OUP user menu

IQGAP1 and IGFBP2: Valuable Biomarkers for Determining Prognosis in Glioma Patients

Kerrie L. McDonald PhD, Maree G. O'Sullivan MSc, Jonathon F. Parkinson MBBS, Janet M. Shaw BSc, Cathy A. Payne BSc, Janice M. Brewer MBBS, Lawrence Young BSc, Dianne J. Reader BSc, Helen T. Wheeler MBBS, FRACP, Raymond J. Cook MBBS, FRACS, Michael T. Biggs MBBS, FRACS, Nicholas S. Little MBBS, FRACS, Charlie Teo MBBS, FRACS, Glenn Stone PhD, Bruce G. Robinson MD, MSc, FRACP
DOI: http://dx.doi.org/10.1097/nen.0b013e31804567d7 405-417 First published online: 1 May 2007

Abstract

Clinical treatment decisions and the survival outcomes of patients with gliomas are directly impacted by accurate tumor classification. New and more reliable prognostic markers are needed to better identify the variable duration of survival among histologically defined glioma grades. Microarray expression analysis and immunohistochemistry were used to identify biomarkers associated with gliomas with more aggressive biologic behaviors. The protein expression of IQGAP1 and IGFBP2, when used in conjunction with the World Health Organization grading system, readily identified and defined a subgroup of patients with grade III gliomas whose prognosis was poor. In addition, in patients with glioblastoma multiforme, in whom IQGAP1 and IGFBP2 were absent, long-term survival of more than 3 years was observed. The use of these markers confirmed a nonuniform distribution of survival in those with World Health Organization grade III and IV tumors. Thus, IQGAP1 and IGFBP2 immunostaining supplements current histologic grading by offering additional prognostic and predictive information.

Key Words
  • Glioma
  • IQGAP1
  • IGFBP2
  • Microarray
  • Microvascular proliferation
  • Necrosis

Introduction

Diffuse infiltrative gliomas are the most common primary tumor of the brain and are associated with very poor survival. High variability exists in survival rates of patients, highlighting the heavy reliance on accurate tumor classification for the clinical management of all glioma types. The World Health Organization (WHO) classification system is widely used for the grading of gliomas (1). In the current WHO 2000 version, diffuse astrocytomas are classified according to a three-tiered grading system (grades II-IV) and include glioblastoma multiforme (GBMs), which are classified as grade IV. Oligodendrogliomas and oligoastrocytomas are divided into 2 grades (grades II and III) with anaplastic (grade III) being the highest grade designation given to these tumors. Grades III and IV are generally regarded as high-grade because of rapid tumor progression and short survival.

Anaplastic oligodendrogliomas vary widely in their clinical behavior, with mean survival ranging from 1 to >10 years (2). The introduction of routine testing of oligodendrogliomas for loss of heterozygosity (LOH) on chromosomes 1p and 19q has helped identify patients with an increased sensitivity to chemotherapy and better prognosis (3,4). However, there is still significant variability in the clinical behavior of the majority of anaplastic oligodendrogliomas that retain the 1p and 19q chromosomal arms (3). In addition, the basis of long-term survival in certain patients with GBMs also remains poorly understood. Long-term survival is defined by the diagnosis of a GBM and subsequent survival for >3 years after initial diagnosis (5). Only younger age and a high Karnofsky performance score have been associated with long-term survival (6). Genetic markers are required to better delineate the nonuniform distribution of clinical behavior that exists within each grade.

There is great enthusiasm for microarray technology and its utility in genetic profiling of individual gliomas. Microarray analysis of high-grade gliomas has helped identify new targets associated with glioma invasion (7-13). For example, a 44-gene expression signature has been used to segregate gliomas into 4 survival-based groups, including a poor prognostic group with a median survival of 293 days (14).

We used microarray analysis to identify prognostic markers for survival in patients with glioma. In addition to the WHO 2000 classification system, we classified all gliomas into a modified grading scheme based on the histologic features typically associated with poor survival, namely, necrosis and microvascular proliferation (MVP). These 2 groups were designated NMVP-positive (necrosis and MVP present in tumor) and NMVP-negative (necrosis and MVP absent in tumor). This partitioning of gliomas allowed for the identification of markers strongly correlated with the most aggressive biologic behavior. Unique to this study was the use of the multivariate analysis methods, GeneRave and Stepwise Diagonal Discriminant Analysis, to identify small sets of genes with better predictive accuracy than the usually much larger sets found by existing methods. The candidate gene sets were validated using quantitative polymerase chain reaction (PCR) and protein expression was investigated by immunohistochemistry.

Materials and Methods

Tumor Sampling and Data Collection

Glioma samples were collected from patients who underwent surgery at Royal North Shore Hospital, North Shore Private Hospital, and Prince of Wales Private Hospital, NSW, Australia. Approval for this study was obtained from the human research ethics committees of the participating institutions. No patients received chemotherapy or radiotherapy before surgery. Surgically removed gliomas were snap-frozen in liquid nitrogen immediately and stored at −80°C until RNA extraction. Frozen tissue was available for 71 gliomas (37 samples included in the microarray analysis and an additional 34 samples included in the quantitative PCR). Formalin-fixed, paraffin-embedded glioma blocks were provided by the Department of Anatomical Pathology, Royal North Shore Hospital. Paraffin-embedded tissue was available for the 71 frozen tumor samples and an additional 72 glioma samples (total of 143 gliomas).

