Genome-Wide Analysis Identifies MEN1 and MAX Mutations and a Neuroendocrine-Like Molecular Heterogeneity in Quadruple WT GIST Free

Approximately 10% to 15% of adult cases of gastrointestinal stromal tumors (GIST) do not harbor mutations in KIT or platelet-derived growth factor receptor alpha (PDGFRA) receptors and are often referred to as KIT/PDGFRA wild-type (WT) GIST (1).

Between 20% and 40% of KIT/PDGFRA WT GIST show loss of function of the succinate dehydrogenase complex (SDH), designated as SDH-deficient GIST or SDHB-negative GIST based on the loss of subunit B (SDHB) protein expression (2–4). The most frequent identifiable molecular events found in SDH-deficient GIST are germline and/or somatic loss-of-function mutations in any of the four SDH subunits (A, B, C, or D), with a prevalence of the subunit A involvement (4–7). Additionally, a genome-wide DNA hypermethylation or miRNA specific profile has been associated with SDH-deficient GIST (8–12). SDH-deficient GIST have distinctive clinicopathologic features, including a predilection for young women, gastric localization, mixed epithelioid and spindle cell morphology, diffuse KIT and ANO1 (DOG1) IHC positivity, frequent lymph node metastases, and an indolent disease, even when metastases are present (2, 3, 13). Moreover, SDHB IHC-negative GIST are characterized by overexpression of the insulin growth factor 1 receptor (IGF1R; refs. 14, 15). Many of these GIST arise in the context of the Carney–Stratakis syndrome (the dyad of GIST and paraganglioma), and are characterized by germline SDHB, SDHC, or SDHD inactivating mutations (16). They also occur in the context of the Carney Triad (gastric GIST, paraganglioma, and pulmonary chondroma) and may be characterized by SDHC hypermethylation (17, 18).

Amongst the KIT/PDGFRA WT GIST, the remaining cases include a subgroup harboring mutations in BRAF/RAS or NF1 and are referred to as RAS pathway (RAS-P) mutant GIST (approximately 15% of cases; refs. 19, 20); lastly, we distinguish a subgroup lacking mutations in the KIT/PDGFRA or RAS pathways, and retaining an intact SDH complex referred to as quadruple WT GIST (approximately 50% of KIT/PDGFRA WT GIST and 5% of all GIST; ref. 21). A massively parallel sequencing and gene expression study on two cases of quadruple WT GIST showed a distinct transcriptome profile profoundly different from SDHA-mutated GIST and KIT/PDGFRA mutated GIST, suggesting a different molecular background (22). Moreover, in recent reports, NF1 mutations, an MYC-associated factor X (MAX) mutation, and the ETV6–NTRK3 fusion have been described as novel molecular events in quadruple WT GIST (23–26).

The aim of this work is to improve the diagnostic process of quadruple WT GIST through a wide comprehensive molecular characterization of this subset of patients, essential for the identification of the driver molecular abnormalities as potential markers and targets of new treatments.

All the patients included in the study were quadruple WT GIST, being negative for mutations in KIT, PDGFRA, SDH, and RAS-P genes. All the cases were reported as being sporadic, lacking any personal or familiar history of a cancer prone disease. This study was approved by the local institutional ethical committee of Azienda Ospedaliero-Universitaria Policlinico S.Orsola-Malpighi (number 113/2008/U/Tess). GIST diagnosis was based on histologic evaluation and on immunohistochemistry of CD117 and DOG1 reviewed by expert pathologists.

Selection of quadruple WT GIST started from 30 KIT/PDGFRA WT GIST from adult patients without any personal or familial history of cancer. BRAF and KRAS mutational status was assessed by Sanger sequencing, while SDH deficiency was assessed by IHC for SDHB. We excluded 15 samples that showed SDH deficiency and one that carried the common BRAF V600E mutation, therefore identifying 14 cases that resulted BRAF–KRAS/KIT/PDGFRA WT and SDH intact. Only nine samples (4 fresh frozen and 5 FFPE samples) had sufficient DNA quality and yield for whole exome sequencing (WES), while whole transcriptome sequencing (WTS) and miRNA profiling were performed on 4 of the quadruple WT cases for which fresh-frozen tissue was available. Fresh tissue specimens were collected during surgical operation, snap-frozen in liquid nitrogen and stored at −80°C until analysis. NF1 alterations were identified directly by WES and mapping of the variants on the HGMD database (www.hgmd.cf.ac.uk). Moreover, WES analysis of matched peripheral blood (PB) as normal counterpart was performed to exclude germinal mutations. Patient's characteristics are listed in Table 1.

