Role of Smad Proteins and Transcription Factor Sp1 in p21Waf1/Cip1 Regulation by Transforming Growth Factor-β*

Transforming growth factor-β (TGF-β) inhibits cell cycle progression, in part through up-regulation of gene expression of the p21WAF1/Cip1(p21) cell cycle inhibitor. Previously we have reported that the intracellular effectors of TGF-β, Smad3 and Smad4, functionally cooperate with Sp1 to activate the human p21 promoter in hepatoma HepG2 cells. In this study we show that Smad3 and Smad4 when overexpressed in HaCaT keratinocytes lead to activation of the p21 promoter. Activation requires the binding sites for the ubiquitous transcription factor Sp1 on the proximal promoter. Induction of the endogenous HaCaTp21 gene by TGF-β1 is further enhanced after overexpression of Smad3 and Smad4, whereas dominant negative mutants of Smad3 and Smad4 and the inhibitory Smad7 all inhibit p21induction by TGF-β1 in a dose-dependent manner. We show that Sp1 expressed in the Sp1-deficient Drosophila SL-2 cells binds to the proximal p21 promoter sequences, whereas Smad proteins do not. In support of this finding, we show that DNA-binding domain mutants of Smad3 and Smad4 are capable of transactivating the p21 promoter as efficiently as wild type Smads. Co-expression of Smad3 with Smad4 and Sp1 in SL-2 cells or co-incubation of phosphorylated Smad3, Smad4, and Sp1 in vitro results in enhanced binding of Sp1 to the p21 proximal promoter sequences. We demonstrate that Sp1 physically and directly interacts with Smad2, Smad3, and weakly with Smad4 via their amino-terminal (Mad-Homology 1) domain. Finally, by using GAL4 fusion proteins we show that the glutamine-rich sequences in the transactivation domain of Sp1 contribute to the cooperativity with Smad proteins. In conclusion, Smad proteins play important roles in regulation of the p21 gene by TGF-β, and the functional cooperation of Smad proteins with Sp1 involves the physical interaction of these two types of transcription factors.

Transforming growth factor-␤ (TGF-␤) 1 is the prototype of a family of multifunctional cytokines that regulate many aspects of cell physiology, including cell growth, differentiation, motility and death, and play important roles in many developmental and pathological processes (1,2). TGF-␤ inhibits cell proliferation of various cell types of epithelial origin by repressing the expression of the proto-oncogene c-myc and by inhibiting the activity of cyclin-dependent kinases (CDKs) which leads to the arrest of the cell cycle at an early G 1 phase (3)(4)(5)(6). The mechanism of action of TGF-␤ in modulating the activity of the CDKs involves the regulation of the CDK inhibitors (CKIs) p15 Ink4B , p21 Waf1/Cip1 , and p27 Kip1 and the repression of the CDK phosphatase cdc25A (4,(7)(8)(9)(10). This regulation consists of both transcriptional induction of the genes for p15 and p21 and partitioning of the CKIs between different complexes with CDKs (4,7,8). Thus, the transcriptional induction of the genes for the two CKIs, p15 and p21, has been postulated at least as partially determining the anti-proliferative action of TGF-␤ (11).
The p21 gene is regulated by a rapidly growing list of physiological and pathological factors, such as tumor suppressors of the p53 family, differentiation factors, growth factors, cytokines, and stress factors (Ref. 12 and references therein). However, the detailed transcriptional mechanisms involved in p21 gene regulation by the above factors still remain poorly understood. In the cases of p53, vitamin D3, interferon ␥, and other signals, factor-specific DNA motifs scattered in the region between Ϫ2,300 and Ϫ210 base pairs upstream from the transcriptional initiation site of the p21 gene have been shown to mediate the response of this gene to the above stimuli (13)(14)(15). In contrast, for a large number of other signaling factors such as, TGF-␤1, progesterone, phorbol esters, and Rho GTPases, the proximal region of the p21 promoter (base pairs Ϫ210 to ϩ1) is the major site for reception of the inducing signal (Ref. 12 and references therein). This short, proximal promoter contains several closely spaced G/C-rich motifs that serve as binding sites for members of the Sp1 family of transcription factors (16).
Sp1 is a ubiquitously expressed transcription factor with a zinc finger DNA-binding domain that recognizes G/C-rich DNA sequences (17,18). Sp1 is required for early embryogenesis and regulates the terminal differentiation state of cells by affecting the methylation of DNA CpG islands (19). The transcriptional activity of Sp1 is regulated by phosphorylation in a cell cyclespecific manner, acetylation by the co-activator p300, and glycosylation, which protects this factor from proteasome-dependent degradation (20 -22). Sp1 exerts its transcriptional properties by interacting directly with factors of the basal transcription machinery and by cooperating with several transcriptional activators (23)(24)(25)(26)(27)(28)(29). Thus, although Sp1 traditionally appeared as a ubiquitous factor primarily serving the core activity of promoters, recent evidence increasingly implicates this protein in several instances of regulated gene transcription.
It is established that transcription factor Sp1 participates in the regulation of the p21 gene by TGF-␤ (16,30). The TGF-␤ signaling pathway utilizes plasma membrane serine/threonine kinase receptors and their cytoplasmic effectors termed the Smad proteins (2,31). Smad proteins are transcriptional activators that bind to DNA and cooperate with a fast growing list of transcription factors in regulating target gene expression (32)(33)(34). In a previous report (35) we have provided evidence for the role of Smad3 and Smad4 in mediating the induction of the p21 gene by TGF-␤1. The Smads were shown to cooperate functionally with Sp1 and depend on the Sp1-binding sites of the p21 promoter for their action. In addition, we have demonstrated that Jun family members, which themselves are induced and activated by TGF-␤, can also regulate the p21 promoter by physically interacting with Sp1 and utilizing the same G/C-rich motifs of the proximal promoter (12). In the present work we demonstrate that Smad3 and Smad4 enhance the level of endogenous HaCaT p21 gene induction by TGF-␤, whereas dominant negative Smads and the inhibitory Smad7 block p21 induction by TGF-␤ in a dose-dependent manner. These results correlate very well with the p21 promoter transactivation studies in the same cells. We show that Smads mediate enhancement of the Sp1 affinity for the p21 promoter, independent from a direct association of Smads to DNA, and Smad2, Smad3, and Smad4 physically interact with Sp1. We have mapped the domain of Smad3 and Smad4 required for this interaction and provide evidence for the involvement of specific Sp1 sequences in the functional cooperation between these two classes of transcription factors.

EXPERIMENTAL PROCEDURES
Materials-The purified baculoviral Smad3, TGF-␤ type I receptorphosphorylated Smad3 and Smad4 proteins were a generous gift from F. M. Hoffman and A. Comer (36). Restriction enzymes and modifying enzymes (T4 DNA ligase, T4 polynucleotide kinase, Klenow fragment of DNA polymerase I, and calf intestinal alkaline phosphatase) were purchased from Minotech, New England Biolabs, or Life Technologies, Inc. Vent DNA polymerase was from New England Biolabs. The Sequenase version 2 kit, poly(dI/dC), acetyl-CoA, dNTPs, protein A, protein G-Sepharose beads, and the GST purification kit were from Amersham Pharmacia Biotech. Isopropyl-␤-D-thiogalactopyranoside was from Calbiochem. [␥-32 P]ATP, [␣-32 P]dCTP, and [ 14 C]chloramphenicol were from Amersham Pharmacia Biotech or NEN Life Science Products. All reagents for cell culture (Dulbecco's modified Eagle medium (DMEM), fetal bovine serum, trypsin-EDTA, and phosphate-buffered saline) were from Life Technologies, Inc. O-Nitrophenyl galactopyranoside and the monoclonal anti-FLAG (M5, F-4042) antibody were from Sigma. The transfection reagent Fugene-6 and the monoclonal anti-hemagglutinin (HA) (12CA5) antibody were from Roche Molecular Biochemicals. The luciferase assay system, the consensus Sp1 oligonucleotide, and purified Sp1 protein were from Promega. All oligonucleotides were synthesized at the microchemical facility of the IMBB, Heraklion, Greece. The monoclonal anti-Myc antibody was produced by the 9E10 hybridoma cell clone. The monoclonal anti-GAL4 DNA-binding domain antibody (RK5C1), the monoclonal anti-Smad1/2/3 (H-2) antibody, the rabbit polyclonal anti-Sp1 (PEP-2) antibody, and the polyclonal anti-hexahistidine (His-probe H-15) antibody were from Santa Cruz Biotechnology. The monoclonal anti-␤-catenin (C19220) and the monoclonal anti-p21 (C24420) antibodies were from Transduction Laboratories. The antiphosphoserine rabbit polyclonal antibody (Poly-Z-PS1) was from Zymed Laboratories Inc. The anti-mouse horseradish peroxidase-conjugated secondary antibody was from Amersham Pharmacia Biotech. All other chemicals were obtained from the usual commercial sources at the purest grade available.
