Molecular subtyping of ependymoma and prognostic impact of Ki-67

Although ependymomas (EPNs) have similar histopathology, they are heterogeneous tumors with diverse immunophenotypes, genetics, epigenetics, and different clinical behavior according to anatomical locations. We reclassified 141 primary EPNs from a single institute with immunohistochemistry (IHC) and next-generation sequencing (NGS). Supratentorial (ST), posterior fossa (PF), and spinal (SP) EPNs comprised 12%, 41%, and 47% of our cohort, respectively. Fusion genes were found only in ST-EPNs except for one SP-EPN with ZFTA-YAP1 fusion, NF2 gene alterations were found in SP-EPNs, but no driver gene was present in PF-EPNs. Surrogate IHC markers revealed high concordance rates between L1CAM and ZFTA-fusion and H3K27me3 loss or EZHIP overexpression was used for PFA-EPNs. The 7% cut-off of Ki-67 was sufficient to classify EPNs into two-tiered grades at all anatomical locations. Multivariate analysis also delineated that a Ki-67 index was the only independent prognostic factor in both overall and progression-free survivals. The gain of chromosome 1q and CDKN2A/2B deletion were associated with poor outcomes, such as multiple recurrences or extracranial metastases. In this study, we propose a cost-effective schematic diagnostic flow of EPNs by the anatomical location, three biomarkers (L1CAM, H3K27me3, and EZHIP), and a cut-off of a 7% Ki-67 labeling index. Supplementary Information The online version contains supplementary material available at 10.1007/s10014-021-00417-y.


Introduction
Ependymomas (EPNs) are uncommon neuroepithelial malignancies, constituting approximately 2% of the central nervous system (CNS) tumors and about 6.8% of all gliomas [1]. They can occur at any age and are the third most common CNS tumors in children [2]. 2016 World Health Organization (WHO) classification subdivided EPNs mostly based on histology and included only one genetically defined EPN subtype, EPN, RELA fusion-positive [3], and it has been noted that the WHO grades do not reflect biological behavior and patient outcomes [4]. RELA fusion-positive EPN has been revised into ZFTA fusionpositive in the 2021 WHO classification [5].
The cIMPACT-NOW update 7 and the 2021 WHO classification of tumors of the CNS classify EPNs by a combination of anatomic site, molecular genetics, epigenetics, and histological features [6,7]. According to these updates, anatomic sites, e.g., ST, PF, and SP, should add to the diagnostic term, EPN. However, the criteria for the grades are not clearly defined in the WHO classification. In general, grade 3 EPN is an infiltrative, highly proliferative tumor with active mitosis and microvascular proliferation, but necrosis and nuclear polymorphisms may be present in both grades of EPN and are, therefore, not the parameters of EPN grade. As we already know, nuclear polymorphism itself is very rare in both grades of EPNs. Also, the mitotic rate is not precisely defined. In 2016 WHO blue book mentioned that the interpretation of the most histopathological variables in EPN grading is subjective, and multivariate analyses of prognostic variables are limited [8]. Therefore, histology-or molecularly defined EPN grades still have pitfalls as there are no clear criteria established for WHO grades 2 and 3.
ZFTA (zinc 119 finger translocation associated, previous term, c11orf95) fusion-positive ST-EPN is considered a grade 3 associated with poor outcomes. YAP1 (yes associated protein 1) fusion-positive ST-EPN [6,7,9] is associated with a good prognosis and grade 2. For the past 5 years, neuropathologists have been applying 2016 WHO classification to the diagnosis of CNS tumors; thus, it is necessary to add EPN-RELA fusion-positive to ST-EPN-ZFTA fusion-positive. PF-EPN is characterized by the absence of somatic mutation of genes. There are two groups of PF-EPN; group A (PFA) and group B (PFB). PFA-EPN commonly occurs in children, exhibits aggressive behavior, is considered to be of grade 3, and is characterized by loss of H3K27me3 and overexpression of Zest 2 Polycom Inhibited Complex (EZHIP) inhibitory proteins [10,11]. EPNs of this subtype show increased CpG methylation at CpG islands, demonstrating a CpG island methylator phenotype (PFA-CIMP + ) [11]. PFB-EPN has no genetic or epigenetic abnormalities, mostly in adults, and is classified as grade 2 because of a good prognosis. In most SP-EPNs, the NF2 gene is inactivated by mutation or deletion and the new subtype with MYCN amplification is associated with a dismal prognosis [12].
However, EPNs grading based on histological parameters is required when molecular studies fail to reveal reliable results or when a molecular genetic study is unavailable. In addition, some exceptional cases of grade mismatch with molecular genetic subtypes such as WHO grade 2 PFA or grade 3 PFB, and grade 2 ZFTA-RELA fusion-positive have been reported [4,7,13]. cIMPACT-NOW update 7 explained that although molecular properties are closely related to prognosis, molecularly defined EPN grades are not mature enough to grade EPNs [7]. In addition, the histological grade is still important because the 5-year overall survival (OS) of grade 3 SP-EPN in a recent large cohort was significantly worse than that of grade 2 SP-EPN [14].
A Ki-67 index greater than 5%, 7%, or 10% has been reported to be significantly associated with poor prognosis [15][16][17][18]. It is also important to determine the treatment options after surgery. However, there has been no consensus on the cut-off in previously published studies.
Here, we reclassified our series of EPNs for 20 years by anatomical site, genetic, and epigenetic classifiers according to the cIMPACT-NOW update 7 and 2021 WHO classification [6,7]. We also studied the association of copy number variation (CNV) with clinical outcomes. We propose a cost-effective schematic diagnostic flow, easy to apply in clinical and pathological practice with a 7% cut-off for the Ki-67 index.
Informed consent for the genetic study of EPNs was obtained from the patients or parents of the children. The institutional review board of our hospital approved this study (Approval No. 2012-034-1179), which was conducted under the ethical standards outlined in the 1964 Declaration of Helsinki and subsequent amendments.

