High-Resolution Genome-Wide Mapping of Genetic Alterations in Human Glial Brain Tumors

Introduction

Adult gliomas encompass a highly lethal group of tumors that includes astrocytomas, oligodendrogliomas, and oligoastrocytomas. Genomic DNA copy number aberrations are key genetic events in gliomagenesis. Recurrent genomic regions of alteration in copy number, including net gains and losses, have been found in these neoplasms. Whereas some of these regions contain known (or candidate) oncogenes and tumor suppressor genes, the biologically relevant genes within other regions remain to be identified ( 1).

Comparative genomic hybridization (CGH) has been used to analyze DNA copy number changes in various human cancers, including gliomas ( 2, 3). This karyotype-based method, however, has limited mapping resolution, and gains or losses must be several megabases in size to be detected. Microarray-based CGH (array-CGH) provides a higher-resolution means to map DNA copy number alterations ( 4). cDNA microarrays in particular permit gene-by-gene analysis of aberrations in gene copy number. Here, we have used 42,000-element array-CGH technology with the aim to generate highly precise and comprehensive gene copy number profiles in a cohort of 54 gliomas of various histogenesis and tumor grade. The generated high-resolution genome-wide maps allowed delineating the precise (gene specific) boundaries of known and new chromosomal alterations, which is not feasible by classic chromosomal CGH. We show that gliomas can be clustered into distinct subgroups based on their genetic profiles, which include recurrent patterns of interrelated chromosomal changes. The alteration of a subset of genes can predict astrocytic and oligodendroglial tumor phenotypes. Finally, we have identified in a subset of gliomas five common deleted regions that involve potential candidate tumor suppressor genes.

Materials and Methods

Tumor specimens. Fifty-four fresh-frozen glioma specimens were collected under Institutional Review Board–approved guidelines and subjected to standard WHO classification ( 5). Specimens included astrocytic [3 juvenile pilocytic astrocytomas, 1 low-grade astrocytic glioma, 3 anaplastic astrocytomas, 31 glioblastomas (of these 3 secondary glioblastomas and 2 gliosarcomas)], oligodendroglial (5 oligodendrogliomas, 3 anaplastic oligodendrogliomas), and seven anaplastic oligoastrocytomas tumors. One tumor had been classified as glioneuronal neoplasm. Human male and female genomic reference DNA was purchased from Promega (Madison, WI). Genomic DNA was isolated using the DNeasy Tissue Kit (Qiagen, Valencia, CA), DPNII (New England Biolabs, Beverly, MA) digested, and purified using the QIAquick PCR Purification Kit (Qiagen).

DNA labeling and microarray hybridizations. Labeling of digested DNA and microarray hybridizations were done essentially as described ( 4), with slight modifications. Two micrograms of DNA were labeled using random primers (Bioprime Labeling Kit, Invitrogen, Carlsbad, CA). Tumor DNA and reference DNA were fluorescently labeled with Cy5 (red) and Cy3 (green) dye (Amersham Biosciences, Piscataway, NJ), respectively. Tumor DNA was hybridized together with sex-matching reference DNA to a Stanford human cDNA microarray containing 41,421 cDNA elements, corresponding to 27,290 different UniGene cluster IDs.

Data analysis. Microarrays were scanned on a GenePix 4000B scanner (Axon Instruments, Union City, CA). Primary data collection was done using GenePix Pro 5.1 software. Raw data were deposited into the Stanford Microarray Database. Measurements with consistent (regression correlation, >0.6) and sufficient fluorescent intensities (reference wavelength channel, >2.5 above background) were considered reliable. Raw element intensities were background corrected and normalized using SNOMAD data analysis tools (http://pevsnerlab.kennedykrieger.org/snomad.htm). Gene copy numbers were reported as a moving average (symmetrical 3-/5-/7-nearest neighbors).

The GoldenPath Human Genome Assembly (http://genome.ucsc.edu, National Center for Biotechnology Information build 34) was used to map log intensity ratios of the arrayed human cDNAs to chromosomal positions. The CaryoScope (http://genome-www5.stanford.edu/cgi-bin/caryoscope/nph-aCGH-dev_update.pl) and TreeView software ( 6) were used to display gene copy number ratios along the human genome. Altered regions were also identified and visualized by the CGH-Plotter MATLAB toolbox, by means of mean filtering, k-means clustering, and dynamic programming ( 7).

