Using single nucleotide polymorphic (SNP) allele arrays, we analyzed 28 pediatric gliomas consisting of 14 high-grade gliomas and 14 low-grade gliomas. Most of the low-grade gliomas had no detectable loss of heterozygosity (LOH) in any of the 11,562 SNP loci; exceptions were two gangliogliomas (3q and 9p), one astrocytoma (6q), and two subependymal giant cell astrocytomas (16p and 21q). On the other hand, all high-grade gliomas had various degrees of LOH affecting 52 to 2,168 SNP loci on various chromosomes. LOH occurred most frequently in regions located at 4q (54%), 6q (46%), 9p (38%), 10q (38%), 11p (38%), 12 (38%), 13q (69%), 14q (54%), 17 (38%), 18p (46%), and 19q (38%). We also detected amplifications of epidermal growth factor receptor (EGFR) or platelet-derived growth factor receptor α (PDGFRα) in a few of the 13 cases of glioblastoma multiforme analyzed. Interestingly, the amplified EGFR and PDGFRα were located within regions of LOH. SNP loci with LOH and copy number changes were validated by sequencing and quantitative PCR, respectively. Our results indicate that, in some pediatric glioblastoma multiforme, one allele each of EGFR and PDGFRα was lost but the remaining allele was amplified. This may represent a new molecular mechanism underlying tumor progression. (Cancer Res 2006; 66(23): 11172-8)

Brain tumors are the most common solid tumors in children and, despite recent advances in multimodality therapy, have the highest mortality rate of all solid tumors in this patient population. The prognosis of patients with a high-grade glioma is very dismal, whereas the optimal management of incompletely resected low-grade glioma is still debated (1). After undergoing maximal tumor resection, children with a high-grade glioma are treated with radiation, although ongoing clinical trials continue to investigate potential contributions of chemotherapy agents, such as temozolomide. For the majority of children with low-grade glioma, on the other hand, only expectant surveillance is required after complete surgical resection; radiation and/or chemotherapy are reserved for those with an incompletely resected or progressive tumor. Given the drastic difference in clinical management, an accurate pathologic diagnosis is of utmost importance. In the Children's Cancer Group high-grade glioma study CCG-945 (2), however, 70 of 250 “high-grade” gliomas were reclassified as “low-grade” tumors after a central consensus review by five independent neuropathologists. These “misdiagnoses” from a multi-institutional trial indicate a need for biomarkers to improve the accuracy of neuropathologic classification of pediatric gliomas; further functional characterizations of such markers potentially could lead to novel therapies.

Loss of heterozygosity (LOH) and deletion of tumor suppressor genes are observed frequently in malignant cells and can be associated with deregulation of cell division and apoptosis (3). Similarly, amplification of chromosomal regions can increase the expression of oncogenes during tumor progression. The recent introduction of oligonucleotide microarrays designed for whole-genome genotyping of single nucleotide polymorphisms (SNP) has facilitated detailed mapping for LOH and measurement of copy number changes at thousands of SNP loci (4, 5). In this study, we describe the use of this technique to characterize LOH and chromosome copy number changes in 14 high-grade and 14 low-grade pediatric gliomas.

Tissue samples and RNA preparation. Under an Institutional Review Board–approved protocol, brain tumor tissues were obtained, after informed consent was given, from patients undergoing tumor resection at Texas Children's Hospital, Baylor College of Medicine (Houston, TX). Portions of the tumors were fixed in 10% formaldehyde and embedded in paraffin for sectioning and histologic review by the pathologists involved in this study, and the residual tissue samples were snap frozen in liquid nitrogen and stored at −80°C for DNA extraction. All tumor tissues, except one glioblastoma multiforme (GBM750), were obtained at initial diagnosis and had no prior exposure to chemotherapy or radiation treatment. Genomic DNA from tumor tissue was isolated using DNeasy spin column (Qiagen, Valencia, CA). DNA quality and purity were assessed by agarose gel electrophoresis and optical absorbance measurement at A260/A280. Fourteen low-grade gliomas [6 pilocytic astrocytomas (JPA323, JPA159, JPA284, JPA317, JPA567, and JPA1099), 5 gangliogliomas (GAN171, GAN251, GAN503, GAN539, and GAN676), 1 astrocytoma (AST145), and 2 subependymal giant cell astrocytomas (SGCA419 and SGCA745)] and 14 high-grade gliomas [13 glioblastoma multiformes (GBM249, GBM397, GBM460, GBM534, GBM572, GBM693, GBM750, GBM903, GBM930, GBM1209, GBM-CU1, GBM-CU3, and GBM-CU4) and 1 anaplastic astrocytoma (AA-CU6)] were analyzed in this study.