Pathology and immunohistochemistry (IHC)
All tumors were reviewed by two pathologists (KY Lim and SH Park). Histological features including histologic subtype, mitotic rate, necrosis, and microvascular proliferation were evaluated. IHC staining was performed on 3-µm-thick formalin-fixed and paraffin-embedded (FFPE) tissues using an automated immunostaining system (BenchMark ULTRA system; Ventana-Roche, Mannheim, Germany). The primary antibodies used in this study are summarized in Supplementary Table 1.
The Ki-67 labeling indices were calculated the positive nuclei per 500 tumor cells by SpectrumPlus Aperio morphometric algorithm in the several hot spots, and the highest index was determined. The mitotic rate was counted in 10 HPFs of the hot spot area with pHH3 IHC.
IHC readings were semi-quantitatively performed in intensity and distribution. Null, ambiguous, weak, moderate, and strongly positive were scored as 0, 1, 2, 3, and 4, respectively. In our cases, H3K27me3 loss was completely negative in tumor cells, EZHIP overexpressing cases showed diffuse strong positive, and EZHIP negative cases were either completely negative or a few positive cells.
The concordance rates of L1CAM and NF-kB with ST-EPNs were analyzed and both H3K27me3 loss and EZHIP overexpression were used for the diagnosis of PFA-EPNs. We also determined an adequate cut-off of the Ki-67 labeling index to differentiate grade 2 from grade 3, and compared it with the genetic subgroups and clinical outcomes.

DNA and RNA extraction and genetic studies by NGS
Representative areas of the tumors with more than 90% tumor cell content were outlined on the FFPE sections for macrodissection. DNA and RNA extractions were performed from serial sections using the Maxwell ® RSC DNA FFPE Kit (Promega, Madison, WI), according to the manufacturer's instructions.
NGS was performed in 60 cases of EPNs (16 ST-, 23 PF-, and 21 SP-EPNs) with the NextSeq™ 550 system (Illumina, San Diego, CA) using a customized brain tumor gene panel (FIRST Brain Tumor Panel of SNUH), which contains 207 genes including MYCN, and 54 fusion genes including ZFTA, RELA, and YAP1. This gene was approved by the Korean Ministry of Food and Drug Safety.
Sequencing data were analyzed using the SNUH FIRST Brain Tumor Panel Analysis pipeline. The quality control of the fastq file was performed and only the data that met the criteria were analyzed. Paired-end alignment to the hg19 reference genome was performed using BWA-mem and GATK Best Practice [19]. Second quality control was performed to ensure that additional variant calls were appropriate. The open-source tools used were GATK UnifedGenotyper, SNVer, and LoFreq for SNV/InDel detection [20], Delly and Manta for translocation discovery [21], THetA2 for purity estimation, and CNVKit for CNV calling [22][23][24][25]. SnpEf annotated the variants detected from various databases, such as RefSeq, COSMIC, dbSNP, ClinVar, OncoKB, and gnomAD. The germline variant was then filtered using the population frequency of these databases (> 0.01%). Finally, the variants were confirmed through a comprehensive review of a multidisciplinary molecular tumor board. Whole exome sequencing was carried out in 5 cases of ependymomas, which were evaluated with similar pipelines of NGS.

