Prognostic and Pathologic Significance of Quantitative Protein Expression Profiling in Human Gliomas¹

Gliomas are among the most aggressive and treatment refractory of all human tumors. Astrocytomas, oligodendrogliomas, and mixed tumors comprise over 80% of primary adult central nervous system tumors. With aggressive treatment, the 2-year survivals are 66, 45, and 9%, respectively, for low-grade astrocytomas, anaplastic astrocytomas, and glioblastoma multiforme, according to the most recent Surveillance, Epidemiology, and End Results data (1).

The molecular pathogenesis of gliomas has been under active investigation as part of the effort to develop more effective therapeutic strategies for these tumors. Efforts have focused primarily on the genomic alterations within tumor cells. Reported chromosomal alterations have included loss of heterozygosity for 1p, 9p, 10, 13q, 17p, 19q, and 22q loci and gain of chromosome 7 (2, 3, 4, 5, 6, 7, 8, 9, 10). Cell cycle regulatory genes and genes encoding for growth factors are commonly affected. These include inactivation of CDKN2A (encoding for p16 and p14^ARF), RB (on 13q), p53 (on 17p), PTEN (phosphatase and tensin homologue on 10q), and amplification of EGFR³ and cdk4, among others (11, 12, 13, 14, 15, 16, 17, 18). Genetic features are thought to distinguish two broad groups of GBMs. Primary GBMs, or de novo GBMs, are diagnosed histopathologically as GBMs at first presentation and are characterized by EGFR gene amplifications without corresponding p53 gene mutations. Secondary, or progressive GBMs, progress from lower-grade astrocytomas, seldom have EGFR gene amplifications, but usually do have p53 gene mutations (4, 12).

However, genetic alterations represent only one of several mechanisms that alter gene expression in tumor cells. Changes in transcription or in protein stability, protein modification, or protein/protein interactions play important roles in tumor cell proliferation but are not detected in studies that rely solely on DNA analysis. Molecular studies of gene expression in gliomas, either at the RNA or protein level, are relatively sparse. Meticulous tissue handling and storage (snap freezing) are required to quantitatively examine mRNA or protein levels (or activation states). Because most centers preserve tissue in formalin-fixed paraffin, efforts to quantitatively analyze gene expression patterns at the RNA or protein level are effectively thwarted. IHC is often used to qualitatively assess protein expression on paraffin-embedded tissues. However, the interpretation of IHC results is subjective, and problems with antigen preservation and integrity mean that there is often substantial variation between tumor samples and between studies. Given these problems, it is not surprising that studies that rely solely on IHC data can be quite variable and can be inconsistent with the genomic information (13, 19). Given that proteins are the final effectors of gene function and that protein expression levels are not predicated solely on genomic alterations, questions of how changes in cell cycle/signaling genes at the protein level (determined quantitatively) impact upon the molecular pathogenesis of gliomas and affect patient outcome remain largely unanswered.

In an effort to obtain further insights into the molecular pathogenesis of gliomas, we have prospectively collected 100 “snap frozen” glioma specimens. We performed quantitative Western blot analysis to measure the levels of important regulatory proteins in gliomas. This method provides a rapid, quantitative, and accurate assessment of the role of cell cycle/signaling pathways in the molecular pathogenesis of gliomas and reveals whether these expression patterns are of prognostic value. We have quantitatively examined the levels of 12 proteins either in the pRB/p53 pathways or in signaling/immortalization pathways (EGFR, PTEN, and telomerase), most of which have been implicated in gliomagenesis at the genomic level.

The results of these experiments show a widespread pattern of loss of tumor suppressor proteins in gliomas and a more selective pattern of overexpressed oncoproteins. It becomes clear that aberrant expression of certain proteins are significantly interrelated, either directly or inversely. The results, described below, also show that the expression levels of five of these proteins are significant prognostic indicators of patient outcome.

One hundred glioma specimens have been collected prospectively (at Brigham and Women’s Hospital) under an Institutional Review Board-approved protocol for the purposes of protein expression analysis of gliomas. The histopathological breakdown of these tumors is given in Table 1. Each specimen has been “snap frozen” after surgical removal, carefully sectioned, and histopathologically reviewed. Histopathological diagnosis was made by WHO (2000) classification. Tissue containing >95% tumor was provided for this analysis. Each tumor tissue was homogenized in RIPA buffer using a Polytron homogenizer (Fisher) to generate protein lysates for Western blot analysis.

