Molecular Stratification of Childhood Ependymomas as a Basis for Personalized Diagnostics and Treatment

Simple Summary The current trend in neuropathology directs to the integrated histo-molecular approach. The traditional concept of histological grade should be complemented by comprehensive diagnostics with the mandatory use of molecular genetic markers. As a consequence, basic types of CNS tumors fall into multiple nosological entities that can be morphologically similar while having fundamentally different pathogenesis and clinical presentation. This trend is particularly evident for ependymal tumors, which harbor molecular markers of decisive importance for the prognosis. This minireview emphasizes recent achievements in ependymoma biology research closely connected with state-of-the-art diagnostics. Abstract Ependymomas are among the most enigmatic tumors of the central nervous system, posing enormous challenges for pathologists and clinicians. Despite the efforts made, the treatment options are still limited to surgical resection and radiation therapy, while none of conventional chemotherapies is beneficial. While being histologically similar, ependymomas show considerable clinical and molecular diversity. Their histopathological evaluation alone is not sufficient for reliable diagnostics, prognosis, and choice of treatment strategy. The importance of integrated diagnosis for ependymomas is underscored in the recommendations of Consortium to Inform Molecular and Practical Approaches to CNS Tumor Taxonomy. These updated recommendations were adopted and implemented by WHO experts. This minireview highlights recent advances in comprehensive molecular-genetic characterization of ependymomas. Strong emphasis is made on the use of molecular approaches for verification and specification of histological diagnoses, as well as identification of prognostic markers for ependymomas in children.


Introduction
Ependymal tumors (ependymomas, EPNs), a common type of malignant neoplasms of the central nervous system (CNS), constitute about 10% of all intracranial tumors and about 20% of spinal cord tumors. EPNs rank third in the prevalence of pediatric CNS tumors (after glial and embryonal tumors) [1]. Despite the use of advanced protocols that include maximal safe surgical resection followed by localized radiotherapy, the mortality remains high due to frequent relapses explained by the strong metastatic potential of EPNs complemented by an efficient spread of metastases with cerebrospinal fluid.
Adverse predictors for EPNs are early age at onset, residual tumor tissue after resection, and metastatic lesions in CNS [2][3][4][5]; however, the detailed prognosis for EPNs is often hampered by (1) clinical and morphological diversity of the tumors and (2) complex relations of histopathological grades with the prognosis [6,7]. In line with the modern trends in neuropathology, the aggressiveness of a tumor and, accordingly, the prognosis is mainly determined by molecular-genetic aberrations, whereas the conventional, histologically apoptosis inhibition, cell growth, and pro-angiogenic signaling-the basic components of oncogenesis and tumor progression. Expression of RELA chimeras results in constitutive activation of NF-κB signaling pathway [17] and associated resistance of the tumor to chemo and radiation therapies [32]. ZFTA chimeric proteins accumulate in the nucleus. A zinc finger domain in the truncated ZFTA protein endows the chimeras with extraordinary high affinity to DNA. The oncoprotein interferes with chromatin structure at ST-EPN-associated loci, enabling the RELA transactivation domain to induce their transcription [30]. Moreover, apart from the canonical NF-κB pathway activation, ZFTA-RELA fusions may trigger other gene expression programs through recruitment of transcriptional co-activators BRD4, EP300, and CBP, which participate in chromatin-related pathways and represent potential druggable targets [31,33].
EPNs with the ZFTA gene fused with a non-RELA partner gene are considerably less common. These tumors have variable histological structures and, apart from the ependymomal component, may additionally involve pleomorphic xanthoastrocytoma-like, astroblastoma-like, malignant teratoma-like, embryonal tumor-like, or sarcoma-like patterns. Despite the heterogeneous morphology, these tumors are (epi)genetically similar and tend to resemble the classic ZFTA-RELA-fused EPNs, as revealed by methylome assay. A detailed analysis of DNA methylation profiles allows subdivision of these tumors into two clusters, one of them comprising tumors with histological features of astroblastomas and xanthoastrocytomas, harboring ZFTA-MAML2 and MN1-ZFTA rearrangements; the second cluster comprises tumors histologically resembling small-cell sarcomatoid carcinomas and undifferentiated sarcomas, harboring ZFTA-NCOA1 and ZFTA-NCOA2 rearrangements [24][25][26][27][28].
