MYCN amplification drives an aggressive form of spinal ependymoma

Tumor tissue and retrospectively collected clinical data from 13 patients with the local diagnosis of spinal ependymal tumors (made between 2003 and 2018) were obtained from multiple international collaborating centers and collected at the Department of Neuropathology of the University Hospital Heidelberg (Heidelberg, Germany). For all cases, a genotype check was performed to exclude the possibility that material from the same patient was received from more than one center. To this end, the Pearson correlation across beta methylation values of 59 rs-loci present on both the Illumina Infinium HumanMethylation450 and the Illumina Infinium HumanMethylation EPIC array were calculated. Samples with a correlation ≥ 0.95 were considered as genotype match.

Written consent by all patients or their legal representative was obtained. Research use of tissues, clinical and radiological data were in accordance with local ethical approvals.

Genome-wide DNA methylation profiling was performed using the Illumina Infinium HumanMethylation450 and the Illumina Infinium HumanMethylation EPIC Kits as previously described and according to the manufacturer’s instructions [31].

All computational analyses were performed in R version 3.4.4 (R Development Core Team, 2019). Raw signal intensities were obtained from IDAT files using the minfi Bioconductor package version 1.24.0 [2, 15]. Illumina EPIC and 450k samples were merged to a combined data set by selecting the intersection of probes present on both arrays (combineArrays function, minfi). Each sample was individually normalized by performing a background correction (shifting of the 5th percentile of negative control probe intensities to 0) and a dye-bias correction (scaling of the mean of normalization control probe intensities to 10,000) for both color channels. Subsequently, a correction for the type of material tissue (FFPE/frozen) and array (450k/EPIC) was performed by fitting univariate, linear models to the log2-transformed intensity values (removeBatchEffect function, limma package version 3.34.5). The methylated and unmethylated signals were corrected individually. Beta-values were calculated from the retransformed intensities using an offset of 100 (as recommended by Illumina).

Before further analysis, the following filtering criteria were applied: removal of probes targeting the X and Y chromosomes (n = 11,551), removal of probes containing a single-nucleotide polymorphism (dbSNP132 Common) within five base pairs of and including the targeted CpG-site (n = 7998), probes not mapping uniquely to the human reference genome (hg19) allowing for one mismatch (n = 3965), and 450k array probes not included on the EPIC array. In total, 428,230 probes were kept for downstream analysis.

To perform unsupervised non-linear dimension reduction, the remaining probes were used to calculate the 1-variance weighted Pearson correlation between samples. The resulting distance matrix was used as input for t-Distributed Stochastic Neighbor Embedding analysis (t-SNE; Rtsne package version 0.13). The following non-default parameters were applied: theta = 0, pca = F, max_iter = 2500 perplexity = 20.

To identify fitting samples for this study, DNA-methylation profiles of 53,468 samples from different tumor entities and experimental data were screened and compared with the reference cohort of the Heidelberg brain tumor methylation classifier which is based on 2682 CNS tumors representing 82 distinct tumor methylation classes (https://​www.​molecularneuropa​thology.​org) [6].

CNV analysis from 450k and EPIC methylation array data was performed using the conumee Bioconductor package version 1.12.0 (Hovestadt V, Zapatka M, 2017).

Hematoxylin–eosin (H&E)-stained slides were evaluated applying the diagnostic criteria provided by the 2016 WHO Classification of Tumors of the Central Nervous System [24]. Tumors were assessed histologically for the following features: cellularity, perivascular pseudorosettes, microvascular proliferation, necrosis, and mitotic activity.

Immunohistochemistry and fluorescence in situ hybridization were conducted on 1-μm thick formalin-fixed, paraffin-embedded (FFPE) tissue sections mounted on StarFrost Advanced Adhesive slides (Engelbrecht, Kassel, Germany) followed by drying at 80 °C for 15 min. Immunohistochemistry was performed on a BenchMark Ultra immunostainer (Ventana Medical Systems, Tucson, AZ, USA).

For Ki67 analysis, tumor areas with the highest Ki67 labelling indices were evaluated for the fraction of positive cell nuclei by counting all cells excluding lymphocytes and vascular cells in one 200 × microscopic field.

Tissue for ultrastructural examination in one patient was retrieved from a paraffin block, rehydrated, post-fixed in 2.5% glutaraldehyde, and processed and stained for electron microscopy as per routine protocol. Thin sections were examined on a Zeiss 910 electron-microscope.

