Copy Number Profiling of Brazilian Astrocytomas

Copy number alterations (CNA) are one of the driving mechanisms of glioma tumorigenesis, and are currently used as important biomarkers in the routine setting. Therefore, we performed CNA profiling of 65 astrocytomas of distinct malignant grades (WHO grade I–IV) of Brazilian origin, using array-CGH and microsatellite instability analysis (MSI), and investigated their correlation with TERT and IDH1 mutational status and clinico-pathological features. Furthermore, in silico analysis using the Oncomine database was performed to validate our findings and extend the findings to gene expression level. We found that the number of genomic alterations increases in accordance with glioma grade. In glioblastomas (GBM), the most common alterations were gene amplifications (PDGFRA, KIT, KDR, EGFR, and MET) and deletions (CDKN2A and PTEN). Log-rank analysis correlated EGFR amplification and/or chr7 gain with better survival of the patients. MSI was observed in 11% of GBMs. A total of 69% of GBMs presented TERT mutation, whereas IDH1 mutation was most frequent in diffuse (85.7%) and anaplastic (100%) astrocytomas. The combination of 1p19q deletion and TERT and IDH1 mutational status separated tumor groups that showed distinct age of diagnosis and outcome. In silico validation pointed to less explored genes that may be worthy of future investigation, such as CDK2, DMRTA1, and MTAP. Herein, using an extensive integrated analysis, we indicated potentially important genes, not extensively studied in gliomas, that could be further explored to assess their biological and clinical impact in astrocytomas.

pathways (Wen and Kesari 2008). Using an integrated genomics approach, the TCGA (The Cancer Genome Atlas) consortium described four different subtypes of GBMs (classical, mesenchymal, proneural, and neural) (Verhaak et al. 2010). Alterations (expression, mutation, and/or copy number) of the genes TP53, IDH1, PDGFRA, EGFR, NF1, and CDKN2A were considered the most important events to distinguish these four subtypes. Additional analysis of glioma-CpG island methylator phenotype (G-CIMP) positive and G-CIMP negative tumors has shown that DNA methylation patterns strongly correspond to the status of IDH1 mutation (Noushmehr et al. 2010). Recently, hotspot TERT promoter gene mutations have been found in gliomas, with the highest incidence in GBMs (60%) (Vinagre et al. 2013;Heidenreich et al. 2014;Killela et al. 2013;Batista et al. 2016). These mutations generate a de novo binding site for GABPA transcription factor, which ultimately leads to high TERT expression (Bell et al. 2015). More recently, a large cohort study described five glioma groups based on 1p/19q codeletion, IDH1/2 and TERT promoter mutational profile, with important clinical impact, with the "triple-negative" group or the only TERTmutated group exhibiting a higher mortality risk (Foote et al. 2015;Eckel-Passow et al. 2015).
Therefore, the aim of this study was to characterize the genomic profile of 65 Brazilian astrocytomas, using aCGH and MSI, as well as to associate these data with the mutational status of the TERT promoter and IDH1 genes, and clinico-pathological features of the patients. Additionally, by extending these analyses using in silico approaches, this study aimed to describe potentially important molecular subgroups with clinical impact and targets that could be the object of future investigation.

Patients
Sixty-five frozen tissue specimens comprising pilocytic astrocytomas (n = 7), diffuse astrocytomas (n = 9), anaplastic astrocytomas (n = 7), and GBMs (n = 41 primary and one secondary GBMs) were evaluated. Overall, there were 62 primary tumors and three recurrences (with the matched primary tumor also present in our analysis): one pilocytic astrocytoma that recurred after the first surgery, one diffuse astrocytoma that progressed to GBM after the surgery, and one GBM that recurred.
Histologic review of the slides was performed by two neuropathologists (A.P.B. and G.C.A.) to confirm the diagnosis, and to select the samples with . 75% of neoplastic cells and an absence of necrosis.  DNA was isolated from frozen tissue and the peripheral blood of each patient and used for further analysis. Clinical data for each patient was obtained, and the summary of the characteristics is shown in Table 1. The present study was approved by the Barretos Cancer Hospital Ethical Committee (ID 408/2010).

