Germline Mutations in the Mitochondrial 2-Oxoglutarate/Malate Carrier SLC25A11 Gene Confer a Predisposition to Metastatic Paragangliomas

Pheochromocytomas and paragangliomas (PPGL) are neuroendocrine tumors with a very strong genetic component. Up to 40% of patients with PPGL carry a germline mutation in one of the 13 susceptibility genes reported so far (two proto-oncogenes and 11 tumor-suppressor genes; for review, see ref. 1). Mutations in SDHx genes (SDHA, SDHB, SDHC, SDHD), encoding the tricarboxylic acid (TCA) cycle enzyme succinate dehydrogenase (or mitochondrial complex II) account for approximately 50% of the germline mutations identified in affected patients and cause multiple or metastatic PPGL (2). These mutations abolish succinate dehydrogenase (SDH) activity, resulting in the accumulation of its substrate, succinate, which acts as an oncometabolite by inhibiting 2-oxoglutarate (2-OG)–dependent dioxygenases. These include hypoxia-inducible factors (HIF) prolyl-hydroxylases (PHD), ten-eleven translocation enzymes (TET) DNA demethylases (which catalyze the conversion of 5-methylcytosine (5-mC) into 5-hydroxymethylcytosine (5-hmC; ref. 3) and JmjC-domain containing histone demethylases (which promote lysine demethylation of histones; ref. 4). Inhibition of PHD, TET and JmjC-domain-containing proteins respectively result in a pseudohypoxic signature and a hypermethylator phenotype (5, 6). We previously reported the first integrated multi-omics study performed on a large collection of 202 PPGL that remarkably classified tumors according to their genotype (7). Unsupervised classifications revealed that SDHx-mutated PPGL systematically clustered together across the genomic platforms. These tumors formed the so-called cluster 1A (pseudohypoxic) in the transcriptome study, the cluster M1 (hypermethylated) following methylome analysis and the cluster Mi1 in the miRnome-based classification (7). Intriguingly, two tumors were classified within the SDHx-related clusters in these data sets (Fig. 1A and B; Supplementary Fig. S1), even though both did not carry any germline or somatic SDHx mutations. In the first tumor, whole-exome sequencing identified the first PPGL-causative germline mutation in the FH gene, encoding fumarate hydratase, another TCA cycle enzyme (6) previously known to predispose to hereditary leiomyomatosis and renal cell carcinoma (HLRCC). Subsequent study identified FH mutations in four additional patients with PPGL, who had a high incidence of the metastatic forms of the disease (8). The aim of the present study was to identify the mutation causing the second tumor with an SDHx-like genomic profile.

The tumor and blood samples were prospectively collected by the French COMETE Network. This study was conducted in accordance with the Declaration of Helsinki and was approved by the local ethics committee (Comité de Protection des Personnes (CPP) Ile de France II, June 2012). Each patient signed a written informed consent for genetic analyses. The procedures used for PPGL diagnosis were in accordance with both internal and international clinical practice guidelines (2, 9). Germline DNA was extracted from leukocytes according to standard protocols. Tumor DNA extraction was performed using the AllPrep Kit (Qiagen) and the DNA Mini Kit (Qiagen) for frozen tumor and paraffin-embedded tissue, respectively. DNA was quantified and its purity assessed with a NanoDrop ND-1000 spectrophotometer (Labtech). Mutation analyses of the major PPGL susceptibility genes were performed as previously described in the Genetic department of Hôpital Européen Georges Pompidou in Paris (2). Two groups of patients for whom no known mutation in a PPGL predisposition gene had been identified were enrolled: a first group of 267 patients with a clinical history suggestive of a hereditary PPGL (multiple tumors, early onset or family history of PPGL) and a second group of 372 patients with apparently sporadic PPGL (Table 1).

Exome sequencing was performed by IntegraGen Genomics (Evry, France) as previously described (6). The germline mutation identified in the SLC25A11 gene was validated by Sanger sequencing.

PPGL susceptibility genes coding regions and exon-intron boundaries (SDHx, VHL, EPAS1, EGLN1, EGLN2, NF1, RET, TMEM127, MAX, FH and MDH2 genes) were amplified using the SDH MASTR Kit V2.0 (Multiplicom, Belgium). Libraries were sequenced on MiSeq platform (Illumina) using v3 chemistry according to the standard protocol. Alignment and variant calling were performed using SeqNEXT (JSI medical systems) and PolyDiag (Paris Descartes university) software.

