BRAF inhibitors (BRAFi) elicit therapeutic responses in metastatic melanoma, but alarmingly, also induce the formation of secondary benign and malignant skin tumors. Here, we report the emergence and molecular characterization of 73 skin and extracutaneous tumors in 31 patients who underwent BRAFi therapy. The majority of patients presented with classic epidermal tumors such as verrucous papillomas, keratoacanthomas, and squamous cell carcinomas (SCC). However, 15 patients exhibited new or rapidly progressing tumors distinct from these classic subtypes, such as lymph node metastasis, new melanomas, and genital and oral mucosal SCCs. Genotyping of the tumors revealed that oncogenic RAS mutations were found in 58% of the evaluable tumor samples (38/66) and 49% of the control tumors from patients not treated with BRAFi (30/62). Notably, proximity ligation assays demonstrated that BRAF–CRAF heterodimerization was increased in fixed tumor samples from BRAFi-treated patients compared with untreated patients. Our findings reveal that BRAF–CRAF complex formation is significantly associated with BRAFi treatment, and may therefore serve as a useful biomarker of BRAFi-induced cutaneous and extracutaneous tumor formation. Cancer Res; 76(6); 1476–84. ©2016 AACR.
Activating BRAF mutations occur in about 7% of human malignancies and in approximately 40% of melanomas (1). BRAF inhibitors, vemurafenib and dabrafenib, induce objective tumor responses in 50% to 60% of the patients with BRAFV600− metastatic melanoma, and prolong their overall survival compared with previous standard chemotherapy (2). However, BRAF inhibitors have two major drawbacks. First, their action is limited over time with most patients developing secondary resistance in 6 to 8 months (3). Second, they are frequently associated with the emergence of multiple benign and malignant skin tumors: papillomas, keratoacanthomas (KA), squamous cell carcinomas (SCC), and more rarely with atypical pigmented nevi and even new melanomas (4). These rather unexpected adverse events had already been reported with the use of a previous less potent and less selective RAF inhibitor, sorafenib (5) that was used for its anti-VEGFR properties. The hypothesis that skin tumors arising with sorafenib were linked to the anti-RAF effect of the drug has already been demonstrated in vitro on keratinocytes that showed paradoxical activation of the MAPK pathway via CRAF activation associated with the presence of BRAF/CRAF heterodimers in the presence of sorafenib (6–8). As might have been expected from these data, more potent and more selective BRAF inhibitors, such as vemurafenib and dabrafenib, which became available some years later, gave rise to a higher incidence of new skin tumors, occurring in up to 15% to 20% of the patients. These secondary tumors harbor a RAS mutation in 21% to 60% of the cases (9, 10). In line with these findings, in vitro data on melanoma cell lines harboring an NRAS mutation but wild type for BRAF, showed paradoxical MAPK pathway activation in the presence of BRAF inhibitors. In these models, the signal was transduced via CRAF, itself transactivated via its dimerization with BRAF (11–13).
As RAS mutations can occur in various benign tumors from diverse tissues and organs, the worrisome question of whether extracutaneous cancer could be induced by a BRAF inhibitor was logically raised. Recently, the report on a patient with NRASmut myelomonocytic leukemia (14) seemingly aggravated by vemurafenib was the first of these dreaded cases. We report here on a series of skin and extracutaneous tumors arising or progressing rapidly under BRAF inhibitors. These cases, as well as tumors originating from patients without BRAF inhibitor treatment, were screened for the presence of oncogenic hotspot mutations in 47 oncogenes of interest, the level of ERK phosphorylation, and the level of BRAF–CRAF dimerization. The mechanism of reactivation of the MAPK pathway leading to ERK phosphorylation was explored with a newly developed test allowing BRAF–CRAF heterodimerization to be visualized directly on patient tumor specimens.
Patients and Methods
All patients had unresectable metastatic BRAFV600− melanoma except for one patient who had a BRAFV600E anaplastic pleomorphic xanthoastrocytoma. They were all treated at Gustave Roussy except patient #20 who was treated in Reims Hospital, and they were all still on treatment with a BRAF inhibitor on the day of the tumor biopsy/excision. They were treated with vemurafenib (960 mg twice daily; N = 28) or with dabrafenib (150 mg twice daily; N = 3) for at least 20 days. Patients were treated in the context of BRIM-3 (NCT01006980) or VE-BASKET (NCT 01524978) clinical trials, or according of the EMA authorization and they all gave their written informed consent for molecular analyses of their biopsied specimens (IRB# 08-027).
