Elucidating Distinct Roles for NF1 in Melanomagenesis

Oncogene-induced senescence (OIS) is an irreversible growth arrest that is triggered by a variety of oncogenic signals (1). This form of senescence functions as a protective response to aberrant cell signaling and has been shown to restrict the progression of benign lesions such as melanocytic nevi, lung adenomas, neurofibromas, and prostatic intraepithelial neoplasia (2). Several mechanisms have been proposed to underlie OIS, including excessive DNA damage (3–5), heterochromatin formation (6), negative feedback pathways (7, 8), and chemokine signaling (8–10). Notably, these mechanisms are not mutually exclusive, and it is likely that they cooperate to establish a senescence response in different tissues.

OIS has been shown to be important for restricting melanoma development in response to activating BRAF mutations (11, 12). BRAF is mutated in 50% to 70% of human melanomas (reviewed in ref. 13). The most frequent BRAF mutation (BRAF^V600E) results in a constitutively active kinase. Analysis of human lesions and mouse models has shown that BRAF^V600E mutations drive the development of benign nevi (14–16). However, in the absence of additional mutations melanocytes within these nevi ultimately become senescent and do not progress to malignancy (11, 12, 15). Notably, a subset of genetic alterations found in human melanoma prevents BRAF-induced senescence, underscoring the importance of OIS as a mechanism of tumor suppression (15, 17). Nevertheless, we still do not have a complete mechanistic or genetic understanding of how OIS is bypassed in melanoma or more generally in cancer.

We have previously shown that oncogenic RAF triggers a potent negative feedback signaling network that suppresses RAS and that this feedback loop plays an important role in OIS in vitro (7). Specifically, in response to constitutively activated RAF and mitogen-activated protein/extracellular signal–regulated kinase kinase (MEK) proteins, RAS becomes suppressed because of the upregulation of several direct negative RAS regulatory proteins and the concomitant inactivation of positive RAS regulators (7). Moreover, RAF-induced RAS suppression substantially attenuates phosphoinositide 3-kinase (PI3K)/AKT signaling, which contributes to OIS in this setting (7). These observations raise the intriguing possibility that mutational events that promote RAS activation might play an important part in preventing RAF-induced senescence. If so, then mutations in such genes might be expected to cooperate with BRAF mutations in human cancer.

The NF1 tumor suppressor gene encodes a RAS GTPase- activating protein (RAS GAP) neurofibromin, which negatively regulates RAS by catalyzing the hydrolysis of RAS-GTP to RAS-GDP (18). Accordingly, RAS and downstream effector pathways are aberrantly activated in NF1-deficient tumors (18–20). NF1 is mutated in the familial cancer syndrome neurofibromatosis type I and has more recently been shown to be mutated or suppressed by proteasomal mechanisms in glioblastoma, lung cancer, and neuroblastoma (21–25); however, the full extent of NF1 loss in sporadic tumorigenesis is unknown. Because of its direct effects on RAS and known involvement in melanocyte biology, we investigated a potential role for NF1 in melanomagenesis. Our studies reveal distinct mechanisms by which NF1 mutations cooperate with different BRAF mutations in melanomas. Moreover, we have found that NF1/neurofibromin loss affects the therapeutic response to BRAF inhibitors.

We previously showed that oncogenic RAF alleles potently suppress RAS and subsequent PI3K/AKT signaling and that this suppression is important for OIS in some settings (Fig. 1A; ref. 7). Because NF1 encodes a direct negative regulator of RAS, we reasoned that the effects of this feedback response might be counteracted by ablating NF1 expression. Wild-type and Nf1^−/− mouse embryonic fibroblasts (MEF) were stably infected with a hydroxy-tamoxifen (4-OHT)-inducible, activated RAF construct (26). 4-OHT substantially suppressed RAS-GTP levels in wild-type cells, consistent with previous findings (Fig. 1B and C; ref. 7). However, RAF activation had minimal suppressive effects on RAS activity in Nf1^−/− cells (Fig. 1B and C). Similar effects were also observed in cells in which Nf1 expression was acutely ablated by short hairpin RNA (shRNA) sequences (Supplementary Fig. S1). Shortly thereafter, AKT phosphorylation became substantially reduced in Nf1 wild-type cells at both Ser473 and Thr308 (Fig. 1D and E); however, Nf1 deficiency significantly ameliorated this suppression (Fig. 1D and E). It should be noted that even in the absence of Nf1, RAF partially inhibited AKT phosphorylation, consistent with the known involvement of several redundant negative feedback signals (7). Notably, Nf1 loss caused a baseline activation of the PI3K/AKT pathway, as previously observed (Fig. 1D and E; refs. 19, 20), and therefore minimized the net suppressive effects on AKT. Most importantly, however, RAF activation exerted differential effects on proliferation in wild-type and Nf1-mutant cells. Consistent with previous observations, oncogenic RAF caused a potent and irreversible growth arrest in Nf1 wild-type MEFs (Fig. 1F; refs. 27, 28). However, RAF activation did not suppress the proliferation of Nf1-deficient cells (Fig. 1G), showing that RAS suppression is a critical mediator of this inhibitory response.

