A caveolin-dependent and PI3K/AKT-independent role of PTEN in β-catenin transcriptional activity

Loss of the tumour suppressor PTEN is frequent in human melanoma, results in MAPK activation, suppresses senescence and mediates metastatic behaviour. How PTEN loss mediates these effects is unknown. Here we show that loss of PTEN in epithelial and melanocytic cell lines induces the nuclear localization and transcriptional activation of β-catenin independent of the PI3K–AKT–GSK3β axis. The absence of PTEN leads to caveolin-1 (CAV1)-dependent β-catenin transcriptional modulation in vitro, cooperates with NRASQ61K to initiate melanomagenesis in vivo and induces efficient metastasis formation associated with E-cadherin internalization. The CAV1-β–catenin axis is mediated by a feedback loop in which β-catenin represses transcription of miR-199a-5p and miR-203, which suppress the levels of CAV1 mRNA in melanoma cells. These data reveal a mechanism by which loss of PTEN increases CAV1-mediated dissociation of β-catenin from membranous E-cadherin, which may promote senescence bypass and metastasis.

M elanomagenesis is a multistep process including initiation and progression. Mutant BRAF-and NRAS-driven mitogen-activated protein kinase (MAPK) signalling promotes proliferation of melanocytes, but this is effectively blunted by the induction of cellular growth arrest known as oncogene-induced senescence (OIS) [1][2][3] . The cell cycle inhibitor p16 INK4A is critical for this process and its expression is induced by the histone demethylase JMJD3 (ref. 4). OIS is bypassed in melanoma via loss of the p16 INK4A gene or suppression of its transcription by nuclear b-catenin 2,3,5,6 . Hemizygous phosphatase and tensin homologue (PTEN) loss is frequent in various cancers. Mutational inactivation and/or deletion of PTEN is found in about 20% of uncultured primary and metastatic melanomas 7-10 and in 30%-40% of melanoma cell lines 9 . In melanoma tissue, loss of PTEN protein expression has been observed in B15% of the cases 7,11 , but hemizygous gene loss has been observed to be occurring more frequently, that is, 34% (ref. 7). PTEN loss in nevi is rare, that is, 2 out of 39 (ref. 12), suggesting that PTEN aberrations in melanocytes are unlikely to contribute to their uncontrolled proliferation. In Dct::Cre mice, the inactivation of both PTEN alleles does not lead to a difference in the number of nevi 13 . Altogether, it is unlikely that altered PTEN expression directly stimulates abnormal proliferation of melanocytes, but the exact contribution of PTEN to melanoma development and progression remains poorly understood.
Epigenetic inactivation or loss of PTEN may occur at different stages of melanomagenesis, but remains controversial for its role in senescence. On one hand, the acute loss of PTEN and APC/FZR1 induces senescence in mouse primary fibroblasts 14 . However, the inactivation of PTEN failed to induce a robust growth arrest in human IMR90 fibroblasts 15 . Moreover, in human BRAF V600E -mutated melanocytes, reducing PTEN expression was sufficient to bypass senescence 16 . In mice, the induction of a BRAF mutation after birth induces nevi formation and melanomas arise harbouring deletion of p16 INK4A or PTEN 1,17 . These results suggest that the lack of PTEN or p16INK4A contributes to the bypass of senescence in vivo. PTEN has different functions depending on its subcellular localization 18 . At the membrane it can dephosphorylate phosphatidylinositol (3,4,5)-triphosphate, thereby regulating AKT phosphorylation and activity. Among other functions, cytoplasmic PTEN has been shown to interact with caveolin-1 (CAV1), a major endocytic protein in mammals 19 . Such PTEN-CAV1 interaction could implicate this phosphastase in cell signalling other than the canonical PI3K-AKT-GSK3b axis.
In this study, we uncovered a signalling mechanism by which PTEN affects nuclear localization and transcriptional activity of b-catenin through a reciprocal interplay with CAV1. We discovered that the lack of PTEN, through CAV1, induces b-catenin transactivation, leading to the repression of p16 INK4A . The co-occurrence of NRAS G183T mutation and PTEN loss was detected in a fraction of human melanoma biopsies, suggesting a non-epistatic mechanism. Indeed, in a mouse melanoma model, hemizygous PTEN loss synergized with NRAS mutation and led to bypass of senescence. Thus, we have identified a novel CAV1-dependent pathway by which PTEN affects b-catenin activity and mediates melanomagenesis.

Results
PTEN affects b-catenin nuclear localization. To explore the possibility that PTEN induces re-localization of b-catenin from the plasma membrane to the nucleus, we transiently re-expressed PTEN in human PTEN null human cells (Hs944T) (Fig. 1a-d).
