Antagonistic regulation of a proline-rich transcription factor by transforming growth factor beta and tumor necrosis factor alpha.

Transforming growth factor β (TGF-β) and tumor necrosis factor α (TNF-α) often exhibit antagonistic actions on the regulation of various activities such as immune responses, cell growth, and gene expression. However, the molecular mechanisms involved in the mutually opposing effects of TGF-β and TNF-α are unknown. Here, we report that binding sites for the transcription factor CTF/NF-I mediate antagonistic TGF-β and TNF-α transcriptional regulation in NIH3T3 fibroblasts. TGF-β induces the proline-rich transactivation domain of specific CTF/NF-I family members, such as CTF-1, whereas TNF-α represses both the uninduced as well as the TGF-β-induced CTF-1 transcriptional activity. CTF-1 is thus the first transcription factor reported to be repressed by TNF-α. The previously identified TGF-β-responsive domain in the proline-rich transcriptional activation sequence of CTF-1 mediates both transcriptional induction and repression by the two growth factors. Analysis of potential signal transduction intermediates does not support a role for known mediators of TNF-α action, such as arachidonic acid, in CTF-1 regulation. However, overexpression of oncogenic forms of the small GTPase Ras or of the Raf-1 kinase represses CTF-1 transcriptional activity, as does TNF-α. Furthermore, TNF-α is unable to repress CTF-1 activity in NIH3T3 cells overexpressing ras or raf, suggesting that TNF-α regulates CTF-1 by a Ras-Raf kinase-dependent pathway. Mutagenesis studies demonstrated that the CTF-1 TGF-β-responsive domain is not the primary target of regulatory phosphorylations. Interestingly, however, the domain mediating TGF-β and TNF-α antagonistic regulation overlapped precisely the previously identified histone H3 interaction domain of CTF-1. These results identify CTF-1 as a molecular target of mutually antagonistic TGF-β and TNF-α regulation, and they further suggest a molecular mechanism for the opposing effects of these growth factors on gene expression.

Growth factors transmit an extended array of regulatory information to eukaryotic cells by binding to specific receptors at the cell surface. With the identification of a great variety of growth factors and receptors in the recent years, it has become apparent that some factors often exert pleiotropic or redundant effects on target cells. For example, identical immune responses can be induced by several different cytokines acting either alone or synergistically in combinations (1). In contrast, other growth factors, such as transforming growth factor ␤ (TGF-␤) 1 and tumor necrosis factor ␣ (TNF-␣) are often functionally antagonistic. For instance, TGF-␤ and TNF-␣ antagonize on the regulation of nitric oxide production in cardiac myocytes (2), the development of cytotoxic T cells (3), and in certain autoimmune diseases such as collagen-induced arthritis (4) and experimental allergic encephalomyelitis (5).
Recent studies have indicated that both factors bind distinct cellular receptors at the cell surface of numerous target cell types. TGF-␤ receptor activation involves serine/threonine autophosphorylation of signaling receptor complexes (reviewed in Ref. 6). In contrast, TNF-␣ receptors have no apparent enzymatic activity, yet they may associate with a set of recently identified putative signal-transducing molecules such as TNF receptor-associated factors (TRAFs) (reviewed in Ref. 7). Thus, the mutual antagonism of TGF-␤ and TNF-␣ is not exerted through competition for the same cell surface receptors but involves differential regulation of specific target intracellular mediators. However, no such mediators have been identified to date, and the molecular mechanisms underlying this mutual antagonism of TGF-␤ and TNF-␣ are still unknown.
Here, we have analyzed the molecular mechanisms involved in the regulation of the procollagen gene expression by TGF-␤ and TNF-␣. Collagen genes are well-known targets of mutually antagonistic TGF-␤ and TNF-␣ regulation (8 -10). For example, TGF-␤ has been shown to activate ␣2(I) procollagen gene expression in mesenchymal cells by acting at the level of transcriptional initiation (8). This activation correlates with increased collagen production and secretion by fibroblasts during wound healing, a process that is potently promoted by TGF-␤ (11,12). The proinflammatory cytokine TNF-␣ is also present in the healing wounds (13), and it has been shown to downregulate expression of the collagen ␣2(I) gene promoter, thereby antagonizing TGF-␤ action on collagen gene expression (9,15). The mechanism of antagonistic ␣2(I) collagen promoter regulation by TGF-␤ and TNF-␣ is unclear at present. However, binding sites for the ubiquitous transcription factors CTF/ NF-I and Sp1 have been proposed to mediate hormone action (8,14,15). This suggests that TGF-␤ and TNF-␣ might antagonistically regulate the activity and/or the expression of these transcription factors. Alternatively, hormone regulation might * 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  be effected on other proteins that interact with the transactivators, such as components of the basal transcriptional initiation machinery, including transcription factor IIB, the TATAbinding protein or transactivator-specific TATA-binding protein-associated factors.
