ATF-2 Is a Common Nuclear Target of Smad and TAK1 Pathways in Transforming Growth Factor-β Signaling*

Upon transforming growth factor-β (TGF-β) binding to its cognate receptor, Smad3 and Smad4 form heterodimers and transduce the TGF-β signal to the nucleus. In addition to the Smad pathway, another pathway involving a member of the mitogen-activated protein kinase kinase kinase family of kinases, TGF-β-activated kinase-1 (TAK1), is required for TGF-β signaling. However, it is unknown how these pathways function together to synergistically amplify TGF-β signaling. Here we report that the transcription factor ATF-2 (also called CRE-BP1) is bound by a hetero-oligomer of Smad3 and Smad4 upon TGF-β stimulation. ATF-2 is one member of the ATF/CREB family that binds to the cAMP response element, and its activity is enhanced after phosphorylation by stress-activated protein kinases such as c-Jun N-terminal kinase and p38. The binding between ATF-2 and Smad3/4 is mediated via the MH1 region of the Smad proteins and the basic leucine zipper region of ATF-2. TGF-β signaling also induces the phosphorylation of ATF-2 via TAK1 and p38. Both of these actions are shown to be responsible for the synergistic stimulation of ATF-2trans-activating capacity. These results indicate that ATF-2 plays a central role in TGF-β signaling by acting as a common nuclear target of both Smad and TAK1 pathways.

Members of the Smad group of proteins mediate TGF-␤, 1 BMP (bone morphogenetic protein), and activin signaling from receptors to nuclei (for review, see Refs. 1 and 2). Smad2 and Smad3 are substrates and mediators of the related TGF-␤ and activin receptors in vertebrates (3)(4)(5)(6)(7). TGF-␤ first directly binds to the TGF-␤ type II receptor and leads to the formation of an oligomeric complex of the type I and type II receptors (8). Upon ligand binding, the C-terminal ends of these Smad proteins, which bind directly to the type I receptor, are phosphorylated by the type I receptor. This results in their release (7) and hetero-oligomerization with Smad4, a common-mediator of Smad (9 -11). Hetero-oligomers of Smad move into the nucleus and directly participate in TGF-␤and activin-dependent transcriptional activation (12)(13)(14). Smad2 and Smad4 interact with FAST-1, a member of the winged-helix transcription factor family, and mediate activin-dependent transcriptional activation (13,14). Recently, the N-terminal regions of Drosophila Mad and mammalian Smad3 and Smad4, which are conserved in the Smad gene family, were shown to interact with specific DNA sequences, and the direct binding of Smad3/4 to DNA is critical for the TGF-␤-induced transcriptional activation (15)(16)(17)(18).
In addition to the Smad group of proteins, another pathway involving a member of the MAPKKK family of kinases, TAK1 (TGF-␤-activated kinase), is also known to be involved in TGF-␤ signaling (19). TAB1 and TAB2 were identified as proteins that directly bind to TAK1 (20). Overexpression of TAB1 enhances the activity of the plasminogen activator inhibitor 1 (PAI-1) gene promoter, which is regulated by TGF-␤, and increases the kinase activity of TAK1, suggesting that TAB1 is an upstream regulator of TAK1. Furthermore, TAK1 activates stress-activated protein kinases (SAPKs), p38 through MKK6 or MKK3 (21) and c-Jun N-terminal kinases (JNKs) via MKK4 (22). Since MKK4 can also activate p38 (23,24), TAK1 may activate p38 via MKK4. However, it is unknown how the Smad and TAK1 pathways function together to synergistically amplify TGF-␤ signaling.
Recently, the cAMP response element (CRE) in the Ultrabithorax (Ubx) gene enhancer was shown to mediate transcriptional activation by Dpp, a Drosophila homologue of TGF-␤/ BMP (25). In addition, mutation of the AP-1 sites of the collagenase promoter eliminated TGF-␤-dependent transcriptional activation (16). The sequences of the CRE and AP-1 sites (12-O-tetradecanoylphorbol-13-acetate response element,) are similar to each other, and ATF/CREB and members of the Jun family of proteins bind to these sites, respectively (26). So far, a number of transcription factors of the ATF/CREB family have been identified. All members of this family contain a DNA binding domain consisting of a cluster of basic amino acids and a leucine zipper region, the so-called b-ZIP (for review, see Ref. 27). They form homodimers or heterodimers through their leucine zipper regions and bind to CRE. Among many of the transcription factors of the ATF/CREB family, two factors, CREB (28,29) and ATF-2 (also called CRE-BP1) (30 -32), are the best characterized. CREB is activated via direct phosphorylation by cAMP-dependent protein kinase (33). On the other hand, SAPKs such as JNKs and p38 phosphorylate ATF-2 at Thr-69, Thr-71, and Ser-90 which lie close to the N-terminal transcriptional activation domain and stimulate its trans-activating capacity (34 -36). Thus, these two groups of factors, CREB and ATF-2, are linked to distinct signaling cascades * The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. § On leave from the Institute of Medical Science, University of Tokyo, 4-6-1 Shirokanedai, Minato-ku, Tokyo 108-0071, Japan.
involving the cAMP-dependent protein kinase and SAPK pathways. ATF-2, ATF-a, and CRE-BPa form a subgroup (30,37,38) and have a transcriptional activation domain containing the metal finger structure located in their N-terminal regions (38,39). These factors bind to CRE with high affinity as a homodimer or heterodimer with c-Jun (26,40). Among these three factors, ATF-2 has been more extensively studied, and shown to be ubiquitously expressed, with the highest level of expression being observed in the brain (41). Mutant mice generated by gene targeting exhibited lowered postnatal viability and growth, in addition to a defect in endochondrial ossification and a reduced number of cerebellar Purkinje cells (42).
