Introduction

Melanoma, the most lethal form of skin cancer, is increasing in incidence and mortality in the United States (Chin et al., 2006). Alterations of three genetic pathways characterize the majority of melanomas: activation of the RAS–RAF–extracellular signal-regulated kinase (ERK) pathway, inactivation of the p16INK4a–CDK4–retinoblastoma (RB) pathway and inactivation of the alternate open reading frame (ARF)–murine double minute 2–p53 pathway. The RAS–RAF–ERK pathway is a mitogen-activated kinase cascade, which induces diverse transcriptional changes associated with increased proliferation, survival, motility and immune surveillance (Pearson et al., 2000; Johansson et al., 2007; Petermann et al., 2007; Shields et al., 2007). Per the Compilation of Somatic Mutations in Cancer (COSMIC, see (Forbes et al., 2006)), activating mutations of N-RAS occur in 21% of melanomas, those of K-RAS in 2% and of B-RAF in 44%, and these lesions seem to be mutually exclusive, consistent with an epistatic relationship. The presence of activating B-RAF mutations in a high proportion of dysplastic nevi indicates that aberrant activation of the RAS–RAF pathway is not sufficient for tumorigenesis and that additional genetic lesions are required to effect malignant transformation (Davies et al., 2002). Strong RAS–RAF activation has been shown to induce p16INK4a expression and ‘oncogene-induced senescence’ in vitro and in vivo (Serrano et al., 1997; Zhu et al., 1998; Michaloglou et al., 2005), providing a tumor biological basis for cooperative interactions between p16INK4a extinction and RAS–RAF activation.

The INK4/ARF (CDKN2a/b) locus at 9p21 encodes three distinct tumor suppressor proteins—p15INK4b, p16INK4a and ARF. A significant body of human and murine genetic data have determined that p16INK4a and ARF have an important role in melanoma suppression (reviewed in (Kim and Sharpless, 2006)), whereas the role of p15INK4b in this disease is less clear. INK4a/ARF deletion occurs in 50% of human melanomas, making it the most common site of genomic loss in melanomas (Curtin et al., 2005), with homozygous deletion being relatively common (Grafstrom et al., 2005). ARF blocks murine double minute 2-induced degradation of p53, whereas p16INK4a and p15INK4b inhibit cyclin-dependent kinases 4 and 6 , leading to RB-hypophosphorylation and cell cycle arrest. Although ARF loss is the most common lesion of the ARF–murine double minute 2–p53 pathway in melanoma, mutational inactivation of p53 is observed in 10–30% of melanomas (Albino et al., 1994; Sparrow et al., 1995; Akslen et al., 1998; Zerp et al., 1999).

Several genetically engineered mouse models of melanoma using constitutively active and inducible transgenes and germline knockout alleles have studied the interaction of Ras activation and p16INK4a/Arf/p53 loss. Mice engineered with a constitutive, melanocyte-specific, mutant H-Ras transgenic allele on an Ink4a/Arf-deficient background (TyrRas Ink4a/Arf −/−) were shown to develop melanoma with high penetrance and short latency (Chin et al., 1997). Using a doxycycline-inducible mutant H-Ras allele on an Ink4a/Arf-deficient background, it was determined that persistent activation of H-Ras is required for both melanoma initiation and maintenance, thus establishing the concept of tumor maintenance (Chin et al., 1999). Melanocyte-specific expression of an activated N-Ras transgene in an Ink4a/Arf-deficient background can also generate melanomas that can metastasize to the lymph nodes, lung and liver (Ackermann et al., 2005). A role for p53 in melanoma was substantiated in TyrRas transgenic mice that were mutant for p53, which develop melanomas that retain strong p16INK4a expression (Bardeesy et al., 2001). It is noteworthy and of relevance to this study that the TyrRas transgene on a germline p16INK4a-null background exhibited a weaker melanoma phenotype when compared with the TyrRas transgene on either a germline Arf or p53-null background (Bardeesy et al., 2001; Kannan et al., 2003). This murine observation is curious in light of the prominent role of p16INK4a extinction in human melanomas, raising the possibility of cross-species differences in the rate-limiting role of this key RB pathway component and/or of the fact that germline vs somatic inactivation of p16INK4a is a critical parameter of its impact on melanocyte biology.

