Introduction

Transforming growth factor-β (TGF-β) mediates diverse processes, such as proliferation, differentiation and apoptosis, during embryogenesis and in adult tissues (Siegel and Massagué, 2003; Massagué, 2008). On binding of TGF-β, the TGF-β type II receptor phosphorylates and activates the TGF-β type I receptor. In turn, TGF-β type I receptor phosphorylates Smad2 and Smad3, which translocate to the nucleus and form complexes with Smad4 and other DNA-binding factors to activate or repress transcription (Shi and Massagué, 2003; Feng and Derynck, 2005). In most types of normal cells, TGF-β inhibits cell proliferation by inducing arrest in G1 phase. The gene responses that are central to the TGF-β cytostatic program are activation of the cyclin-dependent kinase inhibitors p15Ink4B and p21WAF1/Cip1, and repression of the growth-promoting c-myc and ID transcription factors (Reynisdóttir et al., 1995; Chen et al., 2002; Kang et al., 2003). The cytostatic program is tightly integrated by feedback loops that protect against competing mitogenic signals (Siegel and Massagué, 2003). For example, TGF-β activates p15Ink4B and p21WAF1/Cip1 transcription via cooperative interactions between Sp1 and Smad proteins, and also prevents repression of these genes by c-myc (Feng et al., 2000, 2002; Pardali et al., 2000; Gartel et al., 2001).

Resistance to TGF-β-mediated growth inhibition is an important feature in the pathogenesis of a wide variety of tumors (Siegel and Massagué, 2003; Massagué, 2008). In several types of tumors, this resistance has been attributed to mutations or deletion of core components of the TGF-β signaling pathway. Inactivating mutations in the TGF-β type II receptor have been reported in 60–90% of colon cancers associated with microsatellite instability (Markowitz et al., 1995; Watanabe et al., 2001). Mutations or deletions of Smad4 occur in ∼50% of pancreatic and non-microsatellite instability colorectal cancers (Hahn et al., 1996; Woodford-Richens et al., 2001). By contrast, TGF-β type II receptor mutations have not been detected in lung cancers with microsatellite instability (Takenoshita et al., 1997), and the frequency of Smad4 mutations in lung cancers is low (7%) (Nagatake et al., 1996). Similarly, many breast and prostate cancers are resistant to TGF-β-mediated growth inhibition, but rarely contain TGF-β type II receptor or Smad mutations (Schutte et al., 1996; Takenoshita et al., 1998; Bello-DeOcampo and Tindall, 2003). TGF-β receptor mutations occur in 12–31% of ovarian cancers (Wang et al., 2000; Francis-Thickpenny et al., 2001), but many TGF-β-resistant ovarian cancers have been found to express functional receptors (Yamada et al., 1999). Resistance of many tumors to TGF-β-mediated growth inhibition therefore cannot solely stem from mutations or deletions of core components of the TGF-β signaling pathway and likely arise from other aberrations.

Homeobox genes encode transcription factors that control cell differentiation (McGinnis and Krumlauf, 1992). DLX4 (also reported as BP1) (Haga et al., 2000) is a member of the DLX family of homeobox genes that controls many aspects of embryonic development, including bone morphogenesis and skeletal patterning (Panganiban and Rubenstein, 2002). Although absent from most normal adult tissues, DLX4 is widely expressed in leukemias, lung, breast, ovarian and prostate cancers (Haga et al., 2000; Man et al., 2005; Hara et al., 2007; Tomida et al., 2007; Schwartz et al., 2009). Because DLX4 is expressed in diverse types of tumors, we considered the possibility that DLX4 controls a pathogenic mechanism common to multiple organ sites. Our study demonstrates that DLX4 blocks gene responses that are central to the TGF-β cytostatic program by sequestering Smad4 and Sp1, and by inducing c-myc. This blocking activity of DLX4 could explain, in part, why tumor cells are resistant to the antiproliferative effect of TGF-β.

Results

DLX4 blocks TGF-β-induced, Smad-dependent responses

To determine whether DLX4 blocks the antiproliferative effect of TGF-β, we initially used the non-tumorigenic lung epithelial cell line Mv1Lu. Mv1Lu is a well-established model for studying TGF-β-mediated induction of p15Ink4B expression and growth arrest (Reynisdóttir et al., 1995; Feng et al., 2002). Whereas growth of Mv1Lu cells was inhibited by TGF-β in a dose-dependent manner, enforced expression of DLX4 in these cells increased resistance to TGF-β (Figure 1a). Expression of DLX4 in Mv1Lu cells blocked the ability of TGF-β to induce p15Ink4B expression (Figure 1b). Cell cycle analysis revealed that DLX4 expression inhibited the induction of G1 arrest in Mv1Lu cells by TGF-β (Figure 1c). Enforced expression of DLX4 in the non-tumorigenic mammary epithelial cell line NMuMG also inhibited TGF-β-induced G1 arrest (Supplementary Figure 1). In converse experiments, we determined whether knockdown of DLX4 in tumor cells increased TGF-β-mediated growth inhibition using short hairpin RNAs that targeted different sites of DLX4 (sh90 and sh92). The ability of short hairpin RNAs to knockdown DLX4 in MCF-7 breast cancer cells was confirmed by western blot (Figure 1d) and also by immunofluorescence staining and real-time quantitative PCR (Supplementary Figures 2A and B). Knockdown of DLX4 in MCF-7 cells was observed to increase sensitivity to TGF-β in cell viability assays (Figure 1e), and also increased the proportion of cells in G1 phase (Figure 1f).

