Alterations of DNA copy number are believed to be important indicators of tumor progression in human astrocytoma. We used an array of bacterial artificial chromosomes to map relative DNA copy number in 50 primary glioblastoma multiforme tumors at ∼1.4-Mb resolution. We identified 33 candidate sites for amplification and homozygous deletion in these tumors. We identified three major genetic subgroups within these glioblastoma multiforme tumors: tumors with chromosome 7 gain and chromosome 10 loss, tumors with only chromosome 10 loss in the absence of chromosome 7 gain, and tumors without copy number change in chromosomes 7 or 10. The significance of these genetic groups to therapeutics needs further study.

Studies suggest that astrocytic brain tumor behavior is related to genetic defects acquired over the course of tumor development (16). It is believed that understanding these genetic defects will help predict response to treatment and patient outcome. Previous studies using comparative genomic hybridization (CGH; refs. 7, 8) related radiation sensitivity of tumors and patients' survival to copy number aberrations (CNA) on several chromosomes (2, 3). This report describes CNAs identified on a bacterial artificial chromosome (BAC) array of ∼1.4-Mb resolution [array comparative genomic hybridization (aCGH)], in a set of 50 glioblastoma multiforme (GBM).

We compared specificity and sensitivity of the chromosome CGH and aCGH techniques by comparing array and chromosome CGH in the same tumor samples, by counting fluorescence in situ hybridization (FISH) signals generated by BAC clones used in the array, and by using real-time quantitative PCR. The results obtained by aCGH confirmed other characterizations of the GBM genome.

Analysis of high-resolution aCGH data identified genetic subgroups in this set of GBM and loci for candidate oncogenes and tumor suppressor genes, many of which were previously unknown. We expect that future higher resolution studies of these regions will lead to identification of target genes for therapy.

Cell lines. The cell lines used in this study (SF767, SF126, U343MG, and SF210) were obtained from the Tissue Bank at the Brain Tumor Research Center, University of California, San Francisco UCSF; ref. (9). Cells were cultured in minimal essential medium with Earle's buffered salt solution supplemented with 10% FCS and nonessential amino acids.

Tumor DNA. Fifty tumor samples (GBM, grade 4 astrocytoma) were obtained from the tissue bank at the Brain Tumor Research Center, UCSF. All patients signed consent for specimens to be used for research purposes. The samples were originally chosen for evaluation using chromosome CGH (8). Tumor specimens were snap-frozen in liquid nitrogen immediately after resection and stored at −80°C. Tumors were diagnosed by the Division of Neuropathology at UCSF according to WHO classification (10). Sections on both sides contiguous to the processed specimens were histologically assessed to determine the percentage of tumor cells present in the tissue before DNA extraction. Specimens with 50% or more tumor cells in the adjacent sections were selected for DNA extraction.

Bacterial artificial chromosome array. We used an array with 2,246 BAC clones developed at UCSF for aCGH (11, 12). BAC clones were spotted in triplicate on chromium-coated glass slides. Approximately 10% of the genome was represented on the array. The average resolution was ∼1.4 Mb, but the coverage varied from chromosome to chromosome. The Y chromosome was not represented on this version of the array. BAC clones that FISH mapped to more than one locus were excluded from our analysis.

Array comparative genomic hybridization. We used random priming (13) to label 300 to 500 ng tumor DNA with Cy3-labeled deoxynucleotide triphosphate and sex-matched normal DNA with Cy5-labeled deoxynucleotide triphosphate. The labeled probes were mixed with 50 μg human Cot-1 DNA, precipitated with sodium acetate and ethanol, and resuspended in 2.5 μL of water, 2.5 μL of yeast tRNA (100 μg/mL), 10 μL of 20% SDS, and 35 μL of hybridization mix [15% dextran sulfate in 2× SSC and 50% formamide (pH 7.0)]. The resulting solution was denatured at 80°C for 10 minutes, incubated at 37°C for 1 hour, and applied to an array preblocked for 30 minutes with 50 μg of herring sperm DNA in 5 μL water, 10 μL 20% SDS, and 35 μL hybridization mix. Hybridizations were done at 37°C for 48 hours in a sealed chamber on a slowly rocking stage. Arrays were washed for 15 minutes at 45°C in 50% formamide/2× SSC (pH 7.0), and for 10 minutes at room temperature in 0.1 mol/L Na2HPO4 with 0.1% NP40 (pH 8.0). Arrays were mounted in 0.1 μmol/L 4′,6-diamidione-2-phenylindole in 90% glycerol to visualize spotted BACs, and fluorescent images were acquired using a 16-bit charge-coupled device camera with appropriate filter sets (11).

Image analysis and data acquisition. The “spot” image analysis program (14) was used to assign pixels to the foreground and background for each DNA spot. We evaluated data quality and calculated a relative ratio (RR) of tumor copy number to normal DNA copy number at loci represented on the array. The RR was defined as the ratio of background-subtracted signal intensities for tumor (Cy3) and normal (Cy5) DNA, normalized to the median RR of that hybridization. We required data from at least two of three replicates for each BAC, with a SD ≥0.1. RR values were converted to log2 RR values and plotted against the map position of the BACs on each chromosome.

Copy number aberration frequency map and cutoffs. We constructed a histogram from normalized log2 RR values from each hybridization (Fig. 1A and B), and fit a mixture of three Gaussian distributions to the data according to an algorithm described in Hodgson et al. (15). We defined gains and losses as values greater than or less than 3 SD from the mode of the central Gaussian. The average SD of the central Gaussians for our set of samples was 0.1 ± 0.025. Hybridizations with a SD of >0.15 for the central Gaussian were defined as poor quality and rehybridized. Each locus in each tumor was scored as gain, loss, or no change, and this converted data set was used to calculate and plot the frequency of gain and loss at each locus (BAC) on the array.

Fig. 1.

A, log2 RRs from aCGH of a GBM plotted as a function of genome length. The chromosome boundaries are noted above. B, determining cutoffs for a hybridization. We constructed a histogram of log2 RRs for each case. Three Gaussian distributions were fitted to the histogram (fit is shown as a solid line). Central peak, normal copy number portion of tumor DNA. We took 3 SD values calculated from the central peak (−0.169 and 0.173 for this case) as the cutoff. Clones with RR values below or above the cutoff values were scored as lost or gained, respectively.

Fig. 1.

A, log2 RRs from aCGH of a GBM plotted as a function of genome length. The chromosome boundaries are noted above. B, determining cutoffs for a hybridization. We constructed a histogram of log2 RRs for each case. Three Gaussian distributions were fitted to the histogram (fit is shown as a solid line). Central peak, normal copy number portion of tumor DNA. We took 3 SD values calculated from the central peak (−0.169 and 0.173 for this case) as the cutoff. Clones with RR values below or above the cutoff values were scored as lost or gained, respectively.

Close modal

Comparing chromosome comparative genomic hybridization and array comparative genomic hybridization. Chromosome CGH was done as described by Mohapatra et al. (16).We compared aCGH data from the same set of 50 GBM with chromosome CGH data (16). Two investigators independently scored standard CGH profile CNAs as sure loss (−2), possible loss (−1), normal (0), possible gain (1), sure gain (2), or amplification (3). Scoring differences were resolved by mutual agreement. We used 3 SD aCGH cutoff values to score gains, losses, and amplifications and compared them with the chromosomal CGH profile scores. CNAs determined by aCGH were defined as “sure” (−2 or 2) if >50% of BACs in the region were aberrant, “possible” (−1 or 1) if <50% of BACs in the region were aberrant, and “amplified” if log2 RR values of each BAC in that region were ≥2. We scored a region as normal if there were occasional single-aberrant BACs with an otherwise normal profile.

