Identification of a gadd45β 3′ Enhancer That Mediates SMAD3- and SMAD4-dependent Transcriptional Induction by Transforming Growth Factor β*

GADD45β regulates cell growth, differentiation, and cell death following cellular exposure to diverse stimuli, including DNA damage and transforming growth factor-β (TGFβ). We examined how cells transduce the TGFβ signal from the cell surface to the gadd45β genomic locus and describe how GADD45β contributes to TGFβ biology. Following an alignment of gadd45β genomic sequences from multiple organisms, we discovered a novel TGFβ-responsive enhancer encompassing the third intron of the gadd45β gene. Using three different experimental approaches, we found that SMAD3 and SMAD4, but not SMAD2, mediate transcription from this enhancer. Three lines of evidence support our conclusions. First, overexpression of SMAD3 and SMAD4 activated the transcriptional activity from this enhancer. Second, silencing of SMAD protein levels using short interfering RNAs revealed that TGFβ-induced activation of the endogenous gadd45β gene required SMAD3 and SMAD4 but not SMAD2. In contrast, we found that the regulation of plasminogen activator inhibitor type I depended upon all three SMAD proteins. Last, SMAD3 and SMAD4 reconstitution in SMAD-deficient cancer cells restored TGFβ induction of gadd45β. Finally, we assessed the function of GADD45β within the TGFβ response and found that GADD45β-deficient cells arrested in G2 following TGFβ treatment. These data support a role for SMAD3 and SMAD4 in activating gadd45β through its third intron to facilitate G2 progression following TGFβ treatment.

damental roles in development and adult tissue homeostasis. The epithelial response to members of this family is highly varied and includes such diverse cellular processes as proliferation, movement, differentiation, and apoptosis. Indeed, cells harboring mutations within the signal transduction proteins or the TGF␤ target genes either fail to respond or respond inappropriately to the TGF␤ signal, often leading to developmental problems, oncogenesis, fibrotic disease, metastasis, and autoimmune disorders. Greater understanding of how cells interpret the TGF␤ signal will facilitate the prevention, detection, and treatment of various human diseases.
The central elements of TGF␤ signal transduction are now known (1,2). TGF␤ activates the serine/threonine kinase activity of a multimeric receptor complex. Activation of this complex initiates a cascade of intracellular events that culminate in altered gene expression. The SMAD proteins form the foundation of this signaling network, since they are the only proteins directly phosphorylated by the receptor complex. However, these transcription factors are by no means sufficient to impart a TGF␤ response. To specifically target a gene for transcriptional regulation, the SMADs require assistance by accessory factors. Consequently, the presence and activity of these accessory factors is are important to the TGF␤ transcriptional program as are the SMAD proteins. By designing the system in such a way, cell-specific responses to TGF␤ can be achieved. Further, the logic of the TGF␤ signaling network explains how the cell integrates multiple signals to generate highly specific phenotypic responses.
In an attempt to better understand how TGF␤ regulates gene transcription and how those gene products contribute to TGF␤ biology, we have partially defined the TGF␤ transcriptional profile in normal human mammary epithelial cells (HMEC). cDNA microarray expression analysis of TGF␤treated HMEC revealed a set of genes involved in cellular proliferation, differentiation, and apoptosis. One of these genes, gadd45␤/hMyD118, is regulated by TGF␤ in multiple cell types, thus suggesting that this gene is of central importance to the TGF␤ response. GADD45␤ and two similar small acidic nuclear proteins, GADD45␣ and GADD45␥, make up the GADD45 family (3). All three proteins regulate diverse cellular mechanisms including cell growth, DNA repair, differentiation, and apoptosis, four phenotypes that are also controlled by TGF␤ signaling. Aside from sequence similarity, these genes share transcriptional regulation by DNA damage insult and growth factors. gadd45␤ is, however, the only member of this family regulated by TGF␤ (4,5). gadd45␤ was first discovered as a transcript rapidly induced by either TGF␤ treatment or the onset of terminal differentiation in M1 murine myeloid cells (6,7). Subsequent studies employing antisense-mediated silencing established * This work was supported by a predoctoral fellowship from the Pharmaceutical Research and Manufacturers of America Foundation (to M. B. M.), a grant from the Huntsman Cancer Foundation, a grant from the Willard L. Eccles Foundation, and University of Utah Core Facility Technical Support Grant CA42104. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. GADD45␤ as an important regulator of the G 2 /M checkpoint following genotoxic stress (8) and apoptosis during M1 myeloid cell terminal differentiation (6,9). Human GADD45␤, which was first identified in a complex containing the p38-activating kinase MTK1 (MEKK4), is now a well established regulator of p38 activity and consequently p38-regulated biology (5,10,11). TGF␤ activates p38 kinase activity and induces apoptosis in normal murine hepatocytes, but not in hepatocytes derived from gadd45␤ knockout mice (11). An initial characterization of the molecular mechanism by which TGF␤ induces gadd45␤ transcription has recently been reported. First, reconstitution of SMAD4 expression in SMAD4-null pancreatic cell lines restored gadd45␤ induction by TGF␤ (5). The nature of the TGF␤-SMAD-gadd45␤ link appears to be direct; exogenously expressed SMAD2 and SMAD4 or SMAD3 and SMAD4 induce gadd45␤ proximal promoter activity 3-4-fold (11). However, the relative importance and function of each SMAD protein to the transcriptional activation of the endogenous gadd45␤ gene is not known.