DNA isolation
The DNA from patients' blood was isolated using a QIAmp DNA blood Mini Kit (Qiagen), and DNA from frozen tumor tissue was isolated using a DNeasy Blood and Tissue Kit (Qiagen) according to the protocols provided by the supplier. The 260/280 and 260/230 ratios were determined by NanoDrop (Thermo Scientific) and the DNA was quantified using Quant-IT PicoGreen dsDNA (Invitrogen), using the supplier's protocol.
Array-CGH Two-color 60 K array Comparative Genomic Hybridization (aCGH) was performed using the default protocol published by Agilent Technologies (Agilent Oligonucleotide Array-Based CGH for Genomic DNA Analysis Enzymatic Labeling for Blood, Cells, or Tissues, protocol v. 7.2, published in July 2012) as previously described (Bidinotto et al. 2015). DNA of each patient's blood was used as control, in order to exclude copy number variations. AluI and RsaI restriction enzymes were used to digest 400 ng of both tumor and blood DNA, which was then incubated with random primers. Blood DNA was labeled with cyanine-3 (Cy3), whereas tumor DNA was labeled with cyanine-5 (Cy5). Equal quantities of Cy3-and Cy5-labeled DNA was hybridized into Agilent Human Genome CGH 8 · 60 K microarray slides overnight, and washed according to the supplier's default protocol. The slides were scanned and decoded by the software Feature Extraction v. 10.7 (Agilent Technologies), using the protocol CGH_107_Sep09. The signal intensities were log2 transformed, and the spots were mapped in the most recent version of the human genome (hg19). The data were Lowess normalized and smoothing corrected. The data were CBS segmented. The low-level copy gains/losses threshold value was considered 0.1, and the moderate-to-high gene amplification/homozygous deletion threshold value was considered 0.7, in five consecutive probes. aCGH data of the 65 Agilent arrays can be accessed using the Gene Expression Omnibus (GEO) series accession number GSE71538 (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE71538).

Bioinformatics analysis
Genome plots of each case were generated and visually inspected. Next, frequency plots were generated, and genomic regions were considered as frequently altered when they were gained or lost in at least 30% of the same tumor type. Survival analysis was performed to each altered region found in GBM samples. Kaplan-Meier plots were done considering the regions statistically significant (P , 0.05) in log rank tests. In order to validate our findings, we extended our aCGH analysis using publicly available GBM data on the TCGA Research Network dataset (http:// cancergenome.nih.gov). The set consisted of 498 GBMs CNV MSKCC level 1 data of the Agilent Human Genome CGH 244K microarray platform, and it was subjected to the same CNA detection algorithms performed in our samples. Genomic regions frequently amplified or deleted were considered for further in silico analysis using the professional version of the compendium of cancer transcriptome profiles, Oncomine (Compendia Bioscience, Ann Arbor, MI). Eight GBM expression datasets (totalizing 1489 tumors and brain normal samples) were selected from the Oncomine database. The expression of potentially relevant genes was analyzed in these datasets by selecting the genes that were present in frequently amplified or deleted regions of our aCGH experiments. The genes considered relevant presented amplification in our experiment and overexpression in the Oncomine datasets, or homozygous deletion in our experiment and loss of expression in the Oncomine datasets.
The mRNA expression of these relevant genes was further assessed on the TCGA Research Network GBM expression dataset (http:// cancergenome.nih.gov, n = 542 GBMs) from the Oncomine database. The expression of the genes was categorized in terms of positive or n
negative for each patient, based on a median intensity of log2 mediancentered value. If the intensity value of the gene was greater than the median value considering the 542 GBMs, it was considered positive; otherwise, the gene was considered negative in the patient. Furthermore, the expression of each gene was correlated to the overall survival of the patients. Additionally, correlation studies were performed. The genes were PCA ordered, and Pearson correlation coefficientwas assessed on theTCGA dataset. Correlations with P , 0.05 were considered statistically significant.
Finally, the potentially relevant genes were clustered by biological importance and canonical pathways, using the DAVID v6.7 bioinformatics tool (The Database for Annotation, Visualization, and Integrated Discovery) (Huang et al. 2007).

