The samples were incubated at 95°C for 30 min, and piperidine was removed by several rounds of lyophilization and resuspension in H 2 O. The samples were electrophoresed on DNA sequencing gels.
Gal4 Transactivation Assays-The plasmid encoding Smad3 fused to the DNA binding domain of Gal4 has been described (32). To obtain plasmids for Smad3⌬LG, Smad3R74D and Smad3(4A) fused to the Gal4 DNA binding domain, the DNA fragments encoding Smad3⌬LG, Smad3R74D, and Smad3(4A) were generated by PCR and subcloned into the EcoRI-SalI sites of the mammalian expression vector pXF1Gal4, a pRK5 derivative with the Gal4 DNA binding domain (amino acids 1-147) inserted between ClaI/EcoRI sites of pRK5 (32). The abilities of Smad3 and its derivatives to transactivate the heterologous Gal4 promoter reporter, pFR-Luc (Stratagene), were assayed essentially as described (32).

Role of the AP-1 Site and Smad Binding Sequences in TGF-
␤-induced Transcriptional Activation of the Human Collagenase I Promoter-We have shown that c-Jun synergizes with Smad3 to activate TGF-␤-induced transcription (23). To characterize this functional cooperativity at the collagenase I promoter, we used the luciferase reporter plasmid Ϫ73Col-Luc (47, FIG. 1. The ؊73 to ؊61 segment of the human collagenase I promoter is necessary for TGF-␤ and Smad3/4 responsiveness. A, sequence of the Ϫ73 to ϩ63 region of the human collagenase I promoter, illustrating the transcription start site, the TATA box, the AP-1 site, and seven potential Smad binding sites (SB1-SB7). B, deletion of the Ϫ73 to Ϫ61 sequence of the collagenase I promoter abrogates its responsiveness to TGF-␤ or to overexpression of Smad3/4. The Ϫ60Col-Luc reporter contains the Ϫ60 to ϩ63 region of collagenase I promoter. Indicated expression plasmids were transiently transfected into Mv1Lu cells, together with Ϫ73Col-Luc or Ϫ60Col-Luc, as indicated. Luciferase activity was measured in the presence (solid bars) or absence (empty bars) of TGF-␤ treatment. ␤-Galactosidase activity derived from cotransfected pSV-␤-Gal was used to normalize the transfection efficiency. Normalized luciferase activity (mean Ϯ S.D.) is expressed relative to that of Ϫ73Col-Luc in vector control transfected cells in the absence of TGF-␤. C, the reporter assays were performed as described for panel B, except that F9 cells were used, and a c-Jun expression plasmid was cotransfected, as indicated. 48). This plasmid contains the luciferase gene immediately downstream from the Ϫ73 to ϩ63 sequence of the human collagenase I gene, thus containing the proximal 73 bp of the promoter and the adjacent 63 bp of the 5Ј-untranslated sequence. This sequence, shown in Fig. 1A, contains a single AP-1 binding sequence (position Ϫ72 to Ϫ66) and the TATA box (position Ϫ32 to Ϫ25). Deletion of the most 5Ј 13-bp sequence of this promoter in plasmid Ϫ60Col-Luc abolished the responsiveness to TGF-␤ and to Smad3 and Smad4 in Mv1Lu cells (Fig.  1B). Similar results were obtained in F9 embryonic carcinoma cells, which have very low endogenous level of c-Jun and c-Fos (51) (Fig. 1C). These data indicate that this 13-bp sequence is essential for the transcriptional response to TGF-␤ and the transcriptional cooperativity of c-Jun with Smad3.