Figure 1
figure 1

DLX4 blocks TGF-β-mediated growth inhibition. (a) Vector-control (−DLX4) and +DLX4 stable Mv1Lu lines were cultured with the indicated concentrations of TGF-β for 2 days. Changes in cell growth were determined by MTT assay, and expressed relative to growth of cells incubated without TGF-β. Shown are results of two independent experiments each performed in triplicate. (b) Western blot analysis of Mv1Lu lines following treatment without and with TGF-β (10 ng/ml) for 16 h. (c) Mv1Lu lines were treated without and with TGF-β for 18 h. Indicated are the proportions of cells in G1, S and G2/M phases determined by flow cytometric analysis of propidium iodide staining. (d) MCF-7 cells were transfected with empty vector, non-targeting short hairpin RNA (shRNA) and DLX4 shRNAs (sh90 and sh92). At 2 days after transfection, DLX4 levels were assayed by western blot. (e) Transfected MCF-7 cells were cultured with the indicated concentrations of TGF-β for 2 days. Changes in cell growth were determined by MTT assay. (f) Transfected MCF-7 cells were treated without and with TGF-β (10 ng/ml) for 18 h and proportions of cells in cell cycle phases determined thereafter. (g) Vector-control (−DLX4) and +DLX4 stable MDA-MB-468 lines were transfected with Smad4. At 24 h thereafter, cells were cultured without and with TGF-β (10 ng/ml) for 2 days, and changes in cell growth were examined by MTT assay. (h) Western blot analysis of MDA-MB-468 lines following treatment without and with TGF-β for 16 h.

Full size image

Because TGF-β can also inhibit growth by Smad-independent mechanisms (Petritsch et al., 2000), we used the Smad4-deficient MDA-MB-468 breast cancer cell line to confirm that DLX4 blocks Smad-dependent growth inhibition. Growth of MDA-MB-468 cells was not inhibited by TGF-β (Figure 1g). On the other hand, reconstitution of Smad4 in these cells increased responsiveness to TGF-β (Figure 1g). This responsiveness to TGF-β was abrogated by DLX4 (Figure 1g). Expression of p21WAF1/Cip1 was induced by TGF-β in MDA-MB-468 cells reconstituted with Smad4, and this induction was blocked by DLX4 (Figure 1h). Enforced expression of DLX4 induced c-myc levels, irrespective of the absence or presence of Smad4 (Figure 1h). We further investigated the effect of DLX4 on c-myc induction by assaying c-myc promoter activity. Activity of the c-myc promoter was repressed by TGF-β in control Mv1Lu cells (Figure 2a). In contrast, expression of DLX4 in Mv1Lu cells induced c-myc promoter activity irrespective of TGF-β signaling (Figure 2a). Conversely, knockdown of DLX4 in MCF-7 cells inhibited c-myc promoter activity and increased sensitivity to TGF-β-mediated repression (Figure 2b). These results suggest that DLX4 blocks TGF-β/Smad-dependent repression of c-myc expression, and also induces c-myc levels independently of TGF-β/Smad signaling.

Figure 2
figure 2

DLX4 induces c-myc promoter activity and inhibits TGF-β-mediated induction of p15Ink4B promoter activity. (a) Mv1Lu cells were cotransfected with empty vector (gray bar) or DLX4 (black bar), together with empty pBV-Luc vector or with pBV-MYC(Del4) reporter plasmid that contains 900 bp of c-myc P1 and P2 promoter sequences. Transfected cells were cultured without and with TGF-β (10 ng/ml) for 18 h, and assayed for firefly luciferase (F-Luc) activity. (b) Reporter assays for c-myc promoter activity were likewise conducted using MCF-7 cells that were cotransfected with non-targeting short hairpin RNA (shRNA; gray bar) and DLX4 (sh90) shRNA (black bar). (c) Mv1Lu cells were cotransfected with empty vector (gray bar) or DLX4 (black bar), together with reporter plasmids containing no promoter (pGL2 vector), p15Ink4B promoter sequences (−113 to +70; p15-WT) and p15Ink4B promoter with mutated SBEs (p15-SBE-mt). Cells were cultured without and with TGF-β for 18 h, and assayed for F-Luc activity. (d) Reporter assays using the p15-WT reporter plasmid were performed using transfected MDA-MB-468 lines. Shown are relative F-Luc activities in three independent experiments, each performed in duplicate. Values were normalized by activity of co-transfected Renilla luciferase.

Full size image

The first 113 bp of the p15Ink4B promoter are essential for induction by TGF-β and contain two Smad-binding elements (SBEs; Feng et al., 2000). Activity of this minimal promoter region was induced by TGF-β in control Mv1Lu cells, and this induction was abolished by mutation of the SBEs (Figure 2c). Expression of DLX4 in Mv1Lu cells abolished the induction of wild-type p15Ink4B promoter activity by TGF-β (Figure 2c). DLX4 also modestly inhibited activity of the SBE-mutant promoter (Figure 2c). To confirm that DLX4 blocks Smad-dependent transcription, we performed reporter assays using MDA-MB-468 cells. Although wild-type p15Ink4B promoter activity was unresponsive to TGF-β in Smad4-deficient MDA-MB-468 cells, responsiveness was conferred when Smad4 was expressed in these cells. This Smad4-dependent responsiveness was eliminated when DLX4 was coexpressed (Figure 2d).

In addition to its antiproliferative effect, TGF-β is well known to induce epithelial-to-mesenchymal transition (EMT; Siegel and Massagué, 2003; Massagué, 2008). Because DLX4 abrogated TGF-β-mediated growth inhibition, DLX4 might also block the ability of TGF-β to induce EMT. To address this question, we used the NMuMG cell line, a well-established model for studying TGF-β-induced EMT. Smad4 has been demonstrated to be essential for TGF-β-induced EMT in several cell types including NMuMG cells (Deckers et al., 2006). TGF-β treatment of vector-control NMuMG cells induced profound epithelial-to-fibroblastic morphologic transformation (Supplementary Figure 3A), loss of E-cadherin and upregulation of N-cadherin (Supplementary Figures 3B and C). Enforced expression of DLX4 blocked downregulation of E-cadherin and induction of N-cadherin (Supplementary Figures 3B, C). Epithelial morphology was considerably retained in TGF-β-treated+DLX4 NMuMG cells (Supplementary Figure 3A). Together, our findings that DLX4 blocks TGF-β-mediated, Smad-dependent growth inhibition and EMT indicate that DLX4 inhibits a core component of the TGF-β/Smad signaling pathway.