In non-transfected cells, b-catenin was localized in the nucleus. On PTEN expression, the level of b-catenin in the nucleus was significantly diminished, B60% of green fluorescent protein (GFP)-transfected cells compared with 20% for PTEN ( Supplementary Fig. 1a). In addition, we performed subcellular fractionation experiments on GFP-and PTEN-transfected Hs944T cells. Consistent with immunofluorescence assays, the levels of nuclear b-catenin were lower in PTEN-Hs944T cells compared with GFP-Hs944T cells ( Supplementary Fig. 1b). Conversely, small interfering RNA (siRNA)-mediated PTEN knockdown in PTEN wt human Lyse melanoma cells, as shown by western blot analysis ( Supplementary Fig. 1c), resulted in increased translocation of b-catenin into the nucleus from 40% compared with 2% in control cells (Fig. 1e-h and Supplementary  Fig. 1d). These results mimic the observation from murine melanocytes lacking PTEN, which exhibit strong nuclear b-catenin localization (Fig. 1i,j and Supplementary Fig. 1e). One possible explanation for the relationship between PTEN loss and nuclear b-catenin localization is that the latter is a consequence of activation of the PI3K-AKT axis and inhibition of GSK3b. Thus, we evaluated the PI3K-AKT-GSK3b axis in relationship to the level of pThr41-Ser45 b-catenin to explain its nuclear localization (Fig. 1k). Re-expression of PTEN affected the activity of downstream effectors of phosphoinositide 3-kinase (PI3K), as indicated by the reduction of pAKT (Ser473) and pGSK3b (Ser9), but did not affect the level of total AKT and GSK3b. Even though the level of pThr41/Ser45 b-catenin was similar, on PTEN reexpression the total amount of b-catenin was slightly reduced and the quantity of transcriptionally active form of pb-catenin (Ser675) was decreased, explaining the lower b-catenin nuclear staining. This indicated that the observed strong changes in b-catenin localization could not be explained by minor molecular changes, if any, in the destruction complex that targets b-catenin for degradation. These results were confirmed on pharmacological inhibition of PI3K or GSK3b, using LY294002 and LiCl treatment, respectively, in cells that were transfected or not with PTEN. The decrease of pSer675 b-catenin by PTEN transfection was observed even in the presence of these compounds. LY294002 treatment efficacy was demonstrated by a decrease in pAKT (Ser473) and pGSK3b (Ser9) levels. Positive controls for LiCl treatment included lack of modification of the level of pAKT Ser473 in the presence of LiCl (certainly owing to the resultant of two effects, the dephosphorylation of AKT by PTEN and the induction of phosphorylation of AKT by LiCl 20 ) and induction of pGSK3b (Ser9) irrespective of the presence of PTEN. Moreover, we observed a consistent reduction of pb-catenin Ser675 levels after concomitant re-expression of PTEN and wild-type (WT) p110 or constitutive active p110 mutant (E545K) ( Fig. 1l and Supplementary Fig. 1f). Furthermore, similar results were obtained using a different constitutive active form of p110 (p110 CAAX) and a kinase-dead form of p110 (p110 KD) (Fig. 1m). Lastly, we quantified the amount of cells with positive b-catenin nuclear staining after transfection with GFP, WT PTEN, catalytically inert C124S, lipid (G129E) and protein phosphatase (Y138L) mutants, respectively ( Supplementary Fig. 1g). Altogether, these results suggest that in the absence of PTEN, pathways other than PI3K-AKT and GSK3b are involved in the nuclear localization of b-catenin and the accumulation of active pSer675 b-catenin.
