Expression profiling of ependymomas unravels localization and tumor grade-specific tumorigenesis




Ependymomas derive from ependymal cells that cover the cerebral ventricles and the central canal of the spinal cord. The molecular alterations leading to ependymomal oncogenesis are not completely understood.


The authors performed array-based expression profiling on a series of 34 frozen ependymal tumors with different localizations and histologic grades. Data were analyzed by nonsupervised and supervised clustering methods along with Gene Ontology and Pathway Analyzer tools.


Class discovery experiments indicated a strong correlation between profiles and tumor localization as well as World Health Organization (WHO) tumor grades. On the basis of supervised clustering, intracranial ependymomas were associated with high expression levels of Notch, Hedgehog, and bone morphogenetic protein pathway members. In contrast, most of the homeobox-containing genes manifested high expression in extracranial ependymomas. The results also revealed that WHO grade 2 ependymomas differed from WHO grade 3 ependymomas by genes implicated in Wnt/β-catenin signaling, cell cycle, E2F transcription factor 1 destruction, angiogenesis, apoptosis, remodeling of adherens junctions, and mitotic spindle formation.


Taken together, the tumor localization-related gene sets mainly implicated in stem cell maintenance, renewal, and differentiation suggest the dysregulation of localized cancer stem cells during ependymoma development. The WHO grade differentiating pathways suggested that alteration of the Wnt/β-catenin signaling pathway is a key event in the tumorigenesis of WHO grade 3 ependymomas. On the basis of the current data, the authors suggest a developmental scheme of ependymomas that integrates tumor localization and tumor grades, and that pinpoints new targets for the development of future therapeutic approaches. Cancer 2009. © 2009 American Cancer Society.

Ependymomas are primary tumors of the central nervous system and are derived from cells lining the cerebral ventricles and the central canal of the spinal cord. They account for 3% to 5% of all primary brain malignancies and occur predominantly in young adults and children, in whom they account for approximately 10% of all brain tumors. Four major histologic subtypes and 3 different World Health Organization (WHO) grades are recognized; myxopapillary ependymoma and subependymoma (WHO grade 1), ependymoma (WHO grade 2), and anaplastic ependymoma (WHO grade 3).1

Previous studies demonstrated that ependymal tumors may be heterogeneous and established associations between (epi)genetic alterations and clinicopathologic characteristics. Monosomy 22, loss of chromosomes 6 and 16, and gain of chromosome 7 are associated mostly with extracranial ependymomas,2-6 whereas gain of 1q and losses on 6q, 9, and 13 are observed mainly in intracranial tumors.2, 7, 8 Partial trisomy 19 is linked to a variant of supratentorial ependymomas,9 and gain of 1q is linked to an adverse prognosis in intracranial pediatric tumors.10, 11 Methylation of cyclin-dependent kinase inhibitor 2B (CDKN2B) was observed most frequently in myxopapillary ependymomas, and p14ARF methylation was observed in WHO grade 3 tumors.12

Microarray-based expression studies have correlated molecular signatures with clinicohistologic characteristics. Members of the Notch and Sonic Hedgehog pathways have been related to intracranial ependymomas, and genes of the homeobox-containing (HOX) family have been related to extracranial ependymomas.13-15 A similar pattern of expression was observed by Taylor and coworkers in radial glial fibers, leading to the suggestion that ependymomas derive from regionally specific stem cells bearing a radial glial cell phenotype.15 When only WHO grade 2 tumors were analyzed, intracranial tumors differed from extracranial ependymomas by genes implicated in angiogenesis (fms-related tyrosine kinase 1 [FLT1] and the vasoconstrictor endothelin 1 [EDN1]).16 Posterior fossa ependymomas differed from normal brain tissue by vascular endothelial growth factor (VEGF), Zic family 1 (ZIC1), wingless-type MMTV integration site family member 5A (WNT5A), and genes involved in vesicle trafficking and recycling.17 In the supratentorial compartment, genes implicated in cell cycle (cyclin D1 [CCND1], cyclin-dependent kinase 5 [CDK5], and minichromosome maintenance complex component 7 [MCM7]) were associated with WHO grade 3 ependymomas.13

The identification of ependymoma subgroup-specific alterations has improved our understanding of ependymomal oncogenesis, but characterization of the tumorigenic events still is needed for further stratification of clinical trials. To address this issue, we analyzed microarray-based expression profiles of 34 frozen ependymomas. We unraveled an important effect of tumor localization and WHO grade on expression signatures or profiles, and we also identified multiple pathways that characterized these profiles. We linked dysregulation of developmental pathways to localization and linked the hallmarks of cancer18 to anaplastic ependymomas. On the basis of these results, we suggest a scheme for ependymoma tumorigenesis, highlighting pathways that could be targeted in future therapeutic trials.