Utilizing RNA interference and reconstitution of SMAD3 and SMAD4 protein expression in SMAD-deficient cell lines, we exclude SMAD2 and include SMAD3 and SMAD4 as transcription factors involved in the TGF␤ induction of gadd45␤. Additionally, through a genomics-based approach, we identified a SMAD-dependent TGF␤-responsive enhancer encompassing the third intron of gadd45␤. The importance of this enhancer is indicated by a 3-fold greater transcriptional induction following TGF␤ treatment than transcriptional effects mediated by 5Ј promoter sequences. Finally, using a cell system that does not undergo TGF␤-induced apoptosis but does respond to TGF␤ by gadd45␤ transcriptional induction, we establish an apoptosis-independent role for GADD45␤ as an important mediator of G 2 /M progression following TGF␤ treatment.

MATERIALS AND METHODS
Cell Culture and Drug Treatments-The following cell lines were cultured in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum, 2.0 M L-glutamine, 1.0 M sodium pyruvate, penicillin, and streptomycin and split every third day or at 80% confluence: Mv1Lu (CCL64), HaCaT, HeLa, 293, and 10T1/2. HT29 adenocarcinoma colon cells were cultured in McCoy's medium supplemented with 10% fetal bovine serum. The HepG2 and JAR cell lines were cultured in minimal essential medium and RPMI supplemented with 10% fetal bovine serum, respectively. We obtained all of the cell lines from ATCC except for the HaCaT immortalized keratinocyte cell line, which was a kind gift from D. Grossman (University of Utah, Salt Lake City, UT), the Mv1Lu cells were a kind gift from D. Ayer (University of Utah), and the JAR cells were a kind gift from E. Adashi (University of Utah). Human mammary epithelial cells (HMEC) were obtained from BioWhittaker (Walkersville, MD) and cultured in complete mammary epithelial growth medium. HMEC were seeded at passage 7 or 8 and harvested at no greater than 80% confluence for all experiments. For treatments with TGF␤ (isoform type 1; Peprotech, Rocky Hill, NJ), we found little to no difference with respect to gene transcription if the cells had been previously serum-starved. The vehicle control for TGF␤ comprised 4 mM HCl, 1 mg/ml bovine serum albumin. Cyclohexamide and actinomycin D (Calbiochem) were used at 10 and 5 g/ml, respectively, and treated as described in Fig. 2.
RNA Interference-siRNAs were designed to specifically target either smad2, smad3, or smad4 in accordance with the guidelines developed by Tuschl et al. (12). Because the sequence of mink smad2 and smad3 cDNAs is unknown, siRNAs were designed against the human sequences. The human-designed smad2 and smad3 siRNAs efficiently and specifically silenced mink SMAD2 and SMAD3 protein expression, thus indicating that these sequences are conserved in mink. We designed the smad4 siRNA-A and siRNA-B against the mink sequence, and consequently they do not silence human smad4 (data not shown). The control siRNA (scrambled siRNA; siScr) specifically recognizes human smad4 and thus does not affect mink SMAD2, SMAD3, or SMAD4 expression. The sequences of the chemically synthesized and high pressure liquid chromatography-purified RNA oligomers are as follows (sense strand shown): Smad2 5Ј-UCUUUGUGCAGAGCCCCA-Att; Smad3 5Ј-ACCUAUCCCCGAAUCCGAUtt; Smad4-A 5Ј-GGACGA-AUAUGUUCAUGACtt; Smad4-B 5Ј-UUGGAUUCUUUAAUAACAGtt; siScr 5Ј-GGAUGAAUAUGUGCAUGACtt. To silence gadd45␤ expression, three siRNAs were designed (sense strand shown): siGadd45␤-A (5Ј-GUU GAU GAA UGU GGA CCC Att), siGadd45␤-B (AUC CAC UUC ACG CUC AUC Ctt), and siGadd45␤-C (CUU GGU UGG UCC UUG UCU Gtt). Of these three siRNAs, siGadd45␤-A was the most efficacious and was used to generate the data seen in Fig. 8. All RNA oligomers were reconstituted and annealed following the protocol of Tuschl et al. (12). Mv1Lu cells were plated 24 h prior to transfection and transfected at 70% confluence. All siRNAs were transfected using 18 l of LipofectAMINE 2000/10-cm 2 plate according to manufacturer's guidelines (Invitrogen). For the SMAD silencing experiments, total RNA or protein was isolated 40 -48 h after transfection. In time course experiments, we found that maximal silencing occurred 36 h after transfection for all three SMAD proteins (data not shown). For Gadd45␤ silencing, 3 h after the start of Gadd45␤ siRNA transfection, cells were treated with vehicle or TGF␤ for an additional 2 h prior to RNA isolation or 12 h prior to flow cytometry.