area and in the cytoplasm and it was already suggested that CAV1 might be a positive regulator of b-catenin in human gastric cancer cells 21 . CAV1-scaffolding domain interacts with either PTEN or b-catenin 19,[22][23][24] . Thus, we hypothesized that PTEN and bcatenin could compete for CAV1, subsequently affecting different signalling outcomes. We first verified that CAV1 and b-catenin are able to immuno-complex in a reciprocal manner in Rosi human melanoma and HCT116 human carcinoma cell lines ( Fig. 2b and Supplementary Fig. 2a). Similarly, we confirmed that endogenous CAV1/PTEN can immune-complex in a reciprocal fashion in Rosi and HCT116 cells ( Fig. 2c and Supplementary  Fig. 2a). In addition, we validated this CAV1/PTEN interaction in WM852 and WM793 human melanoma cell lines after immuneprecipitation with CAV1 ( Supplementary Fig. 2b).   Furthermore, using glutathione S-transferase-b (GST-b)catenin fusion protein, we precipitated CAV1 on Hs944T cells transfected with GFP and, to a lesser extent, in cells expressing exogenous PTEN (Fig. 2d). Expanding on the GST pull-down assays, co-immunoprecipitation experiments in Hs944T cells expressing exogenous PTEN reveal that re-expression of PTEN significantly abrogates b-catenin/CAV1 interactions (Fig. 2e). In addition, after transfection of Hs944T cells using several PTEN-GFP constructs (WT, C124S, Y138L and G129E) we observed that the PTEN/CAV1 interaction is unaffected by the C124S mutant compared with WT, but highly disturbed by the Y138L and G129E mutants as revealed after immunoprecipitation experiments using CAV1 antibody ( Supplementary Fig. 2c,d). As previously stated ( Supplementary Fig. 1a), b-catenin is observed B60% of the time within the nucleus of Hs944T cells. On overexpression of GFP or CAV1 in Hs944T cells, immunofluorescence analysis revealed nuclear b-catenin staining, similar to controls (Fig. 2f-k and Supplementary Fig. 1a). Treatment of the Hs944T cells with LY294002 did not affect the level of b-catenin in the cytoplasm or in the nucleus when overexpressing CAV1, confirming that PI3K pathway has limited function in the nuclear translocation of b-catenin under these conditions ( Supplementary Fig. 2e-j). In murine epithelial CSG cells, expressing PTEN, CAV1 and b-catenin, the latter is found at cell-cell contacts, in the cytoplasm and in the nucleus, once the cells form small islets. In these conditions, the reduction of PTEN leads to a nuclear b-catenin (7.5% to 46%) and the reduction of CAV1 leads to a recruitment of b-catenin at the cell-cell contacts, and a reduction of nuclear b-catenin, from 7.5% to 2% (Fig. 2l-t and Supplementary Fig. 2k). Similar experiments were performed with murine pancreatic epithelial cells expressing PTEN (KPC1) or not (KCPTEN2). The reduction of PTEN and CAV1 in KPC1 cells was demonstrated by western blot analysis ( Supplementary  Fig. 2a). Concurrently, decreased amount of PTEN led to increased nuclear b-catenin of cells, from 20% to 50%, whereas the diminution of CAV1 resulted in an accumulation at the cellcell contacts of b-catenin and nuclear exclusion, from 20% to 6% of cells ( Supplementary Fig. 3a,b). As expected, b-catenin is mainly nuclear in KCPTEN2 cells ( Supplementary Fig. 3c,d).
Re-expression of PTEN in KCPTEN2 cells led to an accumulation of b-catenin at cell-cell contacts and nuclear exclusion from 18% of PTEN-expressing cells compared with 91% in controls ( Supplementary Fig. 3d). The overexpression of CAV1 in KCPTEN2 cells did not affect the localization of b-catenin. The absence of response of b-catenin is certainly due to the already high level of CAV1 in KCPTEN2 cells. Furthermore, CAV1 overexpression in PTEN-re-expressing cells failed to rescue b-catenin nuclear exclusion, consistently with immunoprecipitation experiments where PTEN significantly abrogates CAV1/b-catenin complex ( Fig. 2e and Supplementary Fig. 1a).
Thus, it appears that PTEN affects b-catenin-CAV1 complex and decreases the level of nuclear b-catenin.

CAV1 regulates the transcriptional activity of b-catenin.
We first assessed whether PTEN and CAV1 expression affects the transcriptional activity of b-catenin, using the TOP flash reporter assay in Hs944T cells. TOP flash activity is significantly reduced in the presence of PTEN and conversely induced with CAV1 (or b-catenin as positive control) (Fig. 3a). Co-expression of either PTEN and CAV1 or PTEN and BCAT resulted in a significant decrease in TOP flash activity, restoring basal-like levels on PTEN and CAV1 expression and, to a lesser extent, on PTEN and beta-catenin (BCAT) expression. Consistently, TOP flash activity decreased on CAV1 or b-catenin downregulation using appropriate siRNAs (Fig. 3b). In addition, we observed a similar effect on TOP flash reporter after modulating the levels of CAV1 and BCAT in a BRAF (BRAF V600E ) PTENnull cell line ( Supplementary Fig. 4a,b). The transcriptional activity of MITF-M, the master gene of the melanocyte lineage and a known b-catenin transcriptional target, was induced after overexpression of CAV1 and BCAT ( Supplementary Fig. 4c).
