Introduction
Thanks to significant advances in genomics, the 2016 then-2021 World Health Organization (WHO) Classification of Tumours of the Central Nervous System (CNS) has defined tumour entities based on histological but also molecular features, like the driver genetic event [
15,
16].
Diffuse midline gliomas (DMG) H3 K27-altered have now been identified as a new type of malignant gliomas which occur in the paediatric and adult populations, although with disparities according to the preferential location,
i.e. brainstem in children and thalamus in adults [
18,
45].
DMG H3 K27-altered are either characterised by the substitution in histone H3 of the lysine at position 27 by a methionine (H3K27M), or the overexpression of
EZHIP [
3,
42]. Both mechanisms lead to Polycomb Repressor Complex 2 (PRC2) inhibition with a global loss of H3K27me3 [
5,
13], and consequently a major epigenetic and transcriptomic remodelling [
1,
3]. According to different molecular and clinical parameters, including specific DNA methylation profiles, DMG H3 K27-altered appeared more heterogeneous than initially thought, and were further classified in four subtypes:
H3-3A K27-mutant (H3.3-K27M),
H3C2 K27-mutant (H3.1-K27M), H3-wild-type (with
EZHIP overexpression) or
EGFR-altered [
3,
4,
16,
19,
33]. Despite this subdivision, we still observed some clinical and molecular heterogeneity within the most common H3.3-K27M subtype. We previously demonstrated that
TP53 co-driver mutations are associated with a worse tumour response to radiotherapy and a poorer outcome [
41] suggesting that additional molecular alterations can deeply modify the phenotype induced by the driver histone H3 mutation. Recently, we and others have described single cases or small studies of tumours with concomitant alterations of H3-K27M and mitogen-activated protein kinase (MAPK) pathway showing a possible longer survival compared to patients with DMG H3 K27-altered,
BRAF and
FGFR1 wild-type. In these studies, diagnoses ranged from H3.3-K27M
BRAFV600E/
FGFR1MUT pilocytic astrocytoma or midline gangliogliomas grade 1–3 [
11,
21‐
23,
26,
29,
44] to diffuse gliomas grade 4 [
25,
31]. As
BRAF and
FGFR1 mutations are typical hallmarks of low-grade gliomas/glioneuronal tumours such as ganglioglioma or pilocytic astrocytoma, the co-occurrence of H3-K27 and these MAPK alterations makes diagnosis and grading difficult [
30]. In order to understand how these alterations could mitigate the prognosis of these neoplasms, we analysed (radiologic, histologic, genomic, transcriptomic and DNA methylation analyses) a larger DMG H3 K27-altered cohort, comprising 29 tumours harbouring
BRAFV600E or
FGFR1MUT complemented by paediatric and adult cases from the literature.
Patients, materials and methods
Patients and tumour samples
The first part of the cohort is composed of 60 patients diagnosed with H3-K27M or EZHIP-overexpressing tumours harbouring a coding mutation in
BRAF or
FGFR1 genes, from the Necker Enfants-Malades/GHU-Sainte Anne Hospital/Gustave Roussy center and Biomede 1 Trial (NCT02233049) (
n = 29), or from other published cohorts (
n = 31) [
3,
18,
23,
25,
30,
31,
37]
. The control cohort includes patients with wild-type
FGFR1/BRAF, in the above-mentioned cohorts of H3K27M or
EZHIP-overexpressing DMG (flowchart in Supplementary Fig. 1, online resource). Tumour tissue and clinical data were collected under informed consent obtained from the parents or guardian according to the IRB approved protocol (CNIL 1176643).
Radiology analysis
All available radiology outcomes at diagnosis (CT and MRI) of patients with H3.3-K27M tumours from our cohort were reviewed centrally by three experts (VDR, NB, and JG). Parameters specifically recorded were: radiological presentation (diffuse, circumscribed or nodular and diffuse), presence of contrast enhancement (yes/no) and calcifications (yes/no). For circumscribed tumours, the pattern at evolution was also analysed.