This 13-bp sequence of the human collagenase I promoter, which is required for TGF-␤ and Smad3/4 responsiveness, contains a consensus AP-1 binding site and an overlapping Smad binding site, which is identical to the proposed optimal Smad3/ 4-binding sequence, AGAC (SB1, Fig. 1A). Since both c-Jun and Smad3 are able to bind this DNA sequence, we examined the requirements of this AP-1 binding site and Smad binding site for TGF-␤-induced transcriptional activation. Inactivating mutations were introduced into each site individually, and the abilities of c-Jun or Smad3 to bind the mutated 13-bp oligonucleotides were assessed by EMSA (Fig. 2). In the mA oligonucleotide, a single base substitution in the AP-1 binding site abolished DNA binding of c-Jun, whereas mutation of the Smad binding sequence in the mS oligonucleotide did not affect c-Jun binding ( Fig. 2A). Conversely, Smad3 did not detectably interact with the mS oligonucleotide, but bound to the mA oligonucleotide similarly to the wild-type sequence (Fig. 2B).
To assess the effects of the mA and mS mutations on TGF-␤-induced transcription, we introduced these mutations individually into the Ϫ73Col-Luc plasmid, thus generating reporter plasmids mA-Luc and mS-Luc, respectively. As shown in Fig. 3A, inactivation of the AP-1 binding site (mA) markedly decreased the inducibility by TGF-␤ in Mv1Lu cells, whereas inactivation of the Smad binding site (mS) had no effect on the responsiveness to TGF-␤. In F9 cells, the expression of c-Jun was able to activate transcription from the mS but not from the mA mutated promoters, and consistent with this result, Smad3  Fig. 1A). A, c-Jun binds the WT and mS probes, but not the mA probe. Gel shift assays were performed using purified c-Jun protein and 32 P-labeled probes. c-Jun binding can be competed with a 25-fold excess of unlabeled WT or mS probes, but not with a 25-fold excess of an unlabeled oligonucleotide containing an AP-2 DNA binding site. B, GST-Smad3NL protein binds the WT and mA probes, but not the mS probe. Gel shift assays using semipurified GST-Smad3NL were performed as in panel A. A 100-fold excess of unlabeled oligonucleotide, including the unrelated Oct1 oligonucleotide, was used as competitor, as indicated.

FIG. 3. Effects of AP-1 site and Smad-binding sites of human collagenase I promoter on TGF-␤ and Smad3 responsiveness.
A and B, the AP-1 site is necessary for TGF-␤ and Smad3 responsiveness of the human collagenase I promoter. Luciferase reporters carrying a mutation in the AP-1 binding site, mA-Luc, or in the Smad binding site (SB1 in Fig. 1A), mS-Luc, were transfected into Mv1Lu cells (A) or F9 cells (B). Expression vectors for c-Jun and Smad3 were cotransfected into F9 cells, as indicated. TGF-␤-induced luciferase activity was measured. Normalized luciferase activity is presented relative to that of Ϫ73Col-luc in untreated cells. C, mutations of the Smad-binding sites of the human collagenase I promoter moderately decrease TGF-␤ or Smad3 responsiveness. Promoter deletion mutant D14 contains the Ϫ73 to ϩ13 region of collagenase I promoter, with three potential Smad-binding sites at positions Ϫ66 (SB1), Ϫ56 (SB2), and Ϫ24 (SB3) (see Fig. 1A). D14 m1m2 contains double mutations in SB1 and SB2, and D14 m3 contains a mutated SB3, while D14 m1m2m3 contains mutations in all three sites. These reporter constructs were transfected into Mv1Lu cells, with or without cotransfection of Smad3, and TGF-␤-induced luciferase activity was measured. Normalized luciferase activity is presented relative to that of Ϫ73Col-Luc in vector control-transfected cells in the absence of TGF-␤. and c-Jun cooperated to induce transcription from the mS promoter, similarly to the wild-type promoter, but not from the mA promoter (Fig. 3B). Together, these results indicate that the AP-1 site is critically important for TGF-␤-induced transcription activation in the context of the collagenase I promoter, but that this Smad binding site is not essential.
