Oncogenic ALK F1174L drives tumorigenesis in cutaneous squamous cell carcinoma.

Cutaneous squamous cell carcinoma (cSCC) is the second most common skin cancer characterized by increased mortality. Here, we show for the first time that anaplastic lymphoma kinase (ALK), a receptor tyrosine kinase of the insulin receptor superfamily, plays a pivotal role in the pathogenesis of cSCC. Our data demonstrate that the overexpression of the constitutively active, mutated ALK, ALK F1174L , is sufficient to initiate the development of cSCC and is 100% penetrant. Moreover, we show that cSCC development upon ALK F1174L overexpression is independent of the cell-of-origin. Molecularly, our data demonstrate that ALK F1174L cooperates with oncogenic Kras G12D and loss of p53, well-established events in the biology of cSCC. This cooperation results in a more aggressive cSCC type associated with a higher grade histological morphology. Finally, we demonstrate that Stat3 is a key downstream effector of ALK F1174L and likely plays a role in ALK F1174L -driven cSCC tumorigenesis. In sum, these findings reveal that ALK can exert its tumorigenic potential via cooperation with multiple pathways crucial in the pathogenesis of cSCC. Finally, we show that human cSCCs contain mutations in the ALK gene. Taken together, our data identify ALK as a new key player in the pathogenesis of cSCC, and this knowledge suggests that oncogenic ALK signaling can be a target for future clinical trials.


Introduction
Anaplastic lymphoma receptor kinase (ALK) alterations have been identified in several human cancers, including neuroblastoma, glioblastoma, lung cancer, anaplastic large cell lymphoma, and renal cell carcinoma (Hallberg & Palmer, 2013). Most cancerassociated rearrangements in the ALK gene are associated with fusions, copy-gain number, or activating ALK mutations (Hallberg & Palmer, 2013). In mice, overexpression of the mutated ALK F1174L gene results in neuroblastoma development (Heukamp et al, 2012). An elevated expression of phosphorylated ALK as well as its ligands, midkine, and pleiotrophin has been found in patients with basal cell carcinoma (BCC) and cSCC (Ning et al, 2013). To investigate the possible role of ALK in the pathogenesis of skin tumors, we overexpressed ALK F1174L in the epithelial cells in the skin. A number of studies has addressed the cell-of-origin of BCC and cSCC. BCC can arise from the progenitor cells of the interfollicular epidermis, cells in the infundibulum of the hair follicle (HF) (Youssef et al, 2010), and HF stem cells (Grachtchouk et al, 2011). Similarly, compelling evidence suggests that cSCC can also arise not only from interfollicular epidermis but also from the HF stem cells (Lapouge et al, 2011;Sanchez-Danes & Blanpain, 2018). Based on these studies, we have decided to overexpress ALK F1174L in HF stem cells using Lgr5-CreERT2 (Barker et al, 2007) and K15-CreERT2 (Morris et al, 2004) mouse lines, and in all basal cells taking advantage of K5-CrePR1 (Zhou et al, 2002) and K14-CreERT2 (Vasioukhin et al, 1999) transgenic strains. demonstrated to promote skin carcinogenesis (Hoste et al, 2015) suggests that whereas ALK F1174L overexpression alone is sufficient to drive tumor formation in ear and tail skin, it might require an additional promoting treatment in the back skin. We nevertheless excluded those mice from further analysis because our study focused on the dissection of the role of ALK F1174L overexpression in the context of skin homeostasis. ALK overexpression was confirmed using Western blot with anti-pALK antibodies ( Fig S1C). The presence of the luciferase (luc) reporter gene in the ALK transgene allowed us to monitor tumor development using IVIS imaging system (Heukamp et al, 2012) (Fig 1D). Based on the histological examination (Gleich et al, 2016), we distinguished four types of skin lesions, including cysts (n = 5 mice), acanthopapilloma (AP) (n = 7 mice), keratoacanthoma (KA) (n = 7 mice), and cSCC type 1 (n = 8 mice) (Figs 