In aggregate, these genetic studies have clearly established the cooperation of Ras activation and Arf/p53 inactivation in melanomagenesis, but have suggested a more modest role for p16INK4a loss in driving melanoma development in the mouse. This modest impact of p16INK4a loss seems to contrast with the prominent role of p16INK4a in human melanoma, as reflected by the robust melanoma-prone condition in the setting of germline or somatic p16INK4a mutations that spare Arf or p15INK4b in many cases (FitzGerald et al., 1996; Smith-Sorensen and Hovig, 1996; Walker et al., 1998; Daniotti et al., 2004). It is important to note that a few cases of ARF germline or somatic mutations that spare p16INK4a have been identified, suggesting that both products of the INK4a/ARF locus have independent tumor suppressor activity in melanoma (Rizos et al., 2001; Hewitt et al., 2002). Several possible explanations for the apparent cross-species differences have been suggested, including differences in the anticancer potency of p16INK4a vs ARF between mouse and man (Gil and Peters, 2006; Evan and d’Adda di Fagagna, 2009), bias derived from studies using transgene-driven overexpression of activated Ras (Tuveson et al., 2004), as well as epigenetic compensation in mice germline null for p16INK4a. With regard to the latter possibility, germline p16INK4a inactivation is associated with increased, presumably compensatory, expression of other negative regulators of the cell cycle, such as p15INK4b and p18INK4c (Krimpenfort et al., 2007; Ramsey et al., 2007; Wiedemeyer et al., 2008). Furthermore, somatic loss of RB in vitro has been shown to transiently induce cell cycle entry that is later compensated by other RB-family proteins (Sage et al., 2003). Therefore, the apparent differences between human and murine melanoma with regard to tumor suppressor pathway inactivation could reflect technical features of the experimental model systems used or species differences between mouse and man.

Conditional activated knock-in and tetracycline-inducible alleles and conditional knockout alleles have enabled the study of somatic mutations in adult mice. The use of conditional strategies are particularly relevant to melanoma research, in which epidemiological studies have suggested that sun exposure during childhood may be more effective at inducing melanoma than sun exposure later in life (Bader et al., 1985; Harrison et al., 1994; English et al., 2005). In accordance with this view, Merlino and colleagues have shown that ultraviolet B treatment of HGF/SF transgenic mice effectively induces melanoma only in the neonatal period (3.5 days), but not in adult mice (Noonan et al., 2001). Similar results were noted in TyrRas Arf−/− mice, which also pointed to the p16INK4a–RB pathway as a key target of ultraviolet carcinogenic actions (Kannan et al., 2003). These results could indicate that ultraviolet B is less efficient at causing oncogenic mutations in adult mice, or that adult melanocytes are intrinsically more resistant to the transforming effects of oncogenic mutations. The latter model would be consistent with the notion of a ‘developmental window’ at young age during which melanocytes are more susceptible to transformation by a given set of oncogenic events.

Building on this previous history of melanoma modeling in mice, we used somatic and inducible alleles, including a newly engineered ‘floxed’ p16INK4a-specific null allele, to study the effects of somatic, melanocyte-specific loss of p16INK4a and/or p53 combined with somatic activation of an oncogenic K-Ras allele on melanomagenesis in neonatal and adult mice.

Results

To determine the effect of somatic loss of p16INK4a and/or p53 and activation of oncogenic K-Ras, we used an established 4-hydroxytamoxifen (4-OHT)-inducible melanocyte-specific CRE allele (Tyr-CRE-ERT2 (Bosenberg et al., 2006)), together with conditional alleles such as Lox-Stop-Lox-KRasG12D (Johnson et al., 2001), a floxed p53-null allele (p53L (Jonkers et al., 2001) and a newly generated floxed p16INK4a-null allele (p16L). In a subset of mice, CRE recombination was confirmed using a β-galactosidase reporter allele that, on Cre-mediated excision of a Lox-Stop-Lox cassette, expresses the enzyme; it was observed that cells stain positively in the β-galactosidase assay (Rosa26-LSL-β-galactosidase, (Soriano, 1999)) (Supplementary Figure 1). The p16L allele was generated by inserting LoxP elements ∼3.5 kb 5′ to exon 1α and immediately 3′ to the p16INK4a exon 1α (Supplementary Figure 2a, see also Supplementary Methods). The 5′ LoxP element was placed in a variable number of tandem repeat region that is not conserved between 129SvJ and C57Bl/6 strains, and therefore not likely to be a critical Ink4a/Arf regulator. Correct targeting of the locus was confirmed by Southern blotting, PCR and sequencing (Supplementary Figure 2b and c). CRE expression in p16L/L murine embryo fibroblasts and melanocytes demonstrated selective p16INK4a exon 1α excision, loss of p16INK4a protein expression (Supplementary Figure 3a and c) without an alteration in Arf expression (Supplementary Figure 3b and c). In contrast to deletion of Arf or p53, excision of p16INK4a with or without concomitant K-Ras activation did not enhance melanocyte growth in vitro (Supplementary Figure 3d).