The presence of recurrent ZFTA-RELA fusions has been repeatedly implicated as an adverse prognostic factor [5,14,18]. Five-year rates of event-free survival (EFS) and overall survival (OS) for ST-EPN-ZFTA tumors never exceed 29% and 75%, respectively [14]. Within the ST-EPN-ZFTA group, additional risks of relapse have been associated with 1q gains [5]. Interestingly, the St Jude Young Children 07 (SJYC07) study (encompassing ependymal tumors diagnosed in <3-year-olds) identified similar 4-year EFS rates for ST-EPN-ZFTA, ST-EPN-YAP1, and PF-EPN group A [7]. Consistently, two other studies conducted independently by the Italian Association of Pediatric Hematology and Oncology AIEOP and the Children's Oncology Group (trial ACNS0121) revealed no difference in survival rates for ST-EPNs with and without ZFTA-RELA fusion [4,34].
ZFTA (non-RELA)-fused EPNs have an especially dismal prognosis, with EFS rates significantly lower compared with classical ZFTA-RELA-fused EPNs, while the corresponding OS rates are comparable [26]. However, these findings are preliminary, given the small number of cases reported so far. In the context of ST-EPN heterogeneity, it might be useful to consider ZFTA fusions with atypical (non-RELA) partners as a distinguishing feature for a separate group, the prognostic and clinical relevance of which is yet to be specified.

ST-EPN-YAP1 Group
ST-EPN-YAP1 tumors show an aberrant activity of transcription co-activator YAP1 (Yes-associated protein 1) related to its abnormal accumulation in the nucleus. With YAP1 being a direct regulator of TEAD and SMAD transcription factors, its escape from Hippo-dependent sequestration through accumulation in the nucleus results in sustained proliferative signaling via WNT and Hedgehog. More accurately, the nucleus accumulates the oncogenic fusion protein YAP1-MAMLD1 transferred from the cytoplasm to the nucleus independently of its YAP1-Ser127 phosphorylation status that limits the nuclear import of YAP1 in normal cells [35]. Apart from the prevalent YAP1-MAMLD1 fusions, ST-EPN-YAP1 may harbor relatively rare structural variants, e.g., YAP1-FAM118B [14,36]. In some cases, the formation of YAP1 fusions involves focal copy number alterations mapping to the 3 portion of the gene (11q22.1-11q21.2) [37]. Unlike ZFTA-positive ependymomas, ST-EPN-YAP1 tumors have balanced genomes with local aberrations in the YAP1 locus and no evidence of chromothripsis. Andreiuolo et al. (2019) reported a multicenter retrospective study on what is so far the largest cohort of patients with YAP1-positive EPNs (n = 14). Overall survival for these patients (median observation time of 4.8 years within the range of 0.6-16 years) constituted 100%. It is important to note that the boy-to-girl ratio for the studied cohort was 1:6.5, and only three of the patients were over three years old at the time of diagnosis (the median age at diagnosis constituted 8.2 months) [37]. The best survival rates for the ST-EPN-YAP1 group among other EPNs were also reported by other authors [7,14]. Careful de-escalation of conventional EPN treatment protocols specifically for ST-EPN-YAP1 patients is currently under scrutiny. An opportunity to exclude (delay or dismiss) radiation therapy alleviates the risks of severe cognitive dysfunctions, endocrinopathies, and secondary tumors [38].

Molecular Profiles of PF-EPNs
PF-EPNs, more prevalent in children than adults, constitute about 2/3 of intracranial ependymal tumors of childhood. Based on high-throughput molecular techniques, PF-EPNs are subdivided into two molecular groups: PF-EPN group A (PF-EPN-A), and PF-EPN group B (PF-EPN-B) [9].

