Advertisement

Acta Neuropathologica

, Volume 136, Issue 1, pp 153–166 | Cite as

Novel, improved grading system(s) for IDH-mutant astrocytic gliomas

  • Mitsuaki Shirahata
  • Takahiro Ono
  • Damian Stichel
  • Daniel Schrimpf
  • David E. Reuss
  • Felix Sahm
  • Christian Koelsche
  • Annika Wefers
  • Annekathrin Reinhardt
  • Kristin Huang
  • Philipp Sievers
  • Hiroaki Shimizu
  • Hiroshi Nanjo
  • Yusuke Kobayashi
  • Yohei Miyake
  • Tomonari Suzuki
  • Jun-ichi Adachi
  • Kazuhiko Mishima
  • Atsushi Sasaki
  • Ryo Nishikawa
  • Melanie Bewerunge-Hudler
  • Marina Ryzhova
  • Oksana Absalyamova
  • Andrey Golanov
  • Peter Sinn
  • Michael Platten
  • Christine Jungk
  • Frank Winkler
  • Antje Wick
  • Daniel Hänggi
  • Andreas Unterberg
  • Stefan M. Pfister
  • David T. W. Jones
  • Martin van den Bent
  • Monika Hegi
  • Pim French
  • Brigitta G. Baumert
  • Roger Stupp
  • Thierry Gorlia
  • Michael Weller
  • David Capper
  • Andrey Korshunov
  • Christel Herold-Mende
  • Wolfgang Wick
  • David N. Louis
  • Andreas von DeimlingEmail author
Original Paper

Abstract

According to the 2016 World Health Organization Classification of Tumors of the Central Nervous System (2016 CNS WHO), IDH-mutant astrocytic gliomas comprised WHO grade II diffuse astrocytoma, IDH-mutant (AIIIDHmut), WHO grade III anaplastic astrocytoma, IDH-mutant (AAIIIIDHmut), and WHO grade IV glioblastoma, IDH-mutant (GBMIDHmut). Notably, IDH gene status has been made the major criterion for classification while the manner of grading has remained unchanged: it is based on histological criteria that arose from studies which antedated knowledge of the importance of IDH status in diffuse astrocytic tumor prognostic assessment. Several studies have now demonstrated that the anticipated differences in survival between the newly defined AIIIDHmut and AAIIIIDHmut have lost their significance. In contrast, GBMIDHmut still exhibits a significantly worse outcome than its lower grade IDH-mutant counterparts. To address the problem of establishing prognostically significant grading for IDH-mutant astrocytic gliomas in the IDH era, we undertook a comprehensive study that included assessment of histological and genetic approaches to prognosis in these tumors. A discovery cohort of 211 IDH-mutant astrocytic gliomas with an extended observation was subjected to histological review, image analysis, and DNA methylation studies. Tumor group-specific methylation profiles and copy number variation (CNV) profiles were established for all gliomas. Algorithms for automated CNV analysis were developed. All tumors exhibiting 1p/19q codeletion were excluded from the series. We developed algorithms for grading, based on molecular, morphological and clinical data. Performance of these algorithms was compared with that of WHO grading. Three independent cohorts of 108, 154 and 224 IDH-mutant astrocytic gliomas were used to validate this approach. In the discovery cohort several molecular and clinical parameters were of prognostic relevance. Most relevant for overall survival (OS) was CDKN2A/B homozygous deletion. Other parameters with major influence were necrosis and the total number of CNV. Proliferation as assessed by mitotic count, which is a key parameter in 2016 CNS WHO grading, was of only minor influence. Employing the parameters most relevant for OS in our discovery set, we developed two models for grading these tumors. These models performed significantly better than WHO grading in both the discovery and the validation sets. Our novel algorithms for grading IDH-mutant astrocytic gliomas overcome the challenges caused by introduction of IDH status into the WHO classification of diffuse astrocytic tumors. We propose that these revised approaches be used for grading of these tumors and incorporated into future WHO criteria.

Keywords

Astrocytoma Glioblastoma IDH Grading CDKN2A/B 

Introduction

Diffuse astrocytic gliomas are the most common brain tumors of adults and their grading has, therefore, been of considerable significance in the management of patients with brain tumors. As a result, over the years, these tumors have been graded according to various systems, each representing an advance over prior approaches. These have included the Kernohan scheme, the St. Anne-Mayo system and, for the past 25 years, primarily the WHO classification [8, 12, 13]. These schemes often resulted in patient groups with significantly varying prognoses that received treatments of differing intensity. In recent years, the WHO classification scheme based on early work from Zulch [38] and regarding astrocytic gliomas from Burger et al. [3] has been most widely used [13, 14].

IDH1 and IDH2 mutations have emerged as early mutations in diffuse astrocytic tumors [2, 21, 35, 37], and determination of the IDH status by immunohistochemistry [6] or sequencing has now become a standard in the diagnosis of these tumors. The 2016 WHO Classification of Tumors of the Central Nervous System [14] introduced molecular parameters for the categorization of diffuse astrocytomas, notably IDH mutation status [14]. In fact, determination of IDH and 1p/19q status is mandatory for diagnosing beyond NOS (not otherwise specified) categories. However, applying the new classification parameters results in patient groups that are different from those classified prior to the 2016 CNS WHO (e.g. according to the 2007 CNS WHO). From the former group of diffuse WHO grade II and III astrocytomas, approximately 70–80% now are classified as diffuse astrocytoma, IDH-mutant (AIIIDHmut) and anaplastic astrocytoma, IDH-mutant (AAIIIIDHmut) while 20–30% are classified as diffuse astrocytoma, IDH-wildtype (AIIIDHwt) and anaplastic astrocytoma, IDH-wildtype (AAIIIIDHwt). In turn, the shifting of diagnostic groups has prognostic implications. For example, multiple studies have demonstrated a close relationship between many AIIIDHwt/AAIIIIDHwt to glioblastoma, IDH-wildtype (GBMIDHwt) [4, 27, 33]. In fact, prior to the 2016 CNS WHO the prognostic power of grading parameters for diffuse astrocytoma WHO grade II (AIINOS) and anaplastic astrocytoma WHO grade III (AAIIINOS) was most likely a result of considerable contamination with unrecognized GBMIDHwt [34]. In addition, a series of recent studies has highlighted that WHO grading of AIIIDHmut/AAIIIIDHmut has lost its prognostic relevance [4, 18, 20, 22, 28, 33, 34]. Given the clinical importance of appropriate grading, the potential prognostic changes brought about by the 2016 CNS WHO classification pose a major challenge. We, therefore, sought to establish novel, improved grading criteria for IDH-mutant astrocytic gliomas.

Methods

Study design and participants

For this multicenter retrospective analysis we collected a discovery series of 211 astrocytic tumors. We defined the following criteria for inclusion: (1) all tumors had either an IDH1 or an IDH2 mutation; (2) all tumors had a 2016 CNS WHO integrated diagnosis of AIIIDHmut (n = 54), AAIIIIDHmut (n = 90) or GBMIDHmut (n = 67), i.e., no tumors had 1p/19q codeletion; (3) sufficient formalin-fixed, paraffin-embedded (FFPE) material for morphological and additional molecular analyses was available; and (4) the patients had an extended follow-up. Tumor material was from the archive of the Department of Neuropathology, University of Heidelberg, Heidelberg, Germany and from patients enrolled in the NOA04 trial [36]. Use of tissue and clinical data were in accordance with local ethical regulations.

