Skin Tumors Induced by Sorafenib; Paradoxic RAS–RAF Pathway Activation and Oncogenic Mutations of HRAS, TP53, and TGFBR1

This article is featured in Highlights of This Issue, p. 1

BRAF is the most frequently mutated protein kinase in human cancer (1) and is the target of several anticancer drugs. The potency and the specificity of BRAF inhibitors available on the market or under clinical development are variable. Sorafenib (Nexavar, Bayer/Onyx) is a pan-RAF inhibitor which also blocks VEGF receptors (VEGFR)-2, -3, platelet-derived growth factor receptor-β, fms-like tyrosine kinase 3 (FLT3), and KIT. Conversely, vemurafenib (Plexxikon/Roche) or GSK2118436 (GlaxoSmithKline) are highly selective and very potent BRAF inhibitors. These latter drugs have a low IC₅₀ for mutated oncogenic BRAF and are less active on wild-type BRAF (wtBRAF) and other RAF kinases. The clinical activities of these inhibitors also differ from one another. Selective BRAF inhibitors like vemurafenib and GSK2118436 are efficient against tumors harboring BRAF mutations and dependent on the RAF/MEK/ERK (mitogen-activated protein (MAP)/extracellular signal-regulated (ERK; MEK) kinase) pathway. Melanoma with V600E BRAF mutation is a case in point (2). Sorafenib activity is not linked to the presence of a BRAF mutation in the tumor and is more likely due to its antiangiogenic effect. Sorafenib is efficient and has been approved for the treatment of metastatic renal carcinoma (3, 4) and advanced hepatocellular carcinoma (4) independently of the mutational status of BRAF.

In spite of their variability in terms of BRAF selectivity and clinical activity, all 3 RAF inhibitors are associated with one and the same intriguing cutaneous side effect which is the emergence of borderline squamous cell neoplasms. Indeed, we and others have reported single or multiple cutaneous keratoacanthomas (KA) and squamous cell carcinomas (SCC) during treatment with sorafenib (5–9). These paradoxic keratinocyte proliferations arise in 6% to 7% of patients (5,10). The induction of KA and SCC is even more frequent with vemurafenib and GSK2118436 and has been described in 15% to 25% of the patients (2, 10).

We undertook a translational research in the aim of understanding the mechanisms of these skin cancer inductions. We first conducted a detailed clinical, pathologic, and molecular study of the skin neoplasms arising in patients treated with sorafenib. We then looked at the biological effects of sorafenib in vivo on normal skin samples from patients treated with sorafenib or a placebo. In parallel, we analyzed the effect of sorafenib in vitro on cultured keratinocytes. We found that sorafenib exposure was associated with increased mitogen-activated protein kinase (MAPK) pathway activation and keratinocyte proliferation ex vivo and in vivo. Sorafenib induced dose-dependent BRAF/CRAF dimerization associated with strong activation of CRAF in cultured keratinocytes. Finally, oncogenic mutations in HRAS, TGFBR1, and TP53 were detected in several benign or malignant skin tumors. Altogether, our results suggest that sorafenib induces a continuous spectrum of keratinocyte changes and proliferation via the paradoxic activation of the MAPK pathway in keratinocytes associated with BRAF/CRAF heterodimerization and subsequent CRAF activation. Additional events such as a HRAS mutation or EGFR activation giving rise to MAPK pathway coactivation might be required for full transformation of keratinocytes.

Thirty-one skin lesions diagnosed as follicular cystic skin lesions, perforating folliculitis (PF), KA, or SCC originating from 17 patients treated with sorafenib from 2005 to 2010 were studied.

For each patient, the tumor type, duration of sorafenib therapy before the emergence of the skin tumor, and the clinical description of the skin lesions are reported. Four-micrometer sections from formalin-fixed, paraffin-embedded biopsy samples of skin lesions were stained with hematoxylin and eosin and independently read by 2 pathologists.

Normal skin samples from patients included in the TARGET study, evaluating sorafenib versus a placebo in patients with renal cell cancer were also studied (3). Normal skin biopsy samples were collected from the forearms of patients who received sorafenib or a placebo for 6 weeks.

