Gene Expression Profiling with Oligonucleotide Microarrays Distinguishes World Health Organization Grade of Oligodendrogliomas

Introduction

Oligodendrogliomas are primary central nervous system neoplasms estimated to represent <10% of all glial tumors (1) . Classic oligodendrogliomas present with a peak incidence in the fifth decade (median age at diagnosis, 47 years) but can be seen in patients from 3 to 76 years of age (2) . Criteria for distinguishing oligodendrogliomas from other glial subtypes remain subjective, and equivocal morphology often results in significant interobserver variability, even among experienced neuropathologists. Although several histological grading schemes exist, the two-tiered WHO system is currently the most widely used. Low-grade (WHO II) oligodendroglial tumors are composed of uniformly round to oval cells with bland nuclear chromatin and often demonstrate perineuronal satellitosis, microcalcifications, mucoid/microcystic degeneration, and dense capillary networks. Subpopulations of cells may exhibit nuclear atypia and occasional mitoses. High-grade (WHO III) oligodendroglial tumors show increased cellularity, cytological atypia, and high mitotic activity. They often demonstrate vascular proliferation and necrosis as well. High WHO grade has been correlated with clinical progression and decreased survival (3) . However, there is still sufficient individual variability within diagnostic categories to warrant the development of additional prognostic markers. Reported molecular alterations in oligodendrogliomas differ greatly from those in astrocytomas, perhaps accounting for their distinct clinical behavior. The most frequent alterations in oligodendrogliomas are deletion of the long arm of chromosome 19, seen in as many as 80% of cases, and the short arm of chromosome 1, seen in the majority of oligodendroglial tumors, but only rarely in astrocytic tumors (4, 5, 6, 7, 8) . Although combined loss of chromosome arms 1p and 19q is a statistically significant predictor of prolonged survival independent of tumor grade (9, 10) , there is little known with regard to how such genetic alterations correlate with more global patterns of gene expression in oligodendrogliomas. It is now evident that gene expression profiling, accomplished using nucleic acid microarrays, is capable of subclassifying hematological as well as solid organ malignancies based not only on cell lineage but also by other clinically relevant parameters (11 , 12) . In this report, we have examined gene expression in seven oligodendrogliomas using oligonucleotide microarray (GeneChip) technology in an initial attempt to classify these tumors based solely upon their expression profiles and define groups of genes that may ultimately predict biological behavior. We demonstrate that a subset of ∼200 genes accurately groups these tumors based on their original WHO histological grade. Many of the gene products from this expression profile represent cytoskeletal proteins, immunomodulatory factors, and glial markers that potentially have important biological and clinical implications for oligodendroglial malignancies. More significantly, these results suggest that gene expression profiling of malignant glial tumors could be used as an ancillary diagnostic tool for categorization, particularly when classification by histopathology is difficult or ambiguous.

Materials and Methods

Tissue Procurement and Sample Preparation.

All tissue samples were collected by the Siteman Cancer Center Tissue Procurement Facility under an approved protocol from the institution’s Human Studies Committee. Resected tumor tissue was immediately snap frozen in liquid nitrogen. Diagnoses were rendered from review of each patient’s formalin-fixed, paraffin-embedded resection material in surgical neuropathology files and graded according to WHO 2000 criteria (13) . Additionally, 1p and 19q chromosomal status was determined using dual-color fluorescence in situ hybridization as described previously (10) . DNA probes localizing to 1p32 (BAC clone 260 I23; Research Genetics, Huntsville, AL; labeled with FITC) and the subtelomeric region of 19q (Vysis, Downers Grove, IL; labeled with SpectrumOrange) were used (data not shown). Tumors interpreted as having combined 1p and 19q deletions demonstrated a single green and a single red signal in the overwhelming majority of tumor nuclei.

