Identification of Necrosis-Associated Genes in Glioblastoma by cDNA Microarray Analysis

In cancer research, the subject of cell death has been extensively studied with the goal of achieving therapeutically induced tumor death. Among the two major forms of cell death encountered in biology, apoptosis has received a proportionately greater degree of attention and emphasis than has necrosis (1, 2). A large body of recent literature has provided an understanding of the physiological and molecular events that lead to apoptosis; yet similar characterization of necrosis cannot be found.

In the field of neuro-oncology, this paucity of interest in necrosis is surprising because it is a hallmark of the most common and most malignant primary brain tumor—glioblastoma (GBM). In fact, GBM is differentially diagnosed from lower grade astrocytomas based on the histological presence of de novo tumor necrosis and associated microvascular proliferation. Clinical studies indicate that of all clinical, neuroimaging, and histopathological characteristics (including age), necrosis that is visible on magnetic resonance imaging scans has the greatest prognostic value and is inversely related to patient survival (3). Based on the clinical implications and potential for novel therapeutic interventions, the molecular factors mediating necrogenesis in GBM warrant investigation. An initial step in this process would be the identification of specific gene products associated with necrosis.

Although there have been a few limited studies of one or two necrosis-associated molecules, there have been no efforts to characterize global gene expression changes associated with varying degrees of necrosis in GBM (4). In this study, we used cDNA expression microarrays to identify the molecular fingerprints of 15 GBM samples with differing degrees of necrosis; this was done with the goal of identifying genes whose mRNA expression is correlated with the formation of necrosis. In hopes of establishing a relationship between necrosis and the identified genes, their expression levels in anaplastic astrocytomas (AAs), which by definition (St. Anne/Mayo criteria) are tumors that are not associated with necrosis, were determined and compared with expression levels in grade III necrosis GBM samples, which are tumors characterized by extensive amounts of necrosis.

All human tumor samples were obtained from the Brain Tumor Center Tissue Bank at The University of Texas M. D. Anderson Cancer Center (Houston, TX). After surgical removal, tissue samples were quickly frozen at −80°C. All tissue samples used for cDNA analysis were evaluated by a neuropathologist; all specimens selected were characterized by dense tumor cellularity and the presence of <10% normal brain tissue. A second neuropathologist confirmed all diagnoses. Grading of astrocytic neoplasms was performed according to the St. Anne/Mayo criteria. A total of 5 AAs and 15 GBM samples were analyzed. The 15 GBM samples obtained were further graded based on the amount of necrosis observed on magnetic resonance imaging scans. Necrosis on magnetic resonance imaging scans appears as a hypointense region of T1 signal surrounded by a contrast-enhanced region representing viable tumor. Necrosis was graded according to the following previously described system: grade 0, no necrosis apparent on the magnetic resonance imaging scan; grade I, amount of necrosis is <25% of the tumor volume; grade II, amount of necrosis is between 25% and 50% of the tumor volume; grade III, amount of necrosis is >50% of the tumor volume (Ref. 3; Fig. 1). The extent of necrosis was confirmed in all cases during surgery. Within our sample population, four tumors were grade 0, two were grade I, five were grade II, and four were grade III. The sample distribution is shown in Table 1.

The frozen tissue samples were ground to a powder in liquid nitrogen using a mortar and pestle and then treated with lysis buffer, TRI Reagent (Molecular Research Center, Cincinnati, OH), to extract nucleic acids and prevent their degradation. Total RNA was isolated as described previously (5). The quality of this RNA was checked by electrophoresis on a denaturing formaldehyde agarose gel. The presence of high-quality undegraded RNA was verified by the observation of a lack of streaking on the lower (anodal) part of the sample lane on the gel. A further indication that high-quality RNA was obtained was the presence on the gel of a 28S rRNA band twice as intense as that of the 18S rRNA band. A poly(dT) mRNA isolation column (Qiagen, Inc., Chatsworth, CA) was used to separate mRNA from high-quality total RNA. With this column, we recovered 1–2 μg of mRNA from approximately 200 μg of total RNA, but only 0.5–1.0 μg of mRNA was required for analysis in the cDNA array. In this study, we obtained high-quality RNA from >90% of the resected samples. Because undegraded RNA is scarce within regions of necrosis, total RNA and mRNA were predominantly isolated from viable nonnecrotic tumor regions.

