Phenotypic and epigenetic heterogeneity in FGFR2‐fused glial and glioneuronal tumours

Abstract

Aims

FGFR‐fused central nervous system (CNS) tumours are rare and are usually within the glioneuronal and neuronal tumours or the paediatric‐type diffuse low‐grade glioma spectrum. Among this spectrum, FGFR2 fusion has been documented in tumours classified by DNA‐methylation profiling as polymorphous low‐grade neuroepithelial tumours of the young (PLNTY), a recently described tumour type. However, FGFR2 fusions have also been reported in glioneuronal tumours, highlighting the overlapping diagnostic criteria and challenges.

Methods

We investigated the FGFR2 fusion landscape in a French national series of tumours sent to the RENOCLIP‐LOC network. We comprehensively analysed histology, radiology and molecular data including DNA‐methylation profiling for 16 FGFR2‐fused glioneuronal tumours.

Results

Most tumours were located in the temporal or parietal lobe with a median age at diagnosis of 7 years [1–44]. Epilepsy was the most frequent symptom. Five patients had tumour progression or recurrence with a median progression‐free survival of 22.6 months. Histological phenotypes corresponding to PLNTY, GG, MVNT or unclassified tumours were recorded. Epigenetic profiling could not properly distinguish epigenetic clusters related to the GG and PLNTY methylation classes among FGFR2‐fused glioneuronal tumours. However, a neuroradiological review identified strikingly distinct neuroradiological patterns.

Conclusion

While delineating tumour types among the FGFR2‐fused glioneuronal tumour spectrum, by histopathology or DNA‐methylation profiling, remains challenging, neuroimaging data revealed two distinct patterns that could correlate to PLNTY and ganglioglioma. However, more series including extensive histo‐radio‐molecular data are needed to confirm this hypothesis.

FGFR2‐fused brain tumours might represent rather a spectrum of paediatric diffuse low‐grade glioma, glioneuronal and neuronal tumours overlapping PLNTY and ganglioglioma than a unique tumour type.Although the differences are very slight in terms of histopathology or epigenetics, neuroradiology identifies two distinct radiological patterns suggesting the existence of two tumour types within the spectrum of FGFR2‐fused central nervous system tumours.

INTRODUCTION

FGFR‐fused central nervous system (CNS) tumours are rare; they account for 6.1% (n = 29/1000) of paediatric low‐grade gliomas (LGG) according to Ryall et al. [1]. They are mainly encountered in the paediatric population, adolescents or young adults. On histology, they are characterised by an oligodendroglioma‐like phenotype [2, 3]. Although a broad panel of pathological diagnoses (from benign tumours to highly malignant, such as glioblastoma, IDH‐wild type) are associated with FGFR3‐fused CNS tumours [1, 3, 4, 5], CNS tumours presenting with FGFR2 fusions are usually benign [4, 5, 6, 7]. FGFR2 fusions are more often encountered in glioneuronal and neuronal tumours but are also seen in tumours of the paediatric‐type diffuse LGG spectrum [2]. Among this spectrum, polymorphous low‐grade glioneuronal tumour of the young (PLNTY) was introduced in the 2021 WHO classification of CNS tumours as a new tumour type, characterised at histology by “a diffuse growth pattern, an oligodendroglioma‐like component, few (if any) mitotic figures, extravascular CD34 expression and MAP kinase alteration including BRAF V600E mutation and FGFR2/3 fusions”. So far, only BRAF V600E mutation and FGFR2 fusion have been documented in tumours classified by DNA‐methylation profiling as PLNTY [8, 9]. In addition, some authors suggested that FGFR2 fusion was specific to the epigenetic class of PLNTY among low‐grade neuroepithelial tumours (LGNET) [9]. However, FGFR2 fusions have also been reported in ganglioglioma (GG), dysembryoplastic neuroepithelial tumour (DNET) and tumours described as multinodular and vacuolating neuronal tumour (MVNT) [10, 11, 12], these three tumour types having many overlapping diagnostic criteria representing a diagnostic challenge. We, therefore, performed a retrospective analysis of a national series of FGFR2‐fused tumours to better characterise this spectrum [13].