Quantitative microsatellite analysis. Quantitative microsatellite analysis (17) was done at three loci on 7q in a set of 14 GBM selected from our DNA inventory. BAC sequences were mined for possible CA repeat elements and flanking primers were designed with “name” software (Applied Biosystems, Foster City, CA). The forward and reverse primer sequences were as follows: locus 1, 5′-ATTATTGGCCAGTTGGTTCT-3′, 5′-ACAAGGAATAATATCCAGAAGA-3′; locus 2, 5′-TGAAGCACCTGTCCAACC-3′, 5′-ACAAGGAATAATATCCAGAAGA-3′; and locus 3, 5′-CTTCAAACTTTAACTTCC, 5′-TTTACATAGCAACCATTGCA-3′. A pool of six primer sets from microsatellite loci served as the control for a diploid locus as previously described (17). Reactions were done in triplicate with 5 ng of genomic DNA using the ABI 7700 (Applied Biosystems). The average Ct value for each set of triplicates was used to calculate copy number. Gains and losses at loci were determined based on a tolerance interval previously calculated from the results of 167 amplification reactions done on DNA obtained from 10 individuals (17). Based on this tolerance interval, values <1.58 were defined as losses, whereas values >2.53 were defined as gains. In this study, because one locus gave a value of 2.7 in a sample of control DNA, we scored more conservatively, defining values >2.7 as gain.

Fluorescence in situ hybridization. BAC DNA was extracted using a Qiagen tip-500 column (Valencia, CA) and labeled with digoxigenin-11-dUTP or biotin-14-dUTP (Roche Molecular Biochemicals, Indianapolis, IN) by nick translation. Touch preparations were made from frozen tumor tissue, and metaphase spreads were prepared from cell lines SF767, SF126, U343, and SF210 and from normal leukocytes (16). All hybridizations were done as described previously (16). BAC clones used for interphase FISH were physically mapped using the normal metaphase spreads, and the absolute copy number of loci was determined by evaluating at least 200 interphase cells.

Unsupervised clustering. We converted the RR value for each BAC clone in each tumor to a score of 1 (gain/amplified), 0 (no change), or −1 (loss) based on the 3 SD standard described above and analyzed the converted data set with Cluster 3.0 (18), choosing mean centered (no filtering) and centroid linkage (correlation uncentered) calculations. We viewed the results in Treeview (18). Gain, loss, and no change were represented in the final heat map as red, green, and black, respectively.

Supervised analysis. We used Significance Analysis of Microarrays (SAM, available at: http://www-stat.stanford.edu/~tibs/SAM; ref. 19) to identify BACs that segregate genetic subgroups in GBM. We removed BACs with an SD of zero across the data set before analysis. The thresholds for “significant” BACs were chosen to keep the apparent false discovery rate <1%.

Clinical data. We obtained age, sex and survival data for the cases (from existing databases in the Brain Tumor Research Center) and used Cox's proportional hazard model to analyze relationships between survival and genetic aberrations, age and sex. All patients on whom we had follow-up had died. Survival was calculated from the day of first surgery to the date of death. We plotted Kaplan-Meier curves for survival in each genetic group. Fisher's exact test was used to analyze relationships between sex and genetic groups.

Validation of array comparative genomic hybridization FISH and chromosome comparative genomic hybridization

Brain tumor cell line SF767.Figure 2 shows representative chromosome 7 results from SF767 that validate aCGH using FISH and chromosome CGH. Chromosome CGH (Fig. 2A) and aCGH (Fig. 2B) profiles were similar, but aCGH precisely mapped a region of loss to a site near 18,000 bp. FISH confirmed copy number at four BAC clones with aberrant RR values on aCGH (Fig. 2C).

Fig. 2.

aCGH validation by chromosome CGH and FISH. Chromosome CGH (A) and aCGH (B) of chromosome 7 in the SF767 human brain tumor cell line. Overall profiles of CGH and aCGH correlate well. aCGH precisely maps a loss near 7pter to ∼18,000 bp. C, FISH with BAC clones used for aCGH. Representative FISH analyses of interphase nuclei in SF767 cells display the absolute chromosome 7 centromere copy number in red and the indicated BAC clones (arrows) in green. The average absolute copy number was determined by counting signals in 200 interphase cells. FISH confirmed the relative copy number as determined by aCGH in each case.

Fig. 2.

aCGH validation by chromosome CGH and FISH. Chromosome CGH (A) and aCGH (B) of chromosome 7 in the SF767 human brain tumor cell line. Overall profiles of CGH and aCGH correlate well. aCGH precisely maps a loss near 7pter to ∼18,000 bp. C, FISH with BAC clones used for aCGH. Representative FISH analyses of interphase nuclei in SF767 cells display the absolute chromosome 7 centromere copy number in red and the indicated BAC clones (arrows) in green. The average absolute copy number was determined by counting signals in 200 interphase cells. FISH confirmed the relative copy number as determined by aCGH in each case.

Close modal

Primary human brain tumors.Figures 3 and 4 show 9p loss and chromosome 7 gain/epidermal growth factor receptor (EGFR) amplification as determined by chromosome CGH (Figs. 3A and 4A), aCGH (Figs. 3B and 4B), and FISH (Figs. 3C and D and 4C) in two different GBM. aCGH defined EGFR amplicon length and relative copy number (Fig. 4B) more precisely than chromosomal CGH (Figs. 3A and 4A) and detected single-copy loss on much of 9p (FISH-validated for RPCI11-229N14; Fig. 3D), as well as homozygous loss at p16/INK4a/ARF (Fig. 3B). FISH with BAC clones containing EGFR and INK4a/ARF genes validated the aCGH results for absolute copy number at these two loci (Figs. 3C and D and 4C).

Fig. 3.

Regions of homozygous and heterozygous INK4a/ARF and 9p loss in a primary GBM. CGH (A), aCGH (B), and FISH (C) results. CGH and aCGH profiles each detected loss on chromosome 9. aCGH detected two levels of loss (arrows, RR >−1.0 for homozygous loss). The homozygous loss in this tumor was at the INK4a/ARF locus. C, FISH with BAC RPCI11-55A19 (green) containing the INK4a/ARF gene confirmed homozygous loss of the INK4a/ARF locus. The control RPCI11-95G21 probe (red) at 9q22.2 identified two copies in the tumor cell. D, one copy of the 9p locus identified by the RPCI11-229N14 probe (red) and two copies of the control 9q22 locus identified by the BAC RPCI11-95G21 probe (green).

Fig. 3.

Regions of homozygous and heterozygous INK4a/ARF and 9p loss in a primary GBM. CGH (A), aCGH (B), and FISH (C) results. CGH and aCGH profiles each detected loss on chromosome 9. aCGH detected two levels of loss (arrows, RR >−1.0 for homozygous loss). The homozygous loss in this tumor was at the INK4a/ARF locus. C, FISH with BAC RPCI11-55A19 (green) containing the INK4a/ARF gene confirmed homozygous loss of the INK4a/ARF locus. The control RPCI11-95G21 probe (red) at 9q22.2 identified two copies in the tumor cell. D, one copy of the 9p locus identified by the RPCI11-229N14 probe (red) and two copies of the control 9q22 locus identified by the BAC RPCI11-95G21 probe (green).

Close modal
Fig. 4.

aCGH-determined copy number and boundaries of the EGFR amplicon in GBM. Chromosome CGH (A) and aCGH (B) of chromosome 7 with EGFR amplification. Both profiles exhibit amplification on chromosome 7. aCGH maps the EGFR amplification with better resolution (arrow). C, FISH with BAC clone RP5-1091E12 (containing EGFR; red) shows amplification.

Fig. 4.

aCGH-determined copy number and boundaries of the EGFR amplicon in GBM. Chromosome CGH (A) and aCGH (B) of chromosome 7 with EGFR amplification. Both profiles exhibit amplification on chromosome 7. aCGH maps the EGFR amplification with better resolution (arrow). C, FISH with BAC clone RP5-1091E12 (containing EGFR; red) shows amplification.