Consistent with the proposed key role for CTF/NF-I in the TGF-␤ induction of the ␣2(I) collagen promoter (8), we have recently shown that CTF-1, a member of the CTF/NF-I family of proline-rich transcription factors (16), mediates TGF-␤-regulated transcriptional activation in NIH3T3 cells (17). TGF-␤ does not induce CTF-1 through direct phosphorylation of its proline-rich domain, suggesting that TGF-␤ induction is mediated by CTF-1-interacting proteins. One candidate target protein is histone H3, which interacts specifically with the TGF-␤-regulated transactivation domain of CTF-1, and which interaction leads to the alteration of the nucleoprotein architecture of chromatin-reconstituted target promoters (17).
Here, we addressed the possibility that specific CTF/NF-I family members might mediate antagonistic TGF-␤ and TNF-␣ responses. Indeed, we demonstrate that CTF-1 mediates TNF-␣-regulated transcriptional repression, and that TNF-␣ antagonizes the TGF-␤-dependent induction of CTF-1 transcriptional activity in mouse fibroblasts. These effects are mediated by the extreme C-terminal portion of the CTF-1 proline-rich transactivation domain, previously termed the TGF-␤ responsive domain (TRD). Although TNF-␣ repression is mimicked by expression of genes regulating protein phosphorylation, such as the proto-oncogenes ras and raf, our results indicate that the TRD is not directly phosphorylated in response to TNF-␣. Rather, our findings imply the interaction of the CTF-1 transactivation domain (TAD) with histone H3 in this regulatory process.

MATERIALS AND METHODS
Reagents-Human TGF-␤1 was purchased from Nacalai-Tesque (Kyoto, Japan). Mouse TNF-␣ and human PDGF-BB were obtained from Boehringer Mannheim. Human TNF-␣ was from Calbiochem. Dulbecco's modified Eagle's medium and fetal bovine serum were supplied by Life Technologies, Inc.. Arachidonic acid, indomethacin, and NDGA were from Sigma.
Plasmids-The mammalian expression vectors for the GAL4-CTF1 fusion proteins contain the DNA binding domain of the yeast transcription factor GAL4 (amino acids 1-147; Ref. 18 (17). GAL 399 -499 S495A contains two point mutations at positions 1550 and 1552 of the wild-type human CTF-1 cDNA sequence changing serine 495 to alanine (TCC 3 GCA). GAL 399 -499, GAL 1-399, GAL 399 -438, and GAL 438 -499 vectors have been described (19). GAL-AP2 and GAL-Sp1 contain amino acids 31-76 and 132-243 of the AP2 and Sp1 TADs, downstream of the GAL-DBD, respectively (17). The mammalian expression vectors pMLV-VP and pMLV-VP-H3 encode the VP16 transcriptional activation domain either unfused or fused to amino acids 10 -136 of the coding sequence of human histone H3.3 (17). pSG5-PPAR␥ (a kind gift of Prof. W. Wahli) is an SV40driven mammalian expression vector encoding the ␥ isoform of the Xenopus laevis peroxisome proliferator-activated receptor, a member of the superfamily of nuclear hormone receptors (20). The p␣CAT-⌬55 and p␣CAT-⌬87-3xAd (16), G5BCAT (21), and 5xTRE-TATA-CAT reporters (22) contain either zero and three high affinity CTF/NF-I binding sites or five GAL4 and five AP-1 binding sites in front of a TATA box and the cat gene, respectively, whereas the ACO-A.TK.CAT vector contains the upstream region of the PPAR-responsive rat acetyl-CoA oxidase gene (from positions Ϫ1273 to ϩ20; Ref. 20) placed in front of a thymidine kinase basal promoter and the cat gene. pRCBX2 (23) and pRSV-Raf BXB (24), expressing oncogenically activated forms of the proto-oncogenes ras and raf downstream of the conalbumin and Rous sarcoma virus promoters, were generous gifts of Drs. B. Wasylyk and U. Rapp, respectively. The Ras empty plasmid containing the conalbumin pro-moter was prepared by excision of the large HindIII fragment encoding the ras sequences and vector religation.
Cell Lines, Transfections, and Gel Mobility Shift Assays-Two clones of NIH3T3 cells were used in this study: clone 63 (which contains constitutive endogenous AP-1 activity; Figs. 1-3, 6, and 7); and clone 4 (in which AP-1 is inducible by Ras and Raf expression, used in Figs. 4 and 5). Both cell lines were grown at 37°C in a humidified incubator containing 5% CO 2 in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum and antibiotics. Cells were transiently transfected by electroporation, essentially as described (17). Briefly, 4.5 ϫ 10 6 cells were mixed with 70 g of total plasmid DNA and pulsed once at 960 microfarads and 250 V at room temperature, according to the instructions of the electroporator manufacturer (Bio-Rad). The contents of one pulsed cuvette were split in four, and cells were plated in DMEM plus 0.5% fetal bovine serum for 3-5 h. Cultures were then induced for a period of 18 -40 h with either ethanol vehicle, 5 ng/ml human TGF-␤, or 50 ng/ml human TNF-␣, as appropriate. (The results presented in the figures have been obtained using human TNF-␣). Cells were lysed in 1 ϫ Reporter Lysis Buffer (Promega Corp.), and CAT activities were determined using standard procedures (25) and normalized according to the activity of ␤-galactosidase resulting from a cotransfected internal control plasmid (CMV␤gal; Clontech). The activity of the SV40 and CMV promoters was not significantly affected by any hormone or pharmacological agent used in this study.