Targeted gRNA was designed in silico by using http://crispr.genome-engineering.org/ as previously described (10). gRNA was transcribed in vitro using the kit PrecisionX Cas9 SmartNuclease RNA system (System Biosciences CAS510A-KIT). Briefly, targeted oligonucleotides were annealed to form double strands, and then cloned into SmartNuclease Linearized T7 gRNA Vector. The resulting recombinant vector was linearized by EcoRI digestion. The generated template was used to produce gRNA by in vitro transcription. gRNA was then purified (miRNeasy minikit, Qiagen) and its purity was assessed using the Experion RNA StdSens Analysis Kit (Bio-Rad).

Wild-type immortalized mouse chromaffin cells (WT imCC), previously generated by our laboratory (6), were transfected with 0.6 μg of gRNA (RNA i-MAX; Life Technologies) and 2μg of Cas9 (System Biosciences CAS940A-1) and YFP plasmids (Lipofectamine 2000; Life Technologies). Fluorescent cells were cloned by FACS sorting. Clones were screened by direct sequencing of the targeted sequence of murine Slc25a11 gene. Two consecutive transfections were performed: the first generated three heterozygous clones out of 70, and the second performed on heterozygous clones generated 51 homozygous clones out of 129. Predicted exonic off-target sequences were screened by direct sequencing (KRT9 exon 2, TMEM260 exon 16, ALDH2 exon 10, KANK1 exon 19, FCRLB exon 2). Mycoplasma contamination was ruled out using the PCR Mycoplasma Test kit I/C (PromKine, PK-CA91-1048). All experiments were performed between passages 15 and 20.

Slc25a11^Δ/Δ clone 6 was transfected with 2 μg of Slc25a11-expressing vector (ORIGENE, MR204477) using Lipofectamine 2000 (Life Technologies). Twenty-four hours later, selection was performed by Geneticin (Sigma, G8168, 800 μg/mL) and selected cells were cloned by limited dilution under Geneticin selection.

200 ng of denatured DNA was added to the Reacti-Bind Coating Solution (Thermo, 17250) and incubated in a 96-well plate at 37°C for one hour. After blocking for 30 minutes with a blocking buffer [PBS1X, 5 g/l BSA, and 0,5% of Kathon CG/ICP (Supelco 5-0127)], anti-5-methylcytosine (1/5,000, Calbiochem, NA81) or anti 5-hydroxy-methylcytosine (1/200, Actif Motif, 39759) and secondary antibody (1/1,000) were added to the wells and incubated at 37°C overnight. The plate was then incubated with Streptavidin-HRP (BD Biosciences, 554066) for 30 minutes. Finally, revelation was performed by adding a solution of 5 mg/mL of TMB (Sigma), citrate 0,1 mol/L pH 5 and 3% hydrogen peroxide. Reaction was stopped with sulfuric acid and the absorbance was read at 450 nm. The experiments were performed three times.

Data were analyzed by one-way ANOVA. The results were considered to be significant if P < 0.05. Statistics tests were carried out using the Graph-Pad software.