Close modal

Quantitative PCR. We assayed copy number using quantitative microsatellite analysis at 3 loci on 7q to validate aCGH data in 14 tumors. Table 1 provides mapping data for the quantitative microsatellite analysis primers and the 5′ BAC clone position (April, 2003 freeze, UCSC genome database). The aCGH and quantitative microsatellite analysis data were scored as normal (N) or gained (G). aCGH and quantitative microsatellite analysis data disagreed in tumors 26, 27, and 45 at locus 1, in tumor 45 at locus 2, and in tumor 5 at locus 3 (see boxes, Table 1). Thus, 37 of 42 comparisons between aCGH and QPCR agreed.

Table 1.

Comparison of quantitative PCR and aCGH

Tumor numberControl number
Base position Loci/Clones 3 5 6 15 17 26 27 31 34 36 37 45 46 48 1 2 
104,127,138 CTB_66D11 (RR)   
103,873,013 QPCR, Locus 3 2.3 (N) 1.9 (N) 3.6 (G) 2.1 (N) 2.3 (N) 2.2 (N) 2.4 (N) 3.2 (G) 2.1 (N) 3.4 (G) 2.2 (N) 2.2 (N) 3.4 (G) 3.7 (G) 2.2 2.3 
103,510,640 QPCR, locus 2 2.3 (N) 3 (G) 3.2 (G) 2.5 (N) 2.5 (N) 2.5 (N) 2.7 (N) 3.3 (G) 2.5 (N) 3.5 (G) 1.9 (N) 2.8 (G) 3.9 (G) 4.2 (G) 2.4 
103,510,610 QPCR, locus 1 2.2 (N) 3.1 (G) 3.6 (G) 2.6 (N) 2.7 (N) 2.9 (G) 3.1 (G) 3.7 (G) 2.5 (N) 3.6 (G) 2.3 (N) 3.1 (G) 4 (G) 4.1 (G) 2.2 2.7 
101,400,208 CTC_305I2 (RR)   
Tumor numberControl number
Base position Loci/Clones 3 5 6 15 17 26 27 31 34 36 37 45 46 48 1 2 
104,127,138 CTB_66D11 (RR)   
103,873,013 QPCR, Locus 3 2.3 (N) 1.9 (N) 3.6 (G) 2.1 (N) 2.3 (N) 2.2 (N) 2.4 (N) 3.2 (G) 2.1 (N) 3.4 (G) 2.2 (N) 2.2 (N) 3.4 (G) 3.7 (G) 2.2 2.3 
103,510,640 QPCR, locus 2 2.3 (N) 3 (G) 3.2 (G) 2.5 (N) 2.5 (N) 2.5 (N) 2.7 (N) 3.3 (G) 2.5 (N) 3.5 (G) 1.9 (N) 2.8 (G) 3.9 (G) 4.2 (G) 2.4 
103,510,610 QPCR, locus 1 2.2 (N) 3.1 (G) 3.6 (G) 2.6 (N) 2.7 (N) 2.9 (G) 3.1 (G) 3.7 (G) 2.5 (N) 3.6 (G) 2.3 (N) 3.1 (G) 4 (G) 4.1 (G) 2.2 2.7 
101,400,208 CTC_305I2 (RR)   

NOTE: Values in bold are QPCR data that do not correlate with aCGH data.

Chromosome comparative genomic hybridization and array comparative genomic hybridization comparisons.Table 2 compares CNAs scored by chromosome CGH and aCGH. Of 1,059 comparisons, 998 agreed and 64 (∼6%) were discordant. Forty-seven “possible” chromosome CGH gains were scored as “no change” by aCGH, 8 possible chromosome CGH losses were scored as “no change” by aCGH, and 2 “sure” chromosome CGH losses were scored as “possible” by aCGH.

Table 2.

Comparison of standard and aCGH scoring in 50 primary grade 4 astrocytoma

aCGH score
CGH score -1 -2 
23 
63 
40 47 
689 
−1 51 
−2 132 
aCGH score
CGH score -1 -2 
23 
63 
40 47 
689 
−1 51 
−2 132 

NOTE: 3, amplification; 2, sure gain; 1, possible gain; 0, normal; −1, possible loss; −2, sure loss.

FISH validation of array comparative genomic hybridization for copy number aberrations of single bacterial artificial chromosome length. We were concerned about the quality of data from individual BAC clones with RRs that did not match their BAC neighbors. Mismapping or other technical or handling errors could lead to apparent single-clone aberrations. We selected nine BAC clones displaying such single-clone aberrations in one or more cell lines and compared copy number by aCGH (Vexp) and FISH (Vobs) in four cell lines (Table 3). There was a high level of correlation between the values of Vexp and Vobs (Pearson's correlation coefficient, 0.93); however, a linear least square fit estimated the slope as 0.66 ± 0.07 (1 SD), which indicates that aCGH underestimates DNA copy number when the copy number is high. The Shapiro-Wilk normality test revealed that the ratios of Vexp/Vobs were normally distributed at a 0.05 confidence level, with an estimated mean value of 0.94 (SD, 0.15).

Table 3

Fish validation of aCGH data of BAC clones that are aberrant and do not behave like their neighbors

Clone nameChromosomeKilobase positionSF126
U343
SF767
SF210
Ex.Ob.Ra.Ex.Ob.Ra.Ex.Ob.Ra.Ex.Ob.Ra.
RP11-96L18 7p 3,000 4.8 5.6 0.85 5.4 0.73    2.72 2.15 1.26 
RP11-183O1 7p 4,525    4.1 1.0 2.8 2.75    
RP11-123E05 7p 17,196 3.6 5.55 0.65       2.2 2.1 
RP11-95N10 7p 33,764 4.9 5.5 0.9 4.1 5.2 0.8 2.5 2.1 1.2    
RP11-48B18 7p 50,884          1.8 1.8 
RP11-97H6 7p 54,384          1.6 1.95 0.82 
GS-208J7 7q 65,697       3.5 3.6 0.97    
RP11-227E4 7q 67,815       2.1 2.1    
RP11-56C16 7q 76,000 3.5 3.9 0.9    1.9 2.1 0.9 2.2 2.3 0.96 
Clone nameChromosomeKilobase positionSF126
U343
SF767
SF210
Ex.Ob.Ra.Ex.Ob.Ra.Ex.Ob.Ra.Ex.Ob.Ra.
RP11-96L18 7p 3,000 4.8 5.6 0.85 5.4 0.73    2.72 2.15 1.26 
RP11-183O1 7p 4,525    4.1 1.0 2.8 2.75    
RP11-123E05 7p 17,196 3.6 5.55 0.65       2.2 2.1 
RP11-95N10 7p 33,764 4.9 5.5 0.9 4.1 5.2 0.8 2.5 2.1 1.2    
RP11-48B18 7p 50,884          1.8 1.8 
RP11-97H6 7p 54,384          1.6 1.95 0.82 
GS-208J7 7q 65,697       3.5 3.6 0.97    
RP11-227E4 7q 67,815       2.1 2.1    
RP11-56C16 7q 76,000 3.5 3.9 0.9    1.9 2.1 0.9 2.2 2.3 0.96 

NOTE: Ratio = value expected / value observed.

Abbreviations: Ex., expected; Ob., observed; Ra., Ratio.

Copy number aberration frequency map.Figure 5 maps the frequency of CNAs at BAC loci in 50 GBM. The map does not distinguish between gain and amplification or between homozygous and heterozygous loss. The maps suggest that whole chromosome loss and gain take place primarily on chromosomes 10 (loss), 13 (loss), and 7 (gain). Approximately 75% of our samples lost the region around INK4a/ARF (9p21) and 94% had gained or amplified EGFR. Approximately 40% had gain of whole chromosome 7 and ∼60% had loss of whole chromosome 10. 6q was lost in ∼15% and chromosome 19 was gained in ∼15%. Small regions (one or more than one consecutive BAC clones) of gain and/or loss occurred on all other chromosomes.

