Integrated multi-omics analysis of oligodendroglial tumours identifies three subgroups of 1p/19q co-deleted gliomas

Abstract

Oligodendroglial tumours (OT) are a heterogeneous group of gliomas. Three molecular subgroups are currently distinguished on the basis of the IDH mutation and 1p/19q co-deletion. Here we present an integrated analysis of the transcriptome, genome and methylome of 156 OT. Not only does our multi-omics classification match the current classification but also reveals three subgroups within 1p/19q co-deleted tumours, associated with specific expression patterns of nervous system cell types: oligodendrocyte, oligodendrocyte precursor cell (OPC) and neuronal lineage. We confirm the validity of these three subgroups using public datasets. Importantly, the OPC-like group is associated with more aggressive clinical and molecular patterns, including MYC activation. We show that the MYC activation occurs through various alterations, including MYC genomic gain, MAX genomic loss, MYC hypomethylation and microRNA-34b/c down-regulation. In the lower grade glioma TCGA dataset, the OPC-like group is associated with a poorer outcome independently of histological grade. Our study reveals previously unrecognized heterogeneity among 1p/19q co-deleted tumours.

ARTICLE

Received 1 Oct 2015 | Accepted 7 Mar 2016 | Published 19 Apr 2016

Aurlie Kamoun1, Ahmed Idbaih2,3,4,5, Caroline Dehais5, Nabila Elarouci1, Catherine Carpentier2,3,4,

Eric Letouz1, Carole Colin6, Karima Mokhtari2,3,4,7, Anne Jouvet8, Emmanuelle Uro-Coste9,Nadine Martin-Duverneuil10, Marc Sanson2,3,4,5, Jean-Yves Delattre2,3,4,5,11, Dominique Figarella-Branger6,12, Aurlien de Reynis1,*, Franois Ducray13,14,15,* & POLA networkw

Oligodendroglial tumours (OT) are a heterogeneous group of gliomas. Three molecular subgroups are currently distinguished on the basis of the IDH mutation and 1p/19q co-deletion. Here we present an integrated analysis of the transcriptome, genome and methylome of 156 OT. Not only does our multi-omics classication match the current classication but also reveals three subgroups within 1p/19q co-deleted tumours, associated with specic expression patterns of nervous system cell types: oligodendrocyte, oligodendrocyte precursor cell (OPC) and neuronal lineage. We conrm the validity of these three subgroups using public datasets. Importantly, the OPC-like group is associated with more aggressive clinical and molecular patterns, including MYC activation. We show that the MYC activation occurs through various alterations, including MYC genomic gain, MAX genomic loss, MYC hypomethylation and microRNA-34b/c down-regulation. In the lower grade glioma TCGA dataset, the OPC-like group is associated with a poorer outcome independently of histological grade. Our study reveals previously unrecognized heterogeneity among 1p/19q co-deleted tumours.

1 Programme Cartes dIdentit des Tumeurs (CIT), Ligue Nationale Contre Le Cancer, 75013 Paris, France. 2 Universit Pierre et Marie Curie Paris 6, Centre de Recherche de lInstitut de Cerveau et de la Moelle Epinire (CRICM), UMR 975, 75013 Paris, France. 3 INSERM U975, 75013 Paris, France. 4 CNRS, UMR 7225, 75013 Paris, France. 5 AP-HP, Groupe Hospitalier Piti-Salptrire, Service de Neurologie 2-Mazarin, 75013 Paris, France. 6 Universit de la Mditerrane, Aix-Marseille, Facult de Mdecine La Timone, CRO2, UMR 911, 13885 Marseille, France. 7 AP-HP, Groupe Hospitalier Piti-Salptrire, Laboratoire de Neuropathologie R. Escourolle, 75013 Paris, France. 8 Dpartement de Pathologie et Neuropathologie, Hpital Neurologique, Hospices Civils de Lyon, 69374 Lyon, France. 9 CHU Toulouse, Hpital de Rangueil, Service dAnatomie et Cytologie Pathologique, 31400 Toulouse, France. 10 AP-HP, Groupe Hospitalier Piti-Salptrire, Service de Neuroradiologie, 75013 Paris, France. 11 Onconeurotek, Groupe Hospitalier Piti-Salptrire, 75013 Paris, France.

12 AP-HM, Hpital de la Timone, Service dAnatomie Pathologique et de Neuropathologie, 13885 Marseille, France. 13 Hospices Civils de Lyon, Hpital Neurologique, Service de Neuro-Oncologie, 69374 Lyon, France. 14 Department of Cancer Cell Plasticity, Cancer Research Centre of Lyon, INSERM U1052, CNRS UMR5286, 69008 Lyon, France. 15 Universit Claude Bernard Lyon 1, 69000 Lyon, France. * These authors jointly supervised this work.

w A full list of consortium members appears at the end of the paper. Correspondence and requests for materials should be addressed to A.d.R. (email: mailto:[email protected]

Web End [email protected] ) or to F.D. (email: mailto:[email protected]

Web End [email protected] ).

NATURE COMMUNICATIONS | 7:11263 | DOI: 10.1038/ncomms11263 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 1

DOI: 10.1038/ncomms11263 OPEN

Integrated multi-omics analysis of oligodendroglial tumours identies three subgroups of 1p/19q co-deleted gliomas

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms11263

Oligodendroglial tumours (OT), that is, gliomas with an oligodendroglial differentiation, account for B20% of adult diffuse gliomas1. They form a heterogeneous

group of gliomas in terms of clinical, histological and molecular proles2. The survival times of OT patients range from a few years to more than 15 years. This clinical heterogeneity reects underlying molecular heterogeneity. From a molecular point of view, three main subgroups of adults diffuse gliomas can be distinguished on the basis of two biomarkers, the 1p/19q co-deletion and the isocitrate dehydrogenase (IDH) mutation status3,4. Gliomas with the 1p/19q co-deletion (which are virtually all IDH mutated) display the best prognosis. The IDH-mutated gliomas, without 1p/19q co-deletion, have an intermediate prognosis. Finally, the non-1p/19q co-deleted and non-IDH-mutated gliomas have a poor prognosis. OT can belong to all three molecular subgroups even though pure oligodendroglial differentiation is strongly associated with the 1p/19q co-deletion1. Several studies have shown that this molecular classication was very robust and superior to the histological classication5,6. Accordingly, the revised World Health Organization (WHO) classication has proposed to use the IDH mutation and the 1p/19q co-deletion status to provide an integrated histo-molecular diagnosis of OT7. The aim of the present study was to assess whether the molecular classication of OT could be further rened on the basis of the integration of data from additional molecular levels.

Here we present an integrated analysis of the transcriptome, genome and methylome of 156 OT. In addition to formerly described subgroups, we report the identication of three subgroups within 1p/19q co-deleted tumours. One group is associated with more aggressive clinical and molecular patterns, including the MYC pathway activation. Our study reveals previously unrecognized heterogeneity among 1p/19q co-deleted tumours.

Results1p/19q co-deleted OT are molecularly heterogeneous. We used a series of 156 primary OT, 14 additional primary glioma samples and 9 normal samples referred henceforth as the Prise en charge des oligodendrogliomes anaplasiques (POLA) cohort. All samples (n 179) were proled on messenger RNA (mRNA)

expression arrays. MicroRNA (miRNA) sequencing was performed on 177 samples, and most of them were further proled on single-nucleotide polymorphism (SNP) arrays (n 161) and DNA methylation arrays (n 104) as described in

Supplementary Table 1.

A preliminary hierarchical consensus clustering of mRNA expression identied a subset of tumours (n 29), which

consistently clustered with normal brain and epilepsy surgery samples. Those tumours were also assigned to the cluster 0 dened by Gravendeel et al.8 as a group of samples with a high amount of non-neoplastic brain tissue. These tumours were considered as too contaminated with normal brain tissue and therefore removed for further analyses.

Unsupervised consensus clustering analysis of the 141 remaining tumour samples was then performed using three types of omics data (transcriptomic arrays (n 141), miRNA

sequencing (n 137) and DNA methylation arrays (n 87))

independently. Transcriptome-based consensus clustering identied ve robust transcriptomic subgroups, while miRNA-based and methylation-based clustering both identied four subgroups (Fig. 1a). The transcriptomic classication was highly associated with the classications on the basis of the miRNA data (w2 P valueo1.0 10 36) and methylation data

(w2 P value o1.0 10 19). A multi-omics classication was

subsequently obtained by consensus clustering of these three partitions (Supplementary Fig. 1b). Remarkably, the ve resulting classes (C1C5) nearly perfectly matched the transcriptomic classication, thereby suggesting that mRNA expression proling would be sufcient to dene robust molecular classes among OTs. We further characterized those ve classes using SNP data and other histological and clinical annotations of the POLA tumours cohort. As expected the ve classes were strongly associated with IDH mutations (w2 P value o3.0 10 16) and with 1p/19q

co-deletion status (w2 P value o5.0 10 23) (Fig. 1b).

