The potential biomarkers for meningiomas identified using microarray gene expression approaches to date differ significantly from report to report(6
). This likely reflects the relatively small numbers of tumors studied, differences in bioinformatics strategies, and other technical artifacts that are difficult to control when many transcripts are assessed at the same time. To minimize the risk of false positive discovery, a number of strategies are typically employed, including statistical tests such as multiple test correction, validation by use of several different platforms (quantitative PCR), and the inclusion of multiple independent tumor cohorts.
In the current study, we applied all of these strategies, filtering genes across 2 independent sets of gene expression microarrays from 2 separate institutions, validating common genes by RT-qPCR, and confirming our findings using a third set of meningioma microarray gene expression data. This stringent approach generated smaller lists of candidate genes that are more likely to be biologically relevant. At the same time we recognize that other biologically relevant markers might have been discarded in the process. By requiring each candidate to be identified in every cohort examined, we almost certainly have missed some of the genes that may only be implicated in smaller subsets of malignant meningiomas. For example, we eliminated a previously identified gene (NDRG2
) reported by our group in clinically aggressive meningiomas (13
). Nonetheless, NDRG2
likely is biologically important because it 1) was previously validated at the protein level, 2) is linked to a cytogenetically relevant site of deletion (chromosome 14q), and 3) was found to have its second allele frequently inactivated by promoter region methylation. Similarly, in the current study, we also identified genes in only 1 cohort that could not be validated in a second cohort, thereby further underscoring the need for multiple independent validation sets. These validation “failures” likely result from the small number of samples in each cohort and the required use of different q value cut-offs when comparing across independently generated microarray data sets. In this regard, the smaller the number of samples in any group, the larger the change in differential gene expression is required to reach statistical significance. Despite these problems inherent in the high stringency filtering approach, it was a useful strategy for identifying the most common and reproducible progression-associated genetic alterations.
The purpose of this study was to identify proteins whose differential expression correlated with aggressive meningioma behavior and poor patient outcome. It was not our intention to replace well-validated markers (e.g. Ki-67) in the routine evaluation of meningiomas, but to demonstrate that the use of an algorithm over multiple independent sets of meningiomas facilitates the identification of other protein biomarkers useful for stratifying meningiomas in clinically relevant subgroups. Ki-67 has proven to be an excellent biomarker for predicting clinical behavior in univariate analyses but it has some limitations. One of us (A.P.) specifically examined this question in a large set of 425 meningiomas using univariate analysis. In that study, Ki-67 was predictive of recurrence-free survival whereas on multivariate analysis after malignancy grade and mitotic index were taken into account, Ki-67 was not independently predictive (5
). Similarly, in the present study Ki-67 exhibited the highest p value in the grade II meningiomas on Kaplan Meier survival curves (p = 0.5754, p = 0.2554; TTD, TTR, respectively). Thus, although Ki-67 is strongly correlated with tumor grade in general, it does not distinguish between clinically aggressive and indolent atypical meningiomas.
Of the 11 genes identified in our study, 4 of them (BIRC5, TOP2a, TIMP3 and KCNMA1
) were previously implicated in meningioma tumor progression, thereby strengthening their association with aggressive meningioma behavior. BIRC5 (survivin) is an anti-apoptotic protein that is highly expressed in meningiomas (18
is located on 17q, a known region of chromosomal gain in grade III meningiomas. Another gene located on 17q is TOP2A
. TOP2A controls DNA folding states during transcription and 2 previous studies have shown that TOP2A
expression is an independent predictor of meningioma recurrence (19
). A third study did not find such an association in benign and atypical meningiomas (26
). Our finding that TOP2A protein expression was associated with poor overall survival in grade II meningiomas underscores the utility of this specific biomarker. TIMP3
is located at 22q12.3, the same chromosomal arm as NF2
(22q12.2). This reg ion is commonly lost in all grades of meningiomas. TIMP3 is a member of a family of proteins that regulate cell motility and invasion by regulating metalloproteinase function. We found that TIMP3
was underexpressed in grade III meningiomas and has been associated with poor survival in both esophageal squamous cell carcinoma (27
)and meningioma (20
). In addition, 2 recent reports found reduced TIMP3 expression in tumors with more complex karyotypes (8
) or clinically aggressive behavior (9
). In the latter study, BIRC5 expression was also found to be increased in grade III meningiomas. Finally, we also identified KCNMA1
as a potential meningioma biomarker. Although we did not analyze KCNMA1
expression at the protein level, expression of this potassium large-conductance calcium-activated channel protein was also reduced in high-grade meningiomas in previous studies (8
Other genes identified in this study have not previously been linked to meningioma: RRM2
overexpression has been linked to angiogenesis, tumor invasion, cell proliferation, and drug resistance in a variety of different cancers (28
, a cyclin-dependent kinase, is important for regulating cell proliferation (34
) and TPX2
, a microtubule-associated protein, is important in normal mitotic spindle formation (31
). Although its role in meningioma tumor progression has not been explored to date, TPX2
overexpression is associated with increased proliferation in malignant salivary gland tumors and with decreased 5 y survival rates in patients with squamous cell carcinoma of the lung (32
, also known as elafin, is an elastase-specific inhibitor with anti-microbial, anti-inflammatory, and immunomodulatory functions (34
); high PI3
levels have recently been correlated with poor survival in glioblastoma (35
overexpression results in increased cell growth and its expression is upregulated in liver and colon carcinomas (36
). Increased NUSAP1
expression is associated with melanoma progression (37
, a sex-determining region-box protein, is a transcription factor that functions in the developing nervous system and has been reported to be overexpressed in medulloblastoma and malignant glioma (38
Using IHC on select grade I and grade III meningiomas from the original “driver set”, we confirmed the predicted differential protein expression of TOP2A, TIMP3, and BIRC5. In particular, our data suggest that TOP2A may be a useful ancillary marker since there is a commercial antibody to it. Thus, our data suggest that IHC may help distinguish clinically aggressive from more indolent WHO grade II meningiomas.
One of the unique features of the approach we employed is the application of a filtering algorithm using a second independently generated microarray dataset to reduce the numbers of false positives. This resulted in the identification of a robust set of 11 genes that were validated by RT-qPCR on a third set of meningiomas and by microarray analysis on a fourth set of tumors. This approach increases the predictive value of the candidate genetic biomarkers discovered. Future studies employing larger and prospectively collected cohorts of tumors will be required to determine whether these differentially expressed genes have prognostic value in the management of meningioma.