Intercrosses produced the Tyr-CRE-ERT2+KRasLSL/+p16L/Lp53L/L experimental cohort (TKp16L/Lp53L/L), as well as indicated control cohorts. TKp16L/Lp53L/L mice and control littermates were topically treated as neonates with 4-OHT to activate CRE and induce recombination in melanocytes as previously described (Bosenberg et al., 2006; see schematic in Supplementary Figure 1). Within 14 weeks, all 4-OHT-treated TKp16L/Lp53L/L mice developed aggressive tumors and hence had to be killed (Figure 1a, Table 1). Nonmelanoma tumors were not observed in 4-OHT-treated TKp16L/Lp53L/L mice (n=15) (Table 1). In contrast, only 1 of 14 of the 4-OHT-treated TKp16+/+p53+/+ mice developed melanoma at 60 weeks post treatment (Table 1). PCR confirmed that all TKp16L/Lp53L/L tumors exhibited a recombination of K-Ras, p16INK4a and p53 alleles (Figure 1b). Western analysis of TKp16L/Lp53L/L tumors demonstrated an absence of expression of p16INK4a and p53, as well as of p21CIP, a p53 target gene (Supplementary Figure 4a). These data demonstrate potent cooperation between activated K-Ras and combined p16INK4a and p53 loss in driving a highly penetrant and aggressive melanoma condition.

Figure 1
figure 1

Somatic K-Ras activation coupled with p16INK4a and/or p53 inactivation induces melanoma. (a) Kaplan–Meier analysis of melanoma-free survival of 4-OHT-treated TK (n=14), TKp16L/Lp53L/L (n=15), TKp16L/L (n=11) and TKp53L/L (n=11) mice. Pairwise statistical comparisons: TK vs TKp16L/L (P<0.001), TK vs TKp53L/L (P<0.001), TKp16L/Lp53L/L vs TKp16L/L (P<0.001), TKp16L/Lp53L/L vs TKp53L/L (P<0.001) and TKp16L/L vs TKp53L/L (P=NS). (b) PCR for the recombined alleles: K-RasLSLG12D, p16L/L, p53L/L in ears and tumors from 4-OHT-treated or untreated TKp16L/Lp53L/L mice. (c) Representative tumors from each cohort are shown. Tumors from mice of all genotypes exhibited a spindle-like morphology by hematoxylin and eosin staining and were stained for tyrosinase-related protein 1, confirming melanocytic origin. Original magnification: × 100.

Full size image
Table 1 Summary of tumor features by genotype
Full size table

To further dissect the contributions of each tumor suppressor in melanoma development, we generated and compared TKp16L/L and TKp53L/L cohorts. These animals were treated neonatally by topical 4-OHT and followed up for tumor development, as outlined in Supplementary Figure 1. Somatic, melanocyte-specific loss of either p16INK4a or p53 alone in this system led to melanoma formation (Figure 1a) with significantly longer tumor latencies than seen in the TKp16L/Lp53L/L cohort but, notably, shorter than the latencies of the TKp16+/+p53+/+ mice. Unexpectedly, the 4-OHT-treated TKp16L/L and TKp53L/L animals showed comparable median latencies of 24 and 31 weeks, respectively. The 4-OHT-treated compound heterozygous mice TKp16L/+ and TKp53L/+ showed accelerated tumorigenesis compared with 4-OHT-treated TK mice (Supplementary Figure 4b and Table 1). As expected, mice with somatic inactivation of p16INK4a and p53 without concomitant Ras activation (Tp16L/Lp53L/L) were only weakly tumor prone (1 of 19 mice, Table 1).