MATERIAL AND METHODS

Tumour samples and clinical data

The inclusion criteria were supratentorial hemispheric IDH‐non‐mutated, H3‐non altered, glial or glioneuronal tumour, associated with an FGFR2 fusion transcript, and with one available formalin‐fixed paraffin‐embedded (FFPE) block and the corresponding stained slides. A total of 16 cases from five French University Hospitals fulfilled these criteria. The FFPE blocks and slides concerned the first surgical excision in 16 cases, the initial excision (#08) and the recurrence (#08bis) in one. These cases were retrieved from university hospital centres: GHU Paris Psychiatry and Neurosciences, Lille, Toulouse, Bordeaux and Angers. Six cases were previously reported in Pagès et al. (#02, #05, #08, 09) and Métais et al. (#03, #14) [4, 7]. Written informed consent to be included in this study was provided by the participants or their legal guardian/next of kin. This study was reviewed and approved by the Aix‐Marseille University ethics committee (2019‐25‐04‐003). Clinical data were retrospectively collected for each case and included sex, history of epilepsy, tumour location, age at diagnosis, extent of surgical resection and follow‐up (including date at last follow‐up and date of progression or recurrence).

Central radiological review

The French Neuroimaging‐RENOCLIP‐LOC consortium, composed of expert neuroradiologists in the field of neuro‐oncology, conducted a comprehensive review of the imaging cases, blinded from the pathological diagnosis. This review, organised by a neuroradiologist (ME), involved the participation of eight neuroradiologists (NB, VDR, QVM, MO, MR, JPC, CP, ME). The following features were assessed: location (lobe, deep location near the ventricles, or peripheral location close to or involving the cortex) and primary morphological characteristics: a well‐circumscribed nodular lesion or an infiltrative appearance, existence of a cystic component and signal characteristics on T1‐weighted and T2‐weighted images relative to normal‐appearing cortex. Additional analyses included susceptibility‐weighted imaging (SWI) and post‐gadolinium T1‐weighted sequences (presence and homogeneity of enhancement), presence or absence of microcysts with contrast enhancement, presence or absence of high signal FLAIR and presence or absence of perilesional oedema. When available, calcification was assessed on CT scans. The preoperative imaging follow‐up, when available, allowed assessment of the progression of the lesion between diagnosis and surgery, while postoperative imaging follow‐up was collected to assess the completeness of surgical resection and to evaluate its evolution.

Histopathological analysis

Samples were stained with haematoxylin‐phloxine‐saffron (HPS) according to standard protocols. Immunohistochemistry was performed when the following immunostains were not available from the initial centre: OLIG2 (Sigma® Polyclonal or Dako® 6F2), CD34 (QBEnd10, Dako®), GFAP (6F2, Dako®), Neurofilament (2F11 Dako®, or 2F11 Menarini®), synaptophysin (DAK‐SYNAP, Dako®), Chromogranin A (LK2 H10, Diagnostic Biosystem®), Ki67 (MIB1, Dako®), Microtubule Associated Protein (MAP‐2) (HM‐2; Sigma®) and BRAF V600E (VE1; Spring Biosciences®). FFPE sections were deparaffinised and processed with an OMNIS or Ventana autostainer (Dako®, Roche®) according to standard protocols. All tissue samples were centrally reviewed by expert neuropathologists (DFB, PV, ATE, AM). For each case, the following pathological features scored as present or absent: calcification, the specific glioneuronal element, eosinophilic granular bodies, perivascular lymphocytic infiltrates, glial component subtype (oligodendroglioma‐like, astrocytic, piloid), Rosenthal fibres, haemorrhage, microvascular proliferation and tumour necrosis. Mitotic activity was recorded on 10 high power fields, that is, 2 mm² and expressed in mitoses per mm². In addition, the growth pattern was assessed as diffuse (at least regionally) or circumscribed on anti‐neurofilament immunostaining. The occurrence of a neoplastic neuronal component was also analysed. We distinguished “neoplastic ganglion cells”, characterised by enlarged neurons often binucleated expressing chromogranin A but not Olig2, from “immature neuronal cells” characterised by synaptophysin and Olig2 co‐expression but lack of chromogranin A immunoreactivity as suggested by the WHO classification [6]. MAP 2 immunostaining (performed on five cases with a neoplastic neuronal component) did not help distinguish these two neoplastic neuronal components in our hands (Figure S1). CD34 immunostaining pattern was recorded as strong and diffuse or patchy and ramified.