Two-color interphase FISH was performed using a target probe for MYCN (2p24; green) and AFF3 probe (red) as a reference (Zytovision SPEC MYCN/2q11 Dual Color Probe). Samples showing sufficient FISH efficiency (> 90% nuclei with signals) were evaluated. Signals were scored in at least 200 non-overlapping, intact nuclei. Specimens were considered amplified for MYCN locus when more than 10% of tumor cells exhibited either more than eight signals of the corresponding probe with a reference/control ratio > 4.0 or innumerable tight clusters of signals of the reference locus probe.

Gene expression profiling was analyzed using two different approaches: RNA sequencing and Affymetrix arrays. RNA sequencing from fresh frozen tumor material of two patients was performed by the High Throughput Sequencing Unit of the Genomics & Proteomics Core Facility at the DKFZ using the Illumina HiSeq 2000 platform (V4: 125 bp paired end reads) and the TruSeq Stranded mRNA Library Prep Kit according to the manufacturer’s instructions. The general processing of RNA-sequencing data (reads alignment, quality control, and counts computation) was performed as previously described [26]. Gene expression profiling on the Affymetrix GeneChip U133 Plus 2.0 array was undertaken for one patient as previously described [43].

Age distribution and median age were calculated using R (R Core Team, 2017). Survival analysis was performed applying the Kaplan–Meier method using GraphPadPrism for Windows (Graphpad Software 8.0.2, La Jolla California USA, www.​graphpad.​com). P values comparing the survival rates of the respective molecular subgroups were calculated using log-rank tests and were rounded to three decimal digits. PFS was defined as the time interval in years between first diagnosis and progression of a local tumor or detection of distant seeding or local recurrence. OS was defined as the time interval between first diagnosis and death. Patients were censored at the point of death or loss of follow-up. For patient 13, no detailed data regarding OS and PFS were available.

Using a screening approach based on unsupervised analysis of DNA methylation profiling data of a large set of CNS tumors, we identified a distinct cohort of thirteen tumors histopathologically diagnosed as ependymoma. When these samples were clustered with an extensive set of 53,455 DNA methylation profiles covering more than 80 molecularly defined classes of CNS tumors, malignancies outside the CNS, and experimental data (i.e. cell lines, mouse models and patient-derived xenograft models) in a t-SNE-analysis, tumors from this cohort formed a distinct and stable cluster (data not shown) [6]. The highest predicted molecular class was posterior fossa ependymoma type A (PFA) (12/13) or posterior fossa ependymoma type B (PFB) (1/13); however, due to low calibrated scores samples were previously returned as “no matching methylation class” (calibrated score < 0.9, MNP Classifier v11b4). Next, we compared the methylation patterns of our cohort with a reference set of 500 ependymomas from all nine major molecular subgroups [31] (Fig. 1a). The new group did not cluster with any of the other previously described ependymoma subgroups.

DNA methylation array-based CNV plots revealed focal high-level MYCN amplification for all 13 samples of the cohort (Fig. 1b, c, Suppl. Figure 1a, online resource) and several additional chromosomal aberrations at various frequencies, e.g. loss of chromosome 10 (3/13) or focal losses on Chromosome 11q (5/13) (Fig. 2d). Patient 1 additionally showed a BRD4 amplification on chromosome 19p which was maintained throughout several relapses (Suppl. Figure 1b, online resource) and patient 6 showed an additional YAP1 amplification on chromosome 11. Since MYCN amplification is characteristic for aggressive neuroblastomas which are often located close to the spine as well as for a distinct subset of pediatric glioblastomas [13, 21], we repeated DNA methylation-based clustering for the distinct spinal ependymoma cohort with two reference sets of 105 neuroblastomas and 11 MYCN-amplified pediatric glioblastomas, confirming the distinct methylation class of these cases (Suppl. Figure 2, online resource). Additionally, CNV plots were generated for six relapses (from patients 1 and 2). The MYCN amplification remained stable in all six relapsed cases (example given in Suppl. Figure 1b, online resource).

MYCN amplification was validated using fluorescence in situ hybridization (FISH) in cases for which FFPE tissue could be obtained (n = 5 primary tumors, n = 7 relapsed tumors). High-level MYCN amplification was confirmed in all available samples (Fig. 2h; Table 1). A limited set of RNA sequencing data (n = 2) and gene expression profiles generated on the Affymetrix U133 Plus2.0 array (n = 1) allowed comparison of MYCN expression levels with a cohort of spinal ependymal tumors (n = 18) comprising molecular subgroups SP-MPE (n = 8) and SP-EPN (n = 10) as well as to a cohort representing all intracranial molecular subgroups of ependymoma (n = 32) (Fig. 3a, b). MYCN amplified samples showed the highest expression of MYCN compared to both other cohorts. RNA sequencing did not provide significant evidence for additional genetic drivers, such as gene fusions (data not shown).