For gel-shift analysis of the endogenous CTF/NF-I or of the transiently expressed GAL4 fusion proteins, cells were lysed in extraction buffer (20 mM Tris, pH 7.5, 20% glycerol, 500 mM KCl, 1 mM dithiothreitol, and protease inhibitors) exactly as described in Ref. 19. Whole-cell lysates were normalized for protein concentration and incubated with end-labeled double-stranded DNA probes containing either the highaffinity CTF/NF-I binding site found within the first 50 base pairs of the adenovirus origin of replication (26) or the 17-base pair GAL4 binding site (27). Protein⅐DNA complexes were separated from free probe on native polyacrylamide gels and revealed by autoradiography.

CTF/NF-I Is Activated by TGF-␤ and Repressed by TNF-␣ in
NIH3T3 Cells-Previous studies had indicated that TGF-␤ activates the collagen ␣2(I) promoter in mouse NIH3T3 cells and identified binding sites for the transcription factor CTF/NF-I as necessary and sufficient mediators of TGF-␤ induction (8). Conversely, in other experiments, TNF-␣ was shown to suppress the basal activity of this promoter and to antagonize TGF-␤ induction (9,15). To address the possibility that CTF/NF-I might also mediate TNF-␣-regulated transcriptional repression, a reporter promoter containing three high affinity CTF/ NF-I binding sites in front of the ␣-globin TATA box and the cat gene was transiently transfected in NIH3T3 fibroblasts. The CTF/NF-I binding sites conferred efficient TNF-␣ transcriptional repression, because TNF-␣ down-regulated both the basal as well as the TGF-␤-stimulated CTF/NF-I transcriptional activity (Fig. 1A, p␣CAT-⌬87-3xAd). In contrast, TNF-␣ had no effect on the activity of a control promoter lacking CTF/NF-I binding sites, whereas it potently induced nuclear factor-B transcriptional activity under these conditions (Fig. 1A, p␣CAT-⌬55 and data not shown). Gel mobility shift analysis of the endogenous CTF/NF-I proteins indicated that TNF-␣ has no detectable effect on the binding activity and/or expression levels of CTF/NF-I species (Fig. 1B). Thus, these data together suggest that the endogenous mouse CTF/NF-I polypeptides specifically mediate antagonistic TGF-␤ and TNF-␣ transcriptional regulation.
The TGF-␤-responsive Domain of CTF-1 Mediates Transcriptional Repression in Response to TNF-␣-To address the possibility that TNF-␣ may regulate the transcriptional activity of specific CTF/NF-I variants such as CTF-1, we used a panel of expression vectors containing various portions of the CTF-1 cDNA fused to the sequences encoding the DNA binding and dimerization domains of the yeast GAL4 protein (amino acids 1-147, hereafter referred to as GAL-DBD). These GAL-CTF1 constructs were transfected in mouse NIH3T3 cells together with a reporter construct containing five GAL4 binding sites in front of the cat reporter gene (G5BCAT; Ref. 21). As shown previously (17), the prototype of the GAL4 fusion proteins, GAL 399 -499, containing the entire proline-rich transactivation domain of CTF-1 (amino acids 399 -499), activated transcription in NIH3T3 cells, and its activity was increased by TGF-␤ ( Fig.  2A). However, both the basal as well as the TGF-␤-induced GAL 399 -499 transcriptional activity were repressed by either human or mouse TNF-␣ with identical potencies ( Fig. 2A and data not shown). This effect was specific for TNF-␣ because interleukin 1␤, serum, and PDGF-BB had no effect on GAL 399 -499 transcriptional activity. Moreover, the TADs of other known transcription factors, such as that of the proline-rich transactivator AP2, the glutamine-rich TADs of Sp1 and Oct2, and the acidic GAL4 and VP16 TADs, all failed to mediate both TNF-␣ repression and TGF-␤ induction when fused to GAL-DBD ( Fig. 2A and data not shown). GAL 1-399, a GAL4 fusion encoding a CTF-1 derivative without the proline-rich TAD, was not detectably repressed by TNF-␣, whereas a chimera containing the minimal GAL-DBD (amino acids 1-93) fused to the CTF-1 TAD did mediate TNF-␣ repression. GAL4 fusions, containing either the proline-rich domain of CTF-2 or the entire CTF-2 coding sequence, were not detectably repressed by TNF-␣ (data not shown), suggesting that TNF-␣ repression is restricted to specific CTF/NF-I species. Moreover, gel mobility shift assays demonstrated that TNF-␣ had no significant effect on the DNA binding affinity or the expression levels of the GAL4 fusion proteins when given either alone or in combination with TGF-␤ ( Fig. 2B; see below). Taken together, the above results demonstrate that the proline-rich domain of CTF-1 is sufficient to confer TNF-␣ transcriptional repression when tethered to an heterologous DNA binding domain in NIH3T3 cells.