PF-EPN-A Group
This highly heterogeneous group comprises 85-90% of infratentorial EPNs. PF-EPN-A tumors are often located laterally within the posterior fossa and occur predominantly in infants and young children, twice more frequently in boys than in girls; the average age at diagnosis constitutes 3.5 years [14]. The patients are at high risk of relapse, even under multimodal therapy and in the absence of extra adverse prognostic markers [5]. Identification with PF-EPN-A represents a strong independent prognostic factor associated with the worst rates of survival. According to Zapotocky et al. (2019), 5-year and 10year EFS for PF-EPN-A constitute, respectively, 43% and 37% [3]. Relapses are typical for PF-EPN-A but not PF-EPN-B and, consistently, 10-year OS rates for PF-EPN-A are significantly lower than for PF-EPN-B (56-58% vs. 88-100%) [4,14,15]. Additional negative clinical predictors for PF-EPN-A are the presence of residual tumor tissue (incomplete resection) and adjuvant radiation therapy refusal [3]. Retrospective evaluation of outcomes for the patients receiving treatment under HIT-2000 protocol implicated residual tumors, 1q gains, and high mitotic activity of tumor cells (>10 mitotic figures per 10 fields of view) as independent adverse predictors for PF-EPNs in general and PF-EPN-A tumors in particular [2]. Cytogenetic prognostic factors for PF-EPNs include 1q gains and 6q losses [43,44]. These cytogenetic abnormalities are detected in 18.9% and 8.6% of PF-EPNs, respectively. At that, the 1q gained PF-EPNs harbor 6q losses at an increased frequency of 17.7% [43]. Both types of copy number variations have been qualified as adverse predictors.
Five-year progression-free survival rates were 50% for EPNs without 1q gain and 6q loss, as opposed to 32% for 1q gain only, 7.3% for 6q loss only, and 0% for both 1q gain/6q loss tumors [43]. The ultra-high risks conferred by the co-occurrence of cytogenetic markers in PF-EPN-A patients should be taken into account for the treatment regimen optimization.
PF-EPN-A tumors reveal characteristic aberrant patterns of DNA methylation, the so-called CpG-island methylator phenotype (CIMP) with extensive hypermethylation of CpG-islands in promoter regions of multiple genes. This effect critically interferes with the function of PRC2 (Polycomb repressive complex 2) [45,46]-a transcription repressor protein complex that facilitates methylation of nucleosome histone H3 at amino acid residues H3K27 and H3K9 thus inhibiting the expression of key regulatory genes responsible for cell fate determination and differentiation. Bayliss et al. (2016) revealed the deficiency or complete loss of H3K27me3 in PF-EPN-A tumors [46]. This finding complements the earlier hypotheses on the central role of epigenetic mechanisms in PF-EPN pathogenesis inferred from the absence of presumably pathogenic mutations in chromatin remodeling genes and enzymes that catalyze post-translational modifications (e.g., methylation) of histones in whole-genome sequencing data [17,45,46]. The recruitment of Polycomb group (PcG) transcription repressors to chromatin requires the presence of non-methylated CpGs; accordingly, the loss of H3K27me3 methylation has been associated with dense hypermethylation of CpG-islands preventing the recruitment of PcG proteins to chromatin by steric hindrance.
An advanced investigation of molecular mechanisms responsible for the observed epigenetic malfunctioning revealed a plausible association of the H3K27me3 deficiency with elevated expression levels of accessory proteins encoded by EZHIP (formerly CXorf67) and EPOP (formerly C17orf96) [47][48][49]. As demonstrated by Hübner et al. (2019), EZHIP is a competitive inhibitor of PRC2. A conservative stretch of amino acids in the C-terminal portion of EZHIP mimics the K27 methylation target in histone H3, albeit with K27M substitution. The binding of methionine M27 (instead of lysine K27) to the active center in the histone-lysine N-methyltransferase subunit of PRC2 blocks its catalytic activity [50]. Somatic missense mutations in EZHIP are detected in a small proportion of PF-EPN-A tumors (<10%) [48]. Jain et al. (2019) demonstrated that such mutations have no influence on H3K27me3 levels thus disproving their functional significance [51]. Noteworthy, no loss-of-function mutations in EZHIP (nonsense substitutions or frameshift indels) have been reported. Elevated expression of EZHIP in tumors may be caused by mutations in cis-regulatory elements; the same effect may be conferred by the formation of fusion genes involving EZHIP locus (for instance, MBTD1-EZHIP fusion described for low-grade endometrial stromal sarcoma [52]). However, no fusions comprising EZHIP or PRC2 subunit-encoding genes (e.g., EED, SUZ12) have been described for EPNs.
