Axl promotes cell invasion by inducing MMP-9 activity through ... - Nature

Oncogene (2008) 27, 4044–4055

& 2008 Nature Publishing Group All rights reserved 0950-9232/08 $30.00 www.nature.com/onc

ORIGINAL ARTICLE

Axl promotes cell invasion by inducing MMP-9 activity through activation of NF-jB and Brg-1 K-Y Tai1, Y-S Shieh2, C-S Lee1, S-G Shiah1 and C-W Wu1 1 Institute of Cancer Research, National Health Research Institutes, Miaoli County, Taiwan and 2School of Dentistry, National Defense Medical Center, Taipei, Taiwan

Activity of the Axl receptor tyrosine kinase is positively correlated with tumor metastasis; however, its detailed role in the mechanism of tumor invasion is still not completely understood. Here, we show that Axl enhances the expression of matrix metalloproteinase 9 (MMP-9), required for Axl-mediated invasion both in vitro and in vivo. We found that the highly selective MEK1/2 inhibitors U0126 and PD98059, and the expressed dominant-negative form of extracellular signal-regulated kinase (ERK), completely block Axl-mediated MMP-9 activation. In contrast, the phosphatidylinositol 3-kinase inhibitor LY294002 and wortmannin had little effect on activation. Interestingly, however, the Axl ligand Gas6 is not involved in Axl-mediated MMP-9 activation. Mutation of Glu59Axl and Thr77Axl dramatically reduced Gas6– Axl binding but continued to induce MMP-9 activation. In addition, overexpression of Axl-activated ERK and enhanced nuclear factor-jB (NF-jB) transactivation and brahma-related gene-1 (Brg-1) translocation. Exposure to the NF-jB inhibitor silibinin, which inhibits IjBa kinase activity, or overexpression of the dominant-negative mutant IjB and Brg-1 strikingly inhibited Axl-mediated MMP-9 activation. These data indicate that coordination of ERK signaling and NF-jB and Brg-1 activation are indispensable to regulation of Axl-dependent MMP-9 gene transcription. Together with previous data, our results provide a plausible mechanism for Axl-mediated tumor invasion and establish a functional link between the Axl and MMP-9 signaling pathways. Oncogene (2008) 27, 4044–4055; doi:10.1038/onc.2008.57; published online 17 March 2008 Keywords: Axl; Brg-1; ERK; MMP-9; NF-kB

Introduction Axl (also known as Ark, UFO and Tyro7) plays two critical roles in cells: as a transforming gene that is Correspondence: Dr C-W Wu and Dr Shine-Gwo Shiah, Institute of Cancer Research, National Health Research Institutes, 35, Keyan Road, Zhunan Town, Miaoli County 350, Taiwan. E-mails: [email protected] and [email protected] Received 19 October 2007; revised 4 January 2008; accepted 1 February 2008; published online 17 March 2008

overexpressed in human tumors, and as a key mediator of the Gas6/Axl signaling system that is activated in the vasculature. Axl was originally identified as a transforming gene in human chronic myelogenous leukemia, and its transforming activity was attested by several sensitive nude mouse tumorigenicity assays (Neubauer et al., 1993). Many studies have shown that Axl is overexpressed in a variety of tumor cells or in several types of human cancers, including renal (Chung et al., 2003), esophageal (Nemoto et al., 1997), thyroid (Ito et al., 1999), lung (Shieh et al., 2005), breast (Meric et al., 2002), gastric (Wu et al., 2002) and colon (Craven et al., 1995) cancers, ovarian carcinoma (Sun et al., 2004), melanoma (van Ginkel et al., 2004) and osteosarcoma (Nakano et al., 2003). Of particular interest, overexpression of Axl can transform fibroblasts even in the absence of ligand. An EAK (named for EGFR-Axl kinase chimeric receptor construct or a fusion of a viral gag gene to Axl kinase domain construct is sufficient to induce Axl’s transforming activity and cause tumors in nude mice (Fridell et al., 1996). Thus, Axl transactivation is probably mediated through mechanisms that do not involve the interaction of Axl with its endogenous ligand, Gas6. The Gas6/Axl system plays an important role in vascular biology. The signaling promotes cell growth, antiapoptosis and migration in vascular smooth muscle cells, endothelial cells and fibroblasts (Melaragno et al., 1999). Recent studies also indicate that Gas6/Axl signaling affects multiple cellular behaviors required for neovascularization and maintaining vascular integrity (Holland et al., 2005). These findings indicate that Axl regulates processes fundamental for both tumorigenesis and angiogenesis. Axl not only behaves as an oncoprotein when it is overexpressed but also plays a role during invasion. Axl is highly expressed in metastatic prostate cancer cells compared with normal prostate cells and other prostatic carcinoma cell lines (Jacob et al., 1999). Inhibition of Axl signaling by a dominant-negative receptor mutant (AXL-DN) suppresses experimental gliomagenesis, migration and invasion, and leads to long-term survival of mice after intracerebral glioma cell implantation compared with Axl wild-type transfected tumor cells (Vajkoczy et al., 2006). Furthermore, Axl is detected at higher levels in metastases or malignant tumors than in normal tissues or primary tumors, and a higher level is