Bradford assays were performed to determine total protein concentrations, which were normalized to 1 μg/μl for all samples. Samples were then prepared in sample buffer and heated to 95°C for 5 min. The percentage of polyacrylamide composition of gels was dependent on the molecular weights of the proteins to be isolated. Twenty μl of protein lysates in sample buffer from each tissue were loaded within each well. Gels were run at constant current (40 mA) for 3–4 h for maximum separation. Wet transfer was performed for 4 h at constant voltage (40 V) using polyvinylidene difluoride membrane presoaked in methanol. The membrane was then blocked for 1 h in 5% milk in 0.2% PVDF membrane (TBST). The membranes were then washed in 0.2% TBST three times for 15 min each. One of several internal controls (actinin, actin, and cdk1) was used to verify equivalent protein loading. The membranes were sectioned to separate the regions of the protein of interest from the internal controls. Positive controls from lysates of normal astrocytes obtained from 16-week-old fetuses were included in each gel. Because normal astrocytes do not express hTRT, positive controls for hTRT gels were from lysates obtained from an hTRT-expressing glioblastoma cell line established previously from a patient’s tumor. The membranes were then incubated overnight with primary antibodies for either p16 (JC-6), pRB (XZ-56), p14^ARF (SC8613), p130 (SC317), E2F-1 (KH 81, 95), PTEN (SC7974), EGFR (AB-4), cyclin D1 (HD 33,61), cyclin E (HE12), cdk4 (SC260), or hTRT (SC7214). The p16, pRB, E2F-1, cyclin D1, and cyclin E antibodies are monoclonal antibodies that were generated at the Massachusetts General Hospital Cancer Center and have been described previously (20, 21, 22, 23, 24). The p14^ARF, p130, PTEN, cdk4, and hTRT antibodies were purchased from Santa Cruz Biotechnology. The EGFR antibody was purchased from Calbiochem. Membranes were washed three times, incubated with secondary antibody, washed again, and incubated with chemiluminescence detection agent, after which film was exposed for detection. Three separate runs/sample/marker were performed to verify reproducibility. p53 levels were detected by immunoprecipitation-Westerns. The antibody used for the immunoprecipitation was specific for wild-type p53 only (oncogene AB-5). This sensitivity and specificity of this antibody were verified by detection of wild-type p53 in normal astrocytic lines and WI-38 cells and no detection of p53 in glioblastoma cell lines known to harbor mutations of p53 or known to lack p53 (LW5, WF, D384, SF126, U138, and U251), all of which were kindly supplied by Dr. Martin Haas (University of California, San Diego, CA; Ref. 25).

The images were then scanned using Scion image analysis software (PC version of NIH image). External calibration was used using a photographic step tablet. The integrated absorbance of each band was used to quantitatively assess protein expression levels. Optical densities of bands were compared with those of corresponding controls.

Histopathological diagnosis was provided by Brigham and Women’s Hospital pathologists. Clinical data were provided by the Brigham and Women’s Hospital tumor registry and included patient age at diagnosis, treatment type, treatment dates, and clinical/survival outcomes. Complete clinical data were available on 94 of the 100 patients. Data analysis was limited to these 94 patients. Patients with glioblastoma and anaplastic astrocytoma were generally managed by surgery and postoperative radiation and chemotherapy. Patients with low-grade gliomas were generally managed by surgery alone, with radiation deferred for progressive disease. Westerns were performed in a blinded manner, without prior knowledge of clinical outcome. Patient follow-ups were updated at the time of manuscript preparation.

The Fisher Exact test was used to determine association between overexpression in a marker and the type of glioma (GBM versus non-GBM). The results were considered significant if a two-sided P was <0.05. Survival curves were calculated using the Kaplan-Meier method. Univariate analysis used the log-rank test to determine the significance of each prognostic factor. Multivariate models were fit using the Cox proportional hazards model. The comparison for the average number of anomalies between the GBM and the non-GBM group was done using t test.

