Oncogene (2012) 31, 2461–2470 & 2012 Macmillan Publishers Limited All rights reserved 0950-9232/12 www.nature.com/onc ORIGINAL ARTICLE Hypoxia and MITF control metastatic behaviour in mouse and human melanoma cells

Y Cheli1,2, S Giuliano1,2, N Fenouille1,2, M Allegra1,2, V Hofman3, P Hofman2,3, P Bahadoran1,2,4, J-P Lacour2,4, S Tartare-Deckert1,2,4, C Bertolotto1,2,4 and R Ballotti1,2,4

1INSERM U895, Centre Me´diterrane´en de Me´decine Mole´culaire, Equipe 1, Biology and Pathologies of Melanocytes, Equipe Labellise´e par la Ligue Contre le Cancer, Nice, France; 2Universite´ de Nice-Sophia Antipolis, UFR me´decine, Nice, France; 3ERI21/EA 4319 and Human Biobank, Nice, France and 4CHU NICE, Department of Dermatology, Nice, France

Melanomas are very aggressive neoplasms with notorious Introduction resistance to therapeutics. It was recently proposed that the remarkable phenotypic plasticity of melanoma cells Although melanoma is a relatively rare cancer, it is the allows for the rapid development of both resistance to leading cause of death from skin cancer, and its chemotherapeutic drugs and invasive properties. Indeed, incidence continues to increase in the Caucasian the capacity of melanoma cells to form distant metastases population. If diagnosed early, a large surgical resection is the main cause of mortality in melanoma patients. is generally sufficient to cure more than 90% of Therefore, the identification of the mechanism controlling melanoma. However, if the primary lesion is too thick melanoma phenotype is of paramount importance. In the (44 mm) or if melanoma cells have metastasised to present report, we show that deletion of microphthalmia- regional lymph nodes or to distant organs, the prognosis associated transcription factor (MITF), the master gene is extremely pejorative, because existing anti-melanoma in melanocyte differentiation, is sufficient to increase the therapeutic regiments have a very low survival rate. The metastatic potential of mouse and human melanoma cells. notorious aggressiveness of melanoma is due to its MITF silencing also increases fibronectin and Snail, two remarkable metastatic propensity. Strikingly, metastatic mesenchymal markers that might explain the increased dissemination from small primary lesions (less than invasiveness in vitro and in vivo. Furthermore, ablation of 1 mm) is not an exceptional clinical event (Shaw et al., this population by Forskolin-induced differentiation or 1987; Bedrosian et al., 2000; Corsetti et al., 2000). These MITF-forced expression significantly decreases tumour observations suggest that melanomas acquire their and metastasis formation, suggesting that eradication of invasive and metastatic behaviours very early and low-MITF cells might improve melanoma treatment. without the absolute necessity to accumulate additional Moreover, we demonstrate that a hypoxic microenviron- oncogenic mutations favouring distant tumour develop- ment decreases MITF expression through an indirect, ment. In fact, it has been proposed that melanoma cells hypoxia-inducible factor 1 (HIF1)a-dependant transcrip- intrinsically possess a genetic programme predisposing tional mechanism, and increases the tumourigenic and them to exacerbated metastatic features (Gupta et al., metastatic properties of melanoma cells. We identified 2005). The same programme is implemented by mela- Bhlhb2, a new factor in melanoma biology, as the nocyte precursors during their differentiation process to mediator of hypoxia/HIF1a inhibitory effect on MITF allow their migration from neural crest to skin during expression. Our results reveal a hypoxia–HIF1a– embryonic development. BHLHB2–MITF cascade controlling the phenotypic As is the case for many other neoplasms, physiolo- plasticity in melanoma cells and favouring metastasis gical differentiation of melanocytes and metastatic development. Targeting this pathway might be helpful in melanoma development share common signalling path- the design of new anti-melanoma therapies. ways, such as ERK and PI3 kinase pathways (Busca and Oncogene (2012) 31, 2461–2470; doi:10.1038/onc.2011.425; Ballotti, 2000). However, the most remarkable connec- published online 26 September 2011 tion between these two processes is the microphthalmia- associated transcription factor (MITF). Indeed, MITF Keywords: metastasis; MITF; mesenchymal transition; was identified as the master gene in melanocyte hypoxia; Bhlhb2; differentiation differentiation, which it does by controlling the expres- sion of melanogenic enzymes (TYR, TYRP1, DCT). Later, MITF was shown to regulate the expression of genes involved in all steps of melanocyte differentiation, Correspondence: Dr R Ballotti, INSERM U895, Equipe 1, Biology including dendricity (DIA), melanosome biogenesis and Pathologies of Melanocytes, Team 1, 151 route de Saint-Antoine (SILVER, OA1) and melanosome transport (RAB27a) de Ginestie` re, Nice, F-06204 Cedex 3, France. E-mail: [email protected] (for review, see Cheli et al. (2010)). Concomitantly, Received 3 April 2011; revised 9 August 2011; accepted 19 August 2011; King et al. (2001) reported that MITF expression published online 26 September 2011 was maintained in more than 85% of melanoma, and Hypoxia controls MITF and invasiveness in melanoma cells Y Cheli et al 2462 a MITF gene amplification has been found in more than Results 20% of metastatic melanomas. MITF was shown to control the expression of genes involved in cell survival MITF silencing exacerbates the metastatic properties in (BCL2, BIRC7) and in cell cycle control (CDK2; Cheli B16 melanoma cells et al., 2010). Moreover, MITF silencing blocked We previously showed that injection of MITF-silenced melanoma cells in G0/G1 (Du et al., 2004; Carreira B16 cells in syngeneic mice (C57bl6) gave rise to larger et al., 2006) and induced senescence (Giuliano et al., tumours compared with cells transfected with scrambled 2010). Taken together, these observations provide small interfering RNA, indicating that MITF silencing compelling support for a key role of MITF in favouring increases the tumourigenic potential of melanoma cells melanoma development. (Cheli et al., 2011). However, the real clinical burden of Remarkably, most primary melanomas and skin melanomas is due to their elevated metastatic potential. metastases are heavily pigmented. Furthermore, Silver Therefore, we studied the effect of MITF depletion on and MART1, two melanoma differentiation markers, the metastatic behaviours of B16 melanoma cells. are used by the pathologists to identify melanoma First, we showed by fluorescence-activated cell sorting lesions. Therefore, in agreement with persistent MITF analysis and western blot (Figures 1a and b) that expression, melanomas frequently appear to be com- siMITF transfection in B16F10 melanoma cells led to an mitted to the differentiation process. almost complete loss of MITF. Then, as previously However, this new dogma qualifying MITF, the first shown in human melanoma cells 501Mel and SKMel28 melanocyte differentiation marker, as a melanocyte (Carreira et al., 2006), MITF depletion increased both lineage-restricted oncogene is somehow paradoxical to the migration and the invasion properties of B16 an older and more universal doctrine stipulating that melanoma cells in vitro. Taken together the above tumour differentiation is inversely correlated with its observations suggest that MITF silencing might favour aggressiveness. It should be mentioned that several metastasis development in vivo, but this hypothesis has reports suggest that MITF might have an anti-melano- never been demonstrated. To verify this hypothesis, lung ma activity. Indeed, MITF overexpression prevented colonisation and visceral metastasis were analysed after xenograft tumour development (Carreira et al., 2006; injection in the tail vein of C57bl6 mice of increasing Cheli et al., 2011). It was also reported that MITF number of siScrambled- or siMITF-transfected B16 upregulated the expression of two cyclin kinase inhibi- cells. Notably, we observed no increase in lung tors, CDKN2A (Loercher et al., 2005) and CDKN1A melanoma tumour development in mice injected with (Carreira et al., 2005), and MITF depletion was 0.25 Â 106 MITF-depleted B16 cells. However, after demonstrated to increase melanoma cell migration injection of lower number of cells, we observed that the in vitro (Carreira et al., 2006). siMITF-transfected B16 cells gave rise to a larger To reconcile these apparently contradictory sets of number of lung tumour. The number of visceral data, a pertinent and elegant model was proposed metastases (ovaries, kidneys, intestine and heart) sig- (Carreira et al., 2006) in which MITF would function nificantly increased in mice that received siMITF- as a rheostat to control melanoma cell phenotype transfected B16 cells, regardless the number of injected and behaviour, allowing a reversible switch from cells (Figure 1d). Histological analysis of these metas- poorly differentiated and highly metastatic cells to fast tases after haematoxylin and eosin staining showed that growing progeny. This model was validated by our in tumours, arising from siMITF-transfected cells, there recent data (Cheli et al., 2011) demonstrating that a was a larger haemorrhagic area, with a presence of transiently slow growing and MITF-low population necrotic cells and a small islet of normal cells around the existed in melanoma cell lines and in melanoma cells, vessels. Higher magnification showed hyperchromatic freshly isolated from human biopsy specimens. This nuclei with a low mitotic rate in control metastases. In MITF-low population has exacerbated tumourigenic siMITF metastases, we observed clear nuclei, with properties, whereas the MITF-high population inconspicuous nucleoli and numerous mitotic figures grew faster, but displayed poor tumourigenicity when (Figure 1e). Therefore, MITF depletion exacerbates the injected in mice. metastatic behaviour of B16 melanoma cells in vivo. In the present study, we show that MITF silencing increased the invasive and metastatic properties of melanoma cells. Furthermore, we demonstrate that MITF silencing favours the mesenchymal phenotype in hypoxia inhibits MITF expression and increases the B16 melanoma cells tumourigenic and metastatic potential of melanoma Cell motility and metastasis are positively correlated cells. Taken together, our results indicate that the with the mesenchymal phenotype (Radisky, 2005; downregulation of MITF by hypoxia is a key molecular Alonso et al., 2007). Therefore, we next examined the event that controls the metastatic properties of melano- effect of MITF on mesenchymal markers. Quantitative ma cells. In contrast with the dogma ascribing oncogenic PCR on MITF-depleted cells showed a two- to threefold activity to MITF and suggesting that inhibition of increase in Snai1, Fn1 and Sparc expression. However, MITF is a promising way to cure melanoma, our data we observed a decrease in Snai2 expression (Figure 2a). show that inhibition of MITF favours tumour and Western blotting confirmed the increase in Snai1, Fn1 metastasis development and might be deleterious in and Sparc expression at the protein level (Figure 2b). melanoma treatment. Consistently, Snai2 protein expression was inhibited in