cDNA fragments, approximately 200- to 500-bp long, representing 588 human genes with known functions and tight transcriptional controls were immobilized in duplicate on a nylon membrane (Clontech Laboratories, Inc.). To minimize cross-hybridization and nonspecific binding of cDNA probes, each fragment was selected as a unique sequence without a poly(A) tail, highly homologous sequences, or repetitive elements. ³²P-labeled cDNA probes were generated through reverse transcription of 0.5–1.0 μg of each analyzed poly(A)⁺ RNA sample in the presence of [α-³²P]dATP; the probes were then hybridized to the array. After a high-stringency wash of the array, the hybridization pattern obtained was analyzed by autoradiography and quantified by phosphorimaging using ImageQuant software and a Storm 840 PhosphorImager. Because the amount of each cDNA fragment on the membrane was in excess (10 ng), cDNA-probe binding interactions were linear.

For this study, membranes from the same production lot were used because our previous work indicates that this makes the results of the technology reproducible. Lastly, tissue samples were submitted to analysis in a blind manner with respect to their degree of necrosis.

Formalin-fixed, paraffin-embedded tissues were cut into 5-μm-thick sections and mounted on positively charged Superfrost slides. Tissue sections were deparaffinized in xylene, followed by a graded series of alcohol washes, and rehydrated in PBS. Sections were then incubated with pepsin (Biomedia, Foster City, CA) for 30 min at 37°C. To block endogenous peroxidases, slides were placed in a humidified chamber and incubated for 12 min at room temperature with a solution of 3% H₂O₂ in methanol. The slides were washed three times with PBS and blocked for 20 min at room temperature in PBS supplemented with 1% normal goat serum and 5% normal horse serum (protein-blocking solution). The slides were incubated overnight at 4°C with a rabbit polyclonal anti-EphA1 (ephrin type A receptor 1) antibody (1:200 dilution; Santa Cruz Biotechnology, Santa Cruz, CA) or a rabbit polyclonal anti-EP4 (prostaglandin E₂ receptor EP4 subtype) antibody (1:100 dilution; Cayman Chemical, Ann Arbor, MI); for negative control slides, nonspecific IgG was used in place of antibody. The slides were then rinsed three times with PBS, incubated for 10 min in protein-blocking solution, and incubated for 1 h at 25°C in peroxidase-conjugated antirabbit IgG solution. Next, the slides were washed and incubated in stable diaminobenzidine solution (DAB; Research Genetics, Huntsville, AL). The slides were washed three times with distilled water, counterstained with Gill’s hematoxylin solution (Sigma Chemical Co., St. Louis, MO), and examined with a bright-field microscope. Reddish-brown precipitate in the cell cytoplasm indicated a positive reaction.

To determine correlation between gene expression and the amount of necrosis present in the tumor samples, gene expression and necrosis were ranked. Necrosis was ranked according to the previously described grading system; gene expression was ranked after the hybridization results for each array had been corrected for background signal, thresholded at 1 (values of <1 were set at 1), and normalized to the median of that array. The values were transformed to log base 10 scale for analysis. Ranking provided the advantage of accounting for outlying data points, thereby facilitating robust data analysis. Once the respective variables were ranked, the rank correlation coefficients were determined, with the spectrum of possible values ranging from −1, a perfect inverse correlation, to +1, a perfect correlation between necrosis and gene expression; scatter plots were also created to visualize gene expression-necrosis correlation. The statistical significance of the correlation coefficients was then calculated using Ps determined by standard correlation coefficient analysis; statistical significance was set at P < 0.05. Expression values for 26 genes that were significantly correlated with necrosis were compared among samples of grade III necrosis GBMs, grade 0 GBMs, and AAs using Wilcoxon’s rank-sum test.

