Molecular Classification and Survival Prediction in Human Gliomas Based on Proteome Analysis Free

The extensive heterogeneity of astrocytic tumors has made their pathological classification rather difficult; some identical low-grade astrocytomas progress to glioblastoma, whereas others persist in a dormant state for many years. Although several genetic aberrations and gene expression changes have been identified as associated with oncogenesis and malignant progression of the astrocytic tumors (1), their contribution to the clinical classification of gliomas has been limited. More comprehensive characterization of the gene expression that may correlate with the clinical behavior of gliomas is needed to establish an accurate molecular classification to predict the prognosis and treatment response of patients.

Molecular expression profiles using oligonucleotides or cDNA-based microarrays have been used to derive a useful molecular-based classification for the choice of treatment in several types of human cancers (2, 3). Gene expression profiles have identified many genes with distinct expression patterns among different histological types and grades of intracerebral gliomas (4, 5, 6). However, biological systems comprise protein components resulting from transcriptional and post-transcriptional control, post-translational modifications, and shifts in proteins among the different cellular compartments. These properties cannot be analyzed by microarray systems at the RNA level, whereas proteome analysis such as high-resolution two-dimensional gel electrophoresis (2DE) allows separation and visualization of the protein contents of a cellular sample (7). The recent introduction of commercially available immobilized ampholyte pH gradients, more standardized 2DE methods, and bioinformatics protein databases now make it more feasible to screen tumor-related global protein changes on a genome-wide scale (8, 9).

In this study, we examined several human glioma specimens, using 2DE and matrix-assisted laser desorption/ionization time-of-flight mass spectrometry to compare proteome profiling patterns between low- and high-grade tumors. The results showed that proteome-based clustering of gliomas significantly correlated with patient survival. We also identified a group of proteins that were differentially expressed in proportion to the degree of histological grading, which might be novel biomarkers to predict the clinical aggressiveness of gliomas and rational targets for antiglioma therapy.

After obtaining written informed consent from the patients or guardians, we analyzed 75 surgically resected human glioma tissues from newly diagnosed cases, along with 10 normal brain tissues. The normal brain tissues were obtained from patients undergoing resection of extra-axial brain tumors or epilepsy surgery. The resected tumors or normal brain samples were snap-frozen in liquid nitrogen and stored at −80°C for protein extraction immediately after surgical resection. The other portion of the tumor was fixed in 10% formaldehyde and embedded in paraffin for histopathological diagnosis. Of the 75 glioma tissues, 52 were classified as glioblastoma multiforme (GM; grade IV), 13 as anaplastic astrocytoma (grade III), and 10 samples as astrocytoma (grade II) by a neuropathologist according to the WHO criteria. The protocol of this experimental study was approved by the Institutional Review Board.

The frozen samples were mechanically homogenized in lysis buffer containing 30 mm Tris-HCl (pH 7.5), 150 mm NaCl, 1% Triton X-100, 10% glycerol, and protease inhibitor cocktail to generate protein lysates. Total protein concentrations were normalized to 1 μg/μl for all samples. Each protein sample (100 μg) was lyophilized and then mixed with sample buffer containing 7 m urea, 2 m thiourea, 4% (w/v) 3-[(3-cholamidopropyl)dimethylammonio]-1-propanesulfonic acid (CHAPS), and 1% (w/v) DTT. Isoelectric focusing for the first dimension of protein separation was performed with a Multiphore II electrophoresis system (Amersham Biosciences, Piscataway, NJ). Immobiline Dry IPG strips (7 cm, pH 3–10 nonlinear IPG strips; Amersham Biosciences) were rehydrated overnight with the sample/rehydration buffer mixture. The strips were subjected to electrophoresis using a ramping IPG strip (200–5000 V) focusing algorithm. After the isoelectric focusing, the gel strips were equilibrated with SDS equilibration buffer and electrophoresed in vertical SDS-PAGE slab gels containing 12.5% acrylamide. The gels were stained with the Silver Quest silver staining kit, which was compatible with mass spectrometry (Invitrogen Japan, K.K., Tokyo, Japan).

Silver-stained gels were analyzed with the Phoretix 2D Advanced software (Version 5.01; Nonlinear Dynamic, Ltd., Newcastle, United Kingdom). Gel images from 10 samples of normal brain tissues were matched to generate a composite reference image including most of the possible spots. The total integrated spot volumes, the average volume of individual spots, and the SD were calculated, with the volume of the internal standard (keratin-1) used for normalization. To maximize the chance of detecting tumor-specific protein alterations, we also directly visually analyzed the gels. We compared the proteome profiles of the reference gel with the tumor gel; only spots that were present on the reference gel and completely absent from the matched gel (on-off changes) were defined as altered.

