Runx1 Is a Co-activator with FOXO3 to Mediate Transforming Growth Factor β (TGFβ)-induced Bim Transcription in Hepatic Cells*

Transforming growth factor β (TGFβ) regulates essential cellular functions such as cellular proliferation, differentiation, and apo pto sis. The Bcl-2 family of proteins has been implicated as mediators of TGFβ-induced apo pto sis. We demonstrated previously that TGFβ induces the expression of Bim (Bcl-2-interacting mediator of cell death), a member of the BH3-only family of pro-apo pto tic Bcl-2 proteins, to induce cell death in B-lymphocytes. Here, we investigated the mechanism of TGFβ-mediated Bim expression in two hepatocyte cell lines that undergo apo pto sis with TGFβ, AML-12 and Hep3B. We show that TGFβ induces Bim protein and mRNA levels, and its expression is sufficient to induce cell death. Gene array results revealed that Runx1, a member of the Runx family of transcription factors, was induced by TGFβ, and this induction was confirmed at the mRNA and protein levels. Interestingly, TGFβ specifically induced the expression of Runx1 protein from an internal ribosome entry site (IRES)-dependent, cap-independent, mRNA transcript, and its overexpression was sufficient to induce hepatocyte apo pto sis. Deletion and mutation analyses of the murine Bim promoter identified a putative forkhead binding element, at position −174 to −168 from the transcription start site, as the mediator of Runx1 induction. Co-immunoprecipitation, electrophoretic mobility shift assays, and chromatin immunoprecipitation assays demonstrated that Runx1 does not bind directly to the identified forkhead binding element but rather binds the transcriptional regulator FOXO3, which occupies this site. Finally, small interfering RNA knockdown of Runx1 or FOXO3 decreased TGFβ-induced Bim expression. Our results support a mechanism in which TGFβ stimulates Bim transcription by up-regulating Runx1 expression, which binds FOXO3, and the two cooperate in the transcriptional induction of Bim.

no reported hepatic phenotype in Bim knock-out mice (21,22), there are reports of Bim being involved in hepatic apoptosis in vivo during fatal hepatitis (23) and Fas-induced liver damage (24).
The pro-apoptotic activity of Bim is highly regulated by prosurvival and pro-apoptotic cytokines through both post-translational and transcriptional mechanisms. Studies on the posttranslational control of Bim have demonstrated that Bim is phosphorylated by ERK, leading to its subsequent ubiquitination and degradation (25,26). In hepatocytes, we have recently demonstrated that TGF␤/Smad3 signaling induces the dual specificity phosphatase DUSP4 (MKP2) to inhibit ERK activity, thereby stabilizing Bim and promoting apoptosis (27). Studies on the transcriptional control of Bim have focused largely on FOXO3, a member of the forkhead family of transcriptional regulators, which directly stimulates Bim transcription (28,29). Multiple reports have shown that the phosphatidylinositol 3-kinase (PI3K)/Akt pathway is a negative regulator of FOXO3stimulated Bim transcription via phosphorylation of FOXO3, leading to the binding and sequestration of this transcription factor by cytosolic 14-3-3 proteins (30,31).
The prominent role of Bim as a mediator of apoptosis is gaining in importance. Thus, an understanding of how Bim expression is regulated under various physiological and pathological conditions is of significance. There are relatively few studies that directly address the transcriptional regulation of Bim. Herein, we examined the mechanism through which TGF␤ mediates the transcriptional induction of Bim using hepatocyte cell lines that undergo apoptosis in the presence of TGF␤. We demonstrate that TGF␤ stimulates Bim transcription by upregulating the expression of the transcription factor Runx1 through an IRES-dependent mechanism. TGF␤-induced Runx1, in turn, binds to FOXO3, residing at its forkhead binding site, and the two cooperate to stimulate Bim promoter transactivation.

EXPERIMENTAL PROCEDURES
Reagents-TGF␤ was a generous gift from Genzyme Inc. (Cambridge, MA). Rabbit anti-Bim antibody was obtained from BD Biosciences. Rabbit anti-phospho-FOXO3 (Ser-253), rabbit anti-total FOXO3, and rabbit anti-cleaved caspase-3 (Asp-175) were purchased from Cell Signaling Technology (Beverly, MA). Mouse anti-Bcl-X L (H-5), rabbit anti-Hsp-90 (H-114), and normal mouse and rabbit IgG were obtained from Santa Cruz Biotechnology (Santa Cruz, CA). Rabbit anti-AML-1 (Runx1) was obtained from Active Motif (Carlsbad, CA) and mouse anti-FLAG antibody from Sigma. Sheep anti-mouse IgG-HRP and donkey anti-rabbit IgG-HRP were purchased from Amersham Biosciences (GE Healthcare). HRP-protein A was from Zymed Laboratories Inc. (San Francisco, CA). LY294002 and rapamycin were obtained from Calbiochem. The mouse anti-HA antibody (12CA5), protease inhibitor mixture tablets, and the cell death detection ELISA Plus kit were purchased from Roche Diagnostics. Tris-buffered saline-casein for Western blotting was obtained from Pierce Chemical Co. ExpressFect was ordered from Denville Scientific (Denville, NJ). The Dual-luciferase reporter system was purchased from Promega (Madison, WI) and the SuperScript III RT kit from Invitrogen. The RNeasy and Oligotex RNA isolation kits were obtained from Qiagen (Valencia, CA). All primers were ordered from IDT Technologies (Coralville, IA). ON-TARGETplus SMARTpool siRNA, non-targeting pool siRNA, and DharmaFECT 4 reagent were obtained from Dharmacon (Lafayette, CO). Reagent chemicals were obtained from Sigma.
Western Blot Analysis-Whole cell lysates were prepared in TNMG buffer (20 mM Tris, pH 8.0, 150 mM NaCl, 5 mM MgCl 2 , 10% glycerol, and 0.5% Nonidet P-40) containing protease inhibitors. Lysates were sonicated briefly and clarified by centrifugation at 4°C for 10 min in a Beckman tabletop microcentrifuge at maximum speed. Protein concentration of the extracts was determined using Bradford's reagent (Pierce). Western blot analysis was performed by standard SDS-PAGE, as described previously (33).