In order to further characterize quadruple WT GIST, we compared their molecular analyses and profiling to other GIST subsets; to this end, SDH-deficient GIST and KIT/PDGFRA-mutant GIST were analyzed. In particular, for gene expression profiles, 4 quadruple WT were analyzed in comparison to 14 mutated GIST (2 SDH, 5 PDGFRA, and 7 KIT), while for miRNA profiling 4 quadruple WT were compared with 4 SDH-deficient and 4 KIT/PDGFRA-mutant GIST.

For WTS analysis, total RNA was extracted from tumor specimens with the RNeasy Mini Kit (Qiagen), then cDNA libraries were synthesized from 250 ng of total RNA with TruSeq RNA Sample Prep Kit v2 (Illumina) according to the manufacturer's recommendations. Briefly, poly(A)-RNA molecules were purified using oligo-dT magnetic beads, then mRNA was fragmented and randomly primed for reverse transcription, followed by second-strand synthesis to create double-stranded cDNA fragments. The generated cDNA fragments went through a terminal-end repair process and ligation using paired-end sequencing adapters, then amplified to create the final cDNA library.

For WES analysis, genomic DNA was extracted from fresh-frozen tumor specimens and from matched PB with QiAmp DNA mini kit (Qiagen) or with QiAmp DNA micro kit (Qiagen) if the tumor sample was from FFPE block. Libraries were synthesized with Nextera Rapid Capture Exome Kit (Illumina) following the manufacturer's recommendations. Briefly genomic DNA (50 ng for fresh frozen and 100 ng for FFPE samples) was tagged and fragmented by the Nextera transposome technique to an average library size of 290 bp. DNA libraries were then denatured to single-stranded DNA and hybridized to biotin-labeled 80-mer probes designed to enrich 214,126 targeted exonic regions, then eluted from magnetic beads and amplified.

WTS and WES libraries were quality checked and sized with Agilent DNA 7500 chips on the Bioanalyzer 2100 (Agilent Technologies), then quantified using a fluorometric assay (Quant-iT PicoGreen Assay, Life Technologies). Paired-end libraries (12 pmol/L) were amplified and ligated to the flowcell by bridge PCR, and sequenced at 2 × 80 bp read length for WTS and 2 × 100 bp for WES, using Illumina Sequencing by synthesis (SBS) technology.

miRNA profiling was performed using TaqMan Low Density Arrays (Applied Biosystems), pools A and B, which allow to analyze 754 miRNA. Total RNA was isolated from tumor samples and retrotranscribed using the TaqMan MicroRNA Reverse Transcription Kit (Applied Biosystems) and MegaPlex RT primers (Applied Biosystems) pools A and B. cDNAs were preamplified using TaqMan PreAmp Master Mix and PreAmp primers (pools A and B; Applied Biosystems). The array cards were loaded with the preamplified sample and run on the 7900HT Fast Real-Time PCR System (Applied Biosystems).

After demultiplexing and FASTQ generation (both steps performed with bcltofastq function developed by Illumina), the paired-end reads were trimmed using AdapterRemoval (https://github.com/MikkelSchubert/adapterremoval) with the aim of removing stretches of low quality bases (<Q10) and Truseq/Nextera rapid capture adapters present in the sequences. The paired-end reads were then aligned on human reference genome HG19 (http://www.genome.ucsc.edu/) and analyzed with two different pipelines for WTS and WES data. Sequences coming from RNA-seq were mapped with the algorithms TopHat/BowTie (27) and the PCR and optical duplicates were removed with the function rmdup of Samtools (http://samtools.sourceforge.net). Gene expression profiling analysis was carried on first by adopting the function htseq-count (Python package Htseq: http://www.huber.embl.de/HTSeq/doc/overview.html) to quantify the number of reads mapped on genes included in the Ensembl release 72 annotation features (http://www.ensembl.org). Second, the evaluation of differential expressed genes was performed with the R-Bioconductor package edgeR and limma (https://bioconductor.org/) respectively to normalize and to compute the statistical analysis of differential gene expression between quadruple WT and mutated GIST.