IDH-mutated non-1p/19q co-deleted OT clustered into cluster C3. Their genomic prole was characterized by chromosome 7 gain (54%), chromosome 11p loss (41%) and copy neutral loss (LOH) of chromosome 17p (68%) as shown in Fig. 2. IDH wild-type OT formed cluster C2 and had a genomic prole as typically observed in glioblastomas, characterized by gains of chromosome 7, EGFR amplications, CDKN2A deletions and losses of chromosome 10 (Fig. 2). As for 1p/19q co-deleted tumours, they were unexpectedly split into three different clusters C1, C4 and C5, thereby revealing previously unrecognized molecular heterogeneity among 1p/19q co-deleted OT.

Molecular characterization of 1p/19q co-deleted OT subtypes. On the basis of the previous results, we decided to focus on 1p/19q co-deleted OT. To check the robustness of the three previously related classes (C1, C4, C5), we repeated a consensus clustering analysis restricted to 1p/19q co-deleted tumours. For sake of clarity we limited this analysis to the transcriptome, which perfectly summarised multi-omics clusters (Supplementary Fig. 1). We identied three robust subgroups O1, O2, O3 matching almost perfectly the previously identied clusters C1, C4 and C5, respectively (w2 P value 1.0 10 30, Fig. 3a,b).

To validate the three subgroups of 1p/19q co-deleted OT, we performed an unsupervised consensus clustering of mRNA data using the 1p/19q co-deleted OT from three additional public cohorts (The Cancer Genome Atlas5 (TCGA), Gravendeel et al.8, and REMBRANDT project9, Supplementary Fig. 2). As in our dataset, unsupervised consensus clustering optimally partitioned each public dataset into three clusters. We analysed the correlation patterns of class centroids in both our dataset and public datasets, and observed a high similarity between all three-group partitions (Fig. 3c), thereby conrming our ndings.

In the POLA cohort, patients in O1 tended to be older than patients in O2 and O3 (48.7 years vs 44.8 years, t-test P value 0.08) and had less frequently seizures at diagnosis

(43 vs 79%, Fisher test P value 0.001). This may be related to

the fact that grade II oligodendrogliomas were mostly present in O2 and O3, while O1 tumours consisted nearly exclusively of anaplastic oligodendrogliomas (Fisher test P value 9.9 10 6).

Consistently, O1 was signicantly enriched in tumours demonstrating microvascular proliferation (89 vs 45%, Fisher test P value 2.0 10 5) and necrosis (36 vs 10%, Fisher test

P value 0.001). The two 1p/19q co-deleted tumours classied as

glioblastomas with oligodendroglioma component (GBMO) according to the 2007 WHO classication clustered with O1. The genomic prole of O1 tumours differed from O2 and O3 tumours with signicantly higher frequencies of chromosomes 4, 9p, 14q and 18q losses, even when considering only grade III tumours (see Fisher tests P values in Supplementary Fig. 3); 66% of O1 tumours showed at least 1 loss of those 4 chromosomal regions, and 32% of O1 tumours had at least 2 or more regions lost. Tumour cellularity was higher in O1 and O2 tumours than in O3 tumours suggesting that this last subtype may have a more inltrative growth pattern.

2 NATURE COMMUNICATIONS | 7:11263 | DOI: 10.1038/ncomms11263 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms11263 ARTICLE

All oligodendroglial tumours (POLA cohort)

mRNA DNA methylation miRNA seq

a

Consensus classification of each data type

Clusters of clusters

Clusters of clusters

hTERT

mRNA

DNA methylation

miRNA seq

b

P value (chi2 test)

1.67e102.43e13

1.97e131.55e40

4.12e231.22e122.54e165.76e118.96e06

4.3e082.23e07

WHO 2007 Histologic grade ATRX IHC

p53 IHC

IGP Gravendeel

C1 C2 C3 C5

C4

Existing classification

systems

WHO 2007

Histogic grade

Verhaak subtype

Gravendeel IGP

p53 accumulation ATRX loss

1p/19q co-deletion

Common

genomic

events

CIMP

IDH1/IDH2

CIC

Protein

IHC

Verhaak

9 17 18 22

II III IV

Oligodendroglioma

Cla

Loss of expression

No or low expression

Oligoastrocytoma

Mes

Protein expression

GBMO

Neu Pro

Astrocytoma

High expression

Glioblastoma

1p/19q co-deletion

Gene alterations

No information

Present

Mutation

Absent WT Hypermethylator phenotype

Figure 1 | Histo-molecular characterization of the ve subtypes of oligodendroglial tumours robustly identied in POLA cohort. (a) An integrative cluster of cluster approach was used to dene robust molecular subtypes of 141 oligodendroglial tumours. Consensus clustering was used to assign molecular classes on the basis of mRNA data, DNA methylation data and microRNA data independently. Consensus clusters of clusters were subsequently identied on the basis of the classes labels resulting from previous independent classications. (b) Clinical annotations and common genomic alterations associated to each subtype. Genomic alterations were identied through the analysis of SNP arrays. For each clinical and molecular characteristic we performed w2 tests to assess the strength of association with the ve-class system.

According to Gravendeels and Verhaaks classications8,10, most tumours within cluster O1 and O2 were classied as IGP 9 and as proneural, while most tumours within O3 were classied as IGP17 and as either proneural, neural or mesenchymal.

Tests for differential gene expression between subgroups and gene-set enrichment analysis demonstrated that the three subgroups were characterized by the expression of specic markers of differentiation (Fig. 4; Supplementary Table 2; Supplementary Data 1). O1 tumours were characterized by a higher expression of oligodendrocyte precursor cell (OPC) markers, especially GPR17 (ref. 11) and CCND1 (ref. 12) which was validated by immunohistochemistry (Fig. 3b); O2 tumours strongly over-expressed neuron markers13,14 and genes implicated in neurogenesis; and O3 tumours specically expressed mature oligodendrocyte markers14,15. Astrocytic markers were overexpressed in both O2 and O3 compared to O1.

O1 tumours also overexpressed cell-cycle genes, genes implicated in glioma angiogenesis, and key epithelial mesenchymal transition markers (for example, TWIST1, SNAI2

and POSTN), a feature associated with tumour progression in gliomas16 and observed in glioblastomas17. The most striking differential activity among the oncogenic pathways was observed in O1 tumours where several gene sets reecting MYC activity were found among the most signicantly deregulated gene sets (GSA score41, P valueo0.05; Supplementary Table 2).

As in our dataset, O1 TCGA tumours tended to originate from older patients (mean age at diagnosis 51.8 vs 42.8 years, t-test P value 0.08). They were also associated to a higher grade and

with more frequent losses of chromosomal arms 9p and 14q (Fisher test P values 0.04 and 0.009, respectively,

Supplementary Fig. 4a). With the exception of NOTCH1, which was almost never found mutated in O3 subtype, TCGA exome data analysis did not identify any mutation signicantly associated with a specic subgroup (Supplementary Fig. 4b). As in the POLA dataset, CIC mutations were found in all subgroups. Gene enrichment analysis in each TCGA class was consistent with O1, O2 and O3 gene expression characteristics in the POLA cohort (Supplementary Fig. 5). In particular, a striking

NATURE COMMUNICATIONS | 7:11263 | DOI: 10.1038/ncomms11263 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 3

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms11263

prole was also higher in O1 tumours (t-test P value r0.013; Supplementary Fig. 6). To determine which molecular mechanisms could trigger MYC activation in O1 tumours, we looked for genomic, epigenetic and post-transcriptional events reported to enhance the MYC pathway activity in both the POLA and TCGA datasets.

At the genomic level, gains of MYC locus and losses of MAX (Myc-associated factor X), a negative regulator of MYC18, were more frequent in O1 than in O2 and O3 tumours (t-test P values 0.02 and 0.0002, respectively; Fig. 5a). At

the epigenetic level, MYC exon 3 hypomethylation19,20 was specically associated with O1 tumours in both datasets (P valueo0.0001) and correlated with a higher MYC expression (Fig. 5a and Supplementary Fig. 7). In addition, two negative regulators of MYC, mir34b and mir34c2123, were down-regulated in O1 tumours and their transcription start sites, lying within the mir34b/c CpG island, were hypermethylated in both POLA and TCGA datasets (Supplementary Fig. 7).