Axl modulates MMP-9 expression by NF-jB and Brg-1 K-Y Tai et al

4045

associated with a poor clinical outcome (Bittner et al., 2000; Nakano et al., 2003; Shieh et al., 2005). The progression and metastasis of malignant tumors are characterized by the ability of tumor cells to produce matrix-degrading enzymes, giving them an advantage in disseminating from the primary tumor and invading surrounding tissue. Of the several families of matrix proteases implicated in cancer invasion and metastasis, the bulk of the available data focus on the matrix metalloproteinases (MMPs), and most evidence implicates increased proteolysis in these events (Fingleton, 2006). MMP-9, a 92 kDa type IV collagenase, degrades type IV collagen, the major structural component of the basement membrane and extracellular matrix, and increases its activity with the degree of malignancy of tumor cells. Multiple signaling transduction pathways are involved in the regulation of MMP-9 production in human cancer cells. The Ras/Raf extracellular signalregulated kinase (ERK)1/2 cascade is essential for induction of MMP-9 expression (Zeigler et al., 1999; Genersch et al., 2000; Lakka et al., 2002), although strong induction of MMP-9 expression requires a combination of different signaling pathways. The signaling molecules and the phosphorylation of cytoplasmic substrates activated by Axl in various cell types include phosphatidylinositol 3-kinase (PI3-K), Akt, S6K, Src kinase, ERK, p38 mitogen-activated protein kinase (MAPK) and nuclear factor-kB (NF-kB, Hafizi and Dahlback, 2006). In addition, the biochemical characterization of Gas6–Axl interactions has revealed that unlike the epidermal growth factor receptor–Axl chimeric and IL-15 receptor–Axl heterologous receptors, the Ras/ERK pathway is not involved in Gas6induced Axl signaling (Budagian et al., 2005). However, the functional relevance of Axl in the regulation of MMP-9 gene expression is still unknown. In investigating how Axl overexpression increases the metastatic potential, and, in particular, the Axlmediated invasive signaling in cancer cells, we found that Axl enhances the expression of MMP-9 and regulates its promoter activity by MAP kinase kinase (MEK)/ERK rather than by PI3-K–Akt-dependent signaling pathways. We also provide several lines of experimental evidence to support the notion that Axl simultaneously modulates NF-kB and brahma-related gene-1 (Brg-1), a chromatin-remodeling factor, to promote MMP-9 activation. Our data establish a molecular mechanism whereby Axl-overexpressing cancer cells improve invasion and provide crucial evidence of a functional link between the Axl and MMP-9 signaling pathways. Results Axl enhances MMP-9 gene expression To investigate the potential effect of Axl on MMP-9 expression, we examined MMP-9 expression by reverse transcription (RT)–PCR in several slightly invasive Axlnegative cell lines (MCF-7, MDA-MB-415, HT1376 and NCIH-520). MMP-9 expression was much higher in the

Axl transfectant cells than that in the vector control cells (Figure 1a, i). This phenomenon is consistent with a previously established Axl-stable transfectant (Shieh et al., 2005) of the CL1-0 human lung adenocarcinoma cell line (Figure 1a, ii). The activation of MMP-9 by Axl is specific because the MMP-2 and tissue inhibitor of metalloproteinases 1–3 (TIMP1–3) were not activated. The increase in MMP-9 expression by Axl was further supported by a luciferase reporter assay (Figure 1c). Cotransfection of MCF-7 cells with increasing doses of Axl resulted in a dose-dependent activation of luciferase activity, suggesting that Axl enhances MMP-9 transcriptional activity. Likewise, Axl-induced MMP-9 enzymatic activity was also activated in the presence of Axl but not in the presence of Gas6 or AXL-DN (Figure 1b). To further investigate whether Axl enhances MMP-9 expression in primary tumor tissues, we performed immunohistochemical staining (Figure 1d). We found a strong positive correlation between Axl and MMP-9 expression. In human breast tumor tissues in which Axl expression was positive, MMP-9 expression was also positive (14 of 60, 23.3%), whereas MMP-9 expression was undetectable in Axl-negative normal breast tissues (32 of 60, 53.3%). It should be mentioned that the two images (case 1, left and right, and case 2, left and right) were derived from nearby sections of the same tumor tissue sample. Taken together, these results indicate that Axl is able to increase the expression of MMP-9 both in vitro and in vivo. MMP-9 is required for Axl-enhanced invasion, and Axl knockdown impairs the invasive capability of cancer cells Because Axl induces MMP-9 expression and MMP-9 plays an important role in metastasis, we examined whether MMP-9 is required for Axl-induced invasion in vitro. The Axl-induced invasive activity in MCF-7 cells was inhibited 70–80% by MMP-9-specific inhibition, but less so by MMP-2 inhibition (Figure 2a). Next, we asked whether endogenous Axl is crucial for both upregulation of MMP-9 activity and the in vitro invasive activity. We used a lentiviral short hairpin RNA (shRNA) to inhibit Axl expression. The shRNA against Axl was used to transfect CL1-5F4 cells, which exhibit a high level of endogenous Axl and invasive activity, and the two stable puromycin-resistant colonies were named A82 and A84. Two similar controls, one transfected with an empty vector and one containing a scramble sequence were selected and named V and C, respectively. As expected, Axl expression was reduced in the A82 and A84 cells, and their invasive activity decreased significantly in comparison with that of V and C (Figure 2b). Interestingly, zymography showed that the basal level of the 92 kDa gelatinolytic activity in CL1-5F4 was decreased by blocking Axl expression in A82 and A84 cells (Figure 2c). To understand whether the suppression of Axl is sufficient to block the secretion of MMP-9 in cancer cells in which the cell invasion is ongoing, we cultured four colonies (V, C, A82 and A84) on a thin layer of Matrigel-coated on plastic dishes. After 24 h, the Oncogene


4046

colonies were collected in the upper cell culture media and analysed in a zymography assay. The level of MMP-9 activity was not stimulated in Axl-depleted cells even when cultured on Matrigel (Figure 2c, lanes 7 and 8). Our data suggest that Axl plays an important role in the specific activation of the expression of

Oncogene

MMP-9 through contact with the extracellular matrix. To gather further support for the idea that Axl catalytic activity is required for modulating MMP-9 expression, we analysed the dose dependence of MMP-9 inhibition by overexpression of the AXL-DN gene that when overexpressed inhibits Axl activation by silencing


4047

the phosphorylation of a conserved tyrosine residue inside the kinase domain (Figure 2d). Transfection of 0.5, 1 or 2 mg of AXL-DN DNA causes a dosedependent decrease in MMP-9 to the basal level of the

vector control in CL1-5F4 cells. Taken together, these data show that Axl expression is important to the regulation of MMP-9 activity when driving cancer cell invasion.

Figure 2 Matrix metalloproteinase 9 (MMP-9) is required for Axl-enhanced invasion. (a) Quantification of the invasion assays. Invasion assays were performed using a Boyden chamber assay coated with Matrigel, which mimics the basement membrane composition. Matrigel invasion assay for MCF-7 cells transfected with Axl and treated with MMP inhibitors. Bars (from left to right): (1) vector alone, (2) Axl alone, (3) AXL-DN alone, (4) Axl plus MMP-9 inhibitor (100 nM) and (5) Axl plus MMP-2 inhibitor (100 nM). The invasiveness was normalized to that observed in MCF-7/Axl cells. Cells were counted in triplicate wells and in three identical experiments. *Po0.05. (b) Invasion analysis of the lentiviral Axl shRNA-transfected CL1-5F4 cells (A82 and A84) and control cells (V, empty vector; C, scrambled control). *Po0.05. Western blot analysis of Axl expression (upper panel). (c) Zymography of the Axl shRNA-transfected CL1-5F4 cells in culture media collected from cells settled on Matrigel (100 mg cm2) after 24 h. (d) Dose-dependent repression of endogenous MMP-9 expression by AXL-DN in CL1-5F4 cancer cells. The zymographic assay was performed with the indicated concentrations of AXL-DN.