1E and F and S1D and E). Similarly, the targeted expression of ALK F1174L using another HF stem cell-specific Cre line, K15-CrePR1 (Morris et al, 2004), resulted in cSCC development ( Fig 1G). Moreover, the crossings of ALK F1174L mice with K5-CrePR1 (Zhou et al, 2002) and K14-CreERT2 (Vasioukhin et al, 1999) lines gave rise to skin lesions strikingly resembling those present in Lgr5-CreERT2 and K15-CrePR1 lines as assessed with hematoxylin/eosin staining as well as with an immunostaining for pan-cytokeratin (pan-Ck) (Figs 1H and I and S1F). We conclude that ALK F1174L expression alone is sufficient to drive tumor initiation, and it induces cSCC independently of the cell-of-origin. (A) Graphical representation of Lgr5-CreER T2 ;LSL-ALK F1174L genotype and experimental design. Topical application of 4OH-tamoxifen (4-OHT) in ears, back, and tail skin of mice resulted in skin lesion development. Then, mice were euthanized when termination criteria were observed (tumor size and ulceration). (B) Tumor-free survival of Lgr5-CreER T2 ;LSL-ALK F1174L mice (n = 15, median 47 d) and controls (n = 12). Log-rank (Mantel-Cox) Test P < 0.0001, HR 28.12. All mice developed tumors. (C) From left to right. Topical administration of 4-OHT and after tumor formation per location. (D) Representative picture of in vivo imaging system (IVIS). Analysis of the LSL-ALK F1174L transgene expression shows strong luminescent signal from the tumors on ears and tail. (E) Pie chart representing the number of mice that developed the listed skin lesions out of total skin lesions diagnosed. (F) Representative hematoxylin and eosin (H&E) staining of such lesions from the ears. Scale bars = 100 µm. To note, each mouse developed several tumors. (G, H, I) Top to bottom. Genotypes, representative pictures of the tumors marked by pan-cytokeratin immune-labeling and tumor distribution per location. To note, because of leakage of the Cre expression, tumor formation was observed when no specific topical administration was performed. Scale bars = 250 µm.
Next, we examined the expression of ALK protein in the normal human skin. ALK expression was readily detected in virtually all basal keratinocytes (Fig S2A). Given the fact that ALK expression in humans has been described in both tumor types, BCC and cSCC, it was surprising that we did not observe any signs of BCC development in our mouse model. To investigate whether mutations in ALK gene are present in the human cSCC samples, we performed the analysis of publicly accessible data resources. Cohen et al (2015) described the landscape of mutations detected in cSCC diagnosed in patients undergoing BRAF inhibitor therapy for advanced metastatic melanoma and among other mutations, they have detected the ALK F1174L mutation (1/29 patients). This prompted us to investigate the frequency of ALK mutations in primary and metastatic human cSCC, and we focused on analysis of human cSCC using exome sequencing (Fig S2B) (Durinck et al, 2011;Lee et al, 2014;Pickering et al, 2014;Li et al, 2015;Yilmaz et al, 2017;Inman et al, 2018). Of 161 human cSCC cases analyzed, we have identified 32 cases (20%) carrying mutations in the ALK gene. PolyPhen-2 software was used to predict damaging effects of identified ALK mutations ( Fig S2C) (Adzhubei et al, 2010). Whereas nearly 50% of cases predicted no effect on the ALK gene function, others were characterized as potentially harmful mutations. Of note, in addition to the p.F1174L mutation identified in cSCC occurring in the melanoma patient (Cohen et al, 2015), two additional cases harbored the p.G1201E (Murugan & Xing, 2011) and p.G1286R (Bresler et al, 2014) mutations, which are known ALK gain-of-function mutations ( Fig  S2C). We also checked for the presence of mutations in the Alk gene in the previously described mouse models of cSCC (Nassar et al, 2015). Two mutations were described, namely, the p.C787S and the p.C1012S, which, however, did not correlate with Alk expression (Nassar et al, 2015).
