Amplified and Homozygously Deleted Genes in Glioblastoma: Impact on Gene Expression Levels

Background Glioblastoma multiforme (GBM) displays multiple amplicons and homozygous deletions that involve relevant pathogenic genes and other genes whose role remains unknown. Methodology Single-nucleotide polymorphism (SNP)-arrays were used to determine the frequency of recurrent amplicons and homozygous deletions in GBM (n = 46), and to evaluate the impact of copy number alterations (CNA) on mRNA levels of the genes involved. Principal Findings Recurrent amplicons were detected for chromosomes 7 (50%), 12 (22%), 1 (11%), 4 (9%), 11 (4%), and 17 (4%), whereas homozygous deletions involved chromosomes 9p21 (52%) and 10q (22%). Most genes that displayed a high correlation between DNA CNA and mRNA levels were coded in the amplified chromosomes. For some amplicons the impact of DNA CNA on mRNA expression was restricted to a single gene (e.g., EGFR at 7p11.2), while for others it involved multiple genes (e.g., 11 and 5 genes at 12q14.1–q15 and 4q12, respectively). Despite homozygous del(9p21) and del(10q23.31) included multiple genes, association between these DNA CNA and RNA expression was restricted to the MTAP gene. Conclusions Overall, our results showed a high frequency of amplicons and homozygous deletions in GBM with variable impact on the expression of the genes involved, and they contributed to the identification of other potentially relevant genes.

Current availability of large-scale whole genome and gene expression profiling (GEP)-arrays provides an invaluable tool for detailed delineation of those genes involved in recurrent CN alterations, and rapid assessment of their impact on the expression levels of the involved genes. Gain-of-function mutations and both silencing and other type of mutations, may lead to oncogene activation and loss of function of tumor suppressor genes, respectively. Similarly, gene amplification and homozygous deletion may also contribute to the development of the tumoral phenotype. Consequently, detailed analysis of the recurrently involved genes in amplicons and homozygous deletions is a particularly attractive approach for identification of relevant targeted genes. Such analysis would be particularly informative when combined with parallel assessment of the impact of these alterations on the levels of expression of the candidate genes, in paired DNA and mRNA samples.
Previous array-based studies in which CN alterations are related to GEP have contributed to the identification of cancer associated genes relevant to GBM (e.g. MYCN, PIK3CA, CCND2, KRAS, CHD5, CXCL12, PTER, LRRN6C, ERRFI1 and TACC3, and at the same time they have confirmed the pathogenic role of other known genes (e.g. EGFR) [8,22,23,24,25,26]. More recently, The Cancer Genome Atlas (TCGA) research network [8] has further reported occurrence of homozygous deletions of the NF1 and PARK2 genes, and amplification of the AKT3 gene -less frequently also of the FGFR2 and IRS2 genes-, in a series of 206 GBM. Other candidate genes for which different mechanisms of alteration (e.g. epigenetic silencing) have been reported in GBM, include the RBBP5 gene (a member of the RB pathway) amplified at 1q32 [23], the CXCL12 (CXCR4 ligand involved in chemoattraction and tumor invasion) and the HK1(a member of the hexokinase family, known to regulate apoptotic pathways) genes [25]. Nevertheless, high-resolution analyses of somatic CN alterations from a large series of cancer specimens (n = 3,131) including GBM, identified 158 regions which are frequently altered across multiple cancer types, many of which could not be explained by previously known cancer-associated genes [27]. Despite all the above, currently there is limited information [8,23,24,25] about the impact of DNA CN alterations on gene mRNA expression levels in GBM..
In order to identify potentially targeted relevant genes, here we investigated the impact of CN alterations on the expression profile of those genes recurrently involved in amplicons and homozygous deletions in GBM. Our results show that most genes for which a high correlation was observed between CN alterations and gene expression levels, are coded in those chromosomal regions for which amplicons were detected, pointing out the potential role of several genes coded in chromosomes 12q14(e.g. RAP1B, MDM2 andGRIP1), 4q12 (e.g. TMEM165, FIP1L1 and EXOC1), in addition to the EGFR gene, in GBM. Conversely, the MTAP gene coded in chromosome 9p21 was the only gene involved in homozygous deletions whose expression levels showed a significant correlation with the CN status.