Pediatric and/or adult ependymomas from 34 patients were collected from 7 centers. These tumors were snap-frozen in liquid nitrogen and kept at −80°C: These samples included 30 primary tumors, 3 recurrent tumors, and 1 tumor of unknown stage. Frozen sections, which were available for most of the lesions, confirmed that the analyzed samples were constituted predominantly of tumor cells. Age at operation varied from <1 year to 69 years, the mean age was 28.5 years, and the median age was 33 years. Thirteen tumors were located in the posterior fossa, 3 were located in the supratentorial compartment, 14 were located in the spinal cord, 2 were located in the filum terminale, and 2 were of unknown localization. Tumor grading was assessed by 3 different neuropathologists (a local pathologist, C.G., and F.S.). WHO grade 2 ependymomas were tumors that presented with low cellular density and low nuclear chromatin intensity in at least 75% of the tumor surface area, absent or only focally present mitosis (corresponding to a Ki-67 labeling index between <2% and 3%), and the absence or only a focal presence of activated angiogenesis, characterized by plumped or proliferating endothelial cells. In WHO grade 3 ependymomas, at least 25% of the tumor surface had cells with high cellular density and increased nuclear chromatin intensity, and there was focally or diffusely increased mitosis (corresponding to a Ki-67 index >5%). Angiogenesis was activated by plumped and/or proliferating endothelial cells.

Array Hybridization

Total RNA samples were extracted from fresh-frozen tumor specimens using Tripure, as described by the manufacturer (Roche, Mannheim, Germany). RNA quantity and purity were determined by using a NanoDrop ND-1000 spectrophotometer (NanoDrop Technologies, Wilmington, Del). Samples with purity ratios (A260/A280) that were not between 1.70 and 1.83 were discarded from further analyses. Agarose gels were used to appreciate the integrity of the extracted RNA; only samples with thin and clearly distinguishable bands of the 18S and 28S ribosomal RNAs were used for further studies. Gene expression profiles were generated using 5 μg of total RNA on the Affymetrix HG-U133 Plus 2.0 arrays, according to the manufacturer's recommendations (Affymetrix, Santa Clara, Calif). The digitalized images were processed with Microarray Suite 5.0 gene expression software to obtain intensity files. The internal quality controls of the Affymetrix HG-U133 Plus 2.0 arrays led us to discard 5 of 39 hybridized samples. For these samples, we observed elevated RNA degradation, reflected by increased “glyceraldehyde-3-phosphate dehydrogenase ratios,” low “percentages of present transcripts,” and increased “scale factors.” Transcripts that were absent in >90% of the samples were excluded from analyses.

Class Discovery Experiment

To limit the effect of chip-to-chip variation, additional data normalization was performed. For each transcript, intensity values were divided by the median intensity calculated from all 34 tumors before log2 conversion. First, 1000 transcripts with the highest standard deviation (SD) were isolated for hierarchical clustering and sample dendrogram construction. Then, the transcripts responsible for dendrogram formation were isolated by analyses of variance (ANOVAs) based on the 4 predominant tumor groups; 1) low-grade spinal cord tumors, 2) low-grade filum terminale tumors, 3) low-grade posterior fossa tumors, and 4) high-grade tumors (α = .01). Significance assessment was realized by 5000 hazardous permutations. Finally, transcripts with statistically significant group associations were isolated and used for hierarchical clustering. Isolated gene clusters were evaluated by using the Gene Ontology tool in the Database for Annotation, Visualization, and Integrated Discovery (DAVID) platform (available at: accessed June 2009).

Correspondence Analysis

Correspondence between tumor samples and transcription profiles was evaluated based on the 3 most informative chi-square values that revealed associations among transcripts and experiments. Tumors were grouped according to tumor localization or WHO grade. Intensity values for a given transcript were divided by the median intensity calculated from all selected tumors before log2 conversion. The 500 transcripts with highest SD were isolated for correspondence analysis (CoA).