SNP GeneChip assay. DNA labeling and hybridization, as well as washing and staining of the 10K SNP mapping arrays, were done according to the standard Single Primer GeneChip Mapping Assay protocol (Affymetrix, Inc., Santa Clara, CA) as described elsewhere (5). Briefly, 250 ng of genomic DNA were digested with XbaI and then ligated to a XbaI adaptor before subsequent PCR amplification using AmpliTaq Gold (Applied Biosystems, Foster City, CA). To obtain enough PCR products, four 100-μL PCRs were set up for each XbaI adaptor-ligated DNA sample. The PCR products from the four reactions were then pooled and purified. A final 20 μg PCR product was fragmented with DNase I and visualized on a 4% Tris-borate EDTA agarose gel to confirm that DNA fragment sizes ranged from 50 to 100 bp. Fragmented PCR products were then end labeled with biotin. Hybridization and detection were done with an Affymetrix Fluidics Station 450 and GeneChip Scanner 3000.

Data analysis. Signal intensity data from the GeneChip Operating software were analyzed by GeneChip DNA analysis software (GDAS; version 3.0). GDAS Mapping Algorithm uses a model-based approach to do allele calling for all SNPs on GeneChip 10K mapping arrays (6). Information about the linear chromosome location and upstream and downstream associated microsatellite markers and genes for each SNP was extracted directly from NetAffx Analysis Center7

(6). Data from SNP array experiments were collated by using OmniViz software8, and all SNPs with LOH (genotype changing from AB in blood DNA to AA or BB in the corresponding tumor DNA) were identified by using a dynamic query tool within the OmniViz software package.

dChipSNP analysis. The cell files and the corresponding SNP typing text files exported from Affymetrix GDAS software were imported into dChip for LOH and copy number analysis as described elsewhere (7, 8). Normalized intensities of the SNP loci in blood DNA samples from the same 28 study patients were used as the baseline to estimate the copy number of the corresponding SNP loci in the tumor samples. dChip uses a hidden Markov model incorporating signal variations in the normal reference group to identify chromosomal alteration regions from SNP array data (8).

Validation of copy number changes by quantitative real-time PCR. Quantitative real-time PCR was done on an ABI Prism 7000 sequence detector (Applied Biosystems) by using a SYBR Green kit (Applied Biosystems). The target locus from each tumor DNA was normalized to the reference, Line-1, a repetitive element for which copy numbers per haploid genome are similar among all human normal and neoplastic cells (9). The relative target copy number level was calculated by using normal human genomic DNA as the calibrator. Copy number change of a target SNP locus relative to that of the calibrator was determined by using the comparative CT method (10). Quantitative real-time PCR for each primer set was done at least in triplicate, and means are reported. Conditions for the reaction were as follows: 1 cycle at 95°C for 10 minutes and 40 cycles at 95°C for 15 seconds and 60°C for 1 minute. At the end of the PCR, samples were subjected to a melting analysis to confirm specificity of the amplicon. Primers were designed by using Primer 39

to span a 100- to 150-bp nonrepetitive region and synthesized by Sigma-Genosys (The Woodlands, TX). Primer sequences for each target used in this study are published as supporting information (Supplementary Table S1).

Validation of LOH by sequencing of SNP loci from both blood and tumor samples. Sequences of interest from the LOH-associated SNP loci were retrieved from the dbSNP database at the National Center for Biotechnology Information. Primers flanking the SNP site were designed by using Primer 3. PCR-amplified SNP loci fragments from both blood and tumor DNA samples were sequenced to confirm the LOH involving that SNP in the tumor.

Detection of epidermal growth factor receptor expression using reverse-transcription PCR. cDNA was synthesized from total RNA and subjected to PCR using previously described primers that flank exons 2 to 7 of epidermal growth factor receptor (EGFR; ref. 11). PCR conditions were as follows: 1 cycle at 94°C for 1 minute; 40 cycles at 94°C for 30 seconds, 55°C for 1.5 minute, and 72°C for 30 seconds; and 1 cycle at 72°C for 5 minutes.

No detectable LOH was found in most of the low-grade tumors. We analyzed 28 pediatric gliomas by SNP array. The average call rate of DNA samples was 96%, and the average number of informative SNP loci with heterozygous calls (AB) in blood DNA was 38% (Supplementary Table S2). Using the genotyping calls from pairs of blood and tumor samples, we were able to identify SNP loci with potential LOH events. A SNP locus that was AB genotype in blood DNA but AA or BB genotype in the corresponding tumor DNA was considered a potential site of LOH. Most of the low-grade gliomas had no detectable LOH or had only a single chromosomal arm affected (Fig. 1; Table 1). No detectable LOH was found in any of the six JPA or the three GAN analyzed. For the other low-grade tumors (GAN676, GAN251, AST145, SGCA419, and SGCA745), LOH involved only a single chromosomal arm. LOH at selected SNP loci of these low-grade tumors were confirmed by DNA sequencing (Supplementary Fig. S1). Both alleles were detected in the blood DNA but one of the alleles was significantly diminished or absent in the corresponding tumor DNA.