Patients and samples
A total of 46 caucasian patients diagnosed with primary GBM in the absence of other known genetic disorders (except for a case -G23-who had a von Willebrandt disease) who were admitted to the University Hospital of Coimbra (Coimbra, Portugal), were included in this study; 21 were males and 25 females with a mean age of 62613 years (range: 30 to 84 years) ( Table 1).
In the present study, only patients with magnetic resonance imaging (MRI), clinical evidence of disease and a histologically confirmed diagnosis of GBM based on the World Health Organization (WHO) criteria [28], were considered.. Other criteria used for patient inclusion in the study were: i) availability of enough highly-infiltrative (.75%) tumor tissue for genetic studies, and; ii) informed consent to participate in the study given by the patient. The study was approved by the University Hospital of Coimbra Ethics Committee, according to the Declaration of Helsinki protocol.
In addition to a tumor tissue specimen, paired peripheral blood (PB) samples were also collected from each patient at diagnosis.
For every tumor sample, representative parts of fresh tumor tissues obtained by surgical resection were immediately (,30 min) snap-frozen in liquid nitrogen and stored at 280uC until used for interphase fluorescence in situ hybridization (iFISH), GEP and single-nucleotide polymorphism (SNP)-array studies. Prior to these studies, a section cut from the tissue block was assessed by conventional histopathological procedures, to estimate its tumor cell contents. Specimens with $75% tumor cells, in the absence of significant contamination by normal brain parenchyma and tumor necrosis, were selected for further DNA and RNA extraction, as well as for iFISH studies.

Identification of CN alterations by SNP-arrays
For the investigation of CN alterations by SNP-arrays, DNA from frozen tumor tissue and their paired fresh PB samples was purified using the QIAamp DNA Mini Kit (Qiagen, Valencia, CA, USA) according to the instructions of the manufacturer. DNA yield and purity were determined with a NanoDrop-1000 spectrophotometer (Nano-Drop Technologies Inc., Wilmington, DE, USA) and DNA integrity was evaluated by conventional electrophoretic procedures in a 1% agarose gel. Briefly, total DNA was digested with restriction enzymes and ligated to the corresponding adaptors, following conventional Affymetrix procedures (Affymetrix Inc., Santa Clara, CA, USA). A generic primer was used in triplicate to amplify adaptor-ligated DNA fragments, through a polymerase chain reaction (PCR). After hybridization with the sample DNA, the chips were washed, labeled with streptavidin-phycoerythrin and scanned using a GeneChip Scanner 3000 (Affymetrix Inc.).
Two different SNP-array chips were used for CN analysis: 1) the GeneChip Human Mapping 500K Array Set, which provides information according to NCBI/hg17 assembly about .500,000 SNPs (262,264 SNPs in the Nsp array and 238,304 SNPs in the Sty array), was applied for the study of 23 GBM, and; 2) the Genome-Wide Human SNP Array 6.0, which contains probes for 906,600 SNPs and 945,826 non-polymorphic probes featuring a total of .1.8 million probes (Affymetrix Inc.), was used in the other 23 GBM (Table 1). Data about a total of 500,568 and 906,600 DNA probes was obtained in duplicate for paired tumor and normal PB DNA samples for each array, and it was analyzed with the Console Genotyping software (version 3.0.2; Affymetrix Inc.). In addition, the dChip 2010 software (http //www.dchip.org; Dana Farber Institute, Harvard, MA, USA) was used to calculate CN values and plot them according to chromosomal localization. Only common SNP probes between the two types of SNP-arrays (n = 481,622) were used in the analysis. Cut-off values #1.30 and $2.50 (arbitrary units) observed in tumor samples versus those obtained in the paired normal PB DNA samples, were used to establish CN losses and gains, respectively. Amplification and homozygous deletions were defined based on CN cut-off values obtained for DNA tumoral tissue of .5 and ,0.8 PB DNA copies (arbitrary units), respectively.