Supervised Analysis: Extracranial Versus Intracranial Tumors

Localization-specific analysis was performed on the 32 ependymomas with known tumor localization. Transcripts were divided by their median intensity value before log2 conversion. Differentially expressed genes were isolated by using the t test, significance analysis of microarrays (SAM), or NetAffx-SAM application. For the t test application, assessment of significance was realized by 5000 hazardous permutations (P < .01). For SAM, the cutoff level was a predicted false discovery rate of <1 transcript (false discovery evaluated on 500 hazardous permutations). In the NetAffx-SAM application, cancer-related sets of transcripts were isolated from the NetAffx platform and were evaluated statistically by using the SAM algorithm, with a predicted false discovery rate of <1 transcript as the cutoff level. Only transcripts that passed at least 1 of the 3 statistical filters were used for further evaluations. Because 14 of 16 extracranial ependymomas were WHO grade 2 tumors, data were “polished” from the WHO grade-induced bias by using 2-factor ANOVAs, differentiating simultaneously between extracranial and intracranial tumors and between WHO grade 2 and grade 3 tumors. Expression profiles of transcripts exclusively linked to WHO grades were discarded from tumor localization-dependant supervised analysis. The selected transcripts were reorganized into pathways or gene sets based on pathway information extracted from the Kyoto Encyclopedia of Genes and Genomes (KEGG) database (available at: accessed June 2009) and the Ambion database (available at: accessed June 2009). When genes were known to play roles in multiple pathways, only the pathway(s) revealed by additional dysregulated transcripts were considered.

Supervised Analysis: World Health Organization Grade 2 Versus Grade 3 Tumors

For the analysis of WHO grade 2 versus grade 3 tumors, we excluded from our series of 34 tumors 1 WHO grade 1 ependymoma and 1 tumor for which no histologic slides were available. We also discarded 2 tumors for which 3 independent neuropathologists did not agree on WHO grades (n = 30). Four additional posterior fossa ependymomas were characterized by the presence of hypercellular foci. Because the impact of this phenomenon on the tumor's clinical evolution is unclear,19, 20 and because posterior fossa ependymomas are characterized by a broad molecular heterogeneity,13, 15 we designated the WHO grading of these tumors as “uncertain” and discarded them from the supervised analysis. On the basis of these selection criteria, we performed this analysis on 17 WHO grade 2 ependymomas and 9 WHO grade 3 ependymomas. Before log2 conversion, transcripts were divided by the average of the median intensity assessed from the WHO grade 2 tumors and from the WHO grade 3 tumors, respectively. The statistical filters of the t test, SAM, and NetAffx-SAM applications were used, as described above. Similar to the tumor localization-dependent supervised analysis, a 2-factor ANOVA algorithm was used to discard transcripts that were linked only to tumor localization.


Biostatistical analyses were carried out at the TIGR Multiexperiment Viewer (MeV) platform (available at: accessed June 2009) using the hierarchical clustering, t test, SAM, ANOVA, 2-factor-ANOVA, and CoA modules. For supervised analyses, differentially expressed transcripts were studied using the “Pathway Analyzer” and the “Gene Ontology” applications from the DAVID platform as well as the pathway databases from KEGG, Biocarta, and Ambion.


Class Discovery Experiment

By using 1000 transcripts with the highest SD within the 34 profiled tumors, we performed hierarchical clustering to construct a dendrogram based on Euclidian distance. Spontaneous cluster division between low-grade and high-grade tumors was observed. The low-grade tumors also were divided according to localization in the spinal cord, posterior fossa, and filum terminale (Fig. 1A). To assess which of the 1000 transcripts were responsible for the observed dendrogram formation, a 1-way ANOVA was performed by opposing the 4 described tumor clusters: 1) low-grade spinal cord tumors, 2) low-grade posterior fossa tumors, 3) low-grade filum terminale tumors, and 4) high-grade tumors. This procedure resulted in 538 selected transcripts, corresponding to 457 different genes that could be divided into 4 distinct gene clusters: I, II, III, and IV (Fig. 1B).

Figure 1.