Fig. 5

Frequency map of CNAs in 50 primary GBMs. We calculated the frequency of gain and loss at each locus, and plotted it as a function of genome position. The X-axis represents clone position from pter to qter, the Y-axis represents frequency of gain (+) or loss (−). Chromosome identifiers are indicated for each profile. Data were available only at indicated points. This map does not distinguish between gain and amplification or homozygous and heterozygous loss.

Fig. 5

Frequency map of CNAs in 50 primary GBMs. We calculated the frequency of gain and loss at each locus, and plotted it as a function of genome position. The X-axis represents clone position from pter to qter, the Y-axis represents frequency of gain (+) or loss (−). Chromosome identifiers are indicated for each profile. Data were available only at indicated points. This map does not distinguish between gain and amplification or homozygous and heterozygous loss.

Close modal

Possible amplifications and homozygous deletions. aCGH measures the boundaries of an aberration (gain, amplicon, or deletion) and provides a quantitative estimate of copy number across that region. Table 4 lists 55 BACs at 33 loci (more than one consecutive BAC can map to a single locus) wherein log2 RRs were ≥2 or ≤−1. In a diploid tumor, this theoretically represents a locus with >8 copies (RR ≥2) or <1 copy (RR ≤−1). These loci included oncogenes (e.g., EGFR, cMYC, PDGFRA, MDM2, CDK4) and tumor suppressors (e.g., INK4a/ARF, PTEN) previously reported in GBM, as well as loci not reported in earlier studies. Among the 55 BAC clones, 18 were single BAC clone losses and 2 were single BAC clone amplifications. Two consecutive BAC clones were lost at one 10q locus in one tumor; 6 consecutive BAC clones were lost at one locus on 9p in another tumor. There were six amplified regions encompassing 2 to 6 consecutive BAC clones at 1q, 3q, 4q, 7p, 7q, and 12q.

Table 4

Loci exhibiting high-level amplification or possible homozygous deletion

Clone nameLocusFrequencyGene
Amplifications    
RP11-80G24 1p31.1-1p31.2 1/47 AK5 
RP11-148K15 1q31 1/39 MDM4, LRRN5 
RP11-243M13 1q31 1/50 CNTN2, RBBP5 
RP11-118F4 3q26.3-3q27 1/49 USP13, PEX5R 
RP11-125E8 3q26.3-3q27 1/49 ECE2, PSMD2, EIF4G1, CLCN2, POLR2H, THPO, CHRD 
RP11-32K21 4q11-4q12 1/49  
RP11-210D19 4q12 1/42  
RP11-175I24 4q11-4q12 1/42 SCFD2 
RP11-98G22 4q12 1/41 CHIC2 
GS-619B4 4q11-q13 1/41 CLOCK 
RP11-80L11 4q12 1/40  
RP11-15L23 7p12 1/49 COBL 
RP11-14K11 7p11.2 11/47 EGFR 
GS-246M20 7p12 8/44 EGFR 
RP11-34J247 p11.2-7p12.1 4/49 LANCL2 
RP11-97P11 7p11.2 5/49 LANCL2 
RP11-251I15 7p11.2 5/48 FKBP9L, MRPS17, GBS 
CTD-2074H8 7q21.2-7q21.3 2/49 CDK6 
CTB-54J7 7q31 1/42 MET 
CTD-2039F11 7q31.3 1/50 MET 
CTB-185C18 7q31.2-q31.31 1/40 ASZ1 
CTD-2041G23 7q31.31-7q31.32 1/49 CTTNBP2 
GS-561N1 12q13-q14 2/48 SAS 
RP11-61P1 12q14 1/40 ARHCL1 
RP11-5J6 12q15 1/49 IFNG, IL26 
CTB-136O14 12q14 2/42 MDM2 
CTB-82N15 12q14.3-q15 2/48 MDM2 
Deletions    
RP11-60J11 1p36.2 1/46 CAMTA1 
RP11-163H6 3q26.2-3q26.3 1/50 FAD104 
RP11-6E9 4q33 1/48 AADAT 
RP11-135M13 5p14.3-5p15.1 1/39 FBXL7 
RP11-31B18 5q33.2 1/43 SGCD 
RP11-224B15 6p22.3 1/44  
RP11-43B19 6q26 2/49 LPA 
RP11-138J2 8p21 1/47 PTK2B 
GS1-261I1 8 q tel 2/49 ZNF16 
CTB-41L13 9 p tel 1/49 DOCK8 
CTB-65D18 9p21 14/49 CDKN2A 
RP11-33O15 9p21 1/34  
RP11-85J5 9p21.3 1/40  
RP11-55P9 9p21 2/46  
CTD-2170E19 9p21 1/50  
RP11-235F7 9p21 1/44 C9orf72 
RP11-17O5 10q22.2 1/49 KCNMA1 
CTB-46B12 10q23 1/47 PTEN 
RP11-129G17 10q23 1/50 C10orf59 
RP11-162A23 10q26.12 1/40 ACADSB, BUB3 
RP11-49K17 11p14.1 1/34  
RP11-15L8 11q13 1/47 PC, LRFN4, SYT12 
RP11-456I15 11q13.4 1/34 PLEKHB1, RAB6A 
RP11-42L18 11q22.3-11q23 1/49 CUL5, ACAT1, NPAT 
RP11-182D6 14q23-14q24 1/18 GPHN 
CTD-2100E13 18q11.2 2/40 RIOK3, C18orf8, NPC1, ANKRD29 
RP11-4H24 22q13.2 2/19  
CTD-2082H4 Xq27 1/41 FMR2 
Clone nameLocusFrequencyGene
Amplifications    
RP11-80G24 1p31.1-1p31.2 1/47 AK5 
RP11-148K15 1q31 1/39 MDM4, LRRN5 
RP11-243M13 1q31 1/50 CNTN2, RBBP5 
RP11-118F4 3q26.3-3q27 1/49 USP13, PEX5R 
RP11-125E8 3q26.3-3q27 1/49 ECE2, PSMD2, EIF4G1, CLCN2, POLR2H, THPO, CHRD 
RP11-32K21 4q11-4q12 1/49  
RP11-210D19 4q12 1/42  
RP11-175I24 4q11-4q12 1/42 SCFD2 
RP11-98G22 4q12 1/41 CHIC2 
GS-619B4 4q11-q13 1/41 CLOCK 
RP11-80L11 4q12 1/40  
RP11-15L23 7p12 1/49 COBL 
RP11-14K11 7p11.2 11/47 EGFR 
GS-246M20 7p12 8/44 EGFR 
RP11-34J247 p11.2-7p12.1 4/49 LANCL2 
RP11-97P11 7p11.2 5/49 LANCL2 
RP11-251I15 7p11.2 5/48 FKBP9L, MRPS17, GBS 
CTD-2074H8 7q21.2-7q21.3 2/49 CDK6 
CTB-54J7 7q31 1/42 MET 
CTD-2039F11 7q31.3 1/50 MET 
CTB-185C18 7q31.2-q31.31 1/40 ASZ1 
CTD-2041G23 7q31.31-7q31.32 1/49 CTTNBP2 
GS-561N1 12q13-q14 2/48 SAS 
RP11-61P1 12q14 1/40 ARHCL1 
RP11-5J6 12q15 1/49 IFNG, IL26 
CTB-136O14 12q14 2/42 MDM2 
CTB-82N15 12q14.3-q15 2/48 MDM2 
Deletions    
RP11-60J11 1p36.2 1/46 CAMTA1 
RP11-163H6 3q26.2-3q26.3 1/50 FAD104 
RP11-6E9 4q33 1/48 AADAT 
RP11-135M13 5p14.3-5p15.1 1/39 FBXL7 
RP11-31B18 5q33.2 1/43 SGCD 
RP11-224B15 6p22.3 1/44  
RP11-43B19 6q26 2/49 LPA 
RP11-138J2 8p21 1/47 PTK2B 
GS1-261I1 8 q tel 2/49 ZNF16 
CTB-41L13 9 p tel 1/49 DOCK8 
CTB-65D18 9p21 14/49 CDKN2A 
RP11-33O15 9p21 1/34  
RP11-85J5 9p21.3 1/40  
RP11-55P9 9p21 2/46  
CTD-2170E19 9p21 1/50  
RP11-235F7 9p21 1/44 C9orf72 
RP11-17O5 10q22.2 1/49 KCNMA1 
CTB-46B12 10q23 1/47 PTEN 
RP11-129G17 10q23 1/50 C10orf59 
RP11-162A23 10q26.12 1/40 ACADSB, BUB3 
RP11-49K17 11p14.1 1/34  
RP11-15L8 11q13 1/47 PC, LRFN4, SYT12 
RP11-456I15 11q13.4 1/34 PLEKHB1, RAB6A 
RP11-42L18 11q22.3-11q23 1/49 CUL5, ACAT1, NPAT 
RP11-182D6 14q23-14q24 1/18 GPHN 
CTD-2100E13 18q11.2 2/40 RIOK3, C18orf8, NPC1, ANKRD29 
RP11-4H24 22q13.2 2/19  
CTD-2082H4 Xq27 1/41 FMR2 