Oncogene Hypoxia controls MITF and invasiveness in melanoma cells YCheliet al 2463

Figure 1 Metastatic properties increase in MITF-depleted B16 mouse melanoma cells. (a) B16 cells were transfected with siMITF or with the siScr. Cells were detached, fixed in 4% paraformaldehyde, stained with MITF antibody and analysed by flow cytometry for MITF content (abscissa) and forward light scatter (ordinate). (b) Western blot analysis of MITF protein expression in cells transfected with siMITF or with siScr. (c, Upper panel) Transwell migration assays of B16 cells transfected with siScr or siMITF. Results are the mean of three independent experiments ±s.d.; Po0.05. (Lower panel) Transwell invasion assays of B16-F10 cells transfected with siScr or siMITF. Results are the mean of three independent experiments ±s.d.; Po0.05. (d) siScr- or siMITF-transfected B16 cells were injected into the tail vein. Fourteen days later, lung tumours and visceral metastases were counted. Results are expressed as mean±s.e.m. (n ¼ 8animals;*Po0.05, **Po0.01). (e) Hematoxylin/Eosin staining of siScr and siMITF visceral metastases, arrows pointing numerous mitotic figures in siMITF metastasis.

Figure 2 MITF silencing induces the mesenchymal transition of mouse melanoma cells. (a) MITF, Snai1, Snai2, Sparc and fibronectin mRNA levels from siScr or siMITF-transfected B16 cells were measured by quantitative PCR, ±s.d. (b) Western blot analysis of MITF, Snai1, Snai2, Sparc and fibronectin protein expression in cells transfected with siScr or siMITF. Hsp60 was used as the loading control. (c) Secreted Sparc and fibronectin were quantified in the medium of siMITF- and siScr-transfected B16 cells, ±s.d.