Figure 1.

Side-by-side whole-genome LOH patterns and copy number changes of 28 pediatric gliomas. Top, the sample clustering tree is based on LOH data in the significant regions. LOH regions in each of tumors are highlighted (blue color). Bottom, the intensities of red color indicate the copy numbers at the corresponding regions. Graph was generated by dChip2006 software.

Figure 1.

Side-by-side whole-genome LOH patterns and copy number changes of 28 pediatric gliomas. Top, the sample clustering tree is based on LOH data in the significant regions. LOH regions in each of tumors are highlighted (blue color). Bottom, the intensities of red color indicate the copy numbers at the corresponding regions. Graph was generated by dChip2006 software.

Close modal
Table 1.

Chromosomal location of SNP loci showing LOH

Tumor IDChromosomal regions
JPA159 ND 
JPA284 ND 
JPA317 ND 
JPA323 ND 
JPA567 ND 
JPA1099 ND 
GAN171 ND 
GAN503 ND 
GAN539 ND 
GAN676 3q 
GAN251 9p 
AST145 6q 
SGCA419 21q 
SGCA745 16p 
AA-CU6 9p 
GBM693 9p, 18p 
GBM-CU3 4q, 6p, 9p, 10q 
GBM-CU4 4q, 6p, 6q, 21q, 22q 
GBM397 9p, 13q, 14q, 18p, 19q 
GBM460 13q, 17p, 17q, 20p, 20q 
GBM249 1p, 1q, 2p, 3p, 3q, 4p, 4q, 5p, 5q, 6p, 6q, 8p, 8q, 9p, 10q, 11p, 11q, 12p, 12q, 13q, 14q, 15q, 16q, 17p, 17q, 18p, 18q, 19p, 19q, 22q 
GBM534 1p, 1q, 2p, 2q, 3p, 3q, 4p, 4q, 5p, 5q, 6p, 6q, 8p, 8q, 9p, 9q, 10p, 10q, 11p, 11q, 12p, 12q, 13q, 14q, 15q, 16p, 16q, 17p, 17q, 18p, 18q, 19p, 19q, 20p, 20q, 21q, 22q, X 
GBM572 1p, 7p, 7q33, 10q, 11p, 11q, 12q, 13q, 14q, 19q 
GBM750 1p, 1q, 2p, 3p, 3q, 4p, 4q, 5p, 6q, 9p, 9q, 12p, 12q, 13q, 14q, 15q, 16q, 18q, 21q, 22q 
GBM902 3p, 3q, 4p, 4q, 5p, 5q, 6p, 6q, 8p, 8q, 11p, 12p, 12q, 13q, 15q, 17p, 17q, 18p, 18q, 22q 
GBM930 3q, 4q, 10q21.3, 12p, 16q, 19q, 21q 
GBM1209 1p, 1q, 2p, 2q, 5q, 6p, 11p, 13q, 14q, 16p, 16q, 17p, 20p, 21q 
GBM-CU1 7p, 8p, 13q, 14q, 15q, 17q, 18p, 20p, 21q 
Tumor IDChromosomal regions
JPA159 ND 
JPA284 ND 
JPA317 ND 
JPA323 ND 
JPA567 ND 
JPA1099 ND 
GAN171 ND 
GAN503 ND 
GAN539 ND 
GAN676 3q 
GAN251 9p 
AST145 6q 
SGCA419 21q 
SGCA745 16p 
AA-CU6 9p 
GBM693 9p, 18p 
GBM-CU3 4q, 6p, 9p, 10q 
GBM-CU4 4q, 6p, 6q, 21q, 22q 
GBM397 9p, 13q, 14q, 18p, 19q 
GBM460 13q, 17p, 17q, 20p, 20q 
GBM249 1p, 1q, 2p, 3p, 3q, 4p, 4q, 5p, 5q, 6p, 6q, 8p, 8q, 9p, 10q, 11p, 11q, 12p, 12q, 13q, 14q, 15q, 16q, 17p, 17q, 18p, 18q, 19p, 19q, 22q 
GBM534 1p, 1q, 2p, 2q, 3p, 3q, 4p, 4q, 5p, 5q, 6p, 6q, 8p, 8q, 9p, 9q, 10p, 10q, 11p, 11q, 12p, 12q, 13q, 14q, 15q, 16p, 16q, 17p, 17q, 18p, 18q, 19p, 19q, 20p, 20q, 21q, 22q, X 
GBM572 1p, 7p, 7q33, 10q, 11p, 11q, 12q, 13q, 14q, 19q 
GBM750 1p, 1q, 2p, 3p, 3q, 4p, 4q, 5p, 6q, 9p, 9q, 12p, 12q, 13q, 14q, 15q, 16q, 18q, 21q, 22q 
GBM902 3p, 3q, 4p, 4q, 5p, 5q, 6p, 6q, 8p, 8q, 11p, 12p, 12q, 13q, 15q, 17p, 17q, 18p, 18q, 22q 
GBM930 3q, 4q, 10q21.3, 12p, 16q, 19q, 21q 
GBM1209 1p, 1q, 2p, 2q, 5q, 6p, 11p, 13q, 14q, 16p, 16q, 17p, 20p, 21q 
GBM-CU1 7p, 8p, 13q, 14q, 15q, 17q, 18p, 20p, 21q 

NOTE: Only chromosomal arms that have two or more SNP loci with LOH are listed.

