Whole genome duplication is an early event leading to aneuploidy in IDH-wild type glioblastoma

Glioblastoma, the most frequent and lethal form of glioma, displays chromosome instability and recurrent somatic copy number alterations (SCNA). Chromothripsis and whole genome duplication (WGD) have been recently identified in cancer. In the present study, we analyzed SCNA and determine the ploidy pattern in 123 IDH-wild-type glioblastomas, using SNP array data. WGD and chromothripsis events were validated using, respectively, FISH and CTLPScanner. WGD was detected in 11.4% glioblastomas (14/123) and was associated with TP53 mutation (p = 0.0068). It was an early event occurring after the recurrent SCNA observed in diffuse high-grade gliomas. Glioblastomas with WGD were more aneuploid compared to glioblastomas without WGD (p < 0.0001). Chromothripsis occurred in 29.3% glioblastomas (36/123) and mostly affected chromosomes 7, 9 and 12, with amplification of oncogenes (EGFR, MDM2/CDK4), and homozygous deletion of tumor suppressor genes (CDKN2A). There was a significant association between chromothripsis and gene rearrangement at a given locus. WGD is an early genetic event significantly associated to TP53 mutation and leading to chromosome instability and aneuploidy in IDH-wild-type glioblastoma. Chromothripsis recurrently targets oncogenes and tumor suppressor genes that are key players in gliomagenesis and tumor progression. The occurrence of chromothripsis points to underlying gene rearrangements (including gene fusions), potential therapeutic targets in glioblastoma.


INTRODUCTION
Glioblastoma (GBM) is the most frequent and most aggressive form of diffuse glioma. The prognosis is very poor, with a median overall survival of 15 months after maximum safe resection and radiochemotherapy [1]. GBM is one of the most genetically unstable cancers. It is characterized by numerous chromosome (chr) copy number alterations (CNA), such as chr 7 gain, chr 9p loss, and chr 10 loss, along with CDKN2A homozygous deletion (chr 9p21) and EGFR amplification (chr 7p11) [2]. Chromosome instability (CIN) may be the cause or the consequence of GBM development. In high-grade diffuse gliomas, CIN may initiate tumorigenesis.
Even if somatic CNA (SCNA) have been well described in GBM, general mechanisms leading to massive rearrangements implicating several, if not all, chrs are not well understood. Among these mechanisms, whole genome duplication (WGD) leads to tetraploidization of cells, which may pave the way to aneuploidy [3]. Several studies have shown that WGD is an early event in tumorigenesis. Tetraploid cells have been observed in the transition from premalignant to malignant disease (e.g., Barrett's oesophagus, cervical intra-epithelial neoplasia), suggesting that a genomedoubling event can be a driver of tumorigenesis [4][5][6]. The frequency of WGD varies across tumor types; WGD is more frequent in carcinomas, with an incidence >50% in colorectal, breast, lung, ovarian and oesophageal carcinomas, but has been reported in 25% GBM [7]. In a recent pan-cancer analysis, tumors with WGD exhibited a greater number of SCNA, with marked chr losses occurring after WGD [8]. Carter et al. showed that WGD increased tolerance to subsequent CIN events [7]. Furthermore, the authors observed that genome-doubled samples were associated with significantly older age at diagnosis and greater incidence of tumor recurrence. No study has correlated WGD with age and overall survival in GBM.
CT is a recently described phenomenon whereby massive chr shattering occurs during a single (mitotic) event [9]. It is a cataclysmic event in which a single or a few chrs are shattered and then randomly reassembled in a derivative chr, sometimes with additional circular double-minute chrs. The latter may include an oncogene whose amplification may confer a growth advantage to the cell. Conversely, some chr pieces may be lost. In addition, chr rearrangements can juxtapose the coding sequences of two genes, leading to novel fusion genes with oncogenic potential. 2-3% of all cancers show evidence of massive remodelling of a single or a few chrs, with tens to hundreds genomic rearrangements [9]. CT has recently been proposed as a novel mechanism for genetic instability and cancer development. It has been reported in a broad range of tumors (prostate carcinomas, multiple myelomas, GBM, medulloblastomas, neuroblastomas) [10,11]. The mechanisms underlying CT are not yet known. To date, only two studies, both using the TCGA-GBM cohort, have analyzed CT occurrence in GBM [8,12]. Zack et al. showed that CT events mostly involved chrs 9 and 12, and were associated, respectively, with deletion of CDKN2A (9p21) and co-amplification of MDM2/CDK4 (12q15/12q14) [8]. A recent study by Furgason et al. on 12 GBM using whole genome sequencing established a link between CT and amplification of known oncogenes (e.g., EGFR, MDM4, MDM2/CDK4) [12].
The occurrence and impact of WGD and CT have never been studied in a large series of primary GBM independent from the TCGA cohort. Herein, we investigated WGD and CT events in a series of 123 primary IDH-wild-type GBM and correlated the results with the survival data.