At the histopathological level, tumors from the four cohorts (TK, TKp16L/L, TKp53L/L and TKp16L/Lp53L/L) exhibited a spindle-like morphology with scant pigmentation and positive staining for TRP1 (Figure 1c, Supplementary Figure 5). These ‘amelanotic’ melanomas are similar to those reported by Chin et al. (1997) in the H-Ras-driven TyrRas Ink4a/Arf−/− model, but different from the heavily pigmented tumors noted in N-Ras transgenic, Ink4a/Arf-deficient mice (Ackermann et al., 2005) and the conditional B-Raf and Pten mice (Dankort et al., 2009). In contrast to the TyrRas Ink4a/Arf−/− model, in which tumors are limited to ears, trunk, tail and uvea, tumors in the conditionally activatable model occurred on the flank, ear and tail, and uveal tumors were not observed. An increased tumor multiplicity was noted in 4-OHT-treated TKp16L/Lp53L/L mice compared with treated TK, TKp16L/L or TKp53L/L animals; however, there were significantly more tumors from treated TKp16L/Lp53L/L mice when compared with TK, TKp16L/L or TKp53L/L animals, suggesting the increase in multiplicity could be due to an increase in the overall number of tumors (Supplementary Figure 4c). These data suggest that p16INK4a and p53 exert important and nonredundant antimelanoma roles in the context of melanocyte-specific Ras activation, a finding that may reflect the prominent involvement of the Ink4a/Arf locus that dually encodes components of the RB and p53 pathway (Sherr, 2001).

Mice from all cohorts (TK, TKp16L/L, TKp53L/L and TKp16L/Lp53L/L) developed melanocytic proliferation in the skin before tumor formation. Paws and tails of treated mice from all cohorts developed pigmented macules that were not present in wild-type or 4-OHT-untreated littermates (Supplementary Figure 6a and b). Wild-type and 4-OHT-untreated animals had little melanin production in sections of the ear when compared with treated mice from other cohorts. The most pronounced effects were observed in skin from 4-OHT-treated-TKp16L/Lp53L/L mice (Supplementary Figure 6c). These findings suggest that K-Ras activation alone is sufficient to produce a modest increase in melanocytic proliferation, and that this excess, nonmalignant proliferation is partially restrained by the combined activities of p16INK4a and p53.

To determine the role of the INK4–cyclin-dependent kinase 4 (Cdk4)–Rb pathway in melanocyte transformation in vivo, lysates of primary tumors were analyzed. In tumors from 4-OHT-treated mice of informative genotypes (TKp53L/L, TKp16L/+p53L/+and TKp16L/+p53L/L), the majority (8 of 11) of tumors demonstrated absent p16INK4a expression (Figure 2a and b). Recently, expression of p15INK4b has been shown to have a ‘back-up’ tumor suppressor role in p16INK4a germline-deficient animals (Krimpenfort et al., 2007; Ramsey et al., 2007). In tumors from animals with somatic p16INK4a and/or p53 inactivation, however, expression of p15INK4b was observed in 8 of 12 tumors (Figure 2c). Therefore, any such compensatory increase in p15INK4b expression either does not occur or is insufficient to prevent tumorigenesis in melanocytes in the setting of somatic deletion of p16INK4a. In aggregate, these results imply that acquired stochastic loss of p16INK4a, but not of p15INK4b, is commonly associated with in vivo tumor progression in melanocytes harboring somatic, but not germline, p53 inactivation.

Figure 2
figure 2

Analysis of INK4–Cdk4/6–Rb and Arf–p53 pathways in representative melanomas. (a) and (b) Western analysis of p16INK4a and Arf in tumors of the indicated genotypes. (c) Western analysis of p15INK4b expression in tumors of the indicated genotypes. (d) Western analysis of phospho-p53 and Arf in tumors of the indicated genotypes.