Each skin sample was cut into 4-μm sections, formalin-fixed, paraffin-embedded, and stained with hematoxylin and eosin. A section from each case was analyzed immunochemically for Ki-67 (monoclonal MIB-1; Dako) and phospho-ERK protein (monoclonal 20G11; Cell Signaling Technology). Slides were deparaffinized and treated with an alcohol gradient to absolute ethanol. Endogenous peroxidase activity was blocked with the Peroxidase-Blocking Solution (Dako; S2001). Antigen retrieval was achieved with Target Retrieval Solution (Dako). Primary antibody was applied at 1:100 dilution and incubated at 4°C for 1 hour. Anti-mouse immunoglobulin G with the Dako EnVision System (horseradish peroxidase) was employed according to the manufacturer's instructions.

Genomic DNA was extracted from 10- to 20-μm thick paraffin-embedded unstained slides of each skin lesion with the DNeasy Blood and Tissue Kit (QIAGEN) according to the manufacturer's instructions. We analyzed full coding sequences of 70 exons including oncogenic mutational hot spots corresponding to AKT1 (NM_005163.2) exon 4, BRAF (NM_004333.4) exons 11 and 15, EGFR (NM_005228.3) exons 18 to 21, HRAS (NM_005343.2) exons 2 and 3, KRAS (NM_033360.2) exons 2 and 3, NRAS (NM_002524.3) exon 2 and 3, PIK3CA (NM_006218.2) exons 10 and 21, PTEN (NM_000314.4) exons 1 to 9, TP53 (NM_000546.4) exons 2 and 11, and TGFBR1 (NM_004612.2) exons 2, 4, 6, and 7.

Sequences were analyzed by Sanger direct sequencing conducted after PCR amplification of target exons. Sequencing reactions were carried out by the BigDye Terminator Cycle Sequencing Kit software according to the manufacturer's recommendations (Applied Biosystems). The primer sequences are available upon request. Sequencing reactions were analyzed on a 48-capillary 3730 DNA Analyzer. Sequence reading and alignment were carried out with SeqScape software (Applied Biosystems). All detected mutations were confirmed following at least 1 independent PCR reaction.

HaCat keratinocytes were cultured in Dulbecco's Modified Eagle's Medium (Invitrogen) containing 10% fetal calf serum (Perbio) and antibiotics (penicillin/streptomycin; Invitrogen). Cellular proliferation was evaluated by the CellTiter Assay (Promega). Sorafenib and PD153035 used for cell treatment was purchased from Alexis Biochemicals and dissolved in dimethyl sulfoxide (DMSO).

Cells were lysed in 50 mmol/L Tris (pH 7.5), 150 mmol/L NaCl, 0,5% NP40, 5 mmol/L NaF, 2 mmol/L Na₃VO₄, 10 μg/mL leupeptin, and 10 μg/mL aprotinin. For immunoprecipitation, samples were incubated with 1 μg BRAF (clones F-7 or H145; Santa Cruz) or CRAF (clone C12; Santa Cruz or clone 53/c-Raf-1; BD Transduction Laboratories) antibodies for 1 hour and immunoprecipitated by protein G-Sepharose (GE Healthcare) for 1 hour at 4°C. The immune complexes were washed 3 times with lysis buffer, proteins were separated by SDS-PAGE, and Western blot analysis was conducted according to standard protocols with the following antibodies: BRAF (F7 or H145); CRAF (C12 or 53/c-Raf-1); phospho-ERK (MAPK-YT; Sigma) and ERK (Upstate/Millipore); MCL1 (Abcam), EGFR, and phospho-EGFR (Cell Signaling); and Actin (Abcam).

CRAF was immunoprecipitated from cellular extract and its activity toward MEK measured in vitro. Briefly, immunoprecipitated CRAF was incubated with 5 μg of purified GST-MEK in 30 mmol/L Tris (pH 7.5), 100 μmol/L EDTA, 10 mmol/L MgCl₂, 0.1% Triton X-100, 5 mmol/L NaF, 1 mmol/L ATP, 0.3% β-mercaptoethanol for 30 minutes at 30°C. The reaction was stopped by adding Laemmli buffer and analyzed by SDS-PAGE with the phospho-MEK (Cell Signaling Technology) and CRAF antibody. Phosphorylated MEK and CRAF were visualized and quantified by fluorescence on an Odyssey imaging system.