To directly assess whether the Alk F1178L (orthologue to ALK F1174L ) mutation is present in mouse cSCC, we have sequenced exon 23 of the Alk gene from six independent cSCC induced by a combination of DMBA (7,12-dimethylbenzanthracene) and TPA (12-Otetradecanoylphorbol-13-acetate) treatment ( Fig S2D). None of the tumors exhibited mutations in the Alk gene. This could be due to the fact that all mouse DMBA/TPA tumors are induced on the back skin, a location devoid of skin lesions in ALK F1174L overexpressing mice. Because DMBA/TPA-induced tumors are predominantly driven by oncogenic mutations in RAS signaling, an alternative explanation may be that ALK and RAS are mutually exclusive drivers of skin tumorigenesis. To test this hypothesis, we have analyzed the co-occurrence and overall frequency of ALK and RAS mutations in human cSCC using cBio Portal (Fig S2E). Among all RAS genes analyzed, we have observed that mutations in ALK gene appear to be mutually exclusive with mutations in KRAS gene (68 cases analyzed and 0 cases display co-occurrence). Mutations in HRAS and NRAS genes co-occurred with mutations in ALK in some cases (68 cases analyzed, 1 case co-occurrence of NRAS and ALK in the same patient, and 3 cases show cooccurrence of ALK and HRAS mutations).
Furthermore, we ruled out the possibility that the back skin of ALK F1174L Lgr5-CreERT2 mice contained early lesions, yet not visible to the naked eye. To this aim, we performed thorough histological and immunohistochemical examination with anti-Ki67 antibodies and quantified the proliferation index of keratinocytes identified by pan-Ck staining ( Fig S3B). Even though some HF appeared dysplastic, no increased proliferation was observed.
To dissect the molecular mechanisms underlying the ALK F1174Ldriven cSCC in the skin, we isolated tumor epithelial cells (TECs) from ALK F1174L lesions and normal keratinocytes from their wild-type littermates by FACS using a combination of Epcam + CD31 − CD45 − CD140a − markers as previously described (Nassar et al, 2015) and independent triplicates were subjected to RNA sequencing (Fig 2A). Analysis of differentially expressed keratins revealed a pronounced decrease in the expression level of Krt77 (Krt 1b) and Krt15, indicative of an impaired epithelial differentiation ( Fig 2C). On the other hand, several other keratins, including Krt6, Krt7, Krt17, and Krt18 were significantly increased. Krt6, Krt16, and Krt17 are associated with aberrant proliferation, and Krt18 is a marker of embryonal keratinocytes and is often associated with poor differentiation and epithelial-to-mesenchymal transition (EMT) in cSCC ( Fig 2C) (Watanabe et al, 1995). Gene ontology (GO) term analysis provided further insights into biological processes enriched in ALK F1174L TECs. Among significantly upregulated processes were cell adhesion/migration and proliferation, along with alterations in metabolic processes (Fig 2B and E). The latter overall indicated a shift towards an anabolic and glycolytic (Warburg-like) metabolic phenotype, an additional hallmark that ALK F1174L cells would gain in support of transformation. In addition, Kyoto Encyclopeida of Genes and Genomes (KEGG) analysis revealed ALK F1174L -associated signature including alterations in several signaling pathways including FAK and ECM, PI3K-AKT, and JAK-STAT3 (Fig 2B and D). Moreover, ALK F1174L expression resulted in an increase of Mek1 (Map2k1) expression (Fig 2D and F). Interestingly to note is that this signaling signature has been described as characteristic of cSCC development (Ratushny et al, 2012). The expression profile of ALK F1174L -driven tumorigenesis reveals the lack of Sonic Hedgehog (Shh) signaling pathway, which is characteristic of the typical BCC, demonstrating that ALK F1174 expression leads to cSCC formation without any cellular and molecular signs of concomitant BCC development.
