Prognostic impact of molecular markers in a series of 220 primary glioblastomas
In contrast to oligodendrogliomas, molecular predictors of prognosis have not been consistently found in glioblastomas. However, genetic studies show that glioblastomas consist of several genetic subtypes and raise the possibility that molecular alterations could be predictive of survival.
A search for loss of heterozygosity (LOH) on chromosome 1p, 9p, 10q, 19q, EGFR (epidermal growth factor receptor), CDK4, and MDM2 (mouse double minute) amplifications, CDKN2A (INK4A/ARF) homozygous deletions, p53 expression, was performed in a series of 220 primary glioblastomas. The molecular alterations were then correlated with each other to identify distinct molecular pathways and with clinical parameters and the course of the disease to identify prognostic markers.
Nonrandom associations were found between EGFR amplification and LOH10q, LOH9p, and INK4A/ARF deletion, LOH1p and LOH19q, and MDM2 and CDK4 amplification, whereas mutual exclusions were found between p53 expression and EGFR amplification, LOH 9p/INK4A/ARF homozygous deletion, and MDM2 and CDK4 amplification. Age (P = 4.10−5) and performance status (P = .003) were the main predictors of outcome. In contrast, molecular markers were of limited impact: MDM2 amplification correlated with poor outcome on both univariate and multivariate analysis (P = .01) and EGFR amplification with good prognosis on multivariate analysis (P = .02).
Despite their limited prognostic impact, the genetic markers investigated here outline distinct molecular pathways involved in glioblastoma tumorigenesis and warrant broader molecular screening. Cancer 2006. © 2006 American Cancer Society.
Glioblastoma (GBM) is the most common primary brain tumor in adults and also the most malignant glioma. With a median survival ranging from 12–15 months, it is one of the most aggressive human neoplasms. The discovery over the last 15 years of some of the main genetic alterations associated with the development of gliomas has made a major contribution to understanding the molecular pathways involved in glial oncogenesis. One of the consequences of this progress is that the genetic profile may partially predict prognosis and response to treatment, as illustrated by 1p and 19q chromosome deletion in low-grade and anaplastic oligodendrogliomas.1, 2 In GBM such markers are still lacking. To date, the analysis of the main molecular alterations, including p53 mutations and expression, epidermal growth factor receptor (EGFR) amplification, INK4A/ARF deletion, and loss of chromosome 10q, have failed to identify markers useful for clinical practice.3–7
In this study the most commonly reported molecular alterations were analyzed in a large series of primary GBM in order to refine knowledge of the molecular “signature” of these tumors, identify age-related subgroups, and detect molecular predictors of survival that could usefully complete existing prognostic factors.8
MATERIALS AND METHODS
Selection of Patients: Inclusion and Exclusion Criteria
From January 1997 to March 2005, clinical information on patients treated in our department for a primary brain tumor was collected in a database. Patients fulfilling the following inclusion criteria were selected for this study: 1) age 18 years or more at onset; 2) histologic diagnosis of GBM according to the World Health Organization classification; 3) detailed clinical data, at diagnosis and during follow-up; and 4) availability of paired blood and fresh-frozen tumor samples, obtained after informed consent, for molecular analysis. In order to limit this study to primary GBM, patients previously treated for a low-grade glioma (secondary GBM) were excluded.
Loss of heterozygosity (LOH) on chromosomes 1p, 19q, 10q, and 9p was detected by microsatellite analysis of blood and tumor DNA, as previously reported.9
P53 expression was detected on 5-μm sections of formalin-fixed and paraffin-embedded tissues.10 Expression of p53 was defined as a moderate to strong (++ to +++) staining of more than 50% of nuclei.