Microsatellite instability (MSI)
The MSI analysis of tumor and blood DNA of the patients was performed according to methodology previously published (Viana-Pereira et al. Campanella et al. 2014Campanella et al. , 2015a. Briefly, five markers (NR27, NR21, NR24, BAT25, and BAT26) were PCR multiplexed, and the products were separated using an ABI Prism 3500 genetic analyzer (Life Technologies). The results were analyzed with GeneScan Analysis software, version 3.7 (Life Technologies).

TERT mutation analysis
The hotspot mutations analysis of the TERT promoter gene was performed by PCR followed by direct Sanger sequencing (Vinagre et al. 2013;Campanella et al. 2015b;Batista et al. 2016). Briefly, the TERT promoter region was amplified by PCR using the primers: 59-AGTGG ATTCGCGGGCACAGA-39 (forward) and 59-CAGCGCTGCCTGA AACTC-39 (reverse), leading to a 235 bp PCR product containing the C228T and C250T mutations. Amplification PCR was performed with an initial denaturation at 95°for 15 min, followed by 40 cycles of 95°d enaturation for 30 sec, 64°annealing for 90 sec, and 72°elongation for Figure 3 Frequency plot representing the gained and lost regions in glioblastomas.
n 30 sec, and 72°final elongation for 7 min. Amplification of PCR products was confirmed by gel electrophoresis. Sequencing PCR was performed using a Big Dye terminator v3.1 cycle sequencing ready reaction kit (Applied Biosystems) and an ABI PRISM 3500 xL Genetic Analyzer (Applied Biosystems).

IDH1 mutation analysis
The analysis of hotspot mutations of IDH1 (exon 4) was performed by PCR followed by direct sequencing. Briefly, the IDH1 region of interest was amplified by PCR using the primers: 59-CGGTCTTCAGAGAAG CCATT-39 (forward) and 59-CACATTATTGCCAACATGAC-39 (reverse). An amplification PCR reaction was performed in a total volume of 15 ml, comprising: 1 ml of DNA, 1 · buffer solution, 2 mM MgCl 2 , 200 mM of each dNTP, 0.3 mM of each set primer, and 0.5 U Taq DNA polymerase (Invitrogen), and was performed in a Veriti 96-well Thermal Cycler with an initial denaturation at 95°for 10 min, amplified for 40 cycles of denaturation at 95°for 45 sec, annealing at 58°for 45 sec, and extension at 72°for 45 sec, and a final extension at 72°for 10 min. Amplification of PCR products was confirmed by gel electrophoresis. Sequencing PCR was performed using a Big Dye terminator v3.1 cycle sequencing ready reaction kit (Applied Biosystems) and an ABI PRISM 3500 xL Genetic Analyzer (Applied Biosystems).

Data availability
The authors state that all data necessary for confirming the conclusions presented in the article are represented fully within the article.
n  Microsatellite analysis MSI status was assessed in 55 cases (Table 4); MSI-H was only observed in four GBMs (11.1%), the remaining samples presenting microsatellite stable or MSI-L phenotypes (Table 4). One MSI case presented a high number of CNAs (total of 81 CNAs) and TERT mutation. Two other samples presented TERT mutation, were wild-type for IDH1, and presented a total of 23 and 24 CNAs, respectively. The remaining sample was wild-type for TERT and IDH1, and presented 25 CNAs.