NOTE: Frequency = no. of observations / no. of cases studied. Gene(s) as located on specified BAC according to July 2003 freeze, UCSC database.

Figure 6A highlights the EGFR amplicon in this set of tumors. We observed EGFR amplification in 11 of 47 cases (Table 4). The amplicon length varied from a single BAC (lower limit ∼200 kb, upper limit ∼2 Mb) to as big as ∼5 Mb. The amplicon could extend over 2 Mb toward 7pter and up to ∼3 Mb toward 7qter. Other amplicons that covered more than 2 BAC clones were located at 1q32.1 (amplicon size ∼1 Mb), 3q26.33 (∼5.8 Mb), 4q11-12 (∼7.8 Mb), 7q31.2 (∼1.2 Mb), and 12q13.3-15 (∼11.7 Mb; Table 4). One tumor's 12q amplification mapped as three different amplicons, and a second tumor contained the 12q centromeric and telomeric amplicons only. Each 12q amplicon contained a previously implicated oncogene. The centromeric amplicon was ∼2 Mb, encompassing CDK4; the middle amplicon was ∼1 Mb, encompassing GLI1; and the telomeric amplicon was ∼4 Mb, encompassing MDM2 and SAS.

Fig. 6

EGFR amplicons and INK4a/ARF homozygous deletions in GBM. A, EGFR amplicon length and copy number differ from case to case. Three cases with different amplicon size and structure are shown here. One case has two separate amplicons. B, INK4a/ARF homozygous deletions in GBM. Length and copy number of deletions vary. Kilobase (Kb) position runs from telomere to centromere.

Fig. 6

EGFR amplicons and INK4a/ARF homozygous deletions in GBM. A, EGFR amplicon length and copy number differ from case to case. Three cases with different amplicon size and structure are shown here. One case has two separate amplicons. B, INK4a/ARF homozygous deletions in GBM. Length and copy number of deletions vary. Kilobase (Kb) position runs from telomere to centromere.

Close modal

Figure 6B displays the region of INK4a/ARF deletion in this set of tumors. Homozygous deletion at INK4a/ARF usually encompasses a single BAC clone, but it can extend up to ∼400 kb toward 9pter and ∼200 kb toward the centromere. We observed homozygous deletion of the BAC containing INK4a/ARF in 12 cases; 1 case had homozygous deletions of three consecutive BAC clones and 1 case had homozygous deletions of six consecutive BAC clones. Overall, the INK4a/ARF region was deleted in 76% of cases, and 28% were homozygous.

We detected presumed heterozygous loss of PTEN in 34 cases and homozygous loss in 1. Our data suggest that additional isolated homozygous losses occur at 1p36.2, 3q26.2-26.3, 4q33, 5p14.3-15.1, 5q33.2, 6q22.3, 8p21, 8qter, 10q22.2, 10q26.12, 11p14.1, 11q13-13.4, 11q22.3-23, 14q23-24, 18q11.2, and 22q13.2 (Table 4).

Genetic subgroups in grade 4 astrocytoma. Two major genetic subgroups (A and B + C) of grade 4 astrocytoma emerged from unsupervised clustering of aCGH data (Fig. 7). SAM analysis indicated that relative copy number at 227 BAC clones (false discovery rate <0.7%) differ in these groups: 136 clones mapped to chromosome 7, and 84 clones mapped to chromosome 10, indicating that gain of chromosome 7 and/or loss of chromosome 10 tends to occur in group B + C. The clones that differ between the genetic subgroups represent ∼70% of the those we assayed on chromosome 7 and ∼60% of those we assayed on chromosome 10. Group A (n = 23) has relatively fewer CNAs (236 + 150, ranging from 45 to 488). Group B + C is subdivided into B and C branches, and SAM analysis identified 130 clones (at a false discovery rate of 0.8%) that segregated group B from C; 112 mapped to chromosome 7, and 7 mapped to chromosome 13. Group B (n = 9) lost genetic material on chromosome 10 in all cases, and group C (n = 18) had chromosome 7 gain in all cases and chromosome 10 loss in all but two cases. Other frequently observed aberrations in grade 4 astrocytoma (e.g., loss of 9p, 13q, and 14q) were present in all three groups, but relatively more frequently in groups B and C. Gain of chromosome 19 and gain of chromosome 20 was found primarily in group C, and loss of 6q was primarily found in group B + C.

Fig. 7

Unsupervised cluster analysis of aCGH data from 50 primary GBM. Each column represents one case and each row represents one BAC clone. We calculated cutoffs, assigned values of 1, 0, and −1 for gain, no change, and loss, respectively, and clustered the tumors keeping the order of BACs intact. Losses are in green and gains are in red. Chromosome identifiers are indicated on the right. The classification tree on top suggests major genetic subgroups for GBM (groups A, B, and C).

Fig. 7

Unsupervised cluster analysis of aCGH data from 50 primary GBM. Each column represents one case and each row represents one BAC clone. We calculated cutoffs, assigned values of 1, 0, and −1 for gain, no change, and loss, respectively, and clustered the tumors keeping the order of BACs intact. Losses are in green and gains are in red. Chromosome identifiers are indicated on the right. The classification tree on top suggests major genetic subgroups for GBM (groups A, B, and C).

Close modal

Clinical data on genetic subgroups. Age and sex data were available for 50 cases, and survival data were available from 47. Figure 8 shows the Kaplan-Meier curve for survival in 47 cases divided into the three genetic groups mentioned above. There was no difference in survival among genetic groups. Average age of patients did not differ by group, although group C tended toward higher mean and median age. In a larger data set (n = 99) of GBMs, including these 50, we observed that patients with gain of chromosome 7 and loss of 10 tend to be older.5

5

A. Misra and B.G. Feuerstein, unpublished observation.

Age was a highly significant determinant for survival (P = 0.0015), whereas gender was not (P = 0.64).

Fig. 8

Kaplan-Meier survival curve for cases in different genetic groups (A, B, and C) shown in Fig. 7. The inset shows the average ± SD for age of patients in each group.

Fig. 8

Kaplan-Meier survival curve for cases in different genetic groups (A, B, and C) shown in Fig. 7. The inset shows the average ± SD for age of patients in each group.

Close modal

This study suggests that GBM can be categorized into genetic subgroups and validates aCGH for detecting CNAs in GBM. Our data also identified candidate loci for homozygous deletions and amplifications. We compared aCGH results with data obtained by chromosomal CGH, FISH, and quantitative PCR, and conclude that RR obtained from aCGH is a reliable estimate of relative copy number. Moreover, unlike chromosome CGH, aCGH detects homozygous loss and amplicon size and maps CNAs to precise locations in the genome.