The transcriptional responsiveness of the mS-Luc promoter to TGF-␤ and Smad3 raised the possibility that the Ϫ73 to ϩ63 region in Ϫ73Col-Luc contain additional Smad binding sites. This DNA segment contains seven sequences that resemble the proposed consensus Smad3/4-binding sequence, i.e. AGAC, GTCT, or CAGA, three of which are located in the promoter (Fig. 1A). To assess the contributions of these potential binding sites to the TGF-␤ responsiveness, we first generated plasmid D14-Luc. This reporter plasmid contains only the Ϫ73 to ϩ13 promoter segment and therefore does not incorporate the ϩ14 to ϩ63 sequence of the 5Ј-untranslated region with its four potential Smad binding sequences. TGF-␤ or overexpression of Smad3 stimulated transcription from the D14-Luc reporter plasmid similarly to transcription from the Ϫ73Col-Luc reporter (Fig. 3C). Mutation of both the first and second putative Smad binding sites in plasmid D14 m1m2, or of the third Smad binding site in plasmid D14 m3, only moderately affected the activation by TGF-␤. In addition, mutation of all three sites (plasmid D14 m1m2m3) still did not abolish, yet further decreased the TGF-␤-and Smad3-induced promoter activation (Fig. 3C).
Since mutation of all three putative Smad binding sites did not abolish, yet strongly decreased, TGF-␤-induced transcription, we assessed Smad3 binding to the mutated promoter sequence used in the D14 m1m2m3 plasmid. Two separate oligonucleotides, each corresponding to one half of the Ϫ73 to ϩ13 region of the mutant promoter were synthesized and used as probes in EMSA assays. As shown in Fig. 4A, Smad3 bound to the mutated promoter sequence without canonical Smad binding sites, albeit with a lower efficiency than to the wildtype sequence (data not shown; Fig. 4B). The affinity of Smad3 for the wild-type and mutant promoter was further examined in competition experiments, whereby wild-type and mutant promoter DNA were used as competitors of Smad3 binding to the wild-type promoter. As shown in Fig. 4B, the wild-type DNA was 5-fold more efficient than the mutant in competing with the DNA binding of Smad3. We also performed DMS methylation interference assays with the wild-type promoter DNA sequence. When the upper strand was radiolabeled, these analyses showed that Smad3 contacted only the guanine residue of the SB1 site within the Ϫ73 to ϩ13 region of the collagenase I promoter (Fig. 4C), and no interference was detected using the m1m2m3 mutant (data not shown). Finally, footprinting analyses using the Ϫ73 to ϩ13 promoter sequence did not reveal any protected nucleotides in the wild-type and mutated promoter sequences, under conditions which detected strong c-Jun binding to the AP-1 binding site (data not shown).
Although two copies of a composite AP-1/Smad binding sequence allowed for footprinting analyses of Smad3 (23), a single copy of such sequence does not confer a sufficiently high affin-

FIG. 4. Smad3 binding to the wild-type and mutated collagenase I promoter sequences.
A, GST-Smad3 proteins bind oligonucleotides, corresponding to D14 m1m2m3, in EMSA assays. Two scanning oligonucleotides, corresponding to Ϫ68 to Ϫ29 (oligo 1) and Ϫ28 to ϩ11 (oligo 2) of the mutated promoter D14 m1m2m3 were used as 32 P-labeled probes. A 200-fold excess of unlabeled oligonucleotide was used as competitors, as indicated with ϩ. B, competition of Smad3 binding to the collagenase I promoter by wild-type and mutated DNA fragments. 32 P-Labeled DNA fragments corresponding to wild-type D14 (nucleotides Ϫ73 to ϩ 13, Fig. 1A) were incubated with GST-Smad3NL. Progressively increasing amounts of unlabeled D14 wild-type or D14 m1m2m3 DNA fragments, as indicated, were used as competitors. C, methylation interference implicates the SB1 site as the major Smad3 binding site. A DNA fragment corresponding to the wild-type D14 promoter was 32 P-labeled at the 3Ј-end of the upper (U) or lower (L) strand, methylated with DMS, and incubated with GST-Smad3NL. After EMSA, free (F) and protein-bound (B) probes were cleaved with piperidine and separated on DNA sequencing gels. The asterisk marks the G nucleotide in the SB1 sequence that shows methylation interference. The same result was obtained in four independent assays. ity of Smad3 binding to allow Smad3 footprinting (data not shown). These data are consistent with the low affinity binding of Smad3 to DNA, and suggest that the SB1 site is the major Smad binding site in the promoter. Collectively, these results suggest that mutation of the SB1 site and other potential Smad binding sites did not abolish but decreased the affinity and efficiency of Smad binding to the promoter DNA.