Related signatures of disrupted epigenetic regulation have been associated with H3 K27M mutations typical for diffuse midline gliomas (DMGs) but rarely found in PF-EPN-A (<5% of the cases). Noteworthy, in EPNs such mutations are harbored by canonical histoneencoding genes HIST1H3C and HIST1H3B, whereas in DMGs they are predominantly found in a replacement histone gene H3F3A (90% of the cases) [48,[53][54][55]. Given the mutually agonistic roles of the onco-histone H3 K27M and EZHIP, it would be natural to expect similar patterns of disease progression and therapy outcomes for H3 K27M-mutant DMG and EZHIP high PF-EPN-A. Indeed, in DMG, disruption of H3K27me3-mediated epigenetic regulation is associated with an extremely aggressive course of the disease, typically presenting with sustained tumor growth and polychemotherapy resistance [56][57][58]. Similarly, effective chemotherapy regimens for PF-EPN-A are missing [59] and therapeutic options for relapses are extremely limited [5,[60][61][62][63].
Despite the uniformity of methylation profiles within PF-EPN-A, tumors of this group show considerable molecular heterogeneity and can be additionally classified into two major subgroups A1 and A2 (and ultimately into nine minor subtypes by using additional markers: gains 1q, deletions 22q, 6q, and 10q, and OTX2 protein expression). PF-EPN-A1 tumors are distinguished by pronounced expression of the homeotic HOX genes (HOXA1/2/3/4, HOXB2/3/4, HOXC4, and HOXD4) which define the segmental (rhombomeric) organization of the hindbrain in early embryogenesis. PF-EPN-A2 tumors hyperexpress EN2, CNPY1, and IRX3-a group of genes involved in the rhombomere differentiation. Expression of A1-and A2-specific genes within the developing hindbrain shows distinct zonality-increased expression of A2 markers is characteristic of the rostral portion at the border with the midbrain, while expression of HOX genes is more pronounced in caudal segments of the brainstem and spinal cord. Differential expression of the spatial patterning genes in A1 and A2 tumors apparently reflects their origin from different hindbrain structures. However, the practical relevance of the advanced A1/A2 subgrouping is questionable. Stratification by clinical factors (gender, age at diagnosis, tumor resection volume, and received therapy) revealed no significant differences between A1 and A2 tumors, except the patterns of relapse (PF-EPN-A1 tumors more often produce local than distant relapses, and vice versa) [48].

PF-EPN-B Group
In contrast to PF-EPN-A tumors which predominantly affect children, PF-EPN-B tumors are more common in adults. In adolescents (aged [10][11][12][13][14][15][16][17], about 45% of newly diagnosed EPNs fall into this group. The prognosis for PF-EPN-B tumors is favorable: 10-year OS rates for the patients after subtotal and gross total resections reach 66.7% and 96.1%, respectively [3,14,15,64]. Thus, the prognosis for this group strongly depends on the extent of surgical resection. The occurrence of delayed relapses (10 years after the onset) underscores the importance of long-term follow-up [64]. Patients with R0 may benefit from chemo-and radiation-sparing strategies; such possibility is being considered [38]. The observed difference in patterns of recurrence between PF-EPN-A and PF-EPN-B adds to the relevance of comprehensive molecular characterization of a tumor as early as possible.
By now, recurrent mutations or fusion genes in PF-EPN-B tumors are missing, and no clear drivers for this group have been identified. Ciliogenesis and microtubule assembly are deregulated only in PF-EPN-B tumors, while several canonical cancer-associated pathways operate in the PF-EPN-A group (VEGF, PDGF, EGFR, RAS signaling, etc.) [14]. PF-EPN-B tumors harbor major cytogenetic aberrations including gains 1q, monosomies 6, 10, and 17, trisomies 5, 8, and 18, and deletions 22q [64]. The diversity of cytogenetic profiles revealed for PF-EPN-B indicate inherent genomic instability and suggest that these tumors emerge from multiple driving events. Similarly with PF-EPN-A, the PF-EPN-B group shows significant heterogeneity, with distinct molecular subtypes of different demographics, copy number alterations, and gene expression signatures. By contrast with PF-EPN-A, gains 1q pose no extra risks for PF-EPN-B tumors. Losses 13q may represent a more reliable negative prognostic marker than gains 1q; however, this assumption requires further substantiation, particularly as the basis for de-escalation of therapy regimens. However, the extent of resection remains the strongest predictor of poor outcomes for this group. Given the patient data scarcity, advanced stratification within PF-EPN-B remains clinically irrelevant [64].