Statistical analysis
The survival rate of patients according to clinical, pathological, and genetic factors was analyzed using Kaplan-Meier survival analysis and log-rank test. Cox regression analysis was used in univariate and multivariate analysis to study the prognostic impacts of clinical and histopathological parameters. First, the influence of each variable was evaluated, and P ≤ 0.05 was interpreted as statistical significance. Second, the prognostic effect and independence were evaluated by examining only the variables with P ≤ 0.05 in the multivariate analysis. All statistical analyses were performed using SPSS 25 (IBM, Armonk, NY), R version 3.5.3 (R Foundation, Vienna, Austria), and survminer 0.4.6 packages. In 66 cases of SP-EPNs, grade 2 and grade 3 were 92% and 8%, respectively and 89% were adults and 11% were children. Grade 3 SP-EPNs occurred in young patients (median 26 years; range 5-49 years, child: adult = 4: 6), whereas grade 2 EPNs occurred in a broad age range of patients (median 49 years; range 8-73 years). The cervical spinal cord was the most frequently affected site by grade 2 SP-EPNs (74%, 45/61).

Epidemiology and subgroup of ST-, PF-and SP-EPN
NF2 alterations were observed in 79% (16/21) of SP-EPNs that had undergone NGS, with exception of three known NF2 patients. Three NF2 patients presented with multiple benign tumors, including meningiomas or schwannomas in addition to WHO grade 2 EPN. Our cohort did not have MYCN amplified SP-EPNs.
NF2 alteration was found in 60% (3/5) of WHO grade 3 SP-EPNs; two pathogenic mutations of NF2 (splicing, c.1122 + 1G > A, and splicing, c.599 + 1G > A), and one NF2 gene deletion. The remained one had ZFTA-YAP1 fusion, which was just reported [36]. The other one had a balanced chromosome with no mutations (Supplementary Table 2). Thus, NF2 status was not related to histological grade or biological behavior.  Table 3).

The results of the immunohistochemical study
All PFB-EPNs were negative for both H3K27me3 and EZHIP (Fig. 2). There was no case of immunoreactive for K27M (antibody for histone Lys27Met). The exceptional EZHIP-negative PFA-EPN, which was reconfirmed as EPN by electron microscopy, showed characteristic ependymal features such as intracytoplasmic and intercellular microrosettes with microvilli and cilia as well as zonula adherens, but methylation study could not be performed due to poor DNA quality. It was WHO grade 3 occurred in the fourth ventricle of an adult (38-year-old).

CNV analysis by NGS was performed (16 ST-EPNs, 23 PF-EPNs, and 21 SP-EPNs) (Supplementary
Monosomy 6 was found in one case of ST-EPN-ZFTA, two PFA, and one PFB. One PFA-EPN with monosomy 6 had multiple recurrences and extracranial metastasis. Monosomy 11 was only observed in ST-EPN-ZFTA and there was no recurrence during the 3-year follow-up period. Chromosome 1p loss was observed in one ST-EPN-ZFTA, which also presented extracranial metastasis. A balanced chromosomal profile (n = 12) was observed at all anatomical locations, including five ST-EPN-ZFTAs, one YAP1-MAMLD1 fusion-positive ST-EPN, three PFA-EPNs, and three SP-EPNs. Among them, two cases showed intracranial recurrent EPN (2/9, 22%) and one case showed extracranial metastasis. Interestingly, an SP-EPN with a balanced copy number and ZFTA-YAP1 fusion recurred 8 months after GTR which was clustered with ZFTA-RELA fusion-positive ST-EPN by the methylation profile [36].
Multiple CNV was observed in two PFB-EPNs with no recurrence.

Kaplan Meier survival analysis according to Ki-67 labeling indices, H3K27me3 loss, and EZHIP overexpression
The Kaplan-Meier survival analysis according to grade, the prognosis was worse in grade 3 than grade 2, but did not reach the statistical significance in ST-, and SP-EPNs (Fig. 5A, B and 5E, F) because the limitation of the number of WHO grade 2 ST-EPNs (6%) and WHO grade 3 SP-EPNs (8%). Both OS and PFS of PF-EPNs were strongly associated with the grades based on 7% Ki-67 (P = 0.021 and 0.001, respectively) (Fig. 5C, D).
Loss of H3K27me3 (P = 0.007) and overexpression of EZHIP (P = 0.002) were significantly associated with a poor prognosis. H3K27me3 and EZHIP were statistically significantly associated by Fisher's exact test (P < 0.0001).

A precise and easily applicable schematic diagnostic flow in clinical practice
In our study, L1CAM and H3K27me3 (and/or EZHIP) IHC and Ki67 labeling indices were sufficient to classify intracranial EPNs, which could be both practical and costeffective. H3K27me3 loss should be considered before EZHIP because the global reduction of H3K27me3 is the main mechanism of PFA-EPNs. For SP-EPNs, most had NF2 gene deletion, monosomy 22, or rarely an NF2 genesplicing mutation but these genetic changes did not affect the patient's outcomes. The Ki-67 index was shown to be a powerful prognostic factor in multivariate analysis, and a lower than 7% cut-off was a good indicator for grades 2 EPNs. We proposed a schematic diagnostic flow of EPNs, which can be a useful, accurate, and cost-effective diagnosis in clinical and pathological practice (Fig. 6).