Three validation series were compiled. The first validation series from Heidelberg (HD) containing 108 tumors differed from the discovery set only in that sufficient material for the complete set of analyses performed on the discovery set was not available. The HD validation series contained AIIIDHmut (n = 32), AAIIIIDHmut (n = 29) and GBMIDHmut (n = 47). In this set diagnosis and grading according to the guidelines of the 2016 CNS WHO was performed (AvD). The second validation set from the European Organization for Research and Treatment of Cancer (EORTC) contained 154 cases and was compiled from patients of the EORTC studies 22,033 (64 cases), 26091 (80 cases) and 26,981 (10 cases) with Illumina 450k or 850k/EPIC data being available. All patients in this set also had an IDH mutation and none of the tumors harbored a 1p/19q codeletion. The EORTC validation series contained AIIIDHmut (n = 94), AAIIIIDHmut (n = 39) and GBMIDHmut (n = 21). In this set diagnosis and grading according to the guidelines of the 2016 CNS WHO was performed by reference centers. The third validation set from The Cancer Genome Atlas consortium (TCGA) contained 224 cases and was retrieved from published data. In this set diagnosis and grading was obtained from the source. Methylome data based on Illumina 450k array was available for all of these cases. All cases in this set had an IDH mutation as provided by TCGA and none contained a 1p/19q codeletion as assessed by copy number profiles which were calculated from the DNA methylation data. Clinical data were known to the local investigators (MS, TO, DSt, DSc, AvD). Cases from the HD validation set also were reviewed by the same neuropathologist (AvD). Clinical data from the HD validation set were known to the lead authors. Clinical data from the EORTC validation set were not known to the lead authors. Clinical data for the TCGA validation set were open access. The institutional funding of this study did not influence study design, inclusion criteria, analyses or interpretation of the data. MS, TO, DSt, DSc and AvD had full access to all raw data except for the EORTC series and had final responsibility for the decision to submit for publication.

Procedures

For the discovery set, 4 µm sections were cut from FFPE blocks from all tumors and stained with hematoxylin/eosin, and examined by immunohistochemistry with antibodies against IDH1 R132H (clone H09, Dianova, Hamburg, Germany), Ki-67 (clone MIB-1, Dako, Waldbronn, Germany) or pHH3 (rabbit polyclonal, BioCare Medical, Pacheco, USA) to determine proliferation and CD31 (clone JC70A, Dako, Waldbronn, Germany) to determine vascular density.

All tumor samples were analyzed by Illumina HumanMethylation450 (450k) or MethylationEPIC (850k) arrays (Illumina, San Diego, CA, USA) as previously described [32]. Methylation data were analyzed by a classifier as previously reported [5]. CNV plots were generated using the R/Bioconductor package conumee version 1.6.0. Automated assessment of copy-number changes was performed using a proprietary algorithm (Stichel D and colleagues, unpublished). This allowed determining CNV load (CNV-L) for each tumor.

The discovery set was subjected to digital image analysis. Parameters assessed included cell count, proliferation activity and microvessel density. The software package Aperio ImageScope was employed for proliferation and microvessel density analysis. The software packages Ilastik [31] and ImageJ [30] were used for cell count analysis. To receive subgroups with distinct OS, conditional trees were created applying the function ctree from the R package party on the discovery set and a set of selected input variables. Here, the global null hypothesis of independence between any of the input variables and the OS was tested. If it could not be rejected, the association of any single input variable with the response was computed as p value from a test for the partial null hypothesis of the variable and OS. The input variable with the highest association to OS was selected and a binary split in this variable was created. This procedure was then repeated to create conditional trees. To create Modelpath all parameters received from molecular analyses but CDKN2A/B status were excluded. Kaplan–Meier estimators were computed using the function survfit from the R package survival. All parameters assessed are listed in Table 1.
Table 1

Morphological and clinical features and their association with OS in the discovery set

Parameter

Min

Max

Split

n

Rate

p value

Age (years)

15

70

54

190

< 54; 92%

< 0.05

Gender male

0

1

na

130

m; 62%

0.01

Cell count (n/mm2)

734

9143

4604

130

< 4604; 62%

< 0.002

Ki67 (%)

0

97

14.5

144

< 14.5; 71%

< 0.01

pHH3 count (n/10HPF)

0

327

No split

   

Microvessel density (%)

0.2

8.7

1.2

82

> 1.2; 39%

< 0.05

Necrosis absent

0

1

na

144

Present; 32%

< 0.000005

Vascular proliferation absent

0

1

na

129

Absent; 61%

< 0.001

Numerical alterations of several genes with established relevance for astrocytic gliomas—comprising CCND1, CCND2, CDK4, CDK6, CDKN2A/B, EGFR, MDM4, MET, MYC, MYCN, NF1, NF2, PDGFR2, PPM1D, PTEN, RB1 and SMARCB1—were evaluated by visual assessment of copy number profiles and by employing our proprietary algorithm. Our series are based on visual assessment for CDKN2A/B analysis. With the cutoff selected in our series, visual and algorithm driven CDKN2A/B evaluation produced the same number of cases scored as having a homozygous deletion. The respective genes and cutoff values are given in Table 2.
Table 2

Genes with amplifications or homozygous deletions found in 211 IDH-mutated astrocytomas in the discovery set and their association with OS

Gene

Alteration

n

p value

CCND1

Amplification (cutoff 0.35)

1

0.17

CCND2

Amplification (cutoff 0.35)

27

0.15

CDK4

Amplification (cutoff 0.35)

18

0.11

CDK6

Amplification (cutoff 0.35)

6

0.13

CDKN2A/B

Homo del (cutoff − 0.415)

38

0.0001

EGFR

Amplification (cutoff 0.35)

4

0.38

MDM4

Amplification (cutoff 0.35)

5

0.09

MET

Amplification (cutoff 0.35)

11

0.39

MYC

Amplification (cutoff 0.35)

13

0.89

MYCN

Amplification (cutoff 0.35)

12

0.001

NF1

Homo del (cutoff − 0.415)

4

0.52

NF2

Homo del (cutoff − 0.415)

4

0.32

PDGFRA

Amplification (cutoff 0.35)

13

0.03

PPM1D

Amplification (cutoff 0.35)

1

0.70

PTEN

Homo del (cutoff − 0.415)

3

0.11

RB1

Homo del (cutoff − 0.415)

12

0.001

SMARCB1

Homo del (cutoff − 0.415)

2

0.26

Univariate analysis

homo del homozygous deletion

Statistical analysis

All patient sets were retrospectively compiled. The size of the respective sets was determined by availability of data and not by a power calculation. OS times were analyzed by the Kaplan–Meier method and compared with a log-rank test. Prediction error plots were based on Brier scores. p values less than 0.05 were considered significant. Software R version 3.4 and packages survival and party were employed for analysis.