The unexplained tumor with a pseudo-SDHx profile (Fig. 1A and B; Supplementary Fig. S1) was a nonsecreting abdominal paraganglioma diagnosed in a 46 years-old patient (Patient #1). Genetic counseling revealed a family history of cancer, with his father deceased at 49 from a cancer of unknown origin discovered by bone metastases, a paternal uncle with bone cancer, and a paternal cousin with a head and neck tumor (Fig. 1C). Unfortunately, it was not possible to retrieve neither more specific clinical information regarding the relatives of the paternal branch suspected to be affected, nor tumors or DNA samples. Patient #1′s tumor was positive for SDHB by immunochemistry (Fig. 1D) confirming the absence of SDHx mutations (11). Whole-exome sequencing performed on leukocyte DNA identified 2,343 germline genetic variations (2,213 SNVs and 130 InDels). After filtration of the common polymorphisms described in the dbSNP, ExAC or 1000 Genome databases, 218 variations remained. We hypothesized that similar to SDHx and FH genes, the causative gene could be a tumor suppressor, and searched for variations associated with a loss of heterozygosity (LOH) using SNP array in tumor DNA (Fig. 1E). We found 54 variations, of which 11 were classified as pathogenic according to in silico analyses (Supplementary Table S1). Among them, we identified a candidate genetic variation [c.715C>A, p.(Pro239Thr)] in the SLC25A11 gene (NM_003562), encoding the mitochondrial 2-oxoglutarate/malate carrier (OGC) protein. The presence of the SLC25A11 variation, heterozygous in germline and homozygous in tumor DNA, was confirmed by Sanger sequencing (Fig. 1F). OGC is part of the malate/aspartate shuttle (MAS) and mediates the transport of 2-OG from the mitochondrial matrix to the cytoplasm in an electroneutral exchange for malate (Fig. 2A). The proline at position 239 is part of a PX[D/E]XX[K/R]X[K/R] signature sequence motif (PROSITE PS50920, PFAM PF00153), which is highly conserved across species, particularly in the SLC25 family of mitochondrial transporters (12). Mutagenesis of this amino acid in bovine OGC results in a severe defect of 2-OG transport activity (12). We performed OGC immunohistochemistry (IHC) on the paraffin-embedded tumor. We observed no OGC protein in the PGL carrying the SLC25A11 mutation in contrast with the 20 PPGL tumors without SLC25A11 mutations used as controls (6 SDHx, 1 FH, 3 RET, 4 NF1, 1 TMEM127, 3 MAX, 2 sporadic cases; Fig. 2B; Supplementary Table S2). 5-hmC and H3K9me3 IHC further validated a hypermethylated phenotype affecting both DNA and histones, as previously described in SDHx-mutated tumors (6, 13). Altogether these data suggested that germline [c.715C>A, p.(Pro239Thr)] SLC25A11 mutation associated with somatic loss of the wild-type allele causes the patient's PGL.

To establish the frequency of SLC25A11 mutations in patients with PPGL, the 8 exons and exon/intron junctions of the gene were sequenced by Sanger method in a large cohort of 639 patients for whom no germline mutation had been identified in the major PPGL susceptibility genes (SDHx, RET, VHL, TMEM127, MAX, FH), nor in the recently described MDH2 gene (14). These patients had a mean age at diagnosis of 34.8 years ± 15.6 (Table 1) and among them, 121 (19%) had a metastatic disease. Eighty-one (13%) had multiple PPGL and 13 (2%) reported a family history of the disease. We identified six patients with germline SLC25A11 mutations (Table 2; Supplementary Table S3). Five patients had a single metastatic abdominal PGL and one a head and neck PGL; none had a known family history of PPGL (Supplementary Fig. S2 and Supplementary Clinical Data). All six mutations (2 missense, 2 frameshifts, one intronic and one silent mutation) were predicted to be deleterious in silico (Table 2). None of these variants was present in the dbSNP nor ExAC databases. The missense mutations affected highly conserved amino acids (Supplementary Fig. S3A) located in the signature protein sequence or in the alpha matrix helix, both well known to be critical for OGC function (Supplementary Fig. S3B; refs. 12, 15, 16). Interestingly, the silent p.Ala236Ala mutation was associated with a marked decrease in SLC25A11 mRNA levels in leukocytes (Supplementary Fig. S3C). LOH was identified in tumor DNA of the three patients for whom tumor tissue was available (Supplementary Fig. S3D) and all evaluable tumors were negative for OGC by IHC (Table 2; Supplementary Fig. S4). Next-generation sequencing (NGS) performed on tumor DNA of patient #1, #4 and #6 showed no somatic mutation in SDHx, VHL, EPAS1, EGLN1, EGLN2, NF1, RET, TMEM127, MAX, FH and MDH2 genes. We had not enough material to performed NGS on PGL of patient #2. We carried out 5-hmC (Supplementary Fig. S5), H3K9me3 (Supplementary Fig. S6) and H3K27me3 IHC (Supplementary Fig. S7) to establish whether these tumors displayed a hypermethylator phenotype, as the one we observed in the initial SLC25A11-mutated case. In all but one of the SLC25A11-mutated tumors (patient #2), 5-hmC immunolabelling was less intense in tumor cells than in endothelial or sustentacular cells (Table 2; Supplementary Fig. S5), whereas H3K9me3 and H3K27me3 were positive in all cases (Table 2; Supplementary Fig. S6 and S7). Hence, SLC25A11 mutations apparently mediate the inhibition of TET and JmjC-domain containing demethylases, as we previously reported in SDHx- and FH-mutated PPGL (6).