The fact that ATF-2 activity is enhanced by SAPK whose activity in turn is stimulated by TAK1 allowed us to hypothesize that ATF-2 might play an important role in the TGF-␤ signal transduction pathway. Our results indicate that ATF-2 not only directly binds to Smad3/4 hetero-oligomers but also that ATF-2 is phosphorylated by TGF-␤ signaling via TAK1 and p38. The two pathways, Smad and TAK1, synergistically enhance the activity of ATF-2 which acts as their common nuclear target.
Co-immunoprecipitation-A mixture of 2 g of the ATF-2 expression plasmid, pact-CRE-BP1, 2 g of the Flag-Smad3 expression plasmid, pact-Flag-Smad3, and 2 g of the Flag-Smad4 expression plasmid, pact-Flag-Smad4, was transfected into the TGF-␤-responsive 293 cells using LipofectAMINE (Life Technologies, Inc.). In some assays, the plasmid to express the constitutively active form of the TGF-␤ type I receptor, pact-ALK5-T204D, was also co-transfected. The total amount of plasmid DNA was adjusted to 8 g by adding the control effector plasmid, pact1, lacking the cDNA to be expressed. TGF-␤ treatment at a final concentration of 7.2 ng/ml was performed for 1 h before lysate preparation. Forty hours after transfection, cells were lysed in lysis buffer (50 mM Hepes, pH 7.5, 250 mM NaCl, 0.2 mM EDTA, 0.5% Nonidet P-40, 50 mM NaF, 2 mM Na 3 VO 4 , 0.1 M okadaic acid, 25 mM ␤-glycerophosphate, and protease inhibitor mixture), and whole-cell lysates were prepared. After decreasing the NaCl concentration to 100 mM by adding the lysis buffer lacking NaCl, lysates were immunoprecipitated using anti-ATF-2 polyclonal antibodies (N-96, Santa Cruz Biotechnology) or normal rabbit IgG as a control. The immune complex was analyzed by Western blotting to detect the co-precipitated Smad3 and Smad4 using anti-Flag monoclonal antibody (Eastman Kodak Co.) and LumiGO chemiluminescent detection reagent (New England Biolabs). To examine the Smad3/4 and ATF-2 proteins expressed, aliquots of cell lysates were also directly used for Western blotting with anti-Flag and anti-ATF-2 antibodies.
Mammalian Two-hybrid Assay-The plasmids used to express the Gal4-Smad fusion protein containing the Gal4 DNA-binding domain (amino acids 1-147) joined to the N-proximal region of Smad3 (amino acids 1-189) or Smad4 (amino acids 1-265) were made by the PCRbased method with the use of the cytomegalovirus promoter-containing expression vector. The plasmids encoding the VP16-ATF-2 fusion protein containing the C-proximal region of ATF-2 (amino acids 291-505) were constructed similarly using the pcDNA3 vector (Invitrogen). The plasmids used to express the Gal4-ATF-2 fusion protein containing the Gal4 DNA-binding domain joined to the C-proximal region of ATF-2 (amino acids 291-414) were made by the PCR-based method with the use of the cytomegalovirus promoter-containing expression vector. The plasmids encoding the VP16-Smad fusion protein containing the fulllength form of Smad3 or Smad4 were constructed similarly using the pcDNA3 vector. Co-transfection assays were performed as described (44) using the firefly luciferase reporter plasmid containing three copies of the Gal4-binding site. A mixture containing 1 g of the luciferase reporter plasmid, 3 g of either the Gal4-Smad3N, Gal4-Smad4N, or Gal4-ATF-2 expression plasmid, 4 g of either the VP16-ATF-2, VP16-Smad3FL, VP16-Smad4FL, or VP16 expression plasmid, and 0.5 g of the internal control plasmid pRL-TK (Promega) was transfected into HepG2 cells. The total amount of plasmid DNA was adjusted to 8.5 g by the addition of the control plasmid lacking the cDNA to be expressed. Luciferase assays were performed using the dual-luciferase assay system (Promega). Experiments were repeated three times, and the data were averaged.
Detection of Phosphorylated Proteins-To examine the phosphorylation of endogenous protein, 293 cells were serum-starved and incubated with TGF-␤ (3 ng/ml). The cells were disrupted in RIPA buffer (50 mM Tris-HCl, pH 7.4, 150 mM NaCl, 1% Triton X-100, 1% deoxycholate, 0.1% SDS, 50 mM NaF, 2 mM Na 3 VO 4 , 0.1 M okadaic acid, 25 mM ␤-glycerophosphate, and protease inhibitor mixture). After centrifugation, the supernatant was analyzed by 10% SDS-PAGE, followed by Western blotting. The phosphorylation of ATF-2, p38, JNK1, and JNK2 were examined using PhosphoPlus ATF-2 (Thr-71), p38MAPK (Thr-180/Tyr-182), and JNK1/2(Thr-183/Tyr-185) antibody kits, respectively (New England Biolab.). To inhibit p38 activity, 293 cells were treated with SB203580 (Calbiochem) for 45 min before the preparation of cell lysates. To analyze the phosphorylated state of ATF-2 expressed by the transfected DNA, 293 cells were transfected using LipofectAMINE (Life Technologies, Inc.) and a mixture made up of 1.5 g of the plasmid expressing the wild type or CT91 mutant lacking the C-terminal 91 amino acids and 1.5 g of a plasmid expressing the activated form of TAK1 (TAK1⌬N) or no protein at all. About 45 h after transfection, cell lysates were prepared, and the phosphorylated state of ATF-2 was examined again as described above.