Gene Expression Profiles (GEP)
In a subgroup of 23 tumors, total RNA was isolated from freshly-frozen tumor tissue samples in two steps, using TRIzol (Invitrogen Life Technologies, Carlsbad, CA, USA) and the RNeasy Mini Kit (QIAGEN). The integrity and purity of the extracted RNA were determined using a microfluidic electrophoretic system (Agilent 2100 Bioanalyzer, Agilent Technologies, Palo Alto, CA, USA) and GEP were analyzed with the Gene Chip Human Genome U133 Plus 2.0 Array (Affymetrix Inc.) according to the instructions of the manufacturer, through the one-cycle cDNA synthesis kit and the Poly-A RNA gene chip control kit (Affymetrix Inc), as previously reported [29]. Datafiles containing expression values were normalized _Robust Multi-array average expression measure (RMA)_ and analyzed using the R (version 2.7.1; http://www.r-project.org) and Bioconductor software tools (http://www.bioconductor.org).

iFISH studies
Confirmatory iFISH studies were performed in every case according to previously described methods [30], using a set of commercial dual-color fluorescence labelled probes obtained from Vysis, Inc. (Downers Grove, IL, USA) and Q-BIOgene (Carlsbad, CA, USA), as previously described in detail [29,30]; these probes included sequences for the TP73, ANGPTL1, EGFR, ELN, TES, p16, ABL1, PTEN, RB1, TP53, ZNF44, GLTSCR1, and BCR genes (Supplementary table S2). An Axioscope fluorescence microscope equipped with a 6100 oil objective (Zeiss, Göttingen, Germany) was used to count the number of hybridization spots per nuclei ($200 nuclei/slide). Only those spots with a similar size, intensity and shape in non-overlapping nuclei, were counted; doublet signals were considered as single spots. Briefly, gains and losses of specific chromosomal regions were considered to occur when $5% and $10% of the nuclei showed an increased and decreased number of fluorescent signals (spots) with respect to normal diploid cells, respectively. Specimens were considered to carry amplification of the EGFR gene when .10% of the tumor cells exhibited either an EGFR:CEP7 ratio .2 or multiple tight clusters of hybridization signals for the EGFR gene probe. Homozygous deletion was defined as $5% of tumor nuclei with centromeric probe signals in the absence of signals for the locus specific probe.

Real Time RT-PCR validation of microarray-based mRNA levels
Microarray mRNA expression levels of four relevant genes (BCAS2, FIP1L1, EGFR, and XIST) coded in 4 different chromosomes (chromosomes 1, 4, 7 and X) were validated by an independent Quantitative Real-Time RT-PCR assay in a total of 14 GBM samples (G8, G10, G14, G17, G30,G34, G37, G40, G42, G44, G45, G46, G52 and G53). For this purpose the SuperScript III first-strand Synthesis System from Invitrogen (www.invitrogen.com) and both the LightCycler carousel-based system and the LightCycler TaqMan Master chemistry (Roche,Mannhein, Germany) were used, according to the instructions of the manufacturers.

Statistical analyses
The statistical significance of differences observed between groups was assessed by the Student T and the Mann-Whitney U tests, for parametric and non-parametric (continuous) variables, respectively; for qualitative variables, the X 2 test was used (SPSS software, SPSS 15.0, SPSS Inc, Chicago, IL, USA). Survival curves were plotted according to the method of Kaplan and Meier, and the log-rank test was used to assess the statistical significance of differences in overall survival between groups of patients (SPSS software). P-values,0.05 were considered to be associated with statistical significance.