Class discovery experiment: (A) On this heat map of the 1000 genes with the highest standard deviations in the series of 34 ependymomas, green indicates low expression, and red indicates high expression. Tumor samples were organized by hierarchical clustering based on Euclidian distance. In the dendrogram green indicates, medullary low grade; blue, posterior fossa low grade; yellow, filum terminale low grade; red, high grade. (B) Heat map of 538 transcripts isolated by analyses of variance: Cluster I and II transcripts were up-regulated in high-grade and low-grade ependymal tumors, respectively. Cluster III and IV transcripts were up-regulated in intracranial and extracranial ependymal tumors, respectively. GO indicates Gene Ontology data from the Database for Annotation, Visualization, and Integrated Discovery (DAVID) platform; BIRC5, baculoviral IAP repeat-containing 5; EPHB2, ephrin receptor B2; ERBB3, v-erb-b erythroblastic leukemia viral oncogene homolog 3; VEGFA, vascular endothelial growth factor A; IGF1, insulin-like growth factor 1; NELL2, neural epidermal growth factor-like 2; NNAT, neuronatin; SHANK2, SH3 and multiple ankyrin repeat domains 2; AGR2, anterior gradient homolog 2; ZIC1-ZIC4, Zic family members 1-4; DACH1, dachshund homolog 1; LFNG, lunatic fringe homolog; HOX; homeobox-containing genes.

Cluster I contained transcripts with high expression levels in WHO grade 3 ependymomas. Some of the isolated genes are implicated in apoptosis and angiogenesis (baculoviral IAP repeat-containing 5 [BIRC5], ephrin receptor B2 [EPHB2], v-erb-b erythroblastic leukemia viral oncogene homolog 3 [ERBB3], and vascular endothelial growth factor A [VEGFA]), although the Gene Ontology tool determined the highest enrichment scores for “central nervous system (CNS) development” (growth-associated protein 43 [GAP43], fibroblast growth factor 13 [FGF13], forkhead box G1B [FOXG1B], chondroitin sulfate proteoglycan 5 [CSPG5], stathmin-like 2 [STMN2], hairy and enhancer of split 4 [HES4], and embryonic lethal, abnormal vision-like 3 [ELAVL3]) and for “extracellular matrix” development (microfibrillar-associated protein 2 [MFAP2], MFAP5, fibulin 2 [FBLN2], collagen type I [COL1] α 1 [COL1A1], COL3A1, COL4A1, COL4A2, COL4A3, a disintegrin-like and metallopeptidase with thrombospondin type 1 motif 5 [ADAMTS5], chondroitin sulfate proteoglycan 4 [CSPG4], and tissue factor pathway inhibitor 2 [TFPI2]). Cluster II contained transcripts that are highly expressed in WHO grade 2 ependymomas. Also for this cluster, Gene Ontology assessment isolated genes that are expressed during CNS development (neurotropic tyrosine kinase receptor type 2 [NTRK2], transient receptor protein 5 [TRPC5], sex-determining region Y/box 11 [SOX11], cell division cycle 2 [CDC2], neuronatin [NNAT], carbohydrate sulfotransferase 9 [CHST9], short basic domain/secreted 3A [SEMA3], and dopamine receptor D1 [DRD1]). High enrichment scores for genes implicated in tumorigenesis (insulin-like growth factor 1 [IGF1], neural epidermal growth factor-like 2 [NELL2], NNAT, SH3 and multiple ankyrin repeat domains 2 [SHANK2], and anterior gradient homolog 3 [AGR3]) also were obtained. Cluster III contained genes that are highly expressed in intracranial ependymomas. These genes were involved mainly in neurogenesis (dachshund homolog 1 [DACH1], lunatic fringe homolog [LFNG], ZIC1, ZIC2, ZIC3, and ZIC4). Finally, Cluster IV contained genes that are highly expressed in extracranial ependymomas. Gene Ontology assessment isolated 15 HOX genes, mostly members of the type A and type B HOX clusters, which are implicated in embryonic development.

Correspondence Analysis

On the basis of the 4 clusters isolated by the class discovery experiment, we separately studied the molecular class differences of extracranial (n = 15), intracranial (n = 10), WHO grade 2 (n = 17), and WHO grade 3 (n = 9) ependymomas (Fig. 2). In addition to the tumors of unknown localization (n = 3) and tumors with discordance in grading (n = 2), we discarded tumors that contained simultaneous areas of WHO grade 2 and grade 3 (n = 4). CoA performed on extracranial ependymomas only (Fig. 2A) revealed different correspondences for spinal cord WHO grade 2 tumors and grade 3 tumors and for the 2 filum terminale tumors. Among the intracranial ependymomas (Fig. 2B), posterior fossa tumors, supratentorial tumors, and posterior fossa WHO grade 2 and 3 tumors clustered separately. CoA performed on WHO grade 2 tumors only (Fig. 2C) separated the only childhood spinal cord ependymoma (Fig. 2C, asterisk) from the adult spinal cord ependymomas. Furthermore, the 3 posterior fossa tumors and the filum terminale ependymoma had different correspondences. For WHO grade 3 tumors (Fig. 2D), high correspondence was observed between the 4 posterior fossa tumors and between the 2 medullary ependymomas. A supratentorial ependymoma that had a tanycytic appearance (Fig. 2D, asterisk) had correspondence values that differed from the 2 other tumors with the same localization.