In conclusion, DNA methylation profiling identified a novel molecular group of spinal tumors with focal MYCN amplification that separates them from previously defined molecular subgroups of spinal ependymal tumors and that is also distinct from other tumor entities with MYCN amplification that may localize in or close to the spinal cord.

All 13 cases of our cohort were independently diagnosed as ependymoma by different neuropathologists at 11 centers in Europe, Australia, and North America. Cases were described as ependymoma, WHO Grade II (n = 3, including one tanycytic ependymoma) or anaplastic ependymoma, WHO Grade III (n = 10). Histopathological evaluation was complemented by electron microscopy for one sample of the cohort (patient 11) identifying intermediate filaments, ciliary structures and zipper-like tight junctions, which are classic ultrastructural features of ependymoma (Suppl. Figure 3, online resource) [17, 28].

For tumors where material was available (n = 12, five primaries and seven relapses), samples were re-evaluated by an experienced neuropathologist from our center (D.E.R.) confirming the initial diagnoses (Table 1). All tumors exhibited histological signs of ependymal differentiation with perivascular pseudorosettes, perivascular GFAP expression, and dot-like positivity for EMA. Microvascular proliferation was also frequently observed. (Fig. 2a–d, Table 1). Most tumors showed brisk mitotic activity and high Ki67 labelling indices (Fig. 2e and g). Tumor necrosis was present in most cases and in all late manifestations (Fig. 2e–g). H3K27me3, which is consistently lost in PF-EPN-A, was retained in all cases (Table 1). All tumors showed widespread expression of MYCN (Fig. 2h). One primary tumor did not show histological high-grade features, but its recurrence showed brisk mitotic activity and overall histological features of anaplasia, i.e. evidence of malignant progression (Suppl. Figure 4, online resource). Interestingly, while an MYCN amplification was detectable in both tumors, the intensity of the immunohistochemical MYCN expression was strongly increased in histological high-grade areas of the recurrent tumor while low-grade areas present in the same FFPE block still showed an only moderate MYCN labelling intensity. Different components with high- and low-grade morphology in tumors of the other patients also provided evidence for a malignant progression during the course of disease. The pattern of differential intensity of the immunohistochemical MYCN expression in histological low- and high-grade components in the same FFPE block was a consistent feature. A strong association between MYCN staining intensity and Ki67 labelling was evident (Table 1; Fig. 2). Taken together, this suggests that MYCN amplification is an early event in tumorigenesis and that malignant progression is associated with a further increase in MYCN protein levels.

Given that all tumors of the cohort showed widespread immunohistochemical MYCN expression, we evaluated whether immunohistochemical MYCN expression may serve as a surrogate marker for MYCN amplification. We stained 20 spinal ependymomas without MYCN amplification, 10 of the methylation group SP-EPN and 10 of the methylation group SP-MPE. None of these tumors showed a strong expression of MYCN. While the majority of cases showed no immunolabelling at all or only occasional positive cells, MYCN-amplified tumors showed a clearly distinct staining pattern with widespread and usually strong immunohistochemical MYCN expression (Fig. 4).

Considering these findings as well as the results of the molecular analysis, which showed MYCN amplification as the characteristic copy number alteration in these tumors, we suggest to designate this new subgroup “Spinal Ependymoma with MYCN amplification” (SP-EPN-MYCN).

Demographic and basic radiological data were available for 13/13 and 11/13 of SP-EPN-MYCN tumors, respectively. Information regarding metastatic spread at diagnosis and during the course of disease could be obtained for 10/13 patients, respectively (Figs. 5 and 6, Suppl. Figure 5, online resource). Median age of onset was 32 years (range 12–56 years) and, therefore, lower than in previous studies on spinal ependymal tumors which reported a median age of about 40 years [7, 18] (Fig. 5a). Patient gender was evenly distributed across the cohort with seven female and six male cases (Fig. 5b). Next we asked whether there is a predilection site for SP-EPN-MYCN similar to SP-MPE and SP-EPN, which are mostly located at the conus medullaris/filum terminale and in the cervical/thoracic spine, respectively [7, 31]. Exact location in relation to the spinal meninges could be determined from radiological data of seven SP-EPN-MYCN patients. For all of these, the primary tumor was located intradurally and extramedullary (Fig. 5c). The majority of the cases arose in the cervical or thoracic spinal cord (n = 10), with only one case showing lumbar localization at initial presentation. Primary lesions were large, with only one case being limited to a single spinal segment. Nine out of ten cases showed multi-locular, diffuse leptomeningeal dissemination at diagnosis, including intracranial metastases in three cases. Diffuse leptomeningeal spread at some point throughout the course of disease was reported in all patients (10/10), including the two cases that did not show metastatic spread at first presentation. However, dissemination was not limited to the leptomeninges, but included nodular lesions as well. Cystic compartments within the malignant lesions were reported in four of the seven patients for whom radiological footage was available. Representative radiological images of an SP-EPN-MYCN tumor in a 46-year-old female patient are given in Fig. 6 (see also: Suppl. Figure 5, online resource). In conclusion, SP-EPN-MYCN tumors were mainly diagnosed in adolescence and early adulthood and showed distinct radiological features, including extramedullary location and diffuse leptomeningeal spread, thus differing strongly from previously described spinal ependymoma cases [37, 45].