Co-transfection Assay-The plasmids to express Smad3 and Smad4 were constructed by inserting the corresponding cDNAs downstream of the cytomegalovirus promoter. The CRE-containing luciferase reporter was constructed using the previously reported CRE-CAT reporter plasmid (39). The ATF-2 and c-Jun expression plasmids containing the chicken cytoplasmic ␤-actin promoter were described previously (39). In the experiments using the CRE-containing reporter, a mixture containing 2 g of the reporter plasmid, 2 g of the ATF-2 expression plasmid, or 1 g of the plasmid to express ATF-2 or c-Jun, 1.5 g of the plasmid to express Smad3 or Smad4, 1.5 g of the activated TAK1 (TAK1⌬N) expression plasmid, and 0.5 g of the internal control plasmid pRL-TK was transfected into HepG2 cells. The total amount of plasmid DNA was adjusted to 9 g by the addition of the control plasmid lacking the cDNA to be expressed. In the experiments using the p3TP-Lux reporter plasmid, a mixture containing 1.5 g of the reporter plasmid, 2 g of the ATF-2 expression plasmid, 1.5 g of the plasmid to express Smad3 or Smad4, 1.5 g of the activated TAK1 expression plasmid, and 0.5 g of the internal control plasmid pRL-TK was transfected.
To examine the effect of various dominant negative forms of ATF-2, Smad3/4, and TAK1, the following mutants were used. The Ala mutant (ATF-2Ala) in which the three SAPK phosphorylation sites (Thr-69, Thr-71, and Ser-90) were replaced by alanine was constructed using the PCR-based method. The N-truncated mutant of ATF-2 (ATF-2⌬107) lacking the N-terminal 107 amino acids was described previously (39). The C-truncated mutant of Smad3 (Smad3⌬C) or Smad4 (Smad4⌬C) lacking the C-terminal 40 or 38 amino acids were made by using the PCR-based method. The Smad3 mutant in which all the three serine residues of the SSXS motif are mutated to alanine (Smad3AAVA) was also constructed by using the PCR-based method. The dominant negative form of TAK1, in which Lys-63 of the ATP-binding site was replaced by tryptophan (TAK1K63W), was a gift from Dr. K. Matsumoto (19). To examine the effect of various dominant negative forms on the TGF-␤induced activity of 3TP-Lux promoter, a mixture containing 1.5 g of the 3TP-Lux reporter plasmid, 2 g of the plasmid to express various forms of dominant negative forms of ATF-2, Smad3/4, or TAK1, and 0.5 g of the internal control plasmid pRL-TK was transfected into HepG2 cells by using the CaPO 4 method. The total amount of plasmid DNA was adjusted to 9 g by the addition of the control plasmid lacking the cDNA to be expressed. To examine the effect of dominant negative form of ATF-2 on the Smad-and/or TAK1-induced activity of 3TP-Lux promoter, a mixture containing 1.5 g of the 3TP-Lux reporter plasmid, 1 g of the plasmid to express Smad3 or Smad4, 1.5 g of the activated TAK1 (TAK1⌬N) expression plasmid, 1 g of the plasmid to express the dominant negative forms of ATF-2, and 0.5 g of the internal control plasmid pRL-TK was transfected into HepG2 cells by using the CaPO 4 method. The total amount of plasmid DNA was adjusted to 10 g. TGF-␤ treatment was performed for 12 h at a final concentration of 2.4 ng/ml before lysate preparation. Luciferase assays were performed using the dual luciferase assay system (Promega). Experiments were repeated 2-4 times, and the data were averaged.

RESULTS
ATF-2 Binds to the MH1 Region of Smad3 and Smad4 -To investigate whether ATF-2 functions in the Smad pathway, we first of all examined for a direct interaction between Smad3/4 and ATF-2 ( Fig. 1). Protein affinity resins in which the GST, GST-Smad3, or GST-Smad4 fusion protein containing the fulllength form of Smad3 or Smad4 was used as a ligand were prepared (Fig. 1, A and B). The full-length form of human ATF-2 was synthesized using the in vitro transcription/translation system and was mixed with this affinity resin. Approximately 17 and 20% of ATF-2 were bound to the resin containing the GST-Smad3 and the GST-Smad4 fusion protein, respectively, but none was bound by the GST resin alone (Fig.  1C). We further examined which region of Smad3 and Smad4 binds to ATF-2. In addition to the GST fusion protein containing the full-length form of Smad3 and Smad4, five fusion proteins containing a series of truncated Smad3 or Smad4 protein were prepared and used in the binding assays (Fig. 1, A and B). Among Smad proteins, there are two homologous regions, the N-terminal MH1 (mad homology domain 1) and the C-terminal MH2, which are conserved in Smad-related proteins in various species ranging from insects to vertebrates (Fig. 1A). The truncated mutants of Smad3 and Smad4 that lacked the region downstream of the MH1 region still retained the ability to interact with ATF-2 ( Fig. 1, A and C). However, the mutants of Smad3 and Smad4 that lacked a part or the whole region of MH1 could not bind to ATF-2 (see CT1 and NT1 of Smad3 and CT1, NT1, and NT2 of Smad4). These results indicate that the N-terminal MH1 region binds to ATF-2.