The DNA Binding Activities of c-Jun and Smad3 Are Required for TGF-␤-induced Transcriptional Activation-As a complementary approach to the site-directed mutagenesis study of the promoter, we tested the effects of two c-Jun mutants, which lack DNA-binding ability. c-JunV273 contains a point mutation at position 273 with an Arg to Val replacement in the basic DNA binding region, whereas c-Jun⌬RK has a small deletion of the DNA binding region (46). As shown in Fig.  5, these mutants were unable to cooperate with Smad3 to induce transcription from the collagenase I promoter. This finding and the inability of TGF-␤ or Smad3/4 to activate transcription from the mA collagenase I promoter with its mutated AP-1 binding site, strongly suggest that DNA binding of c-Jun is essential for the TGF-␤ response through Smad3/4.
To better assess whether the DNA binding activity of Smad3 is required for TGF-␤-induced transcription from the collagenase I promoter, we generated two Smad3 mutants, which, based on the proposed crystal structure of the Smad3 N-domain/DNA interface (21), are predicted to lack DNA binding. In the ⌬LG mutant of Smad3, the 12-amino acid ␤ hairpin sequence which mediates DNA binding (amino acids 71-82) was deleted, whereas in the R74D mutant, Arg-74, which directly interacts with the guanosine in the GTCT Smad binding element, was replaced by Asp. As expected, these Smad3 mutants were unable to bind to the AP1/Smad-binding oligonucleotide from the collagenase I promoter or to the proposed optimal Smad binding sequence (Fig. 6A). We next examined the effects of these mutations on the transcriptional activity of Smad3 using Gal4 transactivation assays, in which the Smad3 mutants were fused with the Gal4 DNA binding domain. As shown in Fig. 6B, TGF-␤ induced transcriptional activation of Gal4-Smad3⌬LG and Gal4-Smad3R74D to a level similar to that of Gal4-Smad3. We also assessed the abilities of these two mutants to interact with c-Jun using GST pull-down assays. Bacterially expressed GST fusion proteins containing wild-type Smad3 or the NL segment of Smad3 bound to c-Jun ( Fig. 6C; Ref. 23). The ⌬LG mutation did not affect Jun binding, whereas the R74D mutation resulted in a reduced interaction with c-Jun (Fig. 6C).
To evaluate the effect of the DNA binding of Smad3 on transcriptional activation, we tested the abilities of Smad3⌬LG and Smad3R74D to cooperate with c-Jun to activate transcription from the collagenase I promoter. As shown in Fig. 6D, Smad3⌬LG or Smad3R74D were unable to activate transcription or to cooperate with c-Jun, which is in contrast to wild-type Smad3. Western blot analyses of transfected cell lysates showed that the wild-type and mutant Smad3 proteins were expressed at comparable levels (data not shown). Together, our findings strongly suggest that DNA binding of Smad3 is required for transcriptional activation in the context of the collagenase I promoter.