These four mechanismsMYC genomic gain (9% of O1 tumours), MAX genomic loss (35% of O1 tumours), MYC exon 3 hypomethylation (20% of O1 tumours) and mir34b/c locus hypermethylation (28% of O1 tumours)were not all required to observe an increase of MYC activity. Consistently, MYC targets mean expression increased in samples having at least one of this events, in both POLA and TCGA datasets (Fig. 5b). Moreover, MYC alterations (genomic gain or exon 3 hypomethylation) tended to be exclusive with mir34b/c locus hypermethylation (binomial test P value 0.003; Fig. 6) and

MAX genomic losses (binomial test P value 8.0 10 4; Fig. 6).

In the TCGA dataset, MYC gain and MYC exon 3 hypomethylation never occurred with the FUBP1 mutation, which is thought to increase MYC activity24.

Taken together, these results suggest that various molecular mechanisms concur to MYC activity in O1 tumours: genomic alterations, hypomethylation and down-regulation of its silencers mir34b and mir34c through hypermethylation of their promoter region.

Association with survival. In the POLA, Gravendeel and REMBRANDT cohorts we did not observe any signicant association between O1/O2/O3 partition and prognosis. However, due to still limited follow-up, median overall survival was not reached in the POLA cohort. As for Gravendeel and REMBRANDT cohorts, their sizes were limited and their median survival (6 years) was not fully representative of the median survival usually observed in 1p/19q co-deleted tumours25 (410 years) (Supplementary Table 3). In contrast, a remarkable association with prognosis was observed within the TCGA cohort (Fig. 7a), and was independent of grade and age (Supplementary Fig. 8). Consistently, 1p/19q co-deleted tumours with the highest MYC activity score had a worse prognosis (log-rank P value 0.01; Supplementary Fig. 9). When pooling survival

data from the four cohorts, there was a trend towards an association of O1 subtype with a worse prognosis (log-rank P value 0.049; Fig. 7b). Moreover, also consistent with the higher

aggressiveness of O1 subtype, analysis of all patients for whom treatment data was available showed that an initial treatment without radiotherapy (that is, with initial follow-up or with chemotherapy alone) was associated with shorter survival in O1 but not in O2 and O3 tumours (log-rank P value 0.052; Fig. 7c).

DiscussionIn agreement with the TCGA low-grade glioma study, we show here a strong correlation between the classication of OT based either on the different omics separately or on the integrated

Genomic profile of oligodendroglial tumour subgroups (POLA cohort)

C1

C2

C3

C4

C5

1

15

17

19

21

3

5

7

9

11

13

2

4

6

90% 60% 30%

30% 60% 90%

90% 60% 30%

30% 60% 90%

90% 60% 30%

30% 60% 90%

90% 60% 30%

30% 60% 90%

90% 60% 30%

30% 60% 90%

8 10 12 14 16 18 20 22 60%

30%

30% 60%

8 10 12 14 16 18 20 22 60%

30%

30% 60%

8 10 12 14 16 18 20 22 60%

30%

30%

60%

4 9p 14q

18q

5

7

9

11

13

1 2

3

15

17

19

21

EGFR

4

6

+7

CDK4

CDKN2A

10

5

7

9

11

13

1 2

3

15

17

19

21

4

6

+7q

LOH

17p

11p 19q

TP53

5

7

9

11

13

1 2

3

15

17

19

21

4

6

8 10 12 14 16 18 20 22 60%

30%

30%

60%

8 10 12 14 16 18 20 22 60% 30%

30%

60%

1

3

15

17

19

21

5

7

9

11

13

2

4

6

CN gain Amp

CN loss Hdel

LOH

Figure 2 | Genomic prole of oligodendroglial tumour subgroups. We analysed SNP data from 131 of the 141 oligodendroglial tumours characterized. Frequency of genomic amplications (Amp), copy number gains (CN Gain), copy number losses (CN loss), homozygous deletions (Hdel) and copy neutral LOH events (LOH) are displayed for each of the ve subtypes. Left axes show frequencies of CN gains, CN losses and LOH events. Right axes show frequencies of amplications and homozygous deletions. The most frequently altered chromosome arms are highlightened in red (gain), blue (loss) or black (copy neutral LOH).

enrichment in gene sets related to the MYC pathway was also observed in TCGA O1 tumours.

Multi-level deregulation of MYC activity in O1 tumours. The MYC pathway activity was assessed in each tumour as the mean expression of MYC target genes. Consistently this measure was higher in O1 than in O2 and O3 tumours (t-test P value o1.0 10 3; Supplementary Fig. 6). MYC expression

4 NATURE COMMUNICATIONS | 7:11263 | DOI: 10.1038/ncomms11263 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms11263 ARTICLE

1p/19q co-deleted oligodendroglial tumours (POLA cohort)

a c

TCGA.1

TCGA.3

TCGA.2

Centroids correlation

(Pearson) 1

0.5

0

0.5

REMBRANDT.3

REMBRANDT.2

REMBRANDT.1

POLA.O3

POLA.O2

POLA.O1

O3

O2

O1

b

P value (association test)

Gravendeel.3

Gravendeel.2

Gravendeel.1

Histologic features

Patient information

1p/19q OT subclass

Clusters of clusters

WHO 2007

Verhaak subtype

CIC mutation

Gravendeel IGP

Seizure at diagnosis Age at diagnosis

Tumour cellularity

4.84e31

3.34e10

0.1450.265

0.445

0.00143

0.308

MVP

Seizure

Necrosis

% Tumour cells 0 100

Histologic Grade

2.33e05

1.64e06

Cyclin D1 IHC

5.39e05

Microvascular proliferation Necrosis

Age<20 >80

No information

WHO 2007 Verhaak IGP Gravendeel Gene alterations Cyclin D1 IHC

GBMO

Oligoastrocytoma

0.0375

0.0198

Histologic grade

O1 O3

O2

Oligodendroglioma II

III IV

Mes Neu Pro

9 MutationWT Low expression

High expression

17

Figure 3 | External validation of the 3 subtypes of 1p/19q co-deleted OT and further characterization. (a) Co-classication matrix resulting from consensus clustering analysis of mRNA data restricted to 1p/19q co-deleted OT samples (n 80). The strength of the blue colour is proportional to the

frequency at which samples have been clustered together. (b) Clinical and histo-molecular annotations associated to each of the O1 (n 47), O2 (n 17),

or O3 (n 16) subtypes of 1p/19q co-deleted tumours. We performed statistical tests for each variable to assess the strength of association with the

three-class partition (Fisher tests for categorical variables and KruskalWallis tests for continuous variables) and displayed the corresponding P values on the right. The clusters of clusters annotation refers to the previous ve-class partition presented in Fig. 1a and gives the correspondence with O1, O2 and O3 classes of 1p/19q co-deleted tumours. (c) Correlation matrix of class centroids derived from an unsupervised consensus classication of each dataset independently. TCGA5, REMBRANDT9 and Gravendeel8 co-deleted tumours could be partitioned into three stable classes on the basis of their expression data. Hierarchical clustering of the resulting class centroids (marked as 1, 2 and 3 for each dataset) on the basis of the Pearson correlation distance identies three meta-clusters, each of them including one of the O1, O2 and O3 class centroids of our POLA discovery cohort, and one class from each other public dataset. We could therefore assign each meta-class to one of our dened O1, O2 and O3 subtypes. The three meta-classes are delimited with black squares with their corresponding O class name.

clustering of the omics (cluster of clusters analysis) and the 1p/ 19q co-deletion, and IDH mutation status5. These ndings further illustrate the robustness of the subgroups dened by these two biomarkers and support their integration into the revised classication of diffuse gliomas6. Moreover, because of its enrichment in 1p/19q co-deleted gliomas, our study identied three expression-based subgroups within these tumours and robustly reproduced this classication in public datasets through unsupervised analysis of 1p/19q co-deleted glioma samples.