A summary of the intracranial EPN classification is given in Figure 1.

Molecular Profiles of Sp-EPNs
Sp-EPNs constitute a heterogeneous group with a generally favorable prognosis. These tumors mostly occur in adult patients and are rare in children. Clinical outcomes for Sp-EPNs are better than for intracranial EPNs, with 5-year OS rates within the range of 60-90% [14]. Three molecular groups of Sp-EPNs were originally identified, including subependymomas, myxopapillary Sp-EPNs (SP-MPE), and Sp-EPNs per se; the molecular subgrouping shows excellent concordance with corresponding histopathological subtypes [14]. Our knowledge on the molecular pathogenesis of Sp-EPN tumors is limited. The groups reveal characteristic somatic copy number aberrations; most Sp-EPNs harbor 22q deletions involving neurofibromin 2 (NF2) tumor suppressor gene, whereas SP-MPEs show chromosomal instability.
SP-MPEs, the most prevalent type of pediatric spinal cord EPNs, predominantly arise in the conus medullaris, cauda equina, and filum terminale regions [66]. Despite their low mitotic index and slow-growing nature, SP-MPEs generally have far more aggressive behavior than other low-grade CNS tumors. Furthermore, pediatric SP-MPEs are especially aggressive, with much higher rates of local recurrence and secondary seeding to distant craniospinal sites or local spinal sites (64% cf. 32% in adults) [67]. As demonstrated by Ahmad et al. (2021), pediatric SP-MPEs exhibit aberrant activity of the mitochondrial metabolic pathways [68]. The only recurrent focal amplification identified for this group involves HOXB gene cluster mapping to 17q. HOXB13 amplification represents a candidate diagnostic marker for SP-MPEs. The elevated expression of HOXB13 enhances tumor cell proliferation and dissemination, playing a critical role in the development of metastasis [68]. Due to the high propensity for local recurrence and distant neural axis dissemination, the summary of the upcoming WHO CNS5 identifies SP-MPEs with grade 2 (rather than grade 1). SP-MPEs have distinctive histopathological features such as well-organized papillary architecture, with vascular cores and abundant mucinous extracellular matrix. Histological examination of tumor tissue is necessary and sufficient for the diagnosis of SP-MPE, whereas genetic testing is accessory [9].
A rare subtype of Sp-EPN in adult patients has been described recently, presenting highly aggressive clinical behavior with early metastasis, diffuse leptomeningeal spread throughout CNS, and resistance to standard treatment protocols. All of them harbored MYCN amplification and no other recurrent pathogenic events [69][70][71]. Importantly, these tumors formed a distinct methylation cluster of their own, and none of them clustered with any of the previously identified nine EPN groups. Recognizing the importance of clinical and molecular data on such tumors, the summary of the upcoming WHO CNS5 reports a novel nosological entity of MYCN-driven Sp-EPNs with dismal outcomes [9].
Genetic alterations found in particular in EPN groups are summarized in Table 1. Subgroup-specific diagnostic and candidate genes in pediatric EPNs are contained in Table 2.