Smad4 Binds to the b-ZIP Region of ATF-2-To determine further which region of ATF-2 interacts with Smad4, we made various mutants of ATF-2 by an in vitro transcription/translation system and used them in the GST pull-down assay (Fig. 2). Among the six mutants, the two mutants lacking the basic region (⌬BR) or containing a mutated leucine zipper (L34V), in which the third and fourth leucine residues were mutated to valine, failed to bind to GST-Smad4. In contrast, all the other ATF-2 mutants bound to GST-Smad4 with an efficiency similar After washing, the bound proteins were released and analyzed on 10% SDS-PAGE followed by autoradiography. In the input lanes, the amount of 35 S-ATF-2 protein was 10% that used for the binding assay.

FIG. 2. Smad3/4 bind to the b-ZIP region of ATF-2.
On the top, the functional domains of ATF-2 are schematically shown. The structures of the various forms of ATF-2 used are shown below. The results of binding assays shown below are indicated on the right. The relative binding activities of the mutants are designated ϩ and Ϫ, which indicate the binding of 14 -40% and less than 0.5% of the input protein, respectively. In the input lanes, various forms of ATF-2 indicated above each lane were synthesized in vitro and analyzed by 10% SDS-PAGE. In the right panel, the 35 S-ATF-2 proteins indicated above each lane were mixed with the GST-Smad4 affinity resin, which contains full-length Smad4, and the bound proteins were analyzed on 10% SDS-PAGE followed by autoradiography. In the input lanes, the amount of protein was 10% that used in the binding assay. Less than 0.5% of the input ATF-2 proteins bound to the control GST resin (data not shown). WT, wild type.
to that of the wild type (approximately 15-30% of the input ATF-2 protein was bound). These results indicated that the b-ZIP region of ATF-2 interacts with Smad4.
TGF-␤ Signaling-induced Association between ATF-2 and Smad3/4 -We investigated the interaction between ATF-2 and Smad3/4 in mammalian cells by co-immunoprecipitation (Fig. 3A). The ATF-2 expression plasmid was co-transfected into the TGF-␤-responsive 293 cells with the two plasmids expressing Flag-linked Smad3 and Smad4. The cell lysates were immunoprecipitated with the anti-ATF-2 polyclonal antibody, and the co-precipitated Smad3 and Smad4 proteins were detected using anti-Flag antibody. Both Smad3 and Smad4 were co-precipitated with anti-ATF-2 antibody (Fig. 3A, lane 2). The Smad3 proteins overexpressed from the transfected DNA were reported to be localized in the nuclei even in the absence of ligand (45), and ATF-2 is constitutively in the nuclei. These facts are consistent with the data described above. To examine whether TGF-␤ signaling enhances the association between ATF-2 and Smad3/4, the constitutively active TGF-␤ type I receptor, in which Thr-204 was replaced by aspartic acid, was co-transfected, and the cells were treated with TGF-␤. Under these conditions higher amounts of Smad3/4 were co-precipitated with the anti-ATF-2 antibody than in the absence of the exogenous TGF-␤ type I receptor and TGF-␤ stimulation (Fig.  3A, compare lanes 2 and 3). The control IgG did not co-precipitate Smad3/4 at all (Fig. 3A, lane 1). These results show that ATF-2 associates with Smad3/4 upon TGF-␤ stimulation in vivo.
To confirm the interaction in mammalian cells between the b-ZIP region of ATF-2 and the MH1 region of Smad3/4, the two-hybrid assay was performed using HepG2 cells (Fig. 3B). In the first experiment, two types of chimeric proteins were employed. In one, the N-proximal region of Smad3 or Smad4 was fused to the DNA binding domain of Gal4, and in the other, the C-proximal region of ATF-2 containing the b-ZIP structure was fused to the transcriptional activation domain of VP16, and the degree of transcriptional activation in HepG2 cells was examined (Fig. 3B, left panel). The VP16-ATF-2 fusion proteins stimulated Gal4-Smad3N and Gal4-Smad4N activity by 8.9and 11-fold, respectively, whereas VP16 alone had no effect. In the second experiment, the C-proximal region of ATF-2 containing the b-ZIP structure was fused to the DNA-binding domain of Gal4, and the full-length form of Smad3 or 4 was fused to the strong transcriptional activation domain of VP16 (Fig. 3B, right panel). The VP16-Smad3 and VP16-Smad4 fusion proteins stimulated Gal4-ATF-2 activity by 17-and 7-fold, respectively, whereas VP16 alone had no effect. These results indicate that the b-ZIP domain of ATF-2 interacts in mammalian cells with the MH1 region of Smad3 and Smad4.