Similarly to the DMBA/TPA model of mouse carcinogenesis, human SCCs are often associated with mutations in the RAS genes (van der Schroeff et al, 1990;Spencer et al, 1995). Overexpression of either Hras or Kras in mouse skin initiates tumorigenesis (Brown et al, 1998;Vitale-Cross et al, 2004;Caulin et al, 2007). The analysis of RNA-seq data obtained from the comparison of WT ear keratinocytes and ALK F1174L TECs did not reveal any significant deregulation in the Ras signature. Moreover, our data on the co-occurrence of mutations in RAS and ALK genes in human patients did not reveal any strong evidence suggesting that ALK and RAS are mutually exclusive drivers of cSCC ( Fig S2E). Thus, we sought to functionally determine whether ALK F1174L can cooperate with Kras in driving SCC development. To answer this question, we have crossed ALK F1174L Lgr5-CreERT2 mice with Kras G12D strain ( Fig 3A) (Jackson et al, 2001). Within less than 40 d, all mice developed ulcerative lesions on the ears and tails (n = 13) (Fig 3B and E). As shown in Fig 3B, the tumor penetrance/incidence remained as high as in the ALK F1174L transgene and reached 92% in ear and tail areas. However, similarly to ALK F1174L overexpression, no tumors were seen in the back skin of ALK F1174L Kras G12D Lgr5-CreERT2 mice. In analogy to ALK F1174L Lgr5-CreERT2 mice, the histological analysis revealed the presence of four distinct histological entities including, cysts, AP, KA, and SCC1 (Figs 3C and S3C-E). The addition of Kras G12D transgene resulted in an increased number of tumors per mouse as compared with ALK F1174L Lgr5-CreERT2 mice ( Fig 3D). Moreover, ALK F1174L Kras G12D Lgr5-CreERT2 mice showed signs of aggressive SCC (size, ulceration, and bleeding) at earlier time points as compared with ALK F1174L Lgr5-CreERT2 mice (median 19 d post induction versus 37, d respectively, P < 0.0001) ( Fig 3F). Taken together, these data strongly suggest that Kras G12D oncogene cooperates with ALK F1174L in driving the pathogenesis of cSCC.
To determine the molecular mechanisms underlying this cooperative effect, we have isolated TECs from tumors of ALK F1174L Lgr5-CreERT2 and ALK F1174L Kras G12D Lgr5-CreERT2 ( Fig 3G) using previously described FACS protocol (Nassar et al, 2015) and performed RNA sequencing. GO and KEGG analysis highlighted several major alterations upon the addition of Kras G12D oncogene (Fig 3H-J). Among these, cell adhesion/migration and ECM remodeling were suggestive of an EMT process initiated upon Kras G12D . To further test it, we performed immunostaining for vimentin, a mesenchymal marker, and quantified the number of cells double positive for pan-CK and vimentin ( Fig 3K). As shown in Fig 3L, double transgenic ALK F1174L Kras G12D Lgr5-CreERT2 mice showed higher number of pan-Ck + /Vim + cells as than single ALK F1174L transgene (n = 5; 5, P = 0.0117, mean ± SD: 4.907 ± 2.904; 8.547 ± 3.885). Furthermore, based on our RNA-seq data, double transgenic TECs displayed an increased proliferation. To verify this, we performed immunostaining for Ki67, a marker of cell proliferation and quantified number of Pan-Ck + / Ki67 + cells within tumors (Fig 3M and N). Indeed, in alignment with RNA-seq data, we show that Kras G12D oncogenic induction resulted in a pronounced increase in the proliferative index of cancer cells. In addition, among the most significantly regulated keratins, we identified Krt4, Krt8, and Krt19 strongly up-regulated in ALK F1174L Kras G12D TECs ( Fig S3F). Krt4, Krt8, and Krt19 were previously described as keratins associated with stem cell potential, dedifferentiation, proliferation, and invasion (Watanabe et al, 1995). To examine the degree of vascularization within individual tumors, we used a well-defined marker of angiogenesis, CD31, and quantified its relative expression (Fig 3O and P). Our results revealed that Kras G12D significantly increased the expression of CD31, suggesting the presence of enhanced vascularization in tumors of ALK F1174L Kras G12D Lgr5-CreERT2 mice. Thus, the cooperation between ALK F1174 and Kras G12D , leading to the more aggressive type of cSCC, manifests through a combination of cellular mechanisms, including induction of EMT, increase in, proliferation and enhanced vascularization.