HaCat cells were grown on glass cover slips, treated with DMSO or sorafenib, fixed with methanol, and permeabilized with acetone. The proximity ligation assay (PLA) was conducted according to the manufacturer's instructions (Olink) with B-Raf (F7) and C-Raf (C12) antibodies.

HaCat cells were transfected with an siRNA control or targeting BRAF or CRAF as previously described (11).

Simple univariate descriptive statistics were used to compare the population, using the t test with a 5% significance level.

The gender, age, cancer type, daily dose of sorafenib, time interval between the start of sorafenib therapy and the first skin symptom, the number and types of lesions as well as the skin sites involved are detailed in Table 1.

Various clinical aspects were observed but all patients developed small cystic follicular lesions that most frequently affected the face and also sometimes other parts of the skin such as the buttocks, axilla, or inguinal areas (Table 1). They occurred 3 to 5 months after the initiation of sorafenib. Cysts on the face were dominantly noninflammatory millimetric small and white lesions. Inflammatory cystic lesions exhibited erythematous hyperkeratotic papules, with a central, cone-shaped plug.

Lesions clinically suggestive of perforating folliculitis, meaning a more inflammatory infiltrate with a keratotic appearance occurred mostly on the buttocks and the upper part of the limbs (patients # 7, 11, 12, 13, and 14).

Patients 1 to 11 developed at least one lesion characteristic of KA or of KA-like SCC. Clinically, they were fast-growing firm, round nodules with a central crateriform keratotic zone. Three of these patients had multiple skin lesions (patients # 4, 6, and 7) occurring 1 to 9 months after the initiation of sorafenib.

A total of 22 lesions were excised and the pathologic examination with hematoxylin and eosin staining revealed a spectrum of lesions (Fig. 1): (i) noninflammatory cystic hair follicles, (ii) cystic folliculitis and perforating folliculitis with rupture of the hair follicle wall and perifollicular infiltrates, (iii) KA defined as a crateriform epidermal neoplasm limited by sharp beaks with no cytologic atypia nor invasion, (iv) crateriform KA-like SCC defined as crateriform epidermal neoplasms with irregular budding borders, cytologic atypia, and invasive nests (Fig. 1). Lymphatic metastatic embols were seen in one SCC (patient #8).

Patient #7 developed the entire spectrum of the skin lesions that were observed: noninflammatory facial cysts, cystic folliculitis, perforating folliculitis, multiple KA, and SCC that emerged 3 to 5 months, respectively, after the initiation of sorafenib. Sorafenib was stopped due to cutaneous intolerance at the request of the patient after 24 months even though it was still effective against the progression of the renal cell carcinoma. All of her skin lesions, including skin tumors disappeared over a 4-month period following sorafenib interruption.

Eight oncogenic mutations were found in benign and malignant tumors. Three HRAS mutations on codon 12 or 13 were found in 3 lesions: 1 KA, 1 KA-like SCC, and 1 case of benign perforating folliculitis (Table 1). TP53 was also found to be mutated in 3 lesions: 1 KA, 1 in-situ SCC, and 1(?) case of perforating folliculitis (exons 8 and 5). Moreover 2 missense mutations of TGFBR1 were observed in 1 KA (c.1120G>A; p.Gly374Arg) and in 1 in-situ SCC (c.248C>T; Pro83Leu; Table 1). Those TGBR1 mutations are somatic mutations because they were not present in other samples for the same patient. No additional mutation of the screened exons of BRAF, NRAS, KRAS, AKT1, PI3KCA, PTEN, nor EGFR genes was found in any of the other malignant or benign lesions. Six of these mutations are known to be typically UV induced (Table 1; refs. 13, 14).

The results of the immunostaining of normal skin samples with Ki-67 and pERK for patients treated with sorafenib or a placebo are shown in Fig. 2.