All gliomas were obtained from the initial surgery and graded according to the WHO 2000 criteria by experienced neuropathologists. The following subtypes were included in this study: astrocytoma grade II (AII), astrocytoma grade III (AIII), GBM, oligodendroglioma grade II (OdgII), oligodendroglioma grade III (OdgIII), and oligoastrocytoma grade III (OAIII). Clinical information collected included patient age, gender, resection type (debulking or biopsy), tumor location, and percent Ki67 staining. Data on the LOH of chromosome 1p and 19q, routinely tested in patients with oligodendrogliomas, was also available. Survival time (days) was counted from the patient admission date. The census date for survival was July 31, 2006. Only gliomas that were surgically debulked were included in the survival analysis. Pathology reports for all patients included in this study were reviewed for the presence of necrosis and MVP. In our data set, 96% of gliomas with described necrosis also had MVP present. Gliomas with both necrosis and MVP were grouped together and referred to as NMVP-positive gliomas, whereas gliomas in which necrosis or MVP were absent were grouped together and referred to as NMVP-negative gliomas. Clinical information for the 143 study patients is summarized in Table 1.

View this table:
TABLE 1.

Microarray analysis using a two-color platform involves comparative hybridization of the sample of interest against a reference, for example, tumor versus normal. It was logistically and ethically not possible to obtain sufficient matched normal brain tissue; therefore, a commercial total brain RNA mix was used (Ambion, Inc., Austin, TX). The commercial RNA was isolated from a donor who had no history of cancer or brain disorders. All tumors were hybridized against this normal total brain reference.

Preparation of RNA Samples

Total RNA was extracted from fresh-frozen brain tumor tissue using QIAZOL reagent (QIAGEN GmbH, Hilden, Germany) and purified by precipitation with 2.5 mol/L Lithium chloride according to the manufacturer's protocols (Ambion, Inc.). Purified RNA was quantitated by UV absorbance at 260 and 280 nm and assessed qualitatively by formaldehyde agarose gel electrophoresis. Adequate amounts of RNA for the microarray experiments were available for 17 NMVP-positive gliomas and 20 NMVP-negative gliomas.

Microarray Experimental Procedure

Microarray slides printed with the Compugen 19,000 human oligonucleotide library were obtained from the Adelaide Microarray Facility, University of Adelaide, Australia. An array list, H19K array list v3.01.gal, was provided with the chips. Slides from one print batch only were used in these experiments.

For each glioma, 20 μg of purified RNA from both the tumor and the normal total brain control were reversed transcribed using Superscript III (Invitrogen, San Diego, CA) and anchored oligo-(dT20VN) in the presence of Cy5-dCTP and Cy3-dCTP (Amersham Inc., Piscataway NJ) as described previously (15). After direct labeling, the 2 probes were hybridized for 20 hours at 50°C to a microarray slide. The slides were then washed, immediately dried, and scanned with a 10-μm resolution on a GenePix 4000A scanner (Molecular Devices, Sunnyvale, CA) at wavelengths of 635 and 532 nm for Cy5- and Cy3-labeled probes, respectively. The resulting TIFF images were analyzed by GenePix Pro 4.0 software (Molecular Devices).

Microarray Data Analysis

Both univariate and multivariate approaches were used to detect differential expression among tumor groups. Before statistical analysis, preprocessing and normalization of the raw GenePix data were required to remove artifacts and systematic effects due to processing of the arrays. Preprocessing of the data was carried out utilizing R statistical software (Version 2.0.1) libraries contained in Bioconductor, the open source software providing tools for the analysis of genomic data (http://www.bioconductor.org/). No filtering or background correction was performed on the data, and normalization was performed by printTipLoess.

The univariate analysis was performed using the package Limma, which has been specifically designed for detecting differential expression in microarray data. The software application was developed by Smyth (16) and involves the use of a moderated t statistic. The top 100 ranked differentially expressed genes were identified for each tumor group comparison.

Multivariate statistical methods are more appropriate for high-dimensional data such as microarray intensities, and use of these methods avoids the need to adjust for multiple testing by controlling for the family-wise error rate or false discovery rate. Two multivariate methods developed by Commonwealth Scientific and Industrial Research Organization (CSIRO), GeneRave (17) and SDDA, were used to select genes that discriminate between NMVP-positive and NMVP-negative gliomas.

GeneRave has a generalized linear model framework and is coupled with a Bayesian approach to variable selection. The GeneRave method involves fitting logistic regression models to build a model that gives the best separation between groups. GeneRave utilizes specialized model-fitting EM algorithms and achieves almost unbiased estimation of error rates and model significance through cross-validation and permutation. The GeneRave algorithms perform well when there is little prior knowledge and have been shown to be computationally fast and scale up well to handle large numbers of variables.

SDDA is a technique based on linear discriminant analysis, which is used to build multiclass classifiers by assuming that the gene intensity measures come from a multivariate normal distribution with means differing by a class fixed (unknown) covariance matrix.

Validation of Genes Using Quantitative PCR

Target genes selected from the microarray analysis were validated using quantitative real-time PCR in 25 NMVP-positive gliomas and 23 NMVP-negative gliomas (for which adequate amounts of RNA were available). cDNA was synthesized from 5 μg of RNA using random hexamers as described previously (15). All quantitative PCR analyses were performed using a 5’ nuclease technique with specific TaqMan Gene Expression Assays (Applied Biosystems, Foster City, CA) and TaqMan Universal PCR Master Mix, NO AmpErase UNG (Applied Biosystems) on a RotorGene 3000 (Corbett Research, Mortlake, NSW, Australia). Ribosomal 18S RNA was chosen as the endogenous control for normalization. Differences between classes were assessed statistically using REST-XL (Relative Expression Software Tool, Version 2) (18) with which relative expression ratios are computed based on the PCR efficiency and crossing point differences. Student t-test analysis was used to evaluate the statistical significance of the mRNA expression levels of the target genes between NMVP-positive and NMVP-negative gliomas (Stata statistical software, Version 8.2; (StataCorp, College Station, TX).