FISH analysis

FISH analysis, to assess FGFR2 rearrangements, was performed on nine cases at the time of diagnosis (#01, #02, #04, #05, #07–09, #13, #16). For three cases (#01, #04, #05) FISH analysis was repeated to analyse specifically the tumoural neuronal component. Experiments were performed on interphase nuclei on paraffin‐embedded sections (4 μm), following standard procedures and using break‐apart probes targeting FGFR2 (Zytovision, Clinisciences). A case was considered FGFR2‐fused when the scored nuclei displayed a break‐apart signal in at least 20% of the counted nuclei. Hybridisations were considered non‐informative if the FISH signals were either lacking or too weak to be interpreted. The results were recorded using a DM6000 imaging fluorescence microscope (Leica Biosystems, Nanterre, France) fitted with appropriate filters, a CCD camera and digital imaging software (CytoVision, v7.4).

DNA sequencing and RNA sequencing data

Molecular data from DNA sequencing or RNA sequencing were retrieved from pathological reports resulting in heterogenous panels targeting MAP kinase pathway genes (such as BRAF, FGFR1, RAF, PTPN11) and mTOR pathway genes (PIK3CA, AKT, PTEN). Most experiments were performed according to previously described methods on different molecular biology platforms [4, 14, 15]. FGFR2 fusion transcripts were detected in 13 cases on FFPE material by next‐generation sequencing using the FusionPlex® Lung kit by ArcherDX (ArcherDX Inc. Boulder, CO, USA) adapted to sequencing on MiSeq (Illumina Inc., San Diego, CA, USA) (n = 10) or by the RNA sequencing method described in Métais et al. [4] (n = 3); and in three cases on fresh frozen material using the mRNA stranded LT Library Prep (RS‐122‐2101, Illumina®) on NextSeq500 (Illumina®). Six cases had an FGFR2::INA fusion transcript, and five cases had an FGFR2::SHTN1 fusion. Other fusion transcript partners were TACC2 (n = 1), OPTN (n = 1), CTNNA3 (n = 1), HDX (n = 1) and FLT3 (n = 1). Information on targeted NGS was available for 7/16 patients. Only one had a concomitant PTPN11 mutation p.(Q510R) mutation with a variant allele frequency of 45%, which is shown in Table S1. Information on BRAF mutation searched by ddPCR was available in two cases (all negative). In addition, BRAF V600E staining was available in four cases and was always negative. In 10 cases, TERT promoter status was available. Detailed molecular information is included in Table S1.

DNA‐methylation profiling

DNA‐methylation profiling was performed in 14 cases including one on initial (#08) and recurrent (#08bis) tumour. DNA‐methylation profiling could not be performed in two cases (#03 and #07) due to low DNA quantity or quality. At least 250 ng of DNA were extracted from each tissue sample (FFPE). DNA bisulfite conversion was undertaken using the ZymoEZ DNA‐methylation kit (Zymo Research, USA), then treated with FFPE DNA Restore kit (Illumina, San Diego, CA, USA) and DNA clean and concentrator‐5 (Zymo Research, USA). Standard quality controls confirmed DNA quantity/quality and bisulfite conversion. All patient samples were analysed using either Illumina Infinium Methylation EPIC or HumanMethylation450 BeadChip arrays according to the manufacturer's instructions. The iScan control software was used to generate raw data files from the BeadChip in .idat format, analysed using GenomeStudio version 2.0 (Illumina, San Diego, CA, USA) and were checked for quality measures according to the manufacturer's instructions. The .idat files were uploaded to the online CNS tumour DNA‐methylation classifier (versions 12.5 and 12.8) from the German Cancer Research Centre (Deutsches Krebsforschungszentrum, DKFZ) at https://www.molecularneuropathology.org, and a report for every tumour was generated, providing prediction scores for methylation classes (MC) and chromosomal copy‐number‐variation (CNV) plots. Additional analyses were performed in R studio (v4.0.2). Raw signal intensities were obtained from .idat files using the minfi Bioconductor package (v1.34.0).