EGFR amplification of tumor DNA was detected by real-time polymerase chain reaction (PCR) with an internal probe. The EGFR primers (forward = GTGCAGATCGCAAAGGTAATCAG; reverse = GCAGACCGCATGTGAGGAT; probe = CCCCTCCCCGTATCTC [FAM Dye Labeled]) amplified a 79-basepair (bp) genomic fragment. The reference primers amplified a genomic fragment from RNase P (TaqMan RNase P Detection Reagents [FAM Dye] Ref: 4316831; Applied Biosystems, Foster City, CA). Real-time PCR cycles were carried out as follows: 50°C for 2 minutes; 95°C for 15 minutes; cycles 1–40 at 95°C for 15 seconds; 60°C for 1 minute. The same technique was used for MDM2 (mouse double minute) with the following primers: forward = TTGGTTTCTAGACCATCTACCTCATCT; reverse = AAAAGCTGTGTGAATGCGTCAAAT; probe = ACCTGTCTCACTAATTGC (FAM Dye Labeled), which amplify a 91-bp genomic fragment, and for CDK4: forward = GCCACTAAGCAGTAACCATTCAACT; reverse = CGGCTTCAGAGTTTCCACAGAA probe = CCTTCTCACCTTAGGCC (FAM Dye Labeled), which amplify a 102-bp genomic fragment.
Homozygous deletion of INK4A/ARF was detected by real-time PCR as previously described.11
Frequency distributions and summary statistics were calculated for all clinical and molecular variables for the entire population. All variables were coded as binary factors, except surgery, which was coded as an unordered categorical variable (biopsy, partial resection, or complete resection). The chi-square test and Fisher exact test for small samples were used to test the association between molecular alterations. Overall survival (OS) was used to study the prognostic impact of the variables analyzed. OS was defined as the time from the date of first surgery until death or last follow-up. Patients lost to follow-up were censored on the last known day of life and patients still alive on the day of the analysis were censored on March 31, 2005. Probability estimates for OS were calculated by the Kaplan-Meier estimator. The log-rank test was used to test for equality of survival distributions. The following variables were investigated: age, sex, Karnofsky performance status (KPS), symptoms at diagnosis, extent of resection, LOH on chromosome 1p, 19q, 10q, and 9p, p53 expression, EGFR, CDK4, and MDM2 amplifications, and INK4A/ARF homozygous deletion. The distribution of molecular markers was investigated by the Wilcoxon test. The following factors were entered as candidate variables in the multivariate Cox proportional hazard regression model analysis: 1) factors that were significant in univariate analysis (P ≤ .05); 2) and factors that did not reach significance on univariate analysis, but were known to be age-related, like EGFR amplification and 10q loss. Multivariate analysis was performed in the stepwise manner. Two-sided P < .05 was considered significant.
In all, 220 of the 642 GBM patients entered in the database fulfilled the inclusion criteria. The most restrictive factor was availability of paired blood and tumor samples for molecular analysis. Median age at surgery was 56 years (range, 22–83 years), the sex ratio (men/women) was 1.8, and the median postoperative KPS was 80 (range, 30–100). Postoperative treatment consisted of radiotherapy with or without adjuvant chemotherapy for 187 patients (85%), whereas the remaining patients (n = 33, 15%) received chemotherapy alone or best supportive care. Median follow-up was 56 months and 182 (83%) patients were dead at the time of the study. Overall median survival was 16 months (95% confidence interval [CI]: 451–527 days) with a 2-year survival rate of 25%. Only 2 patients were still alive 5 years after diagnosis.
Frequency of molecular alterations and correlations between them.
The most frequent alterations, in order of decreasing frequency, were LOH on 10q (155 of 206, 75%), LOH on 9p (96 of 205, 47%), p53 expression (73 of 199, 37%), INK4A/ARF deletion (68 of 188, 36%), EGFR amplification (72 of 210, 34%), LOH on 19q (60 of 209, 29%), LOH on 1p (40 of 210, 19%), CDK4 amplification (11 of 167, 6.6%), and MDM2 amplification (11 of 177, 6.2%). The associations and exclusions between the molecular alterations are shown in Table 1.