Lysine 41 of Smad3 Is Critical for Interaction with c-Jun and Binding to DNA-Our earlier studies show that the N-domain and the linker (L) segment of Smad3 (amino acids 1-221) and the C-terminal segment of c-Jun (amino acids 235-331) are sufficient to mediate the Smad3/c-Jun interaction (23). To assess the role of this interaction in the functional cooperation of Smad3 and c-Jun, we first defined the sequence within Smad3 that is required for association with c-Jun. To this end, we made GST fusion proteins containing various truncations of the N-L segment of Smad3 (Fig. 7A). As shown in Fig. 7B, deletion of only the N-terminal 47 amino acids of Smad3 in the Smad3⌬H2 mutant was sufficient to abolish the direct interaction of Smad3 with c-Jun. The crystal structure of the Smad3 N-domain bound to DNA predicts that the four lysines at positions 40, 41, 43, and 44 are exposed and constitute part of an ␣-helix (21). This configuration suggested that this region may mediate protein-protein interactions. Thus, we made a GST-Smad3 fusion protein with a deletion of residues 40 to 47 (Smad3⌬KQ). This GST fusion protein was unable to interact with c-Jun in pull-down assays (Fig. 7B), suggesting that the deleted 8-amino acid sequence in the N-domain of Smad3 is required for interaction with c-Jun. To assess the role of the four lysines in this interaction, we replaced Lys-40, Lys-41, Lys-43, and Lys-44 with alanines to minimize the structural perturbation of the ␣-helix. As shown in Fig. 7B, replacement of all four lysines in the Smad3(4A) or Smad3NL(4A) mutants completely abolished its association with c-Jun. In addition, the single replacement of lysine 41 with alanine in Smad3(K41A) strongly reduced the interaction of Smad3 with c-Jun, whereas individual mutations of the three other lysines did not significantly affect association with c-Jun (Fig. 7B).
We also tested the effects of these mutations on the DNAbinding activity of Smad3 using EMSA assays. Smad3⌬H2 and Smad3⌬KQ were unable to bind to the proposed optimal Smad binding sequence, and replacement of all four lysines with alanines similarly abolished DNA binding. Finally, Smad3(K41A) showed a strongly decreased affinity for the optimal Smad-binding DNA sequence, whereas the other individual lysine mutations did not significantly affect DNA binding (Fig. 7C).
We next examined whether Smad3(4A) and the Smad3 mutants with individual lysine to alanine mutations could synergize with c-Jun to transactivate the collagenase I promoter. As shown in Fig. 7D, Smad3(4A) showed only a minimal level, if any, of transcriptional cooperativity with c-Jun. This impaired cooperativity of Smad3(4A) was not due to an effect of this mutation on the transactivation capacity of Smad3 per se, since Smad3(4A) fused to a Gal4 DNA binding domain was transcriptionally as active as wild-type fused to the same sequence, both

Smad3/c-Jun Cooperativity in TGF-␤ Response
in the absence or the presence of TGF-␤ (data not shown). In addition to Smad3(4A), Smad3(K41A) was also unable to cooperate with c-Jun and to mediate TGF-␤-induced transcription, whereas Smad3(K40A), Smad3(K43A), and Smad3(K44A) all synergized with c-Jun in a TGF-␤-dependent manner, similar to wild-type Smad3. These Smad3 mutants were expressed at comparable levels in transfected cells as assessed by Western blot analyses (data not shown). Taken together, these data show that lysine 41 is critical for the ability of Smad3 to bind to DNA, interact with c-Jun and functionally cooperate with c-Jun.
The Leucine Zipper Domain of c-Jun Is Required for Interaction with Smad3-We previously found that amino acids 235-331 of c-Jun is sufficient to mediate the direct interaction with Smad3 and that the segment from amino acids 1-221, which contains the transactivation domain, does not interact with Smad3 (23). To better define the sequence required for interaction with Smad3, we made several deletion mutants in the C-terminal segment of c-Jun (Fig. 8A). This segment contains the basic DNA binding domain (amino acids 252-279), followed by the leucine zipper domain (amino acids 280 -315) and a C-terminal 16-amino acid extension. Deletion of the C-terminal 19 amino acids of c-Jun in the c-Jun⌬C19 mutant did not affect its ability to directly interact with Smad3, whereas deletion of the leucine zipper domain and the C-terminal 16 amino acids in c-Jun⌬LZ abolished Smad3 binding (Fig. 8B). This result suggests that the leucine zipper domain of c-Jun is essential for interaction with Smad3. In addition, deletion of the DNA binding segment in c-Jun⌬RK strongly reduced its ability to associate with Smad3 (Fig. 8B). Finally, the DNA-binding defective mutant c-JunV273 still associated with Smad3, albeit with lower efficiency (Fig. 8B). Taken together, these results show that the leucine zipper domain of c-Jun is essential for association with Smad3, and that an intact DNA binding domain greatly contributes to this interaction.