Apoptosis Assays-Apoptosis was demonstrated by DNA ladder formation and was detected by ELISA as described previously (15). In ELISA assays, 100 ϫ 10 3 cells were seeded into wells of a 24-well dish. Cells were lysed in 200 l of buffer supplied with the kit, and 20 l of the clarified lysate was used in the ELISA. ELISA results are expressed as the ratio of the ABS 405 signal of TGF␤-treated samples normalized to the ABS 405 signal of the control sample.
Northern Blot Analysis-Northern blots were performed using mRNA as described previously (15). Total RNA (ϳ200 g) was isolated from a 150-mm dish of cells using an RNeasy Mini kit, and mRNA was purified from equal amounts of total RNA using an Oligotex mRNA Mini kit. Blots were hybridized in NorthernMax Prehyb/Hyb buffer (Ambion, Austin, TX) and typically washed with 0.1% SDS, 0.5ϫ SSC at 55°C. Northern blots were subjected to phosphorimaging analysis to quantitate band intensity (GE Healthcare). The Bim probe used in Northern analyses was obtained by EcoRI/XhoI digestion of pcDNA3.1/BimEL to release the full-length insert (33). The probe used for ␤-actin was generated by PCR as described previously (15). Runx probes were generated by PCR as described below.
Transient Transfection and Luciferase Assays-Transient transfection of cells was accomplished using ExpressFect under serum-free conditions following the manufacturer's instructions. After 5-6 h of transfection, the medium was replaced with complete medium with or without TGF␤ as indicated. For reporter assays, about 100,000 cells/well were seeded into 12-well plates and typically transfected with 0.5 g of the mouse 0.8 Bim luciferase promoter construct (34), 10 ng of pRL-SV40-Renilla luciferase (Promega), and up to 0.5 g of additional plasmid DNA per well. GFP was used as a negative control, to ensure that all cells received equal amounts of DNA. Approximately 24 h after the start of transfection, the cells were collected in 200 l of lysis buffer, and a 20-l aliquot was assayed for luciferase activity using Stop & Glo reagent (Promega). For Western blot and DNA ladder experiments, 4 ϫ 10 6 cells were seeded into 100-mm dishes and transfected with up to 10 g of the appropriate plasmid DNA. GFP was again used to balance the amount of DNA transfected into cells. Cells were collected 24 h later for analyses. The human pLNCX-FLAG-AML1a (Runx1A) and pLNCX-FLAG-AML1b (Runx1B) expression plasmids have been described previously (35), as have the wild-type and triple mutant pECE-HA-FKHRL1 (FOXO3) plasmids (36).
In siRNA experiments, cells were plated into a 24-well dish (50,000 cells/well) for Western blot and ELISA studies or a 6-well dish (400,000 cells/well) for RT-PCR studies. Cells were transfected with specific siRNAs or non-targeting siRNAs in serum-free medium for 5-6 h at a final concentration of 100 nM using DharmaFECT 4 reagent following the manufacturer's protocol. An equal volume of complete medium containing 2ϫ fetal calf serum and antibiotics was then added, and the transfection was continued for 24 h. The next day, TGF␤ was added, and the cells were incubated overnight. Cells were analyzed by Western blot, ELISA, and RT-PCR as described above and below.
Gene Array-Microarray analysis was performed using the Illumina Mouse-6 BeadChip platform as described previously (27). Briefly, total RNA was prepared as described above from control and TGF␤-treated (30 min and 1 h) AML-12 hepatocytes from two experiments. cRNA synthesis, hybridization, and washing were performed using Illumina reagents following the manufacturer's directions. The data output from the Illumina BeadChip scanner was normalized using the rank-invariant method and analyzed using BeadStudio software.
RT-PCR-Total RNA was prepared as described above using an RNeasy Mini kit. RNA was subjected to DNase digestion using DNA-free (Ambion), and cDNA was prepared using random primers supplied with the SuperScript III RT kit (Stratagene). The primers used to amplify the 3Ј-UTR of mouse Runx1 were 5Ј-AAACAAGTTGGTAGGTAACGCAGC-3Ј for the top strand and 5Ј-TGACTGATTCCACTAAGACTGG-GAG-3Ј for the bottom strand. The PCR reaction was annealed at 55.1°C and run for 28 cycles to generate a 524-bp product. The primers used to amplify the 3Ј-UTR of human Runx1 were 5Ј-CGCTGGAAAGCAAACAGGAAG-3Ј for the top strand and 5Ј-GGTCAAAGCAAGAAAGAAGC AAGC-3Ј for the bottom strand. The PCR reaction was annealed at 52.9°C and run for 28 cycles to generate a 319-bp product. The primers for ␤-actin PCR were described previously (15). The PCR reaction was annealed at 66°C and run for 18 cycles. The IRES 5Ј-UTR of Runx1 was amplified from mouse primary B-cell cDNA using a top strand primer of 5Ј-GGCTGGCACTTCCATCCTG-3Ј and a bottom strand primer of 5Ј-CACAACAAGCCGATTGAG-TAAGG-3Ј. An approximate 1.4-kb product was generated after 35 cycles using an annealing temperature of 56.3°C. The IRES PCR product was TA-cloned into the pGEM T-Easy vector (Promega), which was digested with EcoRI to produce a probe for Northern blots. The primary B-cells were isolated and cultured as described previously (27,37). The cap 5Ј-UTR of Runx1 was amplified from cDNA prepared from a human B-cell line (Daudi) using a top strand primer of 5Ј-AACCA-CAAGTTGGGTAGCCTGG-3Ј and a bottom strand primer of 5Ј-AAAATGCTGTCTGAAGCCATCG-3Ј. An approximate 171-bp product was generated after 32 cycles using an annealing temperature of 55.6°C. The cap PCR product was purified using a PCR purification kit (Qiagen) and used for Northern blots. Daudi cells were cultured as described above. All PCR reactions used AccuPrime Taq polymerase using Buffer I (Invitrogen) except for the ␤-actin PCR reactions, which used Hot-Start Taq (Denville Scientific).