Owing to the fact that b-catenin can also act as a co-transcriptional repressor, we performed similar experiments for p16 INK4A . PTEN overexpression significantly induced p16 INK4A luciferase reporter activity (Fig. 3c). Overexpression of CAV1 or BCAT significantly reduced p16 INK4A luciferase reporter activity, whereas inverse knockdown significantly increased reporter activity (Fig. 3c,d). PTEN overexpression completely rescued the inhibitory effect of BCAT on p16 INK4A luciferase reporter (Fig. 3c). However, under these conditions, PTEN overexpression failed to rescue the inhibitory effect of CAV1 on p16 INK4A luciferase reporter. As expected, PTEN overexpression increased p16 messenger RNA transcript, whereas CAV1 and BCAT reduced the levels (Fig. 3e). We also observed an increase in p16 mRNA on co-transfection of PTEN and CAV1 or PTEN and BCAT to similar levels as PTEN alone. In congruency with prior results, knockdown of CAV1 or BCAT resulted in an increase of p16 mRNA levels (Fig. 3f). We then wondered whether the levels of CAV1 could also modulate MYC, another b-catenin target gene.
Indeed, MYC levels were directly affected by altering CAV1 ( Supplementary Fig. 4d,e). Finally, in agreement with our in vitro data, histomolecular analysis of human melanoma biopsies revealed the existence of PTEN-negative, CAV1-positive and P16-negative tumour (Fig. 3g). In conclusion, CAV1 acts on the transcriptional activity of exogenous (TOP) and endogenous (MITF, MYC and p16 INK4A ) b-catenin targets.
NRAS Q61K and PTEN loss cooperate during melanoma initiation.
The oncogenic form of NRAS (NRAS Q61K/R ) and the lack of PTEN are found in B20% and 30% of human melanoma, respectively, and it has generally been assumed that they are mutually exclusive 10 . We found that NRAS mutation and the loss of PTEN may coexist in human melanoma. A series of 105 human melanoma samples was analysed by comparative genomic hybridization for PTEN loss and for the presence or the absence of point mutations affecting NRAS. NRAS G183T mutation, resulting in an amino-acid change Q61K, was found in 16 samples (15%), of which 2 also showed homozygous PTEN loss ( Supplementary Fig. 5a). A second independent series of 101 human melanoma samples was analysed for NRAS G183T mutation and PTEN protein expression. Allele-specific PCR and DNA sequencing revealed that 14 samples harboured NRAS G183T mutation. Immunohistochemistry analysis showed that 39 samples were negative for PTEN; 3 of these also contained the NRAS G183T mutation (Fig. 4a). In addition, we tested the status of NRAS and PTEN in human melanoma cell lines. Lyse and Rosi cells express PTEN, whereas Hs944T and SK29 cells do not; Lyse and Hs944T cells carry NRAS G183T mutation, whereas Rosi and SK29 cells are WT for NRAS ( Supplementary Fig. 5b). Moreover, we determined the status of PTEN and p16 at the genomic, transcript and protein level for several human melanoma cell lines ( Supplementary Fig. 5c-e). In conclusion, the presence of NRAS mutation and PTEN loss is not mutually exclusive in melanoma. A mouse melanoma model for these two mutations was generated. Tyr::NRAS Q61K mice were crossed with Tyr::Cre mice and PTEN f/ þ mice. We produced the following mice: Tyr::Cre/°; PTEN f/ þ (henceforth DPTEN), Tyr::NRAS Q61K /°;Tyr::Cre/°(NRAS) and Tyr::NRAS Q61K /°;Tyr::Cre/°;PTEN f/ þ (NRAS-DPTEN). None of the DPTEN mice developed melanoma during 2 years of follow-up observation (Fig. 4b). Half of NRAS mice spontaneously developed melanomas with a latency period of 71 ± 16 weeks. NRAS-DPTEN mice spontaneously developed melanomas, with a higher penetrance (86%) and a shorter latency (27 ± 13 weeks) than the NRAS mice. Most melanomas appeared in the hairy part of the skin for both genotypes (Fig. 5a,b and Supplementary Table 1). The tumours consisted of irregularly pigmented cells with diverse sizes, large nucleoli and positive S100 immunostaining (Fig. 5c-h). Similar results were obtained with another mouse melanoma model in which the NRAS mutation is NRAS G12D and the activation of the mutation occurred at 10 weeks of age in melanocytes, using the CreERt2-LoxP-tamoxifen system ( Supplementary Fig. 6).