Principal component analysis of gene expression profiling was performed with the function prcomp from stats R packages (https://www.r-project.org), while Multiple Experiment Viewer (http://mev.tm4.org) was adopted to the supervised hierarchical clustering using the Manhattan distance and the average linkage method. In order to identify the pathways overrepresented, we performed a gene set enrichment analysis with the WEB-based GEne SeT AnaLysis Toolkit (http://www.webgestalt.org) using as a priori gene sets the KEGG pathways database. DeFuse (http://compbio.bccrc.ca/software/defuse/), ChimeraScan (https://code.google.com/archive/p/chimerascan/), Tophatfusion (https://ccb.jhu.edu/software/tophat/fusion_index.shtml), and FusionMap (http://www.arrayserver.com/wiki/index.php?title=FusionMap) methods were used to detect chimeric transcripts from RNA-seq data.

miRNA data were analyzed with SDS Relative Quantification Software version 2.4. (Applied Biosystems); and miRNA with Ct values ≥35 were considered as not expressed and excluded from further analysis. Normalization was carried out by subtracting the mean Ct from individual Ct values. R-Bioconductor package limma was adopted to evaluate the differential expression profile between the quadruple WT and mutated GIST. For each of the significant differentially deregulated miRNA, the set of target genes were identified with the aims to reach the miRNA/mRNA network. The validated targets were obtained from the miRTarBase database (http://microrna.sanger.ac.uk/) that contains miRNA–target interactions (MTI) with experimental support. The predicted targets were retrieved from TargetScan (http://www.targetscan.org), DianaLab (http://diana.imis.athena-innovation.gr), miRAnda (http://www.microrna.org), mirDB (http://mirdb.org), and miRTarBase.

Using this information, miRNA array and mRNA from RNA-seq were analyzed to highlight pairs of mRNA/miRNA with opposite trends (UP vs. DOWN and vice versa).

Data from WES were mapped with Burrows-Wheeler Aligner with the default setting (28); the PCR and optical duplicates were removed as previously described for the RNA-seq, Genome Analysis Toolkit (https://software.broadinstitute.org/gatk) were used to locally realign, recalibrate, and call the Ins/del variants, while point mutations were identified with the algorithm Mutect (https://www.broadinstitute.org/cancer/cga/mutect). Single-nucleotide variants (SNV) and ins/del were annotated with a gene and protein alteration using Annovar (http://annovar.openbioinformatics.org); nonsynonymous and nonsense SNV, frameshift/non-frameshift Indels, and splice-site mutations were selected with a threshold read depth ≥15× and a variant allele frequency ≥0.2. All the variants were filtered in order to select novel or rare events (frequency in the population <1%) basing on database of human variability dbSNP (http://www.ncbi.nlm.nih.gov/SNP), 1000 Genomes (http://www.1000genomes.org), ExAC (http://exac.broadinstitute.org), and EVS (http://evs.gs.washington.edu/EVS). In-depth evaluation of high confidence somatic variants was performed by verifying the presence of alternate allele on the normal counterpart and manually visualizing each variation with the tview function of Samtools. Potential candidate drivers were highlighted considering the Catalog of Somatic Mutations in Cancer (http://cancer.sanger.ac.uk/cosmic), pointing out the Cancer Gene Census set, and predicting the effect of the mutations on protein structure and function with SNPeff (29).

Moreover, based on WES data, the analysis of amplifications and large deletions was performed making a consensus between Control FREEC (http://boevalab.com/FREEC) and ADTEX (http://adtex.sourceforge.net) with paired tumor/matched normal samples. Also, a filtering procedure was applied taking into account the uncertainty value given by Control FREEC (<80%) and the polymorphic copy-number variants from the Database of Human Genomic Variants (http://dgv.tcag.ca/dgv/app/home).

Relevant mutations were validated on tumor and PB DNA using specific primers and the Sanger sequencing method, as described previously (5). qPCR amplification of ASCL1, GAPDH, and GUSB was performed with real-time LightCycler 480 instrument (Roche). The ASCL1 mRNA expression level was evaluated on 4 quadruple WT GIST in comparison with 3 SDH-deficient and 5 KIT/PDGFRA-mutated GIST. Relative expression was estimated by the DDCt method, using GAPDH and GUSB genes as housekeeping controls.