To delve further into the question of whether DMBA/TPA tumors exhibit molecular signature similar or distinct of the one observed in ALK-driven cSCC, we compared genetic profiles of ALK F1174L TECs to the previously published signature of DMBA/TPA tumors (Nassar et al, 2015) (Figs S4 and S5). 302 overlapping genes emerged as a result of this comparison, including 85 commonly up-regulated and 206 down-regulated (Fig S4A and C). The analysis of Ras-associated pathways in both mouse models confirms our previous observations and reveals that ALK executes a specific pro-tumorigenic program that is distinct of Ras-driven cSCC (Fig S4B and D).
Functional loss of TP53 gene is a hallmark of human SCC (van Kranen et al, 1995), and loss of p53 gene combined with Kras G12D overexpression promotes invasive SCC in mice (Lapouge et al, 2011;. Our analysis of human cSCC patients (Fig S2E) revealed the co-occurrence between ALK mutations and mutations in p53 gene (68 cases analyzed, 0 cases displayed ALK only mutations, and 19 cases displayed mutations in both genes). To determine whether the loss of p53 can promote metastases in ALK F1174 -driven model of SCC, we have crossed ALK F1174L Lgr5-CreERT2 mice to p53 fl/fl allele (Marino et al., 2000) (Fig 4A). Neither tumor-free survival nor progression-free survival were significantly altered upon p53 loss (Fig 4B and C). However, the frequency of tumors per mouse was increased in p53 fl/fl as compared with ALK F1174 single transgene (Fig 4D). Histological examination revealed several p53 loss-associated alterations. In addition to a higher degree of nuclear atypia, tumors isolated from ALK F1174L p53 fl/fl Lgr5-CreERT2 mice contained not only cysts, KA, AP, and SCC1 but also included more aggressive histological subtypes, AP3 and SCC2 (Figs 4E and F and S6A-C). Because p53 loss leads to an invasive cSCC in several mouse models of cSCC, we decided to investigate whether tumors obtained from ALK F1174L p53 fl/fl Lgr5-CreERT2 mice were characterized by a change in the EMT signature. We have decided to evaluate the expression of vimentin, a protein which is not only expressed at low levels in the dermal cells of the skin but is also one of markers of cells undergoing EMT . The quantification of the relative number of Vim + /Pan-Ck + cells within these tumors highlighted the presence of a greater number of cells with invasive characteristics, particularly in SCC2 (Fig 4G). The recombination in p53 floxed allele was verified using PCR analysis as shown in Fig 4H ( Marino et al., 2000). In summary, these data indicate that the p53 loss resulted in the most aggressive histological subtype of ALK F1174L -driven primary SCC. Our data are in agreement with several other reports demonstrating that Kras G12D when combined with the loss of p53 gene results in a spindle cell, highgrade SCC (Lapouge et al, 2011;White et al, , 2014. To monitor metastases, we took the advantage of the presence of luc reporter gene and subjected ALK F1174L p53 fl/fl Lgr5-CreERT2 mice to an in vivo live imaging at the time point when the termination criteria were met and before euthanizing (n = 7). As shown in Fig 4I, luciferase expression was readily detected in the skin as well as in the gut of ALK F1174L p53 fl/fl Lgr5-CreERT2 mice. However, no signs of distant metastases were detected in the internal organs (Fig 4I). This could be explained either by strict termination criteria of our study or low sensitivity of IVIS imaging, and therefore, we cannot completely rule out the occurrence of micrometastases with parameters under our instrumental detection limits.