Patient characteristics
Median age at diagnosis was 63 years old (range: 22 to 84 years). Male-to-female ratio was 1.56. Median OS was 14.2 months and median PFS was 7.7 months.

Immunohistochemistry
We performed IHC against IDH1-R132H mutant protein on the whole cohort. No GBM expressed IDH1-R132H mutant protein arguing for the diagnosis of primary IDH-wild-type GBM.

Sanger sequencing
No IDH1/2 mutation was detected supporting the diagnosis of IDH-wild-type GBM according to the 2016 WHO classification [2]. Thirty-four cases out of 123 (27.6%) harbored TP53 mutation.

WGD is present in 11.4% of GBM
In most cancers, WGD corresponds to tetraploidy, which is a type of polyploidy in which a single cell has two sets of chromosomes (4n or 92 chromosomes) instead of one (2n or 46 chromosomes).
SNP array data analysis showed that most GBM (109/123, 88.6%) had a near-diploid karyotype with a number of chrs between 39 and 52, and an average estimated ploidy of 2.03. However, some GBM (14/123, 11.4%) presented a near-tetraploid DNA content, with a number of chrs oscillating between 58 and 92, and an average estimated ploidy of 3.46, representing WGD (Supplementary Figure 1).
In comparison with near-diploid GBM, WGD GBM displayed more chr losses (but no gain) and allelic imbalances. In near-diploid GBM, a ploidy status near 2 was observed in 76.7% cases whereas a ploidy status near 4 was observed in only 22.4% WGD GBM (p < 0.0001). There was a tendency for chr losses after genome doubling. This observation is supported by the average estimated ploidy of 3.46 in near-tetraploid GBM. Hence, GBM with WGD presented with a larger overall number of SCNA.
We used a FISH approach to confirm WGD in 6 GBM (out of 14) with a near-tetraploid karyotype (Supplementary Table 2). At least three chrs per sample were tested. For all the chrs tested, the copy number was in accordance with that observed on GAP analysis. For case GBM49-003, the SNP array data showed that there were two copies of chr 2 with a heterozygous allelotype, there were three copies of chr 19 with an allelic imbalance, and there were four copies of chr 17 with a heterozygous allelotype. The same copy numbers were detected using FISH ( Figure 2).
Moreover, analysis of the SNP array profiles helped retrace the chronology of the genetic events in GBM. For example, a near-tetraploid GBM presented two types of chr losses: chr losses with a homozygous allelotype (2 DNA copies instead of 4) (e.g., chrs 10, 13q and 14q) and chr losses with a heterozygous allelotype (e.g., chrs 2 and 4) ( Figure 3). According to these observations, chrs 10, 13q and 14q were lost before WGD occurrence, and chrs 2 and 4 were lost after genome doubling. In our cohort, 73.3% chr losses (11/15 cases) did not present loss of heterozygosity (LOH), hence occurred after WGD. Inversely, chr 10 loss and chr 9p loss occurred before WGD (8/8 cases and 6/6 cases, respectively) in our cohort. This observation suggests that WGD is an early event and most likely contributes to CIN. For 3 WGD GBM (out of 14), the recurrent tumor was analyzed by SNP array and all 3 cases displayed stable WGD profiles, without significant additional losses or gains (Supplementary Figure 2).
We did not find any association between WGD and common SCNA in GBM (Supplementary Table 3). The . WGD is in pink and CT in dark blue (bottom rows). Note that WGD cases present marked aneuploidy with many chr losses. Abbreviations: SCNA, somatic copy number alterations; chr, chromosome; ampl, amplification; HD, homozygous deletion; LOH, loss of heterozygosity; WGD, whole genome duplication; CT, chromothripsis. www.oncotarget.com frequency of specific arm-level SCNA (e.g., 7p gain, 9p loss) was not significantly different in diploid vs tetraploid GBM, except for chr 10q loss, which was less frequent in WGD GBM (p = 0.0044).
In our cohort, even if the difference was not statistically significant, WGD GBM tended to occur at a younger age compared to near-diploid GBM (median age 55 vs 63 years, respectively). There was no association between WGD and survival times (OS and PFS; OS: 13.3 months in WGD cases vs 14.2 months in diploid cases).

