Introduction
Ependymal tumors are classified into four histopathological subtypes including subependymoma (grade I), myxopapillary ependymoma (grade I), ependymoma (grade II), ependymoma,
RELA fusion-positive (grade II or III), and anaplastic ependymoma (grade III) according to the World Health Organization (WHO) Classification of Tumours of the Central Nervous System [
8]. Each of the latter two, which are clinically malignant, is defined as “A circumscribed glioma composed of uniform small cells with round nuclei in a fibrillary matrix and characterized by perivascular anucleate zones (pseudorosettes) with ependymal rosettes also found in about one quarter of cases (ependymoma), a high nuclear-to-cytoplasmic ratio, and a high mitotic count (anaplastic ependymoma)” [
8]. Ependymal tumors, which may arise from any part of the neuroaxis, are identified as supratentorial (ST), posterior fossa (PF) or spinal (SP) ependymomas (EPNs). Malignancy grading for EPN and anaplastic EPN is often inconsistent, and the clinical significance of EPN pathological grading is controversial [
20,
25,
28,
32]. Regardless of location, standard treatment for these tumors involves maximal safe surgical resection followed by local radiation therapy [
29]. Incomplete resection or recurrence predicts a dismal prognosis. At present, there are no reports of chemotherapeutic agents with proven efficacy against these tumors [
11,
35]. Therefore, further clarification of molecular mechanisms underlying the genesis of EPN, as well as development of new treatments for these tumors may be essential.
A series of extensive molecular analyses has demonstrated that supratentorial and posterior fossa EPNs may have distinct molecular profiles and are most likely separate diseases [
19,
25,
27,
29,
37]. Recently, a consensus scheme for the molecular classification of EPNs based on these studies has been proposed [
25]. ST-EPNs were segregated into two molecular subgroups denoted as ST-EPN-
RELA and ST-EPN-
YAP1. ST-EPN-
RELA tumors, which are characterized by the presence of various types of
C11orf95-RELA fusion genes, account for approximately 70% of ST-EPNs [
27]. Some
C11orf95-RELA fusion genes have been experimentally demonstrated as oncogenic. ST-EPN-
YAP1 subgroup is characterized by the
YAP1. ST-EPN-
YAP1 fusion gene, which is much less common than ST-EPN-
RELA. The oncogenic potential of most
YAP1 fusions remain to be determined. However, fusion genes are often intimately implicated in tumorigenesis. Therefore, fusion genes that are highly specific to ST-EPN are likely to be promising therapeutic targets for these particular tumors. However, diagnosis and clinical significance of the
C11orf95-
RELA- or
YAP1-fusion negative ST-EPNs remains controversial. Whether these tumors harbor an unidentified driver event or belong to an entity that is entirely different from EPN, needs to be determined. Thus, a detailed investigation of the molecular profiles of these tumors was felt to be imperative.
According to the methylation pattern, PF-EPNs are segregated into two main molecular subgroups termed PF-EPN-A (PFA) and PF-EPN-B (PFB) [
19,
25]. PFA subgroup tumors are characterized by an increased DNA methylation pattern in the CpG islands, which is different from that seen in PFB ependymomas. PFA patients are mostly infants or young children associated with a poor prognosis whereas PFB patients are older with better prognoses [
19,
25]. No recurrent driver mutation, identifiable as a therapeutic target or diagnostic marker, has been found in either subgroup. Because the presence of biologically distinct subgroups within PF-EPN has significant clinical implications, validation of these sub groups as well as development of robust diagnostic tools for these groups are deemed essential.
For the purpose of molecular and clinical characterization of ependymal tumors and identification of therapeutic targets, we performed molecular analyses on a considerable series of ependymal tumors collected via the Japan Pediatric Molecular Neuro-oncology Group (JPMNG). These cases were examined using detailed clinical information and centrally reviewed histopathology. We confirmed that RELA fusion is a highly specific diagnostic marker for ST-EPN, and that methylation-based classification of PF-EPN is robust and may serve as an independent prognostic marker. We found that a considerable proportion of histopathologically diagnosed ST-EPN does not contain the RELA fusion, and that the molecular pathogenesis of these tumors may be complex.