Analyses on patients' tumor specimens
Seventy-three tumor samples originating from 31 patients treated with BRAF inhibitors were analyzed (Supplementary Table S1). To be able to evaluate the effect of the treatment on their healthy-looking skin, 6 of these 31 patients, plus 3 additional patients who never developed a secondary tumor (patient #9, 10 and 15), agreed to undergo a normal skin biopsy, before and/or after the initiation of BRAF inhibitor treatment, which enabled us to study 16 additional “normal skin” specimens. A series of control tumors were analyzed as controls: 13 verrucous papillomas, 13 KA or SCCs, 10 vulvar SCCs, 10 urothelial carcinomas, 10 NRASmut melanomas, 7 melanocytic congenital nevi, originating from 63 patients who were not treated with a BRAF inhibitor (Supplementary Table S2).
Samples were cut into 4-μm sections, formalin-fixed, paraffin-embedded and stained with hematoxylin and eosin. The phospho-ERK protein was analyzed immunohistochemically (monoclonal antibody clone 20G11; Cell Signaling Technology). Epidermal keratinocytes were used as an internal positive control. A quickscore method (15) for immunohistochemical semiquantitation based on both the intensity and the proportion of brown stained nuclei was defined for each sample. The percentage of tumor cells staining positively throughout the section was termed score A and was assigned scores of 1 to 6 (1 = 0%–4%; 2 = 5%–19%; 3 = 20%–39%; 4 = 40%–59%; 5 = 60%–79%; 6 = 80%–100%). The whole section was scanned at low power to gauge the general level of intensity throughout the section. The average intensity, corresponding to negative, weak, intermediate, and strong staining, was allocated a score of 0 to 3, respectively, and termed score B.
A genetic analysis was performed on 119 evaluable samples, 12 biopsy specimens were too small to allow DNA extraction. The evaluable samples included 57 from patients treated with a BRAF inhibitor, and 62 control spontaneous tumors. Genomic DNA was extracted from 10- to 20-μm thick paraffin-embedded unstained slides of each skin lesion using the DNeasy Blood and Tissue Kit (Qiagen) according to the manufacturer's instructions. A Ion AmpliSeq primer pools targeting was designed to target mutations/deletions within the following gene regions of interest including: HRAS (exons 2 and 3), KRAS (exons 2 to 4), NRAS (exons 2 and 3), NF1, NF2, RAC1, BRAF, MAP2K1, MAP2K2, AKT1, AKT2, AKT3, PI3KCA, PIK3R1, MET, MITF, MLL3, NFE2L2, NOTCH1, NOTCH2, PDGFRA, PTEN, RET, ROS1, STK11, TGFBR1, TP53, ALK, CDKN2A, CTNNB1, EZH2, ABL1, CDK4, EGFR, ERBB2, FGFR1, FGFR2, FGFR3, FGFR4, GNA11, GNAQ, IDH1, IDH2, KIT, FLT3, JAK2, and KIT (exons 11, 13, and 17). The multiplex PCR-based Target sequencing was performed with the Ion AmpliSeq Library Kit 2.0 according to the manufacturer's instructions (Life Technologies). An equal amount of each sample libraries were pooled and Emulsion PCR amplification was carried out with the Ion PGM Template OT2 200 Kit (Life Technologies). After enrichment, the Sequencing was performed using 316 v.2 Ion Sequencing Chip according to the Ion PGM 200 Sequencing Kit with Personal Genome Machine (PGM; Life Technologies). Median coverage depth over 500 reads per amplicon (×500) was obtained for each analysis allowing detection of hot-spot variants down to 5% of allele frequency. The data were analyzed with the Torrent Suite Variant Caller 4.2 software and reported somatic variants were compared with the reference genome hg19. The germline variants found in 1000 Genome database (16) and/or In Exome Sequencing Project (ESP) database (ref: Exome Sequencing Project, ESP) were filter-out, and all remaining variants were confirmed by visualizing the data through Alamut v2.4.2 (the gene browser supporting human genes; Interactive Biosoftware).