29.3% of GBM harbored CT patterns
Out of 123 GBM, 36 (29.3%) presented a CT pattern (Figures 1 and 4). CT events were defined as 1) the presence of at least 10 genomic rearrangements per chr arm with such rearrangements occurring in no more than 4 chrs in a given tumor, 2) a clustering of breakpoints, and 3) interspersed loss and retention of heterozygosity. Thirtyone of the 36 cases (86.1%) presented a CT pattern only on one or two chrs whereas 5 cases presented a CT pattern on three or four chrs ( Figure 4). CT events mostly affected genomic regions harboring common driver alterations in GBM (Table 1). Indeed, CT was detected on chrs 7, 9 and 12, involving EGFR (7p11), CDKN2A (9p21), and MDM2/CDK4 (12q15/12q14) loci, respectively. Seventyfive percent GBM (27/36) displayed CT at least at one of these loci. The presence of a CT pattern in a given genomic region was highly associated with detection of gene amplification or homozygous deletion in that region. CT patterns were found at the EGFR locus in 15 patients (41.6%) and at the CDKN2A locus in 14 cases (38.8%). In our series, 15 of 52 (28.8%) GBM harboring EGFR amplification presented a CT pattern affecting the EGFR locus (p < 0.0001) and 14/72 (19.4%) GBM with CDKN2A homozygous deletion displayed a CT pattern on chr 9p21 (p = 0.0003) ( Figure 5). CT also involved the TP53 locus in two cases harboring TP53 homozygous deletion. We also observed CT patterns at other loci displaying amplification of MDM4 (1q32), MET (7q31) or PDGFRa gene (4q12) (one case each). Finally, 6 GBM (out of 123, 4.8%) had CDK4 amplification (12q14.1) and 13 GBM (13/123, 10.5%) presented co-amplification of MDM2 and CDK4 genes. More than half of those co-amplified cases (7/13; 53.8%) presented a CT pattern at the MDM2 and CDK4 loci (p < 0.0001) (Figure 3). In 5 of these 7 cases, we observed additional amplifications with a CT pattern on chr 12 (Supplementary Table 4). Some amplicons implicated oncogenes such as KRAS (l2p12) and HMGA2 (12q14). We also noted amplification of STAT2 gene (12q13.3) in one case.