To investigate the potential cooperativity of NF1 and RAF mutations in a relevant tumorigenic setting, we generated a mouse model to evaluate the effects of Nf1 loss in the presence of activating Braf mutations. As noted previously, BRAF is mutated in 50% to 70% of human melanomas (13). On the basis of our in vitro findings and the fact that neurofibromin plays a well-established role in melanocytes (29), we evaluated the potential cooperativity of Braf and Nf1 mutations in the context of melanomagenesis.

Mice carrying a conditional inactivating mutation in Nf1 (Nf1^flox/flox mice; ref. 30) were crossed to mice with a conditional activating mutation in Braf (Braf^CA/+ mice; ref. 31). These animals were then crossed to a transgenic mouse strain harboring a tamoxifen-inducible Cre recombinase–estrogen receptor fusion transgene that is under the control of the melanocyte-specific tyrosinase promoter, designated Tyr::CreER^T2 (32). Activation of CreER by tamoxifen in Tyr::CreER^T2; Braf^CA/+; Nf1^flox/flox mice leads to melanocyte-specific conversion of Braf^CA to Braf^V600E and the conversion of the Nf1^flox alleles to Nf1 null alleles. Six genetic cohorts of animals were generated to evaluate the effects of Braf activation in the presence and absence of Nf1.

Mice were treated topically with tamoxifen 2 to 3 months after birth, as previously described (15). After 4 to 5 weeks, significant darkening of the tails, ears, eyelids, perianal regions, and paws was observed in the Tyr::CreER^T2; Braf^CA/+; Nf1^flox/flox mice, as compared with all other genotypes (Fig. 2A and B). Whereas Tyr::CreER^T2; Braf^CA/+ animals exhibited subtle hyperpigmentation, the skin hyperpigmentation in the Tyr::CreER^T2; Braf^CA/+; Nf1^flox/flox mice became dramatically more pronounced over time and was concurrent with visible thickening of the respective tissues (Fig. 2A and B). The histopathologic features of these lesions were consistent with expansion of the dermis, the skin layer in which murine melanocytes reside, and massive melanin deposition (Fig. 2C). An increased number of cells expressing the melanocyte marker S100 were observed, confirming excessive melanocyte hyperproliferation in Tyr::CreER^T2; Braf^CA/+; Nf1^flox/flox mice, as compared with the Braf^CA/+ genotype (Fig. 2C). Importantly, we found that deep dermal lesions derived from control Tyr::CreER^T2; Braf^CA/+ mice stained positive for senescence associated–β-galactosidase (SA-β-gal; Fig. 2D, top), as has been shown in human nevi (12) and in lesions within the Braf^V600E-driven mouse model described by Dhomen and colleagues (15). However, senescence was not observed in Tyr::CreER^T2; Braf^CA/+; Nf1^flox/flox mice (Fig. 2D, bottom). These results are consistent with our cellular studies and indicate that mutations in Nf1 prevent Braf-induced senescence of melanocytes in mice, thereby rescuing the proliferative restriction and triggering excessive proliferation.

PI3K is a well-known effector of RAS (33). We and others have previously shown that NF1 loss triggers activation of the PI3K/AKT/mTOR pathway through RAS (19, 20, 34). Moreover, Nf1 mutations minimized the suppressive effects of Braf mutations on this pathway (Fig. 1D and E). Notably, we found that the PI3K inhibitor GDC-0941 prevented melanocytic hyperplasia in Tyr::CreER^T2; Braf^CA/+; Nf1^flox/flox mice (Fig. 2E), showing that Nf1 loss was mediating its effects in melanocytes, in part, by permitting or enhancing the activation of this pathway.