Progressive deletion from the C-terminal end of the CTF-1 TAD eliminated both basal transcriptional activity and TGF-␤ inducibility ( Fig Fig. 2A). Gel-shift analysis confirmed that TNF-␣ did not significantly affect the levels of these fusion proteins ( Fig. 2B and data not shown). Thus, these results collectively demonstrate that TNF-␣ represses the activity of the CTF-1 prolinerich domain, thereby antagonizing TGF-␤ induction, and that the CTF-1 TRD mediates both responses in NIH3T3 cells.
Inhibition of CTF-1 TAD Activity by Oncogenic Ras or Raf-1-Our subsequent experiments focused on the identification of signal transduction intermediates involved in the antagonistic regulation of CTF-1 transcriptional activity by TGF-␤ and TNF-␣. Several intracellular mediators of TNF-␣ action are known, including arachidonic acid (AA) and its metabolites prostaglandins and leukotrienes, reactive oxygen intermediates (ROIs), and protein kinases (reviewed in Refs. 7 and 28). Since it had been proposed that prostaglandins might at least partially mediate TNF-␣ inhibition of collagen ␣1(I) transcription in fibroblasts (29), we initially assessed whether stimulation of mammalian cells with AA might affect GAL 399 -499 transcriptional activity. However, exogenous AA did not regulate GAL 399 -499, nor did it alter the TNF-␣-mediated repression of basal or TGF-␤-activated GAL 399 -499 transcription (Fig. 3). In contrast, AA efficiently stimulated the transcriptional activity of the previously identified AA-responsive transcription factor PPAR (30) in fibroblastic cells (Fig. 3). Consistent with the lack of effect of exogenous AA on CTF-1 16), as described under "Materials and Methods." Five hours after transfection, cells were induced with ethanol vehicle or with 5 ng/ml human TGF-␤ and/or 50 ng/ml human TNF-␣, as indicated. CAT activities were determined 18 h later and normalized to ␤-galactosidase activity. The mean values (bars, S.D.) of five independent experiments are presented, expressed as normalized CAT activity relative to that obtained with p␣CAT-⌬87-3xAd in the absence of hormones, which was arbitrarily set to 100. B, TNF-␣ does not affect the DNA-binding activity and expression of the CTF/NF-I proteins. Equal protein amounts of whole-cell lysates of NIH3T3 cells that had been treated with or without TNF-␣ for 18 h as in A (lanes 2 and 3, respectively) or lysate extraction buffer (lane 1) were analyzed for CTF/NF-I binding activity in a gel mobility shift assay, as described under "Materials and Methods." The position of migration of the various DNA-bound CTF/NF-I species is indicated by the bracket, whereas the unbound probe is shown by the arrow.

FIG. 2. TNF-␣ down-regulates the CTF-1 transcriptional activation domain in NIH3T3 cells.
In A, the TGF-␤-responsive domain of CTF-1 mediates TNF-␣ transcriptional down-regulation. NIH3T3 cells were transfected with expression vectors for the indicated GAL-DBD fusion proteins (the various GAL-CTF1 derivatives, GAL-AP2 or GAL-Sp1), the internal control plasmid CMV␤gal, and the G5BCAT reporter construct driven by five GAL4 binding sites (21). Five hours after transfection, cells were induced with ethanol vehicle or with 5 ng/ml TGF-␤ and/or 50 ng/ml of TNF-␣, as indicated. CAT activities were determined 40 h later and normalized to ␤-galactosidase activity. The mean values (bars, S.D.) of four independent experiments are presented, expressed as normalized CAT activity relative to that obtained with GAL 399 -499 in the absence of hormones, which corresponds to a mean 45 Ϯ 11 fold activation over the G5BCAT basal level, and was arbitrarily set to 100. In B, NIH3T3 cells were either mock-transfected or transfected with the indicated expression vectors, as in A. Cells treated with or without TNF-␣ for 40 h were lysed, and equal protein amounts of whole-cell lysates were analyzed for GAL4 DNA-binding activity in a gel mobility shift assay, as described under "Materials and Methods." The relevant GAL4 fusion-DNA complexes are indicated by arrows, whereas the closed circle indicates complexes that may result from proteolytic degradation products of the GAL4 fusion proteins. transcriptional activity, we found that overexpression of the 85-kDa cPLA 2 phospholipase, which generates intracellular AA, or specific inhibition of AA metabolism with NDGA or indomethacin, all failed to block or mimic TGF-␤ and TNF-␣ regulation of CTF-1 (data not shown). Taken together, these results thus indicate that AA and its metabolites are not involved in the signal transduction pathways that regulate CTF-1 activity in response TGF-␤ and TNF-␣.