Results

Series validation and determination of histologic and molecular features associated with OS

On the basis of histological review, all tumors in the discovery set were confirmed to be diffuse astrocytic gliomas. Classification according to the 2016 CNS WHO revealed three distinct groups, AIIIDHmut, AAIIIIDHmut or GBMIDHmut, with significantly different OS (Fig. 1a). The IDH status was determined by IHC with H09 [6] or by Sanger sequencing. All immunonegative samples have been subjected to sequencing. IDH status was also confirmed by performing 450k/EPIC analysis and receiving the readout of “methylation group astrocytoma IDH mutant” by a recently developed and published classifier tool [5] which predicts IDH-mutant astrocytoma with very high specificity and sensitivity. Further, combined 1p/19q deletion was excluded for every tumor based on analysis of the copy number profile (CNP) generated from 450k/EPIC data. Therefore, our series included exclusively AIIIDHmut, AAIIIIDHmut or GBMIDHmut, in full agreement with 2016 CNS WHO classification criteria. In the discovery set of 211 patients, 135 were still alive at the time of last follow-up. Median follow-up in these 135 patients was 1772 days (4.9 years); average follow was 2086 days (5.7 years).
Fig. 1

Kaplan–Meier plot stratifying according to WHO in the discovery set (a). b The same patients with additional stratification for homozygous deletion of CDKN2A/B

The morphological parameters of strongest negative prognostic value were vascular proliferation (p < 0.0005) and necrosis (p < 0.00005). Patients aged 55 and older did worse (p < 0.04) than patients aged below this cutoff confirming the influence of age on OS. Female patients fared significantly (p = 0.01) better than male patients. Image analysis showed that cellularity had a major association with OS. Cell count defined as number of tumor cells per square millimeter in the tumor area of highest density emerged as an important parameter for prognostic evaluation. Patients with a cell density of 4605/mm2 or more fared worse than patients with 4604 or fewer tumor cells/mm2 (p < 0.002). Ki-67 immunohistochemistry was significantly associated with worse OS (cutoff 14.5%; p < 0.006), however, mitotic count established by assessing sections treated with pHH3 antibody was not prognostic (Table 1). While pHH3 has been established as a reliable marker for assessing mitotic activity, those studies have mainly been performed prior to further stratification of diffuse astrocytomas by IDH status [7]. In a recent study on IDH-mutant diffuse glioma, pHH3 was found to reliably detect mitotic figures but turned out with a weaker association with survival than Ki-67 [10].

EPIC- or 450k array data were employed to assess several parameters: this included analysis by the brain tumor classifier tool [5] for the confirmation of IDH mutations and evaluation of individual copy number status. Copy number plots were evaluated for amplifications or homozygous deletions, respectively. In the discovery set regions with such alterations contained CCND1, CCND2, CDK4, CDK6, CDKN2A/B, EGFR, MDM4, MET, MYC, MYCN, NF1, NF2, PDGFR2, PPM1D, PTEN, RB1 and SMARCB1 (Table 2). Upon univariate analysis, the strongest association with OS was observed for homozygous CDKN2A/B deletion and for MYCN amplification (Table 2 and Fig. 1b). Furthermore, the total CNV-load (CNVL) was determined, resulting in a split for a value of 349,695,798 base pairs (rounded 350 Mb). This number refers to the sum of all gains or deletions as determined by analysis of the 450k/850k raw data by our proprietary algorithm.

Patients with higher CNVL fared significantly worse than patients with lower CNVL (p < 0.0001). However, the frequency for all chromosomal gains or losses increased with age as demonstrated by summary CNV plots (supplementary figure 1). Further, EPIC/450k array methylation data were used for unsupervised clustering of the discovery dataset. This analysis yielded two sets with highly significant differences in OS (supplementary figure 2a). Likewise, the algorithm underlying our methylation based classifier [5] recognized two sets with significantly different OS (supplementary figure 2d).

Novel grading algorithms based on discovery set

Multivariate analysis including all parameters exhibiting a significant association with OS (Table 1) was performed to develop grading models that could be compared with the current 2016 CNS WHO standard. Three different models were generated. The first approach (Modelpath) aimed at minimizing the required molecular data input, which would presumably allow for the most widespread applicability. In addition to IDH and 1p/19q testing, the first approach only requires the determination of CDKN2A/B status. The second approach, ModelCNVL made use of all molecular data available. Because ModelCNVL in the discovery set placed a set of seven patients with necrosis into the most favorable group, we created a Modelcombined, which shifted these patients to the patient group with intermediate OS (Fig. 2). While this model was not based on a mathematical procedure, it worked well in the discovery and all validation sets.
Fig. 2

Distribution of 211 patients in the discovery set according to WHO grading and to three different grading approaches. N+: necrosis present, N−: necrosis absent, P+: CDKN2A/B is homozygously deleted, P−: CDKN2A/B is not homozygously deleted, C+: high CNV-load, C−: low CNV-load. Parameter P+ is not encountered in AIIIDHmut and, therefore, this potential tumor subset is not depicted in the three alternative grading approaches. Top: grading model applied. Middle: placement of patient groups with distinct features according to the four different approaches. Bottom: OS of patients according to the four approaches

Modelpath demonstrated that an algorithm based on employment of absence/presence of homozygous CDKN2A/B deletion and absence/presence of necrosis formed three groups among the 211 tumors in the discovery cohort. In this Modelpath, patients with homozygous deletion of CDKN2A/B exhibited the worst OS and were termed astrocytoma, IDH-mutant, grade 4, pathology (A4path). This group comprised 38 tumors of the discovery cohort and contained 23 GBMIDHmut and 15 AAIIIIDHmut. Tumors with necrosis but without homozygous CDKN2A/B deletion fared significantly better and were termed astrocytoma, IDH-mutant, grade 3, pathology (A3path). A3path contained 44 tumors which by definition were all GBMIDHmut. The remaining 129 patients with neither homozygous CDKN2A/B deletion nor necrosis were termed astrocytoma, IDH-mutant, grade 2, pathology (A2path). A2path contained 75 AAIIIIDHmut and 54 AIIIDHmut. OS plots of the discovery set assembled according to 2016 CNS WHO and according to Modelpath are shown (Fig. 3a, b). Modelpath performed better in predicting OS of the groups than the 2016 CNS WHO did as illustrated by Brier scores (Fig. 4a).
Fig. 3

Kaplan–Meier plots stratifying according to WHO, Modelpath, ModelCNVL and Modelcombined in the discovery set (ad), the HD validation set (eh) the EORTC validation set (il) and the TCGA set (mp). Coulor schemes for WHO: red GBMIDHmut, blue AAIIIIDHmut, green AIIIDHmut; for Modelpath: red A4, blue A3path, green A2path; for ModelCNVL: red A4, blue A3CNVL, green A2CNVL; for Modelcombined: red A4, blue A3combined, green A2comined

Fig. 4

Brier scores to the grading approaches in the discovery set (a) and HD-validation set (b): reference which was random distribution (black), classifier tool (orange) based on epigenomic classification [5], unsupervised clustering of 850K based methylation data (yellow), WHO diagnosis (red), Modelpath (blue), ModelCNVL (green) and Modelcombined (purple)

ModelCNVL relied on absence/presence of homozygous CDKN2A/B deletion and on a threshold value of 350 Mb for CNVL. Therefore, A4 were determined as in our first approach. However, the remaining patients were subdivided with those having a CNV-L value < 350 Mb being termed astrocytoma, IDH-mutant, grade 2, CNVL (A2CNVL) and those with value > 350 Mb being termed astrocytoma, IDH-mutant, grade 3, CNVL (A3CNVL). A2CNVL featured 86 patients containing 7 GBMIDHmut, 42 AAIIIIDHmut and 37 AIIIDHmut. A3CNVL contained 87 patients, with 37 GBMIDHmut, 33 AAIIIIDHmut and 17 AIIIDHmut. A OS plot of the discovery set according to ModelCNVL is shown (Fig. 3c). ModelCNVL performed better in predicting OS of the groups than the 2016 CNS WHO did as illustrated by Brier scores (Fig. 4a).