Antibodies

A mouse monoclonal antibody against IQGAP1 (BD Transduction Laboratories, Macquarie University Research Park, NSW, Australia) was used at a concentration of 1:300. The negative control, mouse IgG1 was purchased from Dako, Inc. (Carpinteria, CA). A goat polyclonal antibody against IGFBP2 (C-18; Santa Cruz Biotechnology, Inc., Santa Cruz, CA) was used at a concentration of 1:150. Normal goat IgG was purchased from Santa Cruz Biotechnology.

Immunohistochemistry

Immunohistochemical studies were performed on serial 4-μm sections from 143 paraffin-embedded glioma blocks (73 NMVP-positive gliomas and 70 NMVP-negative gliomas) and normal brain samples. One section from each sample was stained with hematoxylin and eosin to facilitate histologic assessment. Sections were deparaffinized, rehydrated, treated with EDTA retrieval solution (pH 9.0) or citrate-based retrieval solution (pH 6.0) for 20 minutes at 95°C, and blocked with 0.3% hydrogen peroxide before the application of primary antibodies (Dako Aust. Pty Ltd, Botany, Australia). The IQGAP1 antibody was detected using the Envision+ Dual Link Peroxidase Detection System (Dako Aust. Pty Ltd) and IGFBP2 was detected using the LSAB+ Streptavidin Peroxidase Detection System (Dako Aust. Pty Ltd). A DAB+ liquid stable substrate system was used for visualization. All immunohistochemical staining was performed on an autostainer (Autostainer Plus; Dako, Inc.). Sections were counterstained with hematoxylin. Two types of negative controls, substituting the matched mouse IgG isotype and goat nonimmune IgG in the staining protocol, were used.

Immunostaining results were evaluated semiquantitatively by 3 independent observers and a neuropathologist. IQGAP1 immunostaining was scored as follows: 0, negative staining; 1, weak cytoplasmic staining (<5% of examined tumor cells); 2, moderate cytoplasmic staining (<20% of examined tumor cells); 3, moderate to strong cytoplasmic staining (<25% of examined tumor cells); and 4, strong cytoplasmic staining (>25% of examined tumor cells). Gliomas that scored ≥3 were regarded as positive. Gliomas that scored ≤2 were regarded as negative for IQGAP1 protein expression. IGFBP2 immunostaining was scored as follows: 0, negative staining; 1, weak cytoplasmic staining (<5% of examined tumor cells); 2, moderate cytoplasmic staining (<25% of examined tumor cells); and 3, strong membranous and cytoplasmic staining (>25% of examined tumor cells). Gliomas that scored ≥2 for IGFBP2 protein expression were regarded to be positive, and gliomas scoring ≤1 were regarded as negative. Chi-square analysis was used to evaluate the statistical significance of the immunostaining results (Stata statistical software, Version 8.2). The agreement (average κ and range) among the observers was assessed using Stata. κ >0.80 indicates excellent interobserver agreement in excess of chance, whereas κ <0.20 indicates poor agreement (19). Logistic regression analysis was used to assess the association between IGFBP2 and IQGAP1 protein expression levels and NMVP status (Stata).

Survival Analysis

Cox proportional hazards regression analysis was performed to assess IGFBP2 and IQGAP1 protein expression levels as prognostic indicators for survival adjusted for age (R statistical software, Version 2.0.1). The proportional hazards assumption was tested and found to be acceptable. Kaplan-Meier survival analysis was used to generate survival curves and estimates of median survival times. The logrank test was used to compare survival curves for samples split by age, NMVP status, tumor type, IQGAP1 protein expression, and IGFBP2 protein expression (Stata).

Results

Survival Analysis of the Glioma Dataset

To determine the correlation between WHO tumor grade and survival, we plotted Kaplan-Meier survival curves of the 143 patients entered in this study (Fig. 1A). These results confirm that histologic (WHO) grade is a good predictor of survival, with significantly shorter survival observed in patients with GBM. There is a subgroup of patients within the OdgIII and OAIII grades who have survival similar to those with GBM in the first year; however, other patients within these same grades show much longer survival (Fig. 1A). To identify genes associated with this more aggressive biologic behavior, we assigned the gliomas (n = 143) into 2 broad groups: NMVP-positive (n = 73) containing 52 GBMs, 8 OAIIIs, and 13 OdgIIIs; or NMVP-negative (n = 70) containing 4 OAIIIs, 13 OdgIIIs, 20 OdgIIs, 20 AIIIs, and 13 AIIs. Kaplan-Meier plots of these 2 major groups show that patients with NMVP-positive gliomas had a median survival of 337 days (Fig. 1B). Only 10% of patients with NMVP-positive gliomas were alive at 5 years compared with 80% of those with NMVP-negative gliomas. Separating the gliomas according to necrosis and MVP has highlighted the separation of OAIIIs and OdgIIIs into 2 distinct groups. The survival curves also confirm that NMVP status is an important independent predictor of survival. The assignment of gliomas to NMVP-positive and -negative groups provides a larger sample size to allow sufficient power for the microarray and multivariate analyses.