Using 94 glioma specimens, which have been “snap frozen” after surgical removal, carefully sectioned, and histopathologically reviewed, we have examined the protein expression levels of each of the components of the pRB pathway that have been linked to tumorigenesis (p16^INK4, cyclin D1, cyclin E, cdk4, pRB, p130, and E2F-1). In addition, we examined components of the p53 pathway (p53 and p14^ARF), PTEN, and two oncoproteins that have been found to be misexpressed in some gliomas (EGFR and hTRT, a catalytic component of telomerase). The level of each protein was determined from three independent Western blots of each sample. The signal strength was quantified and compared with the level of expression in normal astrocytes, and an average was determined. The results of this analysis are summarized in Table 2. For this analysis, we use the term underexpression to refer to samples where there was no detectable protein, or where the level of expression was reduced 5-fold or more compared with controls (Fig. 1,A). Overexpression was defined as samples where the protein level was at least 5-fold greater than the corresponding controls (Fig. 1 B).

Most tumors underexpressed p16^INK4 or pRB (68 of 94; 72%), the two best-characterized tumor suppressors in the RB pathway. All tumors had levels of p130 that were comparable with the wild-type controls, and only one tumor showed elevated levels of cyclin D1. Intriguingly, E2F-1, a pRB-binding protein that acts as a tumor suppressor in mice, was underexpressed in over half of the samples (52 of 94). In contrast, positively acting components of this pathway (cyclin D1, cyclin E, and cdk4), which drive the phosphorylation of pRB and behave as oncogenes, were overexpressed in 22% (21 of 94) of tumors. When these changes are considered together, pRB control of E2F1 is predicted to be lost or deregulated in 91% (86 of 94) of the tumors. Most tumors also displayed some deficiency in the p53 pathway (p53 or p14^ARF deficiency in 70 of 94 tumors; 74%). Fig. 2 illustrates that most gliomas demonstrate some deficiency in the pRB/p53 pathways at the protein level. Reduction in PTEN expression was seen in 18% of all gliomas (17 of 94). EGFR was overexpressed in 26% (24 of 94). Overexpression of hTRT was seen in 17 of 94 samples (18%).

Because GBMs have clinical courses that are characteristically more aggressive than their non-GBM counterparts, we compared the protein expression profiles of GBMs versus non-GBM tumors (Table 3, A and B). When the levels of PTEN, pRB, or E2F1 were examined, no significant difference was observed. However, altered levels of several other proteins showed a tight correlation with the histopathology of the tumor. For example, underexpression of p16^INK4 and p14^ARF was significantly more common among GBMs compared with non-GBM tumors (P = 0.02). Conversely, more non-GBMs had loss of wild-type p53 expression compared with GBMs (P = 0.01).

Strikingly, the overexpression of cyclin E, EGFR, cdk4, and hTRT was significantly more common among GBMs compared with their non-GBM counterparts (P < 0.0001). Indeed, the overexpression of one or more of these proteins was observed, in conjunction with deficits in one or more tumor suppressor genes, in 73% of GBMs compared with 11% of the non-GBMs (P < 0.0001; Table 3C). This raises the possibility that the loss of expression of tumor suppressor genes may be an early event in gliomagenesis, whereas oncogene amplification may provide additional growth and survival advantages for high-grade gliomas during the latter stages of tumorigenesis. When the complete panel of proteins is considered, the average number of expression anomalies was higher in GBM versus non-GBM tumors (4.107 versus 2.6053; P < 0.0001), a result that may reflect increased genomic instability in the higher grade tumors (Fig. 3).

The data set was analyzed for significant associations among patterns of alterations in protein expression (Table 4). This revealed an inverse correlation between overexpression of EGFR and the loss of wild-type p53 (P = 0.03), when all tumors were considered (4). As expected from the close proximity of their genes, a strong and significant association was observed between the levels of p16^INK4 and p14^ARF (P < 0.0001) expression. However, in two tumors, loss of p16^INK4 expression was not accompanied by loss of p14^ARF, and these may be examples in which selective methylation of the p16^INK4 promoter has occurred (26). No other combinations of changes were found that were statistically significant, either among any given histopathological variant of glioma or for the tumors as a whole.

Surprisingly, these tumors showed no significant mutual exclusivity between underexpression of p16^INK4 and pRB (P = 0.52) or between the lack of wild-type p53 and p14^ARF deficiency (P = 0.53), as has been suggested previously (12, 27). In some tumors, these genes are mutated in a mutually exclusive manner, consistent with the idea that they form linear pathways (12, 27, 28). However, in this set of tumors, 16 tumors showed underexpression of both p16^INK4 and pRB, whereas 22 tumors had deficits in both p53 and p14^ARF. Mutations of both p16^INK4 and pRB or of mutation of p53 and p14^ARF in tumors have been reported, and these results are consistent with recent evidence that p16^INK4 can have functions that are not mediated through pRB, and that p14^ARF can act independently of p53 (26, 29, 30, 31, 32, 33).