Abbreviation: ND, not detected.

High-grade gliomas have varying degree of complex LOH. We analyzed 13 glioblastoma multiformes and identified LOH affecting 52 to 2,168 SNP loci either localized in two chromosomes (GBM693) or scattered throughout nearly all the chromosomes (GBM534; Fig. 1; Table 1). The most frequent regions of LOH were located at 4q (54%), 6q (46%), 9p (38%), 10q (38%), 11p (38%), 12 (38%), 13q (69%), 14q (54%), 17 (38%), 18p (46%), and 19q (38%; Fig. 1; Table 1). LOH at selected SNP loci at 4q12 and 7p11.2 of four glioblastoma multiforme cases were confirmed by DNA sequencing (Supplementary Fig. S2). Both alleles were observed in equal amounts in the blood DNA but only one of the alleles was detected in the corresponding tumor DNA.

Genes associated with LOH regions in low-grade glioma are involved in regulation of cell cycle and ubiquitin cycle. For the low-grade astrocytomas, AST145 and SGCA419, LOH was detected at 6q23.3-q26 (∼28.7 Mb) and 21q21.1-q22.3 (∼22.7 Mb), respectively. Genes associated with these regions were subjected to ontology analysis. The results (Table 2) indicated that several genes were involved in the ubiquitin cycle [TTC3 (21q22.2), PARK2 (6q25.2), and UBE2G2 (21q22.3)] and the cell cycle [ESR1 (6q25.1), HECA (6q23-q24), and ERG (21q22.3)].

Table 2.

Ontology analysis of genes located in regions of LOH in low-grade gliomas

Biological processGene located in LOH regionCytobandFunction
Signal transduction ESR1 6q25.1  
 ERG 21q22.3  
 APP 21q21.2  
 HBS1L 6q23-q24  
 RGS17 6q25-q26  
 GRM1 6q24  
 MAP3K71P2 6q25.1-q25.3  
 MAP3K4 6q25.3  
Ubiquitin cycle TTC3 21q22.2 Ubiquitin-protein ligase activity 
 PARK2 6q25.2-q27 Ubiquitin-protein ligase activity 
 UBE2G2 21q22.3  
Biological processGene located in LOH regionCytobandFunction
Signal transduction ESR1 6q25.1  
 ERG 21q22.3  
 APP 21q21.2  
 HBS1L 6q23-q24  
 RGS17 6q25-q26  
 GRM1 6q24  
 MAP3K71P2 6q25.1-q25.3  
 MAP3K4 6q25.3  
Ubiquitin cycle TTC3 21q22.2 Ubiquitin-protein ligase activity 
 PARK2 6q25.2-q27 Ubiquitin-protein ligase activity 
 UBE2G2 21q22.3  

Amplifications of specific alleles of platelet-derived growth factor receptor α and EGFR were detected in glioblastoma multiforme. High-level amplifications (copy number 10) were found in four glioblastoma multiforme samples (GBM-CU1, GBM-CU3, GBM572, and GBM930). Amplified SNP loci were located at 4q12 and 7p11.2, and the amplifications were confirmed by semiquantitative real-time PCR (Table 3). We sequenced the kinase domain of EGFR from exons 18 to 21 for both the blood and the corresponding tumor DNA of GBM572 but detected no mutation. Instead, a SNP located between exon 19 and 20 of EGFR indicated that only one allele of EGFR remained in the tumor (Fig. 2). An additional SNP locus within the intron 1 of EGFR also indicated that only one allele of EGFR was remaining within the genome of GBM572 (Supplementary Fig. S2). Thus, amplification of EGFR was derived from the remaining allele. Similarly, we sequenced exons 12 and 18 of platelet-derived growth factor receptor α (PDGFRα) and detected no mutation. Sequencing analysis of a SNP within exon 18 of PDGFRα indicated that one allele of PDGFRα was lost in the tumor (Fig. 2).

Table 3.