Ki-67 staining was found in the nucleus of a few scattered basal cells in nontreated patients whereas it was found in the nucleus of 1 to 2 whole layers or multiple heaps of basal cells in patients treated with sorafenib (Fig. 2A). Ki-67–stained cells were counted by 2 independent pathologists in 3 representative fields of each specimen (15). There was a significant difference between the numbers of Ki-67–positive keratinocytes in the 2 populations of patients with more proliferating cells in the sorafenib-treated group than in the placebo-treated patients (Rank test, P = 0.03; Fig. 2B).

We also found a tendency for increased phosphorylation of ERK in normal skin of sorafenib-treated patients compared with normal skin of placebo-treated patients. A strong positive p-ERK staining was found in 7 out of 11 normal skin samples from sorafenib-treated patients (64%), whereas only 2 out of 8 samples from placebo-treated patients (25%) showed positive, albeit weak and focal p-ERK staining (Fig. 2C and D; P = 0.1, Fisher exact test). The lack of statistical significance might be due here to the small number of specimen.

We first used sequencing to verify that HaCat cells did not harbor an oncogenic mutation in any of the RAS genes (HRAS, KRAS, and NRAS; data not shown).

HaCat cells were treated with increasing amounts of sorafenib, and BRAF–CRAF heterodimerization was evaluated by coimmunoprecipitation experiments. We did not detect any heterodimers in DMSO-treated cells, whereas sorafenib induced BRAF–CRAF heterodimers in a dose-dependent manner in keratinocytes (Fig. 3A). As BRAF–CRAF heterodimerization was previously shown to activate CRAF, we measured CRAF kinase activity in keratinocytes treated with increasing doses of sorafenib. Of note, sorafenib is a reversible inhibitor and therefore was no longer present during the in vitro kinase assay. Treatment of HaCat cells with sorafenib stimulated CRAF activity in a dose-dependent manner (Fig. 3B), which paralleled the dimerization of BRAF with CRAF (Fig. 3A), confirming that dimerization activates CRAF in keratinocytes. To analyze the biological effect of sorafenib-induced CRAF activation on the MAPK pathway, we measured ERK phosphorylation in response to sorafenib in a time course experiment. Sorafenib transiently activated ERK at 1, 3, and 10 μmol/L, attaining a maximum after 15 minutes (Fig. 4A). A more pronounced effect was observed with the lower doses of sorafenib. To analyze the biological consequence of ERK activation, we measured the proliferation rate of HaCat cells treated with increasing doses of sorafenib. Cellular proliferation was increased with the low dose (1 μmol/L) of sorafenib. Proliferation was inhibited with higher doses of sorafenib (3 and 10 μmol/L; Fig. 4B).

To determine whether this dual effect on cell proliferation could be explained by the action of sorafenib on the antiapoptotic molecule MCL1, MCL1 expression was studied as a function of sorafenib concentration. We found that MCL1 expression decreased with higher sorafenib concentrations, (3 and 10 μmol/L), whereas 1 μmol/L of sorafenib did not influence MCL1 expression (Fig. 4C). To test whether sorafenib-induced heterodimerization could be detected in situ, we used a PLA. In this assay, a pair of oligonucleotide-labeled secondary antibodies (PLA probes) generates a signal only when they have bound in close proximity, thus allowing the detection of protein–protein interaction in situ. HaCat cells were treated or not with sorafenib and fixed. BRAF–CRAF heterodimerization was visualized by PLA with a combination of a CRAF and a BRAF antibody. Figure 5 shows that sorafenib-induced BRAF–CRAF heterodimerization can be directly detected in keratinocytes.

RNA interference confirmed the specificity of the PLA signal as inhibition of BRAF or CRAF expression with siRNA strongly diminished the PLA signal induced by sorafenib (Supplementary Fig. S1).

Targeted anticancer agents are associated with frequent and diverse cutaneous side effects. The description and study of these skin manifestations is of critical importance. Indeed, they frequently have an adverse impact on the patient's quality of life and understanding their underlying mechanisms is certainly the best way to treat these skin manifestations in addition to offering new perspectives on skin biology (5, 6, 12–14).