Discussion
Molecular classification is essential for integrated diagnosis of central nervous system tumors in modern diagnostic pathology. However, using such classification in ependymomas is challenging due to the very limited number of available markers. In this study, the proposed molecular classification of supratentorial and posterior fossa ependymomas was extensively investigated in an independent set of 113 locally diagnosed ependymal tumors in Japan.
Our study confirmed that C11orf95-RELA fusion is a unique genetic feature of ST-EPN and that its presence is consistent with histopathological diagnosis. On the other hand, pathogenesis of ST-EPN in the absence of C11orf95-RELA fusion remains unresolved. There were 9 C11orf95-RELA fusion-negative ST-EPNs including 4 ependymoma grade IIs and 5 anaplastic ependymoma grade IIIs. Thus, those tumors were histologically confirmed as ependymoma, by definition. None of these were diagnosed as subependymoma following central review, where only a single case of YAP1 fusion was identified by FISH (EP117). Instead, 2 novel fusion genes, EP300-BCORL1 and FOXO1-STK24, were detected in single cases (EP3 and EP57).
EP300 (E1A binding protein p300, located at 22q13) is a transcriptional coactivator that binds to a variety of transcription factors and bridges them to basal transcription machinery, and additionally functions as histone acetyltransferase that relaxes chromatin structure [
5].
BCORL1 (BCL6 Corepressor Like 1, located Xq26.1) is a transcriptional corepressor that interacts with histone deacetylases to repress transcription of genes such as E-cadherin [
23]. The
EP300-BCORL1 fusion found in EP3 retained nearly all functional domains of both genes, but
BCORL1 expression was significantly increased in EP3 (Additional file
10 Figure S8). Interestingly, 2 ossifying fibromyxoid tumors with a
CREBBP-BCORL1 fusion have been reported [
12].
CREBBP (CBP) is a paralog of
EP300 and as their functions mostly overlap they are often collectively described as
CBP/EP300) [
5].
BCORL1 was overexpressed in
CREBBP1-BCORL1 fusion-positive tumors, suggesting that activation of
BCORL1 may be a consequence of
CREBBP1-BCORL1. Thus, it is likely that
BCORL1 activation, a consequence of
EP300-BCORL1, may lead to deregulation of chromatin remodeling through recruitment of histone deacetylase. In addition,
BCOR (BCL6 corepressor) internal tandem duplication (ITD), which acts as an activating oncogene [
34], was also found in a single ST-EPN in our cohort (EP116). The DKFZ classifier matched the
BCOR ITD tumor to “CNS high grade neuroepithelial tumor with
BCOR alteration (CNS HGNET-
BCOR)” with a high score (0.99). The
BCORL1-fusion tumor was interpreted by the DKFZ classifier as a ‘no match,’ although it was also classified as CNS HGNET-
BCOR with a low score (0.44) (Fig.
1, Additional file
3 Table S3). Thus it is likely that these tumors may belong to a new entity within the unclassified heterogeneous high grade neuroectodermal/glial tumors of children [
30]. An HDAC inhibitor may potentially be effective for tumors with activated
BCOR/BCORL1.
Much less is known about the
FOXO1-STK24 fusion found in another ST-EPN (EP57).
FOXO1 is a transcription factor that is involved in the maintenance of cellular homeostasis [
36].
PAX3-FOXO1 fusion, which acts as a highly activated transcription factor, is found in 60% of alveolar rhabdomyosarcomas [
36].
STK24 (also known as
MST3) is a serine-threonine kinase that functions upstream of the mitogen-activated kinase (MAK) signaling pathway.
STK24/MST3 is overexpressed in breast cancers and promotes proliferation and tumorigenicity [
30]. Recurrent mutations or fusions of
STK24 have not been reported. The DKFZ classifier found no match for this ST-EPN tumor (classified as PFB, score = 0.44). Interestingly, this tumor showed copy number oscillation compatible with chromothripsis on chromosomes 13, on which
FOXO1 and
STK24 are located, strongly suggesting that this may be the mechanism underlying the gene fusion. Both
FOXO1 and
STK24 were overexpressed in EP57 (Additional file
10 Figure S8), suggesting that either of them may carry an oncogenic property. Although a detailed study of individual cases is beyond the scope of this paper, this tumor may warrant further investigation.