Plasmid Constructions-The p21 promoter plasmid Ϫ2,300/ϩ8 p21 luc has been described previously (35). The p21 promoter deletion construct Ϫ143/ϩ8 p21 luc was constructed by transferring the XbaI to HindIII promoter fragment from the Ϫ143/ϩ8 p21 CAT plasmid (35) to pGL2-basic after digestion with NheI and HindIII. The expression vectors pCDNA3-6myc-Smad2, pCDNA3-6myc-Smad3, pCDNA3-6myc-Smad4, and pCDNA3-HA-CA-ALK5 and the reporter construct pGL3-CAGA 12 -MLP-luc were generously provided by Dr. S. Itoh of the Ludwig Institute, Uppsala, Sweden. The expression vector pCDNA1/ amp encoding the FLAG-tagged human Smad3 was described previously (12). The expression vector encoding the FLAG-tagged Smad4 protein was the generous gift of Dr. J. Massagué (Memorial Sloan-Kettering Institute, New York). The expression vectors encoding the three deletion mutants of Smad3, Smad3-(1-122), Smad3-(1-248), and Smad3-(122-424) were constructed by polymerase chain reaction amplification of the specified fragments using appropriate primers and subsequent subcloning of the amplified fragments into the expression vector pCDNA1/amp. The expression vectors encoding the FLAGtagged human Smad3 with the double mutation (R74K and K81R) and the FLAG-tagged human Smad4 with the double mutation (R81K and K88R) in the Smad DNA-binding domain will be described in detail elsewhere. 2 The GAL4(DBD)-Sp1 fusion constructs pSG424/GAL4-Sp1AϩB, pSG424/GAL4-Sp1B, pSG424/GAL4-Sp1Bn, pSG424/GAL4-Sp1Bc, and the pBXG1 plasmid containing the GAL4 DBD portion only were the generous gifts of Dr. G. Gill, Harvard Medical School, Boston. The pG 5 B-CAT reporter containing five tandem GAL4-binding sites in front of the E1B minimal promoter and the CAT reporter gene was the generous gift of Dr. G. Mavrothalassitis, University of Crete Medical School, Heraklion, Greece. The bacterial expression vectors pGEX-Sp1 (GST-Sp1 83-778), pGEX-Sp1 516C (GST-Sp1 ⌬A), pGEX-Sp1 N619 (GST-Sp1 ⌬D), and pGEX-Sp1 ⌬int 349 (GST-Sp1 ⌬BϩC) were the generous gifts of Dr. E. Flavey, Section of Molecular Genetics, Boston University Medical Center, Boston. The original Sp1 mutants were the generous gifts of Dr. R. Tjian, University of California, Berkeley. The bacterial expression vectors pGEX-Smad3, pGEX-Smad3⌬MH1, pGEX-Smad3⌬MH2, pGEX-Smad3MH1, pGEX-Smad3MH2, pGEX-Smad3-Linker, pGEX-Smad4, and pGEX-Smad4⌬MH2 were the generous gifts of Dr. S. Itoh of the Ludwig Institute, Uppsala, Sweden. The Drosophila expression vectors pPac-Sp1 and pPacO were the generous gifts of Dr. J. M. Horowitz, North Carolina State University, Raleigh, and J. Noti, Guthrie Research Institute, Sayre, PA, respectively. The Drosophila expression vectors pRactH-Smad3 and pRactH-Smad4 were constructed by transferring the corresponding Smad cDNA that includes only the protein-coding region from pCDNA1/amp-Smad3 and pCDNA3-Smad4, respectively, into the BamHI and HindIII sites of the polycloning region of the basic Drosophila expression vector pRactH. The pRactH and hsp-lacZ expression vector used for normalization of transfections in Drosophila SL2 cells were the generous gifts of Dr. C. Delidakis, University of Crete, and IMBB, Heraklion, Greece. The quality of all new DNA constructs was verified by DNA sequencing.
Cell Cultures, Transient Transfections, Adenoviral Infections, Reporter, and Western Blot Assays-Human HaCaT keratinocytes, human hepatoma HepG2 cells, and monkey kidney COS-7 cells were cultured in DMEM supplemented with 10% fetal bovine serum, L-glutamine, and penicillin/streptomycin at 37°C in a 5% CO 2 atmosphere. Drosophila Schneider's SL2 cells were cultured in Schneider's insect medium supplemented with 10% insect culture-tested fetal bovine serum and penicillin/streptomycin at 27°C. Transient transfections of COS-7 cells for co-immunoprecipitation assays and of Schneider's SL2 cells for nuclear extract isolation were performed using the liposome reagent Fugene-6 according to the manufacturer's protocol (Roche Molecular Biochemicals). Transient transfections using the calcium phosphate co-precipitation method, chloramphenicol acetyltransferase, luciferase, and ␤-galactosidase assays were performed as described previously (35). Adenoviral stocks were maintained, and infections were performed as described previously (37). The dominant negative (DN) Smad3 and DN-Smad4-encoding adenoviruses were the generous gifts of Dr. Theodore Fotsis, University of Ioannina, Ioannina, Greece. Under optimal conditions, more than 90% of the cells were infected as determined by the blue, ␤-galactosidase-positive staining using a lacZ control virus or the green fluorescent protein autofluorescence (for the DN-Smad3 and DN-Smad4 viruses). Routine infections were performed at a multiplicity of infection (m.o.i.) of 50 with single viruses. HaCaT cells were seeded at a density of 5 ϫ 10 4 cells/cm 2 in 24-well tissue culture plates. The next day the cells were transfected with the reporter constructs using the Fugene-6 reagent. Twelve hours later the culture medium was changed to DMEM containing 5% fetal bovine serum. Cells were infected 1 h later at the appropriate m.o.i. for 12 h and then washed and fed fresh 5% fetal bovine serum/DMEM. Twenty four hours later the reporter assays were performed, which corresponds to 49 h post-transfection and 36 h post-infection. For p21 protein analysis, HaCaT cells were seeded at a density of 10 6 cells/well in 6-well tissue culture plates. Twelve hours later the culture medium was changed to DMEM containing 5% fetal bovine serum and infected with different doses of each virus (see Fig. 2) 1 h later. Twenty four hours later cells were treated with 10 ng/ml TGF-␤1 for 20 h, and 20 g of detergent-soluble cell extracts were analyzed by Western blotting with the p21-specific antibody. FLAG, HA, and Smad1/2/3-specific antibodies served as controls to measure expression of the co-infected proteins, and ␤-catenin antibody served as control to verify equal protein amount loading. Western analysis of transiently transfected SL2 cell extracts or purified baculoviral Smad3 proteins was performed in the same fashion using the relevant antibodies as described under "Results" and figure legends.
Relative protein expression levels were quantified using the scanning densitometric software of the PhosphorImager Fujix BAS 2000. Ratios of band intensities of the tested protein (p21) over the control protein (␤-catenin) were calculated, and the ground condition ratio was set to 1 or 100% relative to which all other conditions are expressed.