Full size image

Similarly, we analyzed the Arf–p53 pathway in lysates from tumors from mice of informative genotypes. In accordance with previous reports (Chin et al., 1997; Bardeesy et al., 2001; Sharpless et al., 2002), we noted that tumors lacking p53 function almost always exhibited a markedly increased Arf expression (Figure 2b and d). As expected, tumors from contemporaneously analyzed TyrRas Ink4a/Arf−/− mice showed Arf loss and expression of phosphorylated p53 (Figure 2d). Analysis of TKp16L/L tumors demonstrated that four of five tumors exhibited phospho-p53 expression (Figure 2d), whereas three of five tumors exhibited Arf expression (Figure 2a and d). In particular, at least two TKp16L/L tumors (#10 and 11, Figure 2d) exhibited phospho-p53 and retained Arf expression, consistent with the possibility of a retained Arf–p53 pathway function in these tumors. In tumors from 4-OHT-treated mice that were heterozygous for the floxed p53 allele, absent Arf expression was noted in two of five tumors (Figure 2b), suggesting that Arf loss can be associated with tumor progression even in a p53L/+ background. Although some TKp16L/L tumors retained Arf–p53 expression, other TKp16L/L tumors lost Arf expression, suggesting that Arf loss can contribute to progression in some tumors. Overall, these results suggest that somatic inactivation of the Arf–p53 pathway can occur, but is not obligate, for progression of K-Ras-induced, p16INK4a-deficient melanoma.

In an effort to benchmark these new somatic models of melanoma to an earlier, well-characterized melanoma model, we also compared tumors from 4-OHT-treated TKp16L/Lp53L/L mice with contemporaneously analyzed TyrRas Ink4a/Arf−/− mice (Chin et al., 1997). It should be acknowledged that several relevant biological differences limit this comparison, including somatic vs germline genetic events; single copy, endogenous K-Ras mutation vs multicopy, transgenic H-Ras mutation; and p53 vs Arf loss. Although the tumor-inducing genetic events occur later (postnatally) in the TKp16L/Lp53L/L model, the tumor latencies were similar between the somatic and germline models (Figure 3a). Moreover, when the growth rates of comparably sized tumors were analyzed in vivo, melanomas from TKp16L/Lp53L/L mice were noted to grow more rapidly with increased ulceration (Figure 3b). All tumor-bearing TKp16L/Lp53L/L mice had to be killed by day 8 of observation because of advanced tumors, whereas TyrRas Ink4a/Arf−/− mice can routinely be observed for more than 20 days before having to be killed. Cell lines derived from TKp16L/Lp53L/L tumors demonstrated increased S-phase and aneuploidy compared with TyrRas Ink4a/Arf−/− tumors (Figure 3c, Supplementary Table 1). Consistent with loss of p53 vs Arf inactivation, Arf levels were increased and p21CIP levels decreased in TKp16L/Lp53L/L tumors compared with lines from TyrRas Ink4a/Arf−/− tumors (Figure 3d). Further characterization of protein expression in tumor lysates from the different models indicated little difference in the expression of tested downstream mediators of Ras signaling (for example, phospho-ERK and phospho-AKT, Figure 3d). These results are consistent with the possibility of a more aggressive in vivo behavior of TKp16L/Lp53L/L tumors as a result of p53 loss compared with Arf loss in the TyrRas Ink4a/Arf−/− model, although other differences such as strain background or Ras isoform differences could also contribute. Importantly, however, despite technical and biological differences between the models, they share several common features including tumor histological appearance, lack of precursor nevus formation, scant or absent tumor pigmentation and lack of metastasis. The shared features of Ras-driven, p16INK4a-deficient melanoma models are in contrast to the features of the pigmented, metastatic tumors seen in B-Raf mutant, Pten-deficient tumors reported by McMahon and Bosenberg (Dankort et al., 2009).

Figure 3
figure 3

TKp16L/Lp53L/L melanomas are more aggressive than those from TyrRas Ink4a/Arf−/− mice. (a) Kaplan–Meier analysis of melanoma-free survival of TKp16L/Lp53L/L (n=15) and TyrRas Ink4a/Arf−/− (n=12) mice (P=NS). (b) in vivo analysis of tumor growth of TKp16L/Lp53L/L (n=3) and TyrRas Ink4a/Arf−/− (n=6) tumors. *P=0.01 at days 6 and 8. (c) in vitro flow cytometry analysis of the cell cycle in tumor cell lines generated from TKp16L/Lp53L/L and TyrRas Ink4a/Arf−/− tumors. Increased S-phase and aneuploidy are noted in tumors from TKp16L/Lp53L/L mice (see also Supplementary Table 1). (d) Western analysis of indicated proteins in lysates from TKp16L/Lp53L/L and TyrRas Ink4a/Arf−/− tumors.