Among down-regulated genes upon ALK F1174L overexpression, we have identified that several inhibitors of signal transducer and represent gene sets down-regulated in the tumor cells. Node size shows the size of gene sets. Nodes that clustered together are classes with same or similar function indication. Lines between the nodes represent association of the gene sets within the nodes. (F) Graphical representation of genes/pathways regulated by ALK F1174L , based on RNA-seq data. activator of transcription 3 (Stat3) were altered. We have observed down-regulation of the protein inhibitor of activated STAT, Pias3, a specific inhibitor of Stat3, which prevents DNA-binding activity of Stat3 and thereby abolished Stat3-mediated transcription (Chung et al, 1997). Moreover, the expression of another inhibitor of Stat3, Socs7 (Martens et al, 2005), was also decreased (Fig 2D). In addition, the expression of the tyrosine phosphatase Ptpn6 (Shp1), which was shown to dephosphorylate Stat3 but not Stat1 or Stat5 (Demosthenous et al, 2015), was also changed based on our RNAseq results. Previous reports showed that the expression of constitutively active form of Stat3 driven by the K5 promoter (K5.Stat3C mice) resulted in an increased number of tumors as well as a shorter latency period after DMBA/TPA treatment (Chan et al, 2008) and Stat3-deficient mice are resistant to chemically induced carcinogenesis (Chan et al, 2004). Based on these observations, we hypothesized that expression of ALK F1174L may trigger Stat3. To test this, we analyzed the protein homogenates from several ALK F1174L and ALK F1174L Kras G12D tumors with anti-Y705-Stat3 and anti-S727-Stat3 antibodies using Western blot ( Fig 5A). The vast majority of tumors showed phosphorylation of Stat3, whereas Stat3 KO ES cells were devoid of any signal. To evaluate whether ALK F1174L overexpression can lead to phosphorylation of Stat3, we compared HEK293T cells transiently overexpressing the mutated ALK (pCMV-ALK F1174L ) with their corresponding mock control (pCMV6-entry). As shown in Fig 5B, the overexpression of ALK F1174L in HEK293T cells resulted in Stat3 phosphorylation. On the basis of these results, we reasoned that Stat3 might play an essential role in mediating ALK F1174L -driven SCC formation. To test this possibility, we have crossed ALK F1174L Lgr5-CreERT2 mice to conditional Stat3 knockout mice (Jacoby et al, 2003) ( Fig 5C). ALK F1174L Stat3 fl/fl Lgr5-CreERT2 showed prolonged tumorfree survival and progression-free survival as compared with ALK F1174L Lgr5-CreERT2 mice. However, tumor formation still occurred (Fig 5D and E). We verified the presence of the recombined Stat3 floxed allele using PCR strategy (Moh et al, 2007) (Fig 5F). The resulting data revealed that 3/6 analyzed tumors from ALK F1174L Stat3 fl/fl Lgr5-CreERT2 mice were partially recombined and 3/6 tumors revealed the lack of the Stat3 recombined allele (Fig 5F), suggesting that the recombination was partial. To test whether the resulting tumors were still expressing Stat3, we have performed immunostaining for p-Stat3 and observed that 100% of tumors still contained Stat3-positive cells (Fig 5G). This observation is in alignment with the data obtained from the analysis of the recombination efficiency (Fig 5F). We next evaluated the percentage of the recombination and counted the number of HF either positive or negative for p-Stat3 protein (all WT HF were devoid of p-Stat3). Whereas the vast majority of HFs in ALK F1174L Lgr5-CreERT2 mice showed readily detectable p-Stat3 protein, less than 50% of HF were positive for p-Stat3 in ALK F1174L Stat3 fl/fl Lgr5-CreERT2 mice (Figs 5H and S6D). It is plausible that these p-Stat3-positive HFs occurred because of the low recombination efficiency and likely give rise to the tumors observed.