Genetic subgroups. Our data suggest that there are genetically distinct subgroups within GBM (Fig. 7). We identified three provisional genetic subgroups: one with loss of chromosome 10 and gain of chromosome 7 (group C), a second with loss of chromosome 10 only (group B); and a third without chromosome 10 loss or chromosome 7 gain (group A). If the mechanisms that underlie malignant behavior in these genetic subgroups substantially differ, we expect that subgroups may behave differently and be responsive to different therapies. We have observed that subgroup C is associated with typical GBM survivors and that group A contains both typical and long-term survivors6

6

J.M. Nigro, A. Misra, L. Zhang, et al. Integrated CGH-Array and expression array profiles identify clinicallly relevant molecular subtypes of glioblastoma. Cancer Research, 2005. In press.

(20). This suggests that tumor behavior and GBM genetic subgroups are related.

Average survival did not vary among genetic groups in the present study (Fig. 8). This result may reflect a study population with few long-term survivors. Three patients in the present study were long-term survivors and lived between 2 and 3 years after diagnosis. A study of 34 cases by Nigro et al.6 included 10 long-term survivors who lived 131 to 459 weeks (average, 231 weeks) after diagnosis and revealed a relationship between survival and genetic subgroup. Thus, differences in study sample selection are likely an important factor in differences in results from these two studies.

One current hypothesis divides GBM into “primary” and “secondary” types depending on the length of symptoms. Primary GBM progresses quickly and secondary GBM has longer progression times (21). The relationship of primary and secondary tumors to the genetic subgroups we describe is unknown. p53 mutation, 10q (without 10p) loss, and loss of heterozygosity of 19q have been associated with secondary GBM, whereas EGFR amplification, p16 homozygous deletion, and PTEN mutation have been associated with primary GBM (2225). Our current data suggest that EGFR amplification occurs primarily in group C (we identified 2 in group A, 1 in B, and 8 in C) and that p16 homozygous deletion occurs in all subgroups (6 in group A, 3 in B, and 5 in C). The one case with PTEN homozygous deletion was in group C. Of nine cases that had loss of 10q without loss of 10p, 8 were in group A and 1 was in group C. Of 13 cases with substantial 19q loss, 8 belonged to group A, 2 belonged to group B, and 2 belonged to group C. This suggests that the genetic subgroup C may have a relationship to primary designation with respect to EGFR amplification and genetic subgroup A may have a relationship to secondary designation with respect to 10q loss (without 10 p loss) and 19q loss.

Age was a strong predictor of survival (P = 0.0015), as previously reported (4, 26, 27). However, we did not find any relationship between age and genetic subgroups.

Amplifications. aCGH maps amplification events very clearly. Previous reports identified EGFR amplification in 35% to 40% of grade 4 astrocytomas (2629). The differences in amplification frequency could result from different assay sensitivities or from different definitions of amplification. For example, studies using Southern blotting (28) and reverse transcription–PCR in tumor biopsy samples (26) detect a >5-fold increase in EGFR copy number in 49% of cases and a 1.2-fold increase in ∼77%. We found increased EGFR copy number in 94% of our cases. FISH-based studies report high-level double-minute EGFR amplifications in over 10% of cells in 33% of cases (27). aCGH should detect these cases as highly amplified. A recent study (30) reports EGFR amplification in isolated cells at the invasive edge of tumor. aCGH might detect this as simple copy number gain. Because we saw whole chromosome 7 gain in ∼40% of cases but EGFR copy number gain in ∼94% (say 95% and above), an extensive FISH study might identify more cases of single isolated cells with EGFR amplification. Tandem duplications of EGFR might also account for our results. These have been reported for EGFR (31), for other regions on 7q (32), and for loci on other chromosomes7

7

M.E. Law, K.L. Templeton, S. Pase, et al. Molecular cytogenetic mapping of 1p and 19q deletions in human glioma cell lines, submitted for publication.

in glioma cell lines. FISH may not be able to detect this type of genetic aberration because the extra copy could be juxtaposed to its parent, and Southern blotting may not be sensitive enough to detect the copy number difference. The biological effects of such low-level copy number increases have yet to be investigated.

Our data suggest that EGFR amplicon size varies among our tumor set (Fig. 6A), from <1 to ∼5 Mb. Because hundreds of copies of the EGFR amplicon can be present in a single GBM tumor cell (29, 33), other genes located in the amplicon could affect tumor behavior; for example, 7p11.2 contains at least eight known genes. Our data from 12q suggests a similar situation—coamplification of CDK4, GLI1, and MDM2 can depend on amplicon size. This situation also needs to be investigated at loci on 1q32.1 (GAC1; ref. 34), 3q26.32 (PIK3CA; ref. 35), 4q12 (PDGFRA; ref. 36), 7q31.2 (MET; ref. 37), and 12q13.3-15 (CDK4, GLI1, MDM2, SAS, PIKE; refs. 38, 39). The UCSC database (July 2003 freeze) suggests the amplicon at 1q32 contains 3 other known genes, at 3q26 contains 19 known and 3 nonannotated genes, at 4q12 contains 4 known and 4 nonannotated genes, at 7q31.s contains 7 known genes, and at 12q13.3-15 contains 33 known and 6 nonannotated genes. The biological and clinical significance of amplicon size and structure has not been thoroughly investigated.

Our data identify gains and amplifications at several loci linked to oncogenes in other tumors, but not frequently associated with GBM. For example, gain of loci at 1q31.1-31.2 and 7q21 occurs in many tumors, and amplifications at these sites occur in a few (Table 4; Fig. 5). At least two proto-oncogenes, jun (40, 41) and CYP2J (42), are located at 1q31.1-31.2 and CDK6 is located at 7q21 (43).

Deletions. aCGH distinguishes homozygous deletions from single-copy losses (Fig. 3). Our aCGH results suggest that 28% of our GBM had homozygous INK4a/ARF deletions (Fig. 6B), and 48% had single-copy deletions. Other studies have reported homozygous INK4a/ARF deletions in 33% to 55% of GBM (4446) and ∼70% of GBM xenografts (47), hemizygous deletion in ∼12% (45, 48), and possible epigenetic inactivation in 3% to 24% (45, 49). Our measurements evaluate genomic 100- to 150-kb BAC inserts, so that homozygous deletions much smaller than the BAC size might be missed. This may in part account for our relatively low frequency of homozygous deletions. The INK4Aα gene is a tumor suppressor on 9p targeted for homozygous loss in GBM. We find that the size of the homozygous deletion varies. It is possible that other genes located in this region affect tumor behavior, consistent with the hypothesis that more than one tumor suppressor gene is located on 9p (5053).

aCGH indicated that 34 cases lost a single copy of the tumor suppressor gene PTEN, 1 case lost both copies, and 12 cases had no relative loss (data not available in 3 cases). The results agree with previous reports of 70% to 80% loss of the PTEN region (8). Because mutation of PTEN occurs in <10% to 30% of primary GBM (54, 55), other mechanisms of PTEN inactivation or AKT activation may exist. For example, there are likely cases of phosphatidylinositol 3-kinase amplification in our tumor set. We have not found literature that reports the frequency of activated AKT in GBM.

There have been reports suggesting that cancers inactivate tumor suppressor genes at 11p13-14.1 (56) and 14q23.3-24.1 (57). We mapped both heterozygous and homozygous losses in these regions (Table 4; Fig. 5). However, many candidate loci we mapped (e.g., 4q33, 5p14.3-15.1, 5q33.2, 6p22.3, 8p21, 8qter, 10q22.2, 10q26.12, 11q13.4, and 14q23-24) are not associated with known tumor suppressors. These sites need validation and further study.