FIGURE 1.

Survival analysis of histology based classification of gliomas. (A) Disease-specific survival of the patients entered in this studies classified according to histologic grade (World Health Organization 2000): astrocytoma grade II (AII), n =12; astrocytoma grade III (AIII), n = 20; glioblastoma multiforme (GBM), n = 52; oligodendroglioma grade II (OdgII), n = 20; oligodendroglioma grade III (OdgIII), n = 26; and oligoastrocytoma grade III (OAIII), n = 12. (B) Disease-specific survival of patients entered in this study and designated as gliomas with necrosis and microvascular proliferation (NMVP-positive), n = 73, or gliomas without necrosis or microvascular proliferation (NMVP-negative), n = 70 (p < 0.001, logrank test).

Microarray Data Analysis

Gene expression profiles of the 17 NMVP-positive gliomas (12 GBMs, 3 OAIIIs, and 2 OdgIIIs) compared with the 20 NMVP-negative gliomas (5 AIIs, 5 AIIIs, 2 OAIIIs, 3 OdgIIIs, and 5 OdgIIs) showed significant upregulation of 185 genes and downregulation of 42 genes. The 2 algorithms, SDDA and GeneRave, were applied to the full gene dataset to identify smaller sets of expressed genes uniquely associated with the NMVP-positive gliomas. In this study we used SDDA to identify 7 gene combinations with known biologic function (Homer1, IQGAP1, LGALS1, LRRC20, IGFBP2, CARHSP1, and COPZ2) that could discriminate between the 2 broad groups of gliomas (Table 2). GeneRave was utilized to identify 7 gene combinations with known biologic function (IGFBP2, C1QL1, CHI3L1 [commonly referred to by its protein product, YKL40], SERPINA3, SPP1, RBP1, and LGALS1). We chose the gene sets Homer1 and IQGAP1 and IGFBP2 and C1CL1 for further validation and analysis based on their rankings as genes with the highest predictive accuracy of identifying gliomas with necrosis and MVP by GeneRave and SDDA (79% and 84%, respectively) (Table 2). IGFBP2 was of particular interest because both algorithms nominated this gene as the top discriminator of NMVP-positive and -negative gliomas (Table 2). By plotting the gene expression values of the 37 gliomas, the scatter plots demonstrate excellent separation of NMVP-positive gliomas from the NMVP-negative gliomas using IQGAP1 and Homer1 (Fig. 2A) and IGFBP2 and C1QL1 (Fig. 2B).

View this table:
TABLE 2.
FIGURE 2.

Plot of normalized log2 gene expression results demonstrating separation of the 20 gliomas without necrosis and microvascular proliferation (NMVP-negative) from 17 gliomas with necrosis and microvascular proliferation (NMVP-positive). (A) Separation of NMVP-positive (+ve) and NMVP-negative (-ve) gliomas with the gene subsets Homer1 and IQGAP1. (B) Separation of NMVP (+ve and -ve) gliomas with the gene subsets IGFBP2 and C1QL1. The gene subsets were validated by quantitative polymerase chain reaction (PCR). (C) Log-transformed class median values for the quantitative PCR expression (normalized to the ribosomal 18S endogenous control) for each of the 25 NMVP-positive gliomas and 23 NMVP-negative gliomas after comparison to a normal brain reference (columns); SE of the expression across all examined samples (bars). *,Significant at p < 0.005; **, significant at p < 0.001 derived from Student t-test analysis.

Quantitative PCR Validation of Gene Sets

Quantitative PCR was used to quantify mRNA levels of Homer1, IQGAP1, IGFBP2, and C1QL1 in 48 gliomas and normal brain. The validation dataset comprised 14 tumors used in the microarray analysis, complemented with an additional 34 untested tumors (n = 48). The total number used in the validation included 25 NMVP-positive gliomas (13 GBMs, 7 OAIIIs, and 5 OdgIIIs) and 23 NMVP-negative gliomas (1 OAIII, 3 OdgIIIs, 6 OdgIIs, 7 AIII, and 9 AIIs). Upregulation of IQGAP1 and IGFBP2 was observed in the NMVP-positive gliomas (p < 0.001 for both target genes) (Fig. 2C). In this NMVP-positive group the mRNA expression of IQGAP1 was 4-fold higher compared with NMVP-negative gliomas whereas IGFBP2 showed 12-fold higher mRNA expression in the NMVP-positive group compared with NMVP-negative gliomas. No significant difference in mRNA expression was observed between the glioma groups with Homer1 and C1QL1 gene sets (Fig. 2C) (p = 0.465, p = 0.721, respectively).

Immunohistochemical Expression of IQGAP1 and IGFBP2

The interobserver agreement among the 3 observers for IQGAP1 and IGFBP2 expression scores in the 143 gliomas tested were κ = 0.91 and 0.97, for each respective protein scored. Logistic regression analysis confirmed that protein expression scores of IQGAP and IGFBP2 were strongly associated with NMVP-positive gliomas (p < 0.001). For each unit increase in IQGAP1 or IGFBP2 score, the odds that the tumor will have necrosis and MVP increase 2.4- and 4.7-fold, respectively (Fig. 3).

FIGURE 3.