Full size image

As epidemiological and murine modeling results have suggested that ultraviolet B exposure is more oncogenic at young ages, we studied the in vivo transformability of somatic K-Ras activation and combined p16INK4a/p53 loss in adult mice. Toward this end, small (1 cm × 1 cm) patches of flank skin were topically treated with 4-OHT in 8-week-old TKp16L/Lp53L/L mice and then manually depilated to stimulate melanocyte proliferation (see Supplementary Figure 1 and Ruzankina et al., 2007). By 15 weeks post 4-OHT treatment, tumor development was observed in the majority (17 out of 27) of treated TKp16L/Lp53L/L adult mice (Figure 4a and b), with all tumors developing at the site of 4-OHT treatment (Figure 4b and c). The median tumor latency of TKp16L/Lp53L/L mice was modestly longer in animals treated as adults compared with neonates (Figure 4a), but this may be explained by the considerably more extensive CRE-mediated recombination observed in neonatally treated mice (not shown, see also (Bosenberg et al., 2006)). To serially monitor melanoma development in adult mice, a green fluorescent protein CRE-reporter allele was crossed into the TKp16L/Lp53L/L model. Through intravital confocal microscopy of 4-OHT-treated TKp16L/Lp53L/L adult mice, CRE recombination was confirmed in 4-OHT-treated skin, but not noted in untreated skin (Figure 4c). Despite a basal level of autoflourescence in untreated skin, all TKp16L/Lp53L/LRosa26-LSL-GFP+ tumors expressed green fluorescent protein, allowing for visualization of individual tumor cells in anesthetized, living mice. These data are not consistent with the model of a ‘developmental window’ for melanocyte transformability, but instead show that adult melanocytes can be efficiently transformed in vivo.

Figure 4
figure 4

Melanocytes from adult TKp16L/Lp53L/L mice are transformable in vivo. (a) Kaplan–Meier analysis of melanoma survival of 4-OHT-treated TKp16L/Lp53L/L mice as adults or neonatal pups. Control mice (4-OHT-untreated or lacking the K-Ras-LSL allele (n=9)) did not develop tumors. Pairwise comparison of adult-treated (n=27) vs neonatally treated TKp16L/Lp53L/L mice (n=15) (P=0.003). (b) Adult TKp16L/Lp53L/L mice developed tumors in depilated and 4-OHT-treated skin patches. 4-OHT-treated control mice did not develop tumors. (c) Serial intravital confocal imaging of skin from TKp16L/Lp53L/LRosa26-LSL-GFP+ mice that were 4-OHT-treated as adults demonstrates green fluorescent protein expression and tumor growth. Melanocytes of untreated skin or of skin of control mice do not express green fluorescent protein; autoflourescence of hair follicles is noted. Original magnification: × 200.

Full size image

Discussion

In this study, we provide evidence that somatic p16INK4a deletion and oncogenic K-Ras activation promote the rapid onset of aggressive melanomas in vivo. To test somatic loss of p16INK4a in melanoma, a novel p16INK4a conditional allele was generated that exhibits unperturbed expression of other Ink4a/Arf locus genes, p15INK4b and Arf. We observed potent cooperation between somatic Ras activation along with somatic p16INK4a loss, whereas somatic Ras activation and p53 deletion yield a somewhat more modest melanoma-prone condition, a finding consistent the more prominent tumor suppressor role of p16INK4a relative to p53 in human melanoma. These data support the view that conditional inactivation of p16INK4a provides a more faithful platform for the generation of refined mouse models of human melanoma, and possibly of other cancers.

Moreover, as opposed to work using germline alleles (Kannan et al., 2003; Sharpless et al., 2003), the use of these conditional systems provides evidence that direct Arf-p53 inactivation is not required for in vivo murine melanoma formation. That is, some tumors arising in TKp16L/L mice (Figure 4d) exhibited Arf and phospho-p53 expression. Although the analysis of TKp16L/L tumors suggests that the ARF–p53 pathway retained expression, this does not eliminate other potential mechanisms that could abrogate the function of this pathway, such as point mutations or murine double minute 2 amplification. Nevertheless, although direct Arf or p53 inactivation does not seem to be an obligate event for in vivo melanomagenesis, our data and those of others certainly suggest that either lesion should strongly enhance tumor progression.