DISCUSSION
The study presented herein thoroughly describes two distinct genetic events that may play a key role in the initiation and/or progression of GBM. We studied WGD and CT in a large series of 123 primary IDH-wild-type GBM independent from the TCGA GBM cohort. WGD is a common event in human cancer, with an average frequency of 37%. The frequency is highly variable across tumor types; carcinomas from the ovary, breast, lung, and esophagus have an incidence of genome doubling of about 50% whereas WGD has not been observed in leukemia [7]. Two studies reported a frequency of WGD in TCGA GBM of 25% and 11%, respectively [7,8]. Our frequency of 11.4% is consistent with these previous findings. Copy number losses occurring in a diploid cell will result in LOH. This LOH will leave a permanent footprint in the genome, persisting after a genome-doubling event.
In contrast, losses occurring after genome doubling are less likely to exhibit LOH [13]. In our cohort, the small proportion of LOH observed on the SNP array profiles suggests that WGD was an early genetic event, potentially leading to CIN. However, some events such as chr 10 loss and chr 9p loss appeared before genome doubling. This observation suggests that key SCNA in GBM appear before genome doubling; WGD may generate CIN, with occurrence of many subsequent chr losses. This hypothesis is consistent with the results of Carter et al., who showed that most common SCNA in GBM occur before WGD [7]. types of chr losses: chr losses with a homozygous allelotype (2 DNA copies instead of 4; e.g., chrs 10, 13q, 14q) (solid boxes) and chr losses with a heterozygous allelotype (three DNA copies; e.g., chrs 2 and 4) (dashed boxes). According to these observations, chrs 10, 13q and 14q were lost before WGD occurrence, and chrs 2 and 4 were lost after genome doubling. Abbreviations: BAF, B-allele frequency; LRR, LogR Ratio; CEN, centromere. www.oncotarget.com In Zack et al.'s work, WGD was inferred to occur earliest relative to focal SCNA among lineages where WGD was common (ovarian, bladder and colo-rectal cancers) and after most focal SCNA in lineages in which WGD was least common (GBM, renal cell carcinoma) [8].
We showed that GBM with WGD were more aneuploid compared to GBM without WGD (p < 0.0001). Dewhurst et al. showed that tolerance of aneuploidy was enhanced in tetraploid clones from a colon cancer cell line but was a rare event in diploid cells [13]. Daughter cells derived from a parental cell that had undergone a segregation error frequently died or underwent cellcycle arrest. Daughter cells arising from tetraploid clones after a segregation error died or arrested less frequently, with the majority continuing through mitosis in the subsequent cell cycle [13]. WGD is a lot more common in plants compared to animals. Polyploidy is usually an evolutionary dead end, except in abnormal circumstances when polyploids might have an evolutionary edge over non-polyploids [3]. Robust growth of tetraploid cells may be via evolution toward a more favorable karyotype [14]. After genome doubling, massive chr losses occur (leading to intermediate ploidy level), which may allow tumor cells to reach a genomic equilibrium that favors survival and growth. Dewhurst et al. demonstrated that genome doubling is an early event in some colorectal cancers, permitting the evolution of more genomically complex, subtetraploid, higher-stage tumors [13]. Polyploidy confers adaptive potential to the (tumor) cells especially during periods of extreme stress by providing mutational High-level event indicates gene amplification (³ 4 copies) or homozygous deletion (no copy); oscillating pattern indicates alternating copy number states (1, 2 or 3) without amplification or homozygous deletion of genomic regions. High-level events were more frequent than oscillating patterns. CT mostly involved chrs 7, 9, and 12, which harbor key oncogenes or tumor supressor genes in GBM. Abbreviations: CT, chromothripsis. www.oncotarget.com robustness. Polyploidy-related changes in gene expression and epigenetics may facilitate or accelerate adaptation (to a new or changing environment) [3]. WGD may contribute to intra-tumor heterogeneity, a substrate for selection and tumor evolution [15]. Tumor cells with WGD may become resistant to radiation therapy and chemotherapy and hence, give rise to tumor recurrence. Of note, the recurrent tumors we analyzed still displayed WGD suggesting that genome doubling indeed confers a selective growth advantage to the cells.
Different mechanisms may lead to the emergence of tetraploid cells in tumors, such as cellular stress, abnormal cell division, or telomere shortening [16]. Mutations affecting the DNA damage response (e.g., TP53 mutation) enable polyploid cells to further proliferate. The propagation of tetraploid cells is inhibited by p53, which arrests the cell cycle in G1 phase [17]. However, p53 inactivation, such as present in 90% of GBM, and gene overdosage (in tetraploid cells) confer tolerance for SCNA [13], allowing tetraploid cells to survive. This is in accordance with our results showing a statistically significant association between TP53 mutation and WGD occurrence (p = 0.0068).
In our cohort, there was no association between WGD and CT. Only 4 out of 14 WGD GBM displayed CT (28.5%). The size of the cohort may be too small to show such an association. Mardin et al. showed in a model cell line that hyperploidization preceded CT, implicating hyperploidization as a "risk factor" for CT in vivo [18]. They showed that the tetraploid cell lines exhibited a significantly higher level of CNA as those derived from the diploid cell line. They observed highly clustered CNA in nine cases, all of which arose in hyperploid lineages. Their analysis indicated an increased rate of CT events in hyperploids compared to diploids (p < 0.05). Mardin et al. showed in human Sonic-hedgehog pathway-driven CDKN2A homozygous deletion (arrow) and chr 9p loss with (C) or without (D) CT. Note interspersed loss and retention of heterozygosity with no more than three different copy number states except for focal amplification or homozygous deletion. www.oncotarget.com medulloblastoma specimens that CT indeed occurred significantly more often in hyperploid compared to diploid tumors [18]. The higher tolerance for CIN in tetraploids might explain why CT could be associated with hyperploid cells but not with diploid cells. Polyploidy buffers the effects of partially recessive deleterious mutations. Transient tetraploidy can, by generating and buffering aneuploidy, result in long-term adaptation. According to Dewhurst et al., a genome-doubling event could represent a macro-evolutionary leap in tumors, which precipitates and sustains extensive chromosomal rearrangements such as CT [13]. Large phenotypic leaps facilitate rapid evolution to novel selection pressures [19]. CT potentially constitutes a mechanism by which aggressive, spontaneous tumor, such as GBM, could arise in a relatively short period of time [12]. Of interest, we studied 43 cases of IDH-mutated 1p/19q-codeleted oligodendrogliomas using SNP arrays and none displayed WGD or CT (data not shown). The better prognosis of such tumors might be related to a more stable genome and the absence of WGD or CT event.