Expression and localization of IQGAP1 and IGFBP2 in normal brain tissue and gliomas classified by histology and by the modified scheme based on necrosis and MVP (NMVP). (A) Normal brain and astrocytoma grade III (AIII), oligodendroglioma grade III (OdgIII), and oligoastrocytoma grade III (OAIII) gliomas, alternatively classified as NMVP-negative (-ve) with little or no IQGAP1 immunostaining. (B) Glioblastoma multiforme (GBM) 1, GBM 2, OdgIII, and OAIII, alternatively classified as NMVP-positive (+ve). Strong IQGAP1 staining is visible. (C) Normal brain and AIII, OdgIII, and OAIII gliomas, alternatively classified as NMVP-negative with little or no IGFBP2 immunostaining. (D) GBM1, GBM2, OdgIII, and OAIII, alternatively classified as NMVP-positive, demonstrating moderate to strong cytoplasmic IGFBP2 staining. Original magnification: 200×.

Expression and Localization of IQGAP1 in Normal Brain and Gliomas

No cytoplasmic or membranous immunostaining of IQGAP1 was observed in the normal glial tissue or in 6 of 70 (9%) of the NMVP-negative gliomas (Fig. 3A). Some uptake was noted in red blood cells within the vascular spaces and in the endothelial cells. Widespread weak cytoplasmic IQGAP1 (scores 1-2) was observed in 45 of 70 (64%) NMVP-negative gliomas tested (Fig. 3A; Table 3). Interestingly, strong positive immunostaining of IQGAP1 was also observed in the endothelial cells in these tumors.

View this table:
TABLE 3.

The immunostaining of IQGAP1 in the NMVP-positive group was intense throughout the sections examined (Fig. 3B; Table 3). Sixty-three of 73 (86%) NMVP-positive gliomas showed strong IQGAP1 protein expression, with accentuation beneath the cell membrane (scores ≥3). In contrast to the NMVP-negative gliomas, there was no staining in the endothelial cells of the NMVP-positive gliomas (Fig. 3B, GBM 1). Of note was the IQGAP1 immunostaining of the pseudopalisading cells surrounding the central necrotic focus (Fig. 3B, GBM 2).

The immunostaining of IQGAP1 in the 143 gliomas grouped according to their WHO classification is summarized in Table 5. As predicted from the NMVP-positive grouping scheme, 79% of GBMs (n = 52) were positive for IQGAP1 (score ≥3). Only 15% of the low-grade gliomas, AII (n = 13), and OdgIIs (n = 20) showed IQGAP1 protein expression. Large variability in IQGAP1 immunostaining was evident within each of the grade III groupings, AIII, OdgIII, and OAIII glioma types. IQGAP1 immunostaining was observed in 20% of AIII (n = 20), 32% of OdgIII (n = 26), and 50% of OAIII (n = 12). Representative images of the negative and positive IQGAP1 immunostaining of pure oligodendrogliomas (OdgIII) and mixed gliomas (OAIII) are shown in Figure 3A and B.

Expression and Localization of IGFBP2 in Normal Brain and Gliomas

No immunostaining of IGFBP2 was observed in normal brain (glia, endothelium, or red blood cells) or in 43 of 70 (61%) NMVP-negative gliomas (Fig. 3C; Table 4). In 16 (23%) of NMVP-negative gliomas, minor immunostaining represented by faint haloes of accentuated cytoplasmic reactivity surrounding the nuclei was observed in <5% of the tumor sections studied. Strong cytoplasmic immunostaining of IGFBP2 was observed in 64 of 73 (88%) NMVP-positive gliomas tested; however, the distribution was patchy throughout the tumor sample (Fig. 3D; Table 4). In some samples, IGFBP2 immunostaining was visible in <50% of the glioma sections examined. Strongly positive areas were often located close to the necrotic regions and staining was especially prominent in pseudopalisading cells (Fig. 3D, GBM 2).

View this table:
TABLE 4.

Immunostaining of IGFBP2 in the 143 gliomas (WHO classification) is summarized in Table 5. Positive IGFBP2 protein expression was observed in 92% of GBMs (score ≥2). Only 3 of a total of 13 AII and 3 of 20 OdgII low-grade gliomas showed positive IGFBP2 immunostaining. Large variability in the IGFBP2 immunostaining was again evident within each of the grade III groupings, AIII, OdgIII, and OAIII. IGFBP2 immunostaining was noted in 15% of AIII (n = 20), 32% of OdgIII (n = 26), and 58% of OAIII (n = 12). Representative images of the negative and positive IGFBP2 staining observed in the pure oligodendrogliomas (OdgIII) and mixed gliomas (OAIII) are shown in Figure 3C and D.

View this table:
TABLE 5.

Survival Analysis

Univariate analysis (logrank test) of the prognostic significance of the variables, age, pure oligodendroglioma (OdgII and OdgIII) or astrocytic component (AII, AIII, OAIII and GBM), and IQGAP1 and IGFBP2 protein expression scores, showed that all variables were highly significantly related to survival time. Kaplan-Meier curves illustrating the effect of these features on survival are shown in Figure 4. Patient age older than 60 years and gliomas with an astrocytic element were strongly associated with poorer survival (Fig. 4A, B). Patients with high scores of IQGAP1 (score ≥3) and IGFBP2 (score ≥2) experienced shorter median survival times (median 351 and 303 days, respectively) than patients who scored negative for IQGAP1 (scores ≤2) and IGFBP2 (scores ≤1) (Fig. 4C, D).

FIGURE 4.