LOH loci with DNA amplifications as confirmed by quantitative real-time PCR

SNP IDAssociated gene (distance from SNP)CytobandCopy numbers
GBM-CU1TGBM-CU3TGBM-572TGBM-930T
SNP_A-1512510 KIT (299040 U), PDGFRA (60710 D), GSH-2 (256999D), CHIC2 (294333 D) 4q12 5.7 (3.6 ± 0.1) 30.7 (12.4 ± 0.2) 4.5 (1.7 ± 0.3) 13.0 (7.4 ± 0.3) 
SNP_A-1518582 EGFR (0, intron) 7p11.2 40.1 (162 ± 0.5) 2.7 (2.1 ± 1.6) 30.2 (122 ± 0.5) 2.8 (1.4 ± 0.6) 
SNP IDAssociated gene (distance from SNP)CytobandCopy numbers
GBM-CU1TGBM-CU3TGBM-572TGBM-930T
SNP_A-1512510 KIT (299040 U), PDGFRA (60710 D), GSH-2 (256999D), CHIC2 (294333 D) 4q12 5.7 (3.6 ± 0.1) 30.7 (12.4 ± 0.2) 4.5 (1.7 ± 0.3) 13.0 (7.4 ± 0.3) 
SNP_A-1518582 EGFR (0, intron) 7p11.2 40.1 (162 ± 0.5) 2.7 (2.1 ± 1.6) 30.2 (122 ± 0.5) 2.8 (1.4 ± 0.6) 

NOTE: ΔΔCT method, use Line-1 for normalization and calibrated to tonsil DNA as 2 copy reference.

Abbreviations: D, distance downstream from the SNP locus; U, distance upstream from the SNP locus.

Figure 2.

LOH of EGFR and PDGFRα genes were located at 7p11.2 and 4q12 loci with DNA amplifications, respectively.

Figure 2.

LOH of EGFR and PDGFRα genes were located at 7p11.2 and 4q12 loci with DNA amplifications, respectively.

Close modal

Detection of an amplified type III mutation of EGFR transcript in glioblastoma multiforme. The most common mutation of the EGFR gene is called a type III mutation, an in-frame deletion of 801-bp spanning exons 2 to 7 in the mRNA, which confers enhanced tumorigenicity when transfected into glioblastoma multiforme cells expressing a wild-type (WT) EGFR (12). Using reverse-transcription PCR (RT-PCR) with a pair of primers flanking exons 2 to 7 as described previously (11), we detected an amplified type III mutation of EGFR involving a 352-bp DNA fragment from cDNA derived from GBM572 (Fig. 3). Sequencing analysis of this cDNA confirmed a truncated 352-bp DNA fragment (data not shown). However, PCR analysis of the exons 2 to 7 of EGFR in both blood and tumor DNA of GBM572 indicated that there was no loss of any of these exons in the genome (Fig. 4). We speculate that the type III mutation of EGFR in the RNA might be a result of genomic rearrangement between intron 1 and intron 7 as observed previously (13). GBM249 also overexpressed an apparently WT EGFR (Fig. 3) but, interestingly, this increase in expression was not due to gene amplification (Fig. 1). Thus, a different mechanism might exist for the overexpression of EGFR in GBM249.

Figure 3.

RT-PCR of EGFR in glioblastoma multiforme. MW, molecular weight size markers; NTC, no template control; Normal, normal cerebral cortex total RNA; GBM, total RNA from glioblastoma multiforme tumor samples.

Figure 3.

RT-PCR of EGFR in glioblastoma multiforme. MW, molecular weight size markers; NTC, no template control; Normal, normal cerebral cortex total RNA; GBM, total RNA from glioblastoma multiforme tumor samples.

Close modal
Figure 4.

PCR analysis of exons 2 to 7 in GBM572 blood DNA and tumor DNA. B, blood DNA; T, tumor DNA. Left, molecular sizes.

Figure 4.

PCR analysis of exons 2 to 7 in GBM572 blood DNA and tumor DNA. B, blood DNA; T, tumor DNA. Left, molecular sizes.

Close modal

Homozygous deletions at the ELAVL2 (9p21.3) and HNT (11q25) loci. Homozygous deletions at five SNP loci located at 9p21.3 were detected in the tumor GBM750. Using real-time quantitative PCR, we confirmed that the gene, ELAVL2 [embryonic lethal, abnormal vision, Drosophila-like 2 (Hu antigen B)], at this locus was lost. Sequencing analysis of exon 6 in ELAVL2 gene from all the 13 glioblastoma multiforme genomic DNA has shown that one of the glioblastoma multiforme cases, GBM397, had a mutation (C to T) at position 1509 (Fig. 5) of ELAVL2 mRNA sequence (NM_004432). The mutation will result in a change from arginine to cysteine in the peptide sequence (Fig. 5). Another homozygous deletion was detected at a SNP locus (SNP_A-1508978) located at 11q25 in the tumor GBM572. This SNP is located within an intron of the gene neurotrimin (HNT). Its deletion was further confirmed by real-time quantitative PCR as zero copy (data not shown).