Detailed clinical data were collected and subsequently analyzed in 12 of 13 cases. The SP-EPN-MYCN cohort showed aggressive behavior, including early metastases, rapid progression after relapse, dissemination throughout the whole CNS, and resistance to common treatment strategies. Successful gross total resection of the primary tumor was reported for one patient only (1/12) but could not be achieved in others (11/12) due to metastatic spread at diagnosis (9/12) or extended lesions that would have resulted in non-acceptable side effects from surgery (2/12). In the majority of cases, surgery was the initial therapeutic step (9/12) (Fig. 7). All patients relapsed or progressed, often with metastatic spread, at some point during the course of disease. Figure 7 and Suppl. Table 1 summarize applied treatment strategies for all cases. Despite highly intensive treatment regimens, including repeated surgery, radiotherapy, different chemotherapy protocols, and targeted therapy, six patients were deceased and one was at a terminal disease stage at time of data collection. Of the remaining six cases, two were diagnosed in late 2018; thus, data on follow-up are limited. Chemotherapy was applied as single agent or combinatorial treatment including temozolomide, carboplatin, etoposide/cyclophosphamide, etoposide/carboplatin, vincristine/cyclophosphamide, and trofosfamide. In one patient, Imatinib was used due to high c-Kit expression of the tumor cells. Figure 8 shows survival of SP-EPN-MYCN (n = 12) compared to a reference set of cases of the three other molecularly defined spinal ependymal subgroups (Fig. 8a, b) as well as ST-EPN-RELA and PF-EPN-A (Fig. 8c, d). As reference sets for the subgroups SP-SE (n = 5), SP-EPN (n = 9), ST-EPN-RELA (n = 76), and PF-EPN-A (n = 219), data published by Pajtler et al. in 2015 [31] were used. For myxopapillary ependymoma (n = 19), a histologically defined reference set from Kraetzig et al. 2018 [22], was used due to a lack of sufficient numbers for molecularly defined SP-MPE with clinical data. Notably, a limited set of clinical data on molecularly defined SP-MPE that was not included showed identical clinical outcomes to histologically classified myxopapillary ependymoma from Kraetzig et al. 2018 [31]. The median PFS for SP-EPN-MYCN was 17 months and was significantly worse than for SP-SE (p = 0.006), SP-EPN (p = 0.001) and SP-MPE (p = 0.008) (Fig. 8a). No disease-related death was reported for any of the other spinal subgroups, whereas SP-EPN-MYCN showed a median OS of 87 months and dismal outcome compared with SP-SE (p = 0.166), SP-EPN (p = 0.050), and SP-MPE (p = 0.005) (Fig. 8b).

Since spinal ependymal tumors are known to have a relatively benign prognosis in general, we also compared SP-EPN-MYCN tumors with ST-EPN-RELA, and PF-EPN-A, as these two subgroups show the worst prognosis of all so far described molecular ependymoma subgroups [31]. There was no significant difference in OS between SP-EPN-MYCN and ST-EPN-RELA (p = 0.252) or PF-EPN-A (p = 0.353) (Fig. 8c). Notably, PFS was significantly worse in SP-EPN-MYCN compared with PF-EPN-A (p = 0.017), and ST-EPN-RELA (p = 0.047) (Fig. 8d).

In conclusion, SP-EPN-MYCN showed significantly reduced PFS and OS compared to all other spinal ependymoma entities and similar prognosis as unfavorable intracranial subgroups ST-EPN-RELA and PF-EPN-A, thus confirming the highly aggressive nature of this newly defined molecular subgroup of spinal ependymal tumors.