Figure 1 Axl enhances matrix metalloproteinase 9 (MMP-9) gene expression. (a, i) Axl increased the mRNA level of MMP-9 in several cancer cell lines. Cells were transfected with Axl at the indicated dose. Total RNA was prepared and equal amounts of RNA were subjected to reverse transcription (RT)–PCR. The MMP-9 signal is shown as a PCR band of 208 bp, and glyceraldehyde-3phosphate dehydrogenase (GAPDH) mRNA (578 bp) was amplified in the same reaction as a loading control. (ii) Analysis of MMP-9, MMP-2 and tissue inhibitor of metalloproteinases 1–3 (TIMPs) mRNA expression in CL1-0/Axl-stable transfectants. (b) Induction of MMP-9 gelatinolytic activity by Axl overexpression in MCF-7 cancer cells. The zymographic assay was performed with the indicated concentrations of Axl: 5 mg for the dominant-negative AXL-DN and 5 mg for Gas6. Gas6 was detected in cell-conditioned medium and Axl was detected in lysates by western blotting (lower panel). The concentration of Gas6 in cell-conditioned medium (97.06 ng ml1) was quantified by enzyme-linked immunosorbent assay (ELISA; Supplementary Figure 1). (c) Axl induced transcriptional upregulation from MMP-9 promoter in MCF-7 cells. A carrying MMP-9 proximal promoter luciferase reporter plasmid (pMMP9-770-luc) was cotransfected with Axl, AXL-DN or Gas6 expression plasmids. After 48 h transfection, cell lysates were prepared and subjected to the luciferase assay. The luciferase reporter activity was normalized to Renilla activity (mean±s.d., n ¼ 3, *Po0.05). (d) Correlation between Axl and MMP-9 expression in the human tissue array. Representative human primary breast tumor tissue samples with immunohistochemical staining. Case 1: sample from a metastatic carcinoma in the lymph node of a woman with breast cancer. Axl-positive staining (top left) and MMP-9 expression-positive staining (top right). Case 2: normal breast tissue from a woman with breast cancer. ‘F’ denotes the extensive mass of dense fibrous and ‘L’ indicates the lobules. Axl-negative staining (bottom left) and MMP-9-negative staining (bottom right). Original magnification, 400. Oncogene


4048

Gas6–Axl contact is not required for MMP-9 activation Gas6 functions as a biological ligand for Axl receptor tyrosine kinase (RTK). However, both the Gas6–Axl pairing and ligand-independent activation of Axl RTK exist (Hafizi and Dahlback, 2006). To determine whether Gas6 is involved in Axl-mediated MMP-9 activation, we designed a Gas6–Axl major contact site mutant Axl (AxlE59R/T77R) plasmid, which dramatically reduced Gas6 binding in a solid-phase assay (Sasaki et al., 2006), to examine the effect of Gas6 on the activation of MMP-9. As expected, mutation of Axl at the major contact acutely affected the high-affinity interaction of Gas6 and Axl (Figure 3a). Zymographic experiments showed no difference in the induction of MMP-9 activity between the wild-type Axl and the binding-site mutant Axl following either cotransfection of Gas6 or no transfection (Figure 3b). Similar results were also observed using Gas6 overexpression to enhance MMP-9 activity by stimulating endogenous Axl in CL1-5F4 cells (Figure 3c). These results indicate the Axl-mediated MMP-9 activation may occur through a ligand-independent pathway. MEK/ERK signaling pathway is involved in Axl-induced MMP-9 activity We performed western blots to examine the activation status of ERK and Akt over the time course of Axlinduced signaling in MCF-7 cells (Figure 4a, i). Strong activation of ERK and Akt was observed 15 h after Axl transfection. However, while the activation of Akt was not diminished in Axl knockdown cells, the activation of ERK decreased (Figure 4a, ii). To investigate the signaling pathways critical for Axl induction of MMP-9 gene expression, we used chemical inhibitors of these signaling pathways to examine the requirement of activation of each pathway in Axl-mediated MMP-9 upregulation. U0126 and PD98059 selectively block the activity of MEK, an activator of ERK kinase. LY294002 and wortmannin are highly specific inhibitors that block PI3-K-dependent Akt phosphorylation and kinase activity. Inhibition of ERK kinase by U0126 or PD98059 effectively blocked Axl-mediated activation of MMP-9. U0126 and PD980859 inhibited Axl-mediated MMP-9 upregulation in a concentration-dependent manner, and they completely blocked MMP-9 induction by Axl at a concentration of 25 and 20 mM, respectively. Interestingly, LY294002 and wortmannin had little effect on Axl-mediated activation of MMP-9, even at high concentrations of 50 and 40 mM, respectively, which effectively inhibited Axl-mediated activation of Akt, a downstream target of PI3-K (Figure 4b, i–ii). These results indicate that the ERK pathway is more likely to contribute to Axl-induced activation of MMP-9, whereas the PI3-K–Akt pathway, despite its activation by Axl, is not a major contributor to MMP-9 activation. Three parallel MAPK pathways have been identified in mammalian cells. ERK, JNK and p38 constitute major components of these signaling pathways, which regulate many intracellular events, including cell proliferation, differentiation and migration. To further Oncogene

Figure 3 Gas6–Axl contact is not required for matrix metalloproteinase 9 (MMP-9) activation. (a) Mutant AxlE59R/T77R physically abolished Gas6 binding in an immunoprecipitation assay. The assay was performed with the indicated transfectants in MCF-7 (from left to right): (1) transfection of Gas6 alone, (2) cotransfection of Gas6 and AXL-WT and (3) cotransfection of Gas6 and AxlE59R/T77R. The protein complexes were subjected to immunoblot analysis with antibodies as indicated. (b) The zymographic assay was performed with the indicated transfectants in MCF-7 cells. (c) CL1-5F4 cells were transfected with Gas6 and subjected to zymography. The zymographic assay was performed with the indicated concentrations of Gas6. Gas6 was detected in cell-conditioned medium by western blotting (lower panel) and the concentration of Gas6 in cell-conditioned medium was quantified by enzyme-linked immunosorbent assay (ELISA; Supplementary Figure 1).