Nuclear Extract Preparation-Nuclear extract preparation for EMSA, Western blotting, and immunoprecipitation experiments was performed following the method of Schreiber et al. (38). Typically, cells from one 100-mm culture dish were resuspended in 400 l of hypotonic buffer, and the nuclear pellet was extracted in 100 l of a high salt buffer. Protein concentration of the extracts was determined using Bradford's reagent. Typically 5 g of nuclear extract was analyzed in EMSA and 150 g in Western blotting experiments.
Rapamycin Inhibition of Protein Synthesis-Hep3B cells were seeded into a 24-well dish at a density of 100,000 cells/well in serum-containing medium. The next day rapamycin (20 nM) was added for up to 24 h. During the last hour of each treatment [ 35 S]methionine (PerkinElmer Life Sciences) was added to a concentration of 100 Ci/ml in serum-free medium lacking methionine. Cells were washed twice with phosphate-buffered saline and lysed in 200 l of buffer (20 mM Tris, pH 7.4, 1% Triton X-100, 10% glycerol, 130 mM NaCl). The lysate was centrifuged, and 10 l of the supernatant was added to 100 l of bovine serum albumin (1 mg/ml in lysis buffer). Proteins were precipitated by adding 1 ml of ice-cold 10% trichloroacetic acid and collected by membrane filtration. Radioactivity on the membranes was determined by liquid scintillation counting.
Co-Immunoprecipitation-Whole cell lysates for immunoprecipitation (IP) were prepared in IP lysis buffer (20 mM Tris, pH 7.5, 150 mM NaCl, 1 mM EDTA, 1 mM EGTA, 1% Triton X-100) containing protease inhibitors. Cell lysates (250 g) were incubated overnight at 4°C with 20 g of anti-FLAG antibody in 800 l of IP lysis buffer containing protein G-agarose. The immune complexes were collected by centrifugation and washed four times in IP lysis buffer. The presence of forkhead protein in the immune complexes was detected by Western blotting. Nuclear extracts (1 mg) were prepared as described above, diluted 5-fold in IP lysis buffer, and precleared with protein G-agarose and rabbit IgG. After centrifugation, the supernatant was immunoprecipitated overnight at 4°C with 15 ml of anti-Runx1 antibody and protein G. The immune complexes were collected and analyzed for forkhead protein as described above except that protein A-HRP was used as the secondary antibody.
EMSAs-EMSAs were performed following the method of Lewis and Konradi (39). Binding reactions were performed in 10 mM Hepes buffer, pH 7.9, containing 10% glycerol, 0.1 mM EDTA, 5 mM MgCl 2 , 2 mM dithiothreitol, and 1 g of poly(dI-dC) in a 20 l volume. Nuclear extract was added to binding buffer and preincubated on ice for up to 1 h. Radiolabeled probe was then added, and the binding reaction was carried out for 20 min at 30°C. Where indicated, 2 l of supershift antibody or "cold" excess oligonucleotide (200 ng) was added during the preincubation step. Samples were loaded onto 6% nondenaturing acrylamide gels and run at 250 V for 4 -5 h. The gels were dried and subjected to phosphorimaging analysis. The probe used in the EMSA was a 54-bp double-stranded oligonucleotide containing a 3ϫ concatamer of the sequence 5Ј-TCCGG-TAAACACGCCAGG-3Ј from the human Bim promoter. The underlined sequence indicates the consensus forkhead binding site (40). This oligonucleotide was end-labeled with [␥-32 P]ATP using the T4 kinase reaction to generate the EMSA probe. Three unlabeled doublestranded oligonucleotides used as cold competitors in the EMSA binding reactions were as follows: 6X-RBE, a 42-bp oligo containing a 6ϫ concatamer of the consensus Runx binding sequence 5Ј-AAC-CACA-3Ј (41); FasL, a 40-bp oligo from the human FasL promoter containing four forkhead binding sites (underlined) within the sequence 5Ј-TCTCTATTTAAATA-AATAAGTAAATAAATAAACT-GGGCAA-3Ј (42); and a 50-bp sequence taken from the mouse Bim promoter containing the sequence 5Ј-GCTGCCCGCAGGCCAAGA-CACTAGGGTA AACACGCCGG-GGTGGGCGGCGCACGC-3Ј containing one forkhead binding site (the forkhead binding site is underlined).
Chromatin Immunoprecipitation Assays-ChIP and reChIP assays were performed following the method of Shang et al. (43). Briefly, Hep3B cells were seeded into 150-mm culture dishes and transiently transfected with Runx1B overnight. The next day the cells were fixed with formaldehyde, lysed in 500 l of buffer, and sonicated. After centrifugation to clarify the lysates, 200 l of the supernatant was precleared with IgG and then immunoprecipitated overnight at 4°C. The immunoprecipitates were washed at 4°C and eluted twice at room temperature with 100-l aliquots of a freshly made sodium bicarbonate/SDS buffer. In reChIP experiments, the initial Runx ChIP was eluted using 10 mM dithiothreitol at 37°C and then diluted 50ϫ in buffer for subsequent immunoprecipitations. ChIP eluates were heated overnight at 65°C with NaCl to reverse the protein-DNA cross-links and purified on a Qiagen PCR purification column. The column was eluted with 50 l of buffer from which, typically, 2-5 l of eluate was used in each 25-l PCR reaction. PCR was performed using AccuPrime Taq with the supplied Buffer II (Invitrogen). The primers used to amplify the human Bim genomic fragment containing the FOXO3 site were 5Ј-GTAGGTGAGCGGGAGGCTA-3Ј for the top strand and 5Ј-AGGCTCGGACAGGTAAAGG-3Ј for the bottom strand. The annealing temperature of the PCR reaction was 61°C, and the reaction was run for 36 cycles for the FLAG-Runx1 ChIP samples and 38 cycles for the FOXO3 samples. The 155-bp product (Ϫ3135 to Ϫ2980 relative to the initiating ATG) contained the consensus forkhead binding site (Ϫ3055 relative to the initiating ATG). The sequence of this 155-bp product corresponds to the region encompassing nucleotides Ϫ254 to Ϫ96 of the mouse Bim promoter (shown in Fig. 6C). The primers used to amplify a human Bim genomic fragment 1869 bp upstream of the FOXO3 site were 5Ј-TGGCAGAGACA-GAAAGGGACAC-3Ј for the top strand and 5ЈTTTGGGGCA CTAAAAGGAAGC-3Ј for the bottom strand. Identical PCR cycling conditions generated a 229-bp product.