Prognostic value of IQGAP1 and IGFBP2 protein expression scores in 143 patients with gliomas. (A) Disease-specific survival of patients with increasing age: 20 to 39 years (n = 36); 40 to 49 years (n = 23); 50 to 59 years (n = 35); and 60+ years (n = 49). Logrank p < 0.001. (B) Disease-specific survival of patients with astrocytic (n = 80) or oligodendroglial tumors (n = 57). Logrank p < 0.001. (C) Disease-specific survival of patients with IQGAP1 protein expression scores: score 0 (n = 7); score 1 (n = 18); score 2 (n = 36); score 3 (n = 17); and score 4 (n = 65). Logrank p < 0.001. (D) Disease-specific survival of patients with IGFBP2 protein expression scores: score 0 (n = 50); score 1 (n = 18); score 2 (n = 21); and score 3 (n = 54). Logrank p < 0.001.

A Cox proportional hazards regression model incorporating age, IQGAP1 protein expression (≥3), and IGFBP2 protein expression (≥2) showed that the best predictors of poor survival were increased age and IGFBP2 protein expression (Table 6). There was evidence that IQGAP1 protein expression was also predictive of poorer survival, but this did not reach significance (p = 0.095). Compared to the reference age group of 20 to 39 years, patients aged 60+ years had a 3.5 times greater risk of death. Patients with high protein expression of IGFBP2 had a 1.7 times greater risk of death.

View this table:
TABLE 6.

The survival curves shown in Figure 5 illustrate the impact of the IQGAP1 and IGFBP2 protein expression markers on patient outcome when the 143 gliomas were grouped according to their WHO classification. Because of low numbers of AII and OdgII (Table 5), comparisons of these low-grade gliomas were not performed. The median survival for patients diagnosed with a GBM did not differ significantly in regard to whether the sample stained positive (IQGAP1+ve) or negative with IQGAP1 (IQGAP1-ve) (Fig. 5A). However, at 3 years, 22% of GBM/IQGAP1-negative patients in our study were alive compared with <10% of GBM/IQGAP1+ve patients. There were no AIII/IQGAP+ve patients alive after 6 years compared with 90% survival in AIII/IQGAP1-ve patients, after 8 years (Fig. 5A). Significantly shorter survival was observed in OdgIII/IQGAP1+ve patients. Only 20% OdgIII/IQGAP1+ve patients were still alive at 5 years compared with 70% of OdgIII/IQGAP1-ve patients (Fig. 5B). In patients who were OAIII/IQGAP1+ve, 25% were still alive at 2 years compared with 70% survival in patients with OAIII/IQGAP1-ve gliomas (Fig. 5B).

FIGURE 5.

Prognostic value of IQGAP1 and IGFBP2 protein expression when applied as an adjunct to the histologically classified glioma patients. (A) Disease-specific survival of patients diagnosed with a glioblastoma multiforme (GBM) or astrocytoma grade III (AIII) with positive (score ≥3) or negative (score ≤2) IQGAP1 protein expression (IQGAP1+ve or IQGAP1-ve, respectively). GBM/IQGAP1+ve (n = 41) and GBM/IQGAP1-ve (n = 11). Logrank p = 0.550. AIII/IQGAP1+ve (n = 4) and AIII/IQGAP1-ve (n = 16). Logrank p = 0.032. (B) Disease-specific survival of patients diagnosed with oligodendroglioma grade III (OdgIII) or oligoastrocytoma grade III (OAIII) with IQGAP1+ve (score ≥3) or IQGAP1-ve (score ≤2) protein expression. OdgIII/IQGAP+ve (n = 9) and OdgIII/IQGAP-ve (n = 17). Logrank p = 0.038. OAIII/IQGAP+ve (n = 6) and OAIII/IQGAP1-ve (n = 6). Logrank p = 0.336 (C) Disease-specific survival of patients diagnosed with a GBM or AIII with IGFBP2+ve (score ≥2) or IGFBP2-ve (score ≤1) protein expression. GBM/IGFBP2+ve (n = 48) and GBM/IGFBP2-ve (n = 4). Logrank p = 0.518. AIII/IGFBP2+ve (n = 3) and AIII/IGFBP2-ve (n = 17). Logrank p = 0.022 (D) Disease-specific survival of patients with OdgIII or OAIII with IGFBP2+ve (score ≥2) or IGFBP2-ve (score ≤1) protein expression. OdgIII/IGFBP2+ve (n = 9) and OdgIII/IGFBP2-ve (n = 17). Logrank p = 0.019. OAIII/IGFBP2+ve (n = 7) and OAIII/IGFBP2-ve (n = 5). Logrank p = 0.086.

The median survivals of patients with AIII/IQGAP+ve gliomas (1,919 days) (logrank p < 0.001) and OdgIII/IQGAP+ve gliomas (427 days) (logrank p = 0.081) were better than those for GBM (315 days). The median survival of patients with OAIII/IQGAP+ve gliomas was 242 days (logrank p = 0.890) (Fig. 5).

The median survival for patients with GBM who were positive or negative for IGFBP2 protein expression (IGFBP2+ve or IGFBP2-ve, respectively) did not significantly differ (182 days with IGFBP2 expression compared to 186 with no IGFBP2 expression) (Fig. 5C). In a small number of GBM/IGFBP2-ve patients (n = 4), longer survival of >3 years was observed. In patients who were AIII/IGFBP2+ve, 70% were alive after 5 years, compared with 95% of patients who were AIII/IGFBP2-ve (Fig. 5C). Significantly poorer survival was observed in the OdgIII/IGFBP2+ve gliomas. Approximately 30% of OdgIII/IGFBP2+ve patients were alive after 3 years compared with 75% survival in patients with OdgIII/IGFBP2-ve gliomas (Fig. 5D). No patients with OAIII/IGFBP2+ve gliomas were alive compared with 80% survival in patients with OAIII/IGFBP2-ve gliomas after 2 years (Fig. 5D).