Histopathological analyses and immunohistochemistry
Formalin-fixed paraffin-embedded (FFPE) tissue samples for each patient were retrieved and Haematoxylin–Phloxine–Saffron (HPS)-stained slides were analysed by two experienced neuropathologists (PV and ATE) to confirm morphological diagnoses. Micro-calcifications were noted as present or absent as well as granular bodies, ganglion neurons, necrosis, and microvascular proliferation. Mitotic activity (per 2 mm2) and tumour growth architecture were analysed within the inherent limits of a stereotaxic biopsy exploration. The latter was labelled as diffuse, compact tumoral areas or both. Morphological aspects were evaluated as ganglioglioma-like (GG-like), HGG with piloid astrocytic component or DMG-like. The infiltration pattern was also assessed by NF70 immunostaining (i.e. residual NF70 network or not). Immunostaining was performed from 3 μm-thick representative FFPE sections using a Dako OMNIS automate. The following primary antibodies were used: H3K27me3 (1:2500, polyclonal, Diagenode), H3-K27M (1:5000, clone EPR18340, Abcam), Neurofilament Protein NF70 (1:100, clone 2F11), CD34 (1:40, clone QBEnd-10, Dako), BRAFV600E (1:100, clone VE1, Abcam), ATRX (1:200, polyclonal, Diagomics), Ki-67 (1:200, clone MIB-1, Dako). Antigen retrieval was performed at 95 °C, pH9 (GV80011-2, Dako) or pH6 (GV805, Dako). External positive and negative controls were used for antibody validation.
DNA/RNA extraction and sequencing
DNA and RNA were extracted from frozen tumours using Allprep DNA/RNA kit (Qiagen) and were quantified using, respectively, the Qubit Broad Range double-stranded DNA assay (Life Technologies) or the Qubit RNA high sensibility (Life Technologies). When no frozen material was available for DNA methylation profiling, tumour DNA was extracted from formol-fixed paraffin-embedded (FFPE) block sections using dedicated protocols at Diagenode or Integragen. Targeted DNA sequencing (≥ 6000× coverage) or whole exome sequencing (≥ 130× coverage) was performed as previously described [
10,
41]. WES was aligned with BWA according to GATK best practice guidelines, then Mutect2 was used for the DNA calling. RNAseq on DMG primary tumours was performed at Integragen (Evry, France). PolyA mRNA molecules were initially purified from at least 100 ng total RNA (NEBNext® Poly(A) mRNA Magnetic Isolation Module, NEB) and libraries were then prepared using the NEBNext Ultra II Directional RNA Library Prep Kit (NEB). Paired-end reads of 100 bp were generated on an Illumina NovaSeq reaching an average sequencing depth of 60 million reads.
DNA Methylation array processing
Genome-wide DNA methylation analysis was performed using either the Illumina HumanMethylation450 BeadChip (450 k) or EPIC arrays as previously published [
2,
3]. Data were obtained from different platforms (DKFZ Heidelberg; Integragen; Diagenode; published data) and were analysed with R (v4.0.4). For t-Distributed Stochastic Neighbour Embedding (t-SNE) analysis, the
minfi package was used to load idat files and preprocessed with the function
preprocess.illumina for dye bias and background correction. Probes located on sex chromosomes or not uniquely mapped to the human reference genome were removed. Probes containing single-nucleotide polymorphisms or that were not present in both EPIC and 450 k methylation array were also eliminated. A batch effect correction was done with
removebatchEffect function from
limma package, to remove difference between formalin-fixed paraffin-embedded and frozen samples. The probes were sorted by standard deviation. The 10,000 most variable probes were used for subsequent clustering analysis and to compute the 1-variance weighted Pearson correlation between samples. The distance matrix was used as input in t-SNE from
Rtsne package. For the second analysis, DNA methylation-based classification of CNS tumours from DKFZ-Heidelberg was used in order to predict the CNS tumour class based on the V12.7 of the classifier (
www.molecularneuropathology.org).