understand the network of Axl-mediated MMP-9 activation, the kinase activities and the specific chemical inhibitors of a group of MAPKs were examined in Axl transfectants. Zymographic assay demonstrated that only U0126 completely blocked MMP-9 induction by Axl at a concentration of 5 mM, yet JNK inhibitor and p38 inhibitor SB202190 did not differ in their ability to repress the induction (Figure 4c). The MAPK activity


4049

Figure 4 Effect of the Axl/MEK/ERK pathway on matrix metalloproteinase 9 (MMP-9) expression. (a, i) MCF-7 cells were transiently transfected with 5 mg Axl. Western blots show phosphorylated Akt, phosphorylated ERK1/2 and total proteins of Akt or ERK1/2 at the indicated times. a-Tubulin was used as a loading control. (ii) Western blot analysis of Axl shRNA stably transfected CL1-5F4 cells. (b) Effects of Axl-induced activation on MMP-9 expression using the chemical inhibitors U0126 and LY294002 in (i), PD98059 and wortmannin in (ii), and three mitogen-activated protein kinase (MAPK) inhibitors U0126, JNK inhibitor I and SB202190 in (c). MCF-7 cells were transfected with Axl for 12 h and then treated with the indicated concentrations of the chemical inhibitors. After 24 h of incubation, the conditioned media were collected, concentrated and subjected to zymographic assay. Similarly, CL1-5F4 cells were treated with the indicated concentrations of the chemical inhibitors and subjected to zymographic assay (Supplementary Figure 3). (d) Modulation of MAPK activity by Axl. Immune complex kinase assays were performed to measure ERK kinase activity, JNK activity and p38 kinase activity in Axl-overexpressing or Axl shRNA-stable transfectants.

was measured by immune complex kinase assays in Axl-overexpressing MCF-7 cells and Axl knockdown CL1-5F4 cells. Western blot analysis using monoclonal antibodies to phosphor-specific myelin basic protein and activating transcription factor-2 demonstrated that Axl significantly modulated ERK activity through both overexpression and depletion, but Axl did not affect JNK and p38 activities (Figure 4d). The above observation together with the potent inhibition of MMP-9 activation by U0126 suggests that ERK and its activity may play an important role in mediating the induction of MMP-9 by Axl. Axl induces MMP-9 by the NF-kB-dependent pathway in MCF-7 cells To determine whether NF-kB is involved in Axlmediated MMP-9 activation, we used immunoblotting

to examine the effect of Axl on NF-kB regulation. MCF-7 cells were transfected with Axl for up to 24 h, and the nuclear translocation of p65 and p50 was determined (Figure 5a). Axl transfection led to nuclear accumulation of p65 and p50, which peaked 6–18 h after transfection. The degradation of cytoplasmic IkB-a was observed after 18 h following Axl transfection. This confirms that activation of NF-kB in MCF-7 cells results from the specific Axl-mediated downstream signaling. Interestingly, nuclear accumulation of c-Fos and Brg-1 also occurred with the same kinetics as those of p65 and p50. To further define Axl-responsive elements in the 770 bp MMP-9 proximal promoter, site-directed mutation constructs of the promoter were tested (Figure 5b). Using a cotransfection assay, we found that mutation of the NF-kB element inhibited Axl-induced MMP-9 promoter activity, whereas mutation distal or proximal Oncogene


4050

Figure 5 Axl induces matrix metalloproteinase 9 (MMP-9) by the nuclear factor-kB (NF-kB)-dependent pathway in MCF-7 cells. (a) MCF-7 cells were transfected with Axl for the indicated times. Nuclear (right panel) and cytoplasmic (left panel) proteins were extracted and immunoblot analysis was performed as indicated. Sp1 was used as a nuclear loading control and a-tubulin was used as a cytoplasmic control. (b) Site-directed mutant constructs of the MMP-9 promoter were transiently transfected into MCF-7 cells and cotransfected with Axl for 48 h, and the luciferase activity was examined. Data are presented as the mean±s.d. from three experiments. ‘X’ denotes the sites of mutations on the promoter. *Po0.05. (c) A significant increase of NF-kB activity was detected in Axlstimulated cells in a concentration-dependent manner. (i) Nuclear extracts were taken from cells transfected with Axl, AXL-DN or Gas6 for 2 days and then subjected to the enzyme-linked immunosorbent assay (ELISA)-based NF-kB activity assay ( þ , 2.5 mg; þ þ , 5 mg). Nuclear extracts (5 mg) were incubated with a wild-type probe of the NF-kB consensus-binding sequence (blank bars) or a mutant-type probe of the NF-kB consensus-binding sequence (black bars). Cytoplasmic extracts were incubated with a wild-type probe of the NF-kB consensus-binding sequence (gray bars). The data represent the mean±s.d. of three separate experiments. *Po0.05. (ii) The ELISA-based NF-kB activity assay was performed on MCF-7/Axl cells treated with LY294002 (50 mM), U0126 (10 mM) or silibinin (10 mM). Data represent the mean±s.d. of three separate experiments, *Po0.05. (d, i) The zymographic assay was performed with the indicated concentrations of the chemical inhibitors in MCF-7 cells. (ii) Dominant-negative ERK1/2 or IkB-a suppresses MMP-9 activity induction by Axl in MCF-7 cells. The zymographic assay was performed with the indicated dominant-negative protein (1–2 mg) and Axl (3 mg) cotransfectants. MCF-7 cells were cotransfected for 24 h and then placed in serum-free media for 24 h. After 24 h of incubation, the conditioned media were collected, concentrated and subjected to the zymographic assay.

to the AP-1 site had only a modest effect. Furthermore, mutation of the NF-kB site plus any AP1 site caused dramatic loss of the response to Axl-induced MMP-9 promoter activity. These results indicate that a major element contributing to Axl-induced MMP-9 transactivation resides in the NF-kB site. To further support this notion, we used chemical inhibitor of NF-kB to examine the effect of NF-kB on Axl-driven MMP-9 activity. Silibinin, a drug used clinically in the treatment of liver diseases and a variety of cancers, inhibits the upstream kinase cascades involved in NF-kB activation through inhibiting IkB kinase activity and p65/p50 translocation (Dhanalakshmi et al., 2002). If NF-kB is involved in the Axl-enhanced MMP-9 activation, then blocking NF-kB activity and its upstream signals should inhibit Axlinduced MMP-9 expression. Consistent with this notion, a zymographic assay showed there were strong inhibitory effects of both silibinin and U0126 on Axl-mediated Oncogene