The hazard ratios and Ps for survival differences between low (grades 0 and I) and high (grades II and III) necrosis GBM groups were computed using Cox proportional hazards regression analysis.

A statistical analysis confirmed the negative prognostic value of necrosis. Within this study, the median survival times for patients having GBMs with necrosis of grades 0, I, II, and III were 64, 116, 53, and 25 weeks, respectively. This necrosis-survival relationship is shown in Table 1. The hazard ratio for patient groups 0 and I versus groups II and III was 2.2 (P = 0.029). Numerous gene expression changes were also found to be associated with necrosis grade. Hybridization experiments revealed that changes in the expression of 26 genes were significantly correlated with necrosis grade. Of these 26 genes, 9 were positively correlated with necrosis grade (Table 2), whereas 17 had negative rank correlation coefficient values (Table 3). Based on their potential cellular functions, all 26 genes were identified as being associated with several specific molecular pathways (and their corresponding biological functions). These pathways included those described for Ras, Akt, tumor necrosis factor (TNF)-α, nuclear factor (NF)-κB, apoptosis, procoagulation, and hypoxia. Based on the functions known for each gene, Tables 2 and 3 include comments on some of the potential effects on the necrosis cascade of the observed up-regulation or down-regulation of these genes. For example, we believe that the end result of the up-regulation of ephrin type A receptor 1, which natively stimulates Akt, is increased Akt activity.

As indicated in Table 2, the rank correlation coefficients for the nine genes with expression positively correlated with necrosis grade ranged from an extremely strong correlation of 0.78 to a weaker but still significant association of 0.52. Paralleling this, the Ps ranged from 0.004 to 0.054. Within this subset of genes, of particular interest were those for the ephrin type A receptor 1 and prostaglandin E₂ receptor EP4 subtype, both of which showed the strongest positive correlation values. This strong correlation can be appreciated on the scatter plots shown in Fig. 2. Genes for both receptors have correlation coefficients that are >0.7 and Ps of 0.004. Further statistical analysis determined the extent of correlation of patient survival with the expression of the four genes most positively correlated with necrosis grade (prostaglandin E₂ receptor EP4 subtype, ephrin type A receptor 1, DNA ligase IV, and thrombomodulin). Cox proportional regression analysis revealed that combining all four genes in a single model gave an R2̂ value of 0.53 and an overall P of 0.012. Furthermore, combining the prostaglandin E₂ receptor (EP4 subtype) and ephrin type A receptor 1 in a single model gave an R2̂ of 0.53 and an overall P = 0.0034.

In contrast, Table 3 displays the gradient of rank correlation coefficient values for genes whose expression was negatively correlated with the amount of necrosis. The correlation coefficients ranged from −0.65 to −0.52, whereas Ps increased from 0.014 to 0.053. This grouping reveals the importance of the expression of genes for the receptors for interleukin 7 and 12, both of which display the strongest inverse correlation with necrosis grade. Genes for both receptors had the lowest calculated rank correlation coefficient values of approximately −0.65 and Ps close to 0.015. Similarly, the strong negative correlation coefficients can be seen in the scatter plots for the interleukin 7 and 12 receptors in the bottom panels of Fig. 2.

As Table 4 indicates, there were significant differences in the median expression levels of the 26 genes between the grade III necrosis GBM and AA samples analyzed. Of the nine genes positively correlated with necrosis, three genes displayed significant differences in their expression levels between the histological subsets. Two of these, the genes for the ephrin type A receptor 1 and thrombomodulin, had the most significant Ps of 0.0159. On the other hand, only 2 of 17 genes (Table 4B) showed significant (P < 0.05) negative correlation with the extent of necrosis. Genes for the interleukin 7 receptor α subunit and G₁-S-specific cyclin E had Ps of 0.0159 and 0.0317, respectively. Although we noted some differences in gene expression between GBMs of necrosis grade 0 and AAs, the significance of this finding is unclear at present.