TGF␤ Induces Apoptosis and Increases Bim Expression in
Hepatic Cell Lines-We had previously shown that Bim expression is increased during TGF␤-induced apoptosis in B-cells (15). In the present study we used hepatic cell lines to delineate the molecular pathway(s) regulating Bim expression, as these cells are more amenable to transfection and also undergo apo-ptosis in response to TGF␤ treatment (9 -14, 17). As shown in Fig.  1A, mouse AML-12 and human Hep3B cells undergo apoptosis in response to TGF␤, with rat FAO cells showing little response. Similarly, TGF␤ was also shown to induce Bim expression in AML-12 and Hep3B cells (Fig. 1B, top panel), whereas expression in FAO cells was minimal. It was interesting that TGF␤-induced Bim expression in Hep3B cells was greater than in AML-12 cells, even though AML-12 cells demonstrated much greater apoptosis than Hep3B cells. These results are likely explained by the expression level of the major prosurvival protein in hepatocytes, Bcl-X L . TGF␤ had no effect on Bcl-X L expression in Hep3B and FAO cells, but it dramatically reduced the expression level of Bcl-X L in AML-12 cells (Fig. 1B, middle  panel). Western blotting for Hsp-90 showed equal protein loading of the samples (Fig. 1B, bottom panel). The importance of Bim expression in TGF␤-induced apoptosis in AML-12 cells was demonstrated in knockdown experiments. Knockdown of Bim expression with siRNA decreased TGF␤-induced apoptosis by about 70% (Fig. 1C). Control experiments demonstrated effective knockdown of Bim by Western blotting (Fig. 1C). We next determined whether the expression of Bim was sufficient to promote apoptosis in hepatic cells. AML-12 and Hep3B cells were transiently transfected with increasing amounts of FLAG-tagged Bim for 24 h, after which the cells were collected and analyzed for Bim expression and caspase-3 activation. Overexpression of Bim protein (Fig. 1D, top panel, arrow) resulted in a dose-dependent increase in cleaved, active caspase-3 formation (Fig. 1D, middle panel). Hsp-90 showed equal protein loading (Fig. 1D, bottom panel). Overexpression of increasing concentrations of Bim was also associated with increasing cell death in AML-12 and Hep3B cells, as measured by ELISA (Fig. 1E). Collectively, these results show that TGF␤induced Bim expression is an important mediator of cell death in hepatic cells.
TGF␤ Increases Bim mRNA Levels in Hep3B Cells-To determine whether TGF␤ induces Bim expression at a transcriptional level, we examined the effect of the transcriptional inhibitor actinomycin D on TGF␤-stimulated Bim expression. Hep3B cells were stimulated with TGF␤ for 12 or 18 h, and actinomycin D was added during the last 6 h of another 18-h TGF␤-treated culture. In the absence of inhibitor, Bim expres- sion increased between 12 and 18 h of TGF␤ treatment; however, this induction was blocked in the presence of actinomycin D ( Fig. 2A, top panel). These results suggest that transcription is necessary to observe sustained TGF␤-induced Bim expression. Transcriptional induction of Bim by TGF␤, as analyzed by Northern blot, is shown in Fig. 2B (left panel), demonstrating a time-dependent increase in Bim mRNA levels. The increase in Bim mRNA levels was 4-fold at 24 h when normalized to ␤-actin mRNA levels (Fig. 2B, right panel). The addition of actinomycin D inhibited both basal and TGF␤-stimulated levels of Bim while having little effect on ␤-actin mRNA levels (Fig. 2C). The data of Fig. 2D confirm the transcriptional induction of Bim by TGF␤, demonstrating that a 0.8-kb Bim promoter/luciferase reporter construct was transactivated by TGF␤. TGF␤ treatment increased Bim luciferase activity by ϳ3-4-fold after 24 and 48 h (Fig. 2D). As a control, it was shown that TGF␤ increased the luciferase activity of a canonical TGF␤-responsive promoter, 3TP-luciferase, ϳ10-fold in 24 h. Taken together, these results demonstrate that TGF␤ is a transcriptional activator of Bim expression.
TGF␤ Increases Bim Expression and Promoter Activity in the Presence of FOXO3-The forkhead family of transcription factors, in particular FOXO3, has previously been shown to acti-vate Bim expression (28,29,44). To determine whether FOXO3 was involved in TGF␤-stimulated Bim expression, we performed initial experiments using the phosphatidylinositol 3-kinase inhibitor LY294002. LY294002 is an upstream inhibitor of Akt, blocking its phosphorylation and ability to phosphorylate and inhibit FOXO3 transcriptional activity. As shown in Fig. 3A, when LY294002 is added to Hep3B cells in the absence of TGF␤, there is a dose-dependent increase in the expression of Bim protein.
Interestingly, the induction of Bim in the presence of the inhibitor is further augmented by TGF␤ suggesting that, in addition to FOXO3 activity, other transcriptional modulators are required for TGF␤ induction of Bim. Controls, depicted in Fig. 3B, demonstrate that the 50 M LY294002 concentration is sufficient to inhibit Akt activity, decreasing both endogenous and overexpressed phospho-FOXO3 levels while having no effect on total FOXO3 levels. To directly determine the role of FOXO3 in TGF␤-induced Bim expression, we performed Bim luciferase assays in the absence and presence of overexpressed FOXO3 (Fig. 3C). Our results confirm that wild-type FOXO3 can stimulate Bim promoter activity and, in addition, demonstrate that TGF␤ treatment augments FOXO3 Bim promoter transactivation. Furthermore, the stimulatory effect of TGF␤ on Bim luciferase activity was observed even in the presence of a FOXO3 triple mutant (TM) lacking Akt phosphorylation sites. This latter result suggests that the increased Bim luciferase activity with TGF␤ is not due to an effect on the phosphorylation status of FOXO3. Indeed, Western blot experiments demonstrated that TGF␤ had little effect on the phosphorylation of either endogenous or wild-type overexpressed FOXO3 (Fig. 3D). These data indicate that FOXO3 transcriptional activity is not altered during TGF␤-induced Bim expression.