To define the gene expression signature of quadruple WT GIST, we performed WTS in 4 quadruple WT GIST (GIST127, GIST133, GIST400, and GIST401) and compared it with the other GIST molecular subsets (2 SDH-deficient and 7 KIT-mutant and 5 PDGFRA-mutant). Unsupervised principal component analysis showed that the four subgroups cluster separately, highlighting that the quadruple WT GIST are a separate entity than the other molecular subgroups of GIST (Fig. 1A).

Supervised analysis resulted in 224 over and 23 underexpressed genes in quadruple WT GIST (FDR-corrected P < 0.05; Fig. 1B), in which it was possible to confirm the expression of molecular markers characteristic of this subgroup, including the COL22A1 and CALCRL genes (22).

Functional enrichment analysis highlighted 10 significantly enriched pathways (Supplementary Table S1), many related to a neural phenotype. In particular, the Neuroactive ligand–receptor interaction pathway was enriched, and interestingly some of the genes involved in this pathway belong to the Secretin family (Class B) G protein–coupled receptors (PTH2R, CALCRL, CRHR2, and GLP2R) and are overexpressed in the quadruple WT GIST (Fig. 1B). The neural-like background was also supported by the overexpression of the transcription factor ASCL1, a commitment lineage marker of neuroendocrine tumors (30), that was the most differentially expressed (FC = 10.7) between quadruple WT GIST and all the other GIST classes. In an additional cohort of 8 GIST (SDH or KIT/PDGFRA mutated), it was possible to confirm that ASCL1 mRNA expression was a specific feature of only the quadruple WT subgroup (Fig. 1C).

To fully define the signature of quadruple WT GIST we analyzed the miRNA expression profile against KIT/PDGFRA-mutant and KIT/PDGFRA WT-SDH-deficient GIST. A total of 66 differentially expressed miRNA were identified as specific of quadruple WT GIST (Fig. 2A; FDR P < 0.05).

The integration of gene expression levels with the targets of differentially expressed miRNA allowed the identification of a network of interactions where we identified 17 miRNA as putative regulators of the genes of neuroendocrine lineage and Neuroactive ligand–receptor interaction pathway (Fig. 2B). miRNAs predicted as regulators of signature genes are reported in Supplementary Table S2.

WES was performed on 9 quadruple WT GIST (4 fresh frozen and 5 FFPE tumor samples) and on matched normal DNA. An average of 60.5 million of reads per sample was obtained producing an average coverage per sample ranging from 54× to 76×. It is known that KIT/PDGFRA wild-type GIST can carry germline mutations in cancer-predisposing genes, including SDHA (already excluded) or NF1. Therefore, we initially analyzed whole exome data from matched normal DNA and then mapped the resulting NF1 variants on the HGMD and ClinVar databases. In this way, we identified two extremely rare germline variations: a p.R2594L in GIST268 (ExAC frequency = 2/121378) and a p.H1374Y in GIST279 (ExAC frequency = 1/118856), both recorded in ClinVar as variations of uncertain significance, and not present in the HGMD database. The NF1 germline alteration in GIST279 was previously described (24).

Copy-number analysis showed that all the quadruple WT tumor samples carried multiple regions of copy-number gains and losses (3–12 regions with copy-number alterations/sample), which are anyway not recurrent within the cohort. The only frequently altered genomic regions are those characteristic of KIT/PDGFRA-mutant GIST, with 6 of 9 samples showing at least one deletion in either 1p, 14q, or 22q chromosome arms (Supplementary Fig. S1 and Supplementary Table S3).

WES of tumor samples and matched normal DNA identified an average of 18 somatic mutations per sample (range, 6–29) (Supplementary Table S4). No somatic alterations in NF1 were identified in our population. The mutational profile did not show any other highly recurrent alteration shared by the majority of the cases. However, it was possible to identify a relevant oncogenic mutation in 6 of the 9 cases (Table 2 and Fig. 3).

GIST257 carried two distinct somatic mutations on TP53, leading to double inactivation of the protein, a frameshift deletion (p.S227fs*18) and a missense mutation predicted as pathogenic and recurrently mutated in multiple neoplasms (p.R158L; COSM10714).