Matched primary recurrence tumors
The molecular profile of the matched primary and recurrence tumors are summarized in Figure 4. The recurrence of pilocytic astrocytoma presented mutation in TERT. No other molecular differences were found when the primary pilocytic astrocytoma was compared to the recurrence (both presented only chr7q34 gain, with no other gene mutation found).
The primary diffuse astrocytoma presented mainly chr7, chr8q, and chr11q gains, as well as chrXp loss. At recurrence, it progressed to GBM, with the typical features of this tumor type described above (amplification of EGFR, gain at chr7, and losses at chr9p24.3-p21.1, chr10p15.3-q26.3, chr13, and chr22). ChrXp loss and IDH1 mutation were found in both tumors (primary and recurrence).
Finally, the GBM sample that recurred into GBM presented exactly the same molecular features, being the typical chromosomal characteristics presented above, and TERT mutated. All these samples were microsatellite stable.
Clinical impact of the molecular features Following the criteria recently described (Eckel-Passow et al. 2015), we separated the cases based on 1p/19q deletion and IDH1 and TERT promoter mutational status. We found that 1.8% of the cases presented the three alterations "triple positive," 22.8% presented mutation only in IDH1, 49.1% presented mutation only in TERT, and 26.3% did not present alteration in any of these markers and was considered to be "triple negative" (Table 5). Of note, survival curves show that the group of cases with mutation only in TERT presented lower survival than those presenting only IDH1 mutation (mean survival of 8.5 months vs. 29.2 months, respectively, P = 0.024 in log rank test), whereas "triple negative" cases presented a mean survival of 21.3 months ( Figure 5).
Nonsupervised hierarchical cluster analysis of the GBM CNA did not show any association with clinico-pathological features (data not shown). Log rank analysis of all altered regions across the GBM samples pointed to the correlation of EGFR amplification and/or gain of chr7 to better survival of patients (P , 0.05, Figure 6).
Moreover, we extended our findings to the expression level. The genes frequently found amplified/deleted in our GBM cases were investigated through a bioinformatics approach, using the compendium of cancer transcriptome profiles (Oncomine). Once the list of genes encompassed in the amplified or deleted regions of our GBM samples was generated, we inquired whether these genes had gain or loss of expression in eight other GBM expression datasets (totaling 1489 tumor and normal brain samples).
Considering these potentially important genes (Table 6), we found in the TCGA expression dataset that the loss of expression of IFNA13, IFNA21, IFNA6, IFNA8, IFNB1, IFNW1, or PTEN was correlated with poor survival, whereas loss of expression of GPNMB, IGF2BP3, ITGB8, or SEC61G was correlated with better survival (Table 7). Additionally, we found that there is an important positive correlation of expression among the genes IFNB1, IFNA21, IFNW1, IFNA14, IFNA4, CDKN2A,  IFNA7, IFNA5, MTAP, IFNA17, IFNA1, IFNA6, IFNA13, IFNA8  Triple positive represents 1p19q deletion + mutation in TERT promoter and IDH1; Triple negative represents none of the three alterations. Figure 5 Survival curve considering the patients presenting only IDH1 mutation, only TERT mutation, and neither mutation in IDH1 nor TERT, nor loss of 1p19q (triple negative).
Finally, DAVID analysis showed that there are several functional annotation clusters with a high enrichment score related to potentially important biological processes, such as posttranscriptional regulation of gene expression, regulation of translation, regulation of cell proliferation, and the transmembrane receptor protein tyrosine kinase signaling pathway ( Figure 9A). KEGG canonical pathways with a high number of genes include the Jak-STAT signaling pathway, alongside pathways related to the immune response, glioma, and prostate cancer ( Figure 9B).