When the tumor suppressor proteins were examined by univariate analysis, only p14^ARF expression had significant prognostic value with low or no expression of p14^ARF predicting for reduced survival (P = 0.04). This correlation is seen for gliomas as a whole but is not significant when GBMs are considered alone. Deficient expression of p16^INK4, pRB, E2F-1, p53, or PTEN tumor suppressors showed no significant correlation with patient survival. Interestingly, overexpression of cyclin D1 was seen in only 1 patient and does not appear to be a major factor in gliomagenesis.

Overexpression of either EGFR, cyclin E, cdk4, or hTRT was strongly associated with aggressive tumors. Overexpression of any one of these proteins predicted for worse survival on univariate analysis, as did advanced age (P < 0.0001). Fig. 4 A illustrates the actuarial survival of patients with tumors overexpressing one or more of these four oncogenes compared with all other tumors. Multivariate analysis (including all significant variables on univariate analysis) indicates that overexpression of EGFR (P < 0.0001), cyclin E (P = 0.0042), cdk4 (P = 0.0085), and hTRT (P < 0.0001) correlates with reduced survival. Advanced age did not reach significance in the multivariate model.

Although survival is quite limited for most GBM patients, the molecular profiles were still able to distinguish the more treatment-refractory, aggressive tumors from those that run a more indolent course. GBMs that had overexpression of either EGFR, cyclin E, cdk4, or hTRT, in combination with underexpression of one or more tumor suppressor proteins had reduced median survivals of 8 months, compared with 26.5 months for tumors that lacked such features (P < 0.0001). Fig. 4 B illustrates actuarial survival of GBM patients with tumors displaying overexpression of one or more oncoproteins compared with GBMs without overexpression. The overexpressed proteins that were associated with reduced survival for GBMs were EGFR (P = 0.0011), cdk4 (P < 0.0001), and hTRT (P < 0.0001). Cyclin E amplification was of borderline significance (P = 0.088) in predicting survival among GBM patients.

The results of this study highlight four important points:

(a) The results obtained by quantitative Western analysis of protein levels in carefully preserved glioma specimens are, for the most part, consistent with previous DNA-based studies with a few exceptions. For example, consistencies include the observation abnormal levels of key cell cycle regulators are found in almost all tumors, which supports the idea that deregulation of the cell cycle may be an obligatory event for gliomagenesis. Previous DNA-based studies have found allelic loss of 13q at the RB gene, RB gene deletions, and RB gene mutations in ∼30% of gliomas (28, 34). In this study, we observed that a similar percentage of tumors, 35%, had either no detectable wild-type protein or had levels that were at least 5-fold lower than wild-type. In a similar way, the frequency of p16 underexpression detected in this protein analysis (56%) closely matches the rate of homozygous deletion of the CDKN2A locus reported in previous DNA-based studies (13, 28, 35). In this study, 89% of gliomas had deficiencies of one or more tumor suppressor genes in the pRB pathway, and 22% had overexpression of one or more oncogenes in the pRB pathway. Other relationships such as the significant association between loss of both p16 and p14^ARF and the inverse relationship of EGFR amplification and p53 mutation/loss were also consistent with previous reports at the genomic level.

The results also contained a few differences with DNA-based studies. We observed tumors that had low levels of both pRB and p16 or that had low levels of p14^ARF, together with wild-type p53. Previous reports of genomic changes in gliomas had suggested that inactivation of these genes is mutually exclusive (4). However, there is evidence that these proteins have certain functions that are independent from one another. It is possible that in some situations, deficiencies in both (loss of both pRB and p16 protein expression and loss of both p53 and p14^ARF) may be advantageous for gliomagenesis (24, 27, 28, 29, 30, 31).