Oncogene Hypoxia controls MITF and invasiveness in melanoma cells Y Cheli et al 2464 MITF-silenced cells. Of note, Fn1 and Sparc levels were developed more slowly (Figure 3b). Furthermore, after also increased in the cell culture medium (Figure 2c), injection in the tail vein, we observed a decrease in lung indicating that these proteins are correctly processed tumours and fewer visceral metastases in Forskolin- and secreted. MITF silencing increases the mesenchymal treated mice (Figures 3c and d). phenotype in B16 melanoma cells. Analysis of tumour cells showed a modest global increase in MITF expression and a significant decrease in the MITF-negative population in the tumours from Pharmacological depletion of the MITF-negative mice treated with Forskolin (Supplementary Figure 1). population in mouse melanoma cells inhibits tumour and Of note, adenovirus-driven MITF overexpression metastasis development (Figure 3e) was sufficient to decrease the tumour volume Interestingly, a MITF-negative population (ranging after subcutaneous injection (Figure 3f). Collectively, from 1 to 2.5%) exists spontaneously in B16 melanoma these results demonstrate that the level of MITF is a key cell lines as shown by fluorescence-activated cell sorting parameter in determining the tumourigenic and meta- analysis (Figure 3a). Therefore, we asked whether forced static properties of B16 melanoma cells. The low-MITF melanoma cell differentiation by Forskolin treatment population displays exacerbated metastatic behaviours. would ablate the MITF-low population and impair tumour and metastasis development. Treatment of B16 melanoma cells with Forskolin for 48 h led to a MITF regulates mesenchymal phenotype and invasiveness depletion of the MITF-low population (Figure 3a). in human melanoma cells Note that the global level of MITF did not change after We determined whether MITF silencing induced a 48 h of Forskolin treatment. This is in agreement with a mesenchymal phenotype and increased the metastatic previous report (Bertolotto et al., 1998) demonstrating potential in the MeWo human melanoma cells. MITF that a cAMP-elevating agent induced a transient depletion increased mesenchymal markers, such as increase in MITF expression that returned to the FN1, SPARC, SNAI1, at the protein level, whereas basal level after 48 h. When B16 cells were injected SNAI2 was inhibited (Supplementary Figure 2A). into Forskolin-treated mice, subcutaneous tumours Moreover, MITF silencing increased lung colonisation

Figure 3 Forskolin induces depletion of MITF-negative cells and prevents the development of mouse melanoma tumours. (a) B16 cells were treated for 48 h with 20 mM Forskolin (Fsk) or dimethyl sulfoxide (DMSO). Cells were detached, fixed in 4% paraformaldehyde, stained with MITF antibody and analysed by flow cytometry for MITF content (abscissa) and forward light scatter (ordinate). (b) Mice were inoculated subcutaneously with B16 melanoma cells (0.5 Â 106) and were treated with Fsk (82 mg per mice 30 min prior injection, and then every 48 h) or phosphate-buffered saline/5% dimethyl sulfoxide (n ¼ 6 in each group). The tumour growth curves were determined by measuring the tumour volume each day, starting 5 days after injection, ±s.e.m. (c, d) B16 melanoma cells were injected into the tail vein. Fifteen days after, mice were killed; the number of lung tumours (c) and visceral metastases (d) was counted and expressed as the mean±s.e.m. (n ¼ 6 mice per group, *Po0.05). Representative pictures of lung metastases are displayed at the top of graph c.(e) Western blots analysis of MITF and Erk2 in B16 cells infected with Ad (adenovirus) control or Ad (adenovirus) MITF. (f) The same infected cells (0.1 Â 106 cells in 200 ml of phosphate-buffered saline) were injected subcutaneously. Tumour volume was measured 30 days after injection. Results are the mean of volume (in mm3)±s.e.m. (n ¼ 6 in each group); *Po0.05 and **Po0.01.

Oncogene Hypoxia controls MITF and invasiveness in melanoma cells YCheliet al 2465 and the number of visceral metastases (Supplementary creased markedly in the three cell lines (Figure 4a). Figure 2B), when injected in the tail vein. MITF decrease was also observed upon treatment with Next, we observed that Forskolin was also able to cobalt, a hypoxia mimetic that inhibits proline hydro- deplete the MITF-low population in the MeWo cells xylase and consequently increases HIF1a expression after 48 h of treatment (Supplementary Figure 2C). In (Figure 4b). These results allow us to conclude that addition, Forskolin treatment of nude mice injected with hypoxic conditions repress MITF expression in mela- MeWo cells led to a significant decrease in subcutaneous noma cells. Next, we checked whether MITF regulation tumour volume (Supplementary Figure 2D) and metas- under hypoxia was due to transcriptional regulation. tasis number (Supplementary Figure 2E). Overexpres- Treatment with cobalt led to a 60% decrease in the sion of MITF in MeWo cells (Supplementary Figure 2F, transcriptional activity of a 2-kb fragment of the MITF left panel) also drastically inhibited formation of promoter (p-MITF; Figure 4c). Forskolin induced a tumours after injection in nude mice (Supplementary twofold increase in the transcriptional activity of this Figure 2F, right panel). Taken together, these data fragment. Then, using two shorter promoter fragments, demonstrated that loss of MITF leads to a mesenchymal p-MITF D1 (1.5 kb) and p-MITF D2 (1.1 kb), we phenotype and increases the metastatic potential in observed that these fragments were still stimulated by human melanoma cells. Forskolin but failed to respond to hypoxia-mimetic treatment. These data indicate that a hypoxia-respon- sive element is located in the p-MITF between À1.5 and Hypoxic conditions induce a loss of MITF À2 kb (downstream transcription start site). It has been reported that hypoxia is an essential cofactor As HIF1a has a pivotal role in the transcriptional for melanoma development (Bedogni et al., 2005). response to hypoxia, we checked whether the decrease in Therefore, we studied the effect of a hypoxic challenge p-MITF activity was mediated by HIF1a. First, we on MITF expression, and on the metastatic potential of observed that HIF1a silencing prevented the inhibition melanoma cells. Upon exposure of B16, SkMel-28 and of the p-MITF and the activation of the pHRE by Mewo cells to a hypoxic environment (1% oxygen), we cobalt treatment, indicating that the inhibition of MITF observed an increase in hypoxia-inducible factor 1 transcription upon hypoxia is dependant on HIF1a (HIF1)a expression; whereas, MITF expression de- (Figure 4d). However, when we co-transfected p-MITF,

Figure 4 Hypoxia induces an inhibition of MITF expression. (a) Western blot analysis of MITF and HIF1a expression in cells under normal oxygen (Ct) or hypoxia condition (1% oxygen, H), ERK2 was used as loading control. (b) Western blot analysis of MITF and HIF1a expression in cells treated or not with 200 mM of CoCl2 for 24 h. ERK2 was used as loading control. (c) B16 cells were transfected with MITF reporter constructs (2 kb promoter fragment (p-MITF), 1.5 kb promoter fragment (p-MITF D1) and 1.1 kb promoter fragment (p-MITF D2). Cells were then exposed for 24 h to 200 mM CoCl2,or10mM Forskolin (as a positive control). The results are the mean±s.d. of three independent experiments. (d) The effect of CoCl2 on p-MITF activity was assayed after transfection with small interfering RNA to HIF1a or overexpression of a stabilised HIF1a mutant (HIF2m). The results are the mean±s.d. of three independent experiments. Hypoxia-response element plasmid (p-HRE) was used as a positive control for hypoxia mimicry and HIF1a activity.