DISCUSSION
In the present study, we performed a molecular characterization in order to describe the genomic alterations and mutation status of the key TERT and IDH1 genes in astrocytomas arising in the Brazilian population.
Overall, the CNAs found in our aCGH analysis correspond to the alterations found in the TCGA datasets. We found that 85.7% of the GBMs presented gain of whole chr7, and 33.3% presented amplification in the region 7p12.2-p11.2, in which EGFR is included. The PTEN tumor suppressor gene loci exhibited loss in 88.1% of our samples. From the 42 GBM samples, only one did not present alterations in chr7 and/or chr10, showing the importance of these loci in gliomagenesis. Log rank analysis showed that our patients presenting EGFR amplification and/or chr7 gain had better overall survival than patients that did not present these alterations. This finding is in accordance with Smith et al. (2001) and Verhaak et al. (2010), who subdivided the GBMs into four groups and found that patients with the 'classical' subtype, characterized by EGFR amplification and chr10 loss, presented better overall survival when they receive an intensive therapy (concurrent chemo-and radiotherapy or more than three subsequent cycles of chemotherapy).
By analyzing the TCGA expression dataset, we found several other genes correlated with overall survival. In order to determine the possible interference of the coexpression of the genes in these results, we performed correlation tests in potentially important genes in GBM development. More than half of the genes that were statistically significant were located on 9p22.1-p21.3. This locus a frequent target of homozygous deletion during gliomagenesis, and this event was observed in more than half of our patients. This region encompasses the CDKN2A tumor suppressor gene (p16 INK4a /p14 ARF /p15 INK4b locus), a potent regulator of the cell cycle (Li et al. 2011). Of interest, the chr13q region encompassing the RB1 gene presented a loss in 57.1% of our GBM samples. Previous studies have reported that homozygous deletion of p16 INK4a , CDK4 amplification, and loss of RB1 are almost mutually exclusive (Verhaak et al. 2010;Ohgaki and Kleihues 2009), and that these alterations are found in 50% of primary GBMs (Ohgaki and Kleihues 2009).
We have previously determined MSI status in 144 gliomas (71 children and young people and 73 adults). Of the 14 gliomas that were from patients of Brazilian origin, all of them were , 18 years of age (Viana-Pereira et al. 2011). Overall, a total of 13.2% of the samples presented MSI, in which the majority was pediatric (P = 0.02, Chi-square  Recurrent mutations in the promoter region of TERT gene, namely the c.-146:C . T and the c.-124:C . T mutations, were recently reported in several tumors, including melanomas, bladder, hepatocarcinoma, thyroid carcinomas, and gliomas (Huang et al. 2013;Vinagre et al. 2013;Killela et al. 2013;Horn et al. 2013;Bell et al. 2015;Batista et al. 2016). These mutations generate a consensus binding site for ETS/ TCF transcription factors (CCGGAA), resulting in increased activity of the TERT promoter and abnormal telomere size maintenance (Huang et al. 2013;Vinagre et al. 2013;Killela et al. 2013;Horn et al. 2013;Bell et al. 2015;Cancer Genome Atlas Research Network et al. 2015;Eckel-Passow et al. 2015;Koelsche et al. 2013). In accordance, we found a low percentage of pilocytic, diffuse, and anaplastic astrocytomas presenting mutations in the TERT promoter gene. Additionally, we found a high percentage of GBMs presenting either the mutation -124:G . A (52.6%) or -146:G . A (21.1%), which shows the importance of this mutation to GBM development, since it constitutively activates the TERT gene, supporting the maintenance of genomic integrity through telomere elongation (Heidenreich et al. 2014;Walsh et al. 2015).
By analyzing an important dataset of gliomas, Ceccarelli et al. (2016) described distinct glioma subgroups based on methylation and gene expression status, and correlated them with survival, grade, and age at diagnosis. Based on DNA methylation analysis, the authors described six clusters: three clusters presented IDH mutations and were enriched for low-grade gliomas, whereas the clusters with wild-type IDH were enriched for GBMs. In fact, we found 77.8% (7/9) of diffuse and 100% (7/7) of anaplastic astrocytomas presenting IDH1 mutation, whereas 92.9% (3/42) of GBMs were wild-type for IDH1 mutation, corroborating this data. Independent of the tumor grade, we also found a dramatic increase in survival in patients presenting IDH1 mutation (29.2 months), suggesting that this gene is an important biomarker, as the authors have previously found that IDH mutation was the main driver of the clusters (Ceccarelli et al. 2016).
Similarly, other comprehensive studies suggest that the combined analysis of the mutational status of TERT, IDH, and 1p/19q deletion had the ability to define the biological and clinical behavior of gliomas better than analysis based solely in histology (Foote et al. 2015;Eckel-Passow et al. 2015). When we performed this stratification in our samples, we found that the group presenting only TERT mutation had a dramatically reduced survival of 8.5 months vs. 29.2 months of only IDH1-mutated patients. This is consistent with recent data that showed an association of TERT mutation with poor survival, and that of IDH1 mutation with better survival (Eckel-Passow et al. 2015;Foote et al. 2015). Generally, mean age at diagnosis in our groups was also consistent with the literature, with elderly patients presenting only TERT mutation (Eckel-Passow et al. 2015).