If OIS does in fact restrict tumor development, then Braf/Nf1–mutant mice would be expected to be more prone to developing melanomas. It has been previously reported that a subset of Braf^V600E mice develop melanomas, presumably owing to the stochastic acquisition of additional genetic alterations (15). Consistent with this observation 22% (6 of 27) of the Braf^V600E mice in our cohort developed melanomas; however, 57% (16 of 28) of the Tyr::CreER^T2; Braf^CA/+; Nf1^flox/flox mice developed melanomas (P = 0.008; χ² test; Fig. 3A–E). One Braf/Nf1-mutant mouse developed 2 melanomas, an event never observed in Braf-mutant animals; however, melanomas from both genotypes grew at similar rates. These observations support the hypothesis that Nf1 loss prevents Braf-induced senescence in vivoand therefore plays a role early in tumor development rather than in progression. However, effects on metastasis could not be evaluated in this model, as animals from both genotypes needed to be euthanized because of primary tumor size. Although pigmented melanocytes were occasionally observed on the exterior, these melanomas were typically hypopigmented (Fig. 3B). All Braf/Nf1 tumors displayed histologic and cytologic features typical of malignant melanomas. Tumors were highly cellular, with most showing a fascicular growth pattern (Fig. 3C). The degree of pleomorphism was variable. The majority of the tumor cells were amelanotic; however, many tumors showed occasional clusters of pigmented cells and pigment-containing macrophages (melanophages; Fig. 3C, right). All tumors involved the dermis as well as subcutaneous soft tissue, and ulceration of the tumor surface was seen in the majority of cases. Tumors expressed both S100 (Fig. 3D) and microphthalmia-associated transcription factor (MITF; Fig. 3E), 2 markers typically used to diagnose human melanomas. Melanomas from both genotypes were further evaluated by immunoblot (Fig. 3F). One tumor that developed in Tyr::CreER^T2; Braf^CA/+; Nf1^flox/flox mice did not efficiently excise Nf1. However, in general, higher levels of phospho-AKT were typically observed in Braf/Nf1-mutant versus Braf-mutant tumors (Fig. 3F and G), consistent with the observation that Nf1 mutations enhance PI3K/AKT activation and contribute to tumorigenesis, in part via this pathway (refs. 19, 20, 34; Fig. 2E).

The mutant BRAF selective inhibitor PLX4032 (vemurafenib; Plexxikon/Roche) promotes the regression of human melanomas that harbor activating BRAF mutations and has been approved for treating human melanomas (35). To investigate the sensitivity of melanomas harboring compound mutations in Braf and Nf1 to BRAF inhibitors and other targeted agents, we first isolated cells from Braf and Braf/Nf1-mutant tumors. As expected, cells from 2 independently derived Braf-mutant melanomas were sensitive to the PLX4032 analog, PLX4720 (Fig. 4A). However, cells from Braf/Nf1-mutant tumors were insensitive to this agent (Fig. 4A; IC₅₀ > 10 μmol/L). Biochemical studies confirmed that increasing concentrations of PLX4720 effectively suppressed phospho–extracellular signal–regulated kinase (ERK) levels in Braf-mutant cells (Fig. 4B). Notably, however, phospho-ERK was not as effectively suppressed in Braf/Nf1-mutant cells (Fig. 4B).

Next, we evaluated sensitivity to these agents in vivo. Whereas allografts from several independent Braf/Nf1-mutant melanomas were readily established, cells from Braf-mutant mouse melanomas never successfully grew as allografts. As such, we focused on using Braf/Nf1-mutant allografts to determine whether we could identify a more effective therapy for these genetically distinct tumors. Two weeks after injection, when tumors were growing in log phase, mice were randomly divided into different treatment groups. Similar to our in vitro analysis, Braf/Nf1-mutant melanomas were relatively insensitive to PLX4720 (Fig. 4C, red) and phospho-ERK was not efficiently inhibited in tumors in vivo (Fig. 4D). This finding is consistent with the observation that induction of tumor regression by PLX4032 requires almost complete suppression of ERK signaling (36). In contrast, these tumors were more sensitive to the MEK inhibitor PD0325901 (Fig. 4C, blue), which effectively suppressed phospho-ERK in vivo (Fig. 4D). Together with our in vitro studies, this insensitivity suggests that neurofibromin loss may enhance ERK activity via a BRAF-independent mechanism, which will be further discussed later.