DLX4 blocks transcriptional activity of TGF-β-activated receptor-regulated Smads

Smad-dependent transcription is controlled at multiple levels, including phosphorylation and nuclear localization of receptor-regulated Smads (R-Smads) (Shi and Massagué, 2003). DLX4 did not alter levels of Smad2 expression or Smad2 phosphorylation (Figure 3a). DLX4 also did not interfere with nuclear translocation of Smad2 following TGF-β stimulation (Figure 3b). DLX4 was predominantly localized in the nucleus, and its localization was not affected by TGF-β stimulation (Supplementary Figures 4A and B). These findings indicate that DLX4 likely inhibits nuclear events downstream of the TGF-β signaling pathway.

Figure 3
figure 3

DLX4 blocks Smad transcriptional activity. Vector-control (−DLX4) and +DLX4 Mv1Lu lines were serum starved overnight and then treated without and with TGF-β (10 ng/ml) for 30 min. (a) Total and phosphorylated Smad2 were detected by western blot. (b) Intracellular localization of Smad2 was detected by immunofluorescence staining. Bar, 20 μm (c and d) Mv1Lu and HepG2 cells were cotransfected with empty vector (gray bar) or DLX4 (black bar), together with GAL4-driven firefly luciferase (F-Luc) reporter plasmid and with (c) GAL4–Smad2 or (d) GAL4–Smad3. Transfected cells were cultured without and with TGF-β for 18 h, and assayed for F-Luc activity. (e) GAL4–Smad2 and (f) GAL4–Smad3 activities were likewise assayed in MCF-7 cells that were cotransfected with non-targeting short hairpin RNA (shRNA; gray bar) or DLX4 (sh90) shRNA (black bar). (g, h) HepG2 cells were cotransfected with empty vector (gray bar) or DLX4 (black bar), together with (g) pSBE4-Luc or (h) BRE-Luc reporter plasmids, and then cultured without and with TGF-β (10 ng/ml) or BMP-4 (80 ng/ml) for 18 h. Shown are relative F-Luc activities in three independent experiments, each performed in duplicate and normalized by activity of cotransfected Renilla luciferase.

Full size image

We determined whether DLX4 inhibits transcriptional activity of TGF-β R-Smads. A construct was generated in which the GAL4 DNA-binding domain (DBD) was fused to the linker region and Mad homology 2 (MH2) transcriptional activation domain of Smad2 (amino acids 173–467). This chimera was coexpressed in Mv1Lu cells with a firefly luciferase reporter controlled by five tandem GAL4-binding sites. Transcriptional activity of GAL4–Smad2 was induced by TGF-β, and this activation was abolished when DLX4 was expressed (Figure 3c). Similar results were obtained using the hepatoma cell line HepG2 (Figure 3c). Expression of DLX4 also abolished TGF-β-mediated activation of a chimera comprising the GAL4–DBD fused to the linker region and MH2 domain of Smad3 (amino acids 133–425) (Figure 3d). Conversely, knockdown of DLX4 in MCF-7 cells increased TGF-β-mediated induction of GAL4–Smad2 and GAL4–Smad3 activity (Figures 3e and f).

Smad2 and Smad3 serve as substrates for the TGF-β receptors, whereas other R-Smads (Smads 1, 5 and 8) are utilized by the bone morphogenetic protein (BMP) and anti-Mullerian receptors (Shi and Massagué, 2003; Feng and Derynck, 2005). We initially tested the ability of DLX4 to inhibit transcription induced by other members of the TGF superfamily by using a synthetic promoter comprising four tandem SBEs (pSBE4-Luc). DLX4 blocked induction of this promoter by TGF-β and by BMP-4 (Figure 3g). TGF-β- and BMP-specific R-Smads preferentially bind distinct DNA sequences (Kusanagi et al., 2000; Korchynskyi and ten Dijke, 2002). Indeed, BMP-4 was not as effective as TGF-β in inducing pSBE4-Luc activity (Figure 3g). We further tested the effect of DLX4 on BMP-induced transcription by using a reporter plasmid that contained the BMP-responsive promoter of the Id1 gene. DLX4 blocked BMP-induced Id1 promoter activity (Supplementary Figure 5). The blocking effect of DLX4 was confirmed by using a synthetic promoter comprising two tandem copies of Id1 BMP response elements (BRE-Luc) (Figure 3h). Because TGF-β- and BMP-specific R-Smads utilize Smad4 as the common and essential partner for formation of functional transcriptional complexes (Shi and Massagué, 2003), our findings raise the possibility that DLX4 inhibits Smad4.

DLX4 prevents Smad4 from binding R-Smads

We subsequently determined whether DLX4 interferes with the binding of Smad4 to R-Smads. Following transfection with FLAG-tagged DLX4 or empty vector, HepG2 cells were treated with or without TGF-β. Smad2 was immunoprecipitated, and precipitates analyzed by immunoblotting using antibody (Ab) to Smad4. Binding of Smad4 to Smad2 was inhibited when DLX4 was expressed (Figure 4a). We confirmed these findings by immunoprecipitating Smad4 and detecting Smad2 in precipitates (Figure 4a). DLX4 also prevented binding of Smad4 to Smad3 (Figure 4a). DLX4 did not alter Smad expression levels (Figure 4b). These results indicate that DLX4 likely inhibits transcriptional activity of Smad2 and Smad3 by preventing Smad4 from interacting with these R-Smads.