The median survivals of patients with AIII/IGFBP2+ve gliomas (1,919 days) (logrank p = 0.018), OdgIII/IGFBP2+ve gliomas (427 days) (logrank p = 0.218), and OAIII/IGFBP2+ve gliomas (333 days) (log rank p = 0.890) were better than those for patients with GBM (315 days) (Fig. 5).

Discussion

Developing reliable prognostic markers for patients with high-grade gliomas has been challenging, despite overwhelming evidence that wide variability exists in the clinical behavior, and thus in survival times, of morphologically identical tumors. We searched for biomarkers that could be used as an adjunct to the WHO 2000 grading system to improve prognostic accuracy in glioma. The overexpression of IQGAP1 and IGFBP2, at both the gene and protein levels, was highly correlated with gliomas exhibiting necrosis and MVP. These markers were generally associated with significantly shorter adjusted median survival. When either of one of these markers was used in conjunction with the WHO system, both IQGAP1 and IGFBP2 readily identified a subgroup of patients with AIII, OdgIII, and OAIII whose prognosis was relatively poor. These results suggest that IQGAP1 and IGFBP2 are clinically useful prognostic markers and provide additional information to the WHO system.

The multivariate algorithms, GeneRave and SDDA, are designed to find small sets of expressed genes that are biologically meaningful and can act as strong discriminators between 2 sample groups. This is the first study to apply these algorithms to a brain tumor sample set. Broadly separating the gliomas into 2 groups using necrosis and MVP allowed us sufficient power to perform multivariate analysis. Thirteen genes with known biologic function were identified by GeneRave and SDDA. The 2 most significantly overexpressed genes associated with necrosis and MVP were IQGAP1 and IGFBP2. We chose to validate these 2 genes in a large, semi-independent glioma sample set.

Protein expression of IQGAP1 and IGFBP2 was observed in >80 percent of NMVP-positive gliomas and was associated with shorter survival. The application of these markers to the grade III tumors (AIII, OdgIII, and OAIII) clearly identified a subset of IQGAP1/IGFBP2-positive patients who had significantly poorer survival outcome. Although AIII tumors separated into distinct survival groups, overall median survival for AIII with IQGAP1 and IGFBP2 protein expression was 5 years. There was no significant difference between the median survival of patients with OAIII gliomas with IQGAP1 and IGFBP2 expression those diagnosed with GBM. The presence of an astrocytic element in gliomas was associated with significantly shorter survival (Fig. 4B). However, this finding most likely implies that the astrocytic component of the oligoastrocytomas in our data set could be GBMs. IQGAP1 and IGFBP2 protein expression could be used to better identify oligoastrocytomas that behave similarly to GBMs.

Although not statistically significant, the shorter survival observed with OdgIIIs was still better than that for GBMs (Fig. 5). LOH of chromosomes 1p and 19q, most commonly detected in pure and mixed oligodendroglial tumors, has been associated with better survival (3,4). The loss of 1p and 19q was not a frequent occurrence in our sample set. Of the 26 patients with OdgIIIs in our study, 8 gliomas with 1p/19q loss were identified. These gliomas stained both positive (n = 3) and negative (n = 5) for IQGAP1 and IGFBP2, respectively. The 3 gliomas identified with 1p/19q loss, coupled with IQGAP1/IGFBP2 protein expression, did not show any survival advantage over the other OdgIII gliomas. For the OAIII dataset (n = 12), LOH data were available for 50% of samples, and none of these showed 1p19q loss. Because of the small number of samples identified with 1p/19q loss, the power to detect any survival benefit in the oligodendrogliomas and mixed gliomas from this deletion is too low, and its relative impact on the prognostic markers IQGAP1 and IGFBP2 cannot be concluded.

As expected, almost 80% of GBMs were IQGAP1-positive and >90% of GBMs were IGFBP2-positive. However, patients with a GBM who were negative for IQGAP1 and IGFBP2 showed an improved survival outcome. This result is clearly demonstrated in Figure 5A and C. Despite the extensive documentation of long-term survival in small populations of GBM patients, no current histologic parameters have been elucidated that could identify such a group. From our dataset of 52 GBMs, 3 patients were identified with a recorded survival of >3 years. In 2 of these patients, no IQGAP1 or IGFBP2 protein expression was detected. One of these patients is still alive after 5 years. The absence of IQGAP1 and IGFBP2 in gliomas may prove to be prognostically useful, warranting a study of these proteins in a larger sample set of long-term survival patients. This study clearly demonstrates utility for IQGAP1 and IGFBP2 to supplement the more traditional diagnostic markers to offer additional prognostic and predictive information.

IQGAP1 expression has been univariately associated with poor prognosis in other cancers (20-23); however, the association of IQGAP1 with poor survival in gliomas has not previously been reported. Our immunohistochemical analysis revealed strong positive cytoplasmic staining for IQGAP1 in NMVP-positive gliomas; however, in contrast, IQGAP1 expression was specific only to the endothelial cell structures in the NMVP-negative gliomas (astrocytic and oligodendroglial). This observation is consistent with a recent report describing IQGAP1 expression in rat brain and human glioma samples (24). The authors suggested that IQGAP1 protein expression was restricted to gliomas of astrocytic origin (24). However, our study demonstrates that there is a subgroup of pure oligodendrogliomas that express IQGAP1 and behave poorly.