(b) The expression patterns detected in this study illustrate the high degree of heterogeneity between gliomas. In this study, we examined just 12 proteins that are known to play important roles in cell cycle control, apoptosis, cell immortalization, and cell proliferation. These products have been shown to be causally involved in tumorigenesis, and it seems likely therefore that drastic changes in the activity of these proteins will have a significant impact on the properties of the tumor cells. It is striking that the levels of these key regulators were altered in many different permutations in this collection of tumors, with the most aggressive tumors containing the largest number of changes. Lower grade tumors, such as oligodendrogliomas, are reported to demonstrate infrequent microsatellite instability and generally fewer genomic alterations than their higher grade counterparts (36). Potentially, the changes in expression may be markers of genome instability, with the most aggressive tumors being the most unstable. Alternatively, it is possible that gliomas may arise though multiple different routes, with different combinations of changes driving the tumorigenic process in different tumors.

(c) Only a small subset of these 12 proteins are useful for distinguishing between GBMs and non-GBMs. In this regard, the overexpression of proteins, such as EGFR, hTRT, cdk4, and cyclin E, is far more informative than the loss of expression of well-studied tumor suppressors such as pRB, p16, and p53. Overexpression of EGFR was observed in 23 of 56 GBMs (41%) compared with only 1 of 38 non-GBM tumors (2.6%; P < 0.0001). This finding is highly consistent with the frequency of EGFR gene amplification in GBMs versus non-GBMs from previous reports (4, 37). In the present series, there appeared to be an inverse correlation of EGFR overexpression with loss of wild-type p53 protein (P = 0.03), but no significant association was found between EGFR overexpression and the underexpression of p14^ARF. Overexpression of hTRT, the catalytic component of telomerase, was present in 17 of 56 (30%) GBMs, compared with 0 of 38 non-GBM tumors (P < 0.0001). Overexpression of cyclin E was found in 21 of 56 GBMs (38%) versus only 2 of 38 (5%) non-GBMs (P < 0.0001). Overexpression of cyclin E is a frequent event in breast tumors but has not been reported previously for gliomas (38, 39, 40). Overexpression of cdk4 protein expression was evident in 13 of 56 GBMs (23%) compared with only 1 of 38 (3%) non-GBMs (P < 0.0001). This frequency is similar to the 15% cdk4 gene amplification rate reported previously (4). When these four markers are considered together, overexpression of as EGFR, hTRT, cdk4, or cyclin E was observed in 73% of GBM versus 13% of non-GBM tumors (P < 0.0001).

(d) Perhaps most importantly, the expression levels of 5 of these 12 markers provided clear prognostic information. Increased levels of EGFR, cyclin E, cdk4, or hTRT in gliomas are strong predictors of poor survival (P < 0.0001). Of all of the tumor suppressor genes examined, only underexpression of p14^ARF showed a clear correlation with poor survival (P = 0.03). Previous studies have disagreed over the prognostic value of EGFR amplification (37, 41, 42). One study suggests that EGFR gene amplification is not an important prognostic indicator (37), but two studies that examined EGFR immunoreactivity on the protein level suggest that elevated levels of EGFR is prognostically significant (41, 42). Our study provides further evidence that the overexpression of EGFR protein strongly correlates with reduced survival in glioma patients and underscores the importance of complementing genomic data with protein/gene expression data. Likewise, there has been conflicting data on the prognostic significance of telomerase activity with clinical outcome. Several studies using telomeric repeat amplification protocol assays in determining telomerase activity failed to demonstrate enhanced telomerase activity correlating with higher tumor grade or outcome (43, 44, 45). Other studies suggest that enhanced telomerase activity as determined by telomeric repeat amplification protocol correlates with enhanced malignant potential of gliomas (46, 47, 48). The results described here indicate that the levels of hTRT expression serve as an informative marker, and that high levels of hTRT strongly correlate with reduced patient survival.

In summary, these data suggest that expression levels of cell cycle/signal transduction proteins can provide valuable insights into the molecular pathogenesis and clinical behavior of gliomas. Protein expression anomalies were present in 92 of the 94 glioma specimens (98%) in this series, indicating that cell cycle/signaling anomalies at some level are obligatory in gliomagenesis. Because GBMs are routinely managed by adjuvant therapy (e.g., radiation and chemotherapy), the observation that increased levels of EGFR, cdk4, hTRT, and cyclin E strongly correlate with reduced survival may suggest that these proteins are associated with resistance to DNA-damaging agents. Further investigations are needed to better understand how these oncogene amplifications are related to specific upstream signal transduction pathways and how they can be manipulated to influence the natural history of the disease and treatment outcome.