Frozen tumor specimens were embedded in freezing medium, sectioned at 5 μm, and stained with H&E. The histopathology of each collected specimen was reviewed to confirm the adequacy of the sample (i.e., minimal contamination with nonneoplastic elements) and to assess the extent of tumoral necrosis and cellularity. Subsequent 50-μm serial sections from each banked frozen specimen were then cut, placed immediately into Trizol reagent (Life Technologies), and homogenized. Total RNA was isolated using the manufacturer’s protocol. Extracted RNA was then further purified by spin chromatography (RNeasy kit; Qiagen) following the manufacturer’s protocol. Purified RNA was quantitated by UV absorbance at 260 and 280 nm and assessed qualitatively by formaldehyde agarose gel electrophoresis.

Oligonucleotide Array Analysis.

Analysis was performed by the Siteman Cancer Center GeneChip Facility. Ten μg of purified, total RNA were converted to cDNA, purified, and then used as a template for in vitro transcription of biotin-labeled antisense RNA. All protocols were performed exactly as recommended by the manufacturer (Affymetrix) and have been described elsewhere (14) . Twenty μg of each biotinylated antisense RNA preparation was fragmented, assessed by gel electrophoresis, and placed in a hybridization mixture containing four biotinylated hybridization controls (BioB, BioC, BioD, and Cre) as recommended by the manufacturer. Samples were hybridized to an identical lot of Affymetrix Hu6800SubD GeneChip arrays (Lot #92985) for 16 h. GeneChips were washed and stained using the instrument’s standard Eukaryotic GE Wash 2′ protocol, using antibody-mediated signal amplification. The images from the scanned chips were processed using Affymetrix GeneChip Analysis Suite 3.2. The image from each GeneChip was scaled such that the average intensity value for all arrays was adjusted to 1500. SADV 3 and Absolute Call Data from each GeneChip were exported to flat text files and used for statistical analysis. 4

Statistical Analysis.

Distance measures, hierarchical clustering computations, and graphical output were performed using the SAS software package (SAS Institute). Of the total of 1879 sequences represented on the array, hybridization control sequences and sequences scored as “absent” (not detected) in all seven samples were excluded from analysis. For the remaining 1013 genes, all SADVs of less than zero were set to zero. Hierarchical clustering based on Euclidean distance measures was performed using several clustering algorithms (e.g., single-linkage, complete-linkage, and Ward’s method) to generate dendrograms that were then compared with each other to assess the stability of the clustering. The dendrogram displayed in Fig. 1 ⇓ was generated using Ward’s Minimum-Variance method (Ward’s method) and was essentially similar to those generated using other algorithms. Ward’s method defines the distance between two clusters as the ANOVA sum of squares between the two clusters added up overall for the tumor samples. The method then joins clusters to maximize the likelihood at each level of hierarchy while assuming multivariate normal mixture, equal spherical covariance matrices, and equal sampling probabilities. Self-organizing maps were fit using GeneCluster (15) . SADVs for the 1013 genes were normalized to a mean of 0 and SD of 1 before fitting the self-organizing maps. This was done to standardize the expression levels and mitigate any differences in gene expression due to scale.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.

Hierarchical clustering of oligodendrogliomas. The dendrogram demonstrates the relatedness of the seven tumor samples to each other, based on the expression of 1013 genes. For each tumor, clinical covariates are listed as well as the scaling factor used to generate SADVs from each array and the total number of genes detected (“Present” calls) in each sample.

Results and Discussion

We tested the hypothesis of whether gene expression profiling by high-density oligonucleotide array could effectively identify subtypes of malignant glial neoplasms. Furthermore, we wished to determine the subset of genes the expression of which defined each subtype identified. Seven cases of oligodendroglioma were chosen, four of which had been classified previously as WHO grade II and three as grade III. All seven samples were processed, and data were analyzed in a blinded fashion. The data corresponding to each tumor specimen are summarized in Fig. 1 ⇓ . As shown in Table 1 ⇓ , the hybridization signal intensity obtained from hybridization controls in each of the seven samples, expressed as SADVs, showed <2-fold variation at high (Cre coefficient of variation, 0.08) and low (BioB coefficient of variation, 0.34) transcript copy numbers. As scaled data sets, the scaling factor used for all seven samples (indicated in Fig. 1 ⇓ ) was 2.78 ± 0.40. The number of genes scored as detected (“P”) in each sample (Fig. 1) ⇓ ranged from 34 to 39%. These metrics all demonstrate that the quality of each of the labeled targets and the resulting fluorescent signal obtained from each of the seven GeneChips was comparable.