Runx1 Is Specifically Induced by TGF␤ and Activates Bim Luciferase-The results presented indicate that in addition to FOXO3, TGF␤ induction of Bim may require other transcriptional modulators (Fig. 3). Previously, it has been reported that Runx3 is a TGF␤-sensitive transcriptional activator of Bim in gastric epithelial cells (45). However, our results from a recent gene array experiment did not show any basal or TGF␤-stimulated Runx3 expression in AML12 cells (Fig. 4A and Ref. 27). We confirmed these negative results by RT-PCR (data not shown). The array did demonstrate, however, both basal and TGF␤-stimulated Runx1 expression in AML12 hepatocytes, which we confirmed by both RT-PCR (data not shown) and Northern blotting (Fig. 4B). As depicted, multiple Runx1 transcripts in AML12 cells were observed; however, only the largest Runx1 transcript was up-regulated (Fig. 4B, arrow). The presence of multiple Runx1 mRNA transcripts was expected, as the Runx1 message is known to have many alternative spliced variants (46,47). As shown in Fig. 4, TGF␤ also increased both Runx1 mRNA levels (Fig. 4C) as well as Runx1 protein levels (Fig. 4D, top panel) in Hep3B cells. We also showed that TGF␤ increased Runx1 protein levels in primary B-cells (Fig. 4D, middle panel), which also undergo TGF␤-induced apoptosis (27). We next determined the ability of overexpressed Runx1 to stimulate Bim luciferase activity. Two different expression con-structs were employed: a wild-type full-length construct, Runx1B, and a splice variant lacking a C-terminal transcriptional activation domain, termed Runx1A (35). When these two constructs were overexpressed in Hep3B cells, Runx1B showed a dose-dependent stimulation of Bim luciferase activity, whereas Runx1A had no effect (Fig. 4E). In agreement with this observation, we demonstrated by Western blotting that overexpression of Runx1B, but not Runx1A, was also able to induce a dose-dependent increase in endogenous Bim expression in Hep3B cells (Fig. 4F, lower panel) even though Runx1A was expressed at higher levels than Runx1B (Fig. 4F, upper panel). These results suggest that TGF␤ induces Runx1 and that Runx1 serves as a transcriptional mediator of Bim induction.
TGF␤ Specifically Induces IRES-activated Runx1 mRNA Transcripts-Runx1 transcription can produce both capdependent mRNA transcripts and IRES-dependent mRNA transcripts, depending on which promoter region in the Runx1 gene is activated (48,49). Therefore, it was of interest to determine the type of Runx1 mRNA transcripts up-regulated by TGF␤ (shown in Fig. 4B). PCR was used to prepare probes for Northern blotting that specifically recognized the 5Ј-UTR region of cap-and IRES-dependent mRNA transcripts. The results of Northern analyses using these probes demonstrated that TGF␤ had little effect on the expression level of cap-dependent mRNA transcripts (Fig. 5A, left panel), whereas TGF␤ markedly increased the expression level of IRES-dependent mRNA transcripts (Fig. 5A, middle panel). Both cap-and IRESspecific Northern probes identified multiple Runx1 mRNA transcripts. Normalization of a representative cap-and IRES-mRNA transcript (Fig. 5A, left and middle panels, arrows) to that of the corresponding ␤-actin mRNA transcript (Fig. 5A, right panel) confirmed that only IRES-specific Runx1 mRNA transcripts were increased by TGF␤ (Fig. 5B).
If TGF␤ specifically induces only IRES-dependent Runx1 mRNA transcripts, then the observed increase in Runx1 protein following TGF␤ stimulation (Fig. 4D) must be IRES transcriptdependent. To address this question, we utilized the ability of the drug rapamycin to inhibit cap-dependent translation (50,51). In Hep3B cells, we demonstrate that 20 nM rapamycin produced a time-dependent decrease in protein synthesis, which was reduced to 37% of control levels after 24 h of treatment (Fig.  5C). We therefore treated cells with 20 nM rapamycin for 24 h prior to stimulation with TGF␤ for various times and examined Runx1 protein induction by TGF␤. The results (Fig. 5D) demonstrate that even in the presence of rapamycin, TGF␤ induces, in a time-dependent manner, Runx1 protein expression. Control blots demonstrated that rapamycin inhibited cytosolic phospho-p70 S6K levels (Fig. 5D, third panel), indicating that mTOR and therefore cap-dependent protein translation was inhibited by rapamycin. Blotting for nuclear GATA6 and cytosolic Hsp-90 demonstrated equal loading (Fig. 5D, second and  fourth panels). These results support the idea that TGF␤ stimulates Runx1 expression through an IRES-dependent pathway.