Our earlier observations of skin manifestations induced by targeted agents had lead us to formulate the hypothesis that sorafenib-induced KA and skin SCC were due to its effect on RAF kinases. Indeed, these side effects were not seen after blockade of VEGFR, KIT, FLT3, or PDGFR, the other sorafenib targets (5, 12). Our hypothesis was further corroborated by the description of KA and SCC occurring with more specific BRAF inhibitors, vemurafenib and GSK2118436 (2, 15).

In this work, we present a series of 31 skin lesions occurring in 17 patients treated with sorafenib 1 to 9 months after the beginning of the therapy. The lesions were benign and borderline or malignant skin tumors which might seem paradoxic because they occurred while under therapy with an antiproliferative agent.

The entity KA as a distinct tumor at the crossroad between benign and malignant tumors has been debated for decades. KA is a rapidly growing keratotic and crateriform tumor, very similar clinically and pathologically to well-differentiated SCC (16). It is characterized by frequent spontaneous resolution and originates from hair follicles (17). For some authors, it is considered to be low-grade SCC (18). KAs can be observed sporadically, mostly on sun-exposed areas, or in the context of genetic diseases, for example, the multiple self-healing squamous epithelioma (MSSE) also called Ferguson-Smith disease (19). Our pathologic classification between KA and SCC depended on the existence of cytologic atypia, invasive nests or lymphatic metastatic embols. When any of these signs were seen, we classified the crateriform epidermal neoplasms limited by sharp beaks as KA-like SCCs. When neither atypia nor invasion was associated with the crateriform lesion, they were considered as KAs.

The first hypothesis that we put forward following our clinical and pathologic studies of these tumors was that they seemed to belong to a common spectrum of lesions with the possibility of transition from benign lesions, that is, from a cystic more or less inflammatory lesion to KA and SCC. This “continuity” hypothesis is based on several observations. First, several of our patients had associated benign and malignant tumors. All of them had several benign cystic lesions, in addition to the lesions that were biopsied and studied (Table 1). One patient, in particular, developed the complete spectrum of lesions, that is, follicular cysts, perforating folliculitis, KAs, and KA-like SCC (patient # 7). Furthermore, all SCCs were crateriform in shape, resembling KA. We did not observe other clinical subtypes of SCC such as Bowen disease or classic sun-induced, ulcerated, or nodular forms of SCC.

Second, all of these lesions seemed to have developed at the expense of or around hair follicles. Follicular dystrophic lesions and perforating folliculitis are obviously follicular lesions and relics of hair follicles were indeed observed in every biopsy sample. Follicular ontogeny for KA is also well documented (20). The microscopically lamellar structure as well as the type of cytokeratins found in KA suggest that they are hair follicle–derived tumors (21). Thus, our first hypothesis is that all these proliferative epidermal lesions could represent various steps of a same generic process initiated by RAF inhibition in the skin and particularly targeting follicular keratinocytes.

To explore the biological effect of sorafenib on normal skin in vivo, we studied normal skin samples of patients treated with sorafenib or a placebo. Our results showed a significant increase in keratinocyte proliferation and no decrease of ERK phosphorylation, but, on the contrary, a clear tendency to an increase of ERK phosphorylation levels in the skin after treatment with sorafenib. We now have preliminary data on normal skin samples from patients treated with vemurafenib (before treatment and at day 21) showing similar results (data not shown).