Unsupervised hierarchical clustering was done in Cluster ( 6), and two-way complete linkage clustering based on Pearson correlation as distance metric was applied. A correlation matrix representing all gene-to-gene correlations was constructed in MATLAB using the built-in corrcoef function. Supervised class prediction analysis was done using the nearest shrunken centroids method implemented in the prediction analysis of microarrays package ( 8). Class predictive genes were identified based on minimal misclassification error in balanced 10-fold cross-validation.

Real-time PCR. Quantitative real-time PCR reactions were done with the ABI Prism 7900HT Sequence Detection System using SYBR GREEN PCR Master Mix (Applied Biosystems, Foster City, CA). Primers targeting introns of the TOPORS, FANCG, RAD51, TP53BP1, BIK, and ADAR genes were designed with the Primer3 program (http://frodo.wi.mit.edu/cgi-bin/primer3/primer3_www.cgi) and synthesized at the Stanford PAN Facility (for sequences, see Supplementary Fig. S10). Thermocycling for each reaction was carried out in a final volume of 20 μL containing 10 ng of genomic DNA, forward and reverse primers at 300 nmol/L final concentration, and 1× SYBR GREEN PCR Master Mix. After 10 minutes of initial denaturation at 95°C, the cycling conditions of 40 cycles consisted of denaturation at 95°C for 15 seconds, annealing at 55°C for 30 seconds, and elongation at 72°C for 30 seconds. All reactions were done in triplicate. Dissociation curve analysis was done after every run to confirm the primer specificity. Gene quantities were determined using standard curves, constructed by five serial dilutions of normal human genomic DNA (Promega), and gene copy numbers were reported as ratios of quantities of the target gene and ADAR as the reference gene, which was unaltered in all tumors.

Results

High-resolution genome-wide gene copy number profiling. To take full advantage of the high spatial resolution of array-CGH, we generated profiles of gene copy number changes across all 34,538 cDNA elements that passed quality-filtering criteria. Clones with altered fluorescent ratio represented genes of potential interest and also provided precise genomic landmarks for altered chromosomal regions. Figure 1A (right) shows a gene copy number heatmap in a highly compressed manner with clones ordered along the genome and tumors grouped into WHO grade I to III astrocytoma, glioblastoma, anaplastic oligoastrocytoma, and WHO grade II and III oligodendroglioma (see also Supplementary Fig. S6). Numerous genomic alterations consisting of gains and losses of large chromosomal regions were detected. Recurrent regions of DNA copy number alteration were readily identifiable and were consistent with published cytogenetic studies ( 1, 2). Cumulative recurrence frequencies of these chromosomal alterations for all tumors are reported in Fig. 1A (left). For example, whereas losses of chromosomes 10 and of large portions of chromosomes 13 and 22 were particularly present in glioblastomas, losses of chromosomal arms 1p and 19q were primarily apparent in oligodendroglial tumors. Gains of whole chromosomes 19 and 20 or of chromosomal arm 19p were almost exclusively observed in glioblastomas. Gains of chromosome 7 were seen in all four tumor subgroups but only coincided with loss of chromosome 10 in glioblastomas. Losses of terminal 9p were frequently observed in the glioblastoma, anaplastic oligoastrocytoma, and oligodendroglioma subgroups but were particularly associated with concurrent gains of terminal 8q or whole chromosome 8 in the anaplastic oligoastrocytoma subgroup (Supplementary Fig. S7). Gain of whole chromosomes 5 and 6 were almost exclusively present in low-grade astrocytic tumors. In the anaplastic oligoastrocytoma and oligodendroglioma groups, five tumors showed terminal gains on 11q of varying size. In the same subgroups, each three tumors showed extensive gains of 10p and terminal 3p, the latter of which was associated with gains of terminal 12p in two cases.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1.

High-resolution genome-wide map in human gliomas by array-CGH. A, highly compressed genomic heatmap of 54 gliomas (right), in which ∼35,000 cDNA clones are ordered by position along the chromosomes and are mean filtered according to moving 5-mb windows. Genomic profiles are ordered separately for WHO grade I to III astrocytic tumors (A), glioblastomas (GBM), oligoastrocytic (OA), and oligodendroglial (O) tumors. Chromosomal gain (red) and loss (green). Left, corresponding integer value recurrences of chromosomal alterations plotted for all tumors and aligned in genome order. Gain (dark red bars) or loss (dark green bars) of chromosomal material. B, gene copy number profiles of target regions (symmetrical 3-nearest neighbor mean filtering) selected based on previous implication in gliomagenesis. Note the varying extent of the altered regions in different tumors. Presumed target genes are indicated with color-coded text. LGA, low-grade astrocytoma. AA, anaplastic astrocytoma; AOA, anaplastic oligoastrocytoma; AO, anaplastic oligodendroglioma.