TGF-␤ Signaling Induces Phosphorylation of ATF-2 via TAK1-We next examined whether phosphorylation of ATF-2 is enhanced by TGF-␤ treatment (Fig. 4). The TGF-␤-responsive 293 cells were treated with TGF-␤, and ATF-2 phosphorylated at Thr-71 was detected by the phospho-ATF-2-specific antibody at various intervals after TGF-␤ treatment (Fig. 4A). The Thr-71 residue is known to be the phosphorylation site of SAPK (34 -36). The degree of phosphorylation of ATF-2 increased up to a maximum of 4-fold at 15 min after TGF-␤ treatment, whereas the amount of ATF-2 was not affected by TGF-␤ treatment. To confirm that TGF-␤ signaling phosphorylates ATF-2 at the same sites as SAPK, the ATF-2 mutant, whose three SAPK phosphorylation sites (Thr-69, Thr-71, and Ser-90) were replaced by alanine, was used. Since this alanine mutant cannot be recognized by the antibody raised against the peptide containing these phosphorylation sites, we used the C-truncated form of ATF-2 to discriminate from the endogenous protein, and we judged the phosphorylation status of the mutants by their altered migration during SDS-PAGE. The C-truncated form of ATF-2, which was phosphorylated by TGF-␤ signaling via SAPK, migrated more slowly during SDS-PAGE than the non-phosphorylated form (Fig. 4B, compare  lanes 1 and 2). However, the migration of the alanine mutant was the same as that of the wild type even in the presence of TGF-␤ treatment, confirming that at least one of these three sites was phosphorylated by TGF-␤ signaling. To investigate further whether phosphorylation of ATF-2 is mediated by TAK1, we examined the effect of activated TAK1 on the phosphorylation of ATF-2 (Fig. 4C). Co-transfection of the plasmid to express the activated form of TAK1, which lacked its Nterminal 22 amino acids, with the C-truncated ATF-2 expression plasmid increased the amount of ATF-2 phosphorylated at Ser-71, as detected by the phospho-ATF2-specific antibody (Fig. 4C, compare lanes 1 and 3). In addition, the activated form of TAK1 further enhanced the phosphorylation of ATF-2 in the presence of TGF-␤ treatment (Fig. 4C, compare lanes 2 and 4).

FIG. 3. Interaction between ATF-2 and Smad3/4 in mammalian cells.
A, co-immunoprecipitation (Co-IP). Whole-cell lysates were prepared from 293 cells transfected with a mixture of plasmids to express ATF-2 and Flag-linked Smad3 and Smad4, and samples from the lysates were directly used for Western blotting with the anti-Flag or anti-ATF-2 antibodies (Direct Western). Whole-cell lysates were also immunoprecipitated by anti-ATF-2 antibody (Ab), and the immunocomplexes were analyzed by Western blotting using anti-Flag antibodies. In lanes 1 and 3, the plasmid to express the constitutively active TGF-␤ type I receptor was also co-transfected, and the transfected cells were stimulated by TGF-␤ for 1 h before preparation of cell lysates. In lane 1, normal IgG was used as a control for immunoprecipitation. B, mammalian two-hybrid interaction. Left, HepG2 cells were co-transfected with the Gal4 site-containing reporter, the plasmid to express Gal4-Smad3 or the Gal4-Smad4 fusion containing the MH1 region of Smad3/4, and the expression plasmid for VP16-ATF-2 containing the b-ZIP region of ATF-2. The degree of activation is indicated (means S.E.). Right, HepG2 cells were co-transfected with the Gal4 site-containing reporter, the expression plasmid for the Gal4-fusion protein containing the b-ZIP region of ATF-2, and the VP16-Smad3 or VP16-Smad4 expression plasmid containing the full length Smad3/4 or VP16 alone.
These results suggest that TGF-␤ signaling induces the phosphorylation of ATF-2 at the SAPK phosphorylation sites via TAK1.

Involvement of p38 in TGF-␤-induced Phosphorylation of ATF-2-
The results of the ATF-2 phosphorylation assays and the fact that TAK1 activates SAPKs, JNKs, and p38 (21,22) suggest that SAPKs phosphorylate ATF-2 upon TGF-␤ stimulation. To investigate which SAPK is activated by TGF-␤ signaling, we examined the phosphorylation of p38, JNK1, and JNK2. The 293 cells were treated with TGF-␤, and p38 phosphorylated at Thr-180/Tyr-182 and JNK1/JNK2 phosphorylated at Thr-183/Tyr-185 were detected by the phosphorylated form-specific antibody at various intervals after TGF-␤ treatment (Fig. 5A). The degree of phosphorylation of p38, which displayed a timing similar to that of ATF-2, increased up to 4-fold. In contrast, the phosphorylation of JNK1 and JNK2 remained unchanged, suggesting that TGF-␤ signaling leads to phosphorylation of ATF-2 through mainly p38 rather than JNK. To confirm these results further, the effect of the specific inhibitor of p38, the pyridinyl imidazole derivative SB203580 which cannot inhibit JNKs (46,47), on the TGF-␤-induced phosphorylation of ATF-2 was examined (Fig. 5B). SB203580 almost completely blocked TGF-␤-induced phosphorylation of ATF-2. These results indicate that TGF-␤ induces the phosphorylation of ATF-2 through the action of TAK1 and p38.
Synergistic Activation of ATF-2 Activity by Smad and TAK1 Pathways-To investigate whether the trans-activating capacity of ATF-2 is enhanced by Smad3/4 and TAK1 pathways, co-transfection assays were performed using a reporter plasmid containing four copies of the consensus CRE sequence (Fig.  6). This artificial promoter was weakly responsive to TGF-␤ in HepG2 cells (2-fold). When present separately, ATF-2, Smad3/4, and the activated form of TAK1 stimulated this promoter activity by 2-, 10-, and 2-fold, respectively, in the absence of TGF-␤ treatment, and by 5-, 17-, and 7-fold, respectively, in the presence of TGF-␤ treatment. The degree of activation of this promoter by ATF-2 was synergistically increased by coexpression of Smad3/4 or the activated form of TAK1. Furthermore, promoter activity could be strongly enhanced by co-expression of all the three effectors together, resulting in a 145and 203-fold stimulation in the absence and presence of TGF-␤ treatment, respectively. These results support the idea that both the Smad3/4 pathway and TAK1 pathway synergistically activate ATF-2. CRE is recognized by the ATF-2/c-Jun heterodimer with high affinity and the c-Jun homodimer with lower affinity (40). To determine which of these actually contributes to CRE-dependent activation, we transfected the cells with plasmids expressing both ATF-2 and c-Jun or with a plasmid expressing c-Jun alone. As reported previously (39),

FIG. 4. Induction of ATF-2 phosphorylation by TGF-␤ signaling via TAK1. A, time course of phosphorylation of endogenous ATF-2.