Table 1. Correlations between Molecular Alterations (Chi-Square Test)
|LOH 1p||AssocP = .005||NS||NS||NS||NS||NS||NS||NS|
|LOH 19q|| ||NS||NS||NS||NS||NS||NS||NS|
|LOH 9p|| || ||AssocP = .0001||AssocP = .03||NS||AssocP = .01||ExclusionP = .04||ExclusionP = .02|
|LOH 10q|| || || ||AssocP<0.0001||NS||AssocP = .0003||NS||NS|
|EGFR amplif|| || || || ||ExclusionP = .0008||AssocP = .002||NS||NS|
|p53 expression|| || || || || ||NS||NS||NS|
|INK4A/ARF deletion|| || || || || || ||NS||ExclusionP = .05|
|MDM2 amplif|| || || || || || || ||AssocP = .0002|
Correlations between molecular alterations and clinical characteristics of the patients.
The only significant link was a correlation between LOH10q. LOH10q was more frequent in older patients (P<.04; Wilcoxon test). The same trend was observed for EGFR amplification, but the difference did not reach significance.
Factors Predictive of Survival
In univariate analysis (Table 2), increased survival was associated with younger age (≤56 years vs. >56 years; P<.0001), higher KPS (>80 vs. ≤80; P = .003), complete resection of the tumor (vs. biopsy or partial resection; P = .01), and the absence of a neurologic deficit at diagnosis (vs. deficit at diagnosis; P = .02). Among molecular factors, MDM2 amplification was the only predictor of poor outcome (log rank test; P = .01). No correlation was found with survival for the other molecular alterations (LOH 1p and 19q; LOH 9p; LOH 10q; EGFR amplification; CDK4 amplification, INK4A/ARF deletion) analyzed either alone or by groups of alterations.
Table 2. Correlations between Clinical or Molecular Factors and Survival in Primary GBM: Univariate Analysis
|KPS >80||85||611|| |
|Age >56||110||407|| |
|Partial resection||55||472|| |
|Total resection||127||512|| |
|No LOH 1p||170||476||.76|
|LOH 1p||40||511|| |
|No LOH 19q||149||456||.2|
|LOH 19q||60||527|| |
|No LOH 9p||109||476||.54|
|LOH 9p||96||478|| |
|No LOH 10q||51||465||.68|
|LOH 10q||155||496|| |
|P53 expression||73||440|| |
|EGFR amplification||72||504|| |
|No INK4A/ARF deletion||120||473||.33|
|INK4A/ARF deletion||68||582|| |
|MDM2 amplification||11||444|| |
|CDK4 amplification||11||472|| |
In a multivariate analysis including all parameters previously identified as significant, age ≤56 years (P = .002; relative risk [RR] = 1.71), KPS >80 (P = .02; RR = 0.64), larger type of resection (P = .002. RR = 0.64), and MDM2 amplification (P = .05; RR = 2) were factors independent of outcome. Because EGFR amplification is known to be associated with older age12 it was also tested on multivariate analysis and found to be correlated with better outcome (P = .03; RR = 0.69).
The aim of this study was to refine the analysis of molecular signatures of GBM and to correlate them with clinical data and outcome. With a median overall survival of 16 months and a 25% rate of survivors at 2 years, our data are comparable to recently reported results.13 In addition, we confirm that young age, good performance status, and macroscopically complete resections are strongly predictive of better outcome.8 Table 1 clearly shows nonrandom associations and exclusions of molecular changes.