Molecular Profiles of Subependymomas
Subependymomas are exceptionally rare slow-growing benign neoplasms, accounting for only 0.07-0.7% of all intracranial tumors [14,72]. These tumors typically arise in adults (aged 22-76 years) without strict predilection to a particular site in CNS. The most typical locations, the fourth and lateral ventricles, are encountered in up to 85% of the cases, followed by septum pellucidum, brainstem, and spine [73]. DNA methylation profiles of subependymomas differ depending on tumor localization, which provides certain grounds for molecular subgrouping [14]. However, all subependymomas have a favorable prognosis independently of localization. According to the cIMPACT-NOW group recommendations and the summary of the upcoming WHO CNS5, morphological examination provides adequate means for the diagnosis and prognostication; the integrative histo-molecular approach for subependymomas is accessory [9,16]. Recent findings suggest that the process of clonal evolution in subependymomas may give rise to more aggressive tumor clones enriched with pure EPN phenotypes, chromosome 6 losses, and TERT mutations. These markers, associated with increased risks of recurrence, should be considered as an indication for more intensive therapies, especially under conditions of subtotal tumor resection [44].

Laboratory Approaches for EPN Diagnostics
From a histological perspective, EPNs show moderate cellularity and variable mitotic activity; they consist of monomorphic rounded or oval cells with scant cytoplasm and vesicular nuclei containing granular (salt-and-pepper) chromatin. Key histological features of EPNs are perivascular (pseudo) rosettes and ependymal rosettes. EPNs are positive for glial fibrillary acidic protein (GFAP), S100, vimentin, rarely cytokeratin, and epithelial membrane antigen (EMA, positive along the luminal surface of ependymal rosettes or as characteristic dot-like or ring-like intracytoplasmic patterns) and negative for most neuronal antigens [6].
No unified standards for molecular diagnostics of EPNs have been introduced so far. The molecular group for each particular case is determined by whole-genome DNA methylation profiling as a golden standard. Comprehensive analysis of DNA methylation signatures is the method of choice for differential diagnosis within a broad spectrum of glial and embryonal tumors including CNS neuroblastoma with FOXR2 activation, CNS Ewing sarcoma family tumor with CIC alteration, CNS high-grade neuroepithelial tumor with MN1 alteration, and CNS high-grade neuroepithelial tumor with BCOR alteration [24,74]. The advantage of using this technique is the suitability of formalin-fixed paraffin-embedded tumor tissue for the analysis (methylated sites in genomic DNA are preserved during fixation, dehydration, etc.). However, the use of DNA methylation assays in routine laboratory practice has serious limitations, as the data processing is complex and the method itself is expensive, sophisticated, and labor-consuming. Alternative diagnostic algorithms may be based on a combination of economically justified methods including histological examination, immunohistochemistry (IHC) testing, reverse transcription-polymerase chain reaction (RT-PCR), Sanger sequencing, fluorescence in situ hybridization (FISH), and probably also the NanoString nCounter ® platform ( Figure 2).

Differential Diagnosis of ST-EPNs
In many cases, unambiguous determination of the molecular group of a tumor can be afforded by a certain combination of routine morphological and genetic approaches [75]. For instance, nuclear accumulation of p65 (RelA) and the presence of ZFTA-RELA fusion are sufficient for the identification of the sample with the core ST-EPN-ZFTA group harboring the classical ZFTA-RELA transcript. The diversity of fusion transcripts, which results from the variability of breakpoints in RELA and its partner genes, significantly complicates the identification of particular rearrangements and their use as markers. Conventional RT-PCR tests are targeted at the two most frequent fusions ZFTA (exon 2)-RELA (exon 2) and ZFTA (exon 3)-RELA (exon 2) [19,75,76].
A highly efficient way of screening for ZFTA and RELA rearrangements is provided by FISH with break-apart probes for one of the fusion partner genes. This approach allows detecting rearrangement of the gene of interest without accurate determination of the breakpoint [19,21]. Its excellent concordance with other methods including nuclear expression of RelA and DNA methylation profiling should be noted [19].
ZFTA gene rearrangements induce the hyperexpression of L1 cell adhesion molecule (L1CAM) and/or cyclin D1 may be considered as a surrogate marker of the ST-EPN-ZFTA group [18,20,22,23]. However, IHC is not sufficient on its own: the results require confirmation by independent alternative methods since neither of the antibodies (anti-p65, anti-L1CAM, anti-cyclin D1) has enough sensitivity and/or specificity to reliably verify the ST-EPN-ZFTA group [20,22]. The diagnosis of ST-EPN-ZFTA with alternative (non-RELA) fusion gene can be suspected if tumor cells show immunopositivity for L1CAM and negativity for p65 [26].