Oncogene Hypoxia controls MITF and invasiveness in melanoma cells Y Cheli et al 2466 with a stabilised HIF1a mutant, the p-MITF activity was not affected by HIF1a overexpression, whereas p- HRE activity was clearly stimulated. Taken together, these data indicate that MITF downregulation under hypoxia is dependant on HIF1a, but that HIF1a does not directly regulate the p-MITF. To confirm that HIF1a is not a direct regulator of p-MITF, we performed a chromatin immunoprecipitation (ChIP) assay and PCR amplification with specific primers of the p-MITF, on total DNA extract or after immunopreci- pitation with antibody to polymerase II, as a positive control, we amplified a 370-bp fragment (Supplementary Figure 3). After immunoprecipitation with anti-HIF1a, no band was amplified under normal oxygen or hypoxia conditions. Conversely, using specific erythropoietin promoter primers (Supplementary Figure 3, lower panel), we amplified 700-bp fragment corresponding to the erythropoietin promoter in the total extract, the anti-Pol II precipitates, and only under hypoxia condi- tion after immunoprecipitation with anti-HIF1a anti- body. These results confirm that HIF1a does not bind to the p-MITF. Furthermore, we performed immunostaining of hu- man tumour samples for MITF and HIF1a (Supple- mentary Figure 4). We observed no HIF1a expression in the close vicinity of the vessels. HIF1a labelling increased with distance from the blood vessels. Con- versely, MITF labelling was strong around the vessels and decreased in the remote regions, supposed to be hypoxic. These observations indicate that in some tumour regions, an inverse correlation between MITF Figure 5 Bhlhb2 is upregulated by hypoxia and inhibits MITF and HIF1a can be observed, in agreement with a expression. (a) Quantitative PCR analysis of Bhlhb2, Sap30, Atf3 regulation of MITF expression, in vivo, by hypoxia and and Mxi1 mRNA levels from cells under normal oxygen (Ct, black) HIF1a. or after 24 h of hypoxia (H, white), ±s.d. (b) Quantitative PCR analysis of Bhlhb2, Sap30, Atf3 and Mxi1 mRNA levels from cells under control conditions (Ct, black) or treated with CoCl2 (white), ±s.d. (c) B16F10 cells were transfected with small interfering RNA Bhlhb2 is induced by hypoxia and represses MITF against Bhlhb2 (siBhlhb2) at two different concentration or with expression the scrambled sequence (siScr). Bhlhb2 and MITF mRNA from Transcription factors Bhlhb2, Sap30, Atf3 and Mxi1 transfected cells under control conditions (black) or treated with CoCl2 (grey) was quantified by quantitative PCR. (d) B16 cells were were reported to be regulated upon exposure of B16 co-transfected with siBhlhb2 or with siScr and MITF reporter melanoma cells to hypoxic conditions (Olbryt et al., constructs (2 kb promoter fragment, p-MITF). Cells were either 2006). Therefore, we examined the effect of cobalt or then exposed to 200 mM CoCl2 or were not exposed, for 24 h, and hypoxia on these transcription factors. Quantitative p-MITF activity was quantified by luciferase activity. Results are PCR showed a consistent threefold increase in the mean of three independent experiments, ±s.d.; *Po0.05 and P Bhlhb2 expression by hypoxia (Figure 5a) or cobalt ** o0.01. (Figure 5b). Sap30 was slightly upregulated by hypoxia, but not by cobalt, whereas Atf3 was strongly stimulated Figure 5A), and Bhlhb2 overexpression repressed the upon cobalt treatment, but not by hypoxia. No p-MITF activity (Supplementary Figure 5B). significant regulation of Mxi1 was observed in both We also checked for Brn-2, as it was reported to conditions. repress MITF expression. Brn-2 was not affected by Therefore, we focused our attention on the role of cobalt and was slightly downregulated by hypoxia Bhlhb2 in the regulation MITF under hypoxia condi- (Supplementary Figure 6) in B16 cells. This observation tions. Quantitative PCR analysis on siBhlhb2-trans- is in agreement with our previous report demonstrating fected cells showed a strong inhibition of the Bhlhb2 that BRN2 was not increased in the MITF-negative messengers that led to an increase in MITF expression population (Cheli et al., 2011). We also observed that and prevented the downregulation of MITF expression Atf3 silencing (Supplementary Figure 7A) did not under hypoxia conditions (Figure 5c). Knockdown of prevent the inhibition of the p-MITF activity induced Bhlhb2 under hypoxia rescued p-MITF activity by cobalt (Supplementary Figure 7B). Taken together, (Figure 5d). Chromatin immunoprecipitation assay, these data show that the increase in Bhlhb2 expression with antibody to Bhlhb2 showed the binding of this under hypoxia is responsible for the decrease of MITF transcription factor to the p-MITF (Supplementary in melanoma cells.