To understand the molecular mechanisms underlying the differences between NRAS and NRAS-DPTEN mouse melanomas and the role of PTEN in b-catenin nuclear localization, we first studied the MAPK and PI3K signalling pathways. After transformation, NRAS-DPTEN melanomas grew faster and larger than NRAS melanomas (Figs 4c and 5i,j). Western blot analysis revealed minimal differences in MAPK activity in NRAS versus NRAS-DPTEN tumours, whereas the status of the PI3K/PTEN/ AKT signalling pathway was significantly different for the two genotypes (Fig. 6). PTEN was almost absent from NRAS-DPTEN tumour samples, suggesting that the expression of PTEN from the remaining WT allele is inhibited for unknown reasons. The amount of pAKT (Ser473) and pGSK3b (Ser9) was dramatically increased in NRAS-DPTEN compared with NRAS; however, most interestingly, the levels of pS6 remained unaffected. These results suggest that signalling via mammalian target of rapamycin-associated proteins was the same in NRAS and NRAS-DPTEN melanomas.
We proceeded to evaluate the amount of b-catenin and p16 INK4A . In agreement with our results with melanoma cell lines, the level of pSer33/37-Thr41 b-catenin, targeted for  Melanoma appearance is associated with proliferation and bypass of senescence. Before transformation, melanoblasts lacking PTEN do not grow faster than WT (Supplementary Fig. 7). OIS is bypassed efficiently in the absence of PTEN. We established cultures of melanocytes from Tyr::Cre/°; PTEN f/f (Hom), Tyr::Cre/°; PTEN f/ þ (Het) and Tyr::Cre/°; PTEN þ / þ (WT) mice. No obvious difference was observed between Het and WT melanocytes. The initial rates of growth of the Het and Hom melanocytes in vitro were indistinguishable, confirming that the absence of PTEN does not induce proliferation before transformation (Fig. 4d). Cultures of Hom melanocytes divided continuously and rapidly became immortalized. In contrast, Het melanocytes in culture stopped expanding within 4 weeks of explantation and developed a large nucleus and a flattened morphology, and accumulated melanin, hallmark features of senescence. Melanocyte cell lines could be established from 90% (9 of 10) of Hom newborn pup skins, but only from 28% (2 of 7) of their Het littermates, implying that the absence of PTEN from melanocytes increased the efficiency of immortalization. We confirmed that by modulating PTEN levels we affected p16 INK4A expression in primary normal human epithelial melanocytes as well as in transformed Lyse human melanoma cells (Figs 1i,j and 4e).
The absence of PTEN promotes efficient metastasis formation. Autopsy of 7 NRAS and 17 NRAS-DPTEN mice carrying melanoma revealed the presence of lung metastasis in 1/7 and 8/17 mice, respectively ( Supplementary Fig. 8). Molecular analysis of NRAS and NRAS-DPTEN tumours was performed to evaluate the level of b-catenin, PTEN, CAV1 and the cell-cell adhesion molecule and b-catenin interactor, E-cadherin (ECAD). ECAD can be internalized using caveolae 25 , and its levels and localization are affected by interaction with b-catenin 26 . Surprisingly, the amount of ECAD mRNA and protein was higher in the absence of PTEN (Fig. 7a,b). However, in the absence of PTEN, the amount of ECAD located at the cell-cell contact was much lower than the amount of cytoplasmic ECAD, which was dramatically increased (Fig. 7c). Similarly, mRNA and protein levels of CAV1 and b-catenin (total and pSer675 protein) were also increased in NRAS-DPTEN tumours compared with NRAS (Fig. 7a,b). In addition, CAV1 and b-catenin were mostly delocalized from the membrane in NRAS-DPTEN tumours (Fig. 7c). Furthermore, we confirmed that PTEN expression decreases CAV1/BCAT immuno-complex in murine tumour samples and affects the transcription of known b-catenin targets, MYC and CCDN1 (Fig. 7d,e). Lastly, we examined the expression of CAV1 and PTEN in human melanomas. Fifty human melanoma samples were stained for PTEN and CAV1 on consecutive slides (Fig. 7d). PTEN and CAV1 were expressed in 40 and 10 melanomas, respectively. When present, CAV1 was mainly located at the membrane, but could also be found in the cytoplasm and seldom in the nucleus. Interestingly, in 34/50 cases, there is a strong tendency for low expression of CAV1 and high levels of PTEN. Altogether, these results show that melanoma samples lacking PTEN and expressing high level of CAV1 do exist in mouse and human melanomas.
b-Catenin induces CAV1 through repression of miRs. CAV1 is regulated by miR-203 in human breast cancer cells and miR-  199a-5p in lung fibroblasts [27][28][29] . A miRnome was performed on NRAS and NRAS-DPTEN tumours, revealed a decrease in miR-203 and miR-199a-5p in the absence of PTEN (Fig. 8a and Supplementary Data 4). We first validated that these two miRs were able to affect the amount of CAV1 mRNA in the absence of PTEN. Hs944t human melanoma cells were transfected with miR-203 and miR-199a-5p mimics, which led to the reduction of CAV1 mRNA and protein (Fig. 8b,c). Next, we wondered whether the reduction of miR expression in the absence of PTEN was related to the increased activation of b-catenin signalling.