The failure of NDGA to block TNF-␣ repression was particularly interesting because NDGA also possesses potent antioxidant activity, as it protects some cells from the cytotoxic effect of TNF-␣ (28). The well-known compounds pyrrolidine dithiocarbamate and N-acetylcysteine are also potent antioxidants and efficient inhibitors of TNF-␣-induced, ROI-mediated nuclear factor-B activation (31). However, treatment of NIH3T3 cells with these drugs or overexpression of the c-Myc oncoprotein, which potentiates TNF-␣ cytotoxicity (32), did not prevent the regulation of GAL 399 -499 by TGF-␤ and TNF-␣ in transient transfection assays (data not shown), implying that TGF-␤ and TNF-␣ signal on CTF-1 through ROI-and cellular growth-independent pathways.

TNF-␣ Repression of CTF-1 Transcriptional Activity Is Not Mediated by Direct Phosphorylation of Its Proline-rich Domain-Growth factor-induced activation of the Ras-Raf pathway results in MAP/ERK-mediated phosphorylation and activation of transcription factors such as TCF/Elk-1 (reviewed in
Ref. 34). Our data suggested that TNF-␣ might repress CTF-1 transcriptional activity in a similar manner, by regulating a Ras/Raf-dependent direct phosphorylation of the CTF-1 TAD. Since TNF-␣ repression is mediated by the TRD portion of CTF-1 (see above), which contains three potential phosphoacceptor sites, (Fig. 6), we mutated these residues to neutral amino acids and asked whether these mutations might block TNF-␣ repression. However, GAL-CTF1 variants harboring a triple mutation of their TRD (where the tyrosines 491 and 497 are mutated to phenylalanines and serine 495 to alanine) were similarly repressed by TNF-␣, as compared to their parental constructs (Fig. 6, compare GAL 479 -499 3xM and GAL 399 -499 3xM with W.T. constructs). Moreover, other GAL 399 -499 mutants containing a single point mutation of each phosphoac-ceptor (for example GAL 399 -499 S495A, Fig. 6) were also regulated as wild-type GAL 399 -499, thereby ruling out the possibility that the triple GAL 399 -499 mutant might still mediate regulation because of simultaneous removal of positive and negative regulatory determinants. Consistently, "solidphase" kinase assays (41) demonstrated that the phosphorylation of bacterially expressed GST-CTF1 fusion proteins by NIH3T3 cell extracts was not modified by prior treatment of the cells with TNF-␣ and/or TGF-␤ (data not shown). These results demonstrate that TNF-␣ does not regulate CTF-1 through direct phosphorylation of the TRD. Thus, the TNF-␣ regulatory effect might be exerted on another factor(s), which interacts and cooperates with the CTF-1 proline-rich domain in mediating TNF-␣ transcriptional repression.
The Interaction of the CTF-1 Transactivation Domain with Histone H3 Is Down-regulated by TNF-␣ in NIH3T3 Cells-We have recently identified histone H3 as such a CTF-1-interacting protein in the yeast two-hybrid system. In NIH3T3 cells, histone H3 interacts with the TRD portion of the CTF-1 TAD, and TGF-␤ up-regulates this interaction (17). Since the TRD also mediates TNF-␣ transcriptional repression, we asked whether the histone H3-CTF1 interaction might be a target of TNF-␣ regulation. To address this question, we used the previously described two-hybrid assay in NIH3T3 cells (17) that detects protein-protein interactions between specific GAL4 fusions and a chimera consisting of the herpes simplex virus VP16 transactivation domain fused to the coding sequence of mouse histone H3.3 (hereafter referred to as VP-H3). In transiently transfected NIH3T3 cells, such an interaction between the two chimeric proteins recruits the potent VP16 TAD to a  , lanes 1-16), or with the AP-1-responsive 5xTRE-TATA-CAT reporter promoter (22) (right panel, lanes 17-24). Cells were also cotransfected with a vector expressing an oncogenic Ras mutant (EJ-Ha-Ras; Ref. 23) or with the same vector without insert, as indicated. Four hours after transfection, cells were induced with 5 ng/ml TGF-␤ and/or 50 ng/ml TNF-␣, or they were left untreated, as indicated. CAT activities were determined 15 h later and normalized to ␤-galactosidase activity. The mean values (bars, S.D.) of at least three independent experiments are presented, expressed as normalized CAT activity relative to that obtained with GAL 399 -499 in the absence of hormones, which was set to 100.