Although our data suggest that ALK F1174L -mediated tumorigenesis, at least in part, might occur via its downstream effector STAT3, further experimental evidence is required to determine the significance of these findings.
Our findings reveal a previously unknown role of oncogenic ALK signaling in cSCC. We show that the expression of a constitutively active ALK F1174L in mice can lead to the development of aggressive forms of cSCC (Fig 6). On the molecular level, ALK F1174L can cooperate with known cSCC drivers, including Kras G12D as well with the loss of tumor suppressor p53. Moreover, our data demonstrate that Stat3 is essential for mediating the oncogenic effect of ALK F1174L . In addition to our data uncovering the essential role of oncogenic ALK F1174L in mouse cSCC, we show that human samples of cSCC contain ALK mutations as well. Therefore, our data provide a rational for oncogenic ALK as a novel therapeutic target and can serve as a basis for the design of future clinical trials.

Transgenic mice
All animal experiments have been approved by the cantonal veterinary authorities of Zurich, in accordance with Swiss laws. For DNA isolation, mouse biopsies were lysed using lysis buffer (5M NaCl, 2M Tris, pH 88.5, 0.5M EDTA, and 20% SDS) and proteinase K. DNA was  Cells that co-expressed Pan-Ck and vimentin were counted as cells in epithelial-to-mesenchymal transition (EMT). All tumors diagnosed as cSCC type 1 or 2 from different mice were analyzed, and quantification of the relative number of cells in EMT per tumor is represented by a dot. Analysis of tumors showed that no significant changes were observed between SCC1 tumors of three Lgr5-CreER T2 ;LSL-ALK F1174L or three Lgr5-CreER T2 ;LSL-ALK F1174L ;p53 fl/fl mice (4.380 ± 1.425; 7.360 ± 1.996. Mann-Whitney test P = 0.3277). Strong increase in the EMT rate was instead noted in SCC2 tumors when compared with SCC1 tumors of the same cohort of mice (7.360 ± 1.996; 17.10 ± 1.875. Mann-Whitney test P = 0.0056) or to Lgr5-CreERT2;LSL-ALKF1174L mice (4.380 ± 1.425; 17.10 ± 1.875. Mann-Whitney test P = 0.0008). All values are showed as median with 95% confidence interval. (H) Above: the schematic structure of the p53 floxed allele and graphical representation of PCR strategy to determine the recombination efficiency of LoxP sites that drives the removal of exons 2-10 of the p53 gene. White boxes represent exons, arrowheads represent the LoxP sites, and purple lines indicate precipitated in 50% lysate and 50% isopropanol and centrifuged at 20.817g for 30 min. After washing in 70% ethanol, 2 µl of DNA were mixed with 10 µl KAPA Taq ReadyMix with dye (KK1024; Kapa Biosystems) and with 0.8 µl of each primer (10 µM) and finally brought to 20 µl with MilliPore water. The following primers with according melting temperatures were used: 60°C, LSL-ALK_7083_Fw (CCATCAGTGACCTGAAGGAGG) LSL-ALK_7083_Rv (CACGTGCAGAAGG  TCCAGC) . The recombination efficiency PCRs were performed as previously described for p53 (Marino et al., 2000) and Stat3 (Moh et al, 2007).