Characterization of known critical amplicons and deletions. Figure 1B shows selected genomic regions that have been strongly implicated in gliomagenesis (see also Supplementary Fig. S8). In each of the regions, the gene primarily implicated as the “driving” target gene has been color-coded in red (amplicon) or green (deletion). The gene-by-gene nature of our approach permitted the dissection of exact amplicon and deletion boundaries within these regions and thus the identification of coaltered genes, some of which may contribute to tumorigenesis. For example, PDGFRA was coamplified with the oncogene KIT in two tumors and with the vascular endothelial growth factor receptor gene KDR and the IGFBP7 gene (data not shown) in one tumor. Several EGFR amplicons included the GBAS gene ( 9) and two glioblastomas showed (noncontiguous) coamplification of the IGFBP1 and IGFBP3 genes, distal to EGFR. The CDK4 amplicon partly included the GEFT ( 10), OS-9 ( 11), and AKT-stimulating CENTG1 ( 12) genes, and the candidate oncogene CTDSP2 ( 13). The two glioblastomas with MDM2 amplification showed coamplification of the putative oncogene SLC35E3 and were those which also showed the (noncontiguous) CDK4 amplicon. One glioblastoma showed an amplified segment immediately distal of MDM2, involving the GAS41 ( 14) and FRS2 ( 15) genes, and the hypothetical gene MGC13168, the latter of which was amplified in an additional glioblastoma. The CDKN2A locus showed codeletion of the candidate tumor suppressor gene MTAP ( 16). Whereas the MYCN amplicon in one glioblastoma included the MYCNOS ( 17), NAG ( 18), DDX1 ( 19), and NSE1 ( 20) genes, the MYC amplicon involved the NSE2 gene (ref. 20; data not shown). Tumors with deletion of the WDR11 gene, implicated in a subset of glioblastoma ( 21), showed codeletion of the FGFR2 gene. DCC deletions were associated with codeletion of the MBD2 gene, which is underexpressed in various cancers ( 22). The PDGFB amplicon in one tumor showed coamplification of the CBX7 gene, which down-regulates CDKN2A expression ( 23).

Genetic subgrouping and correlation matrix analysis. We used unsupervised cluster analysis to evaluate whether gliomas could be classified into distinct subgroups based on their genomic profiles. Here, the data set was prefiltered to include only those fluorescent ratios indicating genes that were either hemizygously deleted or at least 2.5-fold amplified. Symmetrical 7-nearest neighbors averaging yielded 1,767 clones fulfilling these threshold criteria and generated optimal signal-to-noise relations with regard to the analysis of larger chromosomal segments. This approach revealed a genetic subgrouping of the tumors into five classes ( Fig. 2A ): (I) preferential losses of 1p and 19q; (II) gains of 7 and loss of 10 and preferential EGFR amplification; (III) gains of 7 and loss of 10 and additional genetic alterations; (IV) partial gains of 7 and other genetic alterations; and (V) no gains of 7 but other genetic alterations. Although the −1p/−19q class was significantly enriched for oligodendroglial tumors (P = 0.0002, Fisher's exact test) and the +7/−10/+EGFR class only included astrocytic tumors (P = 0.01), overall, there was striking overlap in tumor phenotypes and tumor grades between the genetic subgroups.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2.

Genetic subgrouping and association mapping in human gliomas. A, results of unsupervised two-way linkage clustering of gliomas based on 1,767 cDNA clones with gene copy number ratios indicating at least hemizygous deletions or ≥2.5-fold amplifications (symmetrical 7-nearest neighbor averaging). B, correlation matrix mapping associations between different genomic alterations. Pearson's correlation coefficients for all possible associations between different genetic events are shown below the diagonal. Red, positive correlations (i.e., coappearance of two genetic events); blue, negative correlations (including coappearance of inversely related coevents and mutual exclusive events). Major associations between different genomic regions are highlighted above the diagonal and are color-transformed to distinguish between inverse coevents and mutually exclusive events.