GIST268 carried a frameshift deletion on MAX (c.100_110del; p.K34fs*31), which leads to a premature stop codon. Interestingly, this patient also carried the very rare missense germline NF1 variant (R2573L) that is recorded in the ClinVar database.

GIST320 was shown to carry a homozygous somatic frameshift deletion in the MEN1 gene (c.249_252delGTCT; p.I83fs*34), already reported in COSMIC as a recurrent event (COSM23398). This patient also showed a somatic missense mutation in TP53 (c.646G>A; p.V216M) that is frequently mutated in different tumor histotypes (74 reported cases in COSMIC: COSM10667).

GIST401 showed a somatic frameshift mutation in CHD4 (c.2659delC; p.R887fs*14), a component of the histone deacetylase NuRD complex that was recently added to the Cancer Gene Census list (http://cancer.sanger.ac.uk/census/).

GIST127 showed a complex Ins/del in the CTNND2 gene (c.2986–2987 del/ins AG>T), which leads to loss of the reading frame and premature protein truncation. CTNND2 was shown to be a mutational hotspot in glioblastoma, linked to loss of expression (31).

This sample also carried several other genetic alterations, detected by fusion transcript analysis from WTS data and validated by Sanger sequencing (Supplementary Fig. S2). All the chimeric transcripts led to inactivation of the proteins, apart from MARK2–PPFIA1, which should cause the deregulation of PPFIA1 expression, controlled by a different promoter region. Among inactivating fusions, we detected a SPRED2-NELFCD chimera that loses all the functional domains of the two proteins, and an event of duplication and inversion of RTKN2 locus that determined the insertion of a lnRNA downstream of the exon 8 of TET1, causing the loss of TET1 reading frame and the introduction of a premature stop codon.

A FGFR1 somatic missense mutation (c.1638C>A; p.N546K) was detected in GIST409. This alteration is already reported in the COSMIC database as a hotspot mutation in CNS and soft-tissue tumors.

The WES and gene expression analysis of quadruple WT GIST reveal that this subset of patients presents a homogeneous signature profile driven by an underlying molecular heterogeneity. Several mutations were also identified in the same tumors, with an average of 18 somatic mutations per sample; interestingly, despite the small samples size, no highly recurrent and shared events were found in all patients. However, excluding the germline variation of uncertain significance in NF1, in 6 out of 9 cases it was possible to identify at least one alteration in relevant potential driver genes.

In two cases we detected three pathogenic TP53 mutations, two of which in the same patient, supporting the likely role of this gene in an important fraction of quadruple WT GIST (>20%). As it is well known, the TP53 gene is involved in DNA repair activation and apoptosis initiation. For several decades, the role of TP53 deregulation in carcinogenesis has been studied in human cancers and associated with both the loss of tumor-suppressing function and the oncogenic function (32). The negative prognostic role of TP53 gene overexpression and its correlation with the increased malignant risk in GIST have been already described (33). However, to our knowledge, specific TP53 mutations have never been described in GIST patients. Therefore, the potential role in GIST development and progression of the novel mutations we identified should be further investigated.

MAX was found mutated in one case. MAX is a transcription factor and belongs to the MYC/MAX/MXD network functionally linked to cell-cycle arrest and differentiation. MAX mutation has been found very rarely involved in hereditary pheochromocytomas/paragangliomas (PCC/PGL) lacking other mutations in susceptibility genes and rarely in sporadic cases. With regard to GIST patients, a previous report firstly identified a MAX somatic truncating frameshift mutation (c.160delC; p.Gln54Lysfs*10) in a quadruple WT GIST and interestingly, both that case and our patient carried also additional events in NF1 (23). The previous patient carried a two-base insertion (c.6781_6782insTT; p.His2240Leufs*4) in the tumor that was not seen in the normal DNA, whereas our patient carried a missense germline NF1 rare variant. Even though this coincidence can be considered very rare and interesting at the same time, no definitive conclusions can be drawn on the association between NF1 and MAX genes in these two patients.

A homozygous MEN1 deletion was detected in one quadruple WT GIST. MEN1 is the tumor suppressor gene implicated in the Multiple endocrine neoplasia type 1 (MEN1) hereditary cancer syndrome, which is characterized by various combinations of endocrine neoplasia, and combinations of these tumors may be different in members of the same family (34, 35). To the best of our knowledge, the identification of MEN1 mutation in GIST is a novel finding.