Discussion
We successfully reclassified our 141 EPNs using three biomarkers of L1CAM, H3K27me3, and EZHIP, and Ki-67. Reclassified ependymomas showed a high correlation with pathological grade, patients' age, and biological behavior. It was found that 70% of ST-EPN-ZFTA and almost all PFA-EPN occurred in children, but almost all PFB and 92% of low-grade SP-EPN were adults. Fusion genes were found in ST-EPN, except for one spinal EPN, which had ZFTA-YAP1 fusion [36]. L1CAM was a practically excellent surrogate marker in predicting ZFTA-fusion, and H3K27me3 loss or EZHIP overexpression were excellent biomarkers to identify PFA-EPN. Since clear cell PFA-EPN showed only H3K27me3 loss, H3K27me3 loss may be a better biomarker than EZHIP for PFA-EPN diagnosis.
In SP-EPNs, NF2 mutation or deletion was the underlying genetic abnormality, but it did not affect biological behavior and did not have a role of surrogate marker for grading. 1q25 gain, monosomy 6, and CDKN2A/2B homozygous deletion were related to poor outcomes including multiple recurrences and extracranial metastases, which was similar to previous reports [37,38]. The prognostic and predictive effects of histological parameters of EPN, including histological grade, are known to be controversial. According to Figarella-Branger et al., STR, loss of histological differentiation, high Ki-67 index (≥ 1%), and age younger than 4 years were associated with inferior PFS [39]. In one multivariate analysis by Horn et al., age (age < 3 vs. age ≥ 3), histologic grade (grade 2 vs. grade 3), and extent of resection affected OS and PFS [40].
In our study, the multivariate analysis showed that Ki-67 was the only independent prognostic factor in both PFS and OS, which can be supported by previous reports. Malgulwar et al.'s study revealed that L1CAM positivity and high Ki-67 labeling indices were associated with poor PFS, but the age and the extent of resection were not significant for survival [15].
In our study, the Ki-67 index (cut-off of 7%) was well correlated with the WHO grades and PFS of our series of EPNs in all anatomical locations. However, our cohort of EPNs did not show statistical significance in OS because our cohort has limitations in patients' number and follow-up period. In addition, the patients in our cohort survived longer, the 5-year survival rates of higher grade (WHO grade 3) EPNs were 93.8%, 73.2%, and 93.8% in ST-EPN-ZFTA, PFA, and SP-EPN-NF2, respectively, and the 10-year survival rates were 81.3%, 70.7%, and 87.5%, respectively, suggesting that EPNs are chronic diseases, despite multiple local recurrence and distant metastases.
Although genetic and epigenetic subtypes were highly associated with the prognosis in intracranial EPNs, the previous studies reported that they do not necessarily determine the grades because there are some exceptional cases, such as grade 2 PFA with good prognosis, and grade 3 PFB and ST-EPN-YAP1 with poor outcomes. [13,41] Reproducible cut-offs for Ki-67 labeling indices have been proposed in astrocytic and oligodendroglial tumors to predict the outcomes [42][43][44][45]. Similarly, our meta-analysis with 11 published papers showed that increased Ki-67 labeling indices were strongly correlated with poor OS and PFS of EPNs (both P < 0.0001).
Significant heterogeneity of staining techniques, antibody clones, and counting methods may affect cut-off. However, we used the same thickness of tissue sections and the same immunostaining method, and AperioSpectrum plus image analyzer to count all cases to maintain the same conditions and to minimize the inter-observer variability.
In our study, we found area-to-area differences of Ki-67 labeling indices, and grading based on the 7% cut-off in the hot spot area showed a strong association with genetic subtypes of EPNs and clinical outcomes. We conclude that a Ki-67 index of 7% is the most reliable cut-off for grading in all anatomical locations. Our cut-off is also supported by the study of Zamecnik et al. in which a Ki-67 labeling index more than and equal to 7% was associated with an inferior outcome in both OS and PFS of EPN [16].
In addition, our results showed that chromosome 1q25 gain and CDKN2A/2B loss were significant poor prognostic factors indicating a dismal prognosis, such as multiple recurrence or extracranial metastasis. The monosomy 6 and balanced chromosomal profile were the next worse prognostic factor, also indicating multiple recurrence and extracranial metastasis. However, multiple CNVs without the aforementioned CNVs were associated with favorable outcomes.