None of the other
RELA fusion-negative ST-EPN were classifiable even with the DKFZ classifier. In summation, our findings suggest that
RELA/YAP1 fusion-negative ST-EPNs may be a heterogeneous group of tumors that consist of a variety of mutations or rare fusion genes, which are unlikely to belong to a single category. Further studies using a vast number of tumors may help in clarifying whether tumors with similar genetic changes and/or DNA methylation profiles truly define a new tumor entity. Considering the high homogeneity of
RELA-fusion positive ST-EPNs, it is doubtful whether these are biologically equivalent to ependymoma. According to the latest WHO Classification [
8], ependymomas are primarily diagnosed via histology. As such, they may be diagnosed as ependymomas, at least for the time being. Nonetheless, it is important to be aware that histologically diagnosed RELA-fusion negative ependymomas may have a biology which is different from that of quintessential RELA-fusion positive ependymomas. Further molecular classification and incorporation into future WHO Classification criteria is warranted.
In contrast to a previous large series, no significant association between the presence of
C11orf95-RELA fusion and patient survival was noticed in our series [
25]. Furthermore, RELA fusion status was reportedly not related to a significant difference in the survival of ST-EPN patients [
9]. In addition, the rate of GTR in RELA fusion-positive ST-EPN was not statistically significant compared to that in RELA fusion-negative ST-EPN (
p = 0.55) in our cohort. The impact of
C11orf95-RELA fusion on patient survival needs to be further investigated. These findings may reflect the fact that
RELA fusion-negative ST-EPNs are a biologically heterogeneous group of tumors. Interestingly, median progression-free or overall survival was not reached for
C11orf95-RELA fusion positive ST-EPNs. Other proposed prognostic molecular markers of ependymomas include
TERT and
EZH2 expression [
18,
21,
31]. Although we confirmed elevated
EZH2 and
TERT expression in
RELA fusion-positive ST-EPNs, they were not associated with patient survival. Nonetheless, it may be of interest that
TERT mRNA expression was elevated in
RELA fusion-positive ST-EPNs, to an extent which far exceeded that in glioblastomas with
TERT promoter mutations (Additional file
9 Figure S4b). None of the ST- or PF-EPNs in this cohort carried the
TERT promoter mutation (data not shown). This phenomenon has also been described elsewhere [
10]. Costelo-Branco et al., found that the methylation status of some CpG sites upstream of transcription starting site of
TERT, were positively correlated with
TERT mRNA expression in childhood malignant brain tumors and were also associated with the prognosis of patients with PF ependymoma [
5]. Although neither
TERT mRNA expression nor
TERT UTSS methylation was associated with patient prognosis in this series,
TERT UTSSs were highly methylated in the RELA fusion-positive ST-EPNs with elevated
TERT mRNA expression. The mechanism of
TERT upregulation appears to be complex and warrants further investigation.
We validated the proposed molecular classification of PF-EPN for efficacy in predicting clinical characteristics including that of patient survival. The 450 K analysis accurately classified the published reference PF-EPN dataset, confirming the robustness of the analysis. PFA showed a minor but significant increase in methylation levels and distinct methylation profiles when compared to PFB (Fig.
2). With a few exceptions, PFA patients were mostly infants and the ages of the PFB patients were significantly higher than those of PFA (Additional file
14 Figure S3a). PFA tumors showed significantly more lateral extension compared to PFB, most of which were medially located (Additional file
14 Figure S3b). DKFZ classifier results were mostly consistent with our analysis with a few exceptions. Two PFAs showed no match. One PFB (EP96) was classified as pituitary adenoma and another PFB (EP86) as myxopapillary ependymoma. These classifications were not compatible with their histology or location.