MMP-9 activation; in contrast, the Akt/mTOR pathway inhibitor rapamycin had no apparent effect (Figure 5d, i). Because NF-kB is essential for Axl-mediated MMP-9 transactivation, we also asked whether Axl can enhance NF-kB DNA-binding activity, measured by an enzymelinked immunosorbent assay (ELISA)-based activity assay. We found that Axl dose dependently enhanced NF-kB DNA-binding activity. The enhancement was specific because there was no significant effect on binding of the mutant DNA probe or the cytoplasmic extracts (Figure 5c, i). Gas6-induced NF-kB DNAbinding activity was higher than that of the vector control. The Gas6-dependent nuclear NF-kB DNAbinding activity may be mediated by other Axl subfamily members, such as Sky (Supplementary Figure 4). In addition, the Axl-induced NF-kB DNA-binding activity was dramatically lower in the presence of U0126 or silibinin compared with that in presence of LY294002


4051

Figure 6 Brahma-related gene-1 (Brg-1) is indispensable for Axl-mediated matrix metalloproteinase 9 (MMP-9) activity in MCF-7. (a, i) Axl dose dependently (1–5 mg) induced Brg-1 translocation. The nuclear protein level of Brg-1 was determined by immunoblotting. (ii) The levels of Axl phosphorylation were assayed by western blot analysis. MCF-7 cells were transfected with Axl as described in (i). Axl was precipitated from the lysates and phosphorylation was analysed using anti-pTyr Abs (Upstate Biotechnology Inc.). (b) MCF-7 cells were cotransfected with Axl and Brg-1 or the ATPase-null mutant Brg-1 (K798R). Supernatants from the transfected cells were subjected to zymography, and the induction is shown. (c) Chromatin immunoprecipitation was performed on MCF-7/Axl or CL1-5F4/shAxl-stable transfectants with the antibodies as shown. (d) Schematic diagram illustrating the molecular mechanism responsible for Axl-induced MMP-9 expression. Axl induces both the PI-3K/AKT and ERK pathways but not other mitogen-activated protein kinase (MAPK) pathways. Stimulation of the ERK signal pathways by Axl leads to the activation of nuclear factor-kB (NF-kB) and Brg-1 chromatin-remodeling factor. Translocation of Axl-activated NF-kB and Brg-1 into the nucleus increases MMP-9 expression. Therefore, Axl regulates the transcriptional activation of MMP-9, which plays an important role in invasion and metastasis of tumors, through coordination of the ERK–NF-kB and Brg-1 pathways.

(Figure 5c, ii). The inhibition of PI3-K pathway by LY294002 did not affect the NF-kB-binding activity in our system. It might be possible that the crossregulation triggered by other pathways may be able to restore the NF-kB-binding activity. To further determine whether the enhancement of MMP-9 activity by Axl is through the ERK/NF-kB pathway, we used DN constructs of ERK1, ERK2 and IkB and examined the consequences by zymographic assay (Figure 5d, ii). Axlinduced MMP-9 activity was reduced by both ERK1 (DN) and ERK2 (DN) in a dose-dependent manner and was abolished completely by IkB (DN). This result suggests that the ERK/NF-kB pathway contributes to the Axl-induced MMP-9 expression.

Brg-1 is indispensable for Axl-mediated MMP-9 activity in MCF-7 Because Brg-1 plays a major role in chromatin remodeling of the MMP-9 promoter (Ma et al., 2004), it raises an interesting possibility that the upregulation of Brg-1 by Axl might mediate the Axl-mediated MMP9 activation. Consistent with this notion, Axl induced a time-dependent (Figure 5a) and dose-dependent (Figure 6a, i) induction of Brg-1 translocation, which correlated well with the increased Axl expression and phosphorylation (Figure 6a, ii). To determine whether Brg-1 is essential for Axl-mediated MMP-9 activation, we then used the wild-type or the ATPase-null mutant Brg-1(K798R) and examined its effect on Axl-mediated Oncogene


4052

MMP-9 activity. Treatment of Axl-overexpressing cells with mutant Brg-1 completely reduced the level of MMP-9 activity, indicating that ATPase-dependent chromatin remodeling mediated by Brg-1 was functionally required for Axl-induced MMP-9 activation. Importantly, the wild-type Brg-1 was able to duplicate the Axl-induced MMP-9 activity (Figure 6b). To monitor whether the critical chromatin-remodeling factor and transcription factors are recruited to the endogenous MMP-9 promoter by Axl signaling, chromatin immunoprecipitation (ChIP) assays were performed with antibodies against p65, p50, c-Fos, Brg-1 and normal mouse immunoglobulin G (as a negative control) in both the Axl-overexpression and -depletion systems (Figure 6c). Input control indicated that the soluble chromatin samples obtained from each transfectant cell had equal amounts of chromatin fragments containing the MMP-9 promoter. Regarding NF-kB subunits, p65 and p50 were strongly associated with the MMP-9 promoter in the abundant Axl cells. The recruitment of chromatin-remodeling factor, as represented by binding of Brg-1 to the MMP-9 promoter, occurred in MCF-7/Axl cells but not in the CL1-5F4/ vector or in CL1-5F4/A82 cells. This observation might imply that Brg-1 is recruited only by Axl to silence the MMP-9 promoter for disrupting nucleosome structure to increase DNA accessibility and its transcriptional activity, but that Brg-1 is not involved in MMP-9 transcriptional repression. ChIP was also performed to show the modification of covalent histone on MMP-9 promoter. Methylation of K9-H3 and K27-H3, generally a suppressive modification, occurred in Axl-null (MCF-7/pCMV) and Axl-depleted (CL1-5F4/A82) cells. Acetylation of H3, associated with relaxation of the compact structure of nucleosomes, increased substantially in the Axl-overexpressing cells but decreased slightly in the Axl-depleted cells. These results indicate that NF-kB and Brg-1 are recruited to the MMP-9 promoter and function to activate MMP-9 gene transcription during the Axl-dependent MMP-9 activation.