Immunohistochemical staining (Fig. 3) provided further data corroborating the microarray results. Staining for the prostaglandin E₂ receptor EP4 subtype and ephrin type A receptor 1 indicated that these gene products were (a) overexpressed in grade III necrosis GBM tumor samples, but not in those with grade 0 necrosis or in AA samples, and (b) localized in perinecrotic areas within the tumor samples.

The different forms of cell death are generally viewed as a continuum in which apoptosis and necrosis comprise the two extremes. Whereas apoptosis occurs according to a pathway of caspase activation, necrosis can occur when the apoptotic pathway cannot be completed (1, 2). In cancer research, there has been great interest in inducing apoptosis to fight tumors; yet necrosis, also a form of cell death, has failed to attract similar attention. In light of the clinical relevance of necrosis in neuro-oncology, it is surprising to encounter so little data on the subject. Thus, we have formulated a hypothesis regarding the molecular events in necrosis and characterized the molecular profiles of malignant gliomas that are associated with varying amounts of necrosis.

As described in our hypothesis, we believe that the Ras and Akt pathways, both of which are integral to GBM formation, may initiate a series of events leading to necrosis (6). Ras, through activation by various growth factors such as epidermal growth factor, could lead to increases in TNF-α levels and the eventual overexpression of transcription factor NF-κB. TNF-α, along with other factors, could initiate procoagulation in the tumor vascular network. Extensive procoagulant activity would then lead to hypoxia and nutrient deprivation, resulting in lethal tumor energy decreases. In coordination with antiapoptotic mechanisms activated by Akt and NF-κB, this could prevent the completion of apoptosis induced by ligand-receptor interactions. As summarized in Fig. 4, one of the many possible mechanisms of necrosis formation could involve activation of Ras, Akt, TNF-α, NF-κB, apoptosis, procoagulation, and hypoxia. Our results indicate that all 26 of the genes whose expression is significantly correlated with necrosis have functions associated with the hypothesized biological factors (4, 5, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 76).

The relevance of these genes to the formation of necrosis was investigated by comparison of the gene expression profiles among the grade III necrosis GBM, grade 0 necrosis GBM, and AA subsets. By comparing gene expression between samples containing high degrees of necrosis (grade III) and tumors that do not have necrosis (AA), genes that are uniquely associated with the formation of necrosis may be identified. Among genes positively correlated with necrosis (Table 4A), all but one had elevated expression levels in the necrosis grade III samples relative to the AA samples. This supports the observation that these genes are up-regulated in highly necrotic GBM samples; however, the difference is significant in only three of these genes. These results may indicate that although a certain number of genes are increasingly transcribed in generating necrosis, only a fraction of them may be required for the process. Yet necrosis formation probably results from the output of a set of genes, rather than from the product of a single gene.

Based on their significant Ps, ephrin type A receptor 1, prostaglandin E₂ receptor EP4 subtype, and thrombomodulin may be necessary in modulating Akt, TNF-α, and procoagulant activity. This pattern is also present among genes whose expression we found to be negatively associated with necrosis grade. Whereas 13 of 17 such genes showed increased expression levels in the AAs relative to the necrosis grade III GBM subset, only 2 genes displayed significant differences in their expression between these histological categories. The significant differences in interleukin 7 receptor α subunit and G₁-S-specific cyclin E gene expression indicate that these gene products may be necessary in the control of Ras, NF-κB, and antiapoptotic actions. Thus, although the 26 genes we identified may have roles in pathways resulting in necrosis, only 5 appeared significantly correlated with the process.