Runx1 Is a Co-activator of the Bim Promoter with FOXO3-To determine whether Runx1 was providing the additional activity required for TGF␤ induction, we performed Bim luciferase assays in the presence or absence of co-transfected FOXO3 and Runx1. Overexpression of FOXO3 and Runx1 individually increased Bim promoter transactivation, which was further augmented with TGF␤ treatment (Fig. 6A). When FOXO3 and Runx1 were co-transfected, Bim promoter transactivation was increased compared with when the two were transfected individually. More importantly, Bim promoter transactivation by the combination of FOXO3 and Runx1 was not increased upon TGF␤ treatment, suggesting that both FOXO3 and Runx1 mediate TGF␤ transactivation of the Bim promoter. To determine the cis-acting elements in the Bim promoter that mediate TGF␤ induction and to further demonstrate the role of FOXO3 and Runx1 in this response, we performed deletion and mutation analyses of the 0.8-kb Bim promoter (Fig. 6C, Ϫ699 to ϩ96). Three truncation mutants were produced by stepwise removal of ϳ200 bp from the 5Ј-end of the 0.8-kb Bim promoter sequence to generate Bim luciferase constructs of ϳ0.6 kb (Ϫ480 to ϩ96), 0.4 kb (Ϫ290 to ϩ96), and 0.2 kb (Ϫ120 to ϩ96; see Fig. 6C). The results demonstrate that all of the constructs responded to Runx1 and FOXO3, transfected individually or in combination, except for the 0.2-kb Bim luciferase (Fig. 6B). Therefore, FOXO3 and Runx1 must act through a binding site(s) located within the first 200 bp of the 5Ј-end of the 0.4-kb Bim promoter sequence. Interestingly, bioinformatic analysis of this 200-bp region (MatInspector, Genomatix, Munich, Germany) identified no Runx binding site and a single FOXO binding site (position Ϫ174 to Ϫ168 relative to the transcription start site) that was also conserved in the human Bim promoter (Fig. 6C). The program also identified other FOXO binding sites as well as a single Runx binding site upstream in the full-length 0.8-kb sequence of the mouse Bim promoter. However, these binding sites were not conserved in the human sequence.
Earlier reports have demonstrated that FOXO3 and Runx3 bind to each other (52). We therefore reasoned that because there are no Runx binding sites in the 0.4-kb Bim luciferase construct, and yet this construct responds to overexpressed Runx, that Runx1 may mediate its effects through its potential interaction with FOXO3. To test this hypothesis, we mutated the FOXO3 binding site in the 0.4-kb Bim luciferase construct (see Fig. 6C) and examined its response to overexpressed Runx1 and FOXO3. The results demonstrated that the mutated 0.4-kb Bim luciferase construct (0.4 M) was much less responsive to Runx1, as well as to FOXO3, compared with the wild-type 0.4-or 0.8-kb Bim luciferase constructs (Fig.  6D). These results support the idea that Runx1 is a co-activator of Bim transcription through its interaction with FOXO3 at the forkhead binding site.
Runx1 Binds the Bim Promoter through FOXO3 Binding Sites-We next tested the ability of Runx1 to bind DNA through FOXO3 binding sites. COS cells were transfected with HA-FOXO3 in the absence and presence of increasing concentrations of FLAG-Runx1. EMSA analysis was used to demonstrate binding of FLAG-Runx1 to a 32 P-labeled oligonucleotide probe containing a 3ϫ concatamerized FOXO3 binding site from the human Bim promoter. Overexpression of FOXO3 led to the appearance of a unique band that was not apparent in control, GFP-transfected cells (Fig. 7A, left panel, arrow). Simultaneous expression of increasing amounts of Runx1 resulted in a dose- dependent disappearance of this band, consistent with the idea that Runx1 binds to FOXO3 and supershifts the band. The levels of overexpressed HA-FOXO3 and FLAG-Runx1 are shown in Fig. 7B. That this band contained FOXO3 was proven true in supershift experiments (Fig. 7A, left panel) demonstrating that anti-FOXO3 and anti-HA tag antibodies were able to supershift the band, whereas anti-IgG antibodies were not. Further experiments, shown in Fig. 7A (right panel), demonstrate that the appearance of this band could be blocked by competition with three different unlabeled oligonucleotide probes, each containing forkhead binding sites (FKHD contains 3ϫ FHBE, FasL promoter contains 4ϫ FHBE, and the 50-bp Bim promoter region contains 1ϫ FHBE) but not by an unlabeled oligonucleotide containing concatamerized Runx binding sites (6ϫ Runx). These data support the idea that Runx1 binds to the Bim promoter via forkhead binding sites.
To verify that Runx1 was able to bind to forkhead sites in the endogenous Bim promoter, ChIP assays were performed following overexpression of FLAG-Runx1 in Hep3B cells. As shown (Fig. 7C, top panel), anti-FLAG antibody was able to immunoprecipitate DNA containing the same forkhead binding site in the human Bim promoter as identified previously in the mouse Bim promoter (Fig. 6C). Control experiments dem-onstrated that little PCR product was amplified from an upstream site in the Bim promoter that contained no FOXO3 or Runx1 sites by bioinformatic analysis. When anti-FOXO3 antibody was used for immunoprecipitations, FOXO3 was also shown to bind to the same forkhead site of the human Bim promoter and not to the upstream site (Fig. 7C, middle panel). The binding of Runx1 and endogenous FOXO3 to the forkhead site was unchanged in the presence of TGF␤ (data not shown). Control experiments demonstrated that anti-IgG was not able to immunoprecipitate DNA containing the forkhead binding site (data not shown) and that the input DNA was similar for all samples (Fig. 7C, bottom panel). The association of Runx1 and FOXO3 on the forkhead site in the Bim promoter was confirmed in reChIP experiments (Fig. 7D). Taken together, these data support a model in which TGF␤ stimulates Bim transcription by up-regulating Runx1 expression, which binds FOXO3 residing at its forkhead binding site and activates Bim promoter activity.

Runx1 and FOXO3 Interact and Are Required for Endogenous Bim
Expression-To directly test whether Runx1 was serving as a Bim transcriptional co-activator by binding FOXO3, we performed coimmunoprecipitation assays. As shown in Fig. 8A (top panel), when COS cells were transfected with an equal amount of HA-FOXO3 and increasing amounts of FLAG-Runx1, there was a dose-dependent increase in co-precipitated FOXO3 in the anti-FLAG (Runx1) immunoprecipitates. Control blots demonstrated increasing concentrations of transfected FLAG-Runx1 (Fig. 8A, second panel) and equal expression of FOXO3 and equal loading using anti-Hsp-90 (Fig. 8A, bottom two panels).
To determine whether endogenous Runx1 and FOXO3 bind, AML-12 cells were treated in the absence and presence of TGF␤, and Runx1 was immunoprecipitated from nuclear extracts. As shown in Fig. 8B (top panel), there was a time-dependent increase in co-precipitated FOXO3 in the anti-Runx1 immunoprecipitates. Control blots demonstrated increasing concentrations of TGF␤-induced Runx1 expression (Fig. 8B,  second panel) and equal expression of FOXO3 (Fig. 8B, third  panel). Equal loading was demonstrated using anti-GATA6 (Fig. 8B, bottom panel). Taken together, these results support a model in which Runx1 activation of Bim transcription requires binding to FOXO3.