To fully demonstrate that SLC25A11 is a new tumor-suppressor gene, we generated a knockout (KO) of the Slc25a11 gene using the CRISPR/Cas9 approach in wild-type immortalized mouse chromaffin cells (WT imCC; ref. 6; Supplementary Fig. S8A). Two Slc25a11^Δ/Δ clones (c4 and c6) were selected for analyses. These clones carried a homozygous c.720delG and a heterozygous c.720_733del14 variant in the Slc25a11 gene (Supplementary Fig. S8B and S8C), both leading to premature stop codons in exon 7. The corresponding mutation results in the loss of the third signature sequence motif and the H56 alpha matrix helix (Supplementary Fig. S8D). A rescue was then performed by stably transfecting an Scl25a11-expressing vector in the Slc25a11^Δ/Δclone 6 (clone c6-R). Western blot analysis of OGC protein levels revealed a loss of the transporter in both Slc25a11^Δ/Δ clones, and its re-expression in the c6-R cells (Supplementary Fig. S8E). Both Slc25a11^Δ/Δ clones exhibited nuclear translocation of HIF2α similar to Sdhb^−/− cells, which was not seen in WT imCC, nor in the c6-R cells (Fig. 3A). We used an ELISA method to quantify the levels of 5-mC and 5-hmC in the different cell types. As expected, 5-mC levels were 2-fold higher in Sdhb^−/− imCCs than in WT cells. Similarly, both Slc25a11^Δ/Δ clones displayed an approximately 2-fold increase in 5-mC content, reflecting a hypermethylated phenotype, which was reversed in the Slc25a11-rescued cells (Fig. 3B). Accordingly, 5-hmC levels in WT imCC and c6-R were higher than in Sdhb^−/− and Slc25a11^Δ/Δ cells (Fig. 3C). Hence, Slc25a11 inactivation promotes both a pseudohypoxic and a hypermethylated phenotype. We next evaluated the proliferation, migration and adhesion capacities of Slc25a11^Δ/Δ imCCs in comparison to those of wild-type, Sdhb^−/− and Slc25a11-rescued cells. Slc25a11-deficient cells displayed a decrease in proliferation, compared to WT cells, which was much less marked than that observed in Sdhb-deficient cells (Supplementary Fig. S8F). They showed increased adhesion as compared to WT cells, which was however inferior to that observed in Sdhb^−/− imCC (Fig. 3D). Slc25a11 KO cells demonstrated a 2-fold increase in collective migration as compared with the WT and c6-R cells and comparable with the Sdhb^−/− imCC (Fig. 3E). Altogether, these in vitro data obtained with two different Slc25a11 KO clones and the reversion of the phenotypes shown by the rescue experiments clearly demonstrate that loss of Slc25a11 gene mediates pseudo-hypoxic signature, hypermethylator phenotype and the acquisition of metastatic properties.

To decipher the mechanism linking SLC25A11 loss of function and this phenotype, we performed metabolomic analysis in the different cell types, which showed normal levels of succinate (Supplementary Fig. S8G) but higher levels of aspartate and glutamate, and lower levels of 2-OG in Slc25a11^Δ/Δ cells (Fig. 4A). Interestingly, treating Slc25a11^Δ/Δ cells with 2-OG reversed the migratory phenotype, demonstrating the role of this metabolite in the SLC25A11 tumorigenic cascade (Fig. 4B). Metabolomic analysis performed on the frozen tumor of patients #1, #6 and #7 showed normal succinate levels (Supplementary Fig. S9) whereas, consistent with MAS dysfunction and in vitro data, aspartate and glutamate concentrations were elevated in all three SLC25A11-mutated tumors compared with control PPGL samples from the cluster 1A (2 SDHB and 1 SDHD-mutated cases) or the cluster 2 (3 RET, 1 TMEM127 and 1 sporadic case; Fig. 4C). 2-OG levels were highly variable between tumors homogenates and did not show such decrease (Supplementary Fig. S9).