In order to investigate the impact of CN alterations on GEP, the relationship between DNA CN values and GEP mRNA levels was assessed for 12,445 genes investigated in common with both the SNP-arrays and the GEP arrays, using the Pearson correlation (Supplementary table S1). The correlation between real Time RT-PCR and microarray-based mRNA levels of the BCAS2, FIP1L1, EGFR, and XIST genes was assessed by the Spearman correlation.
By far, chromosome 7p11.2 was the most frequently amplified chromosomal region (n = 21/46 patients; 46%); two additional cases showed amplicons at chromosome 7q ( Figure 1 and Supplementary Table S3). In turn, a high frequency of amplicons was also noted in the long arm of chromosome 12 with recurrent involvement of the 12q14.  Table S3).
iFISH analysis systematically confirmed the presence of the above described amplicons and homozygously deleted chromosomal regions (Supplementary Table S2); although variable patterns of numerical changes were revealed by iFISH at the single cell level, in line with previous observations [31]; such intratumoral iFISH heterogeneity was not detected by the SNP CN profiles as exemplified in Supplementary Table S5 Table S3). Among other, these included the CYP27B1, METTL1, FAM119B, TSFM and AVIL genes at chromosome 12q14.1 in 8/ 46 patients (17%), and the CDK4 and AGAP2 genes in 7 GBM (15%). In addition, amplification of the 12q13.3 chromosomal region involved the GEFT and B4GALNT1 genes in 6 patients (13%) and recurrent amplification of chromosome 12q15 involved the MDM2 gene together with the SLC35E3 and CPM genes in 4 cases (9%).
The amplified chromosomal region at 1q32.1 (n = 3/46 tumors; 7%) showed a length of between 585 Kb and 864 Kb, and it systematically included the MDM4 gene together with another seven genes (SOX13, ETNK2, REN, KISS1, GOLT1A, PLEKHA6 and PIK3C2B (Figure 1 and Supplementary Table S3). Another 10 chromosomal regions from chromosomes 1, 11 and 17 were amplified in only one tumor each and they involved a variable number of genes (between 1 and 37 genes) (Supplementary Table S3).
The most frequent homozygous deletion found involved chromosome 9p21.3 from the 21,978,443 bp to the 22,119,128 bp position. This segment included the CDKN2A gene in 24 patients (52%), together with the CDKN2B and CDKN2BAS genes in 22 of them (48%). The MTAP (12 cases; 26%), LOC554202 (8 tumors; 17%), ELAVL2 (5 GBM; 11%), and TUSC1 (5 cases; 11%) genes were also frequently lost. Noteworthy, in a subset of 7 of the former 24 cases, homozygous deletion of the 9p21.3 segment extended telomericly to the 21,186,931 bp position including between three (n = 4 cases) and four (n = 3) genes more (IFNA7, IFNA4 and IFNA17 without or with the KLHL9 gene, respectively) ( Figure 2 and Supplementary Table  S4). By contrast, the localization of the deleted regions in chromosome 10 was highly heterogeneous extending from the 13,320,848 bp to the 134,174,753 bp positions, with an overall length of between 4to 8,726 Mbp; among these cases, the PTEN gene was homozygously deleted in 3 tumors (Figure 2 and Supplementary Table S4). Another two genes _DNA2 and  Table S4). Of note, the RB1 tumor suppressor gene coded at chromosome 13q14.2 was homozygous deleted together with the RCBTB2 gene in only one tumor (G97) from our series (Supplementary Table  S4).  correlation was found between CN alterations and gene expression levels ( Table 2). Of note, a high correlation was observed between the microarray expression levels and quantitative RT-PCR analysis of the mRNA levels of 4 selected relevant genes (BCAS2, FIP1L1, EGFR and XIST) with Spearman correlation coefficients of 0.6, 0.8, 0.8 and 0.9, respectively.