Because our in vivo experiments showed that melanocyte hyperproliferation in Tyr::CreER^T2; Braf^CA/+; Nf1^flox/flox mice could be rescued by the PI3K inhibitor GDC-0941, we evaluated the effects of GDC-0941 on tumor growth and found that it had a slight growth-suppressive effect on these melanomas (Fig. 4C, green). However, given that mTOR has been shown to be an important effector in NF1-mutant tumors, we also assessed the mTOR inhibitor rapamycin (20, 34, 37). Rapamycin had a more pronounced effect than GDC-0941, but more importantly it synergized with PD0325901 to promote tumor regression (Fig. 4C, purple). In contrast, rapamycin did not promote tumor regression when combined with PLX4720 (Fig. 4C, violet). Taken together, these results suggest that Nf1 mutations can desensitize Braf-mutant melanomas to PLX4720. On the basis of the biochemical function of neurofibromin and the preclinical data presented here, therapies aimed at targeting both MEK and mTOR may represent an alternative therapeutic approach for BRAF/NF1-mutant tumors.

Given these compelling mouse phenotypes, we next investigated whether NF1 is lost or mutated in human melanomas. Like p53, neurofibromin can be inactivated by both genetic and proteasomal mechanisms (23). We first evaluated neurofibromin expression in a panel of melanoma cell lines. Four of 11 cell lines exhibited little or no neurofibromin expression (Fig. 5A). Sequence analysis confirmed that 3 of these cell lines harbored loss-of-function mutations in the NF1 gene (Supplementary Table S1). It should be noted that 2 of the NF1 alterations (p.K1290K in A375 cells and p.K2307K in WM3670 cells) abolish splice donor sites (c.3870G>A and c.6921G>A, respectively), result in defective splicing, and disrupt the NF1 transcript. Sequence analysis of the cDNA of NF1 enabled the detection of these aberrant splicing events; however, both NF1 mutations may have been categorized as “non-deleterious” silent alterations using exome sequencing approaches. We then mined publicly available databases and identified numerous additional NF1 mutations in human cell lines (Supplementary Table S1). Analysis of primary melanomas also confirmed the presence of somatic NF1 mutations (Supplementary Table S2). In many, but not in all, cases BRAF mutations were also present. These findings suggest that NF1 mutations can cooperate with activating BRAF mutations, but they may also play a broader role in melanomagenesis. Interestingly, however, although we expected to find coincidental mutations with BRAF^V600E, we also observed NF1 mutations in cells that harbored inactivating BRAF mutations (Supplementary Table S1), a point that will be discussed later.

To complement the mouse modeling studies and confirm a functional role for NF1 inactivation in human melanomas, we reconstituted neurofibromin in A375 cells, which harbor compound NF1 and BRAF^V600E mutations. Full-length neurofibromin expression potently suppressed the growth of xenografts in mice (P = 3.246E-004; Mann–Whitney U test), consistent with the notion that NF1 inactivation plays a causal role in tumor development (Fig. 5B). As noted earlier, the NF1 tumor suppressor is frequently inactivated by proteasomal mechanisms in human cancer (23). As such, mutational analysis may underestimate the frequency of NF1/neurofibromin loss in tumors. Therefore, we conducted immunohistochemical analysis on human melanoma tumor arrays. No visible neurofibromin expression was observed in 15% (6 of 39) of melanomas and 18% (6 of 34) of metastatic melanomas (Fig. 5C and Supplementary Table S3). It should be noted that only a complete absence of staining was scored as negative in this analysis. Therefore, excessive but incomplete neurofibromin destruction was not considered, but could still play a role in tumor development.