GAL4-responsive promoter, leading to higher levels of transcriptional activation (superactivation). Accordingly, VP-H3 efficiently superactivated GAL 399 -499, the TRD-containing fusion GAL 486 -499 and GAL-Sp1 in NIH3T3 cells (Fig. 7A). This effect was specific, because GAL-DBD, GAL-AP2, and several of other GAL4 fusion proteins did not mediate significant VP-H3 superactivation, whereas expression of either the VP or histone H3 moieties alone or of VP16 fusions to other histones had no effect on GAL 399 -499 activity ( Fig. 7A; Ref.  17). TGF-␤ up-regulated the interaction of both GAL 399 -499 and GAL 486 -499 with VP-H3, because superactivation was increased in the presence of the factor, but it did not upregulate the interaction of VP-H3 with GAL-Sp1 (Fig. 7A). These results are consistent with the previously proposed specific potentiation of the CTF-1-histone H3 interaction by TGF-␤ in NIH3T3 cells (17). Strikingly, TNF-␣ appeared to downregulate the interaction of VP-H3 with GAL 399 -499 and GAL 486 -499, as superactivation was decreased when TNF-␣ was present, either alone or in combination with TGF-␤ (Fig. 7A). This effect was specific for GAL 399 -499 and GAL 486 -499, as TNF-␣ did not decrease the superactivation of GAL-Sp1 by VP-H3. Control experiments demonstrated that the expression of the GAL4 derivatives was not affected by either TNF-␣ or VP-H3 under these conditions (Fig. 7B and data not shown). Thus, these results imply that TNF-␣ may down-regulate the interaction of CTF-1 TRD with histone H3, and they further suggest that histone H3 might be a direct molecular target of antagonistic TGF-␤ and TNF-␣ regulation. DISCUSSION CTF-1 is the prototype member of the CTF/NF-I family of proline-rich transcription factors that mediate regulation of a variety of viral and cellular genes, including several procollagen genes (8,16,42,43). Collagen synthesis and secretion by mesenchymal cells is potently promoted by TGF-␤ during the process of wound healing, whereas TNF-␣ down-regulates collagen gene expression, thereby antagonizing TGF-␤ action (8,9,14,15). Previous studies of the human ␣2(I) collagen promoter had identified a promoter element termed TbRE (for TGF-␤-responsive element, Ref. 14), which can confer antagonistic TGF-␤ and TNF-␣ regulation on heterologous promoters, and is sufficient to direct correct tissue-specific expression of a marker gene in transgenic mice (8,9,14,15,44). Putative CTF/NF-I-and Sp1-like binding sites were mapped on the TbRE, which led to the proposal that such transcription factors might mediate promoter regulation by TGF-␤ and/or TNF-␣ (8,14,15). However, no difference in the binding affinity of either CTF/NF-I or Sp1 in response to TGF-␤ and TNF-␣ could be demonstrated by these authors, and thus the molecular mechanisms underlying the functional antagonism of TGF-␤ and TNF-␣ on collagen gene expression remained obscure.
Our results provide evidence regarding the potential mechanism of antagonistic regulation of the promoter by TGF-␤ and TNF-␣. Using a reporter promoter containing high affinity CTF/NF-I binding sites, we demonstrate that CTF/NF-I binding sites specifically mediate opposing TGF-␤ and TNF-␣ transcriptional regulation. Both TGF-␤ (8) and TNF-␣ (Fig. 1B) have no apparent effect on the CTF/NF-I levels, suggesting that the hormones might antagonistically regulate the transcriptional activity of individual CTF/NF-I species. Indeed, we have shown recently that TGF-␤ induces the proline-rich transactivation domain of CTF-1 in NIH3T3 fibroblasts (17). We now report that TNF-␣ down-regulates the activity of the CTF-1  , lanes 1-12), or with the AP-1-responsive promoter 5xTRE-TATA-CAT (22) (right panel, lanes 13-18). Some cells received a vector expressing a constitutively active deletion mutant of the c-Raf-1 kinase (Raf-BXB; Ref. 24), or the same vector without insert, as indicated. Four hours after transfection, cells were induced with 5 ng/ml TGF-␤ and/or 50 ng/ml TNF-␣ or were left untreated, as indicated. CAT activities were determined 15 h later and normalized to ␤-galactosidase activity. The mean values (bars, S.D.) of three independent experiments are presented, expressed as normalized CAT activity relative to that obtained with GAL 399 -499 in the absence of hormones, which was set to 100. TAD, thereby antagonizing TGF-␤ induction, and that the TGF-␤-responsive domain of CTF-1 mediates both the transcriptional activation and repression effects. Importantly, the naturally occurring CTF-2 TAD, which lacks a TRD, as well as the TADs of other known transcription factors, does not mediate TNF-␣ repression or TGF-␤ induction in NIH3T3 cells. Moreover, growth factors such as PDGF-BB and interleukin 1␤ have no effect on CTF-1 transcriptional activity, consistent with their lack of effect on the ␣2(I) collagen promoter (8 -10). In contrast, oncogenic Ras expression results in down-regulation of promoter activity in fibroblasts (35), and we find that it also represses CTF-1 transcriptional activity. Thus, our results provide a perfect correlation between the regulation of CTF-1 activity and the regulation of the collagen promoter by growth factors and oncogenes. CTF-1 may thus be a major regulator of ␣2(I) collagen and perhaps also of other genes subjected to antagonistic TGF-␤ and TNF-␣ regulation in vivo.