Plasmids and Genomic Alignments-We electronically cloned the human, murine, and rat gadd45␤ genomic loci from publicly available sequence databases. Approximately 8 kb of the genomic loci, starting at 5000 kb upstream of the transcriptional start site, were aligned using the MAVID alignment algorithm (13,14). The portion of this piece of genomic DNA showing conservation among all three species is shown in Fig. 7A. The G45␤-1 (Ϫ1470 bp, ϩ362 bp), G45␤-2 (Ϫ972 bp, ϩ362 bp), G45␤-3 (Ϫ476 bp, ϩ362 bp), G45␤-A (Ϫ1535 bp, Ϫ1042 bp), G45␤-B (Ϫ572 bp, Ϫ79 bp), and G45␤-C (ϩ941 bp, ϩ1428 bp) reporter constructs were created as follows. The indicated region of the human gadd45␤ genomic locus was PCR-amplified from HMEC genomic DNA and cloned into the pCR2.1-TOPO vector (Invitrogen). These DNAs were then subcloned into pGL3basic, sequence-verified, and utilized in subsequent dual luciferase assays. J. Massague generously provided the 3TPLux reporter construct (Memorial Sloan-Kettering Cancer Center, New York). The murine Smad7 cDNA (generously provided by R. Derynk, University of California, San Francisco, CA) was subcloned into pcDNA3.1. Similarly, the FLAG-tagged Smad expression vectors used in the reporter experiments were created by subcloning the cDNAs from constructs provided by D. Satterwhite into pCMV2-FLAG (University of Utah, Salt Lake City, UT). For luciferase assays, all reporters were co-transfected with an SV40-Renilla luciferase reporter plasmid that was used to normalize transfection efficiencies. For retroviral infections, we PCR-amplified the smad3 or smad4 open reading frames from HMEC cDNA and then cloned them into the pBabe retroviral vector. D. Ayer generously provided the GFP-pBabe vector (University of Utah, Salt Lake City, UT).
Luciferase Assays-Fugene 6 (Roche Applied Science) was used to transfect HaCaT cells as instructed by the manufacturer. We seeded cells at a density of 80,000 cells/well in 24-well plates and transfected them the next day. Transfections were performed using 0.6 g of DNA (including either 0.1 g of normalization vector and 0.5 g of reporter vector or 0.1 g of normalization vector, 0.2 g of reporter vector, and 0.3 g of expression vector) and harvested 20 h after the start of transfection. For TGF␤ treatment, medium containing either TGF␤ (200 pM) or an equal volume of vehicle was added to cells 3 h after the start of transfection. Luciferase values were analyzed using a dual luciferase assay system (Promega). Dividing the firefly luciferase activity from each well by the Renilla luciferase activity from the same well normalized transfection efficiencies. Data in each experiment are presented as the mean Ϯ S.D. of triplicates from a representative experiment. All experiments were performed at least three times, producing qualitatively similar results.
Retroviral Transduction-Expression of the GFP, SMAD3, or SMAD4 retroviral constructs was verified by Western blot in a transient assay prior to virus production. To produce the retrovirus, Phoenix helper cells were seeded in 60-mm 2 plates 24 h prior to transfection with LipofectAMINE 2000. 24 h after transfection began, we split the cells 1:3 to 10-cm 2 plates. 48 h after the cells had been split, viruscontaining medium was removed from the Phoenix cells, filtered (0.22 M; low protein binding filter), and added to a 6-well plate containing the HT29 or JAR target cells (at 60% confluence). We added Polybrene (4 g/ml) to the virus immediately before transduction of the target cells to facilitate infection. 24 h after infection, the target cells were split to 10-cm 2 plates and placed under selection with 750 ng/ml puromycin for 10 days.
PCR was performed in duplicate (or triplicate for 18 S rRNA) with a master mix consisting of cDNA template, buffer (500 mM Tris, pH 8.3, 2.5 mg/ml bovine serum albumin, 30 mM MgCl 2 ), dNTPs (2 mM), TaqStart antibody (Clontech), Biolase DNA polymerase (Bioline), genespecific forward and reverse primers (10 M), and SYBR Green I (Molecular Probes, Inc., Eugene, OR). The PCR conditions are as follows: 35 cycles of amplification with 1-s denaturation at 95°C and 5-s annealing at 57°C for Gadd45␤ and 53°C for 18 S rRNA. A template-free negative control was included in each experiment. We determined the copy number by comparing gene amplification with the amplification of standard samples that contained 10 3 to 10 7 copies of the gene or 10 5 to 10 9 for 18 S rRNA. The relative expression level of each gene was calculated by averaging the replicates and then dividing the average copy number of Gadd45␤ by the average copy number of 18 S rRNA. S.E. of the ratios was calculated using a confidence interval.
Northern and Western Blotting-Total RNA was isolated using Trizol following the manufacturer's protocol (Invitrogen). Where indicated, total RNA isolation was followed by poly(A) RNA selection using a PolyATtract TM mRNA Isolation kit (Promega). Total RNA or poly(A) RNA was fractionated through formaldehyde-containing agarose gels and transferred onto NϩHybond nylon membranes (Amersham Biosciences). Labeled probes were generated using the Rediprime II random prime labeling system (Amersham Biosciences) supplemented with [ 32 P]dCTP (ICN). To generate Northern blot probes, we PCR-amplified gene-specific sequences from human, mink, or murine cDNA. Mink gadd45␤ was PCR-amplified using the following degenerate primers: 5Ј-CTGCARATYCACTTCACSCT and 3Ј-GGRAYCCAYTGGTTDTTGC. Hybridizations with 32 P-labeled probes were carried out using ULTRAhyb buffer (Ambion) as recommended by the manufacturer. For Western blotting, protein lysates were harvested in a buffer containing 25 mM Tris-HCl, pH 7.4, 150 mM NaCl, 1 mM CaCl 2 . 1% Triton X-100, 0.1 mM phenylmethylsulfonyl fluoride, 0.1 mM benzamidine, 1 mg/ml pepstatin A, and 1 mg/ml phenanthroline. The resulting whole cell lysates were centrifuged at 20,800 ϫ g for 10 min at 4°C. Following protein quantitation using the DC Protein Assay (Bio-Rad), equal amounts of protein were fractionated through Tris-glycine 4 -12% gradient Nu-PAGE gels using the MES buffer system (Invitrogen). The following antibodies were used to detect the SMAD proteins in both Mv1Lu and human cell lines: SMAD3 (catalog no. 51-1500; Zymed Laboratories Inc.), SMAD2 (catalog no. S66220; Transduction Laboratories), SMAD4 (catalog no. sc-7966; Santa Cruz Biotechnology), and ␤-catenin (catalog no. 610153; Transduction Laboratories). Immune complexes were visualized using a secondary antibody conjugated to horseradish peroxidase (Amersham Biosciences) and Western Lighting chemiluminescence reagent (PerkinElmer Life Sciences).