Co-immunoprecipitation Assays-Forty hours post-transfection of COS-7 cells, total detergent extracts were prepared by lysing the cells in 20 mM Tris-HCl, pH 7.5, 150 mM NaCl, 10% (v/v) glycerol, 1% (v/v) Triton X-100, 1 mM phenylmethylsulfonyl fluoride (PMSF), 10 mM aprotinin at 4°C. The cell extracts were clarified from the insoluble material by a brief centrifugation at 10,000 rpm and were pre-cleared by incubation with protein A-Sepharose at 4°C for 30 min. The pre-cleared extracts were incubated with the FLAG or GAL4-DBD antibody at 4°C for 2 h, and the immunocomplexes were precipitated with protein-G Sepharose, washed with lysis buffer four times, and dissolved in Laemmli SDS-polyacrylamide gel electrophoresis (PAGE) loading buffer. After 7% SDS-PAGE the resolved proteins were transferred to Hybond-C extra nitrocellulose (Amersham Pharmacia Biotech), and the relevant proteins were detected after incubation with the anti-Myc (9E10) antibody followed by anti-mouse horseradish peroxidase-conjugated secondary antibody and homemade enhanced chemiluminescent assay on x-ray film (Fuji).
In Vitro and Bacterial Expression of Proteins-Expression of proteins in vitro was performed using the coupled in vitro transcription/translation system (TNT) of Promega as described previously (12). The quality of the synthesized proteins was verified by SDS-PAGE and autoradiography or PhosphorImager (Fujix BAS 2000) detection. The glutathione S-transferase (GST) fusion proteins were expressed in Escherichia coli strain DH-10␤ and purified as described previously (12). The solubilization of the expressed proteins was monitored by SDS-PAGE and Coomassie Brilliant Blue staining. All proteins were obtained at rather high levels and in relatively pure form as only the primary protein species were detectable without significant degradation products (Fig. 6, B, E, and H).
GST Protein Interaction Assays-Interaction assays of GST-Sp1 proteins with in vitro synthesized Smad proteins were performed as described previously (12). For the interaction assays of GST-Smad proteins with endogenous HaCaT Sp1, total cell extracts were prepared by lysis in 20 mM Tris-HCl, pH 7.4, 100 mM NaCl, 1 mM DTT, 0.5% Nonidet P-40, 1 mM PMSF, 1% aprotinin, 50 mM NaF, 25 mM ␤-glycerophosphate, 1 mM NaOVO 3 . Aliquots of the extracts corresponding to approximately 10 7 cells were incubated with the glutathione-Sepharose beads carrying 10 g of the GST-Smad fusions for 5 h at 4°C. The bound proteins were washed with the lysis buffer, dissolved in Laemmli SDS loading buffer, resolved by 8% SDS-PAGE, and analyzed by Western blotting using the Sp1 antibody and enhanced chemiluminescence.
Nuclear Extract Preparation and Gel Electrophoretic Mobility Shift Assays-Nuclear extracts from transiently transfected Schneider's SL2 cells were prepared 40 h post-transfection by hypotonic lysis of the cells in 10% (v/v) glycerol, 20 mM Hepes, pH 7.9, 0.1 mM EDTA, 50 mM KCl, 2 mM DTT, 0.5 mM spermidine, 50 mM NaF, 1 mM NaOVO 3 , 1 mM PMSF, 10 mM aprotinin by three sequential freeze (liquid nitrogen)thaw (4°C) cycles. The nuclear pellets obtained by centrifugation at 4,000 ϫ g were solubilized by slow rotation for 45 min in hypertonic buffer, 20% glycerol, 20 mM Hepes, pH 7.9, 0.1 mM EDTA, 600 mM KCl, 2 mM DTT, 50 mM NaF, 1 mM NaOVO 3 , 1 mM PMSF, 10 mM aprotinin at 4°C. The soluble nuclear extracts were collected after centrifugation at 12,000 ϫ g, were aliquoted, flash-frozen, and stored at Ϫ80°C. The corresponding cytoplasmic extracts were used for ␤-galactosidase assays to calibrate the extracts for transfection efficiency. The total protein in the nuclear extracts was measured by Bradford assay (Bio-Rad protein assay kit). The abundance of the transfected proteins in the nuclear extracts was estimated by Western blot analysis after 7% SDS-PAGE, using the relevant antibodies and enhanced chemiluminescence.
Gel electrophoretic mobility shift assays (GEMSAs) were performed as described previously (12). The sequences of the oligonucleotides from the p21 promoter and the consensus Sp1 site used in the GEMSA experiments were described by Kardassis et al. (12). The sequence of the sense strand of the mutant double-stranded Ϫ86/Ϫ70 p21 promoter oligonucleotide is 5Ј-GGGTCGACCCTCCTTGA-3Ј, where the underline indicates the mutated nucleotides. For comparison, the wild type oligonucleotide sequence is 5Ј-GGGTCCCGCCTCCTTGA-3Ј (Fig. 1A). The quadruple Smad-binding element (SBE 4 ) oligonucleotide was described by Jonk et al. (38). Oligonucleotides corresponding to the p21 promoter regions were synthesized, annealed, labeled with Klenow and [␣-32 P]dCTP, and 10 fmol were incubated with 200 ng of each purified protein or with aliquots of nuclear extracts from the transfected SL2 cells that were calibrated for total protein content, ␤-galactosidase activity, and specific protein abundance based on the Western blot signals. For the Smad-binding depletion experiment, purified Smad proteins were first incubated with excess (4 pmol) cold SBE 4 oligonucleotide for 10 min at 4°C, and then the Sp1 protein was added and incubated for another 10 min, and finally the labeled p21 promoter oligonucleotide was included in the reaction that proceeded for 30 min prior to 4% PAGE.
Relative bandshift intensities were quantified using the scanning densitometric software of the PhosphorImager Fujix BAS 2000. The ground condition (Sp1 alone) was set to 1 or 100% relative to which all other conditions are expressed.

Smad3 and Smad4 Can Transactivate the p21 Promoter in
Human HaCaT Keratinocytes-The proximal (Ϫ124/Ϫ42) region of the human p21 promoter is G/C-rich and contains 5 sequence motifs that resemble or match exactly the recognition sequence of the ubiquitous transcription factor Sp1 (5Ј-GGGCGG-3Ј, Fig. 1A, double underline). The constitutive activity and induction of this promoter by extracellular signals in different cell systems depends on some of the Sp1-like motifs (16,35,39). Critical for TGF-␤-mediated induction of the p21 promoter is one of these Sp1 sites (designated T␤RE in Fig. 1A) (16). Smads were found capable of transactivating the proximal p21 promoter in HepG2 cells (35). However, similar experiments performed in HaCaT cells failed to demonstrate Smadmediated transactivation of the p21 promoter (data not shown and see Refs. 11 and 40). A reason for this could be the low Smad expression levels achieved after the inefficient transient transfection of HaCaT cells with various protocols (data not shown). Adenovirus-mediated gene transfer is highly efficient (80 -95% infection rate), and relatively high levels of expression of the encoded protein can be obtained (37,41). We thus combined transient transfections of HaCaT cells with three different p21 promoter-luciferase reporters (Ϫ2,300/ϩ8 p21 Luc, Ϫ2,300/ϩ8⌬Sp1 p21 Luc, and Ϫ143/ϩ8 p21 Luc) with transient infections with Smad3-and Smad4-encoding viruses. To stimulate the TGF-␤ pathway we utilized another adenoviral vector encoding the constitutively active (CA) TGF-␤ type I receptor (also termed activin receptor-like kinase 5, ALK-5). As shown in Fig. 1B, CA-ALK-5 overexpression led to a moderate but distinct 2.2-fold activation of the Ϫ2,300/ϩ8 p21 promoter. Overexpression of either Smad3 or Smad4 by means of the adenovirus system also resulted in the same level (2.2-fold) of transactivation in the absence of activated ALK-5 receptor (Fig. 1B). In the presence of the activated receptor the effect of p21 Regulation by Smads the Smads was further augmented. Co-expression of Smad3 and Smad4 led to a higher level (3.7-fold) of transactivation, which reached levels as high as 5.5-fold after activated receptor co-expression. The transactivation potential of the adenovirally encoded Smads is specific as a control virus encoding for ␤-galactosidase resulted in transactivation levels similar to the uninfected condition. Thus, Smad proteins can transactivate the Ϫ2,300/ϩ8 p21 promoter in HaCaT cells, and their potential is significantly increased by TGF-␤ receptor activation.