These results, which at first glance might seem highly improbable, are in fact in line with recent published data showing that BRAF inhibitors induce activation instead of inhibition of the RAF/MEK/ERK pathway in wtBRAF cells (22–24). The proposed mechanisms underlying the activation of the pathway by a RAF inhibitor imply dimerization of RAF proteins and homodimerization (BRAF–BRAF) or heterodimerization (BRAF–CRAF). Binding of the RAF inhibitor enhances signaling via CRAF engaged in dimerization with BRAF (23). In a second model, binding of the inhibitor induces transactivation of the second element of the homo- or heterodimer (22). In both models, this activation requires wtBRAF together with an additional activation signal of the pathway, as is the case with RAS or EGFR activation. To understand whether this effect could explain keratinocyte proliferation in response to sorafenib, we used HaCat keratinocytes to see whether we could reproduce these results in an in vitro system. We showed that sorafenib indeed regulates the activation of the MAPK pathway and cellular proliferation in keratinocytes in a dose- and time-dependent manner with an increase of both parameters when the dose of sorafenib was low (1 μmol/L). For higher sorafenib concentrations (3 and 10 μmol/L), the effect on ERK phosphorylation was transient and disappeared after 1 hour and the effect on proliferation was rather inhibitory. This is in line with the transactivation model (22) in which RAF inhibitors activate ERK signaling at low concentrations but inhibit it at higher concentrations in wtBRAF cells. The decreased effect of higher concentrations of sorafenib on ERK phosphorylation could be explained by the direct binding and inhibiting activity of the drug on CRAF. As sorafenib has been previously shown to regulate the expression of the antiapoptotic MCL1 molecule (25, 26), we explored the effect of various doses of sorafenib on MCL1 expression. We found that the regulation of MCL1 expression was dependent on sorafenib concentration with inhibited MCL1 expression at higher drug doses. This effect of high doses (3 and 10 μmol/L) of sorafenib on MCL1 expression could explain that, although sorafenib activated ERK at all doses tested, it inhibits cell proliferation at higher doses. Thus, the complex action of sorafenib on the activation of the MAPK kinase pathway and on cell proliferation is dose dependent and results from the combined effects of the drug on RAF kinase inhibition on the one hand and on the control of apoptosis via MCL1 regulation on the other hand. We were able to detect the formation of BRAF–CRAF dimers in vitro by coimmunoprecipitation and directly in cells by PLA for all sorafenib concentrations tested. This dimerization-induced CRAF activation in keratinocytes can explain MAPK activation at least at a low sorafenib concentration which might be the case in the skin. Indeed, the plasma sorafenib concentration in clinical therapeutic use ranges from 3 to 6 mg/L (6.5 to 13 μmol/L) but the drug concentration in the skin is probably much lower than in plasma due to high protein binding (27). In most experimental cell lines models used, RAS activation seems required for RAF dimerization. We show here that RAS is not mutated in the HaCat keratinocytes and is rarely mutated in our biopsy specimen, however, it is possible that RAS is activated by upstream signaling like EGFR activation, which is very frequently observed in the skin. Indeed, we showed that when EGFR activity was inhibited with PD153035 in HaCat cells, sorafenib could not induce RAF heterodimerization anymore (Supplementary Fig. S2). This result shows that activation of the pathway upstream of RAF, presumably including RAS activation, is necessary for sorafenib-induced RAF dimerization in keratinocytes. Our data are, therefore, consistent with models from the literature stating that drug-induced RAF dimerization requires RAS activation either through oncogenic mutation or through RTK activation.

Thus, the model we propose is as follows: BRAF wild-type skin keratinocytes submitted to low sorafenib concentrations activate the MAP kinase pathway via BRAF–CRAF dimerization sparking off CRAF activation which in turn leads to keratinocyte proliferation. Additional events could then precipitate the proliferation and transformation process. In accordance with this model, we found oncogenic mutations (3 HRAS and 3 TP53 mutations and 2 TGFBR1) in benign and malignant skin lesions. Sporadic KA and SC, besides the use of RAF inhibitors, are known to harbor oncogenic RAS and TP53 mutations (30–34). RAS mutations are reported in 30% of KA (HRAS) and in 12 to 46% of 22% of SCC (mostly HRAS; refs. 30–32, 34). Furthermore, TGFBR1 mutations similar to the ones we found in our patients tumors have recently been shown to be germinal mutations directly involved in an autosomal genodermatosis characterized by the occurrence of multiple KAS, the MSSE also known as Ferguson Smith disease (19). Patient 10 had 2 lesions screened for mutations, one KA-like SCC with a TGFBR1 mutation (Pro83Leu) and benign cystic folliculitis that did not harbor any mutation, meaning that the TGFBR1 mutation was somatic and not germinal (Table 1).

The lesions presented by patients with MSSE are clinically and pathologically very similar to lesions presented in the context of RAF inhibitor treatment suggesting an effect of these drugs on the TGFB/TGFBR signaling pathway that warrants further exploration.