Preclinical studies in mouse tumors suggested that Nf1 loss can desensitize Braf-mutant melanomas to BRAF inhibitors. To evaluate this possibility in human tumors, we genetically ablated neurofibromin expression with lentiviral shRNA sequences in human melanoma cell lines and found that NF1 suppression decreased the sensitivity of WM3526 cells to PLX4720 by 11-fold (Fig. 6A and Supplementary Fig. S2). As noted previously, A375 cells harbor a loss-of-function NF1 mutation and express very low levels of neurofibromin. Because these cells remain somewhat sensitive to BRAF inhibitors, we further reduced neurofibromin expression with shRNA sequences and found that NF1 suppression also decreased the sensitivity of these cells to BRAF inhibitors (Fig. 6B and Supplementary Fig. S2). It should be noted that although the acute ablation of NF1 in established melanoma cell lines desensitized these cells to PLX4720, tumors that naturally developed in the absence of Nf1 were much more resistant to this agent. We hypothesize that this difference may reflect inherent differences in preexisting signaling networks and/or cooperating mutations in these established tumor cells, which did not evolve in the absence of NF1. Nevertheless, consistent with observations in mouse tumors, PLX4720 was much less effective at suppressing phospho-ERK in cells in which NF1 was ablated, indicating that NF1 loss promotes BRAF-independent ERK signaling in these tumor cells as well (Fig. 6C).

Finally, we were able to obtain frozen tissue from 5 sets of pre- and posttreatment tumor biopsy specimens from patients treated with a BRAF inhibitor (vemurafenib) or combined BRAF/MEK inhibitors (dabrafenib + trametinib). Notably, 2 of 5 tumors expressed very little or no neurofibromin before treatment, consistent with the observation that NF1/neurofibromin is lost or suppressed in human melanomas at a relatively frequent rate (Fig. 6D). However, in 2 of the remaining 3 tumors in which neurofibromin was robustly expressed before treatment, neurofibromin was no longer expressed in tumors following treatment (Fig. 6D). Taken together with preclinical studies in the mouse model and genetic studies in human melanoma cell lines, these observations further support the hypothesis that NF1/neurofibromin suppression may play an important role in mediating resistance to BRAF inhibitors. Notably, in glioblastoma, neurofibromin seems to be more frequently lost by proteasomal mechanisms (23). Therefore, future studies aimed at assessing NF1/neurofibromin loss in response to therapies will likely require analysis of both protein expression and genetic alterations.

Our central hypothesis is that loss of NF1 alleviates the suppression of RAS imposed by activated BRAF. To identify the RAS isoforms that are critically regulated by neurofibromin in melanomas, we conducted both gain- and loss-of-function studies. In melanoma cells that retain neurofibromin expression, RNA interference (RNAi)-mediated suppression of neurofibromin resulted in the activation of H- and KRAS but not NRAS (Fig. 7A). Conversely, reconstituting melanoma cell lines with an active neurofibromin fragment suppressed H- and KRAS activity but not NRAS activity (Fig. 7B). Finally, shRNA-mediated suppression of either H- or KRAS suppressed the ability of NF1-deficient melanoma cells harboring an active BRAF mutation to proliferate and form colonies in soft agar, whereas NRAS-specific shRNA sequences had no effect (Fig. 7C–E). These results suggest that neurofibromin loss/suppression activates H- and KRAS in melanomas and that both of these isoforms are critical for the tumorigenicity of these cancer cells.

As shown in Supplementary Table S1, NF1 mutations were also detected in cells that harbored inactivating BRAF mutations. Although the kinase activity of these BRAF proteins is compromised, they have been proposed to function as scaffolds that translocate CRAF to the membrane and promote CRAF activation downstream of KRAS (38). Notably, the oncogenic effects of these mutants are significantly enhanced in the presence of KRAS^G12D (38). Similarly, an NF1 mutation might be expected to function as an alternative mechanism of activating KRAS, and also potentiate the effects of these BRAF mutants. To evaluate the contribution of RAS isoforms in human melanomas, we examined WM3629 cells. Interestingly, in addition to harboring the kinase-dead BRAF^D594G mutation and an NF1 deletion, this cell line also possesses an activating NRAS^G12D mutation (39, 40). The WM3670 line similarly harbors an inactivating BRAF mutation, as well as NF1 and NRAS^G12D mutations (39, 40). Consistent with the presence of these mutations and our previous observations, NRAS, KRAS, and HRAS are all activated in WM3629 cells. Moreover, shRNA-mediated ablation of all 3 RAS isoforms— H-, K-, and NRAS—suppressed the ability of WM3629 cells to proliferate and form colonies in soft agar (Fig. 7F–H). These results further show that neurofibromin critically regulates H- and KRAS in melanomas and suggest that in the presence of NRAS mutations all 3 RAS isoforms cooperate with inactivating BRAF mutations to promote tumorigenesis.