gadd45␤ Is a Primary TGF␤-responsive Gene in Normal
Human Mammary Epithelial Cells-TGF␤ induces a G 1 cell cycle arrest and epithelial to mesenchymal transition, but not apoptosis, in primary normal human mammary epithelial cells grown in culture (15) (data not shown). To understand the mechanisms behind these TGF␤-induced phenotypes, we partially defined the TGF␤ transcriptome in normal human mammary epithelial cells (HMEC). Specifically, we used cDNA microarray expression analysis to determine the relative expression of 7000 genes at 2 and 12 h after TGF␤ treatment in HMEC. Data analysis revealed 54 up-regulated and 10 downregulated TGF␤-regulated genes. Genes included in this list had a -fold change of greater than 1.3 or less 0.7 at both time points and a p value of less than 0.05 at both time points (Supplemental Table I and methods therein). Next, we identified genes within this data set that were in common to TGF␤regulated genes identified through transcriptional profiling in other TGF␤-responsive cell systems. We surmised that because genes in this subgroup were regulated by TGF␤ irrespective of cell origin or transformation status, they would be of central importance to the TGF␤ cytostatic program. Plasminogen activator inhibitor-1 (PAI1) is a well established TGF␤-induced gene and was induced 8-fold 2 h after TGF␤ treatment in HMEC (Supplemental Table I). Consequently, PAI1 served as an important positive control in the microarray, in Northern blots, and in subsequent experiments (Fig. 1A). A second common TGF␤ target gene identified in our expression analysis was gadd45␤/hMyD118. In addition to our studies in primary normal HMEC, previous findings indicate that gadd45␤ is a TGF␤-induced gene in transformed cell lines derived from my- FIG. 1. gadd45␤ is a TGF␤-inducible gene. A, various established cell lines that have been previously reported to be TGF␤-sensitive by some measure were treated with TGF␤ for 1 h prior to total RNA isolation and Northern blotting for Gadd45␤ and PAI1. The PAI1 3.2-kb transcript is shown; we could not detect the 2.2-kb PAI1 transcript in Mv1Lu or 10T1/2 using human, mink, or murine PAI1 probe. Extended exposure of the Northern blots and additional experiments not shown verified that the human PAI1 and gadd45␤ Northern probes are capable of recognizing their respective orthologues. The 18 S ribosomal band was visualized in the ethidium bromide-stained gel prior to Northern blotting and serves as the loading control. B, randomly cycling HMEC were treated with either TGF␤ (200 pM) or an equal volume of vehicle. At the indicated time, total RNA was isolated from the cells. Following mRNA purification, Northern blot analysis was performed to visualize the relative transcript abundance of the indicated genes. Both the 3.2and 2.2-kb alternatively spliced forms of the mature PAI1 mRNA are shown. C, HMEC were treated with TGF␤ (200 pM for 2 h) or BMP-2 (4 nM) for the indicated times before RNA isolation and Northern blot analysis for the gadd45␤ transcript. GAPDH serves as a loading control. The gadd45␤ and GAPDH signals were quantitated using a Phos-phorImager, and the resulting gadd45␤/GAPDH ratio was plotted below the Northern blots. eloid, breast, skin, breast, pancreas, and bone (4,5,11,16). Because of its frequent presence in the TGF␤ transcriptional response and because of its previously described role in growth arrest, differentiation, and apoptosis, we chose to characterize the upstream signal transduction pathway necessary for gadd45␤ transcriptional induction and, second, to examine the role of GADD45␤ in the TGF␤ response.
We first determined the scope of gadd45␤ transcriptional activation. Specifically, we monitored its induction by TGF␤ in several cell lines and by other members of the TGF␤ superfamily in HMEC. To determine whether other TGF␤-responsive cell lines responded similarly to HMEC with respect to gadd45␤ transcription, several cell lines were treated with TGF␤ or vehicle for 1 h. The gadd45␤ and PAI1 transcripts were induced by TGF␤ in the following cell lines: HMEC, Ha-CaT, Mv1Lu, PANC-1, primary breast organoid outgrowths, and to a lesser extent in HepG2 and HeLa cells (data not shown) (Fig. 1A). The Madin Darby canine kidney and 293 cell lines did not respond to TGF␤ stimulation by inducing either gadd45␤ or PAI1. TGF␤ treatment of 10T1/2 murine fibroblasts caused a moderate increase in PAI1 transcription but did not affect gadd45␤ mRNA levels. We also asked whether other members of the TGF␤ superfamily of growth factors could regulate gadd45␤ transcription. Fig. 1C illustrates that both TGF␤ and BMP2 induced gadd45␤ transcription. However, the kinetics of gadd45␤ induction as well as the strength of induction differed between the two ligands. Finally, of the three genes that comprise the gadd45 family, only gadd45␤ was found to be TGF␤-inducible in HMEC; gadd45␣ was not affected by TGF␤ treatment, and Gadd545␥ was not detected (Fig. 1B). Gadd153/Chop10, a GADD family member by virtue of its induction by cellular stress, was transiently repressed by TGF␤.