Normalisation was conducted using the IlluminaPreprocess function. Background correction and dye‐bias correction were performed on each sample. Filtering criteria of probes were the removal of probes targeting X or Y chromosomes and the removal of probes containing single nucleotide polymorphisms. Hierarchical clustering was performed using the Complex Heat map package (2.4.3). Clustering of beta values from methylation arrays was performed based on Euclidean distance with a ward algorithm. Methylation heat maps show only the most variable probes (SD > 0.20). t‐Distributed stochastic neighbour embedding (t‐SNE) dimensional reduction analysis was performed using the Rt‐SNE package, with the following parameters: exaggeration factor = 42, normalise = TRUE, pca_scale = TRUE, pca_center = TRUE, eta = 500, theta = 0.01, max_iter = 5000, pca = TRUE. Visualisation was done using ggplot2version 3.3.6 and Plotly version 3.10.0. In order to build the t‐SNE, we used, as a reference series, cases retrieved from our DNA‐methylation profiling database that were classified by the v12.5 version of the Heidelberg classifier with a prediction score >0.9 as follows: DNET MC (20 cases), GG MC (18 cases), diffuse leptomeningeal glioneuronal tumour (DLGNT) MC (17 cases), high‐grade astrocytoma with piloid features (HGAP) MC (15 cases), low‐grade glioma hemispheric pilocytic astrocytoma (LGG_PA_HEMI) MC (12 cases), low‐grade glioma midline pilocytic astrocytoma (LGG_PA_MID) (20 cases), low‐grade glioma posterior fossa pilocytic astrocytoma (LGG_PA_PF) MC (22 cases), low‐grade glioma spinal pilocytic astrocytoma (LGG_PA_SPINE) MC (eight cases), low‐grade glioma rosette‐forming glioneuronal tumour (LGG_RGNT) MC (16 cases), pleomorphic xanthoastrocytoma (PXA) (21 cases) and MC, PLNTY MC (two cases); and from previously published studies [9, 16, 17] such as control tissue cerebral hemisphere (CONTR_HEMI) MC (12 cases), control tissue tumour microenvironment (CONTR_REACT) MC (23 cases), control tissue cerebral white matter (CONTR_WM) MC (nine cases), extraventricular neurocytoma (EVN) MC (10 cases) [17, 18] and low‐grade neuroepithelial tumour FGFR2‐fused (LGNET‐FGFR2) MC (eight cases) [9].

Density‐based spatial clustering of applications with noise (DBSCAN)

DBSCAN was performed on t‐SNE for the GG and FGFR2 cases using the dbscan R package (version 1.1‐12). This algorithm identifies high‐density regions within the data, where density is determined by the number of neighbouring data points within a specified distance (epsilon parameter) and the minimum number of points required to form a cluster (minSamples parameter). To determine the optimal eps value, we computed the k‐nearest neighbour distances and calculated the average of the distances of every point to its k‐nearest neighbours. These k‐distances were plotted in ascending order, and we determined the ‘knee’ of the plot, which corresponds to the optimal eps parameter (Figure 5C).