To pinpoint the importance of the Sp1 G/C-rich motifs of the proximal p21 promoter, we used a deletion mutant of the Ϫ2,300/ϩ8 p21 promoter that lacks sequences between Ϫ124 and Ϫ61 of the promoter (Fig. 1C). This deletion removes four out of five of the Sp1 motifs of the proximal promoter and has been previously shown to support very low basal activity that could not be regulated by TGF-␤ or the Smads (16,35). Indeed, under the present conditions of transient transfection coupled to adenoviral infection, the activity of this promoter remained low and not significantly altered by Smad3/4 or the constitutive receptor. This reconfirms the importance of the Sp1 motifs between Ϫ124 and Ϫ61 of the proximal promoter for TGF-␤ and Smad-mediated transactivation of the promoter.
A similar analysis was performed with the proximal Ϫ143/ϩ8 p21 promoter (Fig. 1D). In this case, overexpression of the CA-ALK-5 receptor transactivated the proximal promoter 10-fold, almost to the same extent as the transactivation achieved by the overexpression of Smad3 and Smad4 proteins alone (11-fold for Smad3 and 7-fold for Smad4). Co-expression of Smad3 and Smad4 with the CA-ALK-5 receptor further increased p21 promoter transactivation by Smads, whereas co-expression of Smad3 with Smad4 and CA-ALK-5 led to even higher level of promoter activation (18-fold). It is worth noting that the CA-ALK-5-stimulated activity of the proximal Ϫ143/ϩ8 p21 promoter is reproducibly higher than the receptor-stimulated activity of the Ϫ2,300/ϩ8 p21 promoter (compare Fig. 1, B and D). This also stands true for the stimulatory effects of TGF-␤ on these promoters (35) and implies the presence of negative regulatory elements in the distal p21 promoter region (see "Discussion").
As an independent control for the specificity of Smad-dependent transcriptional activation measured by the transient coupled transfection-adenovirus infection assay, we used the well established reporter 12ϫ(CAGA) Luc (43) whose transactivation by TGF-␤ depends solely on the Smads (Fig. 1E). As expected, overexpressed Smad3 and Smad4 synergized with the CA-ALK-5 signal resulting in a robust 14-fold activation relative to the basal promoter activity.
In conclusion, by using the adenovirus system we were able to demonstrate Smad-mediated transactivation of the human p21 promoter in HaCaT keratinocytes thus firmly establishing the importance of these factors in p21 gene regulation by TGF-␤ in this and other cell types.
Smad Proteins Contribute to the Regulation of the Endogenous p21 Gene by TGF-␤ in HaCaT Cells-To evaluate further the involvement of Smad proteins in the regulation of the endogenous p21 gene by TGF-␤ in HaCaT cells, we monitored p21 protein expression levels by Western blotting using a p21-specific antibody (Fig. 2). Treatment of HaCaT cells with FIG. 1. A-E, Smad3 and Smad4 transactivate the human p21 promoter via its G/C-rich proximal sequences in HaCaT cells. A, schematic representation of the human p21 promoter region Ϫ2300/ϩ8. The region between nucleotides Ϫ124/Ϫ42, which is important for both the constitutive and the inducible activity of the promoter, is shown as a black bar and its nucleotide sequence is expanded below. The oval represents a Smad-binding element (SBE) identified in the distal part of the promoter (42,50). In the sequence, heavy underlines mark the GEMSA oligonucleotide probes (A, p21Pr(Ϫ86/Ϫ70)) and (B, p21Pr(Ϫ122/Ϫ84)) used in Fig. 3. Binding sites for the ubiquitous transcription factor Sp1 are double-underlined. T␤RE indicates the Sp1 motif shown previously to be important for the stimulation of the promoter by TGF-␤ (16). The TATA box is boxed. B-E, HaCaT cells were transiently transfected with the Ϫ2,300/ϩ8 p21 luc (B), the Ϫ2,300/ϩ8⌬Sp1 p21 luc (C), the Ϫ143/ϩ8 p21 luc (D), the 12ϫ(CAGA) luc, and control ␤-galactosidase reporter constructs and the next day were infected with adenoviruses expressing Smad3, Smad4, CA-ALK-5 or control, LacZ. Thirty six hours post-infection cell lysates were assayed for luciferase and ␤-galactosidase activities. The luciferase activity normalized over the ␤-galactosidase activity was plotted in a bar graph relative to the mock transfection control, which was arbitrarily set at 1. The data represent measurements from two independent experiments that include triplicate samples each.
10 ng/ml TGF-␤1 for 20 h resulted in the robust accumulation of p21 protein (12-fold compared with untreated control, Fig.  2A, lanes 1 and 2) as previously reported (7). The TGF-␤1 effect could be mimicked by the adenovirus-encoded constitutively active type I receptor ALK-5 in a dose-dependent manner. A selected m.o.i. of 50 of this virus resulted to a 3-fold activation ( Fig. 2A, lane 3). Overexpression of adenovirus-encoded Smad3 and Smad4 also resulted in a dose-dependent increase of endogenous p21 protein levels. m.o.i. of 50 for each Smad-virus resulted in a low but reproducible 2-fold increase of p21 protein levels ( Fig. 2A, lane 6). Combination of overexpressed Smad3 and Smad4 together with CA-ALK-5 or TGF-␤1 treatment showed further increase of the p21 protein levels, which correspond to 5-and 14-fold, respectively, under the infection conditions shown in Fig. 2A (lanes 4 and 5).
To enhance the evidence that the Smad signaling pathway is involved in p21 protein accumulation in response to TGF-␤1 in HaCaT cells, we also made use of dominant negative and inhibitory Smad proteins (Fig. 2, B-D). A Smad3 carboxyl-terminal truncated mutant (DN-Smad3), the equivalent truncation of Smad4 (DN-Smad4), and the inhibitory Smad7 all resulted in a dose-dependent decrease of p21 accumulation in response to TGF-␤1. The Smad3 and Smad4 mutants have been previously shown to interfere with TGF-␤ signaling in a dominant negative fashion with respect to specific target gene responses (44,45), and we have shown that the DN-Smad4 mutant interferes with p21 promoter regulation in HepG2 cells (35). Similarly, the Smad7 inhibitor has also been previously reported to inhibit the induction of the p21 promoter-luciferase reporter by TGF-␤1 in HaCaT cells (46). The combined p21 Western blotting results strongly support the involvement of the Smad signaling pathway in endogenous cell p21 regulation by TGF-␤1.
Smad Proteins Do Not Associate with the Proximal p21 Promoter DNA-Smad3 contains DNA binding activity with low affinity toward 5Ј-TCTGAGAC-3Ј (termed the Smad-binding element (SBE)), whereas Smad4 toward both the SBE and G/C-rich motifs (42,43,(47)(48)(49). The SBE is absent from the proximal p21 promoter (Fig. 1A) and exists in an upstream distal segment of the promoter (50) with no apparent functional significance (16,35). On the other hand the p21 proximal promoter is G/C-rich. We thus tested the hypothesis of direct DNA binding of Smads to the proximal p21 promoter, which confers the inducibility of this gene to TGF-␤ and Smads (16,35). Since mammalian cells express high levels of Sp1 protein, it has been rather difficult to identify TGF-␤-specific nucleoprotein complexes by using GEMSA on the p21 or other Sp1containing promoter sequences (16,30). For this reason we used the well established Drosophila Schneider's SL2 cell line that lacks endogenous Sp1 activity (51). Fig. 3A shows representative GEMSA data produced from nuclear extracts of transiently transfected SL2 cells with the indicated combinations of Sp1 and Smad proteins and a radioactively labeled oligonucleotide corresponding to the TGF-␤-responsive element (T␤RE) (p21Pr(Ϫ86/Ϫ70)) of the proximal p21 promoter. Similar results were obtained when the upstream oligonucleotide containing two Sp1 motifs (p21Pr(Ϫ122/Ϫ84)) as shown in Fig. 1A was tested (data not shown). Our results showed the following. (a) Under the conditions used, endogenous SL2 nuclear proteins do not recognize the p21 oligonucleotides (Fig. 3A, lane 1). (b) Overexpression of Smad3 or Smad4 or both gives the same negative result as mock-transfected cells (Fig. 3A, lanes 2-4).