SK-Mel2 and CRL-2885 Schwann cell lines were purchased directly from the ATCC cell bank in 2011 and 2014, respectively. The ATCC cell line characterization method is a short tandem repeat DNA profiling. Both cell lines were passaged in our laboratory for fewer than 6 months after resuscitation. The MEL-10 cell line was obtained from Laurence Zitvogel's laboratory and its mutational status checked using the 316 v.2 Ion Sequencing Chip according to the Ion PGM 200 sequencing Kit with Personal Genome Machine (PGM; Life Technologies). The SK-Mel2 cell line was cultured in EMEM and both the MEL10 melanoma cell line and the CRL-2885 Schwann cell line were cultured in DMEM, containing 10% FCS (Perbio) and antibiotics (penicillin/streptomycin; Invitrogen). Vemurafenib used for cell treatment was purchased from Selleckchem and dissolved in DMSO. All cell lines were regularly verified to be Mycoplasma free using a PCR-based test (Biovalley). Genomic DNA was isolated using a Genomic Tip kit (Qiagen), and the same genomic analysis, as described above, was performed using Ion Torrent PGM to confirm the presence of the NRASQ61R mutation in cell lines.
The proteins from the cell lysates were separated by SDS-PAGE and Western blot analysis was performed according to standard protocols using Erk1/2 (rabbit, 9102), and Erk1/2 phospho Thr202/Tyr204 (rabbit, clone 20G11, 4376) antibodies purchased from Cell Signaling Technology and the GAPDH (mouse, clone 6C5, MAB374) antibody purchased from Millipore.
Proximity ligation assay
Cells were grown on glass cover slips, treated with DMSO or vemurafenib, fixed with methanol or PFA 4%, and permeabilized with acetone or PBS-Tween 0.2%. The BRAF–CRAF interaction was detected by proximity ligation assay (PLA) on cell lines in vitro after treatment with vemurafenib (6 μmol/L, 24 hours) or DMSO.
For PLA on fixed tissue samples, following dewaxing and rehydration of tissue sections, antigen retrieval was performed by heating the slides for 30 minutes at 95°C in Tris-EDTA buffer, pH 9. The tissue sections and cell lines were then treated according to the manufacturer's instructions (Olink Biosciences) using BRAF (F7) and CRAF (C12) antibodies, with incubation at 4°C overnight. PLA minus and PLA plus probes were added and incubated for 1 hour at 37°C. Further oligonucleotides were added, and allowed to hybridize to the PLA probes, and a ligase joined the two hybridized oligonucleotides. The circular DNA was then amplified. The amplicons were detected using the “Brigthfield detection kit” for the chromogenic revelation or the “Far Red detection kit” for the fluorescence revelation. The links were visualized by confocal microscopy (SPE, Leica) or a scanner (Olympus VS120).
The BRAF–CRAF PLA was scored in a semiquantitative fashion incorporating both the number and the distribution of specific visualized dimers in tumor cells. The evaluations were recorded in each of the five intensity categories denoted as zero (no staining), 1 (weak but detectable), 2 (mildly distinct), 3 (moderately distinct), and 4 (strong).
The statistical analysis used was a nonparametric test to compare two unmatched groups (Mann-Whitney) or more than two (Anova one-way).