FIGURE 5.

t‐Distributed stochastic neighbour embedding DNA‐methylation profiling data analysis of FGFR2‐fused glial/glioneuronal tumours and density‐based spatial clustering of applications with noise (DBSCAN) clustering of t‐SNE results for GG and FGFR2 cases. (A) Fifteen samples of the present series (FGFR2_fused_cases) and eight samples (LGNET‐FGFR2) from Gupta et al. [9] were compared to a reference data set built up with institutional cases matching with the DKFZ classifier methylation classes listed below, and from previously published studies [17, 18, 19]. Dysembryoplastic neuroepithelial tumour (DNET); ganglioglioma (GG); diffuse leptomeningeal glioneuronal tumour (DLGNT); high‐grade astrocytoma with piloid features (HGAP); low‐grade glioma hemispheric pilocytic astrocytoma (LGG_PA_HEMI); low‐grade glioma, midline pilocytic astrocytoma (LGG_PA_MID); low‐grade glioma, posterior fossa pilocytic astrocytoma (LGG_PA_PF); low‐grade glioma, spinal pilocytic astrocytoma (LGG_PA_SPINE); low‐grade glioma, rosette‐forming glioneuronal tumour (LGG_RGNT); pleomorphic xanthoastrocytoma (PXA); polymorphous low‐grade neuroepithelial tumour of the young (PLNTY); control tissue, cerebral hemisphere (CONTR_HEMI); control tissue, tumour microenvironment (CONTR_REACT); control tissue, cerebral white matter (CONTR_WM); extraventricular neurocytoma (EVN). (B) selection of the t‐SNE data on which DBSCAN was performed. (C) DBSCAN clustering of t‐SNE results for GG and FGFR2 cases. The plot displays the DBSCAN clustering results applied to the t‐SNE transformed data of GG and FGFR2 cases, using parameters ε = 175 and MinPts = 5.

Chromosomal CNV analysis

Chromosomal CNVs were searched for by visual inspection of CNV profiles generated by the molecularneuropathology.org platform as described [16, 20]. Visual inspection indicated a gain of chromosome seven if a complete gain was present; a deletion of chromosome 10 if a complete deletion was present; an EGFR amplification if a high‐level amplification of EGFR locus was present, CDKN2A homozygous deletion if a log2 value of −0.4 or lower was observed, as suggested in Maragkou et al. and Capper et al. [16, 21].

Statistical analysis

Categorical variables are presented as frequencies and percentages, continuous variables as medians and ranges or means and standard deviations. Fisher's exact test was used to compare categorical variables. Statistical tests were two‐sided, and the threshold for statistical significance was p = 0.05. Statistical analysis was done using XLSTAT software Version 2020.1.01 (Addinsoft, Paris, France).

RESULTS

National series data

The median age at diagnosis was 7 years [1–44]. Eight males and eight females were included. Twelve cases had a clinical history of epilepsy, and one case presented with high intracranial pressure (#16). Clinical information was not available in three patients (Table S1). Twelve cases had undergone gross total resection, and four had a biopsy or partial resection. Follow‐up data were available for 15 cases, with a median follow‐up of 35.6 months [6.1–214.3]. Five patients experienced recurrence or progression with a median time of 22.6 months [4.9–146.4]. No patient died from their disease. Most tumours were located in the temporal (6/16) or parietal lobe (5/16) (Table 1).

TABLE 1.

Main clinico‐pathological features of the series.

Sex
M	N = 8
F	N = 8
Sex‐ratio M/F	1
Age at diagnosis
Mean	11 years +/−12
Median	7 years [1–44]
Tumour location
Frontal	N = 1
Temporal	N = 6
Parietal	N = 5
Occipital	N = 3
Insular	N = 1
Surgical excision
Total	N = 12
Partial	N = 4
Radiological pattern
Type A (PLNTY)	N = 7 + 1 recurrence
Type B (GG and MVNT)	N = 8
Other	N = 1
Median follow‐up	35.6 months [6.1–214.3]
Lost to follow‐up	N = 1
Progression	N = 5
Median time between first surgery and progression	22.6 months [4.9–146.4]
FGFR2 fusion transcripts
FRFG2::INA	N = 6
FGFR2::SHTN1	N = 5 + 1 recurrence
Other	N = 5
Histopathologic review
Oligodendroglial component	N = 16
Microcalcification	N = 13
Diffuse growth pattern	N = 13
Neoplastic ganglion cells	N = 3
Immature neuronal cells	N = 4
Specific glioneuronal element	N = 0
CD34 extravascular staining	N = 15

Abbreviations: F, female; GG, ganglioglioma; M, male; MVNT, multinodular and vacuolating neuronal tumour; PLNTY: polymorphous low‐grade neuroepithelial tumour of the young.