Total cell lysates were prepared from TGF-␤-treated or untreated 293 cells and used for Western blotting. In the upper and lower panels, the ATF-2 proteins phosphorylated at Thr-71, and both the phosphorylated and non-phosphorylated forms are indicated, respectively. B, phosphorylation of exogenous ATF-2 at JNK/p38 phosphorylation sites by TGF-␤ signal. The plasmid to express the ATF-2 protein lacking the C-terminal 91 amino acids but containing either the normal three JNK/p38 phosphorylation sites or these sites mutated to alanines was transfected into 293 cells. Cell lysates were prepared, and the C-truncated ATF-2 was detected by Western blotting using anti-ATF-2 antibody which recognizes both the phosphorylated and non-phosphorylated forms. Since the phospho-ATF-2-specific antibody cannot react with the alanine mutant, the phosphorylated form was detected as a slower migrating band on a long SDS-PAGE gel. C, phosphorylation of ATF-2 through TAK1. The two plasmids to express ATF-2 lacking the C-terminal 91 amino acids and the activated form of TAK1 (TAK1⌬N) or no protein were transfected into 293 cells. Phospho-ATF-2 and ATF-2 were detected as described in A.

FIG. 5. Phosphorylation of ATF-2 by TGF-␤ signaling via p38.
A, activation of p38 by TGF-␤ signaling. Total cell lysates were prepared from TGF-␤-treated or untreated 293 cells and used for Western blotting with the phosphorylated form-specific or nonspecific antibody against ATF-2, p38, JNK1, or JNK2. B, inhibition of the TGF-␤-induced phosphorylation of ATF-2 by the p38 inhibitor. In the presence or absence of TGF-␤ treatment, 293 cells were treated with SB203580. ϩ, 0.1 M; ϩϩ, 10 M. Whole cell lysates were prepared and used for Western blotting to detect phosphorylated ATF-2 and both phosphorylated and non-phosphorylated ATF-2 as described in A.
the ATF-2/c-Jun heterodimer activated more strongly the CREcontaining promoter (8-fold) compared with the ATF-2 homodimer. However, further stimulation of ATF-2/c-Jun heterodimerdependent activation by co-expression of both Smad3/4 and the activated form of TAK1 was inefficient compared with that seen with ATF-2 alone. The trans-activating capacity of the c-Jun homodimer was also not so strongly enhanced by Smad3/4 and TAK1 compared with the marked increase in the capacity of the ATF-2 homodimer. These results suggest that the ATF-2 homodimer is the preferred target for the Smad and TAK1 pathways at least in HepG2 cells, although the activity of the ATF-2/c-Jun heterodimer and the c-Jun homodimer are also stimulated to some extent by both pathways.
Involvement of ATF-2 in TGF-␤-inducible Promoter Activation-To examine the role of ATF-2 in the regulation of the TGF-␤-inducible promoters, co-transfection experiments were performed using a fusion promoter (p3TP-Lux reporter) consisting of PAI-1 and collagenase promoters (48) (Fig. 7A). This promoter was highly responsive to TGF-␤ in HepG2 cells (38fold). Smad3/4 stimulated this promoter activity by 251-fold in the absence of TGF-␤ treatment and by 409-fold in the presence of TGF-␤ treatment. The degree of activation of this promoter by Smad3/4 was slightly enhanced by co-expression of ATF-2 or the activated form of TAK1. Furthermore, promoter activity could be strongly enhanced by co-expression of all the three effectors together, resulting in a 791-and 1104-fold stimulation in the absence and presence of TGF-␤ treatment, respectively. These results support the idea that co-expression of ATF-2, Smad3/Smad4, and the activated form of TAK1 synergistically activated this promoter activity. When these results are com-pared with those with the CRE-containing promoter described above, however, some difference is evident. Unlike the case of CRE-containing promoter, Smad3/4 strongly activated this promoter. In addition, ATF-2 alone did not enhance this promoter activity, and the synergism between ATF-2 and the activated form of TAK1 was not observed using this promoter. This could be due to the fact that Smad3/4 can activate this promoter not only via a complex formation with ATF-2 but also via direct binding to the specific sites in the PAI-I promoter (see "Discussion").
To confirm that ATF-2 plays an important role for TGF-␤induced activation of the p3TP-Lux promoter, we used two ATF-2 mutants as follows: the Ala mutant (ATF-2Ala) in which FIG. 6. Synergistic enhancement of ATF-2 activity by Smad3/4, and TAK1. Transcriptional activation of the CRE-containing reporter by ATF-2. HepG2 cells were transfected by a mixture containing the CRE-containing luciferase reporter, the ATF-2 expression plasmid, the Smad3 and Smad4 expression plasmids, the plasmid to express the activated form of TAK1, and the internal control plasmid. The degree of activation is indicated (means S.E.). The data obtained after TGF-␤ treatment is indicated by shaded bars. In some cases, the c-Jun expression plasmid or a mixture of the ATF-2 and c-Jun expression plasmids was used instead of the ATF-2 expression plasmid.

FIG. 7. Involvement of ATF-2 in TGF-␤-induced transcriptional activation.