Modelcombined was devised to circumvent the provocative shift of patients with necrosis but without CDKN2A/B deletion and with a low CNVL into the most favorable patient group (Fig. 2). Instead, this patient group was placed in the intermediate malignancy group. Notably, Modelcombined is not based on a strict mathematical approach. The number of only seven patients in this group is too small to confirm OS comparable to that of patients in the intermediate malignancy group (supplementary figure 3). In Modelcombined the A4 group defined by homozygous CDKN2A/B deletion was identical to that in Modelpath and ModelCNVL. Modelcombined also performed better in predicting OS of the groups than the 2016 CNS WHO could accomplish as illustrated by Brier scores (Fig. 4a).

Importantly, all three-tiered grading approaches, i.e. WHO, Modelpath, ModelCNVL and Modelcombined were predictive with higher power than two-tiered approaches based on unsupervised clustering and the classifier tool (Fig. 4a).

Validation of the study series

The validation sets supported our findings from the discovery set.

In the HD validation set, WHO-based separation was of borderline significance (p = 0.05) (Fig. 3e). In contrast, Modelpath or ModelCNVL both significantly separated three prognostic patient groups (p < 0.0001) (Fig. 3f, g). Best performance was achieved by Modelcombined (Fig. 3h).

The EORTC validation set produced comparable results (Fig. 3i–l). Here, the 2016 CNS WHO did separate groups with different OS (p = 0.004). Modelpath performed slightly better (p = 0.002), however, ModelCNVL and Modelcombined again provided best separation into groups of different OS (p < 0.0001).

The TCGA dataset was only of limited value for validating our models due to a high number of cases with relatively short observation periods. We observed a good separation of patients recognized by our models as highly malignant due to the presence of a homozygous CDKN2A/B deletion. TCGA patients with CDKN2A/B homozygous deletion also died relatively quickly, and therefore, separated significantly from patients without such deletion. However, the separation between the intermediate and more favorable tumor groups was not possible due to a high number of patients lacking long-term follow-up. Thus, due to its composition, the TCGA validation set only could be used for validating the poor OS of patients with homozygous CDKN2A/B deletion (Fig. 3m–p).

In conclusion, in all three validation sets the tested grading models performed better than the current WHO system.

Discussion

Using combined histological and genetic parameters, notably necrosis and CDKN2A/B deletion, we have established a novel grading system for IDH-mutant diffuse astrocytic tumors. Interestingly, this system does not incorporate morphological estimates of proliferation, which have been a key parameter in WHO grading systems through counting mitotic figures. Indeed, CDKN2A/B homozygous deletion is tightly linked to proliferation because loss of p16 removes the inhibition of complexes of CDK4 with D-type cyclins thereby driving the cell cycle. In our discovery set CDKN2A/B homozygous deletion was clearly associated with higher proliferation. However, a considerable fraction of patients lacking the deletion still exhibited high Ki-67 reads (supplementary figure 4).

By focusing on this marker rather than quantifying proliferation, grading accuracy may improve because the difficulties underlying precise determination of mitotic counts (e.g., interobserver variations, tissue artifacts, interfering staining conditions, sampling, or extended time from tissue removal to fixation) are eliminated.

In our discovery set, mitotic count was not a strong predictor of outcome. This confirms previous reports pointing to the failure of morphological proliferation assessment in predicting clinical outcome in IDH-mutant astrocytic gliomas [20, 28]. Likewise, Ki67 staining ubiquitously used as a parameter for estimating proliferation provides variable readouts dependent on staining or processing conditions and, therefore, is problematic for the determination of cutoff values to discern malignancy grades.

Proliferation assessed by Ki-67 is described as low or absent in AIIIDHmut in the 2016 CNS WHO 2016 [17]. In our discovery set, Ki-67 is significantly associated prognosis, but the split point was determined around 15%. A reason for the failure of Ki-67 as a prognostic parameter in WHO is the low split point (often near 2%) in the previous studies [11, 15]. These studies all were from the pre-IDH era and contained many glioblastomas, thus driving the split points to low values. This may explain why the 2016 CNS WHO stated that Ki-67 was not prognostic for AAIIIIDHmut [17].

The presence of homozygous CDKN2A/B deletion turned out to be the most powerful parameter for inferior clinical outcome. This parameter is currently not employed in WHO grading, but could be construed as a molecular estimate of proliferation given the role of the p16 protein in regulating the cell cycle. Interestingly, CDKN2A/B homozygous deletion in tumors without necrosis, graded AAIIIIDHmut, was associated with OS indistinguishable from that of patients with tumors exhibiting both CDKN2A/B homozygous deletion and necrosis that were diagnosed by definition as GBMIDHmut (Fig. 1b). And in turn, by removing tumors with CDKN2A/B homozygous deletion among AAIIIIDHmut, OS of the remaining patients did not significantly (p = 0.124) differ from that of AIIIDHmut (Fig. 1b). CDKN2A/B homozygous deletion or mutations in the RB pathway have been determined as an unfavorable parameter in previous studies [1, 26]. We therefore, performed an analysis combining cases with either CDKN2A/B homozygous deletion or RB1 homozygous deletion or CDK4 or CDK6 amplification. However, this did not result in an overall improvement of separating groups of patients with different outcome in all tumor sets. We think this is partly due to the more pronounced association of CDKN2A/B homozygous deletion with survival than that of the other RB1 pathway genes analyzed (supplementary figure 5). While combining typical lesions of the RB1 pathway should be further explored, we expect that mainly the assessment of CDKN2A/B status will be highly relevant for future diffuse astrocytoma grading systems. Satisfactory data acquisition can be performed by FISH, by quantitative PCR or by array technology as performed in this study. As shown elsewhere, CDKN2A/B status may be readily determined by FISH analysis [23, 25]. Given the utility of immunohistochemistry for routine pathology diagnosis, we tried to detect homozygous deletion of the CDKN2A gene product p16 by immunohistochemistry. Unfortunately, using p16 antibody cloneG175-405 (BD-Biosciences) we could not show satisfactory correlation between p16 immunohistochemistry and molecular detection of CDKN2A/B homozygous deletion. In our immunohistochemistry, the basal nuclear expression of p16 was too low in many samples to allow definite detection of loss of expression. Our observation is supported by another study observing a correlation of p16 expression loss and CDKN2A deletion, however, only in 85% of the deleted tumors [25]. We consider this correlation not tight enough and, therefore, favor determination of CDKN2A/B deletion by an assay addressing DNA directly. A representative CNP with evident CDKN2A/B homozygous deletion is shown in supplementary figure 6.