The protein spots were excised from the gels, and in-gel digestion was performed with an enzyme solution containing 50 mm NH₄HCO₃, 5 mm CaCl₂, and 12.5 ng/μl trypsin. Aliquots of the purified samples were spotted on matrix crystals of α-cyano-4-hydroxyl-cinnamic acid on a stainless steel target and air dried. Mass determinations were performed on the AXIMA-CFR mass spectrometer (Shimadzu Co. Ltd., Kyoto, Japan). The proteins were identified by the peptide mass fingerprinting method, using Mascot Search on the Web (Matrix Science, Ltd., London, United Kingdom).

Immunohistochemical analysis was performed on formalin-fixed, paraffin-embedded sections. The primary antibodies used were rabbit polyclonal anti-cAMP-responsive element binding protein-1 antibody (1:200 dilution), mouse monoclonal anti-Rho A antibody (1:100 dilution), rabbit polyclonal anti-Rac1 antibody (1:200 dilution), goat polyclonal anti-GRP78 antibody (1:100 dilution), and goat polyclonal antienolase antibody (1:100 dilution), all from Santa Cruz Biotechnology (Santa Cruz, CA). The sections were incubated with the antibodies overnight followed by incubation with biotinylated secondary antibodies (DakoCytomation, Copenhagen, Denmark). The bound antibodies were visualized with avidin–biotinylated peroxidase complex and diaminobenzidine tetrachloride.

We carried out hierarchical cluster analysis for the tissue specimens based on the proteome profiling patterns in the 2DE gels. The squared Euclidean distance based on the standardized variables was used to calculate the distance between any two samples, and agglomerative hierarchical cluster analysis was performed using the method of complete linkage (SPSS, Inc., Chicago, IL). Kaplan–Meier survival curves were generated with StatView software (SAS Institute Inc., Cary, NC), and the survival periods were compared using the log-rank test. To screen out proteins that significantly affected histological grading, we used discriminant analysis (SPSS Inc.).

We comprehensively analyzed the protein expression patterns in 75 glioma specimens by use of a 2DE technology with immobilized pH gradients (Fig. 1). Approximately 350 protein spots were identified on each gel by a conservative spot selection software algorithm based on certain spot intensities. Many protein spots were differentially expressed among samples (Fig. 2). A composite reference gel of the 10 normal brain tissues contained a total of 631 protein spots. Gel comparisons between the reference gel and the gels from each glioma specimen revealed matching means of 64 ± 6.2% for astrocytoma and 47 ± 4.9% for GM (P < 0.05). Among the spots on the reference gel, 181 protein spots (28.7%) did not match with any spot on other gels from glioma tissues. These proteins may be specifically translated and functional in the normal brain. Among the 631 protein spots on the reference gel, 257 (40.7%) were found on more than two gels from glioma tissues, and many of the remaining unmatched protein spots appeared only on one gel, suggesting that the tumor-deregulated proteins were to a large extent sample specific and were proteolytic fragments or post-translationally modified proteins.

We performed hierarchical cluster analysis of the 75 tumor specimens and the 10 normal brain samples based on the presence or absence of the 257 protein spots to characterize the proteome profiling patterns associated with variable biological aggressiveness in the human gliomas. A dendrogram in the sample axis is shown in Fig. 3,A. The samples taken from the normal brain tissues were completely categorized into single tissue-specific branches, indicating the validity and clinical significance of the proteome analysis. Seven astrocytoma samples were clustered in an area close to the normal brain tissues, and the patients manifested favorable clinical courses without recurrence. In contrast, two of the other three patients, with astrocytomas that clustered away from the normal brain tissues, had recurrences within 4 years after surgery. Samples from most patients with GM who had relatively long survival periods (>2 years) were also well categorized to an area near the normal brain tissue in the clustering map (Fig. 3 A).

The 52 patients with GM were divided into two groups based on the branching patterns of the dendrogram and the squared Euclidean distance coefficient with reference to the normal brain tissues. We compared the overall survival periods between the two groups: the group with GM that was located near the normal brain tissues on the clustering map (21 patients) and the group with GM that was located relatively away from the normal brain tissues on the clustering map (31 patients; Fig. 3 B). There were no significant differences between the two groups regarding potential prognostic factors such as age, performance status, extent of surgical resection, and dose administered in radiation therapy. The patients with GM located near the normal brain tissues on the clustering map lived significantly longer than the patients whose tumors were located distant from the normal brain tissues on the clustering map (P < 0.001, log-rank test). In accordance with the survival data, six cases in the latter group manifested leptomeningeal dissemination, whereas none of the former group showed such dissemination patterns (P = 0.0321, χ² test). This significant difference indicates that the tumors located distant from the normal brain tissues are more invasive in nature than the tumors located near the normal brain tissues in the proteome-based clustering.

To identify the proteins that significantly correlated with histological grading, we performed discriminant analysis based on expression of the 257 protein spots. The selected protein spots were excised and analyzed by matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (Fig. 4). Some protein spots could be identified by positive gel matching of protein expression patterns from several existing 2DE gel databases. Twenty-five known proteins and 12 unknown proteins were extracted from the analysis (Table 1). These 25 known proteins were functionally categorized, based on their known or inferred biological functions, into signal transduction-related proteins, molecular chaperones, transcription and translation regulators, cell cycle-mediating proteins, extracellular matrix-related proteins, and cell adhesion molecules.