A, transcriptional activation of the p3TP-Lux reporter by ATF-2. HepG2 cells were transfected by a mixture containing the p3TP-Lux reporter plasmid, the ATF-2 expression plasmid, the Smad3 and Smad4 expression plasmids, the plasmid to express the activated form of TAK1, and the internal control plasmid. The degree of activation is indicated (means S.E.). The data obtained after TGF-␤ treatment is indicated by shaded bars. B, effect of dominant negative forms of ATF-2, TAK1, and Smad3/4 on the TGF-␤-induced activity of 3TP-Lux promoter. HepG2 cells were transfected by a mixture containing the p3TP-Lux reporter, the plasmid to express various dominant negative forms of ATF-2, TAK1, or Smad3/4, and the internal control plasmid. The degree of activation is indicated (means S.E.). The data obtained after TGF-␤ treatment is indicated by shaded bars. C, effect of dominant negative forms of ATF-2 on the Smad3/4 and/or TAK1-induced activity of 3TP-Lux promoter. HepG2 cells were transfected by a mixture containing the p3TP-Lux reporter, the Smad3/4 expression plasmid and/or the activated TAK1 expression plasmid, the plasmid to express either of two dominant negative forms of ATF-2, and the internal control plasmid. the three SAPK phosphorylation sites (Thr-69, Thr-71, and Ser-90) (34 -36) were replaced by alanine and the N-truncated mutant (ATF-2⌬107) lacking the N-terminal 107 amino acids including the SAPK phosphorylation sites (Fig. 7B). These two mutants cannot be phosphorylated by TGF-␤ signaling via p38 and were expected to act as a dominant negative form. Cotransfection of either of these two mutants strongly inhibited the TGF-␤-induced activity of the p3TP-Lux promoter, indicating that ATF-2 is involved in the activation of 3TP-Lux promoter by TGF-␤ signaling. In addition, the dominant negative form of TAK1, in which Lys-63 of the ATP-binding site was replaced by tryptophan (TAK1K63W), inhibited the TGF-␤induced activity of the p3TP-Lux promoter. To confirm the role of Smad3/4 in the TGF-␤-induced activity of the p3TP-Lux promoter, we used two types of mutants. The C-truncated mutant of Smad3 (Smad3⌬C) or Smad4 (Smad4⌬C) lacking the C-terminal transcriptional activation domain was reported to act as a dominant negative form (6). The TGF-␤ type I receptor phosphorylates Smad2 at Ser-465 and Ser-467 in the SSXS motif, and the mutant in which all the three serine residues in the SSXS motif were replaced by alanines acts as a dominant negative form, because this mutant stably binds to the TGF-␤ type I receptor (7,49,50). In addition, the alanine mutant of Ser-464 of Smad2 also act as a dominant negative form, although this site is not directly phosphorylated. Therefore, we constructed second type of putative dominant negative form of Smad3 by replacing all the three serine residues of the corresponding SSXS motif to alanine (Smad3AAVA). Co-transfection of the C-truncated mutant of Smad3 (Smad3⌬C) or Smad4 (Smad4⌬C) or the Smad3 alanine mutant (Smad3AAVA) inhibited the TGF-␤-dependent p3TP-Lux promoter activity. These results indicate that that both Smad and TAK1 pathway are required for the TGF-␤-induced activation of the 3TP-Lux promoter.
To confirm the role of ATF-2 further, we examined the effect of two ATF-2 mutants (ATF-2Ala and ATF-2⌬107) on the Smad3/4-and/or activated TAK1-induced promoter activity of 3TP-Lux (Fig. 7C). Either of these two mutant significantly inhibited the 3TP-Lux promoter activity enhanced by Smad3/4, activated TAK1, or both Smad3/4 and activated TAK1. Thus, a dominant negative form of ATF-2 can inhibit the stimulatory effect of either Smad and TAK1 pathways on the of 3TP-Lux promoter. DISCUSSION Our results indicate that ATF-2 is a common nuclear target of the Smad and TAK1 pathways (Fig. 8). Upon binding of TGF-␤ to the type II receptor, the TGF-␤-bound type II receptor makes a heteromeric complex with the type I receptor, resulting in the activation of the latter's serine/threonine kinase activity. The activated serine/threonine kinase of the type I receptor then phosphorylates the bound Smad3 or Smad2 protein, which results in its release from the type I receptor. The released Smad3 forms a hetero-oligomer with Smad4, which is thought to be localized in the cytosol in the absence of TGF-␤ stimulation, and the hetero-oligomer moves into the nucleus. This hetero-oligomer directly binds to ATF-2 through the MH1 region of Smad3/4 and the b-ZIP region of ATF-2, although the exact number of ATF-2 and Smad3/4 molecules in this complex remains unknown. Binding of the Smad3/4 complex to ATF-2 enhances ATF-2 activity, as suggested by the observation that overexpression of Smad3/4 enhances the trans-activating capacity of ATF-2 (Fig. 6). In this sense, Smad3/4 resembles adenovirus E1A, which stimulates CRE-dependent transcription via binding to the b-ZIP region of ATF-2 (51). In addition to this Smad pathway, another pathway, the TAK1 pathway, is required for TGF-␤ signal transduction. The expression of the dominant negative form of TAK1 inhibits the TGF-␤-induced activation of the PAI-1 promoter (19). Upon TGF-␤ stimulation, the TAB1 protein is thought to be activated, an event that results in its binding to the serine/threonine kinase domain of TAK1 (20). However, the precise mechanism of signal transduction from the TGF-␤ receptor to TAB1 remains unknown. TAK1 is a member of the MAPKKK family and activates MKK3 and MKK6 of the MAPKK family, both of which share striking homology with each other (21). TAK1 also activates MKK4, another member of MAPKK (22). TAK1 activates p38, one member of the SAPK family via