Figure 4
figure 4

DLX4 prevents Smad4 from binding R-Smads and blocks interactions of Smad and Sp1 proteins with the p15Ink4B promoter. HepG2 cells were transfected with empty vector or with FLAG-tagged DLX4. At 24 h thereafter, cells were serum starved overnight and then treated without and with TGF-β (10 ng/ml) for 30 min. (a) Smad2 was immunoprecipitated from nuclear extracts, and precipitates analyzed by immunoblotting using Ab to Smad4. Conversely, Smad4 was pulled down, and precipitates analyzed by immunoblotting using Smad2 Ab. Because HepG2 cells express low levels of Smad3, immunoprecipitation (IP) assays to detect binding of Smad3 to Smad4 were performed using extracts of cells that had been transfected with Smad3. (b) Western blot of DLX4, Smad proteins and Sp1 in nuclear extracts. (c) Sp1 was immunoprecipitated from nuclear extracts, and precipitates analyzed by immunoblotting using Ab to Smad4. Conversely, Smad4 was pulled down, and Sp1 detected in precipitates. (d) Sequence of the minimal p15Ink4B promoter indicating Sp1-binding sites and SBEs (adapted from Feng et al., 2000). Underlined are sequences contained in the oligonucleotides used for oligonucleotide pulldown assays (solid line) and gel-shift assays (dashed line). (e) Biotinylated oligonucleotide containing sequences −108 to −39 of the p15Ink4B promoter was incubated with HepG2 nuclear extracts and pulled down. DNA-bound proteins in precipitates were analyzed by immunoblotting. (f) ChIP analysis of interactions of Smad and Sp1 proteins with the p15Ink4B promoter. The input fraction corresponded to 1% of the chromatin solution of each sample before IP.

Full size image

Binding affinity and selectivity of Smad complexes for target gene promoters are governed by interactions with other DNA-binding factors (Shi and Massagué, 2003; Feng and Derynck, 2005). The minimal p15Ink4B promoter contains two Sp1-binding sites adjacent to the SBEs, and its induction by TGF-β involves binding of Smad4 to Sp1 as well as to R-Smads (Feng et al., 2000). As reported by others (Feng et al., 2000; Pardali et al., 2000), TGF-β stimulation induced binding of Smad4 to Sp1 in −DLX4 cells. This binding was not inhibited by DLX4 (Figure 4c). Surprisingly, we observed increased association of Smad4 with Sp1 in +DLX4 cells in the absence of TGF-β stimulation (Figure 4c). DLX4 did not alter expression of Sp1 (Figure 4b). To determine whether DLX4 alters formation of Smad–Sp1–DNA complexes, we performed DNA pulldown assays using a biotinylated oligonucleotide that contained nucleotides −108 to −39 of the p15Ink4B promoter and included the two Sp1 binding sites and SBEs (Figure 4d). Increased levels of Smads were detected in DNA–protein complexes when −DLX4 cells were stimulated with TGF-β (Figure 4e). In contrast, this induction was inhibited in +DLX4 cells (Figure 4e). DLX4 also inhibited interaction of Sp1 with the p15Ink4B promoter (Figure 4e). To confirm the blocking activity of DLX4 in a more physiological context, we performed chromatin immunoprecipitation (ChIP) assays. As shown in Figure 4f, association of R-Smads, Smad4 and Sp1 with the p15Ink4B promoter was detected by ChIP assays in −DLX4 cells following TGF-β treatment. In contrast, interactions of Smad and Sp1 proteins with the p15Ink4B promoter were abrogated in +DLX4 cells (Figure 4f). Together, these findings raise the possibility that DLX4 binds Smads and/or Sp1.

DLX4 directly binds Smad4

To initially investigate whether DLX4 associates with Smad4 in cells, immunoprecipitation assays were performed using extracts of HepG2 cells that expressed FLAG–DLX4. DLX4 associated with Smad4 in the absence and presence of TGF-β stimulation (Figure 5a). Association of endogenous DLX4 with Smad4 was detected in immunoprecipitation assays using extracts of MCF-7 cells (Figure 5b). This interaction was abrogated when DLX4 short hairpin RNA was expressed in these cells (Figure 5b).

Figure 5
figure 5

DLX4 binds to Smad4. (a) HepG2 cells were transfected with empty vector or with FLAG-tagged DLX4. At 24 h thereafter, cells were serum starved overnight and then treated without and with TGF-β (10 ng/ml) for 30 min. FLAG–DLX4 was immunoprecipitated using FLAG Ab, and precipitates analyzed by immunoblotting using Ab to Smad4. Conversely, FLAG–DLX4 was detected in precipitates following immunoprecipitation (IP) using Smad4 Ab. (b) MCF-7 cells were transfected with non-targeting short hairpin RNA (shRNA) or with DLX4 (sh90) shRNA. Endogenous DLX4 was immunoprecipitated using DLX4 Ab, and precipitates analyzed by immunoblotting using Smad4 Ab. IP using mouse immunoglobulin G was included as a negative control. (c) Expression of GST–Smad2 and GST–Smad4 proteins was confirmed by SDS–polyacrylamide gel electrophoresis (left). GST-fusion proteins were assayed for direct binding to in vitro translated 35S-labeled full-length (FL) DLX4 (right). (d) GST–DLX4 constructs comprising the transactivation domain (TA), homeodomain (HD) and C-terminal tail (C), and Smad4 constructs comprising MH1 and MH2 domains and linker (LK) region. (e) FL DLX4 and portions thereof were expressed as GST-fusion proteins (left), and assayed for binding to 35S-labeled FL Smad4 (right). (f) The 35S-labeled FL and truncated Smad4 were translated in vitro (left), and assayed for binding to FL GST–DLX4 protein (right).