Copy number alterations and gene expression levels
Those genes whose expression was mostly impacted by the most frequent CN alterations were the RAP1B, TSFM, CYP27B1, METTL1, AVIL, CDK4 and FAM119B genes in chromosome 12q14, the EGFR gene in chromosome 7p11 and the TMEM165, FIP1L1, CLOCK, SRD5A3 and SCFD2 genes in chromosome 4q. As mentioned above the MTAP gene was the only gene whose expression was highly correlated to the occurrence of homozygous deletion of chromosome 9p21.
From the prognostic point of view, none of the amplified and homozygous deletions showed a clear impact on patient survival, except for the amplification of the PDGFRA gene in chromosome 4q, which was associated with a significantly shorter overall survival (median overall survival of 2 vs 13 months for the PDGFRA amplified vs non-amplified cases ; p = 0.0002) (Figure 3).

Discussion
Identification of genetic markers involved in the oncogenic mechanisms and molecular pathways driving GBM, still remains a challenge. Among other approaches, detailed characterization of amplicons and homozygous deletions provides a useful tool for the screening and identification of candidate genes, despite pathogenic mechanisms (e.g. gene mutation, epigenetic silencing through gene methylation and altered expression microRNA) exist which are not directly related to CN alterations. Several techniques have been previously used for the identification of CN alterations in GBM, including high-resolution SNP-arrays [23,32,33,34]. Through such approaches, CN gains of chromosome 7p11.2, together with del(9p21) and monosomy 10/del(10q), have been commonly observed in GBM, together with other less frequent abnormalities [10,12,35]. Although the information from such analyses is useful for the identification of both the altered chromosomal regions and the genes they contain, valuable insights into the role of relevant involved genes requires further assessment of the impact of genomic aberrations on gene expression. In this study, we integrated genomic and transcriptional data from paired DNA and RNA samples from the same tumors, in order to identify those genes involved in amplicons and homozygously deleted chromosomal regions which show a parallel change in their mRNA expression levels.
Analysis of whole-genome CN alterations by SNP-arrays confirmed the presence of previously reported recurrent genomic alterations in GBM [2,3,5], which were further validated here by iFISH. Overall, almost every tumor in our cohort showed CN alterations in multiple chromosomes, but their frequency and extent varied significantly among different tumors, confirming the genetic complexity of GBM [31]. Interestingly however, those regions recurrently affected by CN alterations consistent with gene amplification, were restricted to a few chromosomes (chromosomes 1q, 4q, 7p and 12q). Similarly, recurrent homozygous deletions only involved chromosomes 9p21 and 10q23. Since many genes involved in gliomagenesis often reside within amplicons and/or homozygously deleted chromosomal regions, we searched for candidate genes encoded in these regions for which a correlation existed between the DNA CN values and mRNA expression levels in the same tumor sample. Noteworthy, virtually all genes which showed a high correlation (R 2 .0.70) between DNA CN values and mRNA levels corresponded to genes coded in those six chromosomes where amplicons were recurrently observed. Amplified genes included several genes (MDM4, PDGFRA, EGFR, CDK4, and MDM2) whose amplification has been previously associated with the pathogenesis of GBM [8,18,36,37]; in addition, other candidate genes were also found to be amplified at both the DNA and RNA levels and thus, to be potentially relevant in the pathogenesis of GBM.
In detail, the EGFR oncogene whose amplification defines a subset of GBM [3,38,39], was the only gene coded in the 7p11.2 amplicon for which a significantly high correlation between DNA CN values and gene expression levels was detected; although this amplicon contained other bystander genes that may be coamplified with EGFR (e.g. LANCL2 and GASP) due to their genomic proximity [14], amplification of all such genes showed limited impact on gene expression levels. These results support and reinforce the critical role of EGFR in the pathogenesis of a significant fraction of GBM, through activation of the RAS pathway [40,41].