If Runx1 and FOXO3 are required for endogenous Bim expression, then knockdown of these two transcription factors should decrease Bim expression. AML-12 cells were transfected with siRNA to Runx1, FOXO3, or Bim for 24 h and then treated with TGF␤ overnight in the continued presence of the siRNAs. Either whole cell lysates (Fig. 8C, left panel) or RNA extracts (Fig. 8C, right panel) were prepared and analyzed by Western blotting and RT-PCR, respectively. The results demonstrate that siRNA knockdown of Runx1 or FOXO3 decreases Bim expression in response to TGF␤ (Fig. 8C, top left panel). Knockdown of Bim served as a positive control. Western blot-ting of whole cell lysates for FOXO3 (Fig. 8C, middle left panel) and RT-PCR analysis of Runx1 mRNA transcript levels (Fig. 8C, right panel) demonstrated effective knockdown of target proteins and no off-target effects. Hsp-90 served as a loading control for whole cell lysates (Fig. 8C, bottom left panel). These results support a role for Runx1 and FOXO3 in TGF␤-induced Bim expression.

DISCUSSION
The role of Bim in mediating growth factor withdrawal or stress-induced apoptosis is well documented (30,31). Further, since our initial observation that TGF␤ induces Bim expression to promote apoptosis in B-lymphocytes (15), numerous studies have confirmed the role of Bim in TGF␤-mediated apoptosis in many cell lines, including hepatocytes (8,17). Although the studies in hepatocytes did not directly examine the TGF␤-mediated transcriptional regulation of Bim promoter activity, they did show that knockdown of Smad3 and Smad4 expression by RNA interference inhibited both TGF␤-induced Bim protein and mRNA expression. These results in hepatic cells are consistent with our earlier observations that overexpression of dominant-negative Smad2 and Smad3 inhibits, and overexpression of Smad3 stimulates, TGF␤-induced Bim expression and apoptosis in B-lymphocytes (15,53). It was surprising, therefore, that we were unable to demonstrate any stimulation by Smad2, Smad3, or Smad4 on Bim luciferase activity in Hep3B cells (data not shown). Although bioinformatic analysis identified a Smad binding element immediately upstream of the forkhead binding site in the murine Bim promoter (see Fig. 6C), it was not conserved in the human promoter sequence. Certainly part of the Smad effect on Bim expression is post-translational, as we have recently demonstrated rapid, Smad3-dependent up-regulation of DUSP4/MKP2 phosphatase levels by TGF␤ in AML-12 cells and B-lymphocytes, resulting in inhibition of ERK activity and less phosphorylation, ubiquitination, and degradation of Bim (27).
In the present study, we have demonstrated a direct transcriptional activation of the murine Bim promoter by TGF␤, providing evidence that it is due to the induction of Runx1. Runx proteins are a family of transcription factors that share a Runt homology domain (RHD) and play important roles in development and cancer. Knock-out studies in mice have shown that Runx1 is vital for hematopoiesis and neural development, Runx2 for bone formation, and Runx3 for gastric epithelial cell proliferation (54). Previous studies have demonstrated that TGF␤ induces the expression of Runx2 in C2C12 mesenchymal stem cells (55) and Runx3 (AML2) in I.29m B-cells (56). This is the initial observation that TGF␤ induces Runx1. Runx proteins have previously been implicated in mediating growth arrest and apoptosis of TGF␤ family members. For example, in the Saos2 osteosarcoma cell line, bone morphogenic protein induces Runx2 expression, which is a direct transcriptional activator of Bax, increasing its expression and sensitizing the cells to apoptosis (57). Overexpression of Runx3 restored TGF␤-induced p21 expression and growth arrest, but not apoptosis, in a Runx3-null biliary tract cancer cell line (58) and in stomach epithelial cells (59). Gastric cancer cells expressing a dominant-negative form of Runx3 or antisense Runx3 were resist- A, EMSAs of Runx1 binding to forkhead binding sites. COS cells were transfected with a constant amount of HA-TM-FOXO3 and increasing amounts of FLAG-Runx1B. Nuclear extracts were prepared and preincubated with supershifting antibodies (left panel) or excess, unlabeled oligonucleotides (right panel) as indicated, before the addition of the 32 P-labeled 3ϫ-forkhead probe. The probe lane contains no nuclear extract but all other components of the EMSA buffer. Arrows indicate bands specifically labeled in the presence of overexpressed FOXO3. B, control Western blots showing the expression of HA-TM FOXO3 and FLAG-Runx1B in the nuclear extracts used in the EMSAs described in A. C, Hep3B cells were transfected with a low (L, 10 g) or high (H, 30 g) dose of FLAG-Runx1B. Cross-linked lysates were prepared, and equal amounts were immunoprecipitated with either anti-FLAG or anti-FOXO3 antibody. PCR analyses were preformed on the purified ChIP extracts to amplify a band containing the FHBE in the human Bim promoter (Fox site) or an upstream site lacking a forkhead/Runx site (Upstream site). PCR was also performed on the input DNA samples obtained before ChIP analyses. D, reChIP experiments were also performed on anti-FLAG ChIP samples. FLAG-ChIP samples were immunoprecipitated with either anti-IgG or anti-FOXO3, and the FOX site was amplified. Cont., control.
ant to TGF␤-induced apoptosis and, more importantly, were also deficient in TGF␤-induced Bim expression (45). Surprisingly, TGF␤ had no effect on Runx3 expression levels in gastric cancer cells but did promote its nuclear localization.