Identification of a tumor with the ST-EPN-YAP1 group is based on detection of either YAP1 rearrangement by FISH or one of the YAP1-fusions (YAP1-MAMLD1, YAP1-FAM118B) by RT-PCR, while IHC with anti-YAP1 and/or anti-Claudin-1 is noninformative [19,37].  [77]. The methodology was further enhanced by Fukuoka et al. (2018) who attempted at finer quantitation of the IHC data. In about 62% of prediagnosed PF-EPN-A tumors examined by the authors, the H3K27me3 reactivity was shown by 5-50% of the tumor cell nuclei; for the rest of PF-EPN-A tumors (about 40%), the proportion was less than 5%. In contrast, the vast majority of PF-EPN-B tumors showed intact expression of H3K27me3, with 90-95% of tumor cell nuclei stained positive, except two cases of reduced positivity (10-60%) classified as PF-EPN-B by DNA methylation profiling. Overall, the cut-off level of 80% H3K27me3-positive tumor cells allowed distinguishing PF-EPN-B from PF-EPN-A with 100% specificity and 86.7% sensitivity [21]. Antin et al. (2020) suggested complementing the routine IHC panels for CNS tumor diagnostics with anti-EZHIP staining. The intensive diffuse nuclear staining with >90% tumor cells immunopositive was obtained for PF-EPN-A (with the exception of rare H3 K27M-mutant tumors), diffuse midline gliomas with wild-type (non-mutant) H3F3A, and germinomas [78]. Nambirajan et al. (2021) evaluated an extended panel of IHC markers (H3K27me3, acetyl-H3K27, H3K27M, ATRX, EZHIP, EPOP, and Tenascin-C) for the use in differential diagnostics of PF-EPNs. The authors demonstrated that a combination of EZHIP-and H3K27me3-specific antibodies is sufficient for reliable verification of PF-EPN-A; this finding illustrates the consistency between genetic and IHC profiles of a tumor [79].

Differential Diagnosis of PF-EPNs
Given the existence of rare EPN subtypes e.g., PF-EPN with ZFTA rearrangement [65] or ST-EPN harboring ZFTA fusions with atypical (non-RELA) partners, correct identification of a tumor with a particular group requires an extended diagnostic algorithm accounting for gene expression signatures. The creation of a universal diagnostic tool for the determination of molecular subtypes of EPN can be based on the analysis of gene expression signatures using the NanoString nCounter ® platform. The assay must be carried out with customized panels of markers, corresponding to EPN group-specific signatures of upregulated genes, designed specifically for this purpose. Lastowska et al. (2021) identified group-specific marker genes for RELA-fused EPN (RELA, ELL3, FBP2, PCP4L1, and MYO3A), YAP1-fused EPN (MRAP, IGF1, CAPS, and WWC1), PF-EPN-A (LAMA2, ALDH1L1, SLC6A13, IGSF1, and CXorf67), and PF-EPN-B (NELL2, DNAH1, CEP83, C9orf72, and NXNL2) tumors. The NanoString-based approach clearly separated PF-EPNs into two clusters, corresponding to the defined PF-EPN groups, based on the expression of selected genes. Extended cluster analysis allowed subdivision of PF-EPN-A tumors into PF-EPN-A1 and PF-EPN-A2 subgroups using four marker genes (SKAP2, WIF1, and EN2, CNPY1, respectively). The described gene panel for RELA-positive EPN did not allow differentiation of EPN with ZFTA-MAML2 fusion [80]. The potential of the NanoString method for diagnosis and advanced classification of other CNS tumors has been confirmed in several studies [81][82][83]. The approach provides reliable stratification of medulloblastomas into four groups (designated WNT, SHH, group 3, and group 4) based on the evaluation of 22 transcripts expressed differentially among the groups and contributing to characteristic group-specific gene expression signatures [81,82]. Primitive neuroectodermal tumors of the CNS (CNS-PNETs) have been already defined as four new molecularly defined entities (CNS neuroblastoma with FOXR2 activation, CNS Ewing sarcoma family tumor with CIC alteration, CNS high-grade neuroepithelial tumor with MN1 alteration, and CNS high-grade neuroepithelial tumor with BCOR alteration) using a single multi-gene tumor-specific signature [83]. However, differential diagnostics of EPNs in mixed series with other CNS tumors by the NanoString approach is impossible. For correct molecular stratification of ependymal tumors by the NanoString approach, all samples intended for comparison must be morphologically identified as EPN prior to the analysis.