MKK6/MKK3 (21), and also possibly through MKK4 (22)(23)(24). p38 directly phosphorylates ATF-2 at Thr-69, Thr-71, and Ser-90, resulting in stimulation of its trans-activating capacity. In fact, co-expression of activated TAK1 enhanced this trans-activating capacity of ATF-2 (Fig. 6). Thus, the Smad and TAK1 pathways synergistically stimulate TGF-␤-induced transcription by acting on the common nuclear target ATF-2. TGF-␤ has an important role in the regulation of genes involved in cell cycle control and genes encoding the extracellular matrix, and many of them have a CRE in their transcriptional control regions. Recently we found that the expression level of some TGF-␤-inducible genes encoding extracellular matrix was decreased in mouse embryonic fibroblasts lacking ATF-2 and its related gene CRE-BPa 2 supporting that the ATF-2 family is important for the TGF-␤mediated transcriptional activation. The identification of ATF-2 as a common nuclear target of Smad and TAK1 pathways may provide a clue as to how the signal transduction of TGF-␤ is regulated. Recent studies using the Xenopus system indicated that TAK1 and TAB1 also function in the BMP signal transduction pathway in Xenopus embryos (52). Therefore, ATF-2 could also play an important role in BMP signal trans-2 Y. Sano and S. Ishii, unpublished data. duction. In fact, ATF-2 binds to Smad1 in the GST pull-down assay which is a mediator of BMP signal transduction. 2 The b-ZIP region of ATF-2 directly interacts with the Nterminal MH1 region of Smad3/4. The b-ZIP region of ATF-2 and the MH1 region of Smad3/4 can also directly bind to CRE and the recently identified Smad-recognition sequence, respectively (17,30). Therefore, the interaction between ATF-2 and Smad3/4 is mediated by a different type of DNA binding domain. There exist numerous examples of DNA binding domains that can additionally serve as interaction sites for specific proteins. The adenovirus 13SE1A protein binds to multiple DNA binding domains including not only the b-ZIP region of ATF-2 but also the metal finger of Sp1 and the basic-helix-loophelix region of upstream factor (51). Similarly, the helical structure of the b-ZIP region of ATF-2, which would be exposed to the solvent, may serve as a protein surface for interaction with Smads. Several examples of protein-protein interaction being mediated via different DNA binding domains have also been reported (44,53).
Two SAPK family members have been identified in mammalian cells, JNKs (54,55) and p38 homologues (also termed p40, RK, and CSBP) (56 -59). Among these two type of SAPKs, only p38 was activated up to a maximum at 15 min after TGF-␤ treatment via TAK1 in 293 cells (Fig. 5). In contrast to this, TAK1 stimulated by ceramide was reported to activate JNK in COS7 cells at the similar timing (22). Although JNK was also reported to be activated by TGF-␤ stimulation in 293T cells, this activation of JNK activity was observed at 12 h after TGF-␤ treatment (60). This contradiction could be due to the difference of cells used, possibly due to the the cell type-specific expression of some co-factor(s). ATF-2 is a good substrate of both SAPKs, whereas c-Jun is phosphorylated by JNKs but not by p38 (61), suggesting that ATF-2 is a preferable nuclear target of the TAK1 signal transduction pathway at least in 293 cells. This may be consistent with our observation that the trans-activating capacity of the c-Jun homodimer is not so strongly enhanced by Smad3/4 and TAK1 compared with the marked increase in the capacity of the ATF-2 homodimer (Fig.  6). However, we observed that c-Jun directly binds to Smad3 via its b-ZIP region like in the case of ATF-2. 2 Therefore, c-Jun may play a role in the Smad pathway in some types of cells. This could be consistent with the recent report that Jun/Fos interacts with Smad3 and is involved in TGF-␤ signaling (62).
Synergistic activation of the CRE-containing promoter by ATF-2 and Smad3/4 suggests the interaction between ATF-2 bound to CRE and Smad3/4 unbound to DNA in this case. Recently, however, direct binding of Smad3 and Smad4 to a specific DNA sequence was reported (17). In addition, the putative Smad-binding sites in the PAI-1 promoter were demonstrated to be critical for the TGF-␤ inducibility of the promoter activity (18,63). Consistent with these reports, we observed that co-expression of Smad3/4 alone enhanced the 3TP-Lux promoter, which contains the PAI-1 promoter segment, more strongly than the case of CRE-containing promoter. On the other hand, expression of the dominant negative form of TAK1 (19) or ATF-2 (Fig. 7, B and C) also lowered the TGF-␤-induced activity of the 3TP-Lux promoter, indicating that the ATF-2-TAK1 pathway is also important for TGF-␤ responsiveness of this promoter. These results may suggest that both ATF-2 and Smad3/4 directly bind to their own target sequences in the PAI-1 promoter and that ATF-2 and Smad3/4 may interact via protein-protein interaction. This may give a high TGF-␤ responsiveness to this promoter. Synergism of promoter activation by multiple DNA binding transcription factors is well known (64). Promoters containing four copies of CRE (Fig. 6) or two copies of Smad3/4-binding sites (17) exhibited only weak TGF-␤ responsiveness, supporting the idea that only one type of DNA binding factor may not sufficient to exhibit the TGF-␤ responsiveness. Transcription factors other than ATF-2 may also function in TGF-␤-induced activation of many other target promoters of TGF-␤ signaling. In fact, recently, Sp1 was demonstrated to activate p21/WAF1/Cip1 promoter activity by interacting with Smad3/4 (65).