This study establishes several distinct roles for NF1 in melanomagenesis. First, our data suggest that in the presence of activating Braf mutations, Nf1 loss prevents Braf-induced senescence of melanocytes. In mice, this results in excessive melanocyte hyperproliferation and ultimately enhances melanoma development. Importantly, NF1 mutations cooccur with activating BRAF mutations in human melanomas, and neurofibromin reconstitution potently suppresses the growth of human melanoma cells as xenografts, further supporting a causal role for NF1 loss in melanomagenesis. However, we have found that NF1 mutations can also cooperate with inactivating BRAF mutations in melanomas. Although these BRAF mutations are less common, in this setting we found that NF1 mutations cooccur with NRAS mutations and that all 3 RAS isoforms are required for the tumorigenic properties of these cells. Thus, NF1 mutations can contribute to melanoma development in at least 2 genetic settings via distinct mechanisms.

We also found that in the context of activating BRAF alleles NF1 mutations contribute to tumorigenesis, in part by promoting activation of the PI3K/AKT/mTOR pathway. Notably, the PTEN tumor suppressor is commonly mutated or lost in human melanomas (41). Mouse modeling studies and human tumor analysis suggest that BRAF and PTEN mutations cooperate in melanomagenesis (14). More recently, PTEN loss has been shown to prevent BRAF-induced senescence in mice and in human melanocytes, further supporting the notion that coactivation of these pathways is important for preventing OIS (17). Our data show that NF1 loss is another important mechanism by which the PI3K pathway can become activated in melanomas. It should be noted, however, that NF1 and PTEN mutations do not seem to be mutually exclusive in melanomas. Moreover, we have found that NF1 loss also results in the activation of multiple RAS isoforms and potentiates ERK activation independent from activating BRAF mutations. Taken together, these observations suggest that NF1 loss contributes to melanomagenesis by enhancing the activation of both PI3K and ERK signaling, which may be important in the context of selecting effective therapies (model presented in Fig. 7I).

Several mechanisms have been reported to mediate the resistance of BRAF-mutant melanomas to BRAF inhibitors (42–46). Additional mechanisms are likely to be discovered, and currently one prevalent mechanism of resistance has not emerged. We have shown that Nf1 mutations confer resistance to PLX4720 in Braf-mutant mouse melanomas; however, these tumors are sensitive to combined MEK/mTOR inhibitors. Importantly, RNAi-mediated NF1 suppression also decreases the sensitivity of human melanoma cell lines to BRAF inhibitors, and, most notably, NF1/neurofibromin is lost in a subset of relapsing and residual tumors from patients exposed to BRAF inhibitors. The observation that NF1/neurofibromin is also mutated or lost in some naïve primary tumors is consistent with the hypothesis that NF1 inactivation may contribute to both de novo and acquired resistance. However, although NF1 mutations can be detected in human melanomas, the true frequency of NF1 loss may be difficult to assess, because like PTEN and P53, the NF1 protein is frequently inactivated by proteasomal destruction (23). Immunohistochemical analysis of tumor microarrays indicates that neurofibromin expression is completely absent in 15% to 18% of melanomas; however, a more quantitative evaluation of protein levels may be required to accurately evaluate its expression or suppression before and after drug treatment. To date, much of the resistance to BRAF inhibitors seems to be driven by events that activate ERK through mechanisms that circumvent or decrease dependency on BRAF. However, aberrant activation of the PI3K/AKT pathway has also been implicated in resistance to BRAF inhibitors (46–48). In this respect, NF1 (loss) is uniquely poised, as it activates the ERK pathway through its effects on KRAS and HRAS and at the same time enhances PI3K/AKT/mTOR signaling.