Frequency map. We derived a map of CNA frequency (Fig. 5) in GBM with an average resolution of ∼1.4 Mb. The map suggests that several mechanisms of genomic instability operate in GBM. These include whole chromosome gain and loss (e.g., chromosomes 7, 10, and 13) and smaller regions of loss and gain (e.g., loss on chromosome 3, gain on 12q). Other studies have suggested that several mechanisms target whole chromosome or parts of chromosomes in tumors. (58, 59). Our data suggest that certain portions of chromosomes are gained (e.g., 7p11-2 EGFR and 7q35-6) or lost (e.g., regions on 10p, 10q, and Xp) at substantially different frequencies than neighboring regions. This suggests that local genetic instability plays an important role in genetic selection of these tumors. This view is supported by the variability we observe in relationships of clones on chromosome 7 and 10 to genetic subgroups.

Technical issues and data interpretation. It is important to recognize the limitations of the frequency map presented in Fig. 5. Some BAC clones may be mapped inaccurately or hybridize poorly. A priori identification of such clones is not easy, so validation of these findings by other methods is important. We avoided including data from BAC clones with suspicious behavior and questionable mapping information. As a whole, we expect our frequency map to be ∼90% accurate because 90% of the BAC clones used were FISH-mapped. Our validation studies with FISH, quantitative PCR, and chromosome CGH all suggested excellent agreement.

One issue for our frequency map is the reliability of CNAs limited to isolated single BAC clones. To address this issue, we selected nine such BAC clones and validated them by FISH (Table 3). We found that aCGH provides generally reliable estimates of relative DNA copy number, although it tends to underestimate higher copy number. We did find that FISH copy number for BACs RP11-96L18, RO11-123E05, and RP11-95N10 were aberrant. Based on these observations, we believe that isolated single aberrant BAC clones indicate a real loss or gain in >50% of cases.

More than 90% of chromosome CGH scores agreed with aCGH results (Table 2). Disagreements were infrequent and were matters of degree. Obvious differences between chromosome CGH and aCGH that could produce such disagreements include (a) hybridization dynamics—chromosome CGH displays CNAs on single metaphase chromosomes and aCGH displays them on 100- to 200-kb segments of the human genome cloned into BACs and (b) cutoffs—aCGH cutoffs were calculated from objective criteria, whereas analysis of chromosome CGH depended on more subjective observations.

These experiments detail genetic CNAs detected in a series of GBM. The CNAs support previous work that describes oncogenes and tumor suppressors associated with this tumor, and suggest several candidate loci that need to be validated and further investigated. Our data also suggest that there are genetic subgroups within GBM. This raises the possibility that each subgroup may need different therapies for optimal treatment.

Grant support: NIH grants CA85799, NS42927, and NS40958, the National Brain Tumor Foundation, and the North American Brain Tumor Consortium.

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 Sharon Reynolds for editorial support and the array core at the UCSF Cancer Center for providing arrays and analysis infrastructure.

