Significance of molecular classification of ependymomas: C11orf95-RELA fusion-negative supratentorial ependymomas are a heterogeneous group of tumors

Abstract

Extensive molecular analyses of ependymal tumors have revealed that supratentorial and posterior fossa ependymomas have distinct molecular profiles and are likely to be different diseases. The presence of C11orf95-RELA fusion genes in a subset of supratentorial ependymomas (ST-EPN) indicated the existence of molecular subgroups. However, the pathogenesis of RELA fusion-negative ependymomas remains elusive. To investigate the molecular pathogenesis of these tumors and validate the molecular classification of ependymal tumors, we conducted thorough molecular analyses of 113 locally diagnosed ependymal tumors from 107 patients in the Japan Pediatric Molecular Neuro-Oncology Group. All tumors were histopathologically reviewed and 12 tumors were re-classified as non-ependymomas. A combination of RT-PCR, FISH, and RNA sequencing identified RELA fusion in 19 of 29 histologically verified ST-EPN cases, whereas another case was diagnosed as ependymoma RELA fusion-positive via the methylation classifier (68.9%). Among the 9 RELA fusion-negative ST-EPN cases, either the YAP1 fusion, BCOR tandem duplication, EP300-BCORL1 fusion, or FOXO1-STK24 fusion was detected in single cases. Methylation classification did not identify a consistent molecular class within this group. Genome-wide methylation profiling successfully sub-classified posterior fossa ependymoma (PF-EPN) into PF-EPN-A (PFA) and PF-EPN-B (PFB). A multivariate analysis using Cox regression confirmed that PFA was the sole molecular marker which was independently associated with patient survival. A clinically applicable pyrosequencing assay was developed to determine the PFB subgroup with 100% specificity using the methylation status of 3 genes, CRIP1, DRD4 and LBX2. Our results emphasized the significance of molecular classification in the diagnosis of ependymomas. RELA fusion-negative ST-EPN appear to be a heterogeneous group of tumors that do not fall into any of the existing molecular subgroups and are unlikely to form a single category.

Keywords: Ependymal tumors; Fusion gene; Gene rearrangement; Molecular classification.

Fig. 1

Clinical and genomic features of supratentorial ependymomas (ST-EPNs). Central and local histological diagnoses are indicated in the top column. All genotypes examined are shown (un-examined genotypes are left as blank). The results of the DKFZ classifier are shown in the bottom columns. Patients’ ages are indicated below the diagram. C11orf95-RELA fusions were detected among only ST-EPNs diagnosed by consensus diagnosis. ST tumors confirmed by consensus diagnosis without C11orf95-RELA fusions show various genetic alterations including YAP1 fusion

Fig. 2

Classification of posterior fossa ependymomas (PF-EPNs) using genome-wide methylation profiling. A heatmap analyzed by 3086 probes which showed high standard deviations (SD > 0.25) on CpG islands for unsupervised hierarchical clustering of 60 centrally-diagnosed posterior fossa ependymomas shows that the tumors are divided into two clusters as PFA and PFB. The following information is indicated below the heatmap: tumor location, a pattern of PF tumor extension, pathological grading, the presence of 1q gain, age at onset, and the DKFZ classifier results

Fig. 3

Survival of ST-EPNs stratified according to the presence of C11orf95-RELA fusions. a Progression-free survival (PFS), b overall survival (OS). There was no survival difference between the two groups. (c-d) PFS (c) and OS (d) of PFA and PFB. Significant differences in OS (p = 0.009) were observed between PFA and PFB patients. (e-f) PFS (e) and OS (f) of PFA with or without 1q gain. A significant difference in PFS (p = 0.02) but not in OS was observed between them

Fig. 4

Prediction of PF-EPN subgroups using methylation thresholds of CRIP1, DRD4, and LBX2. a Methylation percentages for the three genes in the training dataset. b Likelihoods for each subgroup calculated by presuming beta distribution. The long-dashed lines denote thresholds determined by likelihood ratios. c Confusion matrices of prediction with training and validation datasets, according to the rule that classifies a case as PFB if all three genes suggest PFB

Fig. 5

Immunohistochemistry for H3K27me3 in PFA and PFB tumors. All PFA tumors demonstrated reduced H3K27me3 expression (80% or less). Approximately 38% of them showed reactivity in less than 5% of tumor cells (a, e), and the remaining cases showed labeling in 5–50% of tumor cells (b, e). In contrast, most PFB tumors retained intact H3K27me3 expression (> 80% labeled nuclei) (c, e). A few PFB tumors, however, showed labeling in 10–60% of cells, which were categorized as reduced expression of H3K27me3 (d, e). e, a histogram of the percentage of labeled nuclei in PFA and PFB tumors. f, a confusion matrix for actual and predicted subgroup by H3K27me3 immunohistochemistry when PFB was defined as intact H3K27me3 expression and PFA as reduced expression