TGF-␤ induces the growth of NIH3T3 cells and other embryonic fibroblasts through poorly understood mechanisms that may involve increased synthesis of PDGF, which would then act in an autocrine fashion (reviewed in Refs. 6 and 12). Conversely, long-term exposure of NIH3T3 cells to TNF-␣ eventually leads to cell death (45). However, murine TNF-␣ is more cytotoxic than human TNF-␣ in inducing cell death, and this may result from the inability of the latter to bind to both the 55and 70-kDa TNF-␣ receptors at the cell surface, as does murine TNF-␣ (32,45). TNF-␣ cytotoxicity may further be potentiated by expression of the c-Myc oncoprotein (32), and c-Myc overexpression had been associated with decreased CTF/NF-I transcriptional activity in 3T3-L1 cells (46). These findings raised the possibility that the antagonistic regulation of CTF-1 by TGF-␤ and TNF-␣ might be a consequence of their antagonistic effect on cell growth, and that c-Myc might perhaps mediate this antagonism. However, we have not observed any difference in the transcriptional activity and regulation of CTF-1 in either transiently or stably c-Myc-overexpressing NIH3T3 cells. Although this may simply reflect differential regulation of CTF-1 in 3T3-L1 compared to NIH3T3 cells, several independent lines of evidence further suggest that TGF-␤ and TNF-␣ initiate at least two distinct pathways in NIH3T3 cells, one leading to growth control and the second to CTF-1 regulation. (i) Serum or PDGF-BB treatment potently activates the growth of NIH3T3 cells, yet it does not increase GAL 399 -499 transcriptional activity, and it does not prevent antagonistic CTF-1 regulation by TGF-␤ and TNF-␣. (ii) TGF-␤ induction and TNF-␣ repression are similar in actively growing or quiescent NIH3T3 cells.
(iii) NIH3T3 cells stably transfected with cdk4 plus cyclin D2, or cells transiently overexpressing the p27/Kip1 or p21/Cip1 cdk inhibitors, still mediate TGF-␤ induction. 2 (iv) Pharmacological inhibitors of arachidonic acid and ROI generation such as NDGA and pyrrolidine dithiocarbamate, which have been proposed to interfere with TNF-␣-induced cytotoxicity and nuclear factor-B activation (28,47,48), cannot block TNF-␣ repression. (v) Murine TNF-␣ and human TNF-␣ were equally effective in mediating CTF-1 repression in NIH3T3 cells, yet mTNF-␣ is far more cytotoxic (see above; Refs. 32 and 45). Thus, these data altogether strongly support the notion that the antagonistic regulation of CTF-1 activity by TGF-␤ and TNF-␣ in NIH3T3 fibroblasts occurs independently of the control of cell growth.
Based on the previous observation that oncogenic Ras expression leads to decreased procollagen transcription in fibroblasts (35), we examined the potential role of Ras and Raf in the antagonistic regulation of CTF-1 transcriptional activity by TGF-␤ and TNF-␣. Expression of activated forms of either oncogene had no significant effect on the induction of GAL 399 -499 by TGF-␤, suggesting that their expression does not interfere with TGF-␤ regulation in NIH3T3 cells. However, we found that expression of activated Ras and Raf variants is sufficient to specifically mediate CTF-1 repression in NIH3T3 cells, thereby acting indistinguishably from TNF-␣. In contrast, expression of the JNK/SAPK kinase, which mediates TNF-␣ induction of the c-Jun TAD (49,50), and which is apparently induced by TGF-␤ in MC3T3-E1 cells (51), did not affect CTF-1 regulation. Thus, our results imply Ras and Raf-1 in the TNF-␣ signal transduction to CTF-1, and they further argue that the pathways leading to CTF-1 and c-Jun regulation are distinct. Nevertheless, further experiments are required to determine whether Ras and Raf are part of the TNF-␣ signal transduction pathway to CTF-1.