Although the primary tumors of TKp16L/Lp53L/L mice were aggressive in terms of local invasion, we did not detect metastasis, a hallmark of human melanoma. Although metastasis has been observed in both an N-Ras transgenic model that used a multicopy transgene with a presumed high level Ras activation (Ackermann et al., 2005) and a somatic B-Raf mutant and Pten mouse model of melanoma (Dankort et al., 2009), metastasis has otherwise not been a feature of most murine melanoma models involving Ras activation, p53 loss and p16INK4a loss (Chin et al., 1997; Bardeesy et al., 2001; Sharpless et al., 2003), suggesting that other factors are involved in metastasis in melanoma. One potential explanation could relate to Pten status, which is also inactivated with high frequency in human melanoma (reviewed in (Chin et al., 2006)). Several studies have suggested a prominent role for both the lipid and protein phosphatase activities of PTEN in regulating cellular motility and invasion (Tamura et al., 1998; Marino et al., 2002; Raftopoulou et al., 2004; Lin et al., 2009). Similar to other large tumor suppressor genes that can be inactivated by a variety of mechanisms, studies of PTEN in human cancer have been limited by difficulty in determining whether the gene is inactivated in a given tumor. Ongoing comprehensive studies such as the Cancer Genome Atlas involving multidimensional genome-scale analysis of large clinically annotated samples sets should determine whether PTEN loss is a primary driver of metastasis in human melanoma.

In summary, in this paper, we describe a genetically engineered murine model of melanoma that is based on somatic Ras activation and p16INK4a loss, which are hallmarks of the human disease. This study has shown that somatic inactivation solely of p16INK4a, without the corresponding targeting of Arf, is potently tumorigenic in vivo. Inactivation of p16INK4a and p53 with Ras activation in adult melanocytes was potently oncogenic, suggesting that adult cells are not substantially less transformable than neonatal melanocytes. As novel targeted therapies for the RAS pathway are emerging for human melanoma (Engelman et al., 2008), these refined genetic models harboring genetic alterations encountered in the human disease should enable the more accurate use of such agents in specific melanoma genotypes in the future.

Materials and methods

Mouse colony

Animals were generated and genotyped as previously described (Johnson et al., 2001; Jonkers et al., 2001; Bosenberg et al., 2006) and were N1 in C57Bl/6. Mice were housed and treated in accordance with protocols approved by the institutional care and use committee for animal research at the University of North Carolina. Pups were treated for 3 consecutive days with 4-OHT (Sigma H7904, Sigma, St Louis, MO, USA) at 25 mg/ml in dimethyl sulfoxide starting at day 2 (Bosenberg et al., 2006). Tumor growth and survival were assessed three times per week by caliper measurements of tumor areas (width × length (mm2)) and measurement was taken when tumors were between 20 and 30 mm2. Tumor growth was normalized to day 1 values and analyzed using GraphPad Prism software (GraphPad Software, La Jolla, San Diego CA, USA).

Generation of p16L/L allele and derivative cells

The p16INK4a conditional (floxed) allele (p16L) was generated and validated as described in Supplementary Methods. To evaluate the effects of p16INK4a excision, 1 × 106 murine embryonic fibroblasts were treated with CRE-expressing adenovirus (Iowa Vector Core, Iowa City, IA, USA) at a multiplicity of infection of 100 for 24 h. Cells were split and harvested 5 days after CRE treatment. To avoid the effects of adenoviral CRE expression on Arf–p53 expression, we crossed the p16L/L allele into lentiviral transgenic CRE-ERT2 mice (Ruzankina et al., 2007). Murine embryo fibroblasts containing the CRE-ERT2 transgene and homozygous for the p16L/L allele were generated and cultured on a 3T9 protocol. After four passages, cells were treated with 10 nm 4-OH-tamoxifen (Sigma T5648) for 40 h. After treatment, cell lysates were harvested at passages 1, 3 and 5. Primary melanocyte cultures from mice of the indicated genotypes, including the TyrCRE-ERT2 allele, were prepared as described previously (Bennett et al., 1989; Spanakis et al., 1992) and plated on collagen-coated dishes. Cells were treated with or without 4-OHT at 20 days post isolation for 48 h. Successful recombination was confirmed at 22 days and western blot was performed at 45 days. For growth curve analysis, cells were counted at indicated times and replated at 3 × 104 cells per ml. The mean relative population increase was calculated as the number of doublings.