The abilities of these c-Jun mutants to cooperate with Smad3 were tested in F9 cells again using the TGF-␤-responsive 73-bp collagenase I promoter segment. c-Jun⌬C19 synergized as efficiently as wild-type c-Jun in inducing transcription from the collagenase I promoter, further supporting the notion that this C-terminal segment is not required for association with Smad3. In contrast, c-Jun⌬LZ was inactive and did not cooperate with Smad3 (Fig. 8C). Western blot analyses of transfected cell lysates showed that the c-Jun mutants were expressed at similar levels (data not shown).
FIG. 6. The DNA-binding function of Smad3 is required for functional synergy with c-Jun. A, the NL segments of Smad3⌬LG and Smad3R74D are unable to bind DNA. Gel shift assays were performed using GST-fused Smad3 proteins, containing defined segments of Smad3, and the 32 P-labeled WT probe (shown in Fig. 2) or SBE probe, which contains the proposed Smad consensus binding site (5Ј-TAAAGCATGAGTCTAGACACCTCT-G-3Ј). B, the transcriptional activity of Smad3⌬LG and Smad3R74D is similar to that of Smad3. Plasmid pFR-Luc, which contains five tandem Gal4 DNA-binding sequences, was used as a reporter to score the transcriptional activity of the Gal4 fusion proteins. Expression plasmids encoding the indicated Gal4-Smad3 fusion proteins, together with pFR-Luc, were transfected into HepG2 cells, and the TGF-␤-induced luciferase activity was measured. Gal4 DNA binding domain (Gal4-DBD) was expressed as a control. C, the NL segments of Smad3⌬LG and Smad3R74D interact with c-Jun. The Smad3 mutants fused to GST were expressed in E. coli and purified as described under "Experimental Procedures." Equal amounts of GST or GST-Smad3 fusions, containing defined segments of Smad3, were coupled to glutathione-Sepharose beads and incubated with 35 S-labeled, in vitro translated c-Jun. Associated c-Jun was detected by SDS-polyacrylamide gel electrophoresis and autora-diography. Lower panel, Coomassie Brilliant Blue staining of the same gel demonstrates the similar amounts of GST fusion proteins. D, Sma-d3⌬LG and Smad3R74D do not collaborate with c-Jun to transactivate the collagenase I promoter in F9 cells. Indicated expression plasmids and Ϫ73Col-Luc reporter were transfected into F9 cells, and luciferase activity was measured in the presence or absence of TGF-␤.

DISCUSSION
TGF-␤ family members elicit a wide range of transcriptional responses, which may result from the ability of receptor-activated Smads to cooperate with other transcription factors and thereby provide a high level of versatility in the cellular responses to TGF-␤. We have previously shown that Smad3 and Smad4 cooperate with AP-1, composed of c-Jun and c-Fos, to mediate TGF-␤-induced immediate early transcriptional activation of the human collagenase I promoter, which is apparent after 2 h of TGF-␤ treatment. This cooperation is primarily mediated by the ligand-induced interaction of Smad3 and c-Jun, which both can bind DNA and directly associate with each other (23). Consistent with our observations, JunD and FosB have also been shown to form a complex with Smad3 at an AP-1 binding site (44). We have now evaluated the roles of the DNA binding activities of Smad3 and c-Jun in their functional synergy and have defined the c-Jun and Smad3 sequences that mediate their association. Our results strongly suggest that the functional collaboration between c-Jun and Smad3 depends on the protein-DNA interactions of both c-Jun and Smad3.
Our results demonstrate the essential role of the DNA binding of c-Jun in TGF-␤-induced transcription from the collagenase I promoter. A point mutation or a small deletion in the basic region of c-Jun renders it unable to bind to the AP-1 site, and consequently inactivates the transcriptional activity of c-Jun and its ability to synergize with Smad3. Conversely, a point mutation in the AP-1 binding site that abolishes c-Jun binding, without affecting Smad binding, abolishes activation of this promoter by TGF-␤ or by overexpression of Smad3. These observations are in agreement with a previous report using an artificial promoter with four tandem repeats of a AP-1 sequence from the collagenase I promoter (18).