Discussion In this study, we have shown that Axl can specifically induce MMP-9 expression and promote its activity, which in turn plays an important role in cancer metastasis and upregulation of Axl expression in invading cancer cells. As shown schematically in Figure 6d, Axl regulates cancer cell invasive capabilities, at least in part, by modulating MMP-9 expression through activation of the ERK/NF-kB pathway and Brg-1-mediated chromatin remodeling. Previous studies indicated that Gas6 induces the phosphorylation of Axl in fibroblasts and endothelial cells, which contributes to some survival signaling, including the NF-kB transcription factor system (Demarchi et al., 2001; Hasanbasic et al., 2004). In NIH3T3 fibroblasts incubated in serum-deprived conditioned medium, the Gas6/Axl pathway upregulates Oncogene

Bcl-2 and Bcl-x(L) expression and activates NF-kB to suppress apoptosis. In human umbilical vein endothelial cells, Gas6 treatment and subsequent activation of PI3K/Akt and NF-kB decreases caspase 3 activity and increases the level of Bcl-2 protein. In contrast to the role of NF-kB stimulated by the Gas6/Axl system in vascular biology, the possibility that the Axl-driven NF-kB activity plays a role in tumor cells, especially in cancer metastasis, remains to be determined. A positive correlation between Axl-induced cell invasion and NF-kB activity was identified in human non-small cell lung cancer (Lay et al., 2007). Inhibition of NF-kB activity with the NF-kB inhibitor sulfasalazine in nonsmall cell lung cancer cells disrupts chemoresistance and cell invasiveness. Our present observation provides the first evidence that NF-kB activity is essential for an efficient Axl-mediated invasion of cancer cells by secreting MMP-9. This mechanism is consistent with the aberrant activation of Axl and NF-kB associated with malignant properties in cancer cells (Budagian et al., 2005; Vajkoczy et al., 2006; Lay et al., 2007). We found that Axl-induced MMP-9 transactivation is probably mediated through mechanisms that do not involve the interaction of Axl with endogenous or exogenous Gas6, suggesting that this upregulation can be attributed to a distinct pathway, which differs from the Gas6/Axl survival signaling. A striking finding in our study was that induction of MMP-9 by Axl is not dependent on activation of the PI3-K/Akt/NF-kB signaling pathway, but is instead dependent on the MEK/ERK/NF-kB signaling pathway. The chemical inhibitors and dominant-negative plasmids we used to block the function of target kinases were highly specific and did not cause any apparent phenotypical changes in cells during the assay time. We have demonstrated that blocking either MEK/ERK kinase or NF-kB activity completely inhibits MMP-9 production. Interestingly, inhibition of the PI3-K pathway by LY294002 or wortmannin had little effect on MMP-9 induction by Axl. Previous studies have shown that suberoylanilide hydroxamic acid, a histone deacetylase inhibitor, promotes the activation of the PI3-K pathway to cause phosphorylation of Akt serine 473, which in turn activates p300 through the phosphorylation of serine 1834 (Liu et al., 2006). This specific phosphorylation episode is necessary to acetylate RelA/p65 on lysine 310, which results in the stabilization and recruitment of endogenous acetylated RelA/p65 to NF-kB-dependent promoters. This mechanism may explain why blocking the PI3-K pathway cannot abolish Axl-mediated MMP-9 expression but still partially affects the activation. Brg-1 is a core subunit of the human SWI/SNF complex, a well-characterized ATPase-dependent remodeling complex that disrupts histone–DNA interactions and increases the accessibility of specific DNA regions to the basal transcriptional machinery associated with transcription activity (Wang, 2003; Roberts and Orkin, 2004). Brg-1-mediated chromatin remodeling has been shown recently to be required for MMP-9 gene expression (Ma et al., 2004). Consistent with this observation, our results illustrate a novel mechanism


4053

for Axl-mediated gene transactivation in which both NF-kB and Brg-1 are activated by Axl and are functionally required for Axl-induced MMP-9 expression. Recent evidence indicates that the SWI/SNF complex involves direct interactions with the promoter sequence of affected genes combined with epigenetic regulation (Banine et al., 2005). First, both the CD44 and the E-cadherin genes are activated by Brg-1 and Brm, and this activation involves loss of these promoters’ DNA methylation. Changes in cell adhesion play a critical role in tumor progression because cell adhesion molecules of varying classes and functions, including cadherins, CD44 and integrins, can interact with and modulate the signaling function of RTKs. Second, yeast two-hybrid and coimmunoprecipitation analyses demonstrate that Brg-1 binds directly to Rb, a prototypical tumor suppressor gene involved in regulating the cell cycle by repressing expression of the E2F family, and that Brg-1 cooperates with the Rb/DNMT1 interaction to promote cell-cycle arrest, which is involved in DNA methylation of the E2F-dependent promoter (Roberts and Orkin, 2004). Third, the functional cooperativity between Brg-1 and histone acetyltransferases, such as SRC-1 and CBP, has been observed in estrogen receptor signaling activation (DiRenzo et al., 2000). These studies raise the interesting possibility that Axl serves as the pivot of epigenetic regulation in oncogenesis and metastasis because its downstream factor Brg-1 is a major participator in histone modification and DNA methylation. In summary, Axl enhances cancer cell invasion, at least in part, through the transcriptional regulation of the MMP-9 gene. This upregulation is dependent on activation of the ERK and NF-kB pathways, including nuclear translocation of the chromatin-remodeling factor Brg-1 and its subsequent recruitment to the MMP-9 promoter. Moreover, histone modifications on the MMP-9 promoter switch from the repressive status to the active condition, and this switch can be correlated with Axl appearance. The results of our study provide insight into a novel mechanism in which Axl plays a critical role in tumor progression.