The hierarchical clustering relationship among all seven tumors based on gene expression data are displayed as a dendrogram in Fig. 1 ⇓ . Aligned with this dendrogram are the covariates: age, gender, WHO grade, chromosome 1p/19q loss, and disease status. After unblinding the specimens, it was immediately apparent that tumor clustering based on gene expression information accurately divided the tumors based on their previously assigned WHO grade. Tumors 1102, 1113, and 1114 (grade III/IV) defined one closely related group, whereas tumors 1108, 1111, and 1110 (grade II/IV) formed a second, closely related group. Interestingly, tumor 1109 (histological grade II/IV) was less closely related to the three other grade II tumors. In fact, when alternate clustering methods were used, the classification of this tumor vacillated between the grade II and grade III groups. The total number of genes scored “P” and the overall hybridization intensity (reflected by scaling factor) for tumor 1109 was comparable with the other six tumors (Fig. 1) ⇓ , suggesting that there was no technical aberration associated with this sample. Because tumor 1109 lacked features of a higher grade tumor, such as high proliferative index, endothelial proliferation, or necrosis, it was histologically assigned as WHO grade II. However, this tumor did display increased nuclear pleomorphism and increased numbers of microgemistocytes in comparison with the other grade II specimens. By fluorescence in situ hybridization analysis, this tumor did not harbor 1p or 19q deletions, further suggesting that this tumor may represent a slightly different oligodendroglioma variant at the molecular level. Perhaps of related significance is the fact that the age of patient 1109 is higher than the other three patients in the grade II group (Fig. 1) ⇓ , although typical of the median age seen for oligodendroglial tumors (2) . At a mean follow-up of 31 months, there were no clinical recurrences in the patients with grade II lesions and only one death among the three patients with grade III lesions.

Groups of genes were determined to define the expression profile of grade II versus grade III tumors. The clustering algorithm, GeneCluster (15) , which uses the principle of self-organizing maps, was used to classify all 1013 genes used in the above analysis into 12 groups or “nodes.” One of these groups demonstrated a global pattern of gene expression that was relatively higher in all grade II tumors (cluster 10; 96 genes), and a second group had an average level of gene expression that was relatively higher in all grade III tumors (cluster 2; 100 genes). Fig. 2 ⇓ shows a grayscale contour map displaying the relative changes in gene expression values among the seven tumors and 196 genes in clusters 2 and 10. Again, it is interesting to note that tumor 1109 (grade II) does not definitively fit into one of these two clusters, as was seen in attempts to classify this tumor using hierarchical approaches. A representative subset of the 196 genes identified whose average expression level was at least 3-fold different between grade II and grade III tumors is summarized in Table 1 ⇓ . At the level of individual gene expression, tumor 1109 again demonstrated characteristics of both “grade II-like” and “grade-III-like” tumors. Subsequent to filtering for this representative set of genes, each gene was grouped into a broad functional class based on literature review.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

Contour plot of differentially expressed genes. A total of 196 genes from two clusters generated using self-organizing maps is displayed. Each of the seven tumor samples is represented in columns. Each of the 196 genes is represented as a horizontal line. The relative expression level of each gene in each sample is depicted in gray scale (lower expression corresponds to lighter shade). The scale (−2.2 to 1.8) is the normalized expression such that each gene has a mean of 0 and a SD of 1.