Figure 2.

Correspondence analysis of molecular profiles: The 3 axes in each chart (1-3) correspond to the 3 strongest chi-square–derived associations between samples and expression profiles. This analysis was performed on 500 transcripts with the highest standard deviation and (A) extracranial ependymomas, (B) intracranial ependymomas, World Health Organization (WHO) grade 2 ependymomas (C), and (D) WHO grade 3 ependymomas. For ependymoma subgroups, solid circles, indicate WHO grade 2 spinal cord ependymomas; solid circles with asterisk, childhood WHO grade 2 spinal cord ependymoma; solid squares, filum terminale ependymoma; open triangles, WHO grade 2 posterior fossa ependymomas; crosses, WHO grade 3 posterior fossa ependymomas; pound signs, supratentorial ependymomas; pound signs with asterisk, supratentorial ependymoma with tanycytic appearance; π, WHO grade 3 medullar ependymoma.

Supervised Analysis

Extracranial versus intracranial ependymomas

The analysis of extracranial versus intracranial ependymomas was performed on the 32 ependymomas with known localization (16 intracranial tumors and 16 extracranial tumors). It resulted in a total of 408 differentially expressed transcripts, which corresponded to 323 genes associated with 4 pathways or gene sets (Fig. 3A).

Figure 3.

Supervised analysis of genes that were classified in heat maps of the implicated pathways. Green indicates low expression; red, high expression. A1-A4) Pathways differentially expressed between intracranial and extracranial ependymomas. B1-B7) Pathways differentially expressed between World Health Organization grade 2 and grade 3 ependymomas. HOX indicates homeobox-containing genes; BMP, bone morphogenetic protein; E2F1, E2F transcription factor 1.

1) In extracranial tumors (Fig. 3A-1), statistically higher expression levels were observed in most members of HOX transcription factor clusters A (HOXA1, HOXA2, HOXA3, HOXA5, HOXA6, HOXA7, HOXA9, HOXA10, HOXA11), B (HOXB3, HOXB5, HOXB6, HOXB7, HOXB8, HOXB9), and C (HOXC4, HOXC5, HOXC6, HOXC8, HOXC9, HOXC10) and in 2 members of HOX transcription factor cluster D (HOXD8 and HOXD9). Other HOX genes had similar up-regulation (aristaless-related homeobox [ARX], NK6 homeobox 1 [NKX6-1], transcription factor 2 [TCF2], distal-less homeobox 3 [DLX3], DLX4, myeloid ectotropic viral integration site-related homeobox 2 [MEIS2], and orthodenticle homeobox 2 [OTX2]).

2) In intracranial tumors, various members of the Notch signaling pathway had high expression levels (Fig. 3A-2). The Notch2 receptor, along with its binding modulator LFNG and the positive signaling regulator Deltex1e (DTX1), was highly expressed. Similar expression profiles were detected for the transcriptional Notch coactivators recombination signal-binding protein suppressor of hairless (RBPSUH) and mastermind like 2 (MAML2).

3) Within the members of the Hedgehog pathway (Fig. 3A-3), the ligand Indian hedgehog (IHH), the transcription modulator suppressor of fused (SUFU), and the transcriptional activators glioma-associated oncogene family zinc finger 2 (GLI2) and ZIC2 were highly expressed in intracranial ependymomas. The target genes WNT5A and bone morphogenetic protein 4 (BMP4) also had statistically higher expression levels in this tumor localization.

4) Members of the BMP pathway (Fig. 3A-4), including BMP4, its receptor BMR1B, the transcription modulator “similar to mothers against decapentaplegic” (SMAD) family member 1 (SMAD1), 2 coactivators (protein kinase C α [PRKCA] and calcium channel, voltage-dependent, P/Q type α 1A subunit [CACNA1A]), and the target genes inhibitor of DNA binding 1 (ID1) and ID2, all showed statistically elevated expression levels in intracranial ependymomas.