Therapeutic Targeting of EPNs
Despite the profound insights into ependymal tumor biology, consensus recommendations for the management of patients with EPNs with regard to molecular diagnosis are missing. As tumors are resistant to conventional chemotherapy, the search for drug-gable targets is highly relevant. However, no candidate targets have been identified despite the extensive efforts in molecular profiling. Clinical studies on general cohorts showed no significant objective response to ERBB1/2 (lapatinib) and VEGF inhibitors (bevacizumab, sunitinib), despite the evidence on overexpression of ERBB2, ERBB4, or VEGF in EPNs [84][85][86][87].
In the molecular era, cancer-specific somatic aberrations should be taken into account when choosing a treatment strategy. For PF-EPNs, the pathogenesis of which involves epigenetic mechanisms, the possibility of using pharmaceuticals targeted at epigenetic modifications, including abnormalities of DNA methylation and histone modifications, seems most rational. Despite the promising results of pre-clinical research [88,89], histone deacetylase (HDAC) inhibitors showed no therapeutic activity in patients, apparently due to the intricacies of subcellular localization of HDACs and inability of the inhibitors to accumulate inside the brain tissue in concentrations sufficient for therapeutic response [90]. Nevertheless, the search for new HDAC inhibitors with appropriate brain-penetrating capacities and safety profiles may provide a useful treatment strategy [91]. The possibility of using BET-bromodomain inhibitors as anticancer therapeutics is being investigated in pediatric EPN stem cell models [92]. The upregulation of EZHIP could have important implications for therapeutic approaches. As reported by Han et al. (2020), EZHIP has a conservative PALB2-binding domain which enables its functioning as a competitive inhibitor of BRCA2. Elevated levels of EZHIP prevent the formation of BRCA1-PALB2-BRCA2 complexes thus inhibiting the homologous recombination-mediated DNA repair pathway. The results indicate the potential of PARP inhibitors as targeted therapeutics in PF-EPN-A, especially when combined with radiotherapy [61].
NF-kappa B inhibitors are considered potential therapeutic agents for the treatment of ST-EPN-ZFTA with constitutive activation of the NF-κB signaling pathway. A trial currently in phase II is evaluating the effectiveness of marizomib-a second-generation irreversible proteasome inhibitor, enrolling patients with histologically proven spinal or intracranial EPN, including but not limited to ST-EPN-ZFTA (ClinicalTrials.gov Identifier: NCT03727841, accessed on 20.09.2021). At the moment, the enrollment is complete, and the study is in progress. Specific expression of PD-L1 on both tumor and myeloid cells in ST-EPN-RELA has been demonstrated, accompanied by high levels of PD-1 expressed by tumor-infiltrating T cells (both CD4 and CD8) [93,94]. In the context of immunotherapy, ST-EPN-RELA progression may be controlled with PD-1 inhibitors, such as pembroluzimab or nivolumab [95].
Despite the principal shift in the EPN diagnostics and molecular stratification, its immediate impact on the existing treatment regimens is low. The correction would require preclinical and clinical trials for EPNs with due consideration of the molecular subgrouping.

Conclusions
As demonstrated by advanced studies of the last decade, ependymomas constitute a heterogeneous group of tumors and differ by molecular etiology. This minireview underscores the importance of comprehensive molecular profiling for ependymal tumors aimed at identifying specific expression signatures and/or (epi)genetic variants. Molecular identification of an ependymal tumor with a particular molecular group should follow its anatomical and histopathological assessment. Advanced stratification of patients into risk groups provides a framework for personalized management, e.g., allows de-escalation of the therapy in patients with low-risk tumors (supratentorial ependymomas group YAP1 and infratentorial ependymomas group B). Detailed understanding of causative molecular abnormalities for particular tumors is pivotal for the development of novel therapeutic options.