Several genes normally expressed in mature, differentiated oligodendrocytes and associated with cell adhesion and cell to cell signaling were down-regulated in higher grade tumors. Two examples include the heparan sulfate proteoglycan, epican (16) , and β2-integrin (17) . These proteins are found in mature, differentiated oligodendrocytes and are associated with cell-cell adhesion. As such, they might reasonably be expected to be less abundant in anaplastic tumors. Three genes, the protein counterparts of which are associated with cytoskeletal organization, vimentin, syntrophin, and glycophorin C, were also down-regulated in higher grade tumors. Although expression of these last two genes has not been characterized in glial tumors, vimentin is commonly expressed in gliomas (18) . Conversely, several myosin genes were expressed at significantly higher levels in the anaplastic tumors. Drebrin (19) , a neuronal actin-binding protein, was also more highly expressed in anaplastic tumors. These findings are in general agreement with many previous studies that have defined a role for cytoskeletal proteins in the migration of infiltrative gliomas (20 , 21) . There was also moderately decreased expression of the gap junction protein, connexin43, a mediator of intercellular communication, the expression and cellular localization of which may be disrupted in anaplastic oligodendrogliomas (22) .

Another group of genes with decreased expression in anaplastic tumors included MHC antigen E and two other IFN-inducible genes. Mature human oligodendrocytes have been shown to express multiple MHC class I and class II antigens (23) . Loss of MHC expression and, more generally, loss of IFN-γ responsiveness may be one way that more aggressive glial tumors further evade an already weak host immune response (24) . Other genes that were relatively down-regulated in anaplastic tumors were glial cell differentiation markers, such as myelin basic protein and a novel neuropeptide-like, G protein-coupled receptor. Paradoxically, expression of tissue factor, which has been shown previously to correlate with a higher grade of astrocytic tumors (25) , was actually lower in the grade III oligodendrogliomas. This may be related to the capillary proliferation, which is typical of low-grade oligodendrogliomas and lacking in low-grade astrocytomas. Finally, α2-macroglobulin, a protease inhibitor expressed in astrocytic gliomas (26) , was also down-regulated in the more anaplastic oligodendroglial tumors.

There were far fewer genes that demonstrated consistent “up-regulation” in all grade III tumors as compared with grade II tumors. In addition to the cytoskeletal genes mentioned above, a kex-like endoprotease, the guanine nucleotide-binding protein Rap2, and DNA topoisomerase I were elevated in anaplastic specimens. Of note, elevated topoisomerase I expression in more anaplastic tumors may be an important predictor of response to treatment using camptothecin analogues (27) . The human homeobox gene, HOX-11, was also expressed at relatively increased levels in the anaplastic oligodendrogliomas. Although initially identified as a gene whose dysregulation is an important step in the progression of T-cell leukemia, further studies suggest that HOX-11 may be a more general transcription factor responsible for G₁ progression in the cell cycle (28) .

Of course, the exact biological and clinical significance of these genes as they relate to oligodendroglioma biology will only be determined through additional studies. However, the data presented in this report demonstrate three important points: (a) as has been shown for other malignancies, expression profiling of human oligodendrogliomas may provide an accurate and precise method for tumor classification, at least as compared with the WHO histological classification scheme used currently; (b) in examining a relatively few number of genes (1013 in this study), a molecular portrait emerges of anaplastic oligodendrogliomas with predicted characteristics that include dysregulated cytoarchitecture, decreased immunogenicity, increased migratory potential, loss of differentiation, and potentially increased susceptibility to topoisomerase inhibitors. A more comprehensive and accurate portrait should evolve as additional genes and a larger number of tumor samples are analyzed; and (c) as suggested by tumor 1109 in this study, molecular profiling of oligodendrogliomas may reveal new diagnostic categories that are not realized by current histological classification schemes. In particular, it will be interesting to determine the potential utility of expression profiling in the morphologically equivocal gliomas with less convincing oligodendroglial features. This approach may ultimately result in new diagnostic methods to better manage patients with this disease.