To distinguish whether TGF␤ treatment resulted in increased gadd45␤ transcription or increased gadd45␤ mRNA stability, we measured the gadd45␤ mRNA half-life before and after TGF␤ treatment. HMEC were treated with TGF␤ for 1 h before the addition of the transcription inhibitor actinomycin D for various periods of time. Quantitative analysis of the Northern blot revealed that TGF␤ failed to stabilize the gadd45␤ mRNA (Fig. 2, A and B). The accumulation of gadd45␤ mRNA within 2 h of TGF␤ treatment suggested that it is an immediate early TGF␤-induced target gene. To test this idea, we pretreated HMEC with the protein translation inhibitor cyclohexamide 15 min before a 3-h combined TGF␤/cyclohexamide treatment. We found that the levels of gadd45␤ increased in a TGF␤-dependent manner irrespective of cyclohexamide pretreatment, indicating that new protein synthesis is not required for TGF␤ induction of gadd45␤ (Fig. 2C). These data indicate that gadd45␤ is a direct TGF␤ transcriptional target.
gadd45␤ Is Partly Dependent upon SMAD3 and Independent of SMAD2 in Its Regulation by TGF␤ -We first sought to determine whether specific inhibition of SMAD2, SMAD3, and SMAD4 abrogated gadd45␤ responsiveness to TGF␤. To approach this, we employed siRNA-mediated silencing of the SMAD2, SMAD3, and SMAD4 proteins. Because we were unable to achieve silencing greater than 60% of wild-type levels in HMEC, we chose to use Mv1Lu cells for our siRNA studies. Transfection of Mv1Lu cells with siRNAs specific to SMAD2 or SMAD3 reduced the respective protein expression to nearly undetectable levels (Fig. 3A). Loss of SMAD2 caused a 70% decrease in the induction of PAI1 by TGF␤. In contrast, siRNA silencing of SMAD2 had no significant effect on gadd45␤ induction following TGF␤ treatment (Fig. 3A). SMAD3-deficient cells, however, responded to TGF␤ stimulation with reduced levels of induction for both gadd45␤ and PAI1. Although the decrease in PAI1 induction by TGF␤ observed in the SMAD2 and the SMAD3 single-knockout cells was enhanced in the double-knockout cells, the SMAD2/SMAD3 double-knockout cells behaved similarly to SMAD3-deficient cells with respect to gadd45␤ induction (Fig. 3A). Dose-response curves with the Smad3 siRNA (IC 50 ϳ1 nM) further demonstrated that TGF␤ activates PAI1 and gadd45␤ through a mechanism that is partly dependent upon SMAD3 (Fig. 3B).
SMAD4 Silencing Prevents gadd45␤ and PAI1 Induction by TGF␤ -Of the many proteins involved in mediating the different facets of TGF␤ signal transduction, SMAD4 is considered central to many of the responses. Two different siRNAs were designed against mink smad4, and the efficacy of their silencing was tested in Mv1Lu cells by Western blot (Fig. 4A). Consistent with the central role of SMAD4 in TGF␤ signaling, siRNA silencing of SMAD4 resulted in a dramatic loss of gadd45␤ transcriptional induction following TGF␤ treatment (Fig. 4B). As a confirmation of specificity, a human-specific SMAD4 siRNA, which contains mismatches at two positions relative to the mink sequence, did not affect SMAD4 protein expression or TGF␤-regulated transcription of gadd45␤ or PAI1. siSmad4-A and siSmad4-B both robustly silenced SMAD4 protein expression and did not interfere with SMAD3 protein expression (Fig. 4A). Examination of the gadd45␤ and PAI1 transcript levels in these SMAD4-deficient cells revealed a clear necessity for SMAD4 in targeting these genes for tran- gadd45␤ Enhancer Mediates TGF␤ Induction scription following TGF␤ stimulation (Fig. 4B). The small difference between siSmad4-A and siSmad4-B in silencing SMAD4 protein expression directly reflected the levels of gadd45␤ and PAI1 induction by TGF␤. The siSmad4-A silences SMAD4 protein expression with an IC 50 of less than 1 nM, which is consistent with the IC 50 of silencing imparted by siSmad3 (compare Fig. 4C with Fig. 3B). The induction of gadd45␤ and PAI1 transcripts by TGF␤ in these cells showed close correlation with each other and with the levels of SMAD4 protein (Fig. 4C).