Gene expression analysis
Reads were pre-processed using the nf-core RNAseq pipeline (v3.0), mapped to the reference genome GRC38/hg38 with the STAR tool (v2.6.1d), annotated with GENCODE v36 and counting was performed with the Salmon quantification tool (v1.4.0). Differential gene expression analysis was performed with the DESeq2 package (v1.30.0, minReplicatesForReplace = 7, betaPrior = TRUE) with a threshold of 0.01 for Benjamini–Hochberg adjusted p value (adj-p). For gene set enrichment analysis (GSEA), hypergeometric tests were used to identify overrepresented gene sets from the MSigDB v7.4 database, amongst genes ranked by significance and fold-change in differential expression analysis, with Benjamini–Hochberg multiple testing correction using the package Clusterprofiler. Catalogues considered included Hallmark and C2. Differences were considered as significant when false discovery rate adj-q value was < 0.02. Gene expression comparison was evaluated with Wald test using DESeq2.
Univariate and multivariate survival analyses
Overall survival (OS) was estimated with the Kaplan–Meier method and median overall survival was computed using a log-rank test. OS was obtained from the post-diagnosis until death patient or last known information. The analysis was realised in Prism9 software. Multivariable Cox proportional hazards regression model on OS was performed including histone H3, BRAF, FGFR1, TP53 status, age at diagnosis and tumour location with R software using the function coxph() of the survival package (Version 3.2–13).
Statistical analyses
Distribution of age at diagnosis according to different parameters was accessed by Mann–Whitney test. Presence of macro-calcification, contrast enhancement, radiologic profile and sex ratio were evaluated by Fisher’s exact test. Chi-square test for trend was used to evaluate tumour type and location. All statistic tests were performed using Prism 9 software (GraphPad).
Discussion
The recent description of ‘
DMG, H3 K27-altered’ and its sub-classification into four molecular subgroups does not capture completely the diversity of this disease [
16]. Our data support the individualization of an additional new subtype of DMG with distinct histological, radiological, clinical, genomic, transcriptomic and epigenetic features that we provisionally termed
DMG, H3 K27 and BRAF/FGFR1 co-altered (DMG_K27-BRAF/FGFR1) which may represent 20% of
DMG H3 K27-altered. Using unsupervised analysis of DNA methylation tumour profiles, DMG_K27-BRAF/FGFR1 formed a specific cluster, separated from other DMG_K27 gliomas, others adult/paediatric diffuse gliomas, low-grade glial/glioneuronal tumours and more specifically
BRAFV600E-mutated ganglioglioma even midline located.
This highlights a possible distinct cell of origin for DMG_K27- BRAF/FGFR1, able to exhibit a mixed glial and neuronal differentiation, mostly noticeable in the BRAF subclass [
2,
39]. Schüller et al
. did not mention any of these phenotypes in DMG H3-K27M
FGFR1MUT, due to the limited number of tumours [
31]. The analysis of the H3.3-K27M ancestral clone derived from a DMG
BRAFV600E H3.3K27M harboured this same DNA methylation profile, confirming that the specific DNA methylation profile of DMG_K27-BRAF/FGFR1 is not a strict consequence of MAPK alterations. The fair discrimination of tumours on the tSNE based on the type of the secondary MAPK mutation (
i.e. BRAF vs. FGFR1) even suggests that this entity could be further subdivided.
Genotype–morphotype correlations support the distinction from classical DMG, H3 K27-altered and from glial/glioneuronal tumours MAPK-altered: (i) whilst ependymal differentiation has been described in rare DMG H3-K27 [
36], a mixed glioneuronal differentiation associated with CD34 positivity and eosinophilic granular bodies or a piloid differentiation are not yet described; (ii) only an exceptional subset of ganglioglioma grade 1 present
FGFR1 alteration, more characteristic of other glioneuronal tumours [
24] and (iii) the existence of true malignant transformation in ganglioglioma is a matter of debate. The majority of reported cases were published before the advent of molecular biology, reclassified in a wide spectrum of CNS WHO tumour types without a distinct methylation class [
28] or more interestingly were midline-located with a co-occurring
BRAF and H3-K27M mutations [
11,
21‐
23,
26,
29,
30,
44]. In the unified methylation class that we describe, the radiological and histopathological presentations are thus highly heterogeneous including tumours with mixed glioneuronal or pilocytic differentiation, and do not always fulfil a strict diagnostic criterion of DMG. Indeed, these tumours are less diffuse, with a frequent nodular to circumscribed radiological aspect (91% for
BRAFMUT and 78% for
FGFR1MUT DMG) and calcifications.