We were unable to detect mRNA transcripts for Runx2 or Runx3 by PCR in either Hep3B or AML-12 cells even after TGF␤ treatment (data not shown). These negative results confirmed our gene array analysis in AML-12 cells (Fig. 4A). However, we readily detected basal levels of Runx1 mRNA in both AML-12 and Hep3B hepatocytes, and its levels were induced by TGF␤ (Fig. 4, B and C). These results imply that both Runx1 and Runx3 isoforms are cellspecific mediators of TGF␤-induced Bim expression to elicit cell death. This idea is supported by the finding that both Runx1 and Runx3, but not Runx2, have been shown to be frequently down-regulated, along with their co-activator core-binding factor B, in gastric and hepatocellular cancers (60,61).
Our finding that Runx1 activates the Bim promoter in cooperation with FOXO3 is not surprising given that Runx proteins have been shown previously to serve as transcriptional co-activators with other proteins, including Smads (41,54). The forkhead site at Ϫ174 to Ϫ168, which we identified as being important for Runx1/FOXO3 activation, has also been shown to be important for activation of the rat Bim promoter (28). Other reports showing direct activation of the Bim promoter by FOXO3 have not identified the exact site(s) (29,44). Runx3 has also been shown to directly activate the Bim promoter, although at widely divergent sites. In mouse embryonic fibroblasts, the RBE located from Ϫ609 to Ϫ603 of the mouse promoter (see Fig. 6C) was shown to be activated by Runx3 (52), whereas in gastric cancer cells, two RBEs were identified in the human Bim promoter but were located far upstream of the mouse RBE (45). Interestingly, we always detected a decreased stimulatory response to Runx1 and FOXO3 in our 0.4-kb Bim luciferase construct when compared with the 0.8-kb construct (see Fig. 6, B and D). Thus, the Runx and forkhead binding elements at the 5Ј-end of the mouse Bim promoter may be active in our hepatic cells as well, albeit to a lesser extent than in mouse embryonic fibroblasts.
All Runx proteins are similar in that transcription can be initiated from two different promoters, designated the distal (P1) and proximal (P2) promoters (48). The Runx1 protein produced by the P1 promoter is translated from a cap-dependent mRNA transcript and encodes a longer N-terminal amino acid sequence (19 -32 amino acids) relative to that produced from the P2 promoter (5 amino acids). More importantly, the Runx1 protein produced from the P2 promoter is translated from an IRES-dependent mRNA transcript (62). The diversity of Runx1 proteins expressed by cells is further complicated because of alternative splicing of the primary mRNA transcripts and post-translational modification of the proteins themselves by phosphorylation, acetylation, and ubiquitination (48,49). Our results indicate that TGF␤ specifically induces only IRES-activated Runx1 mRNA transcripts and protein. Although many Runx1 mRNA transcripts could be detected by FIGURE 8. Runx1 and FOXO3 interact and are required for endogenous Bim expression. A, COS cells were transfected with a constant amount of HA-TM-FOXO3 and increasing amounts of FLAG-Runx1B. Runx1-bound protein complexes were immunoprecipitated from whole cell lysates with anti-FLAG antibody and Western blotted for total FOXO3 protein. Lysates were Western blotted for FLAG-Runx1 and HA-FOXO3 expression, as well as Hsp-90 as a loading control. B, AML-12 cells were treated with TGF␤ for the indicated times, and nuclear extracts were prepared. Runx1-bound protein complexes were immunoprecipitated with anti-Runx1 antibody and Western blotted for total FOXO3 protein. Nuclear extracts were Western blotted for Runx1 and FOXO3 expression, as well as GATA6 as a loading control. C, AML-12 cells were transfected with siRNA to Runx1, FOXO3, or Bim for 24 h and then treated with TGF␤ overnight. Either whole cell lysates (left panel) or RNA extracts (right panel) were prepared and analyzed by Western blotting and RT-PCR, respectively. The control (con.) samples received non-targeting siRNAs. In the right panel, RT-PCR was used to amplify Runx1 mRNA with ␤-actin serving as a loading control.
Northern blotting using a probe against the common 3Ј-UTR of Runx1, TGF␤ induction was most apparent for the largest transcript (Fig. 4B). The size of this TGF␤-induced Runx1 mRNA was similar to the largest and most abundant transcript identified with the IRES-specific probe and much greater than any transcript identified with the cap-specific probe (Fig. 5A). As further proof of their identity, the time course of TGF␤ induction of the largest Runx1 mRNA transcript was also similar using either the common 3Ј-UTR or the IRES-specific probes. At present, it is unclear whether the presence of multiple TGF␤-induced Runx1 mRNA transcripts explains the multiple TGF␤-induced Runx1 protein bands identified on Western blots (Figs. 4D and 5D) or if this is due to post-translational modification of the Runx1 protein. Although Runx1 protein modification could be either TGF␤-dependent or -independent, there are reports that TGF␤ stimulates Runx3 acetylation, which prevents its ubiquitination and degradation (63).
Previous studies have demonstrated that the Runx1 5Ј-UTR, used in our study as an IRES-specific probe, does in fact direct cap-independent, IRES-initiated mRNA translation (62). In addition, a gene array study demonstrated that the Runx1 mRNA translation state, a measure of ribosome-bound mRNA, was actually increased in the presence of rapamycin over time in U87 and LAPC-4 cell lines (50). This result is opposite to that expected for a cap-dependent mRNA transcript. Thus, these two independent studies support our claim that the Runx1 mRNA transcript(s) induced by TGF␤ in hepatic cells is indeed IRES-dependent. This is consistent with the hypothesis that during apoptosis there must be a switch from cap-dependent to cap-independent translation because of caspase cleavage of the cap-dependent translation machinery (64). Indeed, many IRESdependent eukaryotic mRNAs have been identified that encode proteins associated with cell proliferation and cell death, such as c-Myc and Apaf-1, respectively (64,65).
Taken together, our data support a model in which TGF␤ stimulates Bim transcription by up-regulating Runx1 expression, which binds FOXO3, residing at its forkhead binding site, and activates Bim promoter activity and expression. Of particular interest is our finding that TGF␤ specifically induces IRESdependent expression of Runx1. Future studies are needed to determine whether the target of TGF␤-induced Runx activation, Bim, is also translated from IRES-dependent mRNA transcripts. Moreover, it will also be of interest to determine whether IRES-dependent mRNA translation is a common feature of TGF␤-induced apoptosis.