Another key parameter, the presence of necrosis, which according to the 2016 CNS WHO results in the diagnosis of GBMIDHmut, appears overrated in terms of estimating prognosis in IDH-mutant diffuse astrocytomas. As a single factor, necrosis is prognostically highly significant in our discovery set (Table 1), but the presence of necrosis in the absence of homozygous CDKN2A/B deletion was associated with clearly better prognosis than that of tumors containing the homozygous deletion (Fig. 1b). To account for these observations, we developed grading models based on those parameters that turned out to predict clinical outcome better than grading according to the 2016 CNS WHO (Fig. 2). Better outcome prediction could be confirmed in two independent validation series and better detection of patients at risk for poor OS could be achieved in the TCGA set with shorter observation periods only (Fig. 3). Our data on the TCGA data are in line with a previous study on the TCGA LGG data set reporting that CDKN2A homozygous deletion, but not expression associates with unfavorable OS [29]. Because all three tiered grading approaches in this study predicted OS with less error than the two-tiered grading schemes emerging from unsupervised clustering or from using the methylation-based classifier [5], the latter were not further considered, although both schemes proved stable in the validation series (Fig. 4a, b and supplementary figure 2a, b, c, d, e, f and 5).

An interesting parameter emerging from our analysis was CNVL. A proportion higher than 350 Mb either lost or gained in the areas covered by the methylation arrays correlated with poorer OS. Previously, the mutational load has been shown to correlate with tumor grade in IDH-mutant glioma [9]. While our data more reflect genomic instability and those data rely on an accumulation of mutations they also support a quantitative approach to tumor grading.

Shifts of patients according to different grading approaches

WHO grading and our grading models result in three different groups, however, with different distributions. The allotment of patients in the discovery set to model-specific sets is provided in supplementary table 1. Our novel approaches lead to a higher overall number of diffuse astrocytomas in the lower grade group. This effect is more pronounced for Modelpath than for ModelCNVL. In Modelpath, the increase from 54 AIIIDHmut to 129 A2path comes with the reduction of median OS from 7053 to 5122 days. In contrast, the increase from 54 AIIIDHmut to 86 A2CNVL did not alter the median OS of 7053 days. The shifts in patient numbers in the comparable groups is mainly due to the strong influence of proliferation rate on the WHO grading algorithm. The typical problem is exemplified by a well differentiated diffuse astrocytoma being currently diagnosed as a WHO grade III tumor because of the detection of a few mitotic figures. In such a setting, the clinician may be surprised by a grade III designation given the lack of neuroradiological or intraoperative evidence for higher grade. The novel grading approaches described here appear to address this clinicopathological problem. It should be noted that the diagnostic considerations underlying WHO grading of the discovery set already deviated from the traditional WHO guidelines due to the experience of the authors who diagnosed and graded the tumors [20, 28]. In particular, more than single mitoses were accepted as compatible with AIIIDHmut. In the setting of IDH mutations, the presence of few mitotic figures did not result in a diagnosis of AAIIIIDHmut. This may be another reason for “standard” grading successfully predicting OS in the discovery set.

Our data demonstrate that AAIIIIDHmut stratified for CDKNA/B homozygous deletion separates into two groups with quite different OS: patients with tumors lacking this alteration follow a moderately aggressive course while patients with tumors having a CDKN2A/B deletion fare more similar to patients with GBMIDHmut with homozygous deletion but much worse than patients with GBMIDHmut lacking CDKN2A/B homozygous deletion (Fig. 1b). Thus, combining patients with AIIIDHmut and AAIIIIDHmut but lacking homozygous deletion of CDKN2A/B into a single group in Modelpath forms a fairly homogenous cohort and may have implications for the nomenclature of these tumors as well as a potential therapeutic relevance.

Modelpath results in reduction of the size of the intermediate grading group. In the discovery set, AAIIIIDHmut with 90 patients constitutes the largest group. The equivalent group according to Modelpath includes 44 patients. This contrasts ModelCNVL which allots 87 patients to the intermediate group. The relation of the A3path and A3CNVL groups is analyzed in Modelcombined based on CDKN2A/B, necrosis and CNVL (Fig. 3b–d). The power of this model is best exemplified by the lowest prediction error rate as shown by Brier score (Fig. 4a for discovery set, Fig. 4b for HD-validation set).

Implication for the use of morphological parameters, proliferation and necrosis

Our data challenge the value of two main criteria that have long been used in the grading of diffuse astrocytomas: proliferation and necrosis. As previously reported [20], mitotic count is of less significance in a set of IDH-mutant diffuse astrocytomas than in a set of diffuse astrocytomas not tested for IDH status (and thereby likely to contain a substantial fraction of biological glioblastomas). In our analyses, mitotic figures did not emerge as a parameter providing the most significant separation of patients with IDH-mutant diffuse astrocytoma in groups with different OS. We, therefore, suggest a more conservative use of proliferation and caution to use this as a sole indicator for high tumor grade.

The other grading parameter in question is necrosis. The presence of necrosis in a diffuse astrocytic neoplasm that had not been pretreated inevitably prompted the WHO diagnosis of GBMIDHmut. However, patients with tumors exhibiting necrosis but not containing homozygous CDKN2A/B deletion survive significantly better than patients with tumors lacking necrosis but containing a CDKN2A/B homozygous deletion (Fig. 1b). This observation in our discovery and validation series (Fig. 3) should prompt a more conditional approach to the presence of necrosis as a parameter for grading of higher-grade diffuse astrocytic tumors.

Clinical perspective of grading according to the novel approaches

Caveats of our study are the heterogeneity of treatment and the potential effect of treatment on OS. Patients with AAIIIIDHmut and AIIIDHmut.have been treated quite heterogeneously with therapies ranging from wait and see to radiotherapy, chemotherapy or combinations thereof. In general, patients with AAIIIIDHmut may have been likely to receive a more intense treatment than patients with AIIIDHmut. We cannot exclude the possibility of treatment being effective in patients with AAIIIIDHmut, thereby blurring the distinction to OS of patients with AIIIDHmut.

Nonetheless, the current treatment for AIIIDHmut and AAIIIIDHmut is similar in most centers. Therefore, the shift of patients between the corresponding groups following grading according to our novel approaches will likely not have an immediate effect on treatment. However, upon applying Modelcombined we subdivide AIIIDHmut into A2combined and A3combined. Patients with A2combined, although more numerous than those in AIIIDHmut, appear to exhibit a slightly better OS than patients with AIIIDHmut. This may prompt future consideration regarding therapy in A2combined patients. In contrast, patients with A3combined show average OS closer to AAIIIIDHmut patients, thereby potentially supporting similar treatments for these groups. In turn, it will be of interest to determine if patients with GBMIDHmut should receive the same treatment as those patients with GBMIDHmut lacking homozygous CDKN2A/B deletion since these patients have OS approaching that of AAIIIIDHmut. Finally, patients assigned to the least favorable A4 group all exhibit homozygous CDKN2A/B deletion; such patients are in need of maximal treatment and could possibly benefit from treatment with CDK antagonists because CDKN2A/B deletion results in increased CDK activity [24].

Implications for the classification and grading of IDH-mutant diffuse astrocytic tumors

The 2016 CNS WHO divided the traditionally termed “glioblastoma” into GBMIDHwt, GBMIDHmut and, in specific situations, diffuse midline glioma, H3K27–mutant. Genetically, these three groups are entirely different. While the diffuse midline glioma, H3K27–mutant is semantically clearly set apart from GBMIDHwt, the latter shares the same “family name” with GBMIDHmut [18]. However, there appears to be greater genetic kinship between GBMIDHmut and AIDHmut and not between GBMIDHmut and GBMIDHwt. This kinship is also supported by the distribution of other mutations such as ATRX or TP53 that are typically seen in these tumors and by the rarity of many mutations typical for GBMIDHwt such as EGFR amplification. Thus a genetic approach to taxonomy would favor a term such as “high-grade AIDHmut” rather than the current term of GBMIDHmut.