In contrast to the inactivating effect of mutations in the AP-1 binding site, mutations in all three potential Smad-binding sites within the collagenase I promoter only moderately reduced the response to TGF-␤ or to Smad3 overexpression. This result suggests that the DNA sequence requirements for Smad3 binding may not be as critical as originally thought. Our observations at the collagenase I promoter resemble those at the minimal enhancer fragment of the Drosophila labial promoter, which confers a robust Dpp responsiveness. This promoter fragment contains four CRE sites and two Mad-binding sites. Mutation of both Mad-binding sites only moderately decreases the transcriptional activation by Dpp, whereas mutation of the CRE sites substantially reduces the activity of this enhancer (36). Similarly, mutation of the FAST-1-binding site in the Xenopus Mix.2 promoter eliminated its response to activin, while mutation of the Smad4-binding site reduced, but did not eliminate the activin response (17,37). Finally, in the artificial 3TP-Lux promoter, which contains AP-1 and Smad4- binding sites, the AP-1 site is critically important for TGF-␤ responsiveness, whereas mutation of the Smad-binding sites has no effect on transcriptional activation by TGF-␤ or overexpressed Smad3 (18). These results have been largely interpreted to mean that DNA binding by Mad or Smads may not be necessary for Mad/Smad-induced transcriptional activation. In contrast, our current studies suggest that the ability of Smad3 to bind DNA is essential for TGF-␤-inducible transcription and functional cooperation with c-Jun, but that the DNA sequence requirements for Smad3 binding are not stringent. This conclusion is based on two sets of data. First, the Smad3 mutant Smad3⌬LG, which is unable to bind DNA, did not synergize with c-Jun in transcription assays, even though it maintained its transcriptional activity (Fig. 6B) and ability to interact with c-Jun (Fig. 6C). These data demonstrate that DNA binding by Smad3 is essential for promoter activation, and that c-Jun/ DNA interaction is not sufficient to confer TGF-␤ responsiveness to the collagenase I promoter. Second, mutation of all three putative Smad binding sites only moderately decreased, but did not abolish, the transcriptional activation by TGF-␤ or overexpressed Smad3. To address whether a non-consensus Smad binding site exists in the collagenase I promoter, we carried out methylation interference, footprinting and gel shift assays. Methylation interference experiments revealed that the AGAC sequence, i.e. the SB1 site, which partially overlaps with the AP-1 binding site (Fig. 1), has the highest affinity for Smad3. Mutation of this sequence no longer allowed for methylation interference by Smad3, yet allowed for low efficiency Smad3 binding. Even mutation of all three potential Smad binding sites did not abolish Smad3 binding to the DNA, although the binding efficiency was strongly decreased (Fig. 4B). Together with the transcriptional assay data, our results sug-gest that a consensus Smad binding site is not absolutely required, but that the interaction of Smad3 with the promoter DNA in a non-sequence-specific manner is essential. Considering the low affinity and specificity of DNA binding by Smad3, this non-sequence-specific DNA contact by Smad3 may serve to facilitate or stabilize the formation of a higher order transcription complex. Consistent with this notion, increased levels of Smad4 enhance transcription from a mutated Mix. 2 promoter, in which the Smad4 binding site was mutated (37).
We previously found that the C-terminal third of c-Jun (amino acids 235-331) is sufficient for interaction with Smad3 (23). We have now extended this observation and show that c-Jun interacts through its basic DNA binding domain and leucine zipper domain (bZip) with the N-domain of Smad3. This finding is similar to previous observations that the bZip domain of AP-1 not only participates in DNA-binding, but also serves as a protein-interacting surface for other transcription factors. For example, c-Jun/c-Fos can interact with the glucocorticoid receptor, and mutations in the bZip domain of c-Jun or in the outward-facing residues of the basic region of c-Fos abolish this interaction (52). Similarly, in the complex of nuclear factor of activated T cells (NFAT) with AP-1, the DNA-binding domains of c-Jun and NFAT are sufficient for physical association of both proteins and necessary for their functional cooperation (53). Our assignment of the sequence in c-Jun that interacts with Smad3 differs from a recent study, which localized the Smad3-interacting sequence of c-Jun to the C-terminal 20 amino acids of c-Jun or JunB, as assessed using GST adsorption assays (45). The basis for this discrepancy is unclear.