In this respect, we quantified the levels of miR-203 and miR-199a-5p after modulating b-catenin. Indeed, b-catenin represses miR-203 and miR-199a-5p transcription. (Fig. 8d). Such regulation could be direct, as chromatin immunoprecipitation (IP)experiments revealed that b-catenin binds the promoter region of miR-203. Finally, we showed that b-catenin controls the level of CAV1 mRNA and protein (Fig. 8e).

Discussion
In this study, we demonstrate the existence of a complex signalling network involving reciprocal interactions among PTEN, CAV1 and b-catenin; regulating molecular and cellular mechanisms that play a critical role in tumour initiation and progression. PTEN is classically known to inhibit the PI3K/AKT signalling axis, but here we show that it also remarkably controls the nuclear levels and transcriptional activity of b-catenin in an alternative PI3K/AKT way. b-Catenin transcriptional activity represses p16 INK4A transcription, leading to bypass of senescence, and the putative tumour suppressors mir-203 and mir-199a-5p, resulting in regulation of CAV1. CAV1 interacts with either PTEN or b-catenin, modulating the localization and cotranscriptional activity of b-catenin. Importantly, PTEN loss, via CAV1 interaction, also leads to the internalization of ECAD, promoting metastasis. These events occur in epithelial (salivary and pancreatic) and non-epithelial (melanocyte) cells, and appear to be independent of the RAS-BRAF context. Our work identifies a novel mechanism by which a subset of melanomas can escape OIS and result in aggressive tumours. This mechanism by which loss of PTEN induces bypass of senescence, allowing an earlier melanoma initiation with a higher penetrance after oncogenic NRAS Q61K -induced senescence, was modelled in a mouse model, relevant for human melanomagenesis. In fact, based on a small melanoma series, it was generally assumed that NRAS mutations and PTEN loss are mutually exclusive events in human melanomagenesis; however, we showed that these two events co-exist in a fraction of human melanomas (5 out of 206 melanoma from 2 independent cohorts), as it was recently showed in one case 30 . Moreover, after analysing The Cancer Genome Atlas (TCGA), we found five melanoma samples with PTEN homozygous deletion and carrying NRAS Q61K mutations. Our study was limited to the PTEN loss-mediated bypass of OIS on the NRAS Q61K background, but the PTEN/CAV1/b-catenin/ p16 INK4A pathway may hold true in BRAF V600E melanomas as well ( Supplementary Fig. 4a-c). Moreover, in primary human fibroblasts and melanocytes, PTEN loss inhibits BRAF V600E (or HRAS V12G )-induced senescence 15,16 . Consequently, the loss of PTEN results in OIS bypass associated with RAS or RAF. Bypassing senescence is classically associated with p53/MDM and Rb/p16 INK4A proteins. In our models, oncogenic NRAS with or without PTEN loss did not affect neither P53, MDM2 nor MDM4 expression levels (data not shown). This supports a p53-independent model of senescence in melanoma cells in the NRAS Q61K background, in which b-catenin-regulated expression of senescence-inducing p16 INK4A is directly affected by the absence of PTEN 6,31,32 . At this point, it has to be noted that the role of b-catenin during melanomagenesis remains controversial [33][34][35][36][37][38] .
The role of CAV1 in tumorigenesis is subject of debate 39,40 ; expression is tissue specific and varies substantially depending on the stage of the disease 41 . In melanomagenesis, its function was only investigated at the level of progression, with controversial results depending likely on the molecular context 39,40 . We demonstrated that CAV1 immuno-complexes with b-catenin and PTEN in the melanocyte lineage. Moreover, CAV1 has been associated with accumulation of b-catenin in gastric cancer and HEK293T cells 21,42 . We showed that in human melanoma NRAS Q61K PTEN-null cells, p16 INK4A is repressed through CAV1/b-catenin; this interaction is ablated on PTEN reexpression. Thus, CAV1 would serve as a promoter of tumour initiation and progression by enhancing b-catenin-related transcription.
In NRAS-DPTEN murine melanoma tumours, western blot analysis revealed that the levels of CAV1, b-catenin and ECAD were higher than in NRAS tumours. Moreover, it appears that ECAD is more abundant in the cytoplasm of NRAS-DPTEN melanoma cells than in NRAS cells, and less at the cell-cell contact (Fig. 7c). ECAD is mainly found at the cell-cell contact and can be internalized using caveolae 25 . The reduction of ECAD at the cell-cell contact is likely a feature of melanoma progression and may induce a pseudo-epithelial to mesenchymal transition.