In our cohort, WGD was associated with a younger age at diagnosis (not statistically significant) but had no impact on survival. In Dewhurst et al.'s study, WGD was independently predictive of poor relapse-free survival in early-stage colo-rectal cancer [13]. Analyses in larger GBM series will be needed to determine the association between WGD and age or prognosis.
The second genetic event studied herein is CT. In our cohort, CT was detected in 29.3% of cases, consistent with the findings of the TCGA studies [9,12,20]. In accordance with previous works [12], CT mostly involved chrs 7, 9 and 12, with associated EGFR amplification, CDKN2A homozygous deletion and CDK4/MDM2 co-amplification, respectively. Our results show that CT underlies amplification of oncogenes and homozygous deletion of tumor suppressor genes. In our cohort, 28  GBM using whole genome sequencing [21]. Four of 12 patients (33%) harboring CDKN2A/B deletion displayed complex rearrangements at the 9p21 locus and 8/14 GBM (57%) with EGFR amplification presented complex rearrangements at 7p11. Taken together, these data suggest that complex rearrangements including CT are responsible for EGFR amplification and CDKN2A/B deletion. Many studies suggest that CT is a phenomenon leading to chr rearrangements (including gene fusions) implicating yet to discover driver genes. Few fusion genes have been described in high-grade gliomas. Singh et al. reported FGFR3-TACC3 and FGFR1-TACC1 gene fusions in 3% GBM [28] The FGFR3-TACC3 fusion might be suspected on SNP arrays by detecting focal gains involving FGFR3 (chr 4p16.3). The identification of such fusion genes allowed the development of targeted therapies, with promising preliminary results [22]. Frattini et al. described recurrent in-frame fusions involving EGFR (chr 7p11) and different partners (SEPT14, PSPH) in 7% GBM [30]. Of note, in our cohort, one case harboring a CT pattern on chr 7p11 and EGFR amplification displayed an EGFR-SEPT14 fusion. Four additional cases displayed gene fusions recurrently implicating EGFR, SEPT14 or VOPP1 genes; those fusions, detected by RNA sequencing, were validated by RT-PCR and Sanger sequencing (data not shown, manuscript in preparation). Recently, a fusion between NAB2 and STAT6 genes (chr 12q13.3) associated with a focal 12q13 amplification has been reported in a 29-year-old patient with an IDH1-mutant GBM [23]. Chr 12 often harbors numerous genomic rearrangements, which are associated with the occurrence of gene fusions [24,25]. In our cohort, out of 13 GBM with MDM2/CDK4 co-amplification, 7 had additional focal amplifications on chr 12q. In view of those results, focal gains or amplifications may indicate the presence of yet-to-identify gene fusions. Of particular interest, Parker et al. described a fusion between RELA and C11orf95 resulting from CT on chr 11q13 in supratentorial ependymomas with a dismal prognosis [26]. CT has also been associated with gene fusions in other tumor types (medulloblastoma, small cell lung carcinoma, diffuse large B cell lymphoma) [27][28][29]. The presence of gene fusions in CT regions remains to be thoroughly investigated as it may open new therapeutic avenues in GBM.
Numerous mechanisms for CT occurrence have been proposed, including telomere attrition (leading to breakage-fusion-bridge cycles), aberrant mitosis producing micronuclei, and premature chr compaction [12,30]. Chr segregation errors during mitotic cell division can entrap DNA within a micronucleus. Micronucleated chrs are susceptible to shattering during the next mitosis generating multiple distinct DNA fragments [30]. In the absence of functional p53, DNA double-strand break repair ensues through error-prone non-homologous end joining, which directly links multiple fragments together in a haphazard manner by ligation [30]. Interestingly, tetraploid cells display frequent chr mis-segregation, supporting the hypothesis that WGD is a risk factor for CT.
With regard to the limitations of the study, we only tested 4 chromosomes per sample by FISH to confirm WGD. Isolation of tumor cells from fresh tissue would have allowed analysis of whole nuclear DNA content by flow cytometry. However, FISH technique based on the use of centromere probes represents a sensitive method for detecting chromosome aneuploidies [31]. Regarding CT, more stringent criteria can be used to identity CT patterns based on whole-genome paired-end DNA sequencing data [32]. Nonetheless, SNP arrays allow for the detection of the major features of CT: clustering of breakpoints, regularity of oscillating copy-number states, and prevalence of regions with interspersed loss and retention of heterozygosity [32]. A consensus on which criteria are to be met to define CT has yet to be reached.
In conclusion, we demonstrated that WGD was an early event associated with TP53 mutation and leading to aneuploidy in IDH-wild-type GBM. Those are novel findings in GBM pathogenesis. Our results concur with those of others that the occurrence of CT points to underlying gene rearrangements, potential key drivers in gliomagenesis. According to Notta et al., SCNA from CT are essentially clonal, suggesting that these events are also sustained early in tumorigenesis [33]. Cancer genome instability drives tumorigenesis and tumor growth. It also fuels intercellular heterogeneity and tumor evolution. Genomic instability may drive branched evolution of tumor cells contributing to drug resistance. According to the Big Bang model proposed by Sottoriva et al. in colorectal cancer, there are uniformly high levels of intratumoral heterogeneity throughout the neoplasm [34]. Both public (clonal) and most private (subclonal) alterations occur early in tumor growth [34]. The Big Bang model implies punctuated clonal evolution (vs sequential clonal evolution) such as mutational bursts or CT. Such a model might explain GBM intra-tumoral heterogeneity seen at diagnosis. Even though the initial event in neoplastic transformation is still unknown, assessing spatial heterogeneity via multiple sampling and genomic profiling of distinct tumor regions within a GBM may help designing effective multidrug therapeutic strategies.