The three subgroups of 1p/19q co-deleted tumours had different patterns of differentiation related to OPC, astrocytic, neuronal and oligodendrocytic marker genes expression. O1 OPC-like gliomas had a more aggressive histological and genomic prole with frequent chromosome 9p and 14q losses. It remains to be determined whether these three subgroups correspond to different oncogenic pathways or to different steps during oligodendrogliomagenesis. Yet, the absence of clear differences

regarding the somatic mutational landscape of the three subgroups rather discards the rst hypothesis. The second hypothesis is supported by the higher age observed among O1 patients in both POLA and TCGA cohorts and by the fact that MYC activation which was frequently observed in O1 tumours, has recently been implicated in the malignant progression of IDH mutant gliomas26. Besides, strong evidence suggests that OPC are the cell of origin of oligodendrogliomas27. These cells can differentiate into oligodendrocytes, astrocytes and may also differentiate into neural cells28. Therefore, in the more differentiated O2 and O3 subgroups, tumour cells could still be able to differentiate, while this differentiation capacity would be lost in the O1 OPC-like tumours as additional genomic alterations are acquired. The study of the gene expression prole of initial and recurrent tumours would be of great interest to determine whether tumours from the differentiated groups can evolve into OPC-like tumours over time. The better

NATURE COMMUNICATIONS | 7:11263 | DOI: 10.1038/ncomms11263 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 5

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms11263

Transcriptomic profile of 1p/19q co-deleted tumour subroups (POLA cohort)

O1 O2 O3 mRNA expression +

Astrocytic markers (e.g. AQP4,

PEA15, CHI3L1)

OPC markers (e.g. OLIG2, CSPG4, GPR17, CCND1)/ cell cycle genes/ MYC targets gene sets from litterature

Neurogenesis (e.g. NEUROD1, FOXG1, SEMA6A)

Neuron markers (e.g. NEFM, NEFL, SNAP25)

Mature oligodendrocyte markers(e.g. MBP, MOBP, MOG)

Immune response/microglia markers(e.g. CD163, CD14)

Angiogenesis (e.g. VEGFA, VEGFR, ANGPT2)

Figure 4 | Heatmap of mRNA expression prole from the top most up-regulated probe sets in each subtype. We performed moderated t-tests to analyse differential expression of Affymetrix HG-U133 Plus 2.0 arrays probe sets in each subtype compared to the others, and selected the top 1,000 probe sets with the highest fold-change among all signicantly deregulated probe sets (P valueo0.05, no adjustments for multiple comparisons). For each of the eight clusters of probe sets highlighted on the heatmap we performed gene-set enrichment analysis and annotated the clusters on the right with the most relevant signicantly enriched gene sets (hypergeometric test P valueo0.05) and corresponding relevant gene markers.

1p/19q co-deleted tumours (POLA + TCGA cohorts)

a

O1 O2 O3

P value

1p/19q codeleted classes

mir34b/c TSS hypermethylation

+

+

O1 vs non-O1

MYC targets mean expression

<1.7e04

5.43e050.04420.126

MYC mRNA expression

MYC exon 3 hypomethylation

MAX genomic loss

MYC mRNA expression

<0.0130.0204

MYC genomic gain

2.27e04

FUBP1 mutation

MYC targets mean expression

Genomic gain

Genetic and epigenetic alterations

Hypermethylation

Hypomethylation

No data

Mutation

Genomic loss

WT

b

POLA TCGA

0.4

0.2

0.0

0.2

0.4

O1

MYC targets mean

expression

MYC targets mean

expression

1.0

0.5

0.0

0.5

O2

O3

1 event

0 event 1 event

0 event

P val = 0.004 P val = 0.008

Figure 5 | Overview of MYC-related genomic and epigenomic alterations frequently observed in O1 tumours. (a) Summary of MYC-related genetic and epigenetic alterations in pooled data from both POLA (80 co-deleted gliomas) and TCGA (131 co-deleted gliomas) cohorts. The pooled sample sizes of O1, O2 and O3 classes are respectively 90, 61 and 60 tumours. The mean expression of MYC targets was computed for each sample to get a measure of MYC activation. (b) Relation between the presence of at least one MYC deregulation event (MYC genomic gain, MAX genomic loss, MYC exon 3 hypomethylation, mir34b/c TSS hypermethylation) and MYC activity measured through the mean expression of MYC targets. For each dataset, y axis show the mean expression values after centring on the samples. For each box and whiskers plot, bottom and top of the boxes are the rst and third quartile of the data and whiskers represent the lowest (respectively highest) data point still within 1.5 interquartile range of the lower (respectively upper) quartile. Bold lines represent median values.

prognosis of the differentiated subgroups also argue towards the use of differentiation therapies in O1 tumours, such as inhibitors of the membrane receptor GPR17, which was highly expressed in the OPC-like group and has been suggested to block OPC differentiation29,30. Interestingly, such inhibitors are being developed to promote myelin repair in multiple sclerosis29.

In the O1 group, the MYC pathway appeared as a particularly important oncogenic pathway. FUBP1 inactivating mutations are thought to activate MYC24. However, they were not signicantly associated with the OPC-like group (Supplementary Fig. 4). Here, we identied four distinct molecular mechanisms that could concur to increase MYC activity in the OPC-like group:

6 NATURE COMMUNICATIONS | 7:11263 | DOI: 10.1038/ncomms11263 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms11263 ARTICLE

MYC locus genomic gain, MAX locus genomic loss, hypomethylation of MYC exon 3, and down-regulation of MYC silencers mir34b and mir34c through promoter hypermethylation. MYC locus genomic gain, together with MAX and FBXW7 locus genomic losses have been suggested to activate the MYC pathway during the malignant progression of IDH-mutated gliomas26. FBXW7 locus genomic loss was not signicantly associated with the OPC-like group, but MAX locus genomic loss at 14q was observed in 35% of O1 tumours. MAX, a MYC-associated factor, is a tumour suppressor gene whose mutations cause hereditary pheochromocytoma31. MYC locus genomic gain at 8q24 was observed in about 10% of O1 tumours. Interestingly, a strong association has been shown between 1p/19q co-deleted IDH-mutated gliomas and SNPs mapping to the 8q24 locus, which is rich in long non-coding RNA that may modulate MYC expression32. Hypomethylation of MYC exon 3 at the same CpG position than in our O1 tumours (Chr8: 128,752,988-hg19, GRCh37) has been reported in myeloma, leukaemia, B-cell malignancies and colorectal cancer19,20,33. This particular site seems to be important for MYC expression auto-regulation. In colorectal carcinoma, partial

MYC pathway deregulation in O1 tumours (POLA + TCGA cohorts)

O1 tumours (POLA+TCGA, 90 samples)

MAX copy number loss

MYC MYC copy number gain

exon 3 hypomethylation

mir34b/c hypermethylation

68%

Mutual exclusion: P val = 0.003

MYC

9% 20%

mir34b/c 27%

Mutual exclusion: P val = 0.0008

MAX

35%

Mutual exclusion: P val = 0.003

Figure 6 | Focus on O1 tumours and their specic MYC signalling related alterations. Percentages refer to the proportion of O1 tumours (pooled data from POLA and TCGA) harbouring the alterations.

P values refer to one-sided binomial tests, which assess the probability that two of the three genetic loci are both altered in the same tumour sample.

Overall survival of 1p/19q co-deleted tumours

a

b

TCGA LGG patients All patients from pooled cohorts (POLA, TCGA, Gravendeel, REMBRANDT)

Pooled cohorts (POLA,TCGA, Gravendeel, REMBRANDT)

O1 patients O2&O3 patients1

0.8

0.6

0.4

0.2

10 20 30 40 50 60

0

Fraction of patients alive

1

0.8

0.6

0.4

0.2

Fraction of patients alive

1

0.8

0.6

0.4

0.2

log-rank testP value = 0.001

Other (n = 79) O1 (n = 39)

log-rank testP value = 0.049

Other (n = 160) O1 (n = 117)

0

0

0

18 42 66 90 114 144 174

0

18 36 54 72 90 108

Months Months

c

Fraction of patients alive

1

0.8

0.6

0.4

0.2

Fraction of patients alive

log-rank testP value = 0.052

No RT (n = 19) RT (n = 72)

log-rank testP value = 0.34

No RT (n = 66) RT (n = 53)

0

0

10 20 30 40 50 60

0

Months

Months

Figure 7 | Overall survival of 1p/19q co-deleted tumours (a) Overall survival of TCGA patients with 1p/19q co-deleted tumours according to O1 subtype membership. We used the available clinical data from 118 patients with co-deleted tumours (59 with grade III tumours and 59 with grade II tumours). (b) Overall survival of all 278 patients with 1p/19q co-deleted tumours after pooling patients with available clinical data from TCGA (n 118 patients),

POLA (n 80 patients), Gravendeel (n 42 patients) and REMBRANDT (n 37 patients) cohorts. (c) Overall survival of O1, and other O2 or O3 patients

who did or did not receive radiotherapy as initial treatment after surgery. We used data from 222 patients with 1p/19q co-deleted tumours that were pooled from the four cohorts: POLA (n 75 patients), TCGA (n 66 patients), REMBRANDT (n 49 patients) and Gravendeel (n 32 patients). Patients were

included if their treatment and survival data were available and if they had not deceased within the rst 3 months after diagnosis so that they could have effectively received radiotherapy. For each subgroup of patients (with O1, O2 or O3 tumours) we compared the ve-year survival of patients treated with an initial radiotherapy (RT, red curves)combined or not with chemotherapyagainst patients who had not received initial radiotherapy and were managed with initial follow-up or with chemotherapy alone (No RT, grey curves).