(c) Overexpression of Sp1 results in a specific nucleoprotein complex as expected (Fig. 3A, lane 8). (d) Overexpression of Smad3 together with Sp1 or Smad4 together with Sp1 results in a small 2-and 2.5-fold enhancement of the Sp1 nucleoprotein complex, respectively (Fig. 3A, lanes 5 and 6). (e) Overexpression of both Smad3 and Smad4 together with Sp1 results in a significant (5-fold) enhancement of the Sp1 nucleoprotein complex on both p21 probes (Fig. 3A, lane 7 and data not shown). In order to prove the specificity of the obtained Sp1 nucleoprotein complex in the SL2 nuclear extracts, we used competition experiments with excess amounts of cold oligonucleotides corresponding to the wild type p21 proximal promoter (Fig. 3B, lanes  3-5), the same oligonucleotides harboring point mutations in the Sp1-like elements (Fig. 3B, lanes 6 -8) or a consensus Sp1 sequence (Fig. 3B, lane 9). These experiments confirmed that the nucleoprotein complexes formed with p21 promoter DNA probes are specific and contain only Sp1. In addition, unrelated, non-Sp1 sequence containing oligonucleotides including the SBE 4 did not show any competition (data not shown). It must be noted that co-expression of Smads with Sp1 results only in relative enhancement of the nucleoprotein complex without any additional higher size complexes. Finally, the enhanced binding of Sp1 to the p21 proximal promoter after co-expression of Smad proteins in the SL2 cells could not result from nonspecific effects of differential protein expression in the transfected cells, as the levels of nuclear Sp1 and Smad proteins appeared rather comparable (Fig. 3C).
Thus the combined data demonstrate that Smad proteins do not associate with the proximal p21 sequences tested and coexpression of Smad3, Smad4, and Sp1 results in an enhanced nucleoprotein complex that retains the binding characteristics of Sp1. Smad Proteins Induce an Increase in Sp1 Affinity for the DNA Sequences of the p21 Proximal Promoter-To examine thoroughly whether Smad proteins alter the Sp1 affinity for DNA, we relied on purified protein factors and in vitro GEMSA. By using purified unphosphorylated or phosphorylated (by CA-ALK-5 receptor) Smad3 and unphosphorylated Smad4 from a baculovirus system (36) and purified Sp1 from bacteria, we confirmed that Smad3, phospho-Smad3, Smad4 alone, or in combinations could not exhibit stable complexes with the p21 promoter DNA (Fig. 3D, lanes 2-4 and 13-15). The same proteins strongly and stably associated with the SBE DNA (data not shown and see Ref. 36). Purified Sp1 formed a strong complex with both p21 promoter probes (12); however, for the purpose of the experiments presented in this figure, a low concentration of Sp1 was used that reproducibly gave a rather weak binding (Fig. 3D, lane 9). Interestingly co-incubation of phospho-Smad3, Smad4, and Sp1 with the p21 probe resulted in a strong nucleoprotein complex of higher molecular mass (labeled Sp1Ј) when compared with the complex obtained by Sp1 alone (Fig. 3D, lane 5). This effect was only obtained when phospho-Smad3 was used, as unphosphorylated Smad3 in combination with Smad4 and Sp1 resulted in the same weak bandshift as Sp1 alone (compare Fig. 3D, lanes 9 and 16). Western blot analysis of the two Smad3 preparations using anti-phosphoserine-specific antibodies confirmed that only phospho-Smad3 contained phosphorylated serines (Fig. 3F). The enhanced Sp1Ј bandshift was also followed by a lower and weaker trailing bandshift. Competition experiments with excess cold oligonucleotides demonstrated that the higher mass Sp1Ј complex is specific for the p21 probe used, whereas the trailing bandshift possibly represents an experimental artifact (Fig. 3D, lanes 6 -8). In addition, these competition assays proved that the Sp1Ј bandshift represented a high affinity complex of Sp1 to the p21 promoter DNA (Fig. 3D, lanes 5-12).  Fig. 3D for the p21Pr(Ϫ86/Ϫ70) probe were also obtained for the upstream p21Pr(Ϫ122/Ϫ84) probe (data not shown). Identical competition profiles were observed by the consensus Sp1 oligonucleotide, whereas the mutant p21Pr(Ϫ86/Ϫ70) oligonucleotide failed to compete like in the SL2 nuclear extract GEMSAs (Fig. 3B), and the unrelated SBE 4 oligonucleotide not only failed to compete but exhibited enhancement of Sp1 binding to the p21 promoter (data not shown and see below). These experiments suggest that Smad proteins increase the affinity of Sp1 for its cognate G/C-rich-binding motif.
To understand further the nature of the induced enhancement of the Sp1 affinity for DNA by the Smads, we performed a preincubation-competition experiment using a quadruple concatamer of the consensus SBE (SBE 4 ) (Fig. 3E). In this experiment the phospho-Smad3 and Smad4 were preincubated with excess (400-fold) cold SBE 4 oligonucleotide to saturate the intrinsic binding of the Smad proteins for DNA, followed by addition of the Sp1 and the labeled p21 promoter probe (Fig. 3E, lane 2). The resulting complex was compared with that obtained when all proteins were mixed together in the absence of excess SBE 4 oligonucleotide (lane 1) or in the absence of Sp1 (lane 3). Surprisingly, this method of treatment   FIG. 4. A and B, Smad3 and Smad4 mutant proteins that cannot associate with consensus Smad-binding elements are fully capable of transactivating the proximal p21 promoter. Transient transfection experiments of HepG2 cells with the indicated wild type (wt) and double mutant (dm) Smad and CA-ALK-5 expression constructs together with the Ϫ143/ϩ8 p21 luciferase (A) or the Smad-sensitive 12ϫ(CAGA) luciferase and control ␤-galactosidase reporter constructs. Forty hours post-transfection cell lysates were assayed for luciferase and ␤-galactosidase activities. The luciferase activity normalized over the ␤-galactosidase activity is plotted in a bar graph relative to the mock transfection control, which is arbitrarily set at 1. The data represent measurements from three independent experiments that included triplicate samples each. The analysis in B serves as control experiment for the results of A. The scale of the relative luciferase activity in B is broken to fit in the figure. p21 Regulation by Smads resulted in a further 3-fold enhancement of the Sp1Ј bandshift (compare lanes 1 and 2) which suggests that when the Smad proteins are provided with their DNA substrate they can still enhance the affinity of Sp1 for the p21 promoter DNA. These findings suggest that the Smad3-Smad4 complex bound or unbound to DNA leads to a significant increase in the affinity of Sp1 for its DNA-binding sequences.
Smad Binding to DNA Is Not Essential for the Activation of the p21 Proximal Promoter-To examine whether Smad binding to DNA is required for the activation of the p21 promoter, we made use of point mutants in the conserved ␤-hairpin of the MH1 domain of Smad3 and Smad4 which is the DNA-binding domain of Smads (47). We used an R74K/K81R double mutant of Smad3 (Smad3dm) and the corresponding R81K/K88R double mutant of Smad4 (Smad4dm). These mutations in the ␤-hairpin of the MH1 domain, as predicted from the crystal structure, completely abolish binding of the Smads to the SBE. 2 The prediction in our experiment was that these mutants should have wild type transactivation activity on the p21 promoter if DNA binding of Smads to the p21 proximal promoter is not important. Indeed, the experiments shown in Fig. 4A, performed in HepG2 cells, demonstrate that both Smad3 and Smad4 mutants can transactivate the proximal p21 promoter as efficiently as the wild type proteins. The constitutive effect of overexpression of the Smad proteins on the proximal p21 promoter was retained by the mutants, as was the inducible effect stimulated by the co-expression of CA-ALK-5.