We then examined the relationship between occurrences of different genetic alterations among the tumors, independent of subgrouping. We created an association map representing all gene-to-gene correlations within the data set, including positive coevents, inversely related coevents, and mutually exclusive events ( Fig. 2B). This analysis revealed highly correlated behavior of clones belonging to the same amplicon or deletion as well as interrelationships among distant genetic aberrations, the latter confirming the nonrandom coappearance of distinct genetic events in different gliomas. The most striking relationships included colosses on 1p and 19q, combined gains on 7 and losses on 10, mutual exclusiveness of losses on 10 and losses on 19q, combined gains on 8(q) and losses on 9p, coamplification of regions on 3p and 12p, and combined losses on 10 and gains on 20.

Genetic alterations predicting glioma phenotype. We did class prediction analysis using nearest shrunken centroids to examine whether the alteration of a subgroup of genes may predict glioma phenotype. The same threshold filtering (<0.5-fold deleted or >2.5-fold amplified) was used as in the unsupervised learning algorithm; however, a smaller moving average window size (symmetrical 3-nearest neighbors; total of 8,433 cDNA clones) was chosen, enabling to report changes for single genes rather than larger chromosomal segments. Tumors were grouped according to pure astrocytic, mixed oligoastrocytic, and pure oligodendroglial histology. This analysis identified a set of 170 genes (Supplementary Table S1) whose genetic alterations accurately predicted the phenotype of 35 of 37 pure astrocytic and seven of eight pure oligodendroglial tumors (class error rates of 0.078 and 0.125; Supplementary Fig. S9). As expected by the shared genetic alterations of mixed oligoastrocytomas with both pure phenotypic counterparts, these tumors were not separately classifiable. Figure 3 shows the 170 genes ordered by position along the genome and the recurrence of their alteration in the three morphologic subgroups. Clusters of altered genes on chromosomes 1p, 7 (including EGFR), 10, and 19q were associated with glioma phenotype. A yet unknown gene (FLJ23129) on 1p31.2 had the highest predictive value.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3.

Phenotype-associated genetic events in human gliomas. Supervised class prediction analysis using the nearest shrunken centroids method identified a subset of 170 genes whose alterations distinguished pure astrocytic (A) from pure oligodendroglial (O) tumor morphology (class error rates of 0.078 and 0.125, respectively). As expected by the shared genetic alterations of mixed oligoastrocytomas (OA) with both pure phenotypic counterparts, these tumors were not separately classifiable. Genes are ordered by map position along the genome (right). Left, recurrence of alteration (positive, amplification; negative, deletion) for each gene in the three morphological subgroups, indicating the importance of genetic changes on chromosomes 1p, 7, 10, and 19q in predicting tumor phenotype. Astrocytic (filled blue bars), oligodendroglial (filled orange bars), and oligoastrocytic (empty red bars) subgroups.

Common minimally altered regions. We then searched the genomes of all the tumors to detect new genetic alterations that could be potentially associated with gliomagenesis. This analysis identified multiple recurrent regions of gene copy number imbalance (see Supplementary Table S2). The identification of many copy number alterations, along with the high degree of structural complexity within each altered locus, prompted the implementation of objective criteria to define and prioritize minimal common regions across the data set. Applying a strict locus identification algorithm, alterations present in at least two tumors and encompassing a minimum of three adjacent cDNA clones with fluorescent ratios indicating ≤0.5- or ≥2.0-fold changes in gene copy number were considered. Each locus is characterized by a peak profile, the width and amplitude of which reflect the contour of the most prominent gene alteration for that locus. Five common deleted regions contained putative tumor suppressor genes within the peak position of the minimally altered segment and were therefore considered as high-priority loci ( Fig. 4 ). Of these, two noncontiguous regions mapped several megabases distal of the CDKN2A locus (9p21.3) to the short arm of chromosome 9 (9p21.1 and 9p13.3) and included the candidate tumor suppressor genes TOPORS and FANCG, respectively. Two noncontiguous minimally deleted regions mapped to 15q15.1 and 15q15.3, involving the RAD51 and the TP53BP1 gene loci. Another minimal common deletion area mapped to the BIK gene locus on 22q13.2 ( Fig. 4). Real-time quantitative PCR precisely mirrored (100% concordance) the gene copy numbers for all five candidate genes relative to the reference gene ADAR in panels of tumors with and without the minimally deleted regions in question, confirming their deletion in representative tumors ( Fig. 5 ).

• Download figure
• Open in new tab
• Download powerpoint

Figure 4.