NATURE COMMUNICATIONS | 7:11263 | DOI: 10.1038/ncomms11263 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 7

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms11263

hypomethylation of this position was associated with the deregulation of cell proliferation. Mir34b and mir34c have been broadly reported to be negative regulators of MYC, and silencing of miR34b/c locus through promoter hypermethylation has been reported to up-regulate MYC expression2123. However, 32% of O1 tumours didnt harbour any of these four alterations despite showing a high MYC activity. Further analysis may identify other alterations impacting the MYC signalling pathway. OPC-like tumours might be candidates for strategies aiming at inhibiting MYC activity such as bromodomain and extraterminal bromodomain (BET) inhibition that has been shown to suppress MYC transcriptional activity in several cancers34 and to inhibit cell growth in IDH1-mutant glioma primary cell cultures26.

The clinical signicance of the three gene expression subgroups of 1p/19q co-deleted tumours remains to be determined since association with survival was only observed in the TCGA dataset and not in the three other cohorts. However, the poorer outcome associated with classication into the O1 group and with high MYC activity would be in line with previous studies showing that (1) necrosis, 9p loss and a high number of genomic alterations are associated with worse prognosis in 1p/19q co-deleted tumours35,36, and (2) increased MYC activity is associated with malignant progression and worse prognosis in IDH-mutated tumours26,37. Identifying patients with 1p/19q co-deleted tumours with poorer outcome is an important issue. Since these tumours are usually chemo sensitive, these patients might be candidates to receive more intensive chemotherapy regimens. On the other hand, patients with a favourable molecular prole might be the best candidates to benet from less intensive treatment strategies, for example, initial treatment with chemotherapy alone to reduce the potential side effects of brain radiotherapy. The present study suggests that such a strategy might be appropriate in O2 and O3 but not in O1 tumours. Future studies will have to determine efcient molecular markers to rapidly label 1p/19q co-deleted patients according to this stratication.

Methods

Patient samples and consent. Samples were obtained with informed and written consent after approval of the institutional review boards of respective hospitals participating in the POLA network. All patients were aged 18 years or older at diagnosis, and tumour histology was centrally reviewed and validated according to WHO guidelines38. A total of 179 samples were included in this study: 156 gliomas with an oligodendroglial phenotype, as well as 11 glioblastomas, 2 diffuse astrocytomas, 9 normal brain samples and 1 NOS sample. A summary of each sample of the tumour cohort and respective pathological information on the patients is provided in Supplementary Table 1.

DNA and RNA extraction. DNA and total RNA were extracted from frozen tumour samples using the iPrepChargeSwitch Forensic Kit and the RNeasy Lipid Tissue Mini Kit (Qiagen), respectively. DNA and RNA integrity and quantity were assessed on the basis of the quality control criteria established by CIT(Cartes dIdentit des Tumeurs) programme protocols (http://cit.ligue-cancer.net

Web End =http://cit.ligue-cancer.net). A 1-mg volume from each DNA and RNA sample was used for SNP array experiments (outsourced to the Integragen Company Paris, France) and to perform the gene expression analysis, respectively.

SNP arrays analysis. Illumina SNP arrays were used to analyse the DNA samples from 161 tumour samples (74 Illumina HumanCNV610-Quad v1.0, 52 HumanCNV370, 34 HumanOmniExpress-12v1 and 1 HumanCore-12v1). Integragen SA (Evry, France) carried out hybridization, according to the manufacturers recommendations. The BeadStudio software (Illumina) was used to normalize raw uorescent signals and to obtain log R ratio (LRR) and B allele frequency (BAF) values. Asymmetry in BAF signals due to bias between the two dyes used in Illumina assays was corrected using the tQN normalization procedure.39 We used the circular binary segmentation algorithm40 to segment genomic proles and assign corresponding smoothed values of log R ratio and B allele frequency. The Genome Alteration Print method was used to determine the ploidy of each sample, the level of contamination with normal cells and the allele-specic copy number of each segment41.

mRNA expression proling and analysis. The IGBMC Microarray and Sequencing Platform performed mRNA expression proling using HumanGeneChip HG-U133 Plus 2.0 arrays (Affymetrix) for the 179 samples from the study. We used the RMA algorithm (Bioconductor affy package) to normalize the data. Probe set intensities were then averaged per gene symbol.

We used the Bioconductor ConsensusClusterPlus package for consensus clustering analysis and identication of homogeneous gene expression clusters. The 5% most variant probe sets were selected to determine the consensus partitions of the data set in K clusters (for K 2, 3, ..., 8). Computations were performed on the

basis of the 1,000 resampling iterations of hierarchical clustering, using Pearsons dissimilarity as the distance metric and Wards method for linkage analysis. To determine the optimal number of clusters, we used the cumulative distribution functions (CDFs) of the consensus matrices and considered both the shape of the functions and the area under the CDF curves, as previously described42.

We used the Bioconductor package limma to test for gene differential expression between different conditions43.

DNA methylation proling and analysis. We analysed whole-genomeDNA methylation in 104 tumour samples using the Illumina Innium Human-Methylation450 Beadchips. Integragen SA (Evry, France) carried out microarray experiments and hybridized to the BeadChip arrays following the manufacturers instructions. Illumina GenomeStudio software was used to extract the beta value DNA methylation scores for each locus together with detection P values.

As described elsewhere44, we replaced data points with detection P value40.05 with NA values. We also masked data points as NA for probes that contained SNPs or overlapped with a repetitive element that was not uniquely aligned to the human genome or regions of insertions and deletions in the human genome. Homogeneous tumour subgroups with similar methylation proles were identied using consensus clustering. We used the Bioconductor package ConsensusClusterPlus as described above, using the 5% most variant CpG sites, the Euclidean distance metric, and R ward.D method for linkage analysis.

We determined CpG Island Methylator Phenotype (CIMP) by restricting the consensus clustering analysis to CpG sites located within CpG islands. Samples with a CIMP phenotype were determined according to the classication results from the partition in two classes. Samples falling within the class showing strong hypermethylation were assigned a positive CIMP status.

miRNA proling and analysis. miRNA proling was performed on 177 samples. A PCR barcoding method45 was used to prepare multiplexed miRNA libraries that were sequenced by Integragen SA (Evry, France) on an Illumina HiSeq 2000 sequencer. Image analysis, base calling, demultiplexing and conversion of BCL to FASTQ format were performed using Illumina CASAVA 1.8.2 software. MirExpress software46 was used to remove adaptor sequences. MiRanalyzer0.3 software47 was used to process FASTA les for each sample and to quantify read counts for each miRNA referenced in mirBase74 v18.

Unsupervised classication was performed using 757 miRNAs that were expressed (410 reads) in at least two samples. The miRNA counts were log2 transformed, divided by the total number of reads in each sample and centred on the mean expression level of each gene. Consensus clustering was performed as described above. Pearsons dissimilarity was used as the distance metric and Wards method for linkage analysis. We determined the optimal number of clusters on the basis of the CDF curves.

We used the Bioconductor package limma to test for microRNA differential expression between different conditions43.

Immunohistochemical staining. Immunohistochemistry was performed on 4-mm-thick sections of formalin-xed parafn embedded blocks with a ventana Benchmark XT Device. The following antibodies were used after antigen retrieval to assess ATRX (anti-ATRX, Sigma, polyclonal, dilution 1/400), p53 (anti-p53, Dako clone DO.7, dilution 1/200) and CCND1 (anti-CCND1, Ventana, clone SP4). p53 protein was dened as highly expressed when we observed a strong nuclear expression in more than 10% of the nuclei.

IDH and CIC mutations. IDH1 codon 132 and IDH2 codon 172 were sequenced using the Sanger method with the following primers: IDH1-Forward: TGTGT TGAGATGGACGCCTATTTG; IDH1-Reverse: TGCCACCAACGACCAAGTC; IDH2-Forward: GCCCGGTCTGCCACAAAGTC and IDH2-Reverse: TTGGCAGACTCCAGAGCCCA.