reverse transcriptase (Invitrogen, Carlsbad, CA, USA) to preserve the relative mRNA profile and to produce a template suitable for PCR. Immunohistochemistry Human breast cancer tissue samples were collected from tissue array CBA2 (SuperBioChips, Seoul, Korea) and stained with anti-human Axl antibody (R&D Systems, Minneapolis, MN, USA) and anti-human MMP-9 antibody (Millipore, Billerica, MA, USA). Protein expression was analysed using immunohistochemistry as described previously (Shieh et al., 2005). Preparation of nuclear extracts, coimmunoprecipitation and western blot analysis Nuclear extracts were obtained according to the manufacturer’s instructions (NE-PER, Pierce, Rockford, IL, USA). A 25 mg sample of each lysate was subjected to electrophoresis using 8% SDS–polyacrylamide gel electrophoresis and immunoblotted with anti-Axl (R&D Systems), anti-phospho-Akt (ser473), anti-Akt, anti-phospho-ERK1/2, anti-ERK1/2, antiphospho-IkB-a, anti-IkB-a (Cell Signaling Technology, Danvers, MA, USA), anti-sp1 (Upstate Biotechnology Inc., Lake Placid, NY, USA), anti-a-tubulin (Lab Vision, Fremont, CA, USA), anti-NF-kB p65/p50, anti-c-Fos, anti-c-Jun, antiBrg-1 and anti-Brm (Santa Cruz Biotechnology, Santa Cruz, CA, USA). For the coimmunoprecipitation assay, 300 mg of cell lysates was prepared using a lysis assay buffer containing protease inhibitors. Immune complexes were collected onto protein G agarose beads (Upstate Biotechnology Inc.) that were prebound with anti-Gas6 antibody (Santa Cruz) and then washed six times with lysis buffer. Immune complex kinase assay An immune complex kinase assay was performed as described previously (Singh and Zhang, 2004). In vitro invasion assay The invasion ability was examined using 24-well insert-based assays (BD Biosciences, Franklin Lakes, NJ, USA) and performed as described previously (Shieh et al., 2005). Reagents Wortmannin, PD98059, SB202190, JNK inhibitor I and rapamycin were purchased from Calbiochem (La Jolla, CA, USA). LY294002 and U0126 were obtained from Cell Signaling Technology (Beverly, MA, USA).

Materials and methods Cell lines, DNA constructs and transfections The human cancer cell lines NCIH-520 (lung), CL1-5F4 (lung), HT1376 (bladder), MDA-MB-415, MCF-7 (breast) and two human lung adenocarcinoma cell lines with different invasive capabilities (CL1-0, weakly invasive; CL1-5F4, highly invasive) were grown and derived as described previously (Tai et al., 2007). Construction of full-length human Axl has been described previously (Shieh et al., 2005). Three clones (V2HS_234991, V2HS_232535 and V2HS_238359) of shRNA targeting Axl (NM_021913) were obtained from GenDiscovery (Open Biosystem, Huntsville, AL, USA) and transfected into CL1-5F4 cells. RNA extraction and RT–PCR Total cellular RNA was extracted, purified and converted to cDNA using oligo d(T)12–18 primers and SuperScript III

Luciferase reporter assay An luciferase reporter assay was performed as described previously (Ma et al., 2004). Zymographic assay Cells were then washed and transfected for 6 h. After 6 h, cells were placed in serum-free media for 24 h. An zymographic assay was performed as described previously (Yao et al., 2001). ELISA-based NF-kB activity assay An ELISA-based NF-kB activity assay was performed as described previously (Jin et al., 2005). Chromatin immunoprecipitation assays The ChIP assay was performed according to the manufacturer’s instructions (EZ ChIP Assay Kit, Upstate Biotechnology Inc.). Oncogene


4054 Immunoprecipitated DNA was amplified by a primer pair corresponding to a 164 bp fragment (658 to 495) from the human MMP-9 promoter.

Statistical analysis All experiments were performed in three or more independent assays, which yielded highly comparable results. Data are summarized as mean±standard deviation (s.d.). Statistical analysis of the results was performed by Student’s t-test for unpaired samples. A P-valueo0.05 was considered significant.

Acknowledgements This study was supported by the National Health Research Institutes (Zhunan, Taiwan) grant no. 96A1-CASP01-014 (to C-W Wu). We thank Dr Christian Muchardt (Institute Pasteur, Paris) and Dr Weidong Wang (National Institute on Aging, Baltimore) for providing the BJ-5-Brg-1 and BJ-5-Brg-1(K798R) plasmids; Dr Anthony N Imbalzano for contacting Dr Muchardt and Dr Wang and Dr Jian Jian Li (Purdue University, Indiana) for providing the pCEP4–ERK1(K71R), pCEP4–ERK2(K52R) and pEGFP–IkBa(S32A, S36A) plasmids.

References Banine F, Bartlett C, Gunawardena R, Muchardt C, Yaniv M, Knudsen ES et al. (2005). SWI/SNF chromatin-remodeling factors induce changes in DNA methylation to promote transcriptional activation. Cancer Res 65: 3542–3547. Bittner M, Meltzer P, Chen Y, Jiang Y, Seftor E, Hendrix M et al. (2000). Molecular classification of cutaneous malignant melanoma by gene expression profiling. Nature 406: 536–540. Budagian V, Bulanova E, Orinska Z, Thon L, Mamat U, Bellosta P et al. (2005). A promiscuous liaison between IL-15 receptor and Axl receptor tyrosine kinase in cell death control. EMBO J 24: 4260–4270. Chung BI, Malkowicz SB, Nguyen TB, Libertino JA, McGarvey TW. (2003). Expression of the proto-oncogene Axl in renal cell carcinoma. DNA Cell Biol 22: 533–540. Craven RJ, Xu LH, Weiner TM, Fridell YW, Dent GA, Srivastava S et al. (1995). Receptor tyrosine kinases expressed in metastatic colon cancer. Int J Cancer 60: 791–797. Demarchi F, Verardo R, Varnum B, Brancolini C, Schneider C. (2001). Gas6 anti-apoptotic signaling requires NF-kappa B activation. J Biol Chem 276: 31738–31744. Dhanalakshmi S, Singh RP, Agarwal C, Agarwal R. (2002). Silibinin inhibits constitutive and TNFalpha-induced activation of NFkappaB and sensitizes human prostate carcinoma DU145 cells to TNFalpha-induced apoptosis. Oncogene 21: 1759–1767. DiRenzo J, Shang Y, Phelan M, Sif S, Myers M, Kingston R et al. (2000). BRG-1 is recruited to estrogen-responsive promoters and cooperates with factors involved in histone acetylation. Mol Cell Biol 20: 7541–7549. Fingleton B. (2006). Matrix metalloproteinases: roles in cancer and metastasis. Front Biosci 11: 479–491. Fridell YW, Jin Y, Quilliam LA, Burchert A, McCloskey P, Spizz G et al. (1996). Differential activation of the Ras/extracellular-signalregulated protein kinase pathway is responsible for the biological consequences induced by the Axl receptor tyrosine kinase. Mol Cell Biol 16: 135–145. Genersch E, Hayess K, Neuenfeld Y, Haller H. (2000). Sustained ERK phosphorylation is necessary but not sufficient for MMP-9 regulation in endothelial cells: involvement of Ras-dependent and -independent pathways. J Cell Sci 113(Part 23): 4319–4330. Hafizi S, Dahlback B. (2006). Signalling and functional diversity within the Axl subfamily of receptor tyrosine kinases. Cytokine Growth Factor Rev 17: 295–304. Hasanbasic I, Cuerquis J, Varnum B, Blostein MD. (2004). Intracellular signaling pathways involved in Gas6-Axl-mediated survival of endothelial cells. Am J Physiol Heart Circ Physiol 287: H1207–H1213. Holland SJ, Powell MJ, Franci C, Chan EW, Friera AM, Atchison RE et al. (2005). Multiple roles for the receptor tyrosine kinase axl in tumor formation. Cancer Res 65: 9294–9303. Ito T, Ito M, Naito S, Ohtsuru A, Nagayama Y, Kanematsu T et al. (1999). Expression of the Axl receptor tyrosine kinase in human thyroid carcinoma. Thyroid 9: 563–567. Jacob AN, Kalapurakal J, Davidson WR, Kandpal G, Dunson N, Prashar Y et al. (1999). A receptor tyrosine kinase, UFO/Axl, and other genes isolated by a modified differential display PCR are Oncogene