Whereas melanoma cell lines clearly demonstrated the causal relationship between PTEN, CAV1, b-catenin and p16 INK4A expression to robustly bypass senescence, immunohistochemical studies of melanoma tissue revealed that this mechanism plays a role in only a fraction of cases (Fig. 7f). In fact, we observed the expected correlation trend (evaluated as P ¼ 0.065, using the Fisher's exact test) of CAV1 high PTEN low or the inverse CAV1 low PTEN high in 12 of 50 samples. Other combinations were found in the remaining samples, indicating a high level of molecular and clinico-pathological complexity that indicates that other mechanisms of OIS escape exist in human melanomagenesis.
Thus, although the validity of our model of PTEN/CAV1/ b-catenin-regulated p16 INK4A repression is supported by our findings in cell culture, mouse models and human samples, the role of PTEN in senescence bypass is intricate and most likely context dependent. Be as it may, our studies indicate that CAV1 and CAV1-related pathways may be a potential therapeutic target for melanoma treatment. On the other hand, our findings predict that PI3K/AKT inhibitors will not block effectively the mechanism of senescence bypass caused by PTEN loss.

Methods
Cell culture and cell lines. Mouse primary melanocyte cell lines were established from mice 1-5 days after birth. Mice were rinsed with 70% ethanol and then in ice-cold PBS. The skin was removed and stored in PBS. Next, the skin was cut into small pieces and incubated with collagenases type 1 and 4 for 40 min at 37°C and 5% CO 2 . Following this incubation, the dermis and epidermis were separated using forceps and the epidermis was washed in wash buffer (1x Hank's balanced salt solution, 1 mM CaCl 2 , 0.005% DNase, 20% FCS). After washing, it was centrifuged for 5 min at 1,100 r.p.m. at room temperature. The resulting cell pellet was resuspended in dissociation buffer (GIBCO) and incubated at 37°C and 5% CO 2 for 10 min in a petri dish. Next, the cells were put through 18-and 20-g needles and washed in a 15-ml tube with wash buffer for 10 min, allowing for the removal of grease and hairs. The supernatant was centrifuged for 5 min at 1,100 r.p.m. at room temperature (RT) and the resulting cell pellet was resuspended in PBS and counted. The cells were then centrifuged again for 5 min at 1,100 r.p.m. (room temperature) and plated in tissue-culture dishes in F12 media supplemented with 10% FCS and 200 nM 12-O-tetradecanoylphorbol-13-acetate.
Western blotting. Whole-cell lysate was prepared from human melanoma cell lines using RIPA buffer and whole-tissue lysate was prepared from mouse melanoma tumour using SDS lysis buffer 53 Supplementary Fig. 9.
Immunofluorescence microscopy. Primary murine melanocytes were grown to near confluence upon which point were counted, 2.5 10 5 cells were seeded in 18mm glass cover slips and allowed 24 h to recover prior immunofluorescence analysis. Similar procedure was followed for CSG, KPC1 and KCPTEN2 cells. Human Hs944T melanoma cells were transfected with CMV::PTEN-GFP (#1031) and allowed 48 h to recover before being fixed in 4% paraformaldehyde (PFA) for 20 min at RT. Human or mouse cells were permeabilized with 0.2% v/v PBS/Triton X-100 for 5 min at RT. Then, cells were washed twice with PBS and blocked with 1% BSA (w/v) and 10% fetal bovine serum in PBS for 20 min at RT. Cells were incubated with the primary antibody anti-b-catenin (dilution 1/100) at 4°C overnight. Alexa 555 anti-rabbit (Sigma) secondary antibody was incubated for 1 h at RT in the dark. Cells were counterstained with 0.5 mg ml À 1 4,6-diamidino-2phenylindole to visualize the nucleus.