Full size image

To determine whether DLX4 interacts with Smad4 by direct binding, we tested the ability of in vitro translated 35S-labeled DLX4 to bind GST–Smad4 protein. GST-pulldown assays demonstrated that DLX4 directly binds Smad4 (Figure 5c). DLX4 also bound Smad2, albeit more weakly (Figure 5c). We sought to identify the Smad4-binding domain of DLX4 by testing truncated GST–DLX4 fusion proteins for their ability to bind in vitro translated 35S-labeled Smad4 (Figure 5d). Deletion of the C-terminal tail of DLX4 only weakly affected its ability to bind Smad4 (Figure 5e). In contrast, deletion of the DNA-binding homeodomain of DLX4 markedly inhibited its Smad4-binding ability. Binding of the DLX4 homeodomain to Smad4 was detected but not as strongly as observed with full-length DLX4 (Figure 5e). We also investigated which domain of Smad4 interacts with DLX4 by testing the ability of GST–DLX4 protein to bind in vitro translated portions of Smad4 protein (Figure 5d). Direct binding of DLX4 was detected to the Smad4 Mad homology 1 (MH1) domain, but not to the MH2 domain (Figure 5f).

DLX4 also directly binds Sp1

Because DLX4 directly binds Smad4, which also binds Sp1, we determined whether DLX4 interacts with Sp1. In immunoprecipitation assays using lysates of transfected HepG2 cells, DLX4 was found to associate with Sp1 (Figure 6a). This association was not dependent on TGF-β stimulation. We confirmed the ability of endogenous DLX4 to interact with Sp1 in immunoprecipitation assays using extracts of MCF-7 cells (Figure 6b). GST–DLX4 protein directly bound to in vitro translated 35S-labeled Sp1 (Figure 6c). Whereas the N-terminal domain of DLX4 did not bind Sp1, binding by the homeodomain was strongly detected (Figure 6c). We also tested the ability of GST–DLX4 protein to directly bind in vitro translated 35S-labeled portions of Sp1 (Figure 6d). DLX4 did not bind the N-terminal domain of Sp1 (amino acids 1–557), but bound to its C-terminal DBD (amino acids 557–778; Figure 6e). In the presence of DLX4, decreased levels of Sp1 were detected in complexes associated with the p15Ink4B promoter in both oligonucleotide pulldown and ChIP assays (Figures 4e and f). In gel-shift assays, we observed that DLX4 did not bind p15Ink4B promoter sequences, but prevented Sp1 from binding the promoter (Figure 6f). These observations are consistent with the ability of DLX4 to partially inhibit activity of the SBE-mutant p15Ink4B promoter that contains intact Sp1-binding sites (Figure 2c). Enforced expression of DLX4 also inhibited activity of a synthetic promoter that comprised tandem Sp1-binding sites (Figure 6g). Conversely, knockdown of DLX4 stimulated Sp1-driven promoter activity (Figure 6g). These results indicate that DLX4 inhibits p15Ink4B transcription by sequestering Sp1, in addition to preventing Smad4 from interacting with R-Smads.

Figure 6
figure 6

DLX4 binds to the DBD of Sp1. (a) Lysates were prepared from HepG2 cells as described in Figure 5a. Interaction of FLAG–DLX4 with Sp1 was detected by reciprocal immunoprecipitation (IP) using FLAG and Sp1 Abs. (b) MCF-7 cells were transfected with non-targeting short hairpin RNA (shRNA) or with DLX4 (sh90) shRNA. Endogenous DLX4 was immunoprecipitated using DLX4 Ab, and precipitates analyzed by immunoblotting using Sp1 Ab. IP using mouse immunoglobulin G was included as a negative control. (c) GST–DLX4 proteins (described in Figure 5d) were assayed for binding to 35S-labeled full-length (FL) Sp1. (d) Sp1 constructs comprising the transactivation domain (TA) and DBD. (e) The 35S-labeled FL and truncated Sp1 were translated in vitro (left), and assayed for binding to FL GST–DLX4 protein (right). (f) Gel-shift analysis using a 32P-labeled oligonucleotide containing nucleotides −88 to −64 of the p15Ink4B promoter (refer Figure 4d). Recombinant Sp1 protein was incubated with increasing amounts of in vitro translated FLAG–DLX4 and FLAG-tag. Gel-shifted DNA-bound Sp1 is indicated. (g) MDA-MB-468 cells were cotransfected with empty vector (−DLX4) or DLX4, together with the Cignal Sp1 reporter construct driven by a synthetic promoter comprising tandem Sp1-binding sites (left). Sp1-driven promoter activity was likewise assayed in MCF-7 cells that were cotransfected with non-targeting shRNA or DLX4 (sh90) shRNA (right). Shown are average relative firefly luciferase activities in three independent experiments, each performed in duplicate. Values were normalized by activity of cotransfected Renilla luciferase.

Full size image

Discussion

The TGF-β cytostatic program is essential for maintaining normal tissue homeostasis, and is tightly regulated by a network of transcription factors that include Smad proteins, Sp1 and c-myc (Siegel and Massagué, 2003; Massagué, 2008). In this study, we report that DLX4, a homeodomain protein that is expressed in a broad range of malignancies, blocks the antiproliferative effect of TGF-β by counteracting key transcriptional control mechanisms of the TGF-β cytostatic program.

Our studies indicate there are several mechanisms by which DLX4 inactivates transcriptional control of the TGF-β cytostatic program. One mechanism is by sequestering Smad4 and preventing Smad4 from binding R-Smads. Smad interactions might also be prevented by binding of DLX4 to R-Smads as we observed binding, albeit weak, of DLX4 to Smad2. Because Smad4 and R-Smads interact with one another via their MH2 domains (Shi and Massagué, 2003), our finding that DLX4 binds the Smad4 MH1 domain is surprising. One explanation could be that binding of DLX4 to the MH1 domain induces a conformational change such that the MH2 domain of Smad4 is unable to interact with R-Smads.