overexpressed in metastatic prostatic carcinoma cell line DU145. Cancer Detect Prev 23: 325–332. Jin S, Lu D, Ye S, Ye H, Zhu L, Feng Z et al. (2005). A simplified probe preparation for ELISA-based NF-kappaB activity assay. J Biochem Biophys Methods 65: 20–29. Lakka SS, Jasti SL, Gondi C, Boyd D, Chandrasekar N, Dinh DH et al. (2002). Downregulation of MMP-9 in ERK-mutated stable transfectants inhibits glioma invasion in vitro. Oncogene 21: 5601–5608. Lay JD, Hong CC, Huang JS, Yang YY, Pao CY, Liu CH et al. (2007). Sulfasalazine suppresses drug resistance and invasiveness of lung adenocarcinoma cells expressing AXL. Cancer Res 67: 3878–3887. Liu Y, Denlinger CE, Rundall BK, Smith PW, Jones DR. (2006). Suberoylanilide hydroxamic acid induces Akt-mediated phosphorylation of p300, which promotes acetylation and transcriptional activation of RelA/p65. J Biol Chem 281: 31359–31368. Ma Z, Shah RC, Chang MJ, Benveniste EN. (2004). Coordination of cell signaling, chromatin remodeling, histone modifications, and regulator recruitment in human matrix metalloproteinase 9 gene transcription. Mol Cell Biol 24: 5496–5509. Melaragno MG, Fridell YW, Berk BC. (1999). The Gas6/Axl system: a novel regulator of vascular cell function. Trends Cardiovasc Med 9: 250–253. Meric F, Lee WP, Sahin A, Zhang H, Kung HJ, Hung MC. (2002). Expression profile of tyrosine kinases in breast cancer. Clin Cancer Res 8: 361–367. Nakano T, Tani M, Ishibashi Y, Kimura K, Park YB, Imaizumi N et al. (2003). Biological properties and gene expression associated with metastatic potential of human osteosarcoma. Clin Exp Metastasis 20: 665–674. Nemoto T, Ohashi K, Akashi T, Johnson JD, Hirokawa K. (1997). Overexpression of protein tyrosine kinases in human esophageal cancer. Pathobiology 65: 195–203. Neubauer A, O’Bryan JP, Fiebeler A, Schmidt C, Huhn D, Liu ET. (1993). Axl, a novel receptor tyrosine kinase isolated from chronic myelogenous leukemia. Semin Hematol 30: 34. Roberts CW, Orkin SH. (2004). The SWI/SNF complex—chromatin and cancer. Nat Rev Cancer 4: 133–142. Sasaki T, Knyazev PG, Clout NJ, Cheburkin Y, Gohring W, Ullrich A et al. (2006). Structural basis for Gas6-Axl signalling. EMBO J 25: 80–87. Shieh YS, Lai CY, Kao YR, Shiah SG, Chu YW, Lee HS et al. (2005). Expression of axl in lung adenocarcinoma and correlation with tumor progression. Neoplasia 7: 1058–1064. Singh RA, Zhang JZ. (2004). Differential activation of ERK, p38, and JNK required for Th1 and Th2 deviation in myelinreactive T cells induced by altered peptide ligand. J Immunol 173: 7299–7307. Sun W, Fujimoto J, Tamaya T. (2004). Coexpression of Gas6/Axl in human ovarian cancers. Oncology 66: 450–457. Tai KY, Shiah SG, Shieh YS, Kao YR, Chi CY, Huang E et al. (2007). DNA methylation and histone modification regulate silencing of epithelial cell adhesion molecule for tumor invasion and progression. Oncogene 26: 3989–3997.


4055 Vajkoczy P, Knyazev P, Kunkel A, Capelle HH, Behrndt S, von Tengg-Kobligk H et al. (2006). Dominant-negative inhibition of the Axl receptor tyrosine kinase suppresses brain tumor cell growth and invasion and prolongs survival. Proc Natl Acad Sci USA 103: 5799–5804. van Ginkel PR, Gee RL, Shearer RL, Subramanian L, Walker TM, Albert DM et al. (2004). Expression of the receptor tyrosine kinase Axl promotes ocular melanoma cell survival. Cancer Res 64: 128–134. Wang W. (2003). The SWI/SNF family of ATP-dependent chromatin remodelers: similar mechanisms for diverse functions. Curr Top Microbiol Immunol 274: 143–169.

Wu CW, Li AF, Chi CW, Lai CH, Huang CL, Lo SS et al. (2002). Clinical significance of AXL kinase family in gastric cancer. Anticancer Res 22: 1071–1078. Yao J, Xiong S, Klos K, Nguyen N, Grijalva R, Li P et al. (2001). Multiple signaling pathways involved in activation of matrix metalloproteinase-9 (MMP-9) by heregulin-beta1 in human breast cancer cells. Oncogene 20: 8066–8074. Zeigler ME, Chi Y, Schmidt T, Varani J. (1999). Role of ERK and JNK pathways in regulating cell motility and matrix metalloproteinase 9 production in growth factor-stimulated human epidermal keratinocytes. J Cell Physiol 180: 271–284.

Supplementary Information accompanies the paper on the Oncogene website (http://www.nature.com/onc).

Oncogene