The strong association between EGFR amplification and LOH10q, LOH9p, and INK4A/ARF deletions characterizes primary GBM.3, 14, 15EGFR amplification and p53 expression were present in one-third of the cases and were mutually exclusive. Because p53 expression/mutation in the absence of EGFR amplification is believed to characterize secondary GBM,16, 17 it is possible that some patients with clinically silent lower-grade tumor become symptomatic only after progressing to GBM. The deletion of CDKN2A (INK4A/ARF) locus is a frequent alteration. It inactivates not only the P16/INK4A but also the P14/ARF gene, encoded by an alternative reading frame of CDKN2A,18 resulting in disruption of both RB1/CDK4/P16 and P53/MDM2/P14 pathways. Both MDM2 amplification and INK4A/ARF deletion were mutually exclusive. MDM2 facilitates ubiquitin-mediated degradation of p53, which is inhibited by P14/ARF.19 The fact that both MDM2 amplification and INK4A/ARF/P14 deletion-LOH9p induce the functional inactivation of p53 and are therefore functionally redundant probably explains why we found both alterations mutually exclusive. We found also MDM2 and CDK4 amplifications tightly associated, suggesting that both genes, which map close together to the same region on 12q13, are frequently included in the same amplicon.20 It is interesting to note that CDK4 amplification and INK4A/ARF deletion were also mutually exclusive. CDK4 promotes the phosphorylation of the Rb protein encoded by RB1 gene on chromosome 13q14, and therefore passage through phase S. CDK4 is inhibited by p16 protein, product of the CDKN2A (INK4A) tumor suppressor gene. Again, both alterations result in a redundant effect.21
Contrasting with the heavy prognostic weight of clinical factors, the genetic factors investigated here appear of little prognostic value. Only MDM2 and EGFR had an impact on survival. MDM2 amplification, although infrequent, is shown here to be predictive of poorer outcome on both univariate and multivariate analysis, confirming previous studies,22, 23 whereas no prognostic value was found by others.24, 25
It is quite intriguing to note that one of the most characteristic molecular markers of a primary GBM phenotype, EGFR amplification, when adjusted for age, is a “good” prognosis factor. Indeed, multivariate analysis suggests that EGFR amplification—although it tends to be more frequent in older patients (which have the poorest prognosis)—is by itself indicative of longer survival. This finding is consistent with previous work suggesting that EGFR amplification is a predictor of increased survival in older patients,4, 26, 27 although it is still debatable.5, 6, 28, 29 The negative consequences of older age in GBM do not reflect a general rule of oncology: in other cancers, tumor aggressiveness is higher in young people, or age has little prognostic importance.30–32 A speculative explanation could be the presence of as-yet unidentified molecular alterations or combination of alterations more frequent in older patients. One of these molecular alterations could be the secreted protein YKL40, which has been recently identified and correlated with both poor outcome and older age on multivariate analysis.33
The association between deletions on chromosomes 1p and 19q is well established and constitutes a strong predictor of increased progression-free survival, survival, and chemosensitivity in anaplastic oligodendrogliomas.1 However, combined 1p and 19q loss has also been found in some GBM, particularly the so-called “GBM with an oligodendroglial component.”9 In our series, 19% (n = 40) of the patients had 1p+/-19q deletion and 9% (n = 18) had both 1p and 19q deletions, but their survival was not improved, indicating that the prognostic value of 1p+/-19q loss does not hold in GBM despite encouraging preliminary reports.5, 34 Studies concerning the prognostic value of the other molecular markers in GBM also gave rise to conflicting results. P53 status, either by expression or mutation analysis, was correlated with a better outcome5, 35 or had no impact.4, 25 LOH 10q and INK4A/ARF status were associated with poor outcome,3, 5, 36 or had no impact.4, 25, 37 Nevertheless, the strong prognostic importance of clinical factors, mainly age and performance status, was unambiguous in those studies, as it is in ours.
In contrast with the consensus rapidly obtained in the literature for oligodendroglial tumors,38–40 these data suggest that the systematic analysis of primary GBM with these genetic markers is of limited prognostic weight as compared with clinical prognostic factors and warrant further studies, based on a broader screening of potential markers as offered now by microarray approaches.33, 41
The authors thank Dr. Philippe Broet for statistical expertise, Drs. Michele Kujas and Marc Polivka for providing histologic samples, and Anne-Marie Lekieffre, Murielle Brandel, Marc Ouzounian, and Tristan Salmon-Legagneur from the ARTC for assistance.