However, this still would be in conflict with a harmonious grading scale. Clearly, GBMIDHmut has a better prognosis than GBMIDHwt. In fact, GBMIDHmut has a better prognosis than the prognosis allotted to the pre-IDH era anaplastic astrocytoma WHO grade III (AAIIINOS) [34]. This circumstance would argue for a lower grade than currently allotted to the GBMIDHmut. However, the novel creation of an IDH-mutant glioblastoma grade 3 would seem awkward [18].

We suggest that future grading approaches include assessment of homozygous CDKN2A/B deletion in these tumors, given its powerful association with poorer prognosis. This accounts for both GBMIDHmut and AAIIIIDHmut—resulting in essential irrelevance of the presence of necrosis on OS in tumors with homozygous CDKN2A/B deletion. In other words, a definable fraction of AAIIIIDHmut fare significantly worse than a fraction of the GBMIDHmut. This cannot be repaired using the current WHO nomenclature without major changes to our historical concepts of nomenclature [18].

One possible solution, which we have considered implementing in Heidelberg, would be to restrict classification to the term AIDHmut and then adapt grading according to molecular lesions, thereby omitting the term GBMIDHmut. The term “glioblastoma” would be reserved for those histologically defined glioblastomas lacking IDH mutation or not having had adequate (not otherwise specified, NOS) or diagnostic (not elsewhere classified, NEC) work-ups [19]. Such a system would lend itself well to the use of layered reports, as discussed elsewhere [16]. In this context, if these data were confirmed in other studies, the WHO classification of such tumors could resemble what is given in supplementary Table 2 (based on Modelpath). Advantages of such an approach would include better prognostic correlations and therefore guidance of therapies, and a “freeing up” of grading from classification that could allow more easy adaptation of grading criteria in the future [18].

Conclusions

The 2016 CNS WHO grading of IDH-mutant astrocytic tumors is not as prognostically meaningful as needed and the histological parameters of proliferation and necrosis appear overrated when incorporating IDH gene status. Significantly, several other morphological and molecular parameters show higher prognostic power and modeling has provided suggestions for grading algorithms, with the proposal that CDNK2A/B status will be important for future grading of these tumors.

Notes

Acknowledgements

This work was supported by German Cancer Aid (70112371) to AvD and German Cancer Aid (110624) to WW and MW. We are indebted to Toshio Sasajima, Masaya Oda and Masataka Takahashi for support regarding clinical data acquisition. We thank Viktoria Zeller, Ulrike Lass, Antje Habel, Ulrike Vogel, Katja Brast, Kerstin Lindenberg, John Moyers and Jochen Meyer for excellent technical assistance. We also thank the Microarray unit of the Genomics and Proteomics Core Facility, German Cancer Research Center (DKFZ), especially Matthias Schick, Roger Fischer, Nadja Wermke and Anja Schramm-Glück, for providing methylation services.

Supplementary material

401_2018_1849_MOESM1_ESM.pptx (353 kb)
Supplementary Fig. 1 Copy number summary plots. Vertical axis indicates percentage of patients affected. Horizontal axis refers to chromosomal localization. Dotted vertical lines indicate border between p and q arms. Data are given for three age groups (PPTX 352 kb)
401_2018_1849_MOESM2_ESM.pptx (281 kb)
Supplementary Fig. 2 Kaplan–Meier plots stratifying by unsupervised clustering in the discovery set (a), the HD validation set (b) and the EORTC validation set (c). Kaplan–Meier plots stratifying by the methylation based classifier in the discovery set (d), the HD validation set (e) and the EORTC validation set (f) (PPTX 280 kb)
401_2018_1849_MOESM3_ESM.pptx (151 kb)
Supplementary Fig. 3 OS of subgroups in Modelcombined. The patient set (n = 7) characterized by necrosis, absence of CDKN2A/B homozygous deletion and low CNVL exhibited only 2 events. While the corresponding curve (red) fits well the group of patients with intermediate OS, the number is too low for a clear statement (PPTX 150 kb)
401_2018_1849_MOESM4_ESM.pptx (166 kb)
Supplementary Fig. 4 Association of proliferation markers with CDKN2A/B status. (a) association with Ki67. (b) association with pHH3. Horizontal bars in box-plots correspond to median values (PPTX 166 kb)
401_2018_1849_MOESM5_ESM.pptx (101 kb)
Supplementary Fig. 5 OS of patients with IDH-mutant astrocytoma in association with RB pathway genes. (a) red -homozygous deletion of CDKN2A/B, black – wild type status. (b) red –homozygous deletion of RB1 or CDK4 amplification or CDK6 amplification, black – wild type status.. (a) red -homozygous deletion of CDKN2A/B or homozygous deletion of RB1 or CDK4 amplification or CDK6 amplification, black – wild type status for all (PPTX 100 kb)
401_2018_1849_MOESM6_ESM.pptx (196 kb)
Supplementary Fig. 6 Examples for CNP from astrocytic tumors exhibiting homozygous deletion of CDKN2A/B (PPTX 196 kb)
401_2018_1849_MOESM7_ESM.docx (66 kb)
Supplementary Table 1 Discovery set data employed in grading schemes for IDH-mutant astrocytoma. For each scheme lowest grade is indicated by green, intermediate grade by blue and highest grade by red color (DOCX 66 kb)
401_2018_1849_MOESM8_ESM.docx (14 kb)
Supplementary Table 2 Possible designation of IDH-mutant astrocytoma based on Modelpath in a future classification scheme (DOCX 13 kb)