Six of the mutations found in our samples are known to be potentially UV induced (patients # 1, 3, 4, 6, 10, and 14), either located on typical UV-sensitive dipyrimidine sites (patients #1, 3, 6, 10, and 14) or on an ACA sequence known to give rise to a mutagenic photoproduct (patient #4; refs. 35, 36). Oncogenic events such as TP53 mutations are found in normal skin keratinocytes on sun-exposed areas (37). Indeed, a recent paper showed that intermittently sun-exposed skin contained an extensive number of intact cells with TP53 mutations, with an estimated annual rate of 35.000 new persistent TP53 mutations per individual (38). In accordance with this hypothesis, the 6 lesions that were harboring mutations with UV signature were located on sun-exposed areas, whereas, the 2 remaining mutations (patients #7 and 12) were found on nonexposed skin areas (palm and buttocks; Table 1). Thus, sorafenib-induced skin tumor might result from the combined effect of paradoxic MAP kinase pathway activation and coexisting or even preexisting oncogenic mutations that could have been generated by UV rays action on normal skin.

Altogether, the fact that we found oncogenic TP53, TGFBR1, and RAS mutations with a UV signature suggests that it is probably a wise move to advise patients receiving sorafenib or other RAF inhibitors to use efficient photoprotective measures.

The fact that we did not find oncogenic mutations in the other malignant lesions could be due to the existence of other mutations that we did not look for or by alternate ways of activating the MAP kinase pathway such as EGFR activation. Indeed, EGFR can also be activated in the skin by various stimuli such as physiologic sun exposure or wound healing (28, 29), and its activation is sufficient to promote RAF dimerization in keratinocytes treated with sorafenib (Supplementary Fig. S2).

It is possible that sun-induced RAF or TP53 mutation or EGFR activation were present in the keratinocyte prior to BRAF inhibition and could have led to keratinocyte transformation in the presence of sorafenib. In some of our samples, a mutational event existing prior to malignant transformation is supported by the fact that we found oncogenic mutations of HRAS and TP53 in 2 benign cystic lesions.

Recently, oncogenic RAS mutations, (mostly HRAS) were found at a higher frequency in 71% of 21 KA and SCC induced by vemurafenib, suggesting that RAS involvement might be more preeminent than other oncogenes in the ontogeny of vemurafenib-induced tumors (39). In conclusion, we propose a model in which paradoxic ERK activation by RAF inhibitors in the skin induces changes in keratinocyte biology with activation of the MAK pathway and cell proliferation associated with a spectrum of skin lesions ranging from more or less inflammatory benign follicular cystic lesions to KA-like SCC. Additional and maybe preexisting somatic genetic events might influence the evolution of the benign lesions to more proliferative and malignant tumors. The fact that these skin lesions occur more frequently with more potent and specific BRAF inhibitors than with the pan-RAF inhibitor sorafenib could result from the differential effect of these drugs on CRAF or from various bioavailabilities of the drugs in the skin. These findings have other important clinical implications. They also suggest that the use of RAF inhibitors might be pointless and even deleterious in normal tissues especially in cells harboring a RAS mutation, as it can be found in lung and colon tissue, for example (40, 41). Indeed, we cannot exclude the hypothesis that RAF inhibitors could promote cell transformation and/or proliferation in these wtBRAF cells with RAS activation and lead to internal tumors. Fortunately, such events have not been reported so far.

Altogether, the effect of RAF inhibitors in the skin illustrates the complexity and the variety of the consequences of the systemic blockade of a universal pathway like the MAPK pathway and reminds us that efficient targeted therapies always have off-target effects that need to be very carefully evaluated and monitored.

C. Robert is participating in advisory boards for Roche and GlaxoSmithKline. J.C. Soria is a consultant for Roche and Bayer. B. Escudier is a consultant for Bayer, Pfizer, GlaxoSmithKline, and Novartis. D. Malka is a consultant for Roche. No potential conflicts of interest were disclosed by the other authors.

The authors thank Lorna Saint Ange for editing.

This work was partly supported by the Institut National du Cancer (INCA 2007-1-PL7) for C. Robert, N. Dumaz, and A. Eychene.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.