Finally, we asked whether Mv1Lu cells lacking SMAD2, SMAD3, and SMAD4 responded differently to TGF␤ with respect to gadd45␤ transcriptional induction than cells deficient in only one or two of the SMADs. Mv1Lu cells were transfected with siRNAs directed against each of the SMADs alone and in all combinations thereof (Fig. 5A). Northern blot analysis of gadd45␤ Enhancer Mediates TGF␤ Induction gadd45␤ again demonstrated a SMAD3 and SMAD4 dependence for TGF␤-induced transcription. Loss of SMAD2 in these SMAD3/SMAD4-deficient cells had no further effect on gadd45␤ induction. Interestingly, although PAI1 depends partly upon SMAD2 for TGF␤-induced transcription (Fig. 3A), loss of SMAD2 did not affect PAI1 induction in cells lacking SMAD3 and SMAD4 (Fig. 5B).
Our data generated with siRNA-mediated silencing have revealed no differences between gadd45␤ and PAI1 with respect to their regulation by SMAD3 and SMAD4 (Figs. 3 and 4). To examine the role of SMAD3 and SMAD4 more closely, we asked whether loss of SMAD3 in a SMAD4-reduced background would further inhibit gadd45␤ and PAI1 induction by TGF␤. Mv1Lu cells were transfected with a constant amount of siSmad4-A (15 nM) in the presence of an increasing concentration of siSmad3 (Fig. 5C). siRNA-mediated silencing of SMAD3 in a SMAD4-reduced background had no effect on PAI1 induction (Fig. 5, C and D). SMAD3 silencing in these SMAD4deficient cells did, however, further repress the transcriptional induction of gadd45␤ following TGF␤ treatment. These data support a transcriptional model that distinguishes Gadd45␤ from PAI1 in their regulation by SMAD3 and SMAD4.
SMAD3 and SMAD4 Expression in SMAD3-and SMAD4null Cancer Cells Reconstitutes TGF␤-mediated Induction of gadd45␤ -The second approach we utilized to study the transcriptional regulation of gadd45␤ by TGF␤ relied upon the preponderance of inactivating mutations within the SMAD proteins in human cancer cell lines. HT29 colon adenocarcinoma cells do not express SMAD4 protein because of a nonsense mutation that renders the transcript unstable (17). JAR cells, on the other hand, do not express SMAD3 (18). TGF␤ treatment of these cell lines results in the phosphorylation of SMAD2, indicating that both cell lines express functional TGF␤ receptor complexes and that SMAD2 phosphorylation is not dependent upon SMAD3 or SMAD4 (Fig. 6A). Retroviral transduction followed by polyclonal selection of these cells with either a GFP-encoding retrovirus or a SMAD3-or SMAD4encoding virus provided an experimental approach to further examine the role of the SMADs in gadd45␤ transcription. Two weeks after the transduced cells were placed under selection, expression of the transduced genes was verified by fluorescence microscopy (for GFP expression; data not shown) and Western blot (Fig. 6B). RNA harvested in parallel to the protein samples analyzed in Fig. 6B was reverse transcribed and used in real time quantitative PCR to measure the gadd45␤ transcript levels. TGF␤ treatment of JAR-SMAD3 cells revealed a small but statistically significant increase in gadd45␤ message levels (Fig. 6C). The SMAD4-HT29 cells responded to TGF␤ through a robust induction of gadd45␤ (Fig. 6C). Northern blot analysis of these RNAs confirmed the quantitative PCR results (data not shown).
gadd45␤ Contains a TGF␤-responsive Enhancer That Encompasses the Third Intron-Next, we analyzed the gadd45␤ genomic locus for transcriptional responsiveness to TGF␤. First, 1500 bp of the proximal promoter of gadd45␤ was cloned upstream of firefly luciferase for use in reporter assays (G45␤-1) (Fig. 7A). TGF␤ stimulation of HaCaT or Mv1Lu cells increased the transcriptional activity of G45␤-1, G45␤-2, and G45␤-3 ϳ2-fold (Fig. 7B) (Mv1Lu data not shown). In contrast, the endogenous gadd45␤ transcript levels increased 8 -15-fold in responsive cell lines following TGF␤ treatment (Fig. 1A). We were unable to see increased reporter activity when other portions of this 5Ј-flanking region were analyzed or in numerous other cell lines or when the cells were treated for different lengths of time with TGF␤ ( Fig. 7A) (data not shown). We reasoned that because the gadd45␤ coding sequence is highly conserved between human, mouse, rat, and zebrafish, the region of the genomic locus mediating TGF␤ responsiveness might also be conserved. To address this possibility, we aligned the human, mouse, and rat gadd45␤ genomic sequences and plotted the degree of conservation utilizing the MAVID algorithm (13,14). In addition to the coding regions, three domains of the gadd45␤ genomic locus demonstrate high conservation between species. Each of these regions was cloned upstream of Firefly luciferase and used in reporter experiments (Fig. 7A). Remarkably, TGF␤ robustly activated transcription from the G45␤-C enhancer, which contains part of the third exon, the complete third intron, and part of the fourth exon, but not from G45␤-A or G45␤-B (Fig. 7B, Supplemental Fig. 1) (Mv1Lu data not shown). We took two approaches to test if SMAD proteins were mediating TGF␤-dependent transcriptional induction off of G45␤-C. First, the inhibitory SMAD7 protein was overexpressed to block SMAD activation. SMAD7 overexpression inhibited TGF␤-induced activation of the 3TPLux reporter, which contains the PAI1 promoter, and the G45␤-C reporter, it but did not affect an SV40-driven luciferase construct (Fig. 7C). Second, overexpression of SMAD3 and SMAD4 greatly enhanced G45␤-C reporter activity in HaCaT cells (Fig. 7C) and in HeLa cells (data not shown). In contrast, SMAD2 expression did not affect the transcriptional activity. Interestingly, the increase in reporter activity was dependent upon both SMAD3 and SMAD4, because neither one alone significantly affected the G45␤-C transcriptional activity. These data support a role for SMAD3 and SMAD4 in regulating Gadd45␤ transcription through a 3Ј enhancer that contains the third intron. Indeed, sequence analysis of G45␤-C revealed four conserved putative SMAD binding elements (SBEs) (Supplementary Fig. 1).