The next morning, the immune complexes were collected by centrifugation at 14,000 r.p.m. for 10 min at 4°C. The IP supernatants were removed and the beads were washed four times with IP buffer, with centrifugation at 10,000 r.p.m. for 10 s between each wash. The beads were then washed four times with EDTA buffer (50 mM Tris (pH 8), 150 mM NaCl, 1 mM EDTA, 1% (v/v) Triton X-100). Following the last wash, the immunoprecipitated proteins were solubilized in SDS sample buffer and boiled for 10 min. Samples were then resolved by SDS-PAGE. Trueblot anti-rabbit IgG HRP (18-8816-33, Rockland Immunochemicals) was used as a secondary antibody for western blotting in Fig. 2b,c and Supplementary Fig. 2a. Nuclear b-catenin quantification. Quantification of all images was performed using the ImageJ imaging software. Briefly, using the Threshold function, all images were equally calibrated to either control GFP or si Scramble. Once a black and white image was obtained after processing through the Threshold function, a merge with 4,6-diamidino-2-phenylindole and other fluorescent channels (that is, GFP or red fluorescent protein) was generated, after which point we manually counted only the cells that were positive for GFP or red fluorescent protein (depending on the construct) and displayed a nuclear signal. Subcellular fractionation. Hs944T cells transfected with complementary DNAs encoding either GFP or PTEN were separated into nuclear and cytoplasmic fractions. Cells were lysed with cytoplasmic extraction buffer (10 mM HEPES pH 7.9, 10 mM KCl, 0.1 mM EDTA, 1.5 mM MgCl 2 , 1 mM dithiothreitol, 0.2% Nonidet P-40, 1 mM NaF, 1 mM Na 3 VO 4 and protease inhibitor cocktail), while rotating on a rocker rotator at 4°C for 15 min. The cells were then centrifuged at 14,000 r.p.m. at 4°C for 5 min and the resulting supernatant (cytoplasmic fraction) was collected. The pellet was resuspended in nuclear extraction buffer (20 mM HEPES pH7.9, 420 mM NaCl, 0.1 mM EDTA pH8, 1.5 mM MgCl 2 , 1 mM dithiothreitol, 0.2%Triton X-100, 1 mM NaF, 1 mM Na 3 VO 4 and protease inhibitor cocktail) and incubated at room temperature for 10 min, after which it was centrifuged at 14,000 r.p.m. at 4°C for 5 min. The resultant supernatant (nuclear fraction) was removed from the pellet (cytoskeleton) and the purity of each fraction was assessed by immunoblotting with antibodies to a-tubulin and lamin B1, respectively.
Detection of NRAS mutations from human tissue. DNA was extracted from 20-mm-thick paraffin-embedded sections using NucleoSpin Extract II (Macherey-Nagel, 740590250) according to the manufacturer's instructions and was amplified by PCR. Allele-specific PCR of the NRAS gene was performed using a 5 0 WT PCR primer (LL1827, 5 0 -CAT ACT GGA TAC AGC TGG AC-3 0 ) and a mutated PCR primer corresponding to the NRAS Q61K mutation (LL1828, 5 0 -CAT ACT GGA TAC AGC TGG GA-3 0 ). The reverse primer (LL1800, 5 0 -TGA CTT GCT ATT ATT GAT GG-3 0 ) was used for all the PCR reactions 60 . The PCR mixture contained Expand High fidelity buffer, 200 mM of each dNTP (dNTP mix, Finnzyme, F560XL), 50 pM of each primer, 2.6 U of Expand High fidelity (Roche, 11732650001) and 100 ng of DNA. PCR was performed for 38 cycles of 30 s at 94°C, 90 s at 56°C and 30 s at 72°C. Samples were incubated for 10 min at 94°C before the cycles. The NRAS gene in DNA extracted from tumours was also sequenced following PCR amplification. The primer sequences were 5 0 -GTT ATA GAT GGT GAA ACC TG-3 0 (LL1901; forward) and 5 0 -GAG GTT AAT ATC CGC AAA TGA CTT-3 0 (LL1918; reverse). The NRAS gene exon 3 sequences were analysed by direct DNA sequencing according to the Sanger technology.
Transgenic mice and tumour collection. The transgenic Tyr::N-RAS Q61K /°mouse line was described previously 3 . Floxed PTEN mice were provided by H. Wu (UCLA, Los Angeles, CA, USA) and were obtained from F. Beermann (EPFL, Lausanne, Switzerland). The characterization of the PTEN flox mice 61,62 and Tyr::Cre mice 55 has been reported previously. All mice were backcrossed onto a C57BL/6 background for more than ten generations. Mice were maintained in the specific pathogen-free mouse colony at the Institut Curie, in line with French and European Union law. Ethical authorization number is P2.LL.029.07. Floxed PTEN heterozygous mice were crossed with Tyr::Cre and Tyr::NRAS Q61K /°to generate Tyr::NRAS Q61K /°;Tyr::Cre; PTEN f/ þ (NRAS-DPTEN mice), Tyr::NRAS Q61K /°; PTEN f/ þ (NRAS mice) and Tyr::NRAS Q61K /°;PTEN f/f (NRAS mice). Mice were genotyped by PCR using DNA extracted from tails. The mice were evaluated weekly for tumour appearance and progression. Once tumours were 1 cm across, the mice were killed and autopsied. Some mice were also killed because of poor health. Tumour samples were fixed in 4% PFA and paraffin-embedded for histological analysis and immunostaining. When sufficient tumour tissue was available, samples were frozen for subsequent western blot analysis.