Coding exons (120) of the CIC gene were rst amplied using primers used by Gleize et al.48. Primers are available in Supplementary Table 4. PCR products were puried conforming to the Agencourt AMPure XP PCR purication protocol (Beckman Coulter) with the Biomek 3000 Automation Workstation. Universal tailed amplicon resequencing approach (454 Sequencing Technology, Roche) was used for the sequencing of coding exons of CIC. Sequences analysis was performed using CLC Genomics Workbench software.

pTERT mutations. The promoter region of TERT gene was amplied as follow: TERT-F: GGCCGATTCGACCTCTCT and TERT-R AGCACCTCGCGGTAGT

8 NATURE COMMUNICATIONS | 7:11263 | DOI: 10.1038/ncomms11263 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms11263 ARTICLE

GG; 3 min at 94 C; 35 cycles at 94 C15 s, 60 C45 s, 72 C1 min, with a nal step at 72 C for 8 min. PCR products were then puried with the Agencourt AMPure XP PCR purication protocol (Beckman Coulter). Puried PCR products were used as templates for the sequencing reaction performed with the Big-Dye Terminator Cycle Sequencing Ready Reaction (Perkin Elmer). Extension products were puried with the Agencourt CleanSEQ protocol according to manufacturers instructions (Beckman Coulter). Puried sequences were analysed on an ABI Prism 3730 DNA Analyzer (Applied Biosystems).

Cluster of clusters analysis. We performed a consensus clustering of 85 and 141 tumours samples on the basis of the results from mRNA, DNA methylation and miRNA consensus clustering analyses. The samples were clustered on binary variables for each of the previously dened classes: ve variables for mRNA classes, four variables for DNA methylation classes and four variables for miRNA classes. For each sample, the class variables had values 1 when the sample was in the class, 0 if in another class of the partition, NA when the sample was not classied at the given molecular level. Pearsons dissimilarity was used as the distance metric and Wards method was used for linkage analysis.

Analysis of public data. We downloaded TCGA Low Grade Glioma data with last update on 17th October 2014. 1p/19q co-deleted status was assigned by using Gistic2 results by chromosome arm as found on the TCGA data portal. A total of 131 gliomas were labelled as 1p/19q co-deleted.

REMBRANDT and Gravendeel mRNA data were downloaded from public databases (accession codes are respectively GSE16011 and E-MTAB-3073). For both datasets, 1p/19q co-deleted status was assigned using mRNA expression data. For each sample we computed the centred mean expression values of probe sets located on chromosome arms 1p and 19q and optimized the two-class partition (1p/19q co-deleted vs non-1p/19q co-deleted) of the samples according to these two values. We labelled respectively 42 and 58 gliomas as 1p/19q co-deleted in Gravendeel and REMBRANDT cohorts.

REMBRANDT survival data was downloaded from NIH (http://rembrandt.nci.nih.gov

Web End =http://rembrandt. http://rembrandt.nci.nih.gov

Web End =nci.nih.gov ) in august 2014 and treatment information from the G-DOC plus portal (http://gdoc.georgetown.edu/gdoc/

Web End =http://gdoc.georgetown.edu/gdoc/).

Validation of classication results on public datasets. We validated our mRNA classication results by applying the same unsupervised classication approach on the 1p/19q co-deleted samples of three additional public sample cohorts(TCGA Low Grade Glioma, Gravendeel cohort and REMBRANDT cohort). Then, for each class of each dataset, we computed a centroid prole on the basis of the samples within the class as the mean expression of the 10% most variant genes within the dataset. For each pair of classes to be compared, the 10% most variant genes were selected among the genes which were measured in both datasets.

We could then compare our initial classication system to the ones achieved on each public dataset using pairwise correlations between centroids to measure the inter-dataset similarity of the classes.

Gene-set enrichment analysis. We used the R package GSA49 to perform gene-set enrichment analysis for each molecular subtype compared to the others. Gene-set members lists were retrieved online from MSigDB, GO and SMD databases. Additional gene lists were added to this main set on the basis of specic publications of interest: OPC markers from Dougherty et al.15, VEGF activity markers from Dieterich et al.50 and MYC targets from Zeller et al.51. Gene list from Zeller et al. was also used to assign to each tumour a score of MYC activation on the basis of the mean expression of its targets.

Alterations in the MYC pathway. Genomic gains of MYC and genomic losses of MAX were estimated from the gain normal loss (GNL) values computed from SNP arrays. In POLA dataset, tumours verifying GNL 1 (resp. GNL 1) for all

SNP positions within MYC (resp. MAX) genomic region were considered to have a genomic gain of MYC. For TCGA dataset we used the public GNL data which are given at gene level only: Tumours with GNL 1 (resp. GNL 1) for MYC

(resp. MAX) were assigned a positive status for MYC (resp. MAX) genomic gain.

Hypomethylation of MYC exon 3 was measured from DNA methylation arrays. For both POLA and TCGA datasets we considered that tumours wererelatively hypomethylated on MYC exon 3 if the beta value at CpG position cg00163372 was o0.5.

Hypermethylation of mir34b/c genomic locus was also measured from DNA methylation arrays. We identied four CpG positions within CpG island on mir34b/C locus promoter region that were hypermethylated in O1 tumours (cg22879515, cg21881253, cg13767940 and cg23211240), and used the cg22879515 position to dene mir34b/c hypermethylation in both POLA and TCGA tumours. For each dataset we dened a tumour as hypermethylated for the locus if the beta value was greater than the mean beta value plus twice the standard deviation.

References

1. Ricard, D. et al. Primary brain tumours in adults. Lancet Lond. Engl 379, 19841996 (2012).

2. Wesseling, P., van den Bent, M. & Perry, A. Oligodendroglioma: pathology, molecular mechanisms and markers. Acta Neuropathol. (Berl.) 129, 809827 (2015).

3. van den Bent, M. J. et al. Adjuvant procarbazine, lomustine, and vincristine chemotherapy in newly diagnosed anaplastic oligodendroglioma: long-term follow-up of EORTC brain tumor group study 26951. J. Clin. Oncol. Off. J. Am. Soc. Clin. Oncol. 31, 344350 (2013).

4. Cairncross, J. G. et al. Benet from procarbazine, lomustine, and vincristine in oligodendroglial tumors is associated with mutation of IDH. J. Clin. Oncol. Off.J. Am. Soc. Clin. Oncol. 32, 783790 (2014).5. Cancer Genome Atlas Research Network et al. Comprehensive, integrative genomic analysis of diffuse lower-grade gliomas. N. Engl. J. Med. 372, 24812498 (2015).

6. Reuss, D. E. et al. ATRX and IDH1-R132H immunohistochemistry with subsequent copy number analysis and IDH sequencing as a basis for an integrated diagnostic approach for adult astrocytoma, oligodendroglioma and glioblastoma. Acta Neuropathol. (Berl.) 129, 133146 (2015).

7. Louis, D. N. et al. International Society Of NeuropathologyHaarlem consensus guidelines for nervous system tumor classication and grading. Brain Pathol. Zurich Switz 24, 429435 (2014).

8. Gravendeel, L. A. M. et al. Intrinsic gene expression proles of gliomas are a better predictor of survival than histology. Cancer Res. 69, 90659072 (2009).

9. Madhavan, S. et al. Rembrandt: helping personalized medicine become a reality through integrative translational research. Mol. Cancer Res 7, 157167 (2009).

10. Verhaak, R. G. W. et al. Integrated genomic analysis identies clinically relevant subtypes of glioblastoma characterized by abnormalities in PDGFRA, IDH1, EGFR, and NF1. Cancer Cell 17, 98110 (2010).

11. Chen, Y. et al. The oligodendrocyte-specic G protein-coupled receptor GPR17 is a cell-intrinsic timer of myelination. Nat. Neurosci. 12, 13981406 (2009).

12. Matsumura, N. et al. Coexpression of cyclin D1 and alpha-internexin in oligodendroglial tumors. Brain Tumor Pathol. 32, 261267 (2015).

13. Cahoy, J. D. et al. A transcriptome database for astrocytes, neurons, and oligodendrocytes: a new resource for understanding brain development and function. J. Neurosci. Off. J. Soc. Neurosci. 28, 264278 (2008).

14. Lein, E. S. et al. Genome-wide atlas of gene expression in the adult mouse brain. Nature 445, 168176 (2007).

15. Dougherty, J. D. et al. Candidate pathways for promoting differentiation or quiescence of oligodendrocyte progenitor-like cells in glioma. Cancer Res. 72, 48564868 (2012).

16. Nevo, I. et al. Identication of molecular pathways facilitating glioma cell invasion in situ. PloS ONE 9, e111783 (2014).

17. Tso, C.-L. et al. Primary glioblastomas express mesenchymal stem-like properties. Mol. Cancer Res. MCR 4, 607619 (2006).

18. Cascn, A. & Robledo, M. MAX and MYC: a heritable breakup. Cancer Res. 72, 31193124 (2012).

19. Stephenson, J., Akdag, R., Ozbek, N. & Mufti, G. J. Methylation status within exon 3 of the c-myc gene as a prognostic marker in myeloma and leukaemia. Leuk. Res. 17, 291293 (1993).

20. Sharrard, R. M., Royds, J. A., Rogers, S. & Shorthouse, A. J. Patterns of methylation of the c-myc gene in human colorectal cancer progression. Br. J. Cancer 65, 667672 (1992).