Patients and tumors
Seventy-three tumors originating from 31 patients that had emerged or progressed significantly during treatment with vemurafenib (N = 28) or dabrafenib (N = 3) between 2010 and 2014, were studied (Supplementary Table S1). The majority of these tumors (79.4%) were the classic types of skin tumors observed in the context of BRAF inhibitor treatment: 36 verrucous papillomas and 22 KA or cutaneous SCCs. However, 15 tumors were more atypical. Five tumors were de novo melanomas, which have also been reported during the course of BRAF inhibitor treatment but much less frequently than squamous keratinocytic tumors (17, 18). Two were congenital melanocytic nevi (CMN) that had undergone atypical transformation under treatment (patients #2 and #30, Fig. 1D–F). One patient had three basal cell carcinomas. Three patients had mucosal carcinomas: one vulvar (patient #32), one gingival (patient #23), and one lingual SCC (patient #33, Fig. 1A–C). One patient had an urothelial carcinoma (patient #20), and one patient rapidly developed a regional lymphatic metastasis from a previously excised cutaneous SCC (patient #6). The majority of these tumors were not present before treatment with BRAF inhibitors. A precursor benign lesion, in the form of a congenital melanocytic nevus (patients #2 and #30), was already present in 2 patients before treatment initiation. There was a pre-existing lesion in 2 other patients. The first patient had a 12 mm urothelial nodular lesion that was present on the initial CT scan but that had rapidly progressed (26 mm) under anti-BRAF therapy after 7 weeks of therapy (patient #20, Fig. 1G–I), and the second patient had an aggressive lymph node relapse of a previously treated cutaneous SCC of the lip that had been considered at low risk for relapse (patient #6, Fig. 1J–M).
Genotyping of the tumors
The mutations revealed in each sample are reported in Supplementary Table S1. Oncogenic RAS mutations were found in 58% of the samples from patients treated with a BRAF inhibitor (38/66) and in 49% of the control tumors (no BRAF inhibitor; 30/62). They were present in verrucous papillomas (59% and 69%) and KA/SCC (62% and 33%) in patients with and without BRAF inhibitor treatment, respectively. An NRASQ61− mutation was found in 2 of 5 of the secondary melanomas that were evaluable for DNA sequencing, including one of the 2 transformed CMN. NRAS mutations were present in 4 of 7 (57%) of the CMN from patients who had not received any a BRAF inhibitor. A KRASG12V mutation was found in the urothelial carcinoma and a HRASG13V mutation was found in the vulvar carcinoma, both carcinomas induced by vemurafenib. Two of the 10 urothelial carcinomas used as controls harbored a RAS mutation, but no RAS mutation was found among the 10 control vulvar carcinomas. The lingual SCC also harbored a HRASG12S mutation. NF1 mutations were found in 2 verrucous papillomas and in one SCC. PTEN deletions were found in 3% of the verrucous papillomas, whereas CDKN2A mutations were found in 5% of the KA/SCC in anti–BRAF-treated patients. Patient #6 carried a germline CDKN2AH66fs mutation that was logically found in the lymph node recurrence (from a formerly excised lip SCC) that had progressed rapidly under treatment. Interestingly, this patient developed multiple eruptive cutaneous SCC under vemurafenib, as shown in Fig. 1L.
ERK phosphorylation in tumor samples
To explore the activation of the MAPK pathway, we searched for ERK phosphorylation in the patients' samples. Among the tumors excised from patients treated with a BRAF inhibitor evaluable for phospho-ERK immunostaining, there were: 28 papillomas, 6 KA, 3 BCC, 9 SCC, 4 SSM, 1 transformed naevus, and 1 urothelial carcinoma. High phosphorylation of ERK was observed (Supplementary Table S1). Only the three basal cell carcinomas had low levels of ERK phosphorylation. In control tumors from patients who had not been treated with BRAF inhibitors, the level of phosphorylated ERK was also high and was not significantly different from that observed in tumors from patients under anti-BRAF treatment (Fig. 2).
Visualization of the BRAF–CRAF heterodimer
As the paradoxical activating effect of anti-BRAF treatment on the MAPK pathway is thought to be related to RAF protein dimerization, we searched for BRAF–CRAF heterodimerization using a sensitive PLA. We tested 2 NRAS mutant cell lines and one NF1-mutated Schwann cell line exposed or not to vemurafenib. This assay generates a signal when the two target proteins, recognized by two antibodies, are in close proximity (Supplementary Fig. S1). We indeed observed the induction of BRAF–CRAF heterodimer formation (Fig. 3A) associated with increased phosphorylated ERK in vemurafenib-treated cells (Fig. 3B and C). This induction of ERK phosphorylation was impaired when melanoma cell lines were exposed to vemurafenib in combination with a MEK inhibitor (e.g., trametinib or PD 0325901; Fig. 3B). As one might expect, MEK inhibition did not block BRAF–CRAF heterodimerization in vitro since the dimer formation is presumably upstream MEK (Supplementary Fig. S2).