References

  1. 1.
    Aoki K, Nakamura H, Suzuki H et al (2018) Prognostic relevance of genetic alterations in diffuse lower-grade gliomas. Neurooncology 20:66–77Google Scholar
  2. 2.
    Balss J, Meyer J, Mueller W, Korshunov A, Hartmann C, von Deimling A (2008) Analysis of the IDH1 codon 132 mutation in brain tumors. Acta Neuropathol 116:597–602CrossRefPubMedGoogle Scholar
  3. 3.
    Burger PC, Vogel FS, Green SB, Strike TA (1985) Glioblastoma multiforme and anaplastic astrocytoma, pathologic criteria and prognostic implications. Cancer 56:1106–1111CrossRefPubMedGoogle Scholar
  4. 4.
    Cancer Genome Atlas Research N, Brat DJ, Verhaak RG et al (2015) Comprehensive, integrative genomic analysis of diffuse lower-grade gliomas. N Engl J Med 372:2481–2498CrossRefGoogle Scholar
  5. 5.
    Capper D, Jones DTW, Sill M et al (2018) DNA methylation-based classification of central nervous system tumours. Nature 555:469–474CrossRefPubMedPubMedCentralGoogle Scholar
  6. 6.
    Capper D, Zentgraf H, Balss J, Hartmann C, von Deimling A (2009) Monoclonal antibody specific for IDH1 R132H mutation. Acta Neuropathol 118:599–601CrossRefPubMedGoogle Scholar
  7. 7.
    Colman H, Giannini C, Huang L et al (2006) Assessment and prognostic significance of mitotic index using the mitosis marker phospho-histone H3 in low and intermediate-grade infiltrating astrocytomas. Am J Surg Pathol 30:657–664CrossRefPubMedGoogle Scholar
  8. 8.
    Daumas-Duport C, Scheithauer B, O’Fallon J, Kelly P (1988) Grading of astrocytomas. A simple and reproducible method. Cancer 62:2152–2165CrossRefPubMedGoogle Scholar
  9. 9.
    Draaisma K, Wijnenga MM, Weenink B, Gao Y, Smid M, Robe P, van den Bent MJ, French PJ (2015) PI3 kinase mutations and mutational load as poor prognostic markers in diffuse glioma patients. Acta Neuropathol Commun 3:88CrossRefPubMedPubMedCentralGoogle Scholar
  10. 10.
    Duregon E, Bertero L, Pittaro A et al (2016) Ki-67 proliferation index but not mitotic thresholds integrates the molecular prognostic stratification of lower grade gliomas. Oncotarget 7:21190–21198CrossRefPubMedPubMedCentralGoogle Scholar
  11. 11.
    Hsu DW, Louis DN, Efird JT, Hedley-Whyte ET (1997) Use of MIB-1 (Ki-67) immunoreactivity in differentiating grade II and grade III gliomas. J Neuropathol Exp Neurol 56:857–865CrossRefPubMedGoogle Scholar
  12. 12.
    Kernohan JW, Mabon RF, Svien HJ, Adson AW (1949) A simplified classification of gliomas. Proc Staff Meet Mayo Clin 24:71–75PubMedGoogle Scholar
  13. 13.
    Louis D, Ohgaki H, Wiestler O, Cavenee W (2007) World Health Organization classification of tumours of the central nervous system. In: Bosman F, Jaffe E, Lakhani S, Ohgaki H (eds) World Health Organization classification of tumours, 4th edn. IARC, LyonGoogle Scholar
  14. 14.
    Louis D, Ohgaki H, Wiestler O, Cavenee WK (2016) World Health Organization classification of tumours of the central nervous system. In: Bosman F, Jaffe E, Lakhani S, Ohgaki H (eds) World Health Organization classification of tumours revised, 4th edn. IARC, LyonGoogle Scholar
  15. 15.
    Louis DN, Edgerton S, Thor AD, Hedley-Whyte ET (1991) Proliferating cell nuclear antigen and Ki-67 immunohistochemistry in brain tumors: a comparative study. Acta Neuropathol 81:675–679CrossRefPubMedGoogle Scholar
  16. 16.
    Louis DN, Perry A, Burger P et al (2014) International Society of Neuropathology-Haarlem consensus guidelines, for nervous system tumor classification and grading. Brain Pathol 24:429–435CrossRefPubMedGoogle Scholar
  17. 17.
    Louis DN, Perry A, Reifenberger G et al (2016) The 2016 world health organization classification of tumors of the central nervous system: a summary. Acta Neuropathol 131:803–820CrossRefPubMedGoogle Scholar
  18. 18.
    Louis DN, von Deimling A (2017) Grading of diffuse astrocytic gliomas: Broders, Kernohan, Zulch, the WHO… and Shakespeare. Acta Neuropathol 134:517–520CrossRefPubMedGoogle Scholar
  19. 19.
    Louis DN, Wesseling P, Paulus W et al (2017) cIMPACT-NOW update 1: not otherwise specified (NOS) and not elsewhere classified (NEC). Acta Neuropathol 135:481–484CrossRefGoogle Scholar
  20. 20.
    Olar A, Wani KM, Alfaro-Munoz KD et al (2015) IDH mutation status and role of WHO grade and mitotic index in overall survival in grade II-III diffuse gliomas. Acta Neuropathol 129:585–596CrossRefPubMedPubMedCentralGoogle Scholar
  21. 21.
    Parsons DW, Jones S, Zhang X et al (2008) An integrated genomic analysis of human glioblastoma multiforme. Science 321:1807–1812CrossRefPubMedPubMedCentralGoogle Scholar
  22. 22.
    Pekmezci M, Rice T, Molinaro AM et al (2017) Adult infiltrating gliomas with WHO 2016 integrated diagnosis: additional prognostic roles of ATRX and TERT. Acta Neuropathol 133:1001–1016CrossRefPubMedPubMedCentralGoogle Scholar
  23. 23.
    Perry A, Nobori T, Ru N, Anderl K, Borell TJ, Mohapatra G, Feuerstein BG, Jenkins RB, Carson DA (1997) Detection of p16 gene deletions in gliomas: a comparison of fluorescence in situ hybridization (FISH) versus quantitative PCR. J Neuropathol Exp Neurol 56:999–1008CrossRefPubMedGoogle Scholar
  24. 24.
    Pfaff E, Kessler T, Balasubramanian GP et al (2017) Feasibility of real-time molecular profiling for patients with newly diagnosed glioblastoma without MGMT promoter-hypermethylation—the NCT Neuro Master Match (N2M2) pilot study. Neuro Oncol.  https://doi.org/10.1093/neuonc/nox216 CrossRefGoogle Scholar
  25. 25.
    Purkait S, Jha P, Sharma MC, Suri V, Sharma M, Kale SS, Sarkar C (2013) CDKN2A deletion in pediatric versus adult glioblastomas and predictive value of p16 immunohistochemistry. Neuropathology 33:405–412CrossRefPubMedGoogle Scholar
  26. 26.
    Reis GF, Pekmezci M, Hansen HM et al (2015) CDKN2A loss is associated with shortened overall survival in lower-grade (World Health Organization Grades II–III) astrocytomas. J Neuropathol Exp Neurol 74:442–452CrossRefPubMedPubMedCentralGoogle Scholar
  27. 27.
    Reuss DE, Kratz A, Sahm F et al (2015) Adult IDH wild type astrocytomas biologically and clinically resolve into other tumor entities. Acta Neuropathol 130:407–417CrossRefPubMedGoogle Scholar
  28. 28.
    Reuss DE, Mamatjan Y, Schrimpf D et al (2015) IDH mutant diffuse and anaplastic astrocytomas have similar age at presentation and little difference in survival: a grading problem for WHO. Acta Neuropathol 129:867–873CrossRefPubMedPubMedCentralGoogle Scholar
  29. 29.
    Roy DM, Walsh LA, Desrichard A, Huse JT, Wu W, Gao J, Bose P, Lee W, Chan TA (2016) Integrated genomics for pinpointing survival loci within arm-level somatic copy number alterations. Cancer Cell 29:737–750CrossRefPubMedPubMedCentralGoogle Scholar
  30. 30.
    Schneider CA, Rasband WS, Eliceiri KW (2012) NIH Image to ImageJ: 25 years of image analysis. Nat Methods 9:671–675CrossRefPubMedPubMedCentralGoogle Scholar
  31. 31.
    Sommer C, Strähle C, Köthe U, Hamprecht FA (2011) Ilastik: interactive learning and segmentation toolkit. In: Eighth IEEE international symposium on biomedical imaging (ISBI). IEEE, Chicago, pp 230–233Google Scholar
  32. 32.
    Sturm D, Witt H, Hovestadt V et al (2012) Hotspot mutations in H3F3A and IDH1 define distinct epigenetic and biological subgroups of glioblastoma. Cancer Cell 22:425–437CrossRefPubMedGoogle Scholar
  33. 33.
    Suzuki H, Aoki K, Chiba K et al (2015) Mutational landscape and clonal architecture in grade II and III gliomas. Nat Genet 47:458–468CrossRefPubMedGoogle Scholar
  34. 34.
    von Deimling A, Ono T, Shirahata M, Louis D (2018) Grading of diffuse astrocytic gliomas: a review of studies before and after the advent of IDH testing. Semin Neurol 38:19–23CrossRefGoogle Scholar
  35. 35.
    Watanabe T, Nobusawa S, Kleihues P, Ohgaki H (2009) IDH1 mutations are early events in the development of astrocytomas and oligodendrogliomas. Am J Pathol 174:653–656Google Scholar
  36. 36.
    Wick W, Hartmann C, Engel C et al (2009) NOA-04 randomized phase III trial of sequential radiochemotherapy of anaplastic glioma with procarbazine, lomustine, and vincristine or temozolomide. J Clin Oncol 27:5874–5880CrossRefPubMedGoogle Scholar
  37. 37.
    Yan H, Parsons DW, Jin G et al (2009) IDH1 and IDH2 mutations in gliomas. N Engl J Med 360:765–773CrossRefPubMedPubMedCentralGoogle Scholar
  38. 38.
    Zülch KJ (1979) Histological typing of tumours of the central nervous system. Who’s Who, LondonGoogle Scholar