World Health Organization grade 2 versus grade 3

In an analysis opposing the expression profiles of 17 WHO grade 2 and 9 WHO grade 3 ependymomas, 1442 transcripts were isolated. They corresponded to 1190 genes. The expression profiles were compared with pathways and data-sets listed in the data-bases as implicated in cancer development. One profile was activated in WHO grade 2 ependymomas, and 6 profiles were activated in WHO grade 3 ependymomas (Fig. 3B).

1) In WHO grade 2 ependymomas (Fig. 3B-1), “functional annotation clustering” reveled that, among the up-regulated genes, the dyneins (DNAH1, DNAH5, DNAH7, DNAH9, DNAH10, DNAH11, DNAI1, DNAI2, DNALI1, DNAHD2, DNAHD3, DYNC1LI1, DYNC2LI1, DYNLL1, and DYNLRB2) had high enrichment scores.

2) In WHO grade 3 ependymomas (Fig. 3B-2), inhibitors of β-catenin degradation were highly expressed: Wnt ligand (Wnt11), frizzled receptors (FZD2, FZD5, and FZD8), and dishevelled genes (DVL2 and DVL3). Furthermore, β-catenin and its associated transcription factor TCF3 as well as the Wnt target genes BIRC5, CCND1, FOS-like antigen 1 (FOSL1), c-MYC and tumor protein 53 (TP53) had increased expression.

3) In addition to Ki-67, several cell cycle regulators demonstrated increased expression levels (Fig. 3B-3): cyclins (CDK2 and CDK4), cyclin-dependant kinases (CDK2 and CDK4), cell division cycle proteins (CDC25A, CDC25B, CDC25C, and CDC2), and minichromosome maintenance proteins (MCM2, MCM3, MCM5, MCM6, and MCM7).

4) The E2F1 transcription factor and its associate, the DP1 transcription factor, as well as CDK2, an E2F1 degradation inhibitor, also demonstrated increased expression (Fig. 3B-4). In contrast, cyclin A, which is implicated in E2F1 degradation, had lowered expression.

5) The proapoptotic tumor necrosis factors (TNF) superfamily members 11A and 21 (TNFRSF11A and TNFRSF21), caspases 1 and 4 (CASP1 and CASP4) and BH3-interacting domain death agonist (BID), as well as the antiapoptotic factor BIRC5 (coding the survivin protein) were highly expressed (Fig. 3B-5).

6) Proangiogenic activators (Fig. 3B-6) included VEGF, the kinase insert domain protein receptor (KDR) VEGFR2, VEGFB, TNFAIP3 interacting protein 2 (TNIP2), docking protein 2 (DOK2), protein-tyrosine kinase domain SH1, p21-associated kinase 3 (PAK3), phospholipase C γ1 (PLCG1), and the neuroblastoma RAS viral oncogene homolog NRAS. The angiogenic inhibitor brain angiogenesis inhibitor-associated protein 3 (BAIAP3) had low expression levels.

7) Proteins involved in the destruction of E-cadherin, including the met proto-oncogene MET (hepatocyte growth factor receptor), nonmetastatic protein-23 homolog 1 (NM23H1), caveolin, the Rab protein 5 guanosine triphosphate hydrolase enzyme (Rab5 GTPase), and the Rab7 GTPase, had high expression in WHO grade 3 ependymomas; whereas E-cadherin itself had low expression levels. The Rab11A GTPase, which is involved in the recycling of internalized E-cadherin to the cell surface, demonstrated low expression (Fig. 3B-7).


On the basis of microarray-generated molecular expression profiles, we demonstrated that ependymomas have localization and grade-specific expression signatures. Pathways that differ between extracranial and intracranial tumor localization are implicated in stem cell maintenance, renewal, and differentiation. The signatures related to WHO grade 3 ependymomas are linked to tumor aggressiveness. The basis for these signatures is not known but probably includes an accumulation of genetic and epigenetic alterations.

Unsupervised Clustering

Class discovery experiments and CoA underscored the findings that differences in expression profiles depend on tumor localization and histologic grade. Furthermore, our CoA data suggest that patient age and particular histologic features also have an impact on the expression profiles of ependymomas. Indeed, CoA separated the only WHO grade 2 childhood spinal cord ependymoma from adult ependymomas with the same grade and location (Fig. 2A,C). Moreover, the supratentorial ependymoma that had a tanycytic appearance was distinguished from the other supratentorial tumors (Fig. 2B,D). These observations suggest that supplementary subgroups of ependymomas may exist; however, additional studies on larger sample cohorts are needed.