The PLA was then performed for the first time on formalin-fixed paraffin-embedded tissue. In normal skin, BRAF–CRAF complexes were not increased in patients under anti-BRAF therapy compared with normal control skin (Fig. 4A). The number of BRAF-CRAF dimers was significantly higher in BRAF inhibitor-induced tumors compared with the same types of tumors in untreated patients (Fig. 4A). This was true for keratinocytic tumors but also for all other tumor types (Fig. 4A). BRAF–CRAF heterodimers were also visualized at high levels in the mucosal and melanocytic tumors (Fig. 4B) from the patient with the vulvar SCC (patient #32), in one of KA (patient #27) and in one of the transformed CMN (patient #2) exhibiting a particularly high level of BRAF–CRAF dimers under BRAF inhibitor treatment. In this last lesion, the dimers were more abundant in the more atypical part of the lesion compared with that seen in another nevus excised from the same patient before treatment. To evaluate the association between RAS mutations and BRAF–CRAF heterodimers, we differentially analyzed the BRAF–CRAF PLA score in RAS-mutant and RAS-wt tumors with their BRAF inhibitor-free control series according to their RAS status. We observed that the BRAF–CRAF PLA score was significantly increased by BRAF inhibitor treatment independently of the RAS mutational status (Fig. 4C).
BRAF inhibitors can, thus, also induce or facilitate the progression of various tumors, apart from the classic keratinocytic tumors reported to date. Until recently, mostly skin tumors had been reported, including some new melanomas (17, 18) and none of them were aggressive or became metastatic. Secondary tumor initiation was demonstrated to be linked to the paradoxical activation of the MAPK pathway via activation of CRAF and that this was facilitated by upstream events such as mutations of RAS proteins, upstream of BRAF. As RAS protein mutations can be found in many organs in premalignant tumors such as colon polyps for example, the emergence of extracutaneous tumors was, in theory, possible, and the recent report on the worsening of RASmut leukemia under an anti-BRAF agent was the first sign of this feared event (14).
We indeed observed the emergence of 3 mucosal tumors, 1 genital, and 2 oral lesions that were not present before BRAF inhibitor therapy as well as rapid progression of a RAS mutant urothelial carcinoma. In terms of skin tumors, we observed 5 new melanomas and the transformation of 2 congenital melanocytic nevi. One patient with a CDKN2A germline mutation rapidly developed an aggressive metastatic relapse from a previously resected cutaneous SCC, together with the efflorescence of cutaneous in situ SCCs. For this last patient, it cannot be definitely concluded that BRAF inhibitor therapy was responsible for the relapse, but this is a reasonable hypothesis as CDKN2A is known to be a negative regulator of CDK4 and cyclin D1, which also interferes with the MAPK pathway (19). The three basal cell carcinomas observed in one patient could be unrelated to BRAF inhibitor treatment because basal cell carcinomas frequently occur after 50 years of age.
All the tumors induced or promoted by BRAF inhibitors exhibited a high level of ERK phosphorylation, with RAS mutations in more than 50% of the cases. A transformed CMN, the vulvar, and the urothelial carcinomas also harbored a RAS mutation. In 42% of the tumors occurring in patients treated with a BRAF inhibitor, no RAS mutations were found. It is possible that other unscreened mutations, also activating the MAPK pathway, could be involved. Such was the case, for example, in three tumors that were found to harbor an NF1-inactivating mutation known to activate the NRAS protein (20, 21). In cell lines harboring RAS-activating mutations or NF1 loss of function, we observed the paradoxical activation of the MAPK pathway with ERK phosphorylation being increased by vemurafenib. This activation was associated with the increased formation of BRAF–CRAF heterodimers and was decreased when vemurafenib was associated with a MEK inhibitor. This is in accordance with a much lower incidence of SCC/KA, observed in less than 5% of the patients treated with such combinations (22–24). To potently inhibit mutant BRAF cells without inducing any paradoxical activation, anti-MEK agents or pan-RAF inhibitors called “paradox-breakers,” such as PLX7904, are also being developed (25–27).