We identified SLC25A11 as a new tumor-suppressor gene implicated in the predisposition to metastatic paraganglioma. This gene encodes a carrier that participates to malate–asparate shuttle (MAS) by mediating the transport of 2-OG from the mitochondrial matrix to the cytoplasm in an electroneutral exchange with malate. MAS is composed of OGC (SLC25A11) and of two aspartate–glutamate carriers: CITRIN (SLC25A13) and ARALAR (SLC25A12; Fig. 2). This shuttle regenerates NADH pool in mitochondrial matrix to allow complex I function (17). After the identification of 7 PPGL predisposing genes encoding TCA cycle enzymes (SDHA, B, C, D, SDHAF2, FH and MDH2), it is the first time that a gene encoding a mitochondrial carrier is implicated in PPGL tumorigenesis. Interestingly, patients with CITRIN deficiency caused by homozygous germline mutation in SLC25A13 have an increased risk of hepatocellular carcinoma following hepatic inflammation and fibrosis (18). Also, heterozygous germline mutations in the SLC25A13 gene have been implicated in hepatocellular carcinoma in Asiatic population, suggesting a potential role for MAS in tumorigenesis (19). Nevertheless, mutations in SLC25A13 reported in the latter study are frequent in Asiatic population raising the question of their actual pathogenicity (20).

More strikingly, a gain-of-function mutation in the GOT2 gene, encoding the mitochondrial aspartate aminotransferase was very recently reported in a PGL patient, further reinforcing the link between MAS dysfunction and PPGL (21). The aspartate aminotransferase catalyzes the interconversion of aspartate and 2-OG to oxaloacetate and glutamate. Our metabolomics data show a marked increase in aspartate and glutamate in both SLC25A11-mutated human samples and KO mouse cells associated with a decrease in 2-OG in the KO cells. Various derivatives of glutamate and aspartate were previously shown to be potent inhibitors of HIF prolyl hydroxylases (22). Furthermore, these changes associated with the decrease in 2-OG levels may suggest an increased GOT2 activity in SLC25A11-deficent cells. This could lead to an alteration of the 2-OG/succinate ratio and to the subsequent inhibition of 2-OG dependent enzymes. This is consistent with the recent data on GOT2 activating mutation, where the aspartate/glutamate ratio is also increased (21). Hence, although further studies will be needed, these observations may explain the link between MAS dysfunction and tumorigenesis.

We show that SLC25A11 gene mutations account for 1% of all PPGL, a frequency equivalent to the recently identified PPGL susceptibility genes (SDHA, TMEM127, MAX, FH; refs. 1, 8, 14). Although these data will need to be confirmed in an independent cohort, our analysis of more than 600 patients does suggest that SLC25A11 gene should now be screened in patients with PPGL, especially for those with a malignant phenotype. It is still unclear what will be the penetrance of the disease in affected families. The information we could obtain on few relatives of the SLC25A11-mutation carriers were suggestive of family history of cancers (cases reported with bone metastases of unknown primitive cancer, leukemia, as well as colon, breast, prostate and throat cancers) but not precise enough to allow drawing definite conclusions. So, it should further be assessed whether SLC25A11-mutations may confer an increased risk to other type of cancers. Moreover, SLC25A11 somatic mutations or copy-number alterations have been reported in various types of cancers in The Cancer Genome Atlas (TCGA) project or the Catalog Of Somatic Mutations In Cancer (COSMIC) database (Supplementary Table S4), and 33 of these 145 samples were shown to present an underexpression of SLC25A11 mRNA, thus strengthening the hypothesis of a role of SLC25A11 in the tumorigenesis of other types of cancers. Interestingly, a low expression of SLC25A11 is associated with a reduced survival in renal and pancreatic cancers, according to the TCGA studies (23).

It is worth noting that SLC25A11 gene mutations are strongly associated with the development of metastatic PPGL as 5% of all metastatic patients in our cohort were SLC25A11 mutations carriers and a malignant phenotype was observed in 5 out of the 7 (71%) SLC25A11 mutation carriers (Table 2). In particular, we found germline SLC25A11 mutations in 5 out of 30 (17%) patients with a single, apparently sporadic metastatic abdominal PGL (Table 1). In PPGL, malignancy diagnosis is made at the time of the diagnosis of the first metastasis (24) and the identification of an SDHB or an FH mutation are a risk factor for malignant disease. This study suggests that similar to SDHB and FH mutations, SLC25A11 could be considered as a new genetic risk factor of metastatic PPGL. This observation is important for the follow-up of mutation carriers and may also have some important consequences on their therapeutic management. SDHx, FH- and now SLC25A11-mutated tumors should be managed as the so-called ‘Cluster 1A’ associated tumors. They display pseudo-hypoxic and hypermethylator phenotypes that probably participate to the acquisition of some metastatic properties (6, 25). Hence, antiangiogenic therapies or demethylating agents such as low-dosed 5-aza-deoxycytidine may be of therapeutic benefit for these specific patients. Similarly, it has recently been shown that MGMT promoter methylation in SDHB-metastatic PPGL confers an increased response to temozolomide (6, 26). Hence, SCL25A11-mutated carriers may also benefit from such therapy.