One important biological pathway implicated in the necrosis cascade is apoptosis, in which the largest percentage of the 26 differentially expressed genes is involved. Our model of necrosis predicts potentially increased stimulation of the apoptotic program, which could then be switched to a necrotic mode of death through energy deprivation and antiapoptotic factors (15). Results of the present study support this concept. Genes such as adenosine A3 receptor, which inhibits apoptosis at low ligand concentrations, and CD40, a member of the TNF superfamily that induces apoptosis, are differentially expressed, resulting in the increased induction of apoptosis (21, 27, 34, 55, 62). These changes are accompanied by the up-regulation of DNase X and the down-regulation of RNA polymerase II SIII p15 subunit, which may induce intrinsic apoptosis through DNA degradation and defective transcription, respectively (37, 51). On the other hand, DNase X could potentially be one of the targets for apoptosis-inducing factor, which triggers nuclear apoptosis after dissipation of the mitochondrial transmembrane potential, an event that is crucial to apoptosis (14).

The results also show reduced expression of genes for other apoptosis-inducing ligand molecules, such as Fas and OX40 (members of the TNF gene superfamily); interestingly, NF of activated T cells (NF-AT), a transcription factor that induces the expression of Fas and OX40, is also down-regulated (52). This apparent paradox could be part of a negative feedback mechanism activated in response to the formation of necrosis. Akt, by inhibiting glycogen synthase, prevents the phosphorylation of NF-AT and its translocation to the nucleus, preventing expression of Fas and OX40 (22). To determine whether these changes in gene expression are part of a negative feedback mechanism, future studies will have to ascertain whether these genes are essential to necrosis formation. The use of pharmacological inhibitors will aid in this process.

The results also provide an interesting glimpse into the possible changes in TNF-α levels in regions of necrosis. Although our hybridization data indicated that TNF-α levels may not be changed, a previous immunohistochemistry study localized TNF-α to the necrotic periphery (4). In our study, the up-regulation of genes that are promoted by or initiate TNF-α synthesis is observed concurrently with the marked decrease in expression of genes that are normally repressed by or inhibitory to TNF-α, such as adenosine A3 receptor (11, 40). However, there are paradoxes, such as that involving the gene for the prostaglandin E₂ receptor EP4 subtype. This receptor can inhibit TNF-α synthesis by interaction with prostaglandin, and expression of the gene is positively correlated with necrosis grade (60). In this study, no change in TNF-α gene expression was detected. This may be caused by a lack of sensitivity of nylon microarray analysis. Further investigation of this issue will require Western blot analysis and more immunohistochemical studies. Nevertheless, the results do suggest that other members of the TNF gene superfamily could play equally important roles in inducing procoagulation and apoptosis.

Aside from indicating a strong association between differentially expressed genes and the biological factors perceived to be crucial to necrogenesis, this study brings additional molecules, such as the prostaglandin E₂ receptor EP4 subtype and ephrin type A receptor 1, into consideration. Interestingly, the immunohistochemical staining experiments indicated that these gene products were not only localized to perinecrotic areas within grade III necrosis GBM tumors but were also overexpressed in such samples in comparison with either grade 0 necrosis GBMs or AAs. These data help to validate a potential connection between the above two genes and necrosis formation.

The prostaglandin receptor, which had the highest positive correlation coefficient of all of the genes considered, actively modulates coagulation in addition to its capacity to inhibit TNF-α synthesis. The EP4 subtype is highly expressed on platelets (49). A stimulator of adenyl cyclase, the EP4 receptor induces coagulant activity by elevating intraplatelet cAMP levels (49, 67). The role of the EP4 receptor in procoagulation and anti-TNF-α activity raises an interesting question: what is the primary function of the prostaglandin E₂ receptor EP4 subtype in the appearance of necrosis? Is it mainly a mediator of the coagulant cascade, and although it has anti-TNF-α activity, are there more potent pro-TNF-α molecules to counterbalance its inhibitory effects? Additional studies will be required to resolve these questions.