As a control experiment we tested the transactivation of an artificial promoter that contains 12 copies of the SBE (CAGA element) in front of the adenovirus major late promoter and that is very sensitive to activation by TGF-␤ in a Smad-dependent way (43). As previously reported, the overexpression of Smad3 with Smad4 significantly increased the basal activity of the promoter, and activation by means of CA-ALK-5 further augmented this response (Fig. 4B). In contrast, Smad3dm failed to increase the basal promoter levels when co-expressed with wild type Smad4, suggesting that Smad3 is the primary DNA-binding factor on this promoter. Smad4dm gave wild type levels of constitutive activation. Importantly, the receptorinduced superactivation showed a 2-3-fold decrease by both mutants. Finally, co-expression of both mutants resulted in unaffected basal promoter levels and reduced receptor-superactivated levels. The negative effects of the mutants on basal level and receptor-induced promoter activation seen with the 12ϫ(CAGA) reporter stand in contrast to the p21 proximal promoter experiments, demonstrating that the transactivation of the p21 promoter is independent from mutations that disrupt Smad-SBE DNA binding and interfere with optimal trans-activation of SBE-dependent promoters.
Smad2 and Smad3 Co-immunoprecipitate with Sp1 in Transfected COS-7 Cells-Our previous work demonstrated a strong functional interaction between Smads and Sp1 (35). To test whether Smads and Sp1 physically interact, we applied several complementary experimental approaches. Fig. 5 shows the results of co-immunoprecipitation experiments performed with protein extracts from COS-7 cells transiently transfected with epitope-tagged Smad and GAL4-Sp1 fusion proteins. As a positive control we used the receptor activation-dependent interaction of Smad3 with Smad4 (Fig. 5A, lanes 2 and 3). Smad3 was found to co-precipitate with Sp1 at measurable levels even in the absence of stimulation of the signaling pathway (Fig. 5B,  lane 2). However, co-expression of the CA-ALK-5 resulted in significant enhancement of co-precipitating Smad3 (Fig. 5B,  lane 3). Essentially the same results were obtained for the co-precipitation of Smad2 with Sp1 (Fig. 5C, lanes 2 and 3). In contrast, we failed to detect co-precipitation of Smad4 with GAL4-Sp1, suggesting that the interaction of these two proteins might be either indirect or rather weak compared with the Smad3-Sp1 interaction (Fig. 5C, lanes 4 and 5).
The combined data presented in Fig. 5 demonstrate that Smad2 and Smad3 can co-precipitate with Sp1, whereas Smad4 is unable to do so under the conditions used.
Smad2, Smad3, and Smad4 Directly Associate with Sp1 via Their Conserved MH1 Domain-In order to map the domains of Smad and Sp1 proteins involved in their physical interaction, we used an in vitro interaction assay with GST-Sp1 and wild type or mutated Smad proteins synthesized in vitro. Fig. 6A shows schematically various Smad and GST-Sp1 constructs used in these experiments. The relative expression levels of affinity-purified Sp1, negative control GST, and in vitro synthesized Smad proteins are shown in Fig. 6B. When the in vitro synthesized Smad2, Smad3, and Smad4 were allowed to interact with GST-Sp1, all three Smad proteins were found capable for the direct interaction (Fig. 6C, lanes 4 -6). However, among the three, Smad3 showed the strongest potential for interaction, whereas Smad2 and Smad4 interacted rather weakly (6 and 10% relative to Smad3 which is arbitrarily set to 100%, Fig. 6A). The interaction was specific as the GST moiety of the fusion protein when tested alone failed to support productive interaction (Fig. 6C, lanes 1-3). It must be noted here that these interactions are constitutive and do not depend on an activated TGF-␤ signaling pathway. In order to map the domain in Smad3 that is responsible for the interaction with Sp1, we used three different deletion mutants of Smad3 (Fig. 6A). Smad3-(1-248) contains the MH1 and linker sequences and showed detectable, albeit weak, (17.8%) interaction with Sp1   lysate (input, B) and then allowed to interact with glutathione beads carrying GST alone (lanes 7-9) or GST-Sp1 (lanes 10 -12), washed thoroughly, resolved by 12% SDS-PAGE, and detected by compared with the wild type Smad3 (Fig. 6C, lane 10). Smad3-(1-122) contains the MH1 domain only, which is also truncated at its carboxyl terminus and exhibited similar weak interaction (23.9%) as the previous MH1-linker mutant (Fig. 6C, lane 11). Finally, Smad3-(122-424), which contains very few amino acids from the MH1 domain, the linker, and the MH2 domains, failed to support detectable interactions (Ͻ1.5%) with Sp1 (Fig.  6C, lane 12). The same Smad3 mutants failed to show nonspecific retention to the GST affinity columns (Fig. 6C, lanes 7-9).
We conclude from the in vitro experiments that all three Smad proteins of the TGF-␤ signaling pathway, Smad2, Smad3 and Smad4, are capable of direct physical interaction with transcription factor Sp1, although Smad3 shows a more pronounced interaction (10 -16-fold, based on densitometric analysis) in the in vitro pull-down assays. The MH1 domain of Smad3 is the primary determinant for this interaction, although the MH2 domain is necessary for fully productive interaction, as MH2 truncation significantly decreased (by 4 -5fold) the interaction potential of the residual Smad3 domains.
Although the previous experiments provide strong evidence for the physical association of Smad proteins and in particular Smad3 with Sp1, in these assays chimeric Sp1 proteins were always used (GAL4-Sp1 and GST-Sp1). To obtain evidence that the natural Sp1 molecule also interacts with Smad proteins, we used a series of GST-Smad fusion proteins that included fulllength Smad3 and Smad4 as well as deletion mutants of these two proteins (Fig. 6D). Total detergent extracts from HaCaT cells were passed through the GST-Smad affinity columns (Fig. 6E) and washed, and the proteins bound to the columns were analyzed by Western blotting using an Sp1-specific polyclonal antibody (Fig. 6F). Endogenous Sp1 was readily detectable in the total HaCaT cell extract (Fig. 6F, lane 9). Sp1 was found to interact with GST fusions of full-length Smad4 (lane 1) and Smad3 (lane 3) but not GST alone (lane 8). The interaction with Smad3 was more efficient than with Smad4, which is in agreement with the in vitro experiments of Fig. 6A-C. Experiments with deletion mutants of the two Smad proteins also corroborated the previous results as they showed that the MH1 plus linker domains of Smad4 were capable of sustaining interaction with Sp1 but less efficiently than full-length protein (Fig. 6F, lane 2). In addition, the MH1 domain of Smad3 was the primary determinant for the interaction with Sp1 (Fig. 6F,  lane 6); however, the presence of the linker (lane 5) and the MH2 domain (lane 3) enhances the interaction considerably. Furthermore, the isolated linker plus MH2 (⌬MH1, lane 4), MH2 (lane 7), and linker (not shown) domains did not support any detectable interaction, excluding the possibility that these domains could provide primary specificity to the intermolecular associations studied.
Therefore, these experiments are in full agreement with those shown in Figs. 5 and 6, A-C, and establish that the natural Sp1 molecule is capable of interacting with Smad proteins via their MH1 domains.