Oncogene Hypoxia controls MITF and invasiveness in melanoma cells YCheliet al 2467 and metastatic potential in vivo when injected in mice. Furthermore, forced differentiation upon activation of the cAMP pathway by Forskolin and ablation of the MITF-negative population drastically reduced melano- ma tumour and metastasis formation. Of course, Forskolin exerts pleiotropic biological effects in mela- noma cells, including the inhibition of the PI3 kinase/ AKT/mTOR pathway and the activation of the ERK pathway (Busca and Ballotti, 2000) that can also participate in the reduction of the tumourigenic potential of melanoma cells. In addition, the anti- melanoma action of Forskolin has been ascribed to a systemic effect of the compound (Agarwal and Parks, 1983). Nevertheless, it should be noted that the PI3 kinase and ERK pathways converge to MITF. As MITF-forced expression in both mouse and human melanoma cells significantly reduces tumour formation, the anti-tumourigenic effect of Forskolin can be attributed, at least in part, to the removal of MITF- negative cells. One can also suggest that very high MITF expression, which might be induced by Forskolin, causes the inhibition of tumour growth (Wellbrock and Marais, 2005; Carreira et al., 2006). However, it should be noted that fluorescence-activated cell sorting analysis of MITF expression in the tumours from mice treated with dimethyl sulfoxide or Forskolin (Supple- mentary Figure 2) showed an almost complete disap- pearance of the MITF-low population, strengthening the role of this population in the development of melanoma. Therefore, this approach which is reminiscent of the differentiation therapy used for leukaemia (Spira and Figure 6 Hypoxia increases the tumourigenic and metastatic Carducci, 2003), may provide a pharmacological strat- potential of melanoma cells. (a) Mice were inoculated subcuta- egy to improve the treatment of metastatic melanoma, neously with B16 melanoma cells (0.1 Â 106) incubated for 24 h at provided that they have a conserved cAMP pathway 1% O2 (hypoxia) or normal oxygen (Ct). The tumour growth was determined by measuring the tumour (mean volume (in from adenylate cyclase to MITF. mm3)±s.e.m.). (b, c) B16 melanoma cells (0.1 Â 106) incubated Furthermore, MITF silencing in melanoma cells for 24 h at 1% O2 (hypoxia) or normal oxygen (Ct) were injected increased the expression of fibronectin and SNAI1 into the tail vein of mice. Seventeen days later, mice were killed and (SNAIL), which are markers of mesenchymal pheno- the number of lung tumours (b) or visceral metastases was counted type, and are key factors in the acquisition of metastatic and expressed as the mean±s.e.m.; *P 0.05 and **P 0.01. o o properties (Alonso et al., 2007). However, SNAI2 (SLUG), another mesenchymal marker, which also has Hypoxia increases tumourigenic and metastatic potential a pivotal role in metastatic behaviour (Martin et al., of melanoma cells 2005), is downregulated in MITF-depleted cells. This Finally, we studied the effect of hypoxic conditions on observation can be explained by the fact that SNAI2 is a the tumourigenic and metastatic potential of B16 direct MITF target gene (Sanchez-Martin et al., 2002). melanoma cells. After subcutaneous injection, melano- Of note, SLUG has been reported to be required for ma cells exposed to hypoxia produced larger tumours, melanoma metastasis (Gupta et al., 2005). In our hands, compared with their counterparts kept in normal oxygen SLUG is not required for migration, invasion and conditions (Figure 6a). Furthermore, when injected in metastasis. Overexpression of SNAIL might compensate the tail vein, hypoxia-exposed cells formed more lung for the loss of SLUG. In agreement with this hypothesis, tumours and visceral metastases (Figures 6b and c). in 501Mel human melanoma cells transfected with Therefore, exposure of melanoma cells to hypoxic siMITF, we observed a loss of E-cadherin expression conditions favours their tumourigenic and metastatic (not shown). SPARC, which has a key role in melanoma properties. development (Ledda et al., 1997; Fenouille et al., 2011) and participates in the reinforcement of the mesenchy- mal phenotype (Robert et al., 2006), is also clearly Discussion upregulated in MITF-depleted cells. These observations demonstrate that MITF silencing favours the mesench- In this study, we showed that MITF-depleted mouse ymal phenotype, and acquisition of migration and and human melanoma cells have increased tumourigenic invasion properties that can explain the increased