Transcription factors encoded by homeobox genes are characterized by their helix-turn-helix DNA-binding homeodomain (McGinnis and Krumlauf, 1992). Our findings that binding of DLX4 to Smad4 is mediated, in part, through its homeodomain raises the question of specificity. Indeed, Hoxc8 interacts with Smad1 through its homeodomain (Yang et al., 2000). DLX1 has been reported to bind Smad4, but its binding region and mechanism have not been defined (Chiba et al., 2003). Although the homeodomain is conserved throughout the homeobox gene family, the specificity of family members for different Smads and different Smad domains is striking. DLX3 binds Smad6 but not Smad4 (Berghorn et al., 2006). Hoxc8 interacts with the MH1 domain of Smad1 (Yang et al., 2000), whereas Hoxa13 binds the MH2 domains of Smad1, Smad2 and Smad5 but does not bind Smad4 (Williams et al., 2005). Within the homeodomain, the residues of the third helix are the most highly conserved. Less conserved residues of the first and second helices might govern preferential binding to a specific Smad protein or Smad domain. Although other homeodomain proteins might potentially block TGF-β-mediated growth inhibition by binding Smads, the specificity of this inhibition could largely depend on the context on their expression. Although most homeobox genes are expressed in a highly tissue-specific manner, DLX4 is expressed across diverse malignancies (Haga et al., 2000; Man et al., 2005; Hara et al., 2007; Tomida et al., 2007; Schwartz et al., 2009). No other homeobox gene has been reported to be commonly expressed in tumors of lung, breast, ovarian, prostate and hematologic origin. Interference by DLX4 of TGF-β-mediated growth inhibition could therefore be a mechanism common to multiple organ sites.

Binding affinity and selectivity of Smad complexes for target gene promoters are principally dictated by interactions with other DNA-binding factors (Shi and Massagué, 2003, Feng and Derynck, 2005). Similarly, binding affinity and selectivity of several homeodomain proteins have been reported to be modulated by interactions with other transcriptional regulators (Shen et al., 1999; Boogerd et al., 2008). Our study is the first report that demonstrates that a homeodomain protein directly interacts with Sp1 and modulates Sp1 activity. Our findings indicate that DLX4 directly binds the DBD of Sp1 and impairs the DNA-binding ability of Sp1, but does not prevent Sp1 from associating with Smad4. Because DLX4 binds to Smad4 and to Sp1, we speculate that DLX4 inhibits p15Ink4B transcription, in part, by forming an inactive DLX4–Smad–Sp1 complex. Because transcription of p21WAF1/Cip1 is also induced by TGF-β via cooperative interactions between Sp1 and Smad proteins (Pardali et al., 2000), DLX4 could likely inhibit p21WAF1/Cip1 transcription by a similar mechanism.

Our finding that DLX4 induces expression of c-myc independently of TGF-β signaling has two implications. Firstly, induction of c-myc provides a competing mitogenic signal against the TGF-β cytostatic program. Secondly, DLX4-induced c-myc expression might lead to downregulated expression of p15Ink4B and p21WAF1/Cip1, as transcription of these genes is repressed by c-myc (Feng et al., 2002; Gartel et al., 2001). The repression of p15Ink4B transcription by c-myc has a notable similarity to our observations with DLX4. Feng et al. likewise reported that c-myc does not compete with Sp1 for interaction with Smads (Feng et al., 2002). These authors speculated that c-myc forms an inactive complex with Sp1 and Smad proteins. However, in contrast to our findings with DLX4, c-myc does not inhibit interactions between Smad4 and R-Smads nor inhibits transcriptional activity of R-Smads per se (Feng et al., 2002). Furthermore, unlike DLX4, c-myc does not affect binding of Sp1 to the p15Ink4B promoter (Feng et al., 2002). DLX4 might therefore repress p15Ink4B and p21WAF1/Cip1 transcription by increasing c-myc expression, and also by directly interfering with Smad4–R-Smad interactions and Sp1 DNA-binding activity.

Homeobox genes have essential roles in controlling cell differentiation during embryonic development. Increasing evidence indicates that aberrant expression of specific sets of homeobox genes in tumors can deregulate cell growth. For example, HOXA5 regulates p53 transcription and is silenced in breast cancers (Raman et al., 2000). Overexpression of HOXB7 and HSIX1 in ovarian and breast tumors induces expression of fibroblast growth factor-2 and cyclin A1, respectively (Naora et al., 2001; Coletta et al., 2004). Aberrant expression of homeobox genes in tumors is thought to reflect an inappropriate recapitulation of embryonic pathways (Abate-Shen, 2002; Samuel and Naora, 2005). The ability of DLX4 to block TGF-β signaling might be related to the functions of DLX genes in controlling bone morphogenesis and skeletal patterning (Panganiban and Rubenstein, 2002). These processes are tightly regulated by members of the TGF superfamily. Because DLX4 binds Smad4, DLX4 also likely blocks signaling emanating from other receptors of the TGF superfamily. Indeed, DLX4 inhibited induction of transcriptional activity by BMP-4 (Figures 3g and h; Supplementary Figure 5). Several cross-regulatory interactions have been reported between BMPs and DLX genes during normal cell differentiation. BMP-2 activates Dlx3 transcription (Park and Morasso, 2002), whereas Smad6, an antagonist of BMP signaling, inhibits DLX3 transcriptional activity (Berghorn et al., 2006). Interestingly, we observed that levels of DLX4 protein decreased in cells following TGF-β stimulation (Figure 1h). DLX4 might be a component of a regulatory loop that blocks TGF-β signaling and is conversely regulated by TGF-β. This feedback mechanism might have an important role in controlling normal embryogenesis and homeostasis.

Resistance to TGF-β-mediated growth inhibition is an important feature in the pathogenesis of most types of tumors. This resistance has been attributed to TGF-β receptor or Smad mutations in several types of tumors, particularly those of gastrointestinal and pancreatic origin (Markowitz et al., 1995; Hahn et al., 1996; Watanabe et al., 2001; Woodford-Richens et al., 2001). Resistance of tumor cells to the antiproliferative effect of TGF-β can also stem from downregulation of TGF-β receptor expression (Tokunaga et al., 1999) and p15Ink4B deletion (Gemma et al., 1996). The ability of DLX4 to disable key transcriptional control mechanisms of the TGF-β cytostatic program could explain why tumors that lack aberrations in core components of the TGF-β signaling pathway can become resistant to the antiproliferative effect of TGF-β.