Several other clinical and biological characteristics support the individualization of this new entity from classical DMG, H3 K27-altered. First, OS is significantly different from classical DMG_K27, with a median around three years for both FGFR1MUT and BRAFMUT H3.3K27M DMG.
Moreover, our multivariate analysis demonstrates for the first time that the presence of these mutations is an independent prognostic factor for improved OS in DMG_K27. Previously, Picca et al
. and Schüller et al. showed in a small populations (
n = 6 or
n = 7), by univariate analysis and without taking into account
TP53 status, that DMG patients with
FGFR1MUT have a better survival [
25,
31]. The identification of a new subtype of
DMG H3 K27-altered with longer survival is also a step forward for clinical research, highlighting the need for patient’s stratification in trials or at least molecular documentation of the cases.
Patients from Necker/Gustave Roussy cohort with BRAF/FGFR1-mutated DMG_K27 received, over a large period of time, quite heterogeneous treatment which did not allow specific statistical conclusion. It thus remains to be defined whether these patients could respond to a targeted therapy against BRAFV600E or FGFR1.
Another meaningful difference is that DMG_K27-BRAF/FGFR1 are more frequent in the thalamus than the brainstem compared to DMG from the H3.3-K27M subtype. The age at diagnosis also differs according to the presence of the MAPK alteration. The age at onset for FGFR1MUT DMG is significantly higher (median 14.8 years) and to our knowledge no adults were affected by a H3.3-K27M BRAF-mutated gliomas. In this new subtype, some heterogeneity remains present at various level between DMG H3K27M BRAF or FGFR1-mutated. This finding cannot be presently explained, but it may also point towards different oncogenesis, which could be individualised in the future studies.
We also investigated gene expression in these tumours and observed a senescence signature including an up-regulation of
CDKN1A (P21) specific to DMG_K27-MAPK which is usually more present in paediatric LGGs. LGGs are characterised by an over-activated MAPK signalling in consequence to oncogenic alteration of
BRAF or
FGFR1 [
15] and the main hypothesis for a slow tumour evolution in LGG is based on the induction of oncogene-induced senescence which gives a growth advantage in a restricted window during brain development [
7,
8,
27,
35,
40]. Senescence triggered via the P53/P21 axis could in part explain the slower tumour evolution in DMG_K27-MAPK. Of note, the DMG_K27-BRAF/FGFR1 share other characteristics with LGGs like calcifications and preferential association of mutations:
FGFR1 with
NF1,
PI3KCA,
PTPN11 [
9,
17,
32,
38]. We also demonstrated in one case from this new DMG subtype harbouring
BRAFV600E, that H3.3K27M was the first mutational event in its oncogenesis. Thus,
BRAF and
FGFR1 mutations would be secondary driver events in the oncogenesis of these tumours and could give a proliferative advantage to the H3.3-K27M ancestral clone in a specific developmental window similarly to paediatric LGGs. Extending the analysis of sequential acquisition of mutations in DMG_K27-MAPK oncogenesis will be essential for designing future therapeutic interventions.
In conclusion, we have identified a fifth subtype of DMG, H3 K27-altered that we named ‘DMG H3 K27 and BRAF/FGFR1 co-altered’ (DMG_K27- BRAF/FGFR1), which harbours specific clinical and biological characteristics. We hypothesise that the better OS of DMG_K27-BRAF/FGFR1 compared to other DMG_K27 could be the result of both a specific cell origin and the oncogene-induced senescence. Individualization of this subtype is of importance for the interpretation of trials and affected patients may deserve specific treatment strategies.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.