Mapping of common minimally deleted regions in human gliomas. CGH profiles (right) for five chromosomal loci are plotted on the y-axis by genomic map positions. Integer value recurrence of deletions for plotted cDNA clones (left). Minimal common regions for displayed loci (megabase map coordinates in the human genome sequence provided in parentheses), the bonds of which delineate the combined physical extent of overlapping genetic alterations across different tumors (square brackets), peak profiles (red vertical bars). Candidate tumor suppressor genes TOPORS, FANCG, RAD51, TP53BP1, and BIK (green). Note the peak profiles of the recurrence plots match mapping positions of these candidate tumor suppressor genes (both sides, green vertical bars). Fluorescent ratios between log₂(−0.3) and log₂(0.3) have been masked.

• Download figure
• Open in new tab
• Download powerpoint

Figure 5.

Gene copy number assessment of candidate tumor suppressor genes by quantitative real-time PCR to confirm array-CGH data. This analysis substantiated the gene copy numbers reduction observed by array-CGH for all five candidate genes TOPORS, FANCG, RAD51, TP53BP1, and BIK relative to the reference gene ADAR (which was unaltered in all tumors) in a panel of representative tumors and normal female genomic reference DNA. Columns, mean of three to six measurements; bars, ±SD.

Discussion

An intriguing characteristic of glial brain tumors is their phenotypic variety. Equally complex genotypic profiles have been identified by molecular genetic studies. The phenotypic and genotypic heterogeneity indicates that no isolated genetic event accounts for gliomagenesis but rather the cumulative effects of a number of alterations that operate in a concerted manner. We here present to our knowledge the first study that has used cDNA array-CGH as a higher-resolution means to accurately map genome-wide DNA copy number alterations in glial brain tumors of various histologies.

We have precisely characterized genomic segments known from previous cytogenetic studies to be involved in gliomagenesis. Such detailed structural information may prove useful in deciphering the mechanisms responsible for the genesis of these chromosomal aberrations. In each of these segments, a particular gene had been assumed as the target. However, it has been increasingly recognized that high-amplitude altered chromosomal regions often comprise multiple genes, raising the possibility of more than one target gene; yet few studies ( 24, 25) have aimed towards dissecting the actual extent of the involved areas and identifying coaltered genes in gliomas. We have detected the concomitant alteration of additional genes in many of the common altered genomic regions. Several of these genes have been implicated in other human cancers or have tumor-promoting biological functions and may therefore potentially also contribute to gliomagenesis. Functional studies will be needed to dissect whether the biologically important outcome of aberration of some of these genomic regions in gliomas may include the coalteration of some of these genes.

Because gliomagenesis is driven by the sequential acquisition of genetic alterations, it is reasonable to subgroup gliomas by their patterns of genomic aberrations. Our study showed that gliomas could be categorized into distinct subgroups based on their genetic profiles and that these genetic profiles do not necessarily follow tumor phenotype and tumor grade. Major genomic alterations showed a recurrent pattern of occurrence that included coincident appearance, inverse coappearance, and mutual exclusiveness. In addition, our results showed that a subset of genetic events strikingly predicted glioma phenotype. Based on the gene copy number pattern of 170 genes, we were able to predict astrocytic and oligodendroglial tumor morphologies with 92% and 88% accuracy, respectively. This set of genes had no overlap with a gene expression–based class prediction model for glioblastoma and anaplastic oligodendroglioma previously built by Nutt et al. ( 26) based on 20 genes, conceivably, because of the two different biological levels (gene copy number versus gene expression) examined, the slightly disparate morphologic subgrouping in both classifiers, and the only partly overlapping gene coverage between the two used genomic platforms. Our predictive system was not able to distinguish oligoastrocytomas from the two pure phenotypic subgroups. This finding is probably attributable to the fact that no genetic aberrations have been detected that distinguish oligoastrocytomas from pure oliodendrogliomas and astrocytomas.