A striking aspect of the TGF-β signaling pathway in tumors is its biphasic function. Many tumors are resistant to the antiproliferative effect of TGF-β but retain other TGF-β-mediated mechanisms that promote EMT and metastasis (Siegel and Massagué, 2003; Massagué, 2008). It has been thought that core components of the TGF-β pathway remain functional in these tumors, whereas downstream aberrations (such as p15Ink4B deletion) disable the growth-inhibitory arm of the pathway (Siegel and Massagué, 2003; Massagué, 2008). By sequestering Smad4, DLX4 inactivates the core pathway and might also block the metastasis-promoting function of TGF-β. Indeed, DLX4 markedly but not completely inhibited TGF-β-induced EMT in NMuMG cells (Supplementary Figures 3A–C). The ability of DLX4 to inhibit TGF-β-induced EMT could explain the association of DLX4 with favorable prognosis in lung cancer patients and its metastasis-suppressive activity reported by Tomida et al. (2007). However, DLX4 levels in ovarian and breast cancers have been reported to correlate with disease progression (Man et al., 2005; Hara et al., 2007). There are several possible explanations for this paradox. TGF-β not only induces EMT by Smad-dependent mechanisms, but also via Smad-independent pathways that involve MAP kinase and RhoA activation (Moustakas and Heldin, 2005). TGF-β-induced, non-Smad pathways that promote cell migration might not be inhibited by DLX4. DLX4 could also promote tumor progression by other mechanisms such as sustained induction of c-myc. In addition, we have found that DLX4 promotes tumor angiogenesis by inducing expression of vascular endothelial growth factor and fibroblast growth factor-2 (Hara et al., 2007). The mechanism that gives rise to overexpression of DLX4 in tumors is unclear. The DLX4 gene maps to the 17q21.3-q22 region, a chromosomal ‘hot spot’ that is amplified in ∼10% of breast and ovarian cancers (Hyman et al., 2002). However, DLX4 overexpression occurs in >50% of these tumors (Man et al., 2005; Hara et al., 2007), indicating that gene amplification is not the sole mechanism underlying this overexpression.

Like the Hedgehog, Wnt and Notch signaling pathways, the homeobox gene network is increasingly thought to be an important hub in the intimate relationship between embryonic development and cancer. Developmental patterning is known to be governed by interactions between homeobox genes and members of the TGF superfamily, but this is the first report that functionally links a homeobox gene that is aberrantly expressed in tumors with resistance to the cytostatic activity of TGF-β. Future studies of pathways controlled by this intriguing class of patterning regulators will provide important insights into the multiple steps of the neoplastic process.

Materials and methods

Plasmids

The DLX4 complementary DNA is described elsewhere (Haga et al., 2000; Hara et al., 2007). DLX4 fragments were subcloned into pET41 GST vectors (Novagen, Gibbstown, NJ, USA) as described in the text. DLX4 and non-targeting short hairpin RNAs were purchased from OriGene Technology (Rockville, MD, USA). GST–Smad2 and GST–Smad4 plasmids were provided by Fang Liu (Rutgers University). Smad complementary DNAs were purchased from OriGene Technology and subcloned into the pFA-CMV plasmid containing the GAL4 DBD (Stratagene, La Jolla, CA, USA) as described in the text. The GAL4-driven pRF-Luc reporter construct was purchased from Stratagene. The pGL2 firefly luciferase reporter vector was purchased from Promega (Madison, WI, USA). The pBV-Luc, pSBE4-Luc and pBV-MYC(Del4) reporter constructs (He et al., 1998; Zawel et al., 1998) were provided by Bert Vogelstein (Johns Hopkins University; Addgene plasmids 16539, 16495 and 16604). Sp1 complementary DNA was provided by Keping Xie (MD Anderson Cancer Center). The Cignal Sp1 reporter construct was purchased from SABiosciences (Frederick, MD, USA). Firefly luciferase constructs containing wild-type and mutant p15Ink4B promoter sequences (Feng et al., 2000) were provided by Xiao-Fan Wang (Duke University Medical Center) and Xin-Hua Feng (Baylor College of Medicine). The Id1 promoter construct was provided by Robert Benezra (Memorial Sloan-Kettering Cancer Center; Addgene plasmid 16048; Tournay and Benezra, 1996). The BRE-Luc reporter construct (Korchynskyi and ten Dijke, 2002) was provided by Peter ten Dijke (Netherlands Cancer Institute).

Antibodies and other reagents

Sources of Abs were as follows: DLX4 (Abcam, Cambridge, MA, USA; Abnova, Taipei, Taiwan); Smad2, phospho Smad2 (Ser465/467), Smad3, Smad4, p15Ink4b and c-myc (Cell Signaling Technology, Danvers, MA, USA); p21WAF1/Cip1 (Calbiochem, San Diego, CA, USA); Sp1 and E-cadherin (Zymed, San Francisco, CA, USA); N-cadherin (BD Biosciences, San Jose, CA, USA); actin and FLAG (Sigma-Aldrich, St Louis, MO, USA); and Smad2/3 and lamin A/C (Santa Cruz Biotechnology, Santa Cruz, CA, USA). Recombinant Sp1 protein, TGF-β and BMP-4 were purchased from Promega, Sigma-Aldrich and R&D Systems (Minneapolis, MN, USA), respectively.