Our multivariate analysis using Cox regression showed that the PFA subgroup was the only molecular marker which was independently associated with patient PFS and OS among all ependymomas. Among PF-EPN, PFA patients showed significantly shorter PFS and OS compared to PFB patients. These findings corroborated previous reports [
19,
29] and consolidated the significance of proposed molecular classification, indicating that PFA and PFB may be biologically distinct subgroups of PF-EPN. The important clinical implication of the PFA/PFB classification is its potential to aid therapeutic decision making. Based on the results of a study conducted on a large series of PF-EPN, Ramaswamy suggested that a substantial proportion of totally resected PFB patients may be treated with surgery alone, without radiotherapy [
29]. Although this suggestion needs to be tested in a randomized clinical trial, it is evident that molecular classification may play an important role in the clinical management of ependymomas. Although resection rate was not significantly associated with survival in our survival analysis, there was a tendency for gross total resection (GTR) to predict longer survival (Additional file
15 Figure S9). This may be due to the relatively small number of cases screened in the study. Retrospectively, the extent of resection although determined locally was not centrally reviewed which may be a limitation of the multi-institutional nature of the study. Data from the Collaborative Ependymoma Research Network (CERN), which was also a multi-institutional study, did not indicate statistically significant differences in PFA survival due to resection rate [
29]. A prospective clinical trial for ST-and PF-EPN with molecular classification and a standardized central review for the extent of resection is on-going in Japan Children’s Cancer Group (JCCG).
In spite of its usefulness, a practical problem associated with methylation classification, is its cost as well as limited availability as a routine diagnostic test. To overcome this issue, a simplified methylation test was developed to determine PFA/PFB for PF-EPNs by examining the 3 most highly methylated regions in PFA via pyrosequencing and rigorously validating it by using an extended PF-EPN cohort of 123 PF-EPNs which combined our cases and an independent set of samples from Toronto (
Results and Fig.
5). With this novel assay, we were able to diagnose PFB with 100% specificity. We also confirmed the efficacy of anti-H3K27me3 immunohistochemistry, recently reported by Panwalkar et al. [
26], by predicting PFB with 100% specificity in a selected Japanese cohort of 44 PF-EPNs (Fig.
5 ). It may be noteworthy, that the proposed cutoff of 80% for H3K27me3 immunohistochemistry, though appropriate, may prove to be somewhat counter-intuitive for judging reduced PFA expression. Among the 4 PFBs examined via both methods, certain single cases were misclassified as PFA in each method. Although methylation assessment at individual CpG sites has its own limitation such as potential heterogeneity of methylation across CpGs as well as masking by co-existing non-neoplastic cells, these assays may serve as clinically applicable techniques for rapid molecular classification of PF-EPN, which are also suitable for risk-grouping in clinical trials. Hopefully, these may lead to better treatment decision making for the ependymoma patients in the future.
It has been suggested that the presence of 1q gain is associated with poor prognosis in ependymomas [
13,
16,
24,
25]. A large cohort study indicated significant differences in PFS between patients with and without 1q gain in both PFA and PFB, whereas significantly shorter overall survival in patients with 1q gain was seen only in PFA patients but not in PFB or ST-EPN-
RELA patients [
6,
24,
25]. In our cohort, 1q gain was highly enriched in a subset of PFA, while only PFB had 1q gain (Fig.
2). PFA patients with 1q gain were older at onset (
p < 0.001; Fig.
3a) and exhibited significantly shorter PFS than those without 1q gain (
p = 0.016, Fig.
4e). However, there was no significant difference of OS between those patients (
p = 0.51, Fig.
4f). The reason for discrepancies related to the impact of 1q gain on OS of PF-EPN between our study and others is currently unknown. It has been recently proposed that PFA and PFB may further be divided into 9 or 5 subgroups [
6,
24]. Although our cohort was too small to validate such subgrouping, it is likely that PF-EPN are a heterogeneous group of tumors. The significance of molecular markers/subgroups in patient prognosis needs to be examined by a prospective study.
Our study demonstrated that histopathological diagnosis of ependymomas such as ST-EPN is often challenging. Eight locally diagnosed ST-EPNs were re-classified as non-EPN tumors following a pathology review. None of them carried
RELA-fusion. The significance of WHO grading of ependymomas is highly controversial [
7,
25]. On the other hand, our histopathological review classified most PFA as WHO grade III, and PFB as WHO grade II. Current WHO Classification bases the diagnosis of ependymomas solely on the histopathology of tumors. Thus the role of histopathology needs to be revisited.