Cell culture

Tumor cell lines were generated and maintained as described (Sharpless et al., 2002). For cell cycle analysis, exponentially growing cells were pulsed for 15 min with 10 μM bromodeoxyuridine (BrdU) in Dulbecco's modified Eagle's medium+10% fetal calf serum before harvest. Cells were fixed, permabilized and stained with anti-BrdU-APC according to BrdU Flow Kit instructions (BD Pharmingen, San Jose, CA, USA) and then resuspended in phosphate-buffered saline+0.2% fetal calf serum+25 μg/ml propidium iodide with RNase and analyzed using a CyAn (Dako) flow cytometer (Cyan ADP, DAKO, Beckman Coulter, Brea, CA, USA) and FlowJo (Treesoft) software (Treestar, Inc., Ashland, OR, USA). Experiments were conducted in triplicate.

4-OHT treatment of adult mice and intravital microscopy

Hair on both flanks of 50- to 55-day-old mice was trimmed (1–2 cm2 section) for treatment. One flank was treated with 20 mM 4-OHT dissolved in ethanol (treated skin) and the other flank was treated with 100% ethanol (control). Treatment was repeated the next day. Five days after treatment, mice were anesthetized with 2% isofluorane and depilated by hand (Ruzankina et al., 2007). For serial confocal imaging of tumors, mice were anesthetized with ketamine/xylazine and ear fur was removed using chemical depilation. Green fluorescent protein fluorescence in the ears was excited with 900-nm light from a Chameleon Ultra Ti-Sapphire pulsed laser (Coherent, Santa Clara, CA, USA) and imaged with a Zeiss LSM 510 NLO inverted 2-photon laser scanning microscope (Thornwood, NY, USA) using × 10 0.3 NA, × 20 0.5 NA and × 40 1.2 NA (water-immersion) objectives. Images were captured using a 12-bit cooled CCD camera (Hamamatsu, Bridgewater, NJ, USA).

Western blots

Western blot assays were performed on tumor lysates in RIPA buffer with protease inhibitors (Roche, Indianapolis, IN, USA) and phosphatase inhibitors (Calbiochem, EMD Chemicals Inc, Darmstadt, Germany) as described (Ramsey et al., 2007). Antibodies used were p21CIP (F-8, Santa Cruz, Santa Cruz, CA, USA), p16INK4a (M-156, Santa Cruz), β-actin (C-1, Santa Cruz), cyclin D1 (DCS-6, Cell Signaling, Danvers, MA, USA), cyclin E (M-20, Santa Cruz), Cdk2 (M2, Santa Cruz), Cdk4 (C-22, Santa Cruz), p15INK4b (K18, Santa Cruz), p27KIP (M20, Santa Cruz), phospho-p53 (Ser15, Cell Signaling), p53 (CM5, Novocastra Laboratories Ltd, Newcastle Upon Tyne, UK), p-ERK (9101S, Cell Signaling), ERK (9102, Cell Signaling), p-AKT (9271S, Cell Signaling), AKT (9272, Cell Signaling), Bax (Santa Cruz) and ARF (ab80, Abcam, Cambridge, MA, USA).

Immunohistochemistry

Assistance in sample processing was provided by the University of North Carolina Center for Gastrointestinal Biology and Disease. Ears and tumors were fixed in 10% formalin, paraffin-embedded, sectioned, deparaffinized and stained using hematoxylin and eosin for histological analysis. To confirm melanocytic origin, paraffin sections were treated with citric acid and stained with a rabbit anti-TRP1 (gift from Vincent Hearing) at a dilution of 1:500, or without primary antibody for control samples. Detection was performed using highly sensitive DAKO EnVision polymerized horseradish peroxidase. β-galactosidase staining was previously described (Bosenberg et al., 2006). Samples were analyzed (Zeiss Axioskop 2) and photographed (Zeiss Axiocam) under bright-field microscopy using × 10 0.25 NA, × 20 0.50 NA and × 40 1.30 NA objectives.

Statistical analysis

Tumor-free survival was analyzed using GraphPad Prism software (GraphPad Software) and comparisons were made using the log-rank test. Tumor sizes were compared using an unpaired, two-tailed t-test. Error bars±s.e.m.