The Glutamine-rich Region of the Sp1 Transactivation Domain Can Sustain Functional Synergism with Smad Proteins-Since a first map of the interaction domain on Smad proteins was established, we were interested in defining the Sp1 sequences that were participating in the Smad-Sp1 interaction. For this reason we relied on the in vitro interaction assay by using a panel of GST-Sp1 deletion mutants (Fig. 6, G  and H) and in vitro synthesized full-length Smad3 protein since this showed the best potential in the interaction assays (Fig. 6,  C and F). The Sp1 mutants included deletions of the conserved and duplicated subdomain A of the major transactivation domain of this protein (23), the second subdomain B of the major transactivation domain together with domain C, which also confers transactivation potential to Sp1 and a carboxyl-terminal deletion of domain D (Fig. 6G). The three mutants and full-length Sp1 were produced as GST fusions (Fig. 6H). Smad3 exhibited direct interaction with full-length Sp1 (Fig. 6I, lane 3) as described above, which was not seen with GST alone (lane 2). Smad3 binding to Sp1 was decreased when subdomain A was deleted (lane 4). However, Smad3 binding was not affected at all by deletion of the subdomains B plus C (lane 5) and was mildly decreased by deletion of domain D (lane 6). Since none of the Sp1 mutants showed complete loss of interaction potential with Smad3, this suggested that the conserved and duplicated glutamine and serine/threonine-rich sequences of subdomains A and B and/or the DNA-binding domain of Sp1 (domain Zn 2ϩ in Fig. 6G), which are included in all the Sp1 mutants, might be involved in the interaction. Alternatively, it is possible that Sp1 contains multiple sequence motifs that contribute to the interaction with Smad proteins.
To characterize in more detail sequences within the conserved domains A and B that might contribute to the functional synergism with Smads, we made use of another panel of deletion mutants of Sp1, which included shorter truncations of the major transactivation domain fused to GAL4 DNA-binding domain (GAL4-DBD, Fig. 7A). Fig. 7B shows that co-expression of Smad3 and Smad4 together with full-length Sp1 fused to GAL4 resulted in almost 20-fold activation of a reporter containing five concatamerized GAL4-binding sites in front of a minimal thymidine kinase promoter, relative to the constitutive level of GAL4-Sp1 alone. Mutants Sp1 A ϩ B, Sp1 B, and Sp1 Bc all exhibited significant transactivation (34 -37- contrast, mutant Sp1 Bn that retains only the serine/threonine-rich region of domain B (amino acids 263-424) was not responsive to the transactivation exerted by the Smad complex (1.2-fold). These data suggest that the glutamine-rich subdomain of Sp1 might be involved in the Smad-Sp1 interaction. Furthermore, a GAL4-Sp1 Bc fusion protein, which contains a triple glutamine to alanine amino acid substitution (GAL4-Sp1 Bc (Q3A), Fig. 7A) showed decreased transactivation (7.5-fold) in the presence of Smad3/Smad4, and the activity of this mutant alone decreased considerably relative to that of GAL4-Sp1 (Fig. 7B). This result enhances the hypothesis that the glutamine residues of this domain are important for the cooperation with Smad proteins, and of course, since not all glutamines were mutated in the triple point mutant, the residual transactivation potential (7.5-fold) can be attributed to the remaining glutamine-rich protein surfaces.
Finally, we tested the functional interaction of the glutamine-rich domain of Sp1 with the three Smad proteins of the TGF-␤ signaling pathway, Smad2, Smad3, and Smad4 (Fig. 7C). The results showed that Smad3 exhibits the strongest cooperative activation (8-fold) together with Sp1-Bc, which can be enhanced further (13-fold) by co-expression of Smad4. Smad2 and Smad4 alone did not show any appreciable level of transactivation (1.1-1.4-fold) in agreement with the low or undetectable constitutive interactions described above. However, the Smad2 plus Smad4 combination resulted in measurable but small transactivation (2.9-fold), which is consistent also with the physical interaction data, since this combination partially mimics the activation of Smad2 by ligand.
In conclusion, these experiments suggest that the glutaminerich subdomain of the transactivation domain of Sp1 plays important roles in the functional synergism between Sp1 and Smad proteins and further strengthen the finding that Smad3 under all experimental conditions tested exhibits the best physical and functional cooperativity with transcription factor Sp1.

DISCUSSION
The experiments of Fig. 1 demonstrate the positive regulatory effect of Smad proteins on the activity of the p21 promoter in human keratinocytes HaCaT. This result is of particular significance as this cell line has been extensively analyzed for the mechanism of growth inhibition by TGF-␤ (4,11,52) and exhibits a rather dramatic induction of its endogenous p21 gene in response to TGF-␤ (7 and Fig. 2). The Smad effect also depends on the integrity of the G/C-rich, Sp1-occupied proximal promoter (Fig. 1C). It must be noted that the adenovirusmediated expression of Smad3 and Smad4 proteins positively up-regulates the Ϫ2,300/ϩ8 and the Ϫ143/ϩ8 p21 promoters, to a lower extent than TGF-␤ (16). Thus, additional regulatory factors may be required for the maximal activation of the p21 promoter by TGF-␤ (see below). In addition, the promoter analysis of Fig. 1 supports our previous analysis in HepG2 cells where a distal inhibitory and a proximal stimulatory promoter segment were functionally defined (35). Removal of the distal sequences results in significant increase of the responsiveness of this promoter to the TGF-␤ signal and Smad proteins (35). For this reason the transcriptional activity of the proximal Ϫ143/ϩ8 p21 promoter in response to TGF-␤ is relatively higher than the activity of the Ϫ2,300/ϩ8 promoter ( Fig. 1 and Ref. 35).
The p21 promoter studies in HaCaT cells (Fig. 1) and in HepG2 cells (35) are in strong agreement with the HaCaT experiments of Fig. 2 in which exogenous Smad3 and Smad4 were found to potentiate the response of the endogenous p21 gene to TGF-␤1. Overexpression of Smad3 and Smad4 by means of adenovirus infection could significantly enhance endogenous p21 accumulation, an effect that could be further augmented by co-expression of the constitutively active type I receptor for TGF-␤ (CA-ALK5) or TGF-␤1. This implies that activation of the Smads by receptors leads to more efficient p21 gene activation. These effects on endogenous p21 protein accumulation are dose-dependent for all tested activators, i.e. the ligand TGF-␤1, the constitutively active type I receptor, and the Smads, and thus, conditions where the cell can tolerate excessive amounts of p21 accumulation can be obtained ( Fig. 2A and data not shown). In addition, carboxyl-terminally truncated dominant negative mutants of Smad3 and Smad4 both inhibited p21 accumulation in response to FIG. 7. A-C, the glutamine-rich regions of the transactivation domain of Sp1 are important for the functional cooperation with Smad proteins. A, diagrammatic representation of the GAL4-Sp1 fusion protein constructs used in the transactivation assays of B and C. The same drawing conventions as in Fig. 6A are used. For mutants Sp1 Bc and its derivative Sp1 Bc (Q3A) the respective wild type and triple glutamine to alanine substitutions in the amino acid sequence of the pertinent region are shown with the substituted amino acids boxed. The relative sizes of the different proteins are not to scale. B, the glutamine-rich region of the transactivation domain B of Sp1 is important for functional cooperation with Smad proteins. HepG2 cells were transiently transfected with the indicated Smad expression constructs and various GAL4-Sp1 deletion mutant fusions (described in detail in A), together with the 5 ϫGAL4 CAT and control ␤-galactosidase reporter constructs. Forty hours post-transfection cell lysates were assayed for CAT and ␤-galactosidase activities. The CAT activity normalized over the ␤-galactosidase activity is plotted in a bar graph relative to the mock transfection control, which is arbitrarily set at 1. The data represent measurements from two independent experiments. Relative fold differences between the plus and minus Smad3/Smad4 groups of data are shown on top of each pair of bars. C, the glutamine-rich region of Sp1 exhibits stronger functional cooperation with Smad3. HepG2 cells were transiently transfected, analyzed, and depicted as in B using the mutant GAL4-Sp1 Bc that contains primarily the glutamine-rich region of repeat B (see A) and the indicated combinations of Smad proteins. TGF-␤1 (Fig. 2, B and C), in a dose-dependent manner. Since these mutants are known to interfere with Smad activation by receptors, oligomerization, nuclear translocation, and cooperation with transcription factors (34,44,45), the Smad signaling pathway must be required for endogenous p21 gene induction by TGF-␤. The same results were obtained with increasing doses of the inhibitory Smad7. The combined data of Figs. 1 and 2 strongly confirm that activation of Smads by the type I receptor is a critical step in endogenous p21 responsiveness to TGF-␤1.