Copyright information

© Springer-Verlag GmbH Germany, part of Springer Nature 2018

Authors and Affiliations

  • Mitsuaki Shirahata
    • 1
    • 2
  • Takahiro Ono
    • 1
    • 3
  • Damian Stichel
    • 4
  • Daniel Schrimpf
    • 1
    • 4
  • David E. Reuss
    • 1
    • 4
  • Felix Sahm
    • 1
    • 4
  • Christian Koelsche
    • 1
    • 4
  • Annika Wefers
    • 1
    • 4
  • Annekathrin Reinhardt
    • 1
    • 4
  • Kristin Huang
    • 1
    • 4
  • Philipp Sievers
    • 1
    • 4
  • Hiroaki Shimizu
    • 3
  • Hiroshi Nanjo
    • 5
    • 6
  • Yusuke Kobayashi
    • 2
  • Yohei Miyake
    • 2
  • Tomonari Suzuki
    • 2
  • Jun-ichi Adachi
    • 2
  • Kazuhiko Mishima
    • 2
  • Atsushi Sasaki
    • 7
  • Ryo Nishikawa
    • 2
  • Melanie Bewerunge-Hudler
    • 8
  • Marina Ryzhova
    • 9
  • Oksana Absalyamova
    • 9
  • Andrey Golanov
    • 9
  • Peter Sinn
    • 10
  • Michael Platten
    • 11
  • Christine Jungk
    • 12
  • Frank Winkler
    • 13
    • 14
  • Antje Wick
    • 13
    • 14
  • Daniel Hänggi
    • 15
  • Andreas Unterberg
    • 12
  • Stefan M. Pfister
    • 16
    • 17
    • 18
  • David T. W. Jones
    • 16
    • 17
  • Martin van den Bent
    • 19
  • Monika Hegi
    • 20
    • 21
  • Pim French
    • 19
  • Brigitta G. Baumert
    • 22
  • Roger Stupp
    • 23
  • Thierry Gorlia
    • 24
  • Michael Weller
    • 25
  • David Capper
    • 1
    • 26
    • 27
    • 28
  • Andrey Korshunov
    • 1
    • 4
  • Christel Herold-Mende
    • 12
  • Wolfgang Wick
    • 13
    • 14
  • David N. Louis
    • 29
  • Andreas von Deimling
    • 1
    • 4
    • 30
    Email author
  1. 1.Department of Neuropathology, Institute of PathologyHeidelberg University HospitalHeidelbergGermany
  2. 2.Department of Neuro-Oncology/NeurosurgerySaitama Medical University International Medical CenterHidakaJapan
  3. 3.Department of NeurosurgeryAkita University Graduate School of MedicineAkitaJapan
  4. 4.Clinical Cooperation Unit NeuropathologyGerman Cancer Consortium (DKTK), German Cancer Research Center (DKFZ)HeidelbergGermany
  5. 5.Akita University Hospital Division of Clinical PathologyAkitaJapan
  6. 6.Medical CenterHidakaJapan
  7. 7.Department of PathologySaitama Medical University International Medical CenterHidakaJapan
  8. 8.Genomics and Proteomics Core Facility, German Cancer Research Center (DKFZ)HeidelbergGermany
  9. 9.NN Burdenko Neurosurgical InstituteMoscowRussia
  10. 10.Department of PathologyUniversity Hospital HeidelbergHeidelbergGermany
  11. 11.Department of NeurologyUniversitätsmedizin Mannheim, Medical Faculty Mannheim, Heidelberg UniversityMannheimGermany
  12. 12.Department of NeurosurgeryHeidelberg University HospitalHeidelbergGermany
  13. 13.Department of NeurologyHeidelberg University HospitalHeidelbergGermany
  14. 14.Clinical Cooperation Unit NeurooncologyGerman Cancer Consortium (DKTK), German Cancer Research Center (DKFZ)HeidelbergGermany
  15. 15.Department of NeurosurgeryUniversity Medical Center Mannheim, University of HeidelbergMannheimGermany
  16. 16.Hopp Children’s Cancer Center at the NCT Heidelberg (KiTZ)HeidelbergGermany
  17. 17.Division of Pediatric NeurooncologyGerman Cancer Consortium (DKTK), German Cancer Research Center (DKFZ)HeidelbergGermany
  18. 18.Department of Pediatric Oncology, Hematology and ImmunologyUniversity Hospital HeidelbergHeidelbergGermany
  19. 19.The Brain Tumor Center at Erasmus MC Cancer InstituteRotterdamThe Netherlands
  20. 20.Laboratory of Brain Tumor Biology and Genetics, Neuroscience Research CenterLausanne University HospitalLausanneSwitzerland
  21. 21.Division of NeurosurgeryLausanne University HospitalLausanneSwitzerland
  22. 22.Department of Radiation-Oncology (MAASTRO Clinic) and GROW (School for Oncology)Maastricht University Medical CentreMaastrichtThe Netherlands
  23. 23.Malnati Brain Tumor Institute of the Lurie Cancer CenterNorthwestern University Feinberg School of MedicineChicagoUSA
  24. 24.EORTC HeadquarterBrusselsBelgium
  25. 25.Department of NeurologyUniversity Hospital and University of ZurichZurichSwitzerland
  26. 26.German Cancer Consortium (DKTK), Partner Site Berlin, German Cancer Research Center (DKFZ) HeidelbergBerlinGermany
  27. 27.Charité-Universitätsmedizin Berlin, Corporate Member of Freie Universität Berlin, Humboldt-Universität zu BerlinBerlinGermany
  28. 28.Department of NeuropathologyBerlin Institute of HealthBerlinGermany
  29. 29.Department of PathologyMassachusetts General Hospital and Harvard Medical SchoolBostonUSA
  30. 30.Department for Neuropathology and CCU NeuropathologyUniversity of Heidelberg and DKFZHeidelbergGermany

Personalised recommendations