Induction of Cre activity, termination criteria, and in vivo imaging system
For activation of Cre activity, 6-12-wk-old mice were shaved on the back skin and 4-Hydroxytamoxifen (4-OHT) 25 mg/ml (≥70% Z isomer H7904; Sigma-Aldrich) was topically administered for three consecutive days, using a small paint brush on both ears (inside and outside), on the tail and on the shaved patch of the back skin. The mice were euthanized when termination criteria was reached. In most cases, size of tumors (>0.5 cm of diameter, n = 2) was the adopted termination criteria. Secondary, mice were euthanized if developed one tumor bigger than 1 cm of diameter or if tumor was ulcerative or bleeding. Mice that showed wounds (signs of fights) were withdrawn from experiment. On the day of euthanizing, the animals were anesthetized with isoflurane (B506; Abbott) and shaved with a waxing cream (Veet). Then mice have been injected with 150 mg/kg body weight of XenoLight D-Luciferin-K+ Salt Bioluminescent Substrate (122799; Perki-nElmer) and imaged with IVIS Lumina III (PerkinElmer). In vivo and ex vivo analysis was performed, and tissue was collected according with the signal acquired for histological analysis. Skin, organs, and tumors were embedded in paraffin for immunohistochemistry and hematoxylin and eosin staining.

Quantifications
Quantification of Vimentin + and pan-cytokeratin + cells was performed using a machine learning bio-image analysis software: QuPath (Bankhead et al, 2017). The software was trained to identify tumor cells, stromal cells, EMT cells, and false positives. Quantification of CD31 signal and ki67 + cells over pan-cytokeratin area was performed with CellProfiler (www.cellprofiler.org), a cell image analysis software. Proliferation rate was calculated by identification of tumor area and subsequent count of proliferating nuclei within the mask made out of the tumor. Relative enrichment of vessels within the tumors was performed by identification of tumor and vessel areas and subsequent quantification with an inverse mask out of tumor area for CD31 + signal. The correlation of the number of vessels with the CD31 + signal was statistically significant, out of three tumors quantified (Pearson r = 0.9990; two-tailed P = 0.0289). For the quantification of vimentin staining, we have used five different Lgr5-CreER T2 ;LSL-ALK F1174L ;LSL-Kras G12D mice from which we analyzed 21 independent tumors (one section per tumor) and five different Lgr5-CreER T2 ;LSL-ALK F1174L mice from which we analyzed 12 independent tumors. For CD 31 quantification, we used 60 independent Lgr5-CreER T2 ;LSL-ALK F1174L ;LSL-Kras G12D tumors (one section per tumor) and 28 independent Lgr5-CreER T2 ;LSL-ALK F1174L tumors (one section per tumor). For Ki67 staining, we used 70 independent Lgr5-CreER T2 ;LSL-ALK F1174L ;LSL-Kras G12D tumors (one section each) and 46 independent Lgr5-CreER T2 ;LSL-ALK F1174L tumors (one section per tumor).

Statistical analysis
Statistical analysis and graphical representation were performed with GraphPad Prism 5 software. The n reported in the figures or figure legends always refer to the number of mice used for each analysis. Column analysis was always performed with t test or Mann-Whitney test accordingly if the group of data considered in examination had a normal distribution or not (D'Agostino & Pearson omnibus normality test). Log-rank (Mantel-Cox) test was performed for statistical analysis of survivals.

DNA extraction from paraffin tissue and sequencing of Alk exon 23
Paraffin-embedded cSCC from DMBA/TPA-treated mice were kindly gifted by the Laboratory of Prof. Sabine Werner. A few sections of tumors were collected in an Eppendorf tube together with 500 μl lysis buffer, which was incubated at 95°C for 10 min, and then centrifuged at 4°C, 20.817g for 5 min and the paraffin was removed. DNA extraction was performed as described above. Exome 23 of Alk was then amplified using KAPA Taq ReadyMix with dye (KK1024; Kapa Biosystems) with specific primers (msAlk-ex23_fw: CTATG-CATCGCCCCAGGAAG, msAlk-ex23_Rv: GGCTGACTCCCAGGAGCCCA; MT = 60°C), and amplicons were sent for Sanger sequencing to Microsynth. Sequencing results were analyzed with Sequencer 5.1 (Genecode) and compared with reference sequence downloaded from Ensembl (www.ensembl.org).