Along with the prostaglandin E₂ receptor, the ephrin type A receptor 1 and its ligand, ephrin-A1, are molecules not previously considered as contributing to the formation of necrosis. Identified as an early response growth factor to TNF-α stimulation, ephrin-A1 (which, unlike other ephrins, is not normally found in the nervous system) is a proangiogenic factor and a chemoattractant for migrating endothelial cells (7, 16). Ephrin-A1 is transcribed at increased levels in tumor cells of epithelial origin, such as advanced melanomas (16). The capability of ephrin-A1, in contrast to other members of the ephrin family, to promote tumor cell survival and growth suggests that it has antiapoptotic activity. Ephrin activation of phosphatidylinositol 3′-kinase, an inducer of Akt and the resulting antiapoptotic pathways, could mediate such activity (7, 16). Therefore, ephrin type A receptor 1, which shows a high positive correlation with necrosis grade, may have angiogenic and antiapoptotic roles induced by TNF-α.

In addition to the high correlation between necrosis grade and the expression of the ephrin type A receptor 1 and prostaglandin E₂ EP4 receptor subtype, statistical analysis revealed that when data for these two genes were combined in a single model, these genes showed a positive and statistically significant correlation with patient survival. Because the extent of necrosis is negatively correlated with patient survival, any genes correlated with necrosis grade should accordingly be correlated with survival. Interestingly, a statistically significant positive correlation with survival was shown when data for the top four genes positively correlated with necrosis grade (prostaglandin E₂ receptor EP4 subtype, ephrin type A receptor 1, DNA ligase IV, and thrombomodulin) were combined into a single model, although individually these genes did not show a significant correlation with survival. This suggests an association among these genes in necrosis formation and indicates that it is an intricate tapestry of gene products, rather than the action of a single gene, that leads to necrosis in GBMs.

There are several limitations to the present study. First is the small sample size of 15 GBMs; for more definitive results, future studies will have to examine a larger population of these tumors. Moreover, because the genes identified in this study have many functions, it is possible that the differential expression we monitored in some of them is related to processes other than necrosis that also correlate with GBM severity. Nevertheless, it is impressive that such a strong correlation between necrosis grade and the expression of several genes was obtained in this study. This is evident in the scatter plots of Fig. 2 that show clustering of gene expression values within specific necrosis grades. Additional limitations are the relative inability of nylon microarray analysis to detect changes in genes with low expression and the possibility of false negatives and positives. With regard to the latter, complementary and confirmatory techniques such as Western blotting and/or immunohistochemistry are always required to confirm and extend the findings of gene expression microarray analysis, as demonstrated in this study in the case of the ephrin type A receptor 1 and the prostaglandin E₂ receptor.

Our initial cDNA microarray analysis of 15 GBMs, in parallel with 5 AAs, has provided a glimpse of molecular events in the formation of necrosis in the highest grade of astrocytoma. The results indicated that Ras, Akt, TNF-α, NF-κB, apoptosis, procoagulation, and hypoxia may be important to necrogenesis because all of the 26 differentially expressed genes identified had functions associated with these factors. Of the 26 genes that may play a role in necrogenesis, only the ephrin type A receptor 1, prostaglandin E₂ receptor EP4, interleukin 7 receptor α subunit, G₁-S-specific cyclin E, and thrombomodulin gene products showed a significant correlation with the process. The data also revealed that necrosis formation may be dependent on a delicate balance between pro- and antiapoptotic activities. Associated with this revelation was the identification of genes not previously implicated in necrosis as possible active participants in the necrosis cascade. Together with these conclusions, it is important to keep in mind the possibility of negative feedback mechanisms. This study highlights the concept that end biological events, such as necrogenesis, are determined by the balance of pro- and counter-regulatory mechanisms. Future studies will be aimed at confirming the findings of the present study through complementary methodologies and extending the investigation of necrogenesis with more comprehensive gene expression analyses.

We thank Dr. David M. Wildrick for editorial assistance in preparing the manuscript.