Growth factor-regulated activation of the Ras-Raf pathway results in ERK-mediated phosphorylation of target transcription factors, which is often associated with changes in their transactivation potential (34). We, therefore, tested whether TNF-␣ might similarly repress CTF-1 through direct phosphorylation of its proline-rich domain. Our analysis demonstrated that TNF-␣ targets the same regulatory domain of CTF-1 previously shown to confer TGF-␤ responsiveness, the TRD (17). 2 A. Alevizopoulos, unpublished observations. FIG. 6. TNF-␣ does not regulate CTF-1 through direct phosphorylation of the TGF-␤ responsive domain. NIH3T3 cells were cotransfected with the G5BCAT reporter construct, the internal control plasmid CMV␤gal, and GAL-CTF1 expression vectors encoding either the wild-type CTF-1 TRD (amino acids 479 -499) or full-length CTF-1 TAD (amino acids 399 -499), or their mutated derivatives harboring either a single mutation of tyrosines 491 or 497 to phenylalanine (Y/F) or a serine 495 to alanine (S/A) mutation, or all three mutations of these amino acids (3xM), as indicated. Four hours after transfection, cells were induced with 5 ng/ml TGF-␤ and/or 50 ng/ml TNF-␣, or left untreated, as indicated. CAT activities were determined 40 h later and normalized to ␤-galactosidase activity. Values are the means of two representative independent experiments and are expressed as normalized CAT activity relative to that obtained with GAL 399 -499 in the absence of hormones, which was set to 100. The primary sequence of the last 20 amino acids of CTF-1 TAD representing the TRD is displayed at the bottom, and altered amino acids in the mutant GAL-CTF1 derivatives are indicated below the wild-type sequence.
However, the TRD is not the direct target of regulatory phosphorylation by either growth factor (Ref. 17 and Fig. 6), suggesting that TGF-␤ and TNF-␣ antagonistically regulate the binding of accessory proteins on the TRD. We have previously identified histone H3 as such a TRD-interacting protein in the yeast two-hybrid system and have shown that TGF-␤ up-regulates this interaction in NIH3T3 cells (17). We, therefore, tested whether TNF-␣ down-regulates the interaction of GAL  H3)), or the MLV-VP vector, which expresses the VP16 TAD (VP), or the same vector without insert (17), as indicated. Four hours after transfection, cells were induced with 5 ng/ml TGF-␤, 50 ng/ml TNF-␣, or left untreated. CAT activities were determined 40 h later and normalized to ␤-galactosidase activity. The mean values (bars, S.D.) of three independent experiments are presented, expressed as normalized CAT activity relative to that obtained with GAL 399 -499 in the absence of hormones and in the presence of VP, which was set to 100. In B, VP-H3 expression does not significantly affect GAL-fusion protein levels. NIH3T3 cells were cotransfected with vectors expressing GAL 399 -499 (lanes 2-9) and plasmids expressing either the VP (lanes 2-5) or the VP-H3 (lanes 6 -9) superactivators or mock-transfected (lane 1), as in A. Four hours after transfection, cells received 5 ng/ml TGF-␤ and/or 50 ng/ml TNF-␣, or left untreated, as indicated. Forty hours later, the cells were collected and lysed. Equal protein concentration of cell lysates were analyzed for GAL4 DNA-binding activity using a gel mobility shift assay, as described under "Materials and Methods." The GAL 399 -499-DNA complex is indicated by an arrow, whereas the closed circle indicates complexes that may result from proteolytic degradation products of the GAL4 fusion proteins. 399 -499 with histone H3. Indeed, we find that two TRDcontaining GAL-CTF1 variants are able to mediate TNF-␣responsive histone H3 interaction, in contrast to GAL-Sp1, the interaction of which with histone H3 is constitutive. Thus, although a role for other CTF-1-interacting proteins cannot be excluded, our results are consistent with the notion that histone H3 might be implicated in the antagonistic regulation of CTF-1 transcriptional activity. Our data further argue that the CTF-1-histone H3 interaction might be down-regulated, directly or indirectly, in response to a Ras-Raf signaling cascade engaged by the TNF-␣ receptors. Interestingly, other growth factors known to activate Ras, such as epidermal growth factor, have been shown to induce histone H3 phosphorylation, and this modification correlates with increased gene expression (52).
The proline-rich domain of CTF-1 mediates TNF-␣ transcriptional repression (this study), and it is capable of binding to histone octamers, leading to reconfiguration of nucleosomal architecture in vitro (17). This suggests that TNF-␣ might regulate chromatin remodeling through CTF-1 in vivo. Indeed, in the case of the human immunodeficiency virus 1 promoter (HIV-1 LTR), TNF-␣ treatment induces disruption of a target nucleosome that is translationally positioned (53) close to CTF/ NF-I binding sites (54). Thus, disruption of the HIV LTR nucleosome may occur through a TNF-␣-regulated interaction of CTF-1 with histone H3.
In summary, the results presented in this study identify CTF-1 as a novel molecular target of TNF-␣ and show that CTF-1 mediates antagonistic transcriptional regulation in response to TGF-␤ and TNF-␣. Given the wide range of opposing TGF-␤ and TNF-␣ actions and the broad expression of the various CTF/NF-I proteins, these findings uncover novel mechanisms that may regulate gene expression in response to the two growth factors, and they illustrate how multiple signaling pathways are integrated by converging to common target regulatory proteins.