Supervised Analyses

In supervised analyses, we identified Notch, Hedgehog, BMP, and HOX gene clusters to differentiate between extracranial and intracranial ependymomas. These pathways are involved in the control, maintenance, proliferation, and differentiation of stem cells, and they are well known in oncogenesis (Fig. 3A). These differentiating signatures may be caused by tumor cells and/or normal/reactive neighboring cells. Their respective impact could be studied by using microdissected tumor.

In extracranial ependymomas, we identified high expression of most of the members of the HOX transcription factor family (Fig. 3A-1). In previous reports, mostly HOXB513 and HOXA915 were associated with extracranial ependymoma development. HOX genes have been implicated in stem cell renewal and in cell fate determination.21 Furthermore, they are reactivated in hematologic malignancies21 and in solid tumors.22

In intracranial tumors, we confirmed previous studies that reported high expression of members of the Notch and Hedgehog pathways14, 15 (Fig. 3A-2,A-3). Both pathways have been associated with cancer development and control the fate of stem cells.23, 24 Notch overexpression promotes cell proliferation and formation of neural stem cell-like colonies in astrocytomas and gliomas.25 In Drosophila, Notch expression enhances stem cell renewal and survival in the stem cell niche.26 In medulloblastoma, Sonic Hedgehog-induced tumors need Notch signaling to ensure their progression and survival.27 Taken together, these results suggest that different cancer stem cells need simultaneous reactivation of both pathways to ensure their maintenance and proliferation.

In the current study, for the first time to our knowledge, we also linked the activation of BMP signaling to intracranial ependymomas (Fig. 3A-4). Stimulation of the BMP pathway can have oncogenic and/or tumor-suppressor potential. BMP4 has been detected in advanced stages of colorectal cancer28; whereas, in glioblastomas, BMP4 overexpression was associated with tumor regression.29 Conversely, the BMP4 target genes ID1 and ID2 were highly expressed in aggressive astrocytic tumors.30

Comparisons between WHO grade 2 and grade 3 ependymomas led to the identification of 7 differentially expressed pathways or gene-sets. One pathway was stimulated in WHO grade 2 ependymomas, and 6 pathways were stimulated in WHO grade 3 ependymomas. Gene Ontology analysis of genes that were highly expressed in WHO grade 2 ependymomas isolated a group of dynein genes (Fig. 3B-1) known to be implicated in the formation of cytoskeleton and mitotic spindles.31 It is noteworthy that gross chromosomal alterations are observed more frequently in WHO grade 2 ependymomas than in WHO grade 3 ependymomas (personal data; results not shown). Thus, it is possible to hypothesize that the dysregulation of dyneins may be at the origin of this elevated frequency. Finally, the identification of only dynein genes implicated in the specific development of WHO grade 2 ependymomas may indicate a more extensive heterogeneity within this group.

The 6 pathways that were stimulated in WHO grade 3 ependymomas conferred to these tumors several of the “hallmark” characteristics necessary for cancer development and progression18 (Fig. 3B). Among the differentially expressed pathways, we observed the simultaneous dysregulation of Wnt/β-catenin signaling (Fig. 3B-2) and several of its downstream or interacting pathways (Fig. 3B-3-B-7). Therefore, the activation of the Wnt/β-catenin signaling pathway in WHO grade 3 ependymomas is likely to be of primary importance for their tumorigenesis and, thus, an important novel target for ependymoma treatment. It is a major controller of cell fate, proliferation, migration, cell polarity, and cell death; and it is implicated in a variety of cancers.32, 33 In the molecular signature of WHO grade 3 ependymomas, we identified high expression of c-MYC, 1 of the Wnt/β-catenin signaling pathway genes. It stimulates key effectors of cell cycle progression and genome replication. In our series, the acquired cell growth potential of WHO grade 3 ependymomas was confirmed by immunohistochemical analysis, which indicated increased Ki-67 expression (data not shown). These data are in good concordance with several studies suggesting Ki-67 as an independent marker.34, 35 Moreover, c-MYC is implicated in the regulation of the cell cycle-promoting transcription factor E2F1,36 the destruction pathway of which was inactivated in our tumor series. In high-grade ependymomas, E2F1 may induce cell proliferation by the activation of genes involved in progression of the cell cycle from G1 phase into S phase.37 A similar oncogenic stimulation has been described in fibroblast cultures.38-40 Moreover, transgenic mice that had aberrant E2F1 expression developed brain tumors, including medulloblastoma, choroid plexus carcinoma, primary neuroectodermal tumors, and malignant gliomas.41