GADD45␤ Regulates G 2 Progression following TGF␤ Stimulation-To examine the contribution of GADD45␤ to the TGF␤ phenotype, a siRNA was designed to silence gadd45␤ expression. Dose-response analysis revealed potent (IC 50 ϳ1 nM) and specific knockout of TGF␤-induced gadd45␤ expression (Fig.  8A). TGF␤ rapidly induces a G 1 cell cycle arrest, but not apoptosis, in Mv1Lu cells. We asked whether Mv1Lu cells deficient in GADD45␤ would undergo a G 1 cell cycle arrest. Introduction of a scrambled siRNA had no detectable effect on TGF␤induced cell cycle arrest (Fig. 8B). However, cells containing reduced levels of gadd45␤ demonstrated a slight reduction in G 1 accumulation and failed to progress through G 2 following TGF␤ treatment (Fig. 8B). Loss of the gadd45␤ transcript did not affect cell cycle progression in the absence of TGF␤ treatment. Dose-response analysis further verified this finding; 0.01 and 0.1 nM siRNA did not significantly affect gadd45␤ transcript levels or cell cycle progression following TGF␤ stimulation. These findings indicate that GADD45␤ is an important regulator of cell cycle progression following TGF␤ treatment.

DISCUSSION
The intracellular domain of a ligand-bound TGF␤ receptor complex ignites an intertwined cascade of signaling events that induces one of many possible phenotypic responses (1, 2). Consequently, the mechanism by which a cell decides how to respond to TGF␤ is fundamental to many aspects of eukaryotic life. One approach to decipher the cellular interpretation of the TGF␤ signal and how that interpretation might be altered in a diseased tissue is to define and utilize the TGF␤ target genes as a starting point in a retrograde molecular characterization of the upstream transcriptional program. Concurrent studies would assess the gene function as it contributes to the phenotypic response. We have employed this approach to the gadd45␤ gene. We found that gadd45␤ transcriptional induction by TGF␤ was dependent upon SMAD4 and to a lesser extent on SMAD3 but independent of SMAD2. Further, SMAD3 and SMAD4 mediated the transcriptional induction of gadd45␤ through an enhancer that encompasses the third intron of the gadd45␤ gene. Finally, TGF␤ stimulation of  7. gadd45␤ is activated by TGF␤ through a 3 enhancer. A, schematic representation of the gadd45␤ genomic locus. The relative position of the exons and introns are indicated above the graph. The MAVID algorithm was used to determine the relative degree of sequence conservation between human gadd45␤ (x axis) and rat gadd45␤ (bottom half of plot; y axis) and murine gadd45␤ (top half of plot; y axis). The degree of genomic conservation is indicated by the height of the black curve. Below the graph is a schematic diagram of six pieces of the gadd45␤ genomic locus that were cloned upstream of firefly luciferase for use in subsequent reporter assays. Of note, G45␤-C contains 93 bp of exon 3, the complete third intron (237 bp), and 98 bp of exon 4 of gadd45␤ ( Supplementary Fig. 1) gadd45␤-deficient cells, but not of gadd45␤-expressing cells, resulted in the activation of a G 2 /M checkpoint.
We used RNA interference as a tool to probe the upstream signal transduction components necessary for gadd45␤ and PAI1 transcriptional induction following TGF␤ stimulation. We chose Mv1Lu cells as a cell system for these studies rather than a human cell line such as HMEC or HaCaTs, because we found that in these cells our siRNAs were more efficacious as compared with a panel of TGF␤-responsive human cells. In fact, silencing SMAD protein expression in HaCaT cells to 30% of wild type levels resulted in no detectable effect on PAI1 or gadd45␤ transcription following TGF␤ treatment. 2 These results and our findings presented in Fig. 4 argue that with respect to the transcriptional activation of gadd45␤ and PAI1, the SMAD proteins are expressed in excess. Coupled to the immediate early transcriptional induction of gadd45␤ and PAI1 by TGF␤, these data suggest that the gadd45␤ and PAI1 promoters share a relatively high affinity for the SMAD proteins. Further, this provides a possible molecular mechanism explaining how gadd45␤ and PAI1 are regulated by TGF␤ irrespective of tissue type. Analogous findings have recently been discovered in Caenorhabditis elegans where the FoxA protein, PHA4, achieves transcriptional discrimination among target genes through a differential affinity to gene promoter sequences (19). Consequently, high affinity PHA4 promoters are responsive to relatively low levels of PHA4 protein expression. Further studies are in progress to classify TGF␤ transcriptional targets by their sensitivity to changes in SMAD protein expression.