From now on, most of the patients with metastatic melanoma harboring a BRAFV600 mutation are going to be treated with a combination of anti-BRAF and anti-MEK agents. However, the addition of MEK inhibitor in vivo significantly decreases but does not entirely abrogate the risk of BRAF inhibitor-mediated oncogenesis. This is illustrated by the recent report on the progression of a presumably dormant KRASmut metastatic colon cancer in a patient receiving the combination of dabrafenib and trametinib (28). In addition, vemurafenib as a single agent is currently being evaluated in several other BRAF-mutant tumor types as well as in the adjuvant setting, in patients having undergone a complete resection of melanoma lymph node metastases. The risk of inducing internal tumors raises a safety issue in this patient population, especially because 50% of them are not going to develop a recurrence of their melanoma.
It is, thus, of critical importance to be aware that various types of tumors can occur, and if possible, to be capable of identifying patients who might be at higher risk of developing them. For example, CMN commonly observed in about 1% to 6% of newborns (29) frequently harbor NRAS mutations (50%–80% for the largest lesions). The highest degree of caution is, thus, recommended before initiating anti-BRAFV600− therapies in patients presenting with this type of benign lesion because of the major risk of transformation. RAS mutations also occur other premalignant congenital or acquired premalignant lesions like nevus sebaceus or colon polyps for respective examples (30, 31). In rare genetic backgrounds like the Costello syndrome with constitutional activation of a RAS protein the risk of developing a secondary tumor is also substantial.
The physical proximity between BRAF and CRAF could be visualized in our patients' tumor samples. The rate of dimers was significantly increased by BRAF inhibitors both in the RASmut and RASwt secondary tumors. Of note with the sequencing method used here, the detection of RAS amplifications among the RASwt secondary tumors was not possible. We found that BRAF–CRAF dimers were low in the three basal cell carcinomas, suggesting that they might not be linked to the treatment. Neither the level of RAS mutations, nor ERK phosphorylation could help differentiate between a BRAF inhibitor-induced tumor and a tumor unrelated to this treatment. Our findings demonstrate that the level of BRAF/CRAF dimerization is the only biomarker significantly associated with BRAF inhibitors–induced oncogenesis in both cutaneous and extracutaneous tumors. The combination of anti-BRAF with anti-MEK, the standard of care for patients with BRAF-mutant metastatic melanoma, significantly decreases the risk of iatrogenic tumors. However, this risk is not totally abrogated by the addition of an anti-MEK agent. Furthermore, because anti-BRAF monotherapy is still presently evaluated in several other tumor types as well as in the adjuvant setting in patients with high risk of melanoma relapse, it is critical to be aware that the spectrum of iatrogenic tumors is wider than the one previously reported.
Disclosure of Potential Conflicts of Interest
A.M. Eggermont is a consultant/advisory board member for GSK. C. Robert is a consultant/advisory board member for Roche, BMS, Novartis, MSD, Amgen. No potential conflicts of interest were disclosed by the other authors.
Conception and design: L. Boussemart, H. Malka-Mahieu, S. Vagner, C. Robert
Development of methodology: L. Boussemart, I. Girault, M. Rubington, L. Lacroix, J. Adam, S. Vagner, C. Robert
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L. Boussemart, I. Girault, C. Mateus, E. Routier, M. Thomas, L. Lacroix, A. Cavalcanti, F. Grange, C. Robert
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): L. Boussemart, I. Girault, M. Rubington, G. Tomasic, M. Laporte, L. Lacroix, S. Vagner, C. Robert
Writing, review, and/or revision of the manuscript: L. Boussemart, I. Girault, L. Lacroix, A.M. Eggermont, F. Grange, S. Vagner, C. Robert
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): N. Kamsu-Kom, M. Thomas, S. Agoussi, M. Breckler, J. Adam
Study supervision: S. Vagner, C. Robert
Other (did some genomic experimentations): M. Breckler
The authors thank the patients for enabling them to present these rare cases to a scientific audience in the form of an article as well as the following Gustave Roussy platforms: Module de Développement en Pathologie, SIRIC SOCRATE (J. Adam), Translational Research Laboratory and Biobank (M. Breckler and L. Lacroix), S. Roy for patient data collection as well as “Vaincre le Melanome” and the Collectif “Ensemble Contre le Mélanome.”
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.