21. Lujambio, A. et al. A microRNA DNA methylation signature for human cancer metastasis. Proc. Natl Acad. Sci. USA. 105, 1355613561 (2008).

22. Leucci, E. et al. MYC translocation-negative classical Burkitt lymphoma cases: an alternative pathogenetic mechanism involving miRNA deregulation.J. Pathol. 216, 440450 (2008).23. Hermeking, H. The miR-34 family in cancer and apoptosis. Cell Death Differ. 17, 193199 (2010).

24. Hsiao, H. et al. Quantitative characterization of the interactions among c-myc transcriptional regulators FUSE, FBP, and FIR. Biochemistry (Mosc.) 49, 46204634 (2010).

25. Eckel-Passow, J. E. et al. Glioma groups based on 1p/19q, IDH, and TERT promoter mutations in tumors. N. Engl. J. Med. 372, 24992508 (2015).

26. Bai, H. et al. Integrated genomic characterization of IDH1-mutant glioma malignant progression. Nat. Genet. advance online publication 48, 5966 (2015).

27. Persson, A. I. et al. Non-stem cell origin for oligodendroglioma. Cancer Cell 18, 669682 (2010).

28. Tsoa, R. W., Coskun, V., Ho, C. K., de Vellis, J. & Sun, Y. E. Spatiotemporally different origins of NG2 progenitors produce cortical interneurons versusglia in the mammalian forebrain. Proc. Natl Acad. Sci. USA. 111, 74447449 (2014).

29. Fumagalli, M. et al. The ubiquitin ligase Mdm2 controls oligodendrocyte maturation by intertwining mTOR with G protein-coupled receptor kinase 2 in the regulation of GPR17 receptor desensitization. Glia 63, 23272339 (2015).

NATURE COMMUNICATIONS | 7:11263 | DOI: 10.1038/ncomms11263 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 9

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms11263

30. Hennen, S. et al. Decoding signaling and function of the orphan G protein-coupled receptor GPR17 with a small-molecule agonist. Sci. Signal. 6, ra93 (2013).

31. Comino-Mndez, I. et al. Exome sequencing identies MAX mutations as a cause of hereditary pheochromocytoma. Nat. Genet. 43, 663667 (2011).32. Enciso-Mora, V. et al. Deciphering the 8q24.21 association for glioma. Hum. Mol. Genet 22, 22932302 (2013).

33. Crossen, P. E. & Morrison, M. J. Methylation status of the 3rd exon of the c-MYC oncogene in B-cell malignancies. Leuk. Res. 23, 251253 (1999).34. Mertz, J. A. et al. Targeting MYC dependence in cancer by inhibiting BET bromodomains. Proc. Natl Acad. Sci. USA. 108, 1666916674 (2011).

35. Figarella-Branger, D. et al. Mitotic index, microvascular proliferation, and necrosis dene 3 groups of 1p/19q codeleted anaplastic oligodendrogliomas associated with different genomic alterations. Neuro-Oncol. 16, 12441254 (2014).

36. Alentorn, A. et al. Allelic loss of 9p21.3 is a prognostic factor in 1p/19q codeleted anaplastic gliomas. Neurology 85, 13251331 (2015).

37. Odia, Y., Orr, B. A., Bell, W. R., Eberhart, C. G. & Rodriguez, F. J. cMYC expression in inltrating gliomas: associations with IDH1 mutations, clinicopathologic features and outcome. J. Neurooncol. 115, 249259 (2013).

38. Louis, D. N. et al. The 2007 WHO classication of tumours of the central nervous system. Acta Neuropathol. (Berl.) 114, 97109 (2007).

39. Staaf, J. et al. Normalization of Illumina Innium whole-genome SNP data improves copy number estimates and allelic intensity ratios. BMC Bioinformatics 9, 409 (2008).

40. Venkatraman, E. S. & Olshen, A. B. A faster circular binary segmentation algorithm for the analysis of array CGH data. Bioinforma. Oxf. Engl 23, 657663 (2007).

41. Popova, T. et al. Genome alteration print (GAP): a tool to visualize and mine complex cancer genomic proles obtained by SNP arrays. Genome Biol. 10, R128 (2009).

42. Wilkerson, M. D. & Hayes, D. N. ConsensusClusterPlus: a class discovery tool with condence assessments and item tracking. Bioinforma. Oxf. Engl 26, 15721573 (2010).

43. Ritchie, M. E. et al. limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res. 43, e47 (2015).44. Hinoue, T. et al. Genome-scale analysis of aberrant DNA methylation in colorectal cancer. Genome Res. 22, 271282 (2012).

45. Vigneault, F. et al. High-throughput multiplex sequencing of miRNA. Curr. Protoc. Hum. Genet. Chapter 11, Unit 11.12.110 (2012).

46. Wang, W.-C. et al. miRExpress: analyzing high-throughput sequencing data for proling microRNA expression. BMC Bioinformatics 10, 328 (2009).

47. Hackenberg, M., Rodrguez-Ezpeleta, N. & Aransay, A. M. miRanalyzer: an update on the detection and analysis of microRNAs in high-throughput sequencing experiments. Nucleic Acids Res. 39, W132W138 (2011).

48. Gleize, V. et al. CIC inactivating mutations identify aggressive subset of 1p19q codeleted gliomas. Ann. Neurol. 78, 355374 (2015).

49. Subramanian, A. et al. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression proles. Proc. Natl. Acad. Sci. USA 102, 1554515550 (2005).

50. Dieterich, L. C. et al. Transcriptional proling of human glioblastoma vessels indicates a key role of VEGF-A and TGFb2 in vascular abnormalization.

J. Pathol. 228, 378390 (2012).51. Zeller, K. I., Jegga, A. G., Aronow, B. J., ODonnell, K. A. & Dang, C. V. An integrated database of genes responsive to the Myc oncogenic transcription factor: identication of direct genomic targets. Genome Biol. 4, R69 (2003).

Acknowledgements

This work is part of the national programme Cartes dIdentit des Tumeurs (CIT) (http://cit.ligue-cancer.net

Web End =http://cit.ligue-cancer.net), Prise en charge des oligodendrogliomes anaplasiques (POLA) Network funded by Institut National du Cancer and POLA Tumor Bank. The results published here are partly on the basis of data generated by The Cancer Genome Atlas (TCGA) pilot project established by the NCI and NHGRI. Information about TCGA and the investigators and institutions that constitute the TCGA research network can be found at http://cancergenome.nih.gov/

Web End =http://cancergenome.nih.gov/ . The research leading to these results has received funding from the program investissements davenir ANR-10-IAIHU-06, from the SIRIC-Marseille Glioma program (Grant INCa-DGOS-INSERM 6038) and from the ARTC (Association pour la Recherche sur les Tumeurs Crbrales). Frozen specimens were stored at Onconeurotek (GH Piti-Salptrire, Paris, France), NeuroBioTec (Groupement Hospitalier Est, 69677 Bron Cedex, France), AP-HM tumour bank (authorization number AC-2013-1786), Hpital Haut Levque CRB (33604, Pessac, France) CHU Montpellier (CCBH-M, 34825, Montpellier, France) IRCNA tumour bank (CHU Nantes, Institut de Cancrologie de louest, 44800 Saint-Herblain, France), CHU Saint-Etienne (CRB 42, 42055 Saint-Etienne, France), Tumorothque de Picardie (CHU Amiens, 80054 Amiens, France).

Author contributions

A.I., C.D., D.F.B., J.Y.D., A.d.R. and F.D. conceived the study; A.K., A.I., N.M.D., M.S., J.Y.D., D.F.B., A.d.R. and F.D. wrote the manuscript; A.K., E.L., A.d.R. and F.D. designed and reviewed statistical and bioinformatics analyses; C.C, C.C, K.M., A.J., E.U.C. and D.F.B. performed experiments; A.K. and E.L. performed bioinformatics analyses; C.C andC.C performed sample preparation; C.D., N.E. and C.C. reviewed samples annotations and conducted data management; all authors reviewed and contributed to the manuscript.

Additional information

Accession codes: The mRNA expression data, DNA methylation data and miRNA sequencing data have been deposited in ArrayExpress database under accessioncodes E-MTAB-3892, E-MTAB-3903 and E-MTAB-3901, respectively. The SNP array data has been deposited in the ArrayExpress database under accession codes E-MTAB-3905(Illumina Human610 Quad), E-MTAB-3907(Illumina HumanCNV370), E-MTAB-3896(Illumina HumanCore) and E-MTAB-3902(Illumina Human OmniExpress).

Supplementary Information accompanies this paper at http://www.nature.com/naturecommunications

Web End =http://www.nature.com/ http://www.nature.com/naturecommunications

Web End =naturecommunications

Competing nancial interests: The authors declare no competing nancial interests.