BIRC5, another Wnt/β-catenin target gene, had high expression levels in WHO grade 3 ependymomas. In those tumors, we observed that high expression levels of various members stimulated intrinsic and extrinsic apoptosis. Because BIRC5 (survivin) is a downstream inhibitor of apoptosis, this may explain why high expression levels of BIRC5 were correlated with an unfavorable outcome in patients with intracranial ependymomas.42 Because the inhibition of apoptosis may play a role in the lack of radiosensitivity and chemosensitivity, survivin may be an additional therapeutic target to be considered.

Angiogenic stimulation may be a result of direct VEGFA activation through the stimulation of Wnt/β-catenin signaling. Such a mechanism has been suggested in colon cancer.43 In WHO grade 3 ependymomas, we observed increased expression levels of VEGFA and other proangiogenic stimulators, such as VEGFB and KDR. VEGFA up-regulation has been described in WHO grade 3 ependymomas, but correlations with overall survival, as reported in astrocytomas, were controversial.44, 45 We hypothesize that angiogenesis of anaplastic ependymomas needs both increased expression of proangiogenic stimulators as well as lowered expression of the angiogenic inhibitor BAIAP3. If so, then BAIAP3 may be a useful marker for prognostic evaluation.

Another important aspect of tumor development is escape from growth-limiting cell-cell interactions/contacts. In our tumor series, WHO grade 3 tumors demonstrated down-regulation of CDH-1 (E-cadherin) and activation of it's destruction machinery. This suggest weakening of intercellular junctions and cytoskeletal dissociation, facilitating tumor growth and progression.46 An inverse correlation between E-cadherin expression and the clinical outcome of cancer patients has been reported for bladder tumors.47 Although, most cancers decrease E-cadherin expression levels by methylation of the promoter region, this was not observed in a series of patients with intracranial ependymomas,48 suggesting the existence of an alternative mechanism.

In addition to the oncogenic role of each of the identified pathways, we were interested in deciphering the “molecular concert” leading to tumorigenesis of ependymomas (Fig. 4). Taylor and coworkers suggested that ependymal tumors recapitulate gene expression profiles of regionally specified stem cells that bear a radial glial phenotype.15 In light of our current results, we suggest that, in developing intracranial ependymomas, simultaneous reactivation of the Notch and Hedgehog pathways maintains the stem cell state and directs stem cell proliferation, whereas BMP directs specific, morphogenetic cell differentiations.25, 49 Furthermore, Taylor and coworkers suggested that, compared with intracranial ependymomas, spinal cord ependymomas arise from different populations of progenitor cells. It is well known that HOX genes coordinate the development of the CNS50 and are implicated in the maintenance, proliferation, and differentiation of stem cells. Our data suggest that the dysregulation of HOX transcription factors is the driving force for extracranial ependymoma development. These data correspond with the “oncology recapitulates ontology” hypothesis, in which genes that are implicated in stem cell fate decisions also may be important for supporting cancer stem cells.

Figure 4.

Tumorigenesis scheme of ependymomas. WHO indicates World Health Organization; BMP, bone morphogenetic protein; E2F1, E2F transcription factor 1; HOX, homeobox-containing genes.

The transformed cancer progenitor/stem cells may accumulate additional genetic hits, which would confer them with further growth advantages.51 Our results demonstrated that ependymomas originating from any compartment acquire the anaplastic phenotype by dysregulation of the same oncogenic pathways, mostly downstream of Wnt/β-catenin. This pathway becomes a target of first choice in therapeutic trials and further implicates an unknown, primary activator event responsible for its stimulation. One clinical impact of our current findings should be the development of targeted chemotherapies pointing to both the pathways implicated in the maintenance of the (cancer) stem cell population and the pathways specific to WHO grade 3 ependymomas.


The authors thank Ms. Liliana Niculescu for excellent secretarial help.

Conflict of Interest Disclosures

This project was supported by Fonds Maisin (UCL), Fédération belge contre le cancer, F.S.R.-FNRS and Télévie. M. Vikkula was “Maître de recherches F.S.R.-FNRS until 30/09/2008”. T. Palm was supported by fellowships from “Fonds pour la formation à la recherche dans l'industrie et l'agriculture” (F.R.I.A.) and patrimoine UCL.