With the exception of a few genes, such as p15 (20) and MMP2 (21), most well characterized immediate early TGF␤regulated genes appear to depend upon SMAD3 and SMAD4, but not SMAD2, for TGF␤ transcriptional regulation. Our work places gadd45␤ within this SMAD2-independent, SMAD3/ SMAD4-dependent class of TGF␤-responsive genes. Our conclusion that gadd45␤ is a SMAD4-dependent TGF␤ target gene agrees with the findings of Yoo et al. (11) and Takekewa et al. (5), who have also reported SMAD4 dependence in gadd45␤ regulation, although through different experimental approaches. Conversely, our findings that TGF␤ regulated gadd45␤ independently of SMAD2 contradict previous findings. Yoo et al. (11) recently reported that overexpression of SMAD2 and SMAD4 together, but not separately, induced a gadd45␤ reporter construct in a TGF␤-dependent fashion. Although additional work is necessary to reconcile these outcomes, they could result from cell type-specific responses to TGF␤ (hepatocytes versus keratinocytes and fibroblasts). It is important to note, however, that SMAD2-deficient fibroblasts show gadd45␤ transcriptional induction following TGF␤ with kinetics and efficacy similar to that of wild-type cells, an observation that is consistent with a SMAD2-independent model of gadd45␤ regulation (22).
In contrast to gadd45␤, we found that SMAD2, SMAD3, and SMAD4 all contributed to TGF␤ regulation of PAI1, although to varying degrees (Figs. [3][4][5]. Extensive research on PAI1 has not implicated SMAD2 (23)(24)(25) in its regulation with the notable exception that fibroblasts derived from smad2 knockout mice failed to induce PAI1 following TGF␤ treatment (21). The ability of smad2 siRNAs to phenocopy the Smad2 knockout fibroblasts in this respect strongly supports the use of siRNAmediated gene silencing in future TGF␤ transcriptional studies. Clearly, genome-wide analysis of TGF␤ responsiveness in SMAD-silenced or SMAD knockout cells will be of great importance.
Yoo et al. (11) have recently shown that 220 bp of the gadd45␤ proximal promoter is activated by TGF␤ and that this activation is enhanced by overexpression of SMAD2, SMAD3, and SMAD4, but not dominant negative forms of SMAD2 or SMAD3. Our data support their results in that we have also found the 5Ј promoter sequence to be TGF␤-responsive (Fig.  7B). However, through a genomics-based alignment strategy, we identified a second TGF␤-responsive domain encompassing the highly conserved third intron of the gadd45␤ gene. In contrast to the 2-fold activation we observed with 5Ј promoter sequences, the 3Ј enhancer is activated 5-7-fold following TGF␤ treatment. It will be important to determine which of the conserved transcription factor binding sites within this enhancer account for induction by TGF␤. Notably, we identified four conserved SMAD binding elements, three of which are located in exonic sequence ( Supplementary Fig. 1). The endogenous gadd45␤ gene may likely respond to TGF␤ through a concerted action of the 3Ј enhancer and 5Ј promoter sequences. A similar transcriptional model has been reported for the gadd45␣ gene where highly conserved sequences within the third intron or fourth exon facilitate transcriptional induction following genotoxic stress (26,27) and vitamin D3 (28), respectively. Thus, in addition to primary sequence and genomic organization, gadd45␣, gadd45␤, and gadd45␥ might also 2  share an intronic/exonic enhancer as an important transcriptional regulatory element.
Last, utilizing the power of siRNA mediated gene silencing, we discovered that Mv1Lu cells made deficient for gadd45␤ arrested at the G 2 /M checkpoint following TGF␤ treatment. Previous research has established GADD45␤ as a negative regulator of cell cycle progression, and several molecular mechanisms behind this inhibition have been put forth (3). Following genotoxic stress, GADD45␤ acts to inhibit Cdc2/cyclin B1 kinase to induce a G 2 /M cell cycle checkpoint in RKO lung carcinoma cells (8). In contrast, normal fibroblasts microinjected with a GADD45␤ expression vector fail to undergo a G 2 /M arrest, although GADD45␤ was found to associate with Cdc2 in these cells (29). Our findings support these previous data in that we also see a GADD45␤-dependent effect on the G 2 /M cell cycle checkpoint. However, we show that GADD45␤ acts to promote G 2 /M progression following TGF␤ treatment in Mv1Lu cells (Fig. 8). This finding supports the notion that GADD45␤ does not act to modulate cell cycle progression in isolation, but rather the presence of other proteins might ultimately determine how cells respond to increases in GADD45␤ protein levels (3). Indeed, GADD45␤ associates with many nuclear proteins involved in cell cycle progression, including proliferating cell nuclear antigen, p21, GADD45␣, and Cdc2/ cyclin B2 (8,30,31).
Perhaps the most well understood function of the GADD45 family of proteins is their ability to regulate apoptosis through the activation of MTK1 (MEKK4) (32) and subsequently p38 kinase (5,10). Although gadd45␤ is rapidly induced by TGF␤ in Mv1Lu and HMEC, we have not detected an apoptotic response following TGF␤ treatment in these cells. 2 An apoptosis-independent cellular response to GADD45␤ induction was recently shown, where tumor necrosis factor ␣ signaling through NF-B induced gadd45␤ transcription to prevent c-Jun N-terminal kinase activation and cell death (33,34). Further, several research laboratories have been successful in generating gadd45␤ overexpression systems and have not observed cell death (35). Future studies utilizing siRNA silencing of gadd45␤ following transcriptional agonists other than TGF␤ will be invaluable in determining the functional consequences of GADD45␤ expression.