Reprints and permission information is available online at http://npg.nature.com/reprintsandpermissions/

Web End =http://npg.nature.com/ http://npg.nature.com/reprintsandpermissions/

Web End =reprintsandpermissions/

How to cite this article: Kamoun, A. et al. Integrated multi-omics analysis of oligodendroglial tumours identies three subgroups of 1p/19q co-deleted gliomas. Nat. Commun. 7:11263 doi: 10.1038/ncomms11263 (2016).

This work is licensed under a Creative Commons Attribution 4.0 International License. The images or other third party material in this article are included in the articles Creative Commons license, unless indicated otherwise in the credit line; if the material is not included under the Creative Commons license, users will need to obtain permission from the license holder to reproduce the material. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/

Web End =http://creativecommons.org/licenses/by/4.0/

POLA networkClovis Adam16, Marie Andraud17, Marie-Hlne Aubriot-Lorton18, Luc Bauchet19, Patrick Beauchesne20, Franck Bielle21, Claire Blechet22, Mario Campone23, Antoine F. Carpentier24, Ioana Carpiuc25,Dominique Cazals-Hatem26, Marie-Pierre Chenard27, Danchristian Chiforeanu28, Olivier Chinot29, Elisabeth Cohen-Moyal30, Philippe Colin31, Phong Dam-Hieu32, Christine Desenclos33, Nicolas Desse34, Frederic Dhermain35, Marie-Danile Diebold36, Sandrine Eimer37, Thierry Faillot38, Mlanie Fesneau39, Denys Fontaine40, Stphane Gaillard41, Guillaume Gauchotte42, Claude Gaultier43, Franois Ghiringhelli44, Joel Godard45, Edouard Marcel Gueye46, Jean Sebastien Guillamo47, Selma Hamdi-Elouadhani48,

10 NATURE COMMUNICATIONS | 7:11263 | DOI: 10.1038/ncomms11263 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms11263 ARTICLE

Jerome Honnorat49, Jean Louis Kemeny50, Touk Khallil51, Franois Labrousse52, Olivier Langlois53,Annie Laquerriere54, Delphine Larrieu-Ciron55, Emmanuelle Lechapt-Zalcman56, Caroline Le Gurinel57, Pierre-Marie Levillain58, Hugues Loiseau59, Delphine Loussouarn60, Claude-Alain Maurage61, Philippe Menei62, Marie Janette Motsuo Fotso63, Georges Noel64, Fabrice Parker65, Michel Peoch66, Marc Polivka67,Isabelle Quintin-Rou68, Carole Ramirez69, Damien Ricard70, Pomone Richard71, Valrie Rigau72,Audrey Rousseau73, Gwenaelle Runavot74, Henri Sevestre75, Marie Christine Tortel76, Fanny Vandenbos77, Elodie Vauleon78, Gabriel Viennet79, Chiara Villa80.

16Hpital Bictre, Pathology Department, 94275 Le Kremlin-Bictre, France; 17CHU Saint-Pierre de la Runion, Pathology Department, 97410 Saint-Pierre de la Runion, France; 18CHU Dijon, Pathology Department, 21000 Dijon, France; 19CHU de Montpellier, Neurosurgery Department, 34925 Montpellier, France;

20CHU Nancy, Neuro-oncology Department, 54035 Nancy, France; 21Groupe Hospitalier Piti-Salptrire, Neuropathology Department, 75013 Paris, France;

22CHR Orlans, Pathology Department, 45000 Orlans, France; 23Centre Ren Gauducheau, Medical Oncology Department, 44805 Saint-Herblain, France;

24Hpital Avicenne, Neurology Department, 93001 Bobigny, France; 25Clinique des Cdres, Medical Oncology Department, 31700 Cornebarrieu, France;

26Hpital Beaujon, Neurosurgery Department, 92110 Clichy, France; 27CHU Strasbourg, Pathology Department, 67098 Strasbourg, France; 28CHU Rennes, Pathology Department, 35033 Rennes, France; 29Hpital de la Timone, Assistance PubliqueHpitaux de Marseille, Neuro-oncology Department, 13385 Marseille, France; 30Institut Claudius Regaud, Radiotherapy Department, 31059 Toulouse, France; 31Clinique de Courlancy, Radiotherapy Department, 51100 Reims, France; 32Hpital de la cavale blanche, CHU Brest, Neurosurgery Department, 29609 Brest, France; 33Hpital Nord, CHU Amiens, Neurosurgery Department, 80054 Amiens, France; 34HIA Sainte-Anne, Neurosurgery Department, 83800 Toulon, France; 35Institut Gustave Roussy, Radiotherapy Department, 94805 Villejuif, France; 36CHU Reims, Pathology Department, 51092 Reims, France; 37CHU de Bordeaux-GH Pellegrin, Pathology Department, 33000 Bordeaux, France; 38Hpital Beaujon, Neurosurgery Department, 92110 Clichy, France; 39CHR Orlans, Radiotherapy Department, 45000 Orlans, France; 40CHU Nice, Neurosurgery Department, 06002 Nice, France; 41Hpital Foch, Neurosurgery Department, 92151 Suresnes, France; 42CHU Nancy, Pathology Department, 54035 Nancy, France; 43CH Colmar, Neurology Department, 68024 Colmar, France; 44Centre Georges-Franois Leclerc, Medical Oncology, 21079 Dijon, France; 45Hpital Jean Minjoz, CHU Besanon, Neurosurgery Department, 25030 Besanon, France; 46Hpital Dupuytren, CHU de Limoges, Neurosurgery Department, 87042 Limoges, France; 47CHU de Caen, Neurology Department, 14033 Caen, France; 48Hpital Lariboisire, Neurosurgery Department, 75475 Paris, France; 49Hospices Civils de Lyon, Hpital Neurologique, Neuro-oncology Department, 69677 Bron, France; 50CHU Clermont-Ferrand, Pathology Department, 63003 Clermont-Ferrand, France; 51CHU Clermont-Ferrand, Neurosurgery Department, 63003 Clermont-Ferrand, France; 52Hpital Dupuytren, CHU de Limoges, Pathology Department, 87042 Limoges, France; 53CHU Charles Nicolle, Neurosurgery Department, 76000 Rouen, France; 54CHU Charles Nicolle, Pathology Department, 76031 Rouen, France; 55CHU Poitiers, Neurology Department, 86000 Poitiers, France; 56CHU de Caen, Pathology Department, 14033 Caen, France; 57Hpital Henri Mondor, Neurosurgery Department, 94010 Henri Mondor, France; 58CHU Poitiers, Neurosurgery Department, 86000 Poitiers, France; 59CHU de Bordeaux-GH Pellegrin, Neurosurgery Department, 33000 Bordeaux, France; 60CHU Nantes, Pathology Department, 44093 Nantes, France; 61CHU de Lille, Pathology Department, 59037 Lille, France; 62CHU Angers, Neurosurgery Department, 49933 Angers, France; 63Hpital Nord, CHU Saint-tienne, Neurosurgery Department, 42270 Saint-Priest en Jarez, France; 64Centre Paul Strauss, Radiotherapy Department, 67065 Strasbourg, France; 65Hpital Bictre, Neurosurgery Department, 94275 Le Kremlin-Bictre, France; 66Hpital Nord, CHU Saint-tienne, Pathology Department, 42270 Saint-Priest en Jarez, France; 67Hpital Lariboisire, Pathology Department, 75475 Paris, France; 68Hpital de la cavale blanche, CHU Brest, Pathology Department, 29609 Brest, France; 69CHU de Lille, Neurosurgery Department, 59037 Lille, France; 70HIA du Val de Grce, Neurology Department, 75230 Paris, France; 71Clinique des Cdres, Pathology Department, 31023 Cornebarrieu, France; 72CHU de Montpellier, Pathology Department, 34295 Montpellier, France; 73CHU Angers, Pathology Department, 49933 Angers, France; 74CHU Saint-Pierre de la Runion, Neurology Department, 97410 Saint-Pierre de la Runion, France; 75Hpital Nord, CHU Amiens, Pathology Department, 80054 Amiens, France; 76Hpital Beaujon, Pathology Department, 92110 Clichy, France; 77CHU Nice, Pathology Department, 06002 Nice, France; 78Centre Eugne Marquis, Medical Oncology, 35042 Rennes, France; 79Hpital Jean Minjoz, CHU Besanon, Pathology Department, 25030 Besanon, France; 80Hpital Foch, Pathology Department, 92151 Suresnes, France.

NATURE COMMUNICATIONS | 7:11263 | DOI: 10.1038/ncomms11263 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 11