We also defined a sequence in Smad3 that is of critical importance for association with c-Jun and may represent the interface of protein interaction. The sequence from amino acid 40 to 47 in the N-domain of Smad3 has a predicted ␣-helical structure and contains four lysine residues at positions 40, 41, 43, and 44. Deletion of amino acids 40 -47 or replacement of these four lysine residues with alanine abolishes the ability of Smad3 to interact with c-Jun. These same mutations unexpectedly also abolished the ability of Smad3 to bind the "optimal" Smad-binding DNA sequence. Of particular interest is that a single amino acid replacement of lysine 41 with alanine, presumably with minimal effect on the helical structure of this region, impaired the interaction of Smad3 and c-Jun and their functional cooperativity, and also decreased Smad3 DNA-binding activity. These data suggest that the same sequence in the Smad3 N-domain, particularly lysine 41, plays an important role in both interaction with c-Jun and DNA binding. Since the sequence required for the c-Jun/Smad3 interaction colocalizes with the sequence involved in DNA binding, it is not possible to pinpoint the role of the physical association of c-Jun and Smad3 in their transcriptional cooperativity. The colocalization of sequences that serve as protein-protein interfaces and DNAbinding domains is not uncommon for transcription factors. For example, in the nuclear complexes of heat shock factor 3 (HSF3) and c-Myb, the winged helix-turn-helix DNA-binding domain of HSF3 associates directly with the DNA binding domain of c-Myb (54). Other examples include the interaction between the homeodomain protein, MAT␣2, and the MADS box protein, MCM1 (55), and the association of the Ets factor, Pu.1, with the lymphoid-restricted interferon regulatory factor, IRF-4 (56).
Although c-Jun and Smad3 are able to interact with each other and with the oligonucleotide comprising the AP-1 binding site and SB1 sequence of the collagenase I promoter, we were unable to detect the formation of a TGF-␤-induced complex of endogenous Smad3 and c-Jun at this promoter sequence (data not shown). This inability may reflect the limited quality of the Smad3 antibodies available to us, the low levels of endogenous Smad3 protein, and the low affinity of Smad3 for a single Smad binding sequence in the collagenase I promoter. In contrast, binding of endogenous Smad3 to promoter sequences with multiple Smad binding elements has been demonstrated (19,22,57,58), suggesting that multiple binding sites confer a higher affinity to the oligomeric Smad3 complex than a single Smad binding DNA sequence. Although the previous (23,44) and current studies strongly support the formation of a transcriptionally active c-Jun/Smad3 complex at the collagenase I promoter, our inability to demonstrate a complex of endogenous Smad3 and c-Jun at the collagenase I promoter raises the possibility that TGF-␤-or Smad3-induced expression of c-Jun (22) or JunB (20) could also contribute to the activation of collagenase I transcription in response to TGF-␤. Such indirect response mechanism may then amplify or prolong the TGF-␤induced immediate early transcription of the collagenase I promoter.
Smads have the ability to cooperate with several other transcription factors in response to TGF-␤ and related factors. Whereas our results are limited to the TGF-␤-induced c-Jun/ Smad3 interaction, the same principles are likely to hold for the interaction of receptor-activated Smads with a variety of transcription factors. We expect that, also in those cases, the specificity of interaction with the DNA is primarily provided by the transcription factor which interacts with the Smad. The Smad, with its modest binding affinity and specificity, then provides further selectivity for a subset of promoter sites, yet allows increased stability/or affinity of binding of the transcription complex with DNA. In these systems, the functional cooperativity then depends on the physical interaction of the Smad with its transcription factor partner and on the ability of the Smad to interact with DNA. Further characterization of other systems of Smad-mediated transcription will be required to test the generality of these principles.