The MAX and MEN1 gene mutations further extend the list of genes characterizing the neuroendocrine tumors (NET) family, along with NF1 and SDH, which are also implicated in the pathogenesis of KIT/PDGFRA wild-type GIST. NETs are a large family of diseases that generally occur as sporadic isolated tumors, sometimes also as part of complex familial endocrine cancer syndromes (35). Somatic and germline mutations of several susceptibility genes may lead to the development of NETs and many of them are shared by GIST lacking KIT/PDGFRA mutations. In fact, the current molecular knowledge underlines that among the KIT/PDGFRA WT GIST, a subgroup presents a deregulation of the SDH complex, another subgroup harbors germline and/or somatic mutations in NF1 associated or not to a clinical NF1 syndrome, and a subgroup of quadruple WT GIST harbors mutations in MAX and MEN1; therefore, all these subgroups can be associated with the neuroendocrine family. The four quadruple WT GIST studied in the present work showed a marked upregulation of ASCL1, an early immunohistochemical marker of neuroendocrine lineage, also playing a role in neural commitment, and of the genes belonging to the Class B Secretin family of G-protein–coupled receptors. In addition, a high relative expression of neural markers in a subset of SDH-intact WT GIST has been reported (36). Moreover, in the past we found that the gene expression profiles of 4 KIT/PDGFRA WT GIST—2 of them SDH deficient—profoundly differed from that of KIT/PDGFRA-mutated GIST, especially in the expression of those genes primarily restricted to neural tissues (37). Therefore, all these findings may reinforce the hypothesis that a great majority of KIT/PDGFRA WT GIST may derive from a cell at a different step of differentiation towards neural features or from a different cell of origin showing neuroendocrine commitment.

In the present study, other additional events in several genes were found. A frameshift deletion was detected in CHD4, a component of the Mi2–NuRD complex, which couples chromatin remodeling and histone deacetylation involved in transcriptional regulation, replication, DNA repair, and cell fate determination. Recently, CHD4 and several other components of the chromatin remodeling process were found recurrently mutated in several tumors (38, 39). Interestingly, MEN1 and MAX are also involved in epigenetic regulation and chromatin modifications (40, 41).

One truncating mutation in CTNND2 (δ-Catenin) was detected in one case. This gene is expressed in normal brain and is commonly overexpressed in several cancers. However, several somatic frameshift mutations are recorded in the COSMIC database, and loss-of-function mutations were identified in glioblastoma and pancreatic adenocarcinoma (31, 42). Moreover, in the same patients, several chromosomal rearrangements were detected that lead to premature stop of relevant genes, such as TET1 and SPRED2. Accumulating evidence indicates SPRED2, an inhibitor of the Ras/ERK signal transduction, as a tumor suppressor, and decreased levels of SPRED2 were associated with increased tumor invasiveness and metastasis (43). However, whether these alterations have an impact on GIST biology still needs to be assessed.

Finally, a FGFR1 somatic missense mutation was detected in one patient. FGFR1 is involved in several soft-tissue sarcoma and central nervous system tumors, and the same mutation we detected was also found in Ewing sarcoma (44). In GIST, the involvement of FGFR1 has been reported as one missense mutation and two fusion events (FGFR1–HOOK3 and FGFR1–TACC1) in quadruple WT GIST (45, 46); however, no other detailed reports on the FGFR1 deregulated pathway are already available in this disease. The FGFR1–TACC1 is the second fusion event associated with GIST along with ETV6–NTRK3 (25, 45, 46). In particular, in a recent report, these oncogenic translocations were observed in two of five cases, and the authors suggested to routinely test the translocation in quadruple WT GIST. However, these events were not found in our quadruple WT GIST patients; overall, the great difference observed regarding gene fusions between these two GIST populations confirms that the quadruple WT GIST can be considered a greatly heterogeneous cancer group that, in the future, could not be considered as a unique family anymore.

As a general clinical consideration, our findings do not help to profile the clinical characteristics of quadruple WT GIST. The only common aspect is the “non-gastric” site of the disease in a context with different gender and variable age of patients, variable disease presentation, and outcome, so a genotype–phenotype correlation cannot be hypothesized. For this reason, one challenging perspective in quadruple WT GIST is to enlarge the sample series as much as possible to collect more patients' data helpful to define a clinical classification and disease outcome.