Amplification of chromosome 12q13-15 was also found in a significant proportion of primary GBM, in line with previous reports [17]. In contrast to 7p11.2 amplicons, 12q13-15 amplicons typically involved multiple genes whose expression was increased at the RNA level in parallel to the greater DNA CN values.. However, from all these genes listed in table 2, only AVIL, FAM119B, METTL1, CYP27B1 and TSFM were systematically amplified in tumors displaying 12q amplicons..Co-amplification of the CDK4 and MDM2 genes at the 12q13-15 amplicon is frequently observed in GBM and it has been previously suggested to confer a tumor growth advantage [42]. However, in our series this amplicon showed no impact on the mRNA gene expression levels of these two genes. By contrast, relatively little is known about the role of other co-amplified and overexpressed genes which are coded in chromosome 12q13-15, even if several of them are also amplified in other tumors, e.g. in lung cancer [43]. From these genes, special attention should be paid to the AVIL and CYP27B1 genes. AVIL encodes for advillin, a member of the gelsolin/villin family of actin regulatory proteins which is almost exclusively expressed by peripheral sensory neurons, and that has been recently identified as a new candidate driver gene in GBM [44,45]. In turn, CYP27B1 (P450 Cytochrome 25-Hydroxyvitamin D3 1,a-Hydroxylase) catalyzes the conversion of calcidiol to calcitriol, the most active vitamin D metabolite, involved in cell proliferation with both anti-proliferative and cell differentiating effects [46,47,48]. Conversely, the role of FAM119B (a gene of unknown function, which has been associated to multiple sclerosis) [49], METLL1 _a nuclear protein that catalyzes the formation of N(7)-methylguanine at position 46 (m7G46) in tRNA, inactivated in response to agonists of the PI3-kinase pathway or the classical MAP kinase cascade_, and TSFM (a gene that encodes the mitochondrial translation elongation factor EFTs) in GBM, deserves further investigations.
Amplification of chromosome 4q12 has been previously described to involve the PDGFRA gene in 8-15% of GBM [21,50], as also found here. However, CN alterations at 4q12 did not show a high correlation with PDGFRA mRNA levels from the same tumor. Conversely, expression of other amplified genes coded at 4q12 in the vicinity of the PDGFRAgene (e.g. FIP1L1 and SCFD2), appeared to be significantly modulated by the 4q12 amplicon, suggesting their potential relevance in GBM. In this regard, it should be noted that the FIP1L1 protein coded by the FIP1L1 gene functions as a component of the cleavage and polyadenylation specificity factor (CPSF), which participates in processing of mRNA. Although, it has been observed that fusion of the FIP1L1 gene to the PDGFRA gene generated by interstitial 4q12 deletion, results in a constitutively activated FIP1L1-PDGFRA fusion protein with tyrosine kinase activity in chronic eosinophilic leukemiar [51] the role of the FIP1L1 gene in gliomagenesis, remains to be dilucidated. Other potentially relevant genes coded at chromosome 4q12 whose CN alterations mostly impacted on their mRNA expression levels included the TMEM165 and the CLOCKgenes. TMEM165 (also named TPARL), encodes a putative transmembrane 324 amino acid protein whose cellular functions are unknown, although intronic splice mutations of the protein have been related to congenital disorders of glycosylation [52]. Of note,T-MEM165 has been found to be up-regulated in invasive GBM cells with transcriptional differences between these and the other core tumor cells, supporting a role for this gene in tumor invasion [53]. In turn, theCLOCK protein contributes to activate transcription of the Period (PER1, PER2, and PER3) and Cryptochrome (CRY1 and CRY2) proteins two proteins that are involved in the circadian system. Deregulation of the circadian clock protein has been implicated in many types of cancer, in both animal and humans [54,55,56,57]. In a recent study on gliomas including GBM, tumor cell expression of PER1 and PER2 was significantly lower than in the surrounding normal/reactive cells [58], suggesting that deregulated expression of these two genes may result in disruption of the control of the normal circadian rhythm and a stimulatory effect on survival and proliferation of gliomas cells. However, the molecular mechanisms of genes controlling circadian rhythm in glioma cells have not been explored so far.