Figure 5.

Sequencing chromatograms of exon 6 of ELAVL2 from GBM397 blood DNA and tumor DNA indicated a C to T mutation that will result in an arginine to cysteine change in the ELAVL2 peptide.

Figure 5.

Sequencing chromatograms of exon 6 of ELAVL2 from GBM397 blood DNA and tumor DNA indicated a C to T mutation that will result in an arginine to cysteine change in the ELAVL2 peptide.

Close modal

This is the first report of genomic analysis of pediatric gliomas by SNP array, which allows detection of both DNA copy number changes and LOH. The 28 pediatric gliomas analyzed consisted of 14 high-grade and 14 low-grade tumors. In this section, the chromosomal regions of these pediatric gliomas with LOH, amplifications, and homozygous deletions are discussed.

Loss of heterozygosity. Most of the low-grade gliomas exhibited no detectable LOH in any of the 11,562 SNP loci; the exceptions are two gangliogliomas (3q and 9p), one astrocytoma (6q), and two subependymal giant cell astrocytomas (16p and 21q). Ontology analysis of genes associated with LOH regions in these low-grade gliomas indicated that a possible defect in the ubiquitin cycle might be involved (Table 2). On the other hand, the high-grade gliomas had complex LOH involving 52 to 2,168 SNP loci on numerous chromosomes. LOH regions identified with high frequency in glioblastoma multiforme include 4q (54%), 6q (46%), 9p (38%), 10q (38%), 11p (38%), 12 (38%), 13q (69%), 14q (54%), 17 (38%), 18p (46%), and 19q (38%). A previous analysis of 13 malignant astrocytomas in children and young adults identified extensive genomic loss and gain by comparative genomic hybridization (14). In that study, the most common recurrent copy number aberrations were losses of 16p (54% of cases), 17p (38%), 19p (38%), and 22 (38%). Using SNP array, we detected similar frequencies of genomic losses in terms of LOH at 17p and 19p but lower occurrence of losses at 16p (15%) and 22 (23%). Additionally, we identified more regions with LOH than the previous study. This is probably because LOH detection by SNP array is based on genotyping calls of matched normal blood DNA and tumor DNA from the same patient and therefore can detect loss of one allele followed by reduplication of the remaining allele as LOH. In contrast, comparative genomic hybridization is based on DNA copy number ratios to a reference DNA and thus cannot detect reduplication of the remaining allele for LOH.

Although low-grade gliomas in children rarely progress to glioblastoma multiforme, a phenomenon of “malignant transformation” more frequently seen in adult patients, several of the most common recurrent regions of LOH (4q, 9p, 10q, 11p, and 13q) we identified in pediatric glioblastoma multiforme are associated with malignant progression in adult glioblastoma multiforme (15). Microsatellite markers have identified novel regions of allelic deletion on chromosome 18p in tumors of the lung, brain, and breast (16). Similarly, a region of common deletion in 22q13.3 in human glioma has been associated with astrocytoma malignant progression (17). In this study, 18p (46%) was one of the regions of most frequent LOH, whereas LOH in 22q13.3 was found in three glioblastoma multiforme. Apparently, LOH at these regions may be related to the aggressive biology of glioblastoma multiforme. Future prospective studies are needed to show any potential correlations between specific LOH and clinical outcomes.

Amplification of EGFR and PDGFR. SNP loci located at 4q12 and 7p11.2 were found to be highly amplified in glioblastoma multiforme. The amplifications of EGFR and PDGFR at these loci were confirmed by semiquantitative real-time PCR (Table 3). A previous study has shown that amplification of EGFR is the most common genetic alteration in adult glioblastoma multiforme (>40%) but occurs rarely in pediatric glioblastoma multiforme (18). In our study of pediatric glioblastoma multiforme, amplification of EGFR is indeed a rare event and was detected in only 2 of 13 glioblastoma multiformes. In adult glioblastoma multiforme, the expression of a truncated receptor with constitutive tyrosine kinase activity (19) and aberrant receptor signaling (20) is identified frequently. The most common rearrangement of the EGFR gene is a type III mutation, an in-frame deletion of 801-bp spanning exons 2 to 7 in the mRNA (12). RT-PCR and sequencing analysis of the EGFR transcript in GBM572 confirmed a type III mutation (Fig. 3). Whereas both GBM249 and GBM572 overexpressed EGFR, only GBM572 also had DNA amplification. Although pediatric glioblastoma multiforme must be considered independently from adult glioblastoma multiforme, demonstration of some cases of pediatric glioblastoma multiforme with amplification and rearrangement of EGFR provides the rationale for clinical trials with agents targeting EGFR (21).