In conclusion, we show, using a large cohort of affected patients and an in vitro experimental model generated by CRISPR/Cas9 technology, that SLC25A11 is a new tumor-suppressor gene conferring a predisposition to metastatic PPGL. The data reported here show how a mitochondrial dysfunction driven by a transporter inactivation can lead to tumorigenesis, and further broadens the field of mitochondrial genetic defects and cancer.

No potential conflicts of interest were disclosed.

Conception and design: A. Buffet, N. Burnichon, A.-P. Gimenez-Roqueplo, J. Favier

Development of methodology: A. Morin, J. Favier

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L.-J. Castro-Vega, F. Habarou, C. Lussey-Lepoutre, E. Letouzé, H. Lefebvre, I. Guilhem, M. Haissaguerre, M. Padilla-Girola, T. Tran, L. Tchara, J. Bertherat, L. Amar, C. Ottolenghi, A.-P. Gimenez-Roqueplo

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Buffet, L.-J. Castro-Vega, F. Habarou, E. Letouzé, N. Burnichon, A.-P. Gimenez-Roqueplo, J. Favier

Writing, review, and/or revision of the manuscript: A. Buffet, L.-J. Castro-Vega, F. Habarou, I. Raingeard, J. Bertherat, N. Burnichon, A.-P. Gimenez-Roqueplo, J. Favier

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A.-P. Gimenez-Roqueplo

Study supervision: A.-P. Gimenez-Roqueplo, J. Favier

We thank Profs. Pierre-François Plouin and Xavier Bertagna for making this work possible through the COMETE Network. We also thank Jean-Michael Mazzella, Mélanie Menara, Aurélien de Reyniès, Nabila Elarouci, Dr. Rossella Libé, Prof. Antoine Tabarin, Prof. Cécile Badoual, Dr Tchao Méatchi, Dr. Mathilde Sibony, Dr Frédérique Tissier, Prof. Nathalie Rioux-Leclercq, Dr. François Le Gall, Dr. Françoise Gobet, Prof. Béatrice Vergier, Dr. Isabelle Pellegrin, Dr. Joel Edery, and Dr. Nathalie Carrere for their help, Estelle Robidel, Maëva Ruel, Irmine Ferrara, Stefani Mazurkiewics, Christophe Simian and Marie Fontenille for technical assistance, Catherine Tritscher for administrative assistance and Pr. Bertrand Tavitian for helpful discussion. We also thank all members of the Genetics Department and Biological Resources Center and Tumor Bank Platform, Hôpital européen Georges Pompidou (BB-0033-00063), all the members of the Tumor BioBank-biological resource center of Rouen University Hospital, all the members of the Centre de Ressources Biologiques Plurithématique Bordeaux Biothèques Santé (BB-0033-00036) and the Mass spectrometry platform of the Biology Department of the Necker Hospital (Christophe Merlette). This work has received funding from the Agence Nationale de la Recherche (ANR-2011-JCJC-00701 MODEOMAPP), the European Union Seventh Framework Program (FP7/2007-2013) under grant agreement n° 259735, the Plan Cancer: Appel à projets Epigénétique et Cancer 2013 (EPIG201303 METABEPIC), the European Union's Horizon 2020 research and innovation program under grant agreement No 633983 and by the Institut National du Cancer and the Direction Générale de l'Offre de Soins (PRT-K 2014, COMETE-TACTIC, INCa-DGOS_8663). A. Buffet received a financial support from ITMO Cancer AVIESAN (Alliance Nationale pour les Sciences de la Vie et de la Santé, National Alliance for Life Sciences & Health) within the framework of the Cancer Plan and from la Fondation pour la Recherche Médicale (FDT20170436955). C. Lussey-Lepoutre and N. Burnichon are funded by the Cancer Research for Personalized Medicine - CARPEM project (Site de Recherche Intégré sur le Cancer - SIRIC). The group is supported by the Ligue Nationale contre le Cancer (Equipe Labellisée). This work is part of the "Cartes d'Identité des Tumeurs (CIT) program" funded and developed by the ‘Ligue Nationale contre le Cancer’ (http://cit.ligue-cancer.net).

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.