Wild-type and Nf1 null MEFs were generated as described (49). Cells were stably infected with a pBabe retroviral vector containing the 4-OHT–inducible estrogen receptor RAF-1 (ΔRAF:ER) construct (28). This construct is an estrogen receptor–RAF-1 fusion molecule, and contains the ligand-binding domain of the estrogen receptor fused to the activated RAF kinase domain of RAF-1. Cells were split to an equal density 16 hours before harvesting in Dulbecco modified Eagle medium (DMEM) containing 1% fetal calf serum (FCS) and different concentrations of 4-OHT (0, 5, 50, or 500 nmol/L). For proliferation studies, cells were seeded at a density of 25 × 10⁴ in a 12-well format in DMEM containing 10% FCS and different concentrations of 4-OHT (0, 5, and 50 nmol/L). Cells were counted the day after plating and every other day after, until uninduced cells reached confluency. Media containing fresh 4-OHT were changed every 48 hours. The melanoma cell lines were obtained from the American Type Culture Collection and the Garraway laboratory (46). No additional authentication was done by the authors. All melanoma cell lines were grown in RPMI medium supplemented with 10% FCS, penicillin, and streptomycin. For dose–response curves, cells were seeded in triplicate in a 6-well format. Cells were counted the day after plating, when PLX4720 (Calbiochem) was added, as well as after 3 days, when dimethyl sulfoxide (DMSO)–treated cells reached confluency. To allow comparison of the IC₅₀ between cell lines, growth inhibition was calculated at 5 population doublings for every cell line. Neurofibromin activity was reconstituted using the previously described NF1-GAP–related domain (GRD) construct (20) or an expression vector containing full-length NF1 cDNA. Neurofibromin expression was abolished using shRNAs specific to NF1 (20).

RAS-GTP levels were detected using a RAS activation assay, following the manufacturer's instructions (EMD Millipore). Tumor lysates were homogenized and extracted with boiling 1% SDS buffer. Resulting protein lysates were quantified and run according to validated immunoblot procedures with the following antibodies: phospho-AKT (Ser473, #4060; Cell Signaling Technology), phospho-AKT (Thr308, #9275; Cell Signaling Technology), AKT (#9272; Cell Signaling Technology), phospho-ERK (Thr202/Thr204, #4370; Cell Signaling Technology), ERK (#9102; Cell Signaling Technology), phospho-S6 (Ser235/236, #2211; Cell Signaling Technology), S6 (#2317; Cell Signaling Technology), NF1 (#A300-140A, Bethyl Laboratories), p120 (#G12920; Trans Labs), RAS (#05-516; Upstate), KRAS (#sc-30; Santa Cruz Biotechnology), HRAS (#sc-520; Santa Cruz Biotechnology), NRAS (#sc-519; Santa Cruz Biotechnology), and MITF (#M3621; Dako). Immunoblots were quantified with ImageJ software.

Animal procedures were approved by the Center for Animal and Comparative Medicine in Harvard Medical School (Boston, MA) in accordance with the NIH Guild for the Care and Use of Laboratory Animals and the Animal Welfare Act. A breeding scheme was set up in which mice carrying a conditional Nf1 allele (Nf1^flox/flox mice; ref. 30) were crossed to mice with a conditional activating mutation in Braf (Braf^CA/+ mice; ref. 31) and a transgenic mouse strain harboring a tamoxifen-inducible Cre recombinase–estrogen receptor fusion transgene that is under the control of the melanocyte-specific tyrosinase promoter, designated Tyr::CreER^T2 (32). Genotyping was conducted by PCR, as described for every model (30–32). To activate Tyr::CreER^T2, mice were treated topically with freshly prepared tamoxifen (T5648; Sigma) 2 to 3 months after birth. Tamoxifen (20 mg/mL in 100% ethanol) was applied to a small section of skin on the shaven backs, and the treatment schedule consisted of 4 treatments over 7 days. For all genotypes, pigmentation levels were quantified weekly on a scale from 0 (no pigmentation) to 8 (high pigmentation). To study the contribution of the PI3K/AKT pathway to the observed hyperpigmentation phenotype in the Tyr::CreER^T2; Braf^CA/+; Nf1^flox/flox mice, animals were treated daily with the PI3K inhibitor GDC-0941 (150 mg/kg, oral gavage) for 8 weeks after Tyr::CreER^T2 induction.