Independently of the precise impact of the molecular mechanisms involved, amplification at the 4q12 chromosomal region was associated with a very short patient overall survival.
Amplification of other chromosomal regions such as chromosome 1q32 was found in a relatively limited number of cases, in line with previous findings [3,8,24]., Controversial results exist as regards the impact of 1q32 amplicon on MDM4 and CNTN2, as independent versus combined targets for amplification [20]. Our results support an independent and more relevant role for the former gene as the target of 1q32 amplification, since a higher frequency of MDM4 amplification in the absence of involvement of CNTN2 was observed in our patients; despite this, in our series, 7 other candidate genes were amplified together with MDM4. These included the PIK3C2B gene involved in the PI3K/AKT signaling pathway, which was also amplified and overexpressed in other series of GBM [20], and the SOX13 gene, which has been reported to be up-regulated in oligodendrogliomas [59]. Despite this, none of the amplified genes at chromosome 1q32 was associated with simultaneously increased RNA expression, further investigations being required to confirm their relevance in the pathogenesis of GBM.
Commonly deleted segments at chromosome 9p21, almost systematically involve the CDKN2A tumor suppressor gene, in association with both the CDKN2B gene, and frequently also the MTAP gene [60,61,62,63,64], in line with our results. CDKN2A/B gene products are involved in the p53 pathway through a protein_p14 arf _ encoded by an alternate reading frame, which binds to the p53/mdm complex and inhibits mdm-mediated degradation of p53. Thus, homozygous deletion of the CDKN2A/B locus could affect both the Rb and p53 pathways [6,9]. Interestingly, it has been shown that MTAP can be lost independently of CDKN2A, which suggests that loss of MTAP may indeed play a role in tumor biology. Noteworthy, MTAP was the single homozygously deleted gene at chromosome 9p21 for which a high correlation was found between DNA CN values and its mRNA expression levels. These results point out the potential relevance of MTAP as a tumor suppressor gene. The MTAP protein is an enzyme that plays a major role in polyamine metabolism and that is essential for salvaging both adenine and methionine. This enzyme is expressed in all normal human tissues, but MTAP protein deficiency or MTAP gene deletion have been previously found in several tumors, including gliomas [63]. This is particularly relevant if we consider that the deleted segments at 9p21 encompassed, not only the CDKN2A/B tumor suppressor genes, but also numerous other potentially relevant genes, such as the ELAVL2 (its absence in Drosophyla causes multiple structural defects and hypotrophy of the CNS), KLHL9 [encodes a substratespecific adapter of a BCR (BTB-CUL3-RBX1) E3 ubiquitinprotein ligase complex, required for mitotic progression and cytokinesis], and IFNA (IFNA7, IFNA4, and IFNA17) genes. Monosomy 10 in association or not with del(10q) is a frequent finding in GBM. Despite multiple chromosomal segments were homozygously deleted in a small fraction of our cases, they rarely involved the same chromosomal regions and none of the deleted genes encoded in these areas, including the PTEN gene, showed a high correlation between its DNA CN values and mRNA gene expression levels.
In summary, here we confirm the high frequency of gene amplification in GBM, which mainly involves the 7p11.2, 12q13-15, and to a less extent also the 4q12 and 1q32 chromosomal regions. Most interestingly, amplification of those genes coded in these chromosomal segments showed a variable impact on their mRNA levels, depending on the specifically targeted gene. Conversely, recurrent homozygous deletions were restricted to chromosome 9p21 and to multiple variable segments of chromosome 10q, the MTAP gene being the only gene whose mRNA levels were significantly affected by such homozygous deletions.

Supporting Information
Table S1 Number of probes analyzed with the two different types of SNP-arrays (6.0 and 500K SNP-arrays) and the GEParray (U133 Plus 2.0), grouped according to their chromosomal localization. DNA CN alterations were assessed in all GBMs samples (n = 46) while mRNA expression levels were evaluated in a subset of 23 of these tumors. (DOC)