Oncogene Hypoxia controls MITF and invasiveness in melanoma cells Y Cheli et al 2468 metastasis development observed within the MITF- and ChIP experiment confirms that HIF1a does not depleted cells. The mesenchymal phenotype was recently bind to this p-MITF region strengthening the likelihood demonstrated to be linked to the acquisition of cancer of an indirect effect of HIF1a on the p-MITF. stem cell properties (Mani et al., 2008), and we recently We identified Bhlhb2 (Bhlh40, Dec1, Stra13) as the showed that invalidation of MITF lead to an increase of mediator of hypoxic effect on MITF expression. Bhlhb2 stem cells markers, such as Oct4 and Nanog (Cheli et al., was consistently upregulated under hypoxic condition, 2011). Therefore, we conclude that MITF silencing and Bhlhb2 knockdown prevents the decrease of MITF generates the full spectrum of molecular responses that expression caused by hypoxia. Moreover, the À1.5 and are required to generate melanoma-initiating cells with À2 kb region of the p-MITF region contains three stem cell-like features. E-boxes that could bind Bhlhb2; one of these E-box is Noteworthy, MITF depletion causes a G0/G1 arrest conserved in the mouse promoter. Bhlhb2 has a known (Carreira et al., 2005; Giuliano et al., 2010).Therefore, it transcriptional repression activity towards numerous appears that slow growing or arrested melanoma cells target genes (Choi et al., 2008; Cho et al., 2009; Bhawal have an increased tumourigenic and metastatic poten- et al., 2011). Bhlhb2 is involved in the control of tial. This apparent contracditory observation is perfectly circadian rhythms (Honma et al., 2002; Nakashima in agreement with the report demonstrating that a slow- et al., 2008), and has an anti-apoptotic activity (Ehata growing population is required for melanoma develop- et al., 2007). Our data will open new research avenue to ment (Roesch et al., 2010; Cheli et al., 2011). However, investigate the role of this transcription factor in MITF was expressed in the smallest detectable tumours, melanoma development. even those arising from MITF-depleted cells (not Taken together, our data demonstrate that the loss of shown). These observations support the phenotype- MITF initiated by hypoxia, is a pivotal event in the switching hypothesis (Hoek et al., 2008; Hoek and development of melanoma tumours and metastases. The Goding, 2010), which suggests the existence of a effect of hypoxia may be due on one side to its reversible transition between invasive, slow-growing proangiogenic action and on the other side to its melanoma cells, and their more differentiated and fast- prometastatic action on melanoma by favouring phe- growing progeny. The switch between the two pheno- notype switching and acquisition of invasive properties. types is controlled by MITF, the master gene in melanocyte differentiation. The key question, then, was, how MITF is regulated Materials and methods within the melanoma tumours? Numerous extracellular signals, including WNT/DKK1, cAMP-elevating Cell culture agents, endothelin, transforming growth factor-b and B16-F10 mouse melanoma cells, and SkMel-28 human stem cell factor, regulate MITF expression. All of these melanoma cells were cultured in Dulbecco’s modified Eagle’s agents can therefore modulate the number of MITF-low medium (Invitrogen, Carlsbad, CA, USA) supplemented with cells and thus the metastatic behaviour of the melanoma 7% foetal bovine serum (Invitrogen) and penicillin/streptomy- cells. However, one of the most frequent tumour cin (100 IU/50 mg/ml (Invitrogen)). MeWo human melanoma conditions is hypoxia. Indeed, tumour regions, distant cells were grown in RPMI medium 1640 (Invitrogen), with the from supporting vasculature are hypoxic. We observed same supplements. Cells were transfected with small interfering RNA against MITF (50-r(GGUGGAUCGGAUCAUCAA that exposure of melanoma cells to hypoxic conditions 0 lead to an increase in HIF1a and a decrease in MITF GTT)d(TT)-3 ; Cheli et al., 2011), small interfering RNA against Bhlhb2 (Dharmacon, Lafayette, CO, USA), small expression. The hypoxia mimetic, CoCl2, inhibits MITF interfering RNA against Atf3 (Dharmacon), or a scrambled expression and decreases p-MITF activity. This effect is sequence using Lipofectamine RNAiMax (Invitrogen) accord- dependant on HIF1a, as small interfering RNA to ing to the manufacturer’s protocol as previously described HIF1a prevents the effect of hypoxia on both MITF- (Cheli et al., 2011). When indicated, cells were treated with and hypoxia-responsive promoters. cobalt (Sigma-Aldrich, St Louis, MO, USA) to mimic hypoxia. The key role of hypoxia and HIF1a in the regulation Hypoxic conditions were produced by incubation of cells in a of MITF expression is strengthened by immunochem- Whitley H35 hypoxystation (Don Whitley Scientific, West istry studies on human melanoma tumours, disclosing an Yorkshire, UK). The oxygen in this workstation was main- inverse correlation between HIF1a and MITF expres- tained at 1%, with the residual gas mixture being 93–94% sion in the perivascular region of two melanoma samples nitrogen and 5% CO2. All media for the hypoxic experiment were buffered with 25 mM Hepes (Invitrogen). out of three. However, further studies, with a larger number of samples, are needed to assess the extent to which this phenomenon exists. Note that in agreement Antibodies and reagents with this result, it has been reported that MITF showed a Antibodies against Bhlhb2, Erk2, GAPDH, Hsp60, Snail and consistent anti-correlation with Glut1, a hypoxia marker, SLUG were from Santa Cruz Biotechnology (Santa Cruz, CA, and a HIF1a target gene (Eichhoff et al., 2010). USA). Anti-fibronectin was from BD Biosciences (Franklin Lakes, NJ, USA). Anti-MITF (C5) was from Abcam However, HIF1a overexpression is not sufficient to (Pleasanton, CA, USA) and anti-HIF1a was from R&D inhibit the p-MITF activity. The hypoxia-responsive systems (Minneapolis, MN, USA). Secondary alexa 488 was element in the p-MITF is located between À1.5 and from Invitrogen. The HRE–LUC reporter vector contains À2 kb from the transcription start site. This region does three copies of the HRE from the erythropoietin promoter not contain an obvious HIF1a binding site (ACGTG), cloned upstream of a minimal thymidine kinase promoter and