Cell lines and transfection

Sources of cell lines were as follows: Mv1Lu, MDA-MB-468 and NMuMG (American Type Culture Collection, Manassas, VA, USA); HepG2, Ampho-293 and MCF-7 (Michelle Barton, Douglas Boyd and Francois-Xavier Claret, MD Anderson Cancer Center). For transient expression, cells were transfected using FuGENE6 reagent (Roche, Indianapolis, IN, USA). For generating stable lines, FLAG-tagged DLX4 complementary DNA was subcloned into the pRetroQ vector (Clontech, Palo Alto, CA, USA), and the retroviral construct was used to transfect Ampho-293 cells. Supernatants were harvested 2 days thereafter and used to infect target cells. Stable lines were selected by puromycin (0.5 μg/ml).

Growth assays and cell cycle analysis

Cells (4000 per well) were seeded in 96-well plates, and cultured for 2 days in complete medium containing TGF-β at concentrations ranging from 0 to 100 ng/ml. Proliferation was measured by the 3-(4,5-dimethylthiazolyl-2)-2,5-diphenyltetrazolium bromide (MTT) assay (Roche). For cell cycle analysis, cells were seeded in 10 cm dishes to reach 30% confluence the following day. Cells were serum starved overnight and then cultured in complete medium with and without addition of TGF-β for 18 h. Cells were harvested, fixed in 70% ethanol, and distribution throughout the cell cycle was determined by flow cytometric analysis of propidium iodide staining.

Reporter assays

Cells were plated in 12-well plates and cotransfected with expression plasmid (400 ng), firefly luciferase reporter plasmid (100 ng) and pRL-CMV Renilla luciferase reporter plasmid (0.5 ng; Promega) for normalizing transfection efficiency. At 24 h after transfection, cells were cultured for an additional 18 h without and with TGF-β or BMP-4. Luciferase activities were assayed using the Dual-luciferase reporter assay kit (Promega).

Immunofluorescence staining

Cells were fixed in 4% paraformaldehyde and permeabilized in 0.1% Triton X-100. Cells were stained for 1 h with Abs to Smad2, DLX4, E-cadherin or FLAG (1:200), and staining detected by Alexa Fluor 594-conjugated secondary Ab (Invitrogen, Carlsbad, CA, USA).

Immunoprecipitation and immunoblotting

Whole-cell extracts were prepared by lysing cells in M-PER buffer (Pierce Biotechnology, Rockford, IL, USA). Nuclear extracts were prepared by lysing cells in cell lysis buffer (10 mM HEPES (4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid), pH 7.9, 1.5 mM MgCl2, 10 mM KCl, 1 mM dithiothreitol plus protease and phosphatase inhibitors), followed by centrifugation at 800 g for 10 min. Nuclear pellets were lysed in nuclear buffer (20 mM Tris–HCl, pH 8.0, 100 mM NaCl, 1% nonidet P-40 (NP-40), 10% glycerol and 2 mM EDTA), centrifuged at 12 000 g for 10 min and supernatants collected. Nuclear extracts were precleared with protein G-agarose. In all, 500 μg of nuclear extract was incubated with Ab for 4–12 h at 4 °C. Immunoprecipitates were washed and subjected to SDS–polyacrylamide gel electrophoresis and immunoblotting.

Oligonucleotide pulldown assay

Nuclear extracts were precleared with streptavidin agarose for 30 min at 4 °C and incubated overnight with biotinylated oligonucleotide corresponding to nucleotides −108 to −39 of the p15Ink4B promoter (Figure 4d). Beads were washed with binding buffer (10 mM HEPES, pH 7.9, 100 mM KCl, 5 mM MgCl2, 1 mM EDTA, 10% glycerol, 1 mM dithiothreitol and 0.5% NP-40). DNA-bound proteins were eluted and subjected to SDS–polyacrylamide gel electrophoresis and immunoblotting.

ChIP

ChIP assays were performed using the ChIP Assay kit (Upstate Biotechnology, Temecula, CA, USA) following the manufacturer's protocol with modifications. Cells were cross-linked using 1% formaldehyde and then neutralized with glycine. Cells were lysed in SDS lysis buffer and sonicated to generate DNA fragments of 200–1000 bp in length. Sheared chromatin was precleared with protein G-agarose and incubated overnight with 4 μg of Sp1 and Smad Abs. Protein–DNA complexes were precipitated with protein G-agarose, followed by elution of immunoprecipitated complexes and reversal of cross-links. Purified DNA was used in PCR reactions to amplify a 535 bp fragment of the p15Ink4B promoter using the following primers: sense, 5′-TATGGTTGACTAATTCAAACAG-3′; antisense, 5′-GCAAAGAATTCCGTTTTCAGCT-3′. The amplified fragment was confirmed by DNA sequencing.

In vitro binding assays

GST–DLX4 and GST–Smad fusion proteins were produced in Escherichia Coli. The 35S-labeled DLX4, Smad4 and Sp1 were synthesized by in vitro transcription/translation (Promega) from the T7 promoter, and incubated for 2 h with 1 μg of GST-fusion protein bound to glutathione-sepharose beads (GE Healthcare, Piscataway, NJ, USA). Beads were washed with binding buffer (20 mM Tris–HCl, pH 8.0, 100 mM NaCl, 0.1% NP-40 and 2 mM EDTA). Associated proteins were subjected to SDS–polyacrylamide gel electrophoresis and visualized by autoradiography.

Gel-shift assay

Recombinant Sp1 (100 ng) was incubated with increasing amounts of in vitro translated FLAG–DLX4 and FLAG-tag at 23 °C for 20 min in binding buffer (10 mM Tris–HCl, pH 7.5, 1 mM MgCl2, 50 mM NaCl, 0.5 mM EDTA, 4% glycerol, 0.5 mM dithiothreitol and 1 μg poly(dI-dC).poly(dI-dC). In all, 1 ng of 32P-labeled oligonucleotide containing sequences −88 to −64 of the p15Ink4B promoter (Figure 4d) was added to the binding reaction and incubated for an additional 20 min. DNA–protein complexes were separated by polyacrylamide gel electrophoresis and visualized by autoradiography.