IQGAP1 is a scaffolding protein that has a multifunctional role in normal tissue (25). It has been shown to be a target molecule of Cdc42 and Rac1 small GTPases and negatively regulates E-cadherin-based cell-cell adhesion (20). Abrogation of cell-cell adhesion is a key event in the invasive phenotype of many cancers, and the cadherin superfamily of adhesion molecules (E-, P-, and N-cadherin) has been associated with glioma invasion (26). In addition, IQGAP1 plays a role in cellular motility and morphogenesis by interacting directly with cytoskeletal, cell adhesion, and signal transduction proteins (25,27-32). Overexpression of IQGAP1 in the breast cancer cell line MCF-7 results in significant increases in cell invasive capacity, whereas downregulation of IQGAP1 in ovarian cancer cells by IQGAP1-specific small interfering RNAs leads to a loss of migratory ability in these cells (33). It seems likely that IQGAP1 may also play a significant role in glioma migration and could be involved in the rapid dissemination of glioma cells throughout the brain. There is an increasing body of evidence from mouse models supporting glioma initiation as a result of neural progenitor cell transformation (34). In a recent study, neoplastic IQGAP1-positive cells were isolated from rat glioblastoma and subsequently expanded in culture (24). These IQGAP1-positive cells possessed cancer stem-like progenitor cell characteristics and were highly aggressive. A better understanding of the precise mechanisms of IQGAP1 and its interaction with other pathways is needed before pharmacologic interventions could be developed.

IGFBP2 protein expression has been shown to be a key signature marker for GBM, and there have been numerous studies linking IGFBP2 with poor prognosis (35-41). However, there have been no reports demonstrating the use of IGFBP2 protein expression as a marker of aggressive biologic behavior in WHO grade III gliomas. In addition, this is the first report suggesting a better survival outcome for patients with GBMs who do not express IGFBP2.

In our study we found the distribution of IGFBP2 immunostaining to be very patchy and associated with the pseudopalisading cells surrounding the necrotic foci and in areas where cellular degeneration was evident in both astrocytic and oligodendroglial tumors. The localization of IGFBP2 immunostaining to the pseudopalisading cells has previously been reported in whole tissue sections (8). Laser capture microscopy has been used to isolate pseudopalisading cells from GBMs and demonstrated upregulation of gene transcripts involved in glycolysis and cell-cycle control in the pseudopalisading cells (40). A role in angiogenesis has also been suggested for pseudopalisading cells because of the high expression levels of vascular endothelial growth factor as a result of increased transcriptional levels of HIF1-α (42). In addition, the chemokine receptor, CXCR4, has been found to consistently colocalize with HIF1-α expression in pseudopalisading glioma cells surrounding areas of necrosis (43).

Overexpression of IGFBP2 is typically observed in advanced stages of cancer, and it seems plausible that its heightened expression is related to the increasing abundance of necrosis. An immunoprecipitation study showed that IGFBP2 binds to integrin α5, suggesting that IGFBP2 functions to enhance elevated migration rates via an integrin-mediated pathway (44). Tissue RNA levels of integrin α5 have been reported to be significantly higher in hypoxic conditions than under normoxic conditions (45). It will be interesting to determine whether IGFBP2 expression is intricately involved in the dynamic biologic behavior of the pseudopalisading cells or whether its observed expression is merely a consequence of the hypoxic environment.

In conclusion, we have shown that the protein expression of IQGAP1 and IGFBP2 is strongly associated with poor survival in astrocytoma and oligodendroglioma. Although the association of IGFBP2 with poor prognosis has previously been described, the association of IQGAP1 is a novel finding. The findings of our study are significant as IQGAP1 and IGFBP2 protein expression markers could complement the WHO classification system to permit more precise delineation of the existing tumor grades and identify potentially more aggressive glioma subtypes. Importantly, these markers were able to identify a subset of GBM patients who had shown long-term survival of >3 years after initial diagnosis. At present, there are no markers with the capacity to predict long-term survival in such patients. Another clinically useful role of these markers would be to improve accuracy in the prediction of biologic grade and aggressive potential, particularly in small biopsies of gliomas that are surgically inoperable. Functional studies of IQGAP1 and IGFBP2 and their relationship with each other and the relative roles played in glioma biology may lead to a better understanding and may provide potential targets for antitumor therapy.

Acknowledgments

We express our gratitude to Dr. Robert Markham, Obstetrics and Gynecology, Faculty of Medicine, University of Sydney, for his kind assistance and the laboratory space provided to us for the immunohistochemistry analysis. We also thank Dr. Marinella Messina and Dr. Diana Benn for critically reviewing the manuscript and providing suggestions.

Footnotes

  • This work was supported by the Sydney Neuro-oncology Group and the Andrew Olle Memorial Trust.

References

  1. 1.
  2. 2.
  3. 3.
  4. 4.
  5. 5.
  6. 6.
  7. 7.
  8. 8.
  9. 9.
  10. 10.
  11. 11.
  12. 12.
  13. 13.
  14. 14.
  15. 15.
  16. 16.
  17. 17.
  18. 18.
  19. 19.
  20. 20.
  21. 21.
  22. 22.
  23. 23.
  24. 24.
  25. 25.
  26. 26.
  27. 27.
  28. 28.
  29. 29.
  30. 30.
  31. 31.
  32. 32.
  33. 33.
  34. 34.
  35. 35.
  36. 36.
  37. 37.
  38. 38.
  39. 39.
  40. 40.
  41. 41.
  42. 42.
  43. 43.
  44. 44.
  45. 45.
View Abstract