Oncogene Hypoxia controls MITF and invasiveness in melanoma cells YCheliet al 2469 the luciferase reporter gene (Berra et al., 2003). Human p- NaCl and 1% Triton X-100, the secondary antibody coupled MITF cloned upstream of luciferase were previously described with horseradish peroxidase (Dakopatts, Glostrup, Denmark) (Bertolotto et al., 1998). The plasmid encoding a stabilised was then incubated for 1 h at 25 1C. After three additional HIF1a mutant was a generous gift from Dr Pouysse´gure washes, proteins of interest were revealed by enhanced (Ginouves et al., 2008). Flow cytometry was performed as chemiluminescence (Amersham, Uppsala, Sweden). previously described (Cheli et al., 2011). Reverse transcription and quantitative PCR In vivo mouse tumour model Total cell RNA was extracted using the RNAeasy miniprep kit Animal experiments were carried out in accordance with the (Qiagen, Valencia, CA, USA), according to the manufacturer’s Declaration of Helsinki Principles, and were approved by a instructions, and 1 mg of RNA was reverse amplified using Local Ethics Committee. Female immune-deficient athymic oligo dT, using the reverse transcription system (Promega, nude FOXN1nu (nude) mice and C57BL6J mice were obtained Madison, WI, USA) according to the manufacturer’s instruc- at 6 weeks of age from Harlan Laboratory (Gannat, France). tions. PCR was performed using the StepOnePlus real-time Animals were maintained in a temperature-controlled facility PCR system and the power SYBR green PCR master mix (22 1C) on a 12-h light/dark cycle, and were given free access to reagent (Applied Biosystems, Foster City, CA, USA). Relative food (standard laboratory chow diet from UAR, Epinay-S/ quantification of the amplicons was performed by 2(ÀDD CT) Orge, France). Mice were inoculated subcutaneously with method. Primer details are available upon request. melanoma cells (0.25  106 to 0.05  106 B16 F10 cells per injection in 200 ml of phosphate-buffered saline). The growth Chromatin immunoprecipitation assay tumour curves were determined by measuring the tumour B16F10 cells were cultured in 150-cm2 culture dishes volume using the equation V ¼ (L  W2)/2 as previously under normoxy, or treated with 200 mM of CoCl for 18 h. described (Botton et al., 2009). Mice were killed by CO 2 2 Briefly, cells were fixed, collected and ChIP was carried out using inhalation, and tumours were removed and frozen for EZ–ChIP Chromatin Immunoprecipitation Kit (Upstate, Jaffrey, immunofluorescence study or analysed by western blot. The NH, USA). After sonication, the sheared chromatin was metastasis model was obtained by intravenous injection into immunoprecipitated using the indicated antibody. Then, the the caudal vein of 0.25 106 B16 F10 cells or 1 106 MeWo   protein/DNA crosslinking complexes were reversed by heat per mouse in 100 ml of phosphate-buffered saline. Mice were treatment (65 1C overnight) and by proteinase K digestion. The treated with 82 mg Forskolin (Sigma-Aldrich; 4 mg/kg) by genomic-captured fragments were purified using spin columns. intraperitoneal injection 30 min before intravenous or sub- Identification of the captured DNA fragments was performed by cutaneous cells injection, and Forskolin injection was repeated PCR analysis using the MITF or erythropoietin promoter every 48 h. The growth tumour curves were determined by primers (primer details available upon request). Thirty-five cycles measuring the tumour volume during 15 days. Tumours were of PCR were performed, and the amplified products were dissociated for 2 h with a mix of collagenase A (0.33 U/ml), analysed on a 2% agarose gel. Dispase (0.85 U/ml), DNase I (144 U/ml) in RPMI medium, and analysed by fluorescence-activated cell sorting as pre- viously described (Cheli et al., 2011). Immunohistological staining Tissue were fixed with 4% paraformaldehyde, permeabilised Cell migration and invasion assay with 0.1% Triton-X 100 and stained with MITF-specific The assay was carried out using the Cell Invasion and antibodies (C5, Abcam), or HIF1a (BD Bioscience). Anti- Migration Assay kit (Chemicon International, Temecula, CA, bodies were detected using the horseradish peroxidase-coupled antibody and revealed using Vectastain ABC kit form USA). Briefly, B16-F10 cells were transfected with 50 nM of small interfering RNA for 48 h. Cells (100 000) were added to Vectorlabs (Burlingame, CA, USA). each well in serum-free Dulbecco’s modified Eagle’s medium and allowed to migrate for 24–48 h at 37 1Cin5%CO2. The Statistical analysis lower compartment of the chamber was filled with culture Results are presented as means±s.d. or means±s.e.m. for the medium containing 7% fetal bovine serum. Cells at the lower in vivo assay, with experiment numbers indicated in the figure membrane surface were fixed in 1% paraformaldehyde in legends. Statistical significance was assessed using the Stu- phosphate-buffered saline (150 mM sodium phosphate, 154 mM dent’s t-test. Po0.05 was accepted as statistically significant. sodium chloride, pH 7.2), stained with 0.1% crystal violet and counted (five random fields/well). Invasion assay was per- formed similarly as the migration assay, except that the upper Conflict of interest chamber of the Transwell was coated with matrigel. The authors declare no conflict of interest. Western blot Cells were solubilised for 10 min at 4 1C in buffer containing 50 mM Tris, pH 7.4, 150 mM NaCl, 1% Triton X-100 and EDTA-free protease inhibitor cocktail from Roche Applied Acknowledgements Science (Indianapolis, IN, USA). Then, 30 mg of proteins were separated by electrophoresis on 10% polyacrylamide SDS– We thank Giorgetti-Peraldi Sophie, Regazzetti Claire and Karine PAGE and transferred onto a polyvinylidene fluoride mem- Bille for their help. Bhlhb2 reporter plasmid and expression brane (Immobilon, Millipore, Molsheim, France). The mem- vector were a kind gift from Dr S Egan (Toronto, Canada). This brane was blocked for 1 h at 25 1Cin10mM Tris–HCl, pH 7.4, work was supported by the ‘Fondation de France’, and Societe´ 150 mM NaCl, 1 mM EDTA, 0.1% Tween 20, 3% bovine serum Franc¸aise de Dermatologie. YC has been supported by ARC’s albumin (weight/volume) and 5% gelatin (weight/volume). fellowships. The authors greatly acknowledge the C3M Imaging The primary antibody was incubated for 18 h at 4 1C. After core facility (microscopy and imaging platform Coˆte d’Azur, three washes, each of 10 min, in 10 mM Tris, pH 7.4, 150 mM MICA) and the C3M animal room facility.

Oncogene Hypoxia controls MITF and invasiveness in melanoma cells Y Cheli et al 2470 References