In conclusion, our study showed that quadruple WT GIST present a homogeneous expression profile profoundly different from other GIST subsets. However, quadruple WT GIST show a great molecular heterogeneity, driven by different mutational events in several genes. So, the quadruple WT GIST that until now was a group defined as the subset of GIST that lack abnormalities of KIT, PDGFRA, SDH, and the RAS signaling pathway today can be considered as a large number of heterogeneous single entities with different molecular alterations. Among the mutational events identified in our population, the MEN1 and MAX gene involvement seems very interesting. These findings, together with the high number of tumor susceptibility genes, indicate that quadruple WT GIST seem to behave as neuroendocrine tumors. In fact, this picture resembles that of pheochromocytomas or other neuroendocrine tumors, characterized by a great variability in oncogenic driver genes, that ends up in a rather specific and characteristic gene expression profile. Further efforts are needed to understand if all these genomic events may represent secondary molecular hits implicated in tumor progression or the early causative event in quadruple WT GIST pathogenesis, in order to devise new targeted therapeutic strategies in this heterogeneous subset of GIST.

M.A. Pantaleo has expert testimony from an entity. A. Gronchi has received speakers bureau honoraria from and is a consultant/advisory board member of Novartis, Pfizer, and Bayer. G. Grignani is a consultant/advisory board member for Novartis, Pfizer, Bayer, Lilly, and Pharmamar. P.G. Casali has received speakers bureau honoraria from Bayer, Novartis, and Pfizer, is a consultant/advisory board member for Bayer, Blueprint Medicines, Novartis, and Pfizer. No potential conflicts of interest were disclosed by the other authors.

Conception and design: M.A. Pantaleo, M. Urbini, V. Indio, G. Ravegnini, M. Nannini, B. Vincenzi, S. Stacchiotti, A. Astolfi, G. Biasco

Development of methodology: M.A. Pantaleo, M. Urbini, V. Indio, G. Ravegnini, M. De Luca, G. Tarantino, P. Paterini, C. Ceccarelli, A. Altimari, E. Gruppioni, A. Astolfi

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M.A. Pantaleo, M. Nannini, A. Gronchi, B. Vincenzi, G. Grignani, C. Colombo, E. Fumagalli, L. Gatto, M. Saponara, M. Ianni, P. Paterini, D. Santini, M.G. Pirini, S.L. Renne, P. Collini, S. Stacchiotti, G. Brandi, P.G. Casali, A.D. Pinna

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M.A. Pantaleo, M. Urbini, V. Indio, G. Ravegnini, M. De Luca, G. Tarantino, A. Gronchi, M. Saponara, D. Santini, S.L. Renne, P. Collini, A. Astolfi, G. Biasco

Writing, review, and/or revision of the manuscript: M.A. Pantaleo, M. Urbini, V. Indio, G. Ravegnini, M. Nannini, S. Angelini, A. Gronchi, G. Grignani, E. Fumagalli, M. Saponara, S.L. Renne, P. Collini, S. Stacchiotti, P.G. Casali, A. Astolfi

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M.A. Pantaleo, M. Urbini, V. Indio, G. Ravegnini, M. De Luca, G. Tarantino, S.L. Renne, P. Collini, A. Astolfi

Study supervision: M.A. Pantaleo, M. Urbini, V. Indio, M. Nannini, M. De Luca, A. Gronchi, M. Saponara, A. Astolfi

The authors extend special thanks to the GIST Study Group members, University of Bologna, Bologna, Italy: Maria Astorino, Rita Casadio, Paolo Castellucci, Fausto Catena, Massimo Del Gaudio, Monica Di Battista, Antonia D'Errico, Giorgio Ercolani, Stefano Fanti, Michelangelo Fiorentino, Cristian Lolli, Anna Mandrioli, Pier-Luigi Martelli, Nico Pagano, Paola Tomassetti, Valerio Di Scioscio, and Maurizio Zompatori

The present study was supported by Italian Association for Cancer Research AIRC, My First Grant 2013 and Fondazione Isabella Seragnoli, Bologna, Italy (M.A. Pantaleo).

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.