Amplification of the PDGFRα gene has been shown recently by real-time quantitative PCR to occur in ∼15% of adult astrocytomas (22), and PDGFRα overexpression has been hypothesized to be an early event in malignant transformation of low-grade astrocytoma (23). In this study, we detected amplification of the PDGFRα gene in 2 of the 13 glioblastoma multiformes. From the expression profiles of GBM930, acquired with Affymetrix U133A chip, genes located at the 4q12 region (KIT, PDGFRα, and CHIC2) were expressed at very high levels (data not shown), indicating the concordance between DNA amplification and mRNA overexpression of these genes in GBM930.

Interestingly, both amplified EGFR and PDGFRα genes are located in regions of LOH, which implies amplification of a specific allele. Sequencing analysis and RT-PCR analysis of the EGFR transcript in GBM572 confirmed that the amplified EGFR allele carries a type III mutation and is the only allele being amplified in the region of LOH (Figs. 2 and 3). This may represent a new mechanism during tumor progression. A previous study of acute lymphoblastic leukemia also provided evidence that duplication of the remaining chromosome can occur in the event of an LOH involving one chromosome (24). Moreover, another recent study on 100 cases of lung cancer samples by SNP array has also revealed that DNA amplification is monoallelic or from the remaining alleles located at LOH regions (25).

Homozygous deletions. 9p is one of the most common sites for LOH in high-grade gliomas. In this study, we detected a homozygous deletion of a region located at 9p21, involving the ELAVL2 gene, in one of the glioblastoma multiformes. ELAVL2 (He1-N1) encodes a predicted 359-amino acid protein that shows significant similarity to the product of the Drosophila elav gene. Absence of the elav gene in Drosophila causes multiple structural defects and hypotrophy of the central nervous system of the fly. Homozygous deletion of the region centering around D9S126 (9p21) and ELAVL2 were found previously by microsatellite marker and fluorescence in situ hybridization analysis in a small cell lung cancer cell line, and ELAVL2 is suggested to be a candidate tumor suppressor gene in both small cell lung cancers and non–small cell lung cancers (26). The identification of a mutation in one of the glioblastoma multiforme cases in this study further supports the potential tumor suppressor role of ELAVL2. Another region of homozygous deletion is located at 11q25, which also has a high frequency of recurrent LOH in glioblastoma multiforme. One of the SNPs exhibiting homozygous deletion is located within an intron of the HNT gene. HNT is a member of the subfamily of immunoglobulin-like cell adhesion molecules and is differentially regulated during development of the central nervous system (27). Cell adhesion molecules are involved in cell dispersal and invasion (28). Future functional studies are needed to confirm whether a loss of HNT function would enhance the invasive and metastatic potential of tumor cells.

In summary, analysis of LOH and copy number changes by SNP microarrays in pediatric gliomas has identified significant areas of allelic imbalance. Novel loci with LOH, amplifications, and homozygous deletions will be of particular interest for the discovery of oncogenes and tumor suppressor genes involved in the pathogenesis of gliomas.

Note: Supplementary data for this article are available at Cancer Research Online (http://cancerres.aacrjournals.org/).

Grant support: Pediatric Brain Tumor Foundation (K-K. Wong); John S. Dunn Research Foundation; Robert J. Kleberg, Jr. and Helen C. Kleberg Foundation; Gillson Longenbaugh Foundation; and Cancer Fighters of Houston, Inc.

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.

We thank the individuals and families who agreed to take part in these studies, the members of the Cancer Genomics Group for helpful discussions, and Kathryn Hale for editing the article.