Patients and tumor samples
One hundred and twenty-three patients were selected according to the following criteria: 1) histopathological diagnosis of primary GBM, IDH-wild-type according to the 2016 WHO classification [2], 2) available freshfrozen tissue, 3) available clinical and survival data, and 4) written informed consent, with approval of the research ethics committee of Angers University Hospital. Freshfrozen tumor samples, with at least 60% of tumor cells, www.oncotarget.com were obtained from Angers Hospital biobank (CRB, Center for Biological Resources).

SNP array and copy number analysis
SNP array was performed on the 123 samples. Tumor DNA was extracted using the Nucleospin Tissue Kit (Macherey Nagel) and quantified using Qubit dsDNA BR Assay Kit (Life Technologies). Tumor DNA was hybridized with Infinium CytoSNP-850 K Illumina Beadchips (Illumina) according to the manufacturer's instructions. SNP arrays were scanned on an iScan (Illumina) and data were processed using the genotyping module in Genome Studio v2011.1 (Illumina) to calculate B-allele frequencies (BAF) and logR ratios. GAP method was used to call SCNA and to determine the number of chrs and the ploidy pattern (near-diploid, near-tetraploid) for each tumor [35]. The GAP method relies on two parameters to estimate tumor ploidy: 1) the DNA index for a tumor genome estimated as an average copy number in a genome divided by 2, and 2) the chromosome counts in a tumor genome estimated as a sum of copy numbers at pericentric regions of each chromosome arm. The calling of SCNA was based on the analysis of logR and allelic ratios [36]. The purity of the samples was assessed using BAF values. SNP array data are deposited in public repository ArrayExpress under the accession number E-MTAB-6371.

CT identification
CT events were detected using segmented data (LogR ratio, BAF value) from the SNP arrays. They were identified following the three criteria described by Korbel and Campbell [32]: 1) There were at least 10 genomic rearrangements per chr arm with such rearrangements occurring in no more than 4 chrs in a given sample, 2) There was a clustering of breakpoints, 3) There was interspersed loss and retention of heterozygosity with no more than two to three different copy number states (except for focal amplification or homozygous deletion (involving key cancer genes)).
After manual screening according to the abovementioned criteria, a validation analysis was performed on segmented data using CTLPScanner, a web server for detection of CT patterns [37]. Scanner parameters and thresholds were as follows: genome assembly = hg19; copy number status change times ≥ 10; log10 of likelihood ratio ≥ 6; minimum segment size = 5 Kb.

Statistical analyses
The khi² test was used to compare the distribution of categorical variables and unpaired t-test or Mann-Whitney test was used to compare continuous variables. Overall survival (OS) was defined as the time between histopathologic diagnosis and death or last followup. Patients who were still alive at last follow-up were considered as a censored event. Progression-free survival (PFS) was defined as the time between histopathologic diagnosis and recurrence or last follow-up. Patients who were recurrence-free at last follow-up were considered as a censored event. In all analyses, a p value < 0.05 (twosided) was considered statistically significant.

Author contributions
BB performed the Sanger sequencing, analyzed all the data and wrote the manuscript. AR supervised the study and wrote the manuscript. FD and MABR performed