1
Cairncross JG, Ueki K, Zlatescu MC, et al. Specific genetic predictors of chemotherapeutic response and survival in patients with anaplastic oligodendrogliomas.
J Natl Cancer Inst
1998
;
90
:
1473
–9.
2
Huhn SL, Mohapatra G, Bollen A, Lamborn K, Prados MD, Feuerstein BG. Chromosomal abnormalities in glioblastoma multiforme by comparative genomic hybridization: correlation with radiation treatment outcome.
Clin Cancer Res
1999
;
5
:
1435
–43.
3
Kunwar S, Mohapatra G, Bollen A, Lamborn KR, Prados M, Feuerstein BG. Genetic subgroups of anaplastic astrocytomas correlate with patient age and survival.
Cancer Res
2001
;
61
:
7683
–8.
4
Simmons ML, Lamborn KR, Takahashi M, et al. Analysis of complex relationships between age, p53, epidermal growth factor receptor, and survival in glioblastoma patients.
Cancer Res
2001
;
61
:
1122
–8.
5
Ino Y, Betensky RA, Zlatescu MC, et al. Molecular subtypes of anaplastic oligodendroglioma: implications for patient management at diagnosis.
Clin Cancer Res
2001
;
7
:
839
–45.
6
Bissola L, Eoli M, Pollo B, et al. Association of chromosome 10 losses and negative prognosis in oligoastrocytomas.
Ann Neurol
2002
;
52
:
842
–5.
7
Mohapatra G, Kim DH, Feuerstein BG. Detection of multiple gains and losses of genetic material in ten glioma cell lines by comparative genomic hybridization.
Genes Chromosomes Cancer
1995
;
13
:
86
–93.
8
Mohapatra G, Bollen AW, Kim DH, et al. Genetic analysis of glioblastoma multiforme provides evidence for subgroups within the grade.
Genes Chromosomes Cancer
1998
;
21
:
195
–206.
9
Rutka JL, Giblin JR, Dougherty DY, et al. Establishment and characterization of five cell lines derived from human malignant gliomas.
Acta Neuropathol (Berl)
1987
;
75
:
92
–103.
10
Burger PC, Scheithauer BW. Tumors of the central nervous system. In: Atlas of tumor pathology. Washington (DC): Armed Forces Institute of Pathology; 1994. p. 160–378.
11
Snijders AM, Nowak N, Segraves R, et al. Assembly of microarrays for genome-wide measurement of DNA copy number.
Nat Genet
2001
;
29
:
263
–4.
12
Misra A, Pellarin M, Shapiro JR, Feuerstein BG. A complex rearrangement of chromosome 7 in human astrocytoma.
Cancer Genet Cytogenet
2004
;
151
:
162
–70.
13
Sambrook J, Fritsch EF, Maniatis T. Preparation of radiolabeled DNA and RNA probes. In: Molecular cloning. 2nd ed. Cold Spring Harbor: Cold Spring Harbor Laboratory; 1989. p. e3–4; 10.8–10.10;10.14–10.15.
14
Jain AN, Tokuyasu TA, Snijders AM, Segraves R, Albertson DG, Pinkel D. Fully automatic quantification of microarray image data.
Genome Res
2002
;
12
:
325
–32.
15
Hodgson G, Hager JH, Volik S, et al. Genome scanning with array CGH delineates regional alterations in mouse islet carcinomas.
Nat Genet
2001
;
29
:
459
–64.
16
Mohapatra G, Moore DH, Kim DH, et al. Analyses of brain tumor cell lines confirm a simple model of relationships among fluorescence in situ hybridization, DNA index, and comparative genomic hybridization.
Genes Chromosomes Cancer
1997
;
20
:
311
–9.
17
Nigro JM, Takahashi MA, Ginzinger DG. Detection of 1p and 19q loss in oligodendroglioma by quantitative microsatellite analysis, a real-time quantitative polymerase chain reaction assay.
Am J Pathol
2001
;
158
:
1253
–62.
18
Eisen MB, Spellman PT, Brown PO, Botstein D. Cluster analysis and display of genome-wide expression patterns.
Proc Natl Acad Sci U S A
1998
;
95
:
14863
–8.
19
Tusher VG, Tibshirani R, Chu G. Significance analysis of microarrays applied to the ionizing radiation response.
Proc Natl Acad Sci U S A
2001
;
98
:
5116
–21. Epub 2001
20
Nigro JM, Misra A, Zhang L, et al. Molecular subgroups of human glioblastoma multiforme correlate with survival.
Neuro-oncol
2003
;
4
:
313
.
21
Kleihues P, Ohgaki H. Primary and secondary glioblastomas: from concept to clinical diagnosis.
Neuro-oncol
1999
;
1
:
44
–51.
22
Biernat W, Debiec-Rychter M, Kordek R, Liberski PP. TP53 mutations in malignant astrocytomas.
Pol J Pathol
1997
;
48
:
221
–4.
23
Nakamura M, Yang F, Fujisawa H, Yonekawa Y, Kleihues P, Ohgaki H. Loss of heterozygosity on chromosome 19 in secondary glioblastomas.
J Neuropathol Exp Neurol
2001
;
59
:
539
–43.
24
Fujisawa H, Reis RM, Nakamura M, et al. Loss of heterozygosity on chromosome 10 is more extensive in primary (de novo) than in secondary glioblastomas.
Lab Invest
2000
;
80
:
65
–72.
25
Reis RM, Herva R, Brandner S, et al. Second primary glioblastoma.
J Neuropathol Exp Neurol
2001
;
60
:
208
–15.
26
Burger PC, Green SB. Patient age, histologic features, and length of survival in patients with glioblastoma multiforme.
Cancer
1987
;
59
:
1617
–25.
27
Lamborn KR, Chang SM, Prados MD. Prognostic factors for survival of patients with glioblastoma: recursive partitioning analysis.
Neuro-oncol
2004
;
6
:
227
–35.
28
Olson JJ, James CD, Krisht A, Barnett D, Hunter S. Analysis of epidermal growth factor receptor gene amplification and alteration in stereotactic biopsies of brain tumors.
Neurosurgery
1995
;
36
:
740
–8.
29
Sauter G, Maeda T, Waldman FM, Davis RL, Feuerstein BG. Patterns of epidermal growth factor receptor amplification in malignant gliomas.
Am J Pathol
1996
;
148
:
1047
–53.
30
Okada Y, Hurwitz EE, Esposito JM, Brower MA, Nutt CL, Louis DN. Selection pressures of TP53 mutation and microenvironmental location influence epidermal growth factor receptor gene amplification in human glioblastomas.
Cancer Res
2003
;
63
:
413
–6.
31
Ciesielski MJ, Fenstermaker RA. Oncogenic epidermal growth factor receptor mutants with tandem duplication: gene structure and effects on receptor function.
Oncogene
2000
;
19
:
810
–20.
32
de Silva MG, Kantharidis P, Scherer SW, et al. Physical mapping of a tandem duplication on the long arm of chromosome 7 associated with a multidrug resistant phenotype.
Cancer Genet Cytogenet
1999
;
110
:
28
–33.
33
Liu W, James CD, Frederick L, Alderete BE, Jenkins RB. PTEN/MMAC1 mutations and EGFR amplification in glioblastomas.
Cancer Res
1997
;
57
:
5254
–7.
34
Almeida A, Zhu XX, Vogt N, et al. GAC1, a new member of the leucine-rich repeat superfamily on chromosome band 1q32.1, is amplified and overexpressed in malignant gliomas.
Oncogene
1998
;
16
:
2997
–3002.
35
Shayesteh L, Lu Y, Kuo WL, et al. PIK3CA is implicated as an oncogene in ovarian cancer.
Nat Genet
1999
;
21
:
99
–102.
36
Smith JS, Wang XY, Qian J, et al. Amplification of the platelet-derived growth factor receptor-A (PDGFRA) gene occurs in oligodendrogliomas with grade IV anaplastic features.
J Neuropathol Exp Neurol
2000
;
59
:
495
–503.
37
Soman NR, Wogan GN, Rhim JS. TPR-MET oncogenic rearrangement: detection by polymerase chain reaction amplification of the transcript and expression in human tumor cell lines.
Proc Natl Acad Sci U S A
1990
;
87
:
738
–42.
38
Khatib ZA, Matsushime H, Valentine M, Shapiro DN, Sherr CJ, Look AT. Coamplification of the CDK4 gene with MDM2 and GLI in human sarcomas.
Cancer Res
1993
;
53
:
5535
–41.
39
Ahn JY, Rong R, Kroll TG, Van Meir EG, Snyder SH, Ye K. PIKE (phosphatidylinositol 3-kinase enhancer)-A GTPase stimulates Akt activity and mediates cellular invasion.
J Biol Chem
2004
;
279
:
16441
–51.
40
Hattori K, Angel P, Le Beau MM, Karin M. Structure and chromosomal localization of the functional intronless human JUN protooncogene.
Proc Natl Acad Sci U S A
1988
;
85
:
9148
–52.
41
Haluska FG, Huebner K, Isobe M, Nishimura T, Croce CM, Vogt PK. Localization of the human JUN protooncogene to chromosome region 1p31-32.
Proc Natl Acad Sci U S A
1998
;
85
:
2215
–8.
42
Ma J, Ramachandran S, Fiedorek FT Jr, Zeldin DC. Mapping of the CYP2J cytochrome P450 genes to human chromosome 1 and mouse chromosome 4.
Genomics
1998
;
49
:
152
–5.
43
Costello JF, Plass C, Arap W, et al. Cyclin-dependent kinase 6 (CDK6) amplification in human gliomas identified using two-dimensional separation of genomic DNA.
Cancer Res
1997
;
57
:
1250
–4.
44
Moulton T, Samara G, Chung WY, et al. MTS1/p16/CDKN2 lesions in primary glioblastoma multiforme.
Am J Pathol
1995
;
146
:
613
–9.
45
Izumoto S, Arita N, Ohnishi T, et al. Homozygous deletions of p16INK4A/MTS1 and p15INK4B/MTS2 genes in glioma cells and primary glioma tissues.
Cancer Lett
1995
;
97
:
241
–7.
46
Fulci G, Labuhn M, Maier D, et al. p53 gene mutation and ink4a-arf deletion appear to be two mutually exclusive events in human glioblastoma.
Oncogene
2000
;
19
:
3816
–22.
47
Jen J, Harper JW, Bigner SH, et al. Deletion of p16 and p15 genes in brain tumors.
Cancer Res
1994
;
54
:
6353
–8.
48
Hegi ME, zur Hausen A, Ruedi D, Malin G, Kleihues P. Hemizygous or homozygous deletion of the chromosomal region containing the p16INK4a gene is associated with amplification of the EGF receptor gene in glioblastomas.
Int J Cancer
1997
;
73
:
57
–63.
49
Costello JF, Berger MS, Huang HS, Cavenee WK. Silencing of p16/CDKN2 expression in human gliomas by methylation and chromatin condensation.
Cancer Res
1996
;
56
:
2405
–10.
50
Giordani L, Iolascon A, Servedio V, Mazzocco K, Longo L, Tonini GP. Two regions of deletion in 9p22-p24 in neuroblastoma are frequently observed in favorable tumors.
Cancer Genet Cytogenet
2002
;
135
:
42
–7.
51
Grady B, Goharderakhshan R, Chang J, et al. Frequently deleted loci on chromosome 9 may harbor several tumor suppressor genes in human renal cell carcinoma.
J Urol
2001
;
166
:
1088
–92.
52
Perinchery G, Bukurov N, Nakajima K, Chang J, Li LC, Dahiya R. High frequency of deletion on chromosome 9p21 may harbor several tumor-suppressor genes in human prostate cancer.
Int J Cancer
1999
;
83
:
610
–4.
53
Kim SK, Ro JY, Kemp BL, et al. Identification of three distinct tumor suppressor loci on the short arm of chromosome 9 in small cell lung cancer.
Cancer Res
1997
;
57
:
400
–3.
54
Liu W, James CD, Frederick L, Alderete BE, Jenkins RB. PTEN/MMAC1 mutations and EGFR amplification in glioblastomas.
Cancer Res
1997
;
57
:
5254
–7.
55
Sano T, Lin H, Chen X, et al. Differential expression of MMAC/PTEN in glioblastoma multiforme: relationship to localization and prognosis.
Cancer Res
1999
;
59
:
1820
–4.
56
Bepler G, Garcia-Blanco MA. Three tumor-suppressor regions on chromosome 11p identified by high-resolution deletion mapping in human non-small-cell lung cancer.
Proc Natl Acad Sci U S A
1994
;
91
:
5513
–7.
57
Hu J, Jiang C, Ng HK, Pang JC, Tong CY. Chromosome 14q may harbor multiple tumor suppressor genes in primary glioblastoma multiforme.
Chin Med J (Engl)
2002
;
115
:
1201
–4.
58
Charames GS, Bapat B. Genomic instability and cancer.
Curr Mol Med
2003
;
3
:
589
–96.
59
Storchova Z, Pellman D. From polyploidy to aneuploidy, genome instability and cancer.
Nat Rev Mol Cell Biol
2004
;
5
:
45
–54.