Besides the alterations in the genes extensively studied in GBMs, there may exist some less-studied regions/genes that could help in the understanding of GBM development and/or could be potential targets in GBM treatment. To identify these genes, we selected those present in regions frequently amplified or deleted in our GBMs and exploring their expression in Oncomine datasets. Hodgson et al. (2009) assessed the gene expression data of TCGA-derived GBMs and found overexpression of genes related to cellular assembly and organization and, among other genes, the authors found CDK2 (located at 12q13), which was found amplified in four of our GBM samples. In fact, this gene interacts with others (found overexpressed in TCGA-derived samples), such as AURKB, BIRC5, CCNB1, CCNB2, CDC2, and FOXM1, and forms a transcriptional network important for G2/M progression and/or checkpoint activation (Hodgson et al. 2009). Still related to cell proliferation and differentiation, DMRTA1 (chr9p21.3) and DMRTA2 (chr1p32.3) were found deleted in 21 and two samples, respectively. These genes are highly expressed in the early developing telencephalon of rodent embryos (Kikkawa et al. 2013;Konno et al. 2012). Studies show that DMRTA1 n  (Konno et al. 2012).
The MTAP gene was codeleted with CDKN2A in 21 GBMs. The protein coded by MTAP cleaves MTA (generated during polyamine biosynthesis) in adenine, and it is converted to AMP and 5-metiltioribose-1phosphate. Then, 5-metiltioribose-1-phosphate is converted to methionine (Bertino et al. 2011). Therefore, this protein is responsible for the recycling of adenine and methionine in the normal metabolism (Bertino et al. 2011). The role of MTAP in gliomas is poorly characterized. High frequency of MTAP deletion has been described in high-grade gliomas (Sasaki et al. 2003), namely in GBMs (Nakahara et al. 2004;Suzuki et al. 2004), in agreement with the present study, and also our recent report of MTAP protein expression in more than 85% of pilocytic astrocytomas (Becker et al. 2015b). Other studies have reported that homozygous deletion of MTAP is highly associated with loss of expression (Crespo et al. 2012), and that its expression is associated with lifetime-and progression-free survival in GBMs (Serao et al. 2011). Interestingly, we previously described that MTAP expression is possibly disrupted through intragenic breakpoints in pediatric high-grade gliomas (Carvalho et al. 2014).
Through bioinformatics approaches, we found that a family of several interferon (IFN) genes are coexpressed with MTAP and CDKN2A. In fact, these genes are located at the same cytoband and frequently deleted in GBM. Exogenous IFN has been used for biotherapy in several malignancies (Dillman 2011), since IFN treatment may induce apoptosis in tumor cells (Sgorbissa et al. 2011). Additionally, studies have evaluated the contribution of autocrine IFN production in the apoptotic response to IFNa in U87MG and T98G cells. They found that endogenous IFN production is responsible for sustaining high levels of TRAIL, and that loss of IFN genes confers an adaptive advantage to cancer cells, since they confer resistance to IFNa-induced apoptosis (Sgorbissa et al. 2011). In line with this, we found in the TCGA dataset that loss of expression of IFNA13, IFNA21, IFNA6, IFNA8, IFNB1, or IFNW1 was correlated to poor survival, increasing the evidence for the importance of the tumor-stroma microenvironment interaction in gliomagenesis.
Despite the extensive molecular characterizations published worldwide, all patients are been treated using the same standard protocols and the outcome of high-grade gliomas remains poor (Malmstrom et al. 2012). To date, there are very few predictive biomarkers, with MGMT methylation status the only one to have been accepted by consensus and in clinical use (Hegi et al. 2005;Malmstrom et al. 2012). In 2014, the International Society of Neuropathology recommended, in the "ISN-Haarlem Consensus Guidelines," the support of molecular analysis in the determination of tumor entities (Louis et al. 2014), showing the emerging importance of molecular analyses in diagnosis.
In conclusion, we performed, for the first time, an integrated characterization of chromosomal CNA, microsatellite instability, and TERT/IDH1 mutational analysis in astrocytomas arising in the Brazilian population. Besides the expected similar pattern of alterations described worldwide, the combination of our findings with in silico analysis of the Oncomine and TCGA data has led to the identification of genes for further investigation in glioma, such as CDK2, DMRTA1, MTAP, and IFN. This study contributes to the molecular profiling of astrocytomas, and constitutes an important step towards future personalized medical approaches for the treatment of patients diagnosed with astrocytomas.