For cancer cell xenograft experiments, nude mice were inoculated subcutaneously with 3 × 10⁶ human or mouse melanoma cells. Tumor volumes were calculated by measuring length and width of the lesions and with the formula [(length) × (width)² × 0.52]. Murine Braf/Nf1-mutant melanoma cells formed rapidly growing tumors. Two weeks after injection, when tumors were growing in log phase, mice were randomly divided into different treatment groups that were administered the PI3K inhibitor GDC-0941 (150 mg/kg, oral gavage), the mTOR inhibitor rapamycin [5 mg/kg, intraperitoneal (i.p.) injection], the MEK inhibitor PD0325901 (10 mg/kg, oral gavage), the BRAF inhibitor PLX4720 (2.5 mg/kg, i.p.), or the combination of both PD0325901 or PLX4720 and rapamycin. Mice were treated daily for 7 days.

Tissues were fixed in formalin, embedded in paraffin, and stained with hematoxylin and eosin (H&E) when indicated. Standard immunohistochemistry and immunofluorescence protocols were followed for S100 (Z0311, 1:400; Dako) staining. An alkaline phosphatase staining method was used for NF1 (ab30325, 1:300; Abcam) immunohistochemistry. Antigen unmasking was conducted by pressure cooking in citrate buffer (pH 6). A human melanoma tissue microarray (ME804a; US Biomax Inc.) was used for evaluation of NF1 expression in malignant and metastatic melanomas. SA-β-gal staining was conducted on fresh mouse tissue. Small pieces were fixed for 30 minutes in 3% paraformaldehyde, incubated in SA-β-gal staining solution as described (50), and then embedded and sectioned.

The entire NF1 coding region of 11 human melanoma cell lines (Fig. 5A) was amplified in 5 overlapping reverse-transcription PCR (RT-PCR) fragments and used as the template for direct sequencing, essentially as described (51). Copy number analysis by multiplex ligation-dependent probe amplification (MLPA) was conducted as described (52). The nomenclature of the mutations is based on NF1 mRNA sequence NM_001042492.2, with 1 being the first nucleotide of the ATG start codon. Next, publicly available resources providing information on somatic mutations implicated in human melanoma cell lines (53, 54) and primary tumors (55–57) were mined.

Patients with metastatic melanoma containing the BRAF^V600E mutation (confirmed by genotyping) were enrolled in clinical trials for treatment with a BRAF inhibitor (vemurafenib) or combined BRAF/MEK inhibitors (dabrafenib + trametinib), and their consent was obtained for tissue acquisition per Institutional Review Board (IRB)–approved protocol. Tumor biopsies were conducted pretreatment (day 0), at 10 to 14 days on treatment, and/or at time of progression if applicable. Formalin-fixed tissue was analyzed to confirm that viable tumor was present via H&E staining. Additional tissue was snap frozen and stored in liquid nitrogen.

M. McMahon has commercial research grants from Novartis and Plexxikon. R. Marais is Secretary General of the European Association for Cancer Research, has received honoraria from service on the speakers' bureau for Roche, and is a consultant/advisory board member for Novartis and Servier Research Institute. No potential conflicts of interest were disclosed by the other authors.

Conception and design: O. Maertens, B. Johnson, K. Cichowski

Development of methodology: O. Maertens, B. Johnson, P. Hollstein, R. Marais

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): O. Maertens, B. Johnson, P. Hollstein, D.T. Frederick, Z.A. Cooper, L. Messiaen, M. McMahon, K. Flaherty, J.A. Wargo, R. Marais, K. Cichowski

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): O. Maertens, B. Johnson, P. Hollstein, L. Messiaen, R.T. Bronson, M. McMahon, K. Flaherty, J.A. Wargo, K. Cichowski

Writing, review, and/or revision of the manuscript: O. Maertens, D.T. Frederick, Z.A. Cooper, M. McMahon, K. Flaherty, J.A. Wargo, K. Cichowski

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): O. Maertens

Study supervision: K. Cichowski

Conducted immunohistochemical analysis: S. Granter

The authors thank M. McMahon and K. Haigis for providing the Braf^CA/+ mice, L. Chin for providing the Tyr::CreER^T2 mice, and L. Garraway for cell lines.

K. Cichowski and the majority of this work was supported by the National Cancer Institute (R01 CA111754) and the Ludwig Center at Dana-Farber/Harvard Cancer Center. O. Maertens is a postdoctoral fellow of the Research Foundation Flanders (FWO).