Smad proteins are known to associate directly with DNA elements containing the TCGTAGAC or G/C-rich sequences, although with relatively low affinity (36,42,43,(47)(48)(49). The p21 proximal promoter lacks any obvious SBEs but contains a G/C-rich region between nucleotides Ϫ124 and Ϫ42 (Fig. 1A). GEMSA experiments (Fig. 3) showed that Smad3 or Smad4 does not bind to the G/C-rich region of the p21 promoter. The transactivation experiments using Smad3 and Smad4 proteins containing point mutations in their DNA-binding domains, which cannot recognize the SBE (Fig. 4), confirmed that p21 promoter regulation by Smads does not require the DNA-binding function of Smads. Control experiments using the multimerized SBE promoter confirmed the defective nature of the mutant Smads in HepG2 cells; however, their absolute negative effects could not be estimated since HepG2 cells contain endogenous Smad3 and Smad4. The mutant Smad3 and Smad4 proteins failed to transactivate the same SBE promoter in cells that lack the genes for Smad3 or Smad4. 2 Finally, it is worth noting that despite the lack of SBE sequences in the proximal p21 promoter, such elements have been described in the distal segment of the promoter (42,50). However, previous deletion analyses have shown that both TGF-␤ and Smad-dependent activation of the p21 promoter do not require this distal SBE (16,35). The exact role of the distal SBE on the basal and inducible activity of the p21 promoter remains to be elucidated.
The GEMSA analyses illustrated in Fig. 3 showed that Smad3 and Smad4 proteins enhanced the formation of a nucleoprotein complex between Sp1 and p21 promoter oligonucleotides. This observation could support a mechanism of cooperativity between Smads and Sp1 in p21 promoter transactivation. On the other hand, binding of Smads to SBE sequences resulted in a stronger cooperativity with Sp1 (Fig. 3E). Such a mechanism would imply that TGF-␤-responsive promoters that contain both SBE sequences and G/C-rich Sp1-binding motifs would provide more optimal substrates for a cooperative function between Smads and Sp1 in transcriptional regulation. Such examples might be the p15 Ink4B , the Smad7, the TGF-␤1, and the TGF-␤ type I and type II receptor promoters (55)(56)(57).
In contrast to previously published examples of nucleoprotein complex formation between Smads and other transcription factors (53,58), we observed the presence of a distinct nucleoprotein complex with only slightly slower mobility than the Sp1-DNA complex (Fig. 3D). This implies that the Sp1-Smad nucleoprotein complex cannot withstand GEMSA conditions or that alternatively high affinity Sp1-DNA complex formation induced by Smad proteins may rapidly lead to Smad dissociation from the complex. A qualitatively similar result has been observed in studies of interaction and cooperation between the co-activator p300 and Sp1 (21). Thus, Smads could induce the following: (a) oligomerization of Sp1, which results in higher affinity binding to the p21 promoter; (b) recruitment of additional cooperating factors (such as p300 or c-Jun) that could possibly stabilize or enhance the transcriptional activity of Sp1; and (c) modulation of the phosphorylation or acetylation status of Sp1 with concomitant effects in DNA binding and transac-tivation potencies. Recent reports indicate that the mitogenactivated protein kinase (MAPK) pathway is also activated by TGF-␤ and contributes positively to the regulation of the p21 gene (59 -61). Thus, it is of interest to examine whether MAPKs directly modulate the transcriptional activity of Sp1, which is phosphorylated in a cell cycle-specific manner (20). Alternatively, transcription factor targets which themselves are activated by the MAPK cascade and which are known to bind to the p21 promoter sequences, such as the Ets-like factor E1AF (62) might also cooperate with the Sp1-promoter complex.
The obvious corollary of the above results has been that Smads may physically interact with Sp1. This hypothesis was tested by co-immunoprecipitation and GST pull-down analyses (Figs. 5 and 6), which led to the conclusion that Smad2, Smad3, and Smad4 proteins all can directly interact with Sp1. Smad3 showed the strongest constitutive association with Sp1 (Fig. 6,  C and F). This result, combined with the strong transactivation potential of this protein on various TGF-␤-inducible promoters, can explain a series of previous data attesting to a dominant and ligand-independent function of Smad3 in activating Sp1dependent transcriptional events (35). However, activation of the TGF-␤ signaling pathway by means of a constitutively active type I receptor (ALK-5) increased the levels of Smad3 species that co-immunoprecipitated with Sp1 in transiently transfected COS-7 cells (Fig. 5B). In contrast, the constitutive association of Smad2 and Smad4 with Sp1 is much weaker (Fig. 5B, 6C, and 6F). Type I receptor activation leads to much stronger enhancement of Smad2 association with Sp1 when compared with the enhancement seen for Smad3 (Fig. 5, B versus C). Although the constitutive association of Smad4 with Sp1 was readily detectable with two independent techniques (Fig. 6), the co-immunoprecipitation assay failed to measure constitutive or ligand-dependent association (Fig. 5C). Since the constitutive Smad-Sp1 interactions were detected using in vitro synthesized Smad proteins and bacterially purified Sp1 (Fig. 6C), these interactions must be direct and do not require any additional intermediates. However, this result does not exclude the possibility that additional factors participate in the Smad-Sp1 nuclear complex in vivo.
The same set of experiments resulted in mapping the Smad interaction domain with Sp1 as the amino-terminal conserved MH1 domain (Fig. 6, C and F). Interestingly, although the MH1 domain is the primary determinant for interactions with Sp1, the linker and MH2 domains contribute positively to the interaction in a progressive manner, making the full-length protein more capable of associating with Sp1 (Fig. 6, C and F). The MH1 domain of Smad proteins is known to associate specifically with several other transcription factors, such as Jun family members, ATF-2, and vitamin D receptor (34). In addition, the MH1 domain contains the DNA-binding domain of the Smad proteins (47). In our efforts to finely map the interaction domain in Smad3, we have collected preliminary data suggesting that sequences proximal to the Smad3 DNA-binding ␤-hairpin domain may be responsible for the specific interaction with Sp1 (data not shown). Thus, the structural model proposed by Shi et al. (47) for the cooperation between Smad3 and Jun family members (leucine zipper proteins) might also apply for transcription factor Sp1 (a prototype for zinc finger proteins).
On the other hand, one possible domain of Sp1 that could confer specificity to the Sp1-Smad cooperativity maps to the conserved and duplicated glutamine-rich region of the Sp1 transactivation domains A and B (Fig. 7). This hypothesis is in agreement with previous results showing the importance of the glutamine-rich sequences in TGF-␤1-mediated activation of GAL4-Sp1 fusion proteins (30). However, additional sequences in Sp1 might play roles in the interaction with Smad proteins (Fig. 6I), a possibility that deserves more detailed analysis. Recently, we reported on the direct association of transcription factors c-Jun and Sp1 (12). Similar to Smad proteins, the Sp1 glutamine-rich segment exhibits functional cooperativity with c-Jun. On the other hand, c-Jun and Smad proteins interact directly via the MH1 domain of the latter (53,54). Thus, one can envision complex intermolecular interactions between these three classes of transcription factors, whereby complexes of all three factors might simultaneously occur as discussed in our previous work (12). The importance of such higher order complexes in the regulation of the p21 promoter by TGF-␤ requires future investigation.
In conclusion, the present data provide strong evidence for the involvement of the Smad signaling pathway in both p21 promoter and endogenous p21 gene regulation by TGF-␤ in HaCaT cells. In addition, we demonstrate the physical association of Smad proteins with transcription factor Sp1. One of the consequences of such interactions is the apparent enhancement of the affinity of Sp1 for its cognate G/C-rich DNA element. Finally, the protein-protein interactions between Smads and Sp1 can account for the synergistic regulation of the p21 promoter by these factors in response to TGF-␤ only partially, as additional nuclear cofactors are most probably participating. Thus, the complexity of the mechanism by which the TGF-␤ signal is integrated on the p21 promoter is gradually uncovered and deserves further attention.