The results of the proteome-based analysis for a subset of proteins related to tumor aggressiveness was confirmed by immunohistochemistry. Among the 37 specific proteins identified, antibodies to 5 were readily available for immunohistochemical analysis on paraffin-embedded sections. We observed high levels of cAMP-responsive element binding protein-1 in the nuclei of grade IV tumors, whereas other grade II tumors hardly stained with the antibody (Fig. 5). GRP78, Rho A, and Rac1 were also more abundantly expressed in the cytoplasm and processes of the tumor cells in the grade IV tumor cells compared with the lower grade tumors. In contrast, we observed higher levels of enolase in the lower grade tumors than the grade IV tumors, as indicated by the proteome analysis.

We examined the proteomes of human gliomas to establish a novel molecular classification that can distinguish aggressive tumors that are resistant to various treatments from the less-aggressive tumors among gliomas of the same histological grade. The results of hierarchical clustering analysis showed that tumor classification based on proteome profiling patterns could generate an accurate patient stratification that is more clinically relevant than the conventional histological classification. We consider proteome-based classification an efficient class-discovery tool for intracerebral gliomas, including survival prediction.

Because the functional molecules in the cell are proteins, proteome analysis based on 2DE would have several advantages over cDNA/oligonucleotide microarray systems for clinical use. First, the system measures primarily high-abundance proteins, which are ideal tumor biomarkers because they can be easily measured and targeted. mRNA levels are not directly associated with the amounts of functional proteins (10). Second, by use of proteomics, we can analyze the post-translational modifications that are essential for protein formation and function. For many proteins, post-translational modifications are important for their functions, and they can be analyzed only by the methods used in proteomics (11, 12). In contrast, 2DE analysis usually cannot detect hydrophobic membrane proteins, proteins larger than 100 kDa, or proteins with pI values outside the range 3–10. These limitations of standard 2DE could account for the failure of proteome analysis to identify proteins known to be overexpressed in glioma tissues, such as epidermal growth factor receptor, insulin-like growth factor binding protein 2, or vascular endothelial growth factor (6, 13).

Many small G-proteins have been reported to be involved in the malignant transformation of gliomas (14, 15, 16). The G-proteins bind to GTP and then activate a cascade of serine-threonine kinase, such as Raf and other so-called ras effectors, which transduce signals by cAMP-responsive element binding protein-1, a transcription regulator, resulting in alteration of the gene expressions (17). In the present study, eight small G-proteins were identified as related to malignant transformation. Among them, RalA, Rab3B, and nucleolar GTP-binding protein, recently recognized as a novel G-protein family, have not been reported as associated with the oncogenesis of gliomas. The up-regulated Rho family (RhoA and Rac1) may induce actin filament-mediated tumor cell migration as well as prevent apoptosis, in part through the ezrin/radixin/moesin group of proteins (14, 18). Recently, Senger et al. (15) emphasized that Rac1 regulates a major survival pathway in most human gliomas and that suppression of Rac1 activity induces death in glioma cells. These results suggest that despite the lack of oncogenic ras mutations in astrocytomas, aberrant G-protein signaling plays a crucial role in malignant transformation of gliomas and would be a rational target for antiglioma therapy.

Chaperone proteins reduce stress-induced denaturation and aggregation of intracellular proteins and are reported to exert protective actions by interfering with the stress-induced apoptotic pathway in healthy neuronal cells (19). The results obtained in our study show that chaperone proteins work diversely in glioma cells; GRP78, prohibitin, and heat shock protein 27 were associated with high-grade tumors, whereas protein disulfide isomerase and α-crystallin were associated with low-grade tumors. Various proteases and their inhibitors were also found to be important in glioma invasion and angiogenesis. The plasminogen activator system, including plasminogen activator inhibitor type 1, was previously demonstrated to be associated with higher grades of glioma (20). This system interacts with annexin II, which is known to be an independent prognostic factor for poor outcome in gliomas (21, 22). We ascertained in the present study that the expression of plasminogen activator inhibitor type 1, annexin II, annexin IV, and cathepsin D proteins was increased in high-grade gliomas.

We have shown a novel system of classification based on protein expression by analyzing 75 glioma specimens with 2DE-based proteome analysis. Our results also demonstrate that the biological characteristics of intracerebral glioma are defined by numerous proteins, which implies that the aggressiveness of gliomas can not be predicted by single or small numbers of protein markers but rather by combinations of many proteins. Many of the identified proteins work in signal transduction and as molecular chaperones. These proteins could be predictive markers for the aggressiveness of gliomas and could be direct and rational targets for antiglioma therapy.

We thank Noriyuki Ozima and Minoru Yamaguchi (Shimadzu Co., Ltd.) for technical assistance in the matrix-assisted laser desorption/ionization time-of-flight mass spectrometry analysis using AXIMA-CFR.