Agarwal KC, Parks Jr RE. (1983). Forskolin: a potential antimeta- of SPARC on melanoma cell cycle progression. Pigment Cell static agent. Int J Cancer 32: 801–804. Melanoma Res 24: 219–232. Alonso SR, Tracey L, Ortiz P, Perez-Gomez B, Palacios J, Pollan M Ginouves A, Ilc K, Macias N, Pouyssegur J, Berra E. (2008). PHDs et al. (2007). A high-throughput study in melanoma identifies overactivation during chronic hypoxia ‘desensitizes’ HIFalpha epithelial-mesenchymal transition as a major determinant of and protects cells from necrosis. Proc Natl Acad Sci USA 105: metastasis. Cancer Res 67: 3450–3460. 4745–4750. Bedogni B, Welford SM, Cassarino DS, Nickoloff BJ, Giaccia AJ, Giuliano S, Cheli Y, Ohanna M, Bonet C, Beuret L, Bille K et al. Powell MB. (2005). The hypoxic microenvironment of the skin (2010). Microphthalmia-associated transcription factor controls the contributes to Akt-mediated melanocyte transformation. Cancer DNA damage response and a lineage-specific senescence program in Cell 8: 443–454. melanomas. Cancer Res 70: 3813–3822. Bedrosian I, Faries MB, Guerry Dt, Elenitsas R, Schuchter L, Mick R Gupta PB, Kuperwasser C, Brunet JP, Ramaswamy S, Kuo WL, Gray et al. (2000). Incidence of sentinel node metastasis in patients with JW et al. (2005). The melanocyte differentiation program predis- thin primary melanoma (o or ¼ 1 mm) with vertical growth phase. poses to metastasis after neoplastic transformation. Nat Genet 37: Ann Surg Oncol 7: 262–267. 1047–1054. Berra E, Benizri E, Ginouves A, Volmat V, Roux D, Pouyssegur J. (2003). Hoek KS, Eichhoff OM, Schlegel NC, Dobbeling U, Kobert N, HIF prolyl-hydroxylase 2 is the key oxygen sensor setting low steady- Schaerer L et al. (2008). In vivo switching of human melanoma cells state levels of HIF-1alpha in normoxia. EMBO J 22: 4082–4090. between proliferative and invasive states. Cancer Res 68: 650–656. Bertolotto C, Abbe P, Hemesath TJ, Bille K, Fisher DE, Ortonne JP et al. Hoek KS, Goding CR. (2010). Cancer stem cells versus phenotype- (1998). Microphthalmia gene product as a signal transducer in cAMP- switching in melanoma. Pigment Cell Melanoma Res 23: 746–759. induced differentiation of melanocytes. JCellBiol142: 827–835. Honma S, Kawamoto T, Takagi Y, Fujimoto K, Sato F, Noshiro M Bhawal UK, Sato F, Arakawa Y, Fujimoto K, Kawamoto T, et al. (2002). Dec1 and Dec2 are regulators of the mammalian Tanimoto K et al. (2011). Basic helix-loop-helix transcription factor molecular clock. Nature 419: 841–844. DEC1 negatively regulates cyclin D1. J Pathol 224: 420–429. King R, Googe PB, Weilbaecher KN, Mihm Jr MC, Fisher DE. Botton T, Puissant A, Bahadoran P, Annicotte JS, Fajas L, Ortonne (2001). Microphthalmia transcription factor expression in cuta- JP et al. (2009). In vitro and in vivo anti-melanoma effects of neous benign, malignant melanocytic, and nonmelanocytic tumors. ciglitazone. J Invest Dermatol 129: 1208–1218. Am J Surg Pathol 25: 51–57. Busca R, Ballotti R. (2000). Cyclic AMP a key messenger in the Ledda MF, Adris S, Bravo AI, Kairiyama C, Bover L, Chernajovsky regulation of skin pigmentation. Pigment Cell Res 13: 60–69. Y et al. (1997). Suppression of SPARC expression by antisense Carreira S, Goodall J, Aksan I, La Rocca SA, Galibert MD, Denat L RNA abrogates the tumorigenicity of human melanoma cells. Nat et al. (2005). Mitf cooperates with Rb1 and activates p21Cip1 Med 3: 171–176. expression to regulate cell cycle progression. Nature 433: 764–769. Loercher AE, Tank EM, Delston RB, Harbour JW. (2005). MITF Carreira S, Goodall J, Denat L, Rodriguez M, Nuciforo P, Hoek KS links differentiation with cell cycle arrest in melanocytes by et al. (2006). Mitf regulation of Dia1 controls melanoma prolifera- transcriptional activation of INK4A. J Cell Biol 168: 35–40. tion and invasiveness. Genes Dev 20: 3426–3439. Mani SA, Guo W, Liao MJ, Eaton EN, Ayyanan A, Zhou AY et al. Cheli Y, Guiliano S, Botton T, Rocchi S, Hofman V, Hofman P et al. (2008). The epithelial-mesenchymal transition generates cells with (2011). Mitf is the key molecular switch between mouse or human properties of stem cells. Cell 133: 704–715. melanoma initiating cells and their differentiated progeny. Oncogene Martin TA, Goyal A, Watkins G, Jiang WG. (2005). Expression of the 30: 2307–2318. transcription factors snail, slug, and twist and their clinical Cheli Y, Ohanna M, Ballotti R, Bertolotto C. (2010). Fifteen-year significance in human breast cancer. Ann Surg Oncol 12: 488–496. quest for microphthalmia-associated transcription factor target Nakashima A, Kawamoto T, Honda KK, Ueshima T, Noshiro M, genes. Pigment Cell Melanoma Res 23: 27–40. Iwata T et al. (2008). DEC1 modulates the circadian phase of clock Cho Y, Noshiro M, Choi M, Morita K, Kawamoto T, Fujimoto K gene expression. Mol Cell Biol 28: 4080–4092. et al. (2009). The basic helix-loop-helix proteins differentiated Olbryt M, Jarzab M, Jazowiecka-Rakus J, Simek K, Szala S, Sochanik embryo chondrocyte (DEC) 1 and DEC2 function as corepressors A. (2006). Gene expression profile of B 16(F10) murine melanoma of retinoid X receptors. Mol Pharmacol 76: 1360–1369. cells exposed to hypoxic conditions in vitro. Gene Exp 13: 191–203. Choi SM, Cho HJ, Cho H, Kim KH, Kim JB, Park H. (2008). Stra13/ Radisky DC. (2005). Epithelial-mesenchymal transition. J Cell Sci 118: DEC1 and DEC2 inhibit sterol regulatory element binding protein- 4325–4326. 1c in a hypoxia-inducible factor-dependent mechanism. Nucleic Robert G, Gaggioli C, Bailet O, Chavey C, Abbe P, Aberdam E et al. Acids Res 36: 6372–6385. (2006). SPARC represses E-cadherin and induces mesenchymal Corsetti RL, Allen HM, Wanebo HJ. (2000). Thin o or ¼ 1 mm level transition during melanoma development. Cancer Res 66: III and IV melanomas are higher risk lesions for regional failure and 7516–7523. warrant sentinel lymph node biopsy. Ann Surg Oncol 7: 456–460. Roesch A, Fukunaga-Kalabis M, Schmidt EC, Zabierowski SE, Du J, Widlund HR, Horstmann MA, Ramaswamy S, Ross K, Huber Brafford PA, Vultur A et al. (2010). A temporarily distinct WE et al. (2004). Critical role of CDK2 for melanoma growth subpopulation of slow-cycling melanoma cells is required for linked to its melanocyte-specific transcriptional regulation by continuous tumor growth. Cell 141: 583–594. MITF. Cancer Cell 6: 565–576. Sanchez-Martin M, Rodriguez-Garcia A, Perez-Losada J, Sagrera A, Ehata S, Hanyu A, Hayashi M, Aburatani H, Kato Y, Fujime M et al. Read AP, Sanchez-Garcia I. (2002). SLUG (SNAI2) deletions in (2007). Transforming growth factor-beta promotes survival of patients with Waardenburg disease. Hum Mol Genet 11: 3231–3236. mammary carcinoma cells through induction of antiapoptotic Shaw HM, McCarthy WH, McCarthy SW, Milton GW. (1987). Thin transcription factor DEC1. Cancer Res 67: 9694–9703. malignant melanomas and recurrence potential. Arch Surg 122: Eichhoff OM, Zipser MC, Xu M, Weeraratna AT, Mihic D, Dummer 1147–1150. R et al. (2010). The immunohistochemistry of invasive and Spira AI, Carducci MA. (2003). Differentiation therapy. Curr Opin proliferative phenotype switching in melanoma: a case report. Pharmacol 3: 338–343. Melanoma Res 20: 349–355. Wellbrock C, Marais R. (2005). Elevated expression of MITF Fenouille N, Robert G, Tichet M, Puissant A, Dufies M, Rocchi S counteracts B-RAF-stimulated melanocyte and melanoma cell et al. (2011). The p53/p21(Cip1/Waf1) pathway mediates the effects proliferation. J Cell Biol 170: 703–708.

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

Oncogene