Our gene-by-gene mapping approach has identified novel common minimally altered regions in a subset of gliomas. The focal and informative nature of these recurrent regions can be best appreciated by consideration of the peak amplitude of recurrence of individual gene alterations within a region. In validation studies, we have focused on five minimally deleted regions, harboring in the peak profile deletions in candidate tumor suppressor genes, not previously reported in gliomas. These alterations were chosen based on their potential biological function and putative implication in human carcinogenesis. Some of the involved genomic regions have been also associated with gliomagenesis. Such coincidence with previous reports provided an additional measure of validation for our experimental approach. Our methodology was highly effective in delineating more discrete genetic alterations within previously described larger chromosomal aberrations. For example, structural abnormalities involving the short arm of chromosome 9 are frequently associated with gliomas. Imbalances on 9p in gliomas have been primarily linked to alteration of CDKN2A within 9p21.3 as a tumor suppressor gene. For several malignancies, however, 9p has been shown altered without loss of CDKN2A, and additional tumor suppressor genes have been implicated to reside in this chromosomal region ( 27). Most recently, Wiltshire et al. ( 2) have suggested multiple tumor suppressor genes on 9p to be involved in the progression of malignant gliomas, with 9p22-p21 and 9p13-p10 being consistently lost. We have identified two small deleted noncontiguous regions within 9p21.1 and 9p13.3, ∼10.5 and 13 megabases distal to CDKN2A, respectively. The high resolution of our technique enabled us to identify the genes TOPORS and FANCG, respectively, residing within the peak profiles of these minimally deleted areas. TOPORS codes for a topoisomerase I– and p53-binding ubiquitin ligase ( 28), that has recently been implicated as a tumor suppressor by inhibiting cellular proliferation and inducing accumulation of cells in the G₀-G₁ phase of the cell cycle ( 27). p53 has been suggested as a ubiquitination substrate for TOPORS, whose overexpression, similar to MDM2, leads to a proteasome-dependent decrease in p53 ( 28).

The FANCG gene codes for a putative tumor suppressor protein that may operate in a post-replication DNA repair or a cell cycle checkpoint function ( 29). Although the role of this protein remains to be fully elucidated, it may not only be involved in the genomic integrity of cells and maintenance of normal chromosome stability, but also seems to participate in interstrand DNA cross-link repair as caused by DNA-damaging agents ( 29). Recent data suggest the association of mutations in this gene with young-onset pancreatic cancer ( 30).

Within chromosome 15q, we have found two noncontiguous minimally deleted regions in a subset of gliomas, mapping to 15q15.1 and 15q15.3 and involving deletions of the RAD51 gene and the TP53BP1 gene. Whereas no mutations or polymorphisms of RAD51 have been detected in brain metastases ( 31) and gliomas ( 32), respectively, our analysis suggests that this gene may be deleted in some gliomas. TP53BP1 has been shown to bind to the central domain of wild-type p53 but not to mutant p53 in human tumors ( 33). Expression of TP53BP1 enhances the transactivation function of p53 and induces the expression of p21 ( 34). In addition, TP53BP1 has been implicated as a critical element in the DNA damage response ( 35) and plays an integral role in maintaining genomic stability ( 36).

Up to 30% of astrocytomas have been shown to carry loss of heterozygosity (LOH) 22q and a role of distal deletions on 22q12.3-q13.2 in glioma progression has been suggested ( 37), although candidate tumor suppressor genes remain to be identified. We have found that a minimally deleted segment in this region involved the BIK gene, a prototype member of the BH3-only Bcl-2 subfamily, which induces apoptosis in various cell types ( 38), including human glioma cells ( 39), and which has been shown to be frequently mutated in human peripheral B-cell lymphomas ( 40).

The degree of change in gene copy number detected by our analyses for different genomic loci has to be viewed in the context of analyzing crude tumor tissue. Because gliomas are known to show a high degree of genetic heterogeneity, our array-CGH and real-time PCR results describe average characteristics of heterogeneous cell populations, where certain clones may harbor homozygous and/or hemizygous deletions and others do not. Laser microdissection studies or fluorescence in situ hybridization analysis in addition to genotyping and promoter methylation analysis will aid in ultimately distinguishing LOH from homozygous deletion in candidate genomic regions. Functional validation will be necessary to definitely assign glioma relevance of the genes targeted by the recurrent alterations described here.

In summary, we have created a high-resolution genome-wide map of genetic aberrations in human gliomas using array-CGH. This gene-by-gene analysis showed the power of this map to precisely localize and size target regions and new recurrent regions of gene copy number change in these neoplasms. The salient results of our study included the identification of five common minimally deleted regions that involve putative tumor suppressor genes. The actual individual or cumulative role of these genes in gliomagenesis has to be evaluated by functional studies. Association mapping of altered chromosomal regions showed a high interrelationship of distinct genetic events in gliomas, which were classifiable into distinct genetic subgroups. In addition, a subset of genetic changes in gliomas was predictive of astrocytic and oligodendroglial tumor phenotypes, suggesting that some gene alterations en route to gliomagenesis may be primarily shared within histologic subgroups, whereas others may be beyond the morphologic boundaries of tumor phenotype.