1
Fisher BJ, Leighton CC, Vujovic O, Macdonald DR, Stitt L. Results of a policy of surveillance alone after surgical management of pediatric low grade gliomas.
Int J Radiat Oncol Biol Phys
2001
;
51
:
704
–10.
2
Fouladi M, Hunt DL, Pollack IF, et al. Outcome of children with centrally reviewed low-grade gliomas treated with chemotherapy with or without radiotherapy on Children's Cancer Group high-grade glioma study CCG-945.
Cancer
2003
;
98
:
1243
–52.
3
Lengauer C, Kinzler KW, Vogelstein B. Genetic instabilities in human cancers.
Nature
1998
;
396
:
643
–9.
4
Matsuzaki H, Loi H, Dong S, et al. Parallel genotyping of over 10,000 SNPs using a one-primer assay on a high-density oligonucleotide array.
Genome Res
2004
;
14
:
414
–25.
5
Wong KK, Tsang YT, Shen J, et al. Allelic imbalance analysis by high-density single-nucleotide polymorphic allele (SNP) array with whole genome amplified DNA.
Nucleic Acids Res
2004
;
32
:
e69
.
6
Liu G, Loraine AE, Shigeta R, et al. NetAffx: Affymetrix probesets and annotations.
Nucleic Acids Res
2003
;
31
:
82
–6.
7
Lin M, Wei LJ, Sellers WR, Lieberfarb M, Wong WH, Li C. dChipSNP: significance curve and clustering of SNP-array-based loss-of-heterozygosity data.
Bioinformatics
2004
;
20
:
1233
–40.
8
Zhao X, Li C, Paez JG, et al. An integrated view of copy number and allelic alterations in the cancer genome using single nucleotide polymorphism arrays.
Cancer Res
2004
;
64
:
3060
–71.
9
Wang TL, Maierhofer C, Speicher MR, et al. Digital karyotyping.
Proc Natl Acad Sci U S A
2002
;
99
:
16156
–61.
10
Tsang YT, Chang YM, Lu X, Rao PH, Lau CC, Wong KK. Amplification of MGC2177, PLAG1, PSMC6P, and LYN in a malignant mixed tumor of salivary gland detected by cDNA microarray with tyramide signal amplification.
Cancer Genet Cytogenet
2004
;
152
:
124
–8.
11
Zhang R, Tremblay TL, McDermid A, Thibault P, Stanimirovic D. Identification of differentially expressed proteins in human glioblastoma cell lines and tumors.
Glia
2003
;
42
:
194
–208.
12
Nishikawa R, Ji XD, Harmon RC, et al. A mutant epidermal growth factor receptor common in human glioma confers enhanced tumorigenicity.
Proc Natl Acad Sci U S A
1994
;
91
:
7727
–31.
13
Frederick L, Eley G, Wang XY, James CD. Analysis of genomic rearrangements associated with EGRFvIII expression suggests involvement of Alu repeat elements.
Neuro-oncol
2000
;
2
:
159
–63.
14
Warr T, Ward S, Burrows J, et al. Identification of extensive genomic loss and gain by comparative genomic hybridisation in malignant astrocytoma in children and young adults.
Genes Chromosomes Cancer
2001
;
31
:
15
–22.
15
Weber RG, Sabel M, Reifenberger J, et al. Characterization of genomic alterations associated with glioma progression by comparative genomic hybridization.
Oncogene
1996
;
13
:
983
–94.
16
Tran Y, Benbatoul K, Gorse K, et al. Novel regions of allelic deletion on chromosome 18p in tumors of the lung, brain and breast.
Oncogene
1998
;
17
:
3499
–505.
17
Oskam NT, Bijleveld EH, Hulsebos TJ. A region of common deletion in 22q13.3 in human glioma associated with astrocytoma progression.
Int J Cancer
2000
;
85
:
336
–9.
18
Weiss WA. Genetics of brain tumors.
Curr Opin Pediatr
2000
;
12
:
543
–8.
19
Voldborg BR, Damstrup L, Spang-Thomsen M, Poulsen HS. Epidermal growth factor receptor (EGFR) and EGFR mutations, function, and possible role in clinical trials.
Ann Oncol
1997
;
8
:
1197
–206.
20
Nagane M, Lin H, Cavenee WK, Huang HJ. Aberrant receptor signaling in human malignant gliomas: mechanisms and therapeutic implications.
Cancer Lett
2001
;
162
Suppl:
R17
–21.
21
Mellinghoff IK, Wang MY, Vivanco I, et al. Molecular determinants of the response of glioblastomas to EGFR kinase inhibitors.
N Engl J Med
2005
;
353
:
2012
–24.
22
Arjona D, Bello MJ, Alonso ME, et al. Real-time quantitative PCR analysis of regions involved in gene amplification reveals gene overdose in low-grade astrocytic gliomas.
Diagn Mol Pathol
2005
;
14
:
224
–9.
23
Guha A, Dashner K, Black PM, Wagner JA, Stiles CD. Expression of PDGF and PDGF receptors in human astrocytoma operation specimens supports the existence of an autocrine loop.
Int J Cancer
1995
;
60
:
168
–73.
24
McEvoy CR, Morley AA, Firgaira FA. Evidence for whole chromosome 6 loss and duplication of the remaining chromosome in acute lymphoblastic leukemia.
Genes Chromosomes Cancer
2003
;
37
:
321
–5.
25
Laframboise T, Weir BA, Zhao X, et al. Allele-specific amplification in cancer revealed by SNP array analysis.
PLoS Comput Biol
2005
;
1
:
e65
.
26
Cairns P, Okami K, King P, et al. Genomic organization and mutation analysis of Hel-N1 in lung cancers with chromosome 9p21 deletions.
Cancer Res
1997
;
57
:
5356
–9.
27
Struyk AF, Canoll PD, Wolfgang MJ, Rosen CL, D'Eustachio P, Salzer JL. Cloning of neurotrimin defines a new subfamily of differentially expressed neural cell adhesion molecules.
J Neurosci
1995
;
15
:
2141
–56.
28
Bogenrieder T, Herlyn M. Axis of evil: molecular mechanisms of cancer metastasis.
Oncogene
2003
;
22
:
6524
–36.

Supplementary data