Histopathologic review

All cases had an oligodendroglioma‐like component. Five had an additional astrocytic, and two had a piloid component, although Rosenthal fibres were not seen. No case had the specific glioneuronal element. Eosinophilic granular bodies and perivascular lymphocytic infiltrates were present in 5/16 and 6/16 cases, respectively. Calcification was observed in 13/16 cases (Figure 1G). Only one case (#16) had haemorrhagic changes accompanied by siderophagic infiltrates, microvascular proliferation and necrosis. Mitotic activity was observed in four cases ranging from 0.6 to 1.7/mm². Thirteen cases had a diffuse (at least regionally) growth pattern, whereas three were circumscribed. Neoplastic ganglion cells were observed in three cases (Figure 1A–B). An immature neuronal component was observed in four cases (Figure 1D–F). Strong and diffuse extravascular CD34 expression was observed in nine cases (Figure 1I), and six cases had a patchy and ramified pattern of CD34 expression (Figure 1C). One case was CD34 negative. OLIG2 was expressed in all cases.

FIGURE 1.

Histopathological features of FGFR2‐fused glial and glioneuronal tumours. Histopathological features of ganglioglioma (A–C) with neoplastic ganglion cells indicated by a black arrowhead (A) (scale bar 100 μm), expression of chromogranin A (B) (scale bar 100 μm) and patchy ramified extravascular CD34 expression (C) (scale bar 500 μm). Histopathological features of multinodular vacuolating neuronal tumour (D–F) (scale bar 200 μm) displaying an immature eosinophilic neuronal component indicated by a black arrowhead (D), chromogranin A negativity (E) with synaptophysin and OLIG2 co‐expression (F). Histopathological features of polymorphous low‐grade neuroepithelial tumour of the young (G–I) (scale bar 100 μm): an oligodendroglioma‐like neoplasm with microcalcification (G), OLIG2 expression (H) and strong and diffuse CD34 extravascular expression (I). GG, ganglioglioma; MVNT, multinodular and vacuolating neuronal tumour; PLNTY, polymorphous low‐grade neuroepithelial tumour of the young.

Cases were classified histopathologically according to the WHO diagnostic criteria [6]. Three cases with neoplastic ganglion cells (#01, #02 and #06) were classified histopathologically as GG (Figure 1A–C). Four cases with an immature neuronal component were classified histopathologically as MVNT (#03, #04, #05 and #15) (Figure 1D,E,F). Five cases (#07‐11), without a neuronal component, fulfilled the essential diagnostic criteria for PLNTY such as an oligodendroglioma‐like tumour cells component with strong and diffuse extravascular CD34 expression, a diffuse growth pattern (at least focally), without mitotic activity (Figure 1G,H,I). Four cases did not match diagnostic criteria for a specific tumour type according to the WHO 2021 and remained unclassified histopathologically. Three of these did not have a neuronal component and had a solid growth pattern (cases #12–14), whereas the other case (#16) was CD34 negative, without a neuronal component.

FGFR2 rearrangement is present in the neuronal component

FISH analyses were positive in 8/9 cases at the time of diagnosis (one was noncontributory), in full accordance with RNA sequencing results. The rearrangement was observed in at least 30% of the tumour cells. FISH performed on tumours with a neoplastic neuronal component (either ganglionic or immature, cases #01, #04, #05) confirmed the presence of the FGFR2 rearrangement in both the glial and the neoplastic neuronal component as the presence of non‐specific granular probe attachment, resembling vesicles, was highly suggestive of neuronal tumoural cells (Figure 2) [22].

FIGURE 2.

FISH analysis of FGFR2‐fused glial/glioneuronal tumour. FISH analysis performed on a histological multinodular and vacuolating neuronal tumour (A) (scale bar = 100 μm), shows the presence of an FGFR2 rearrangement (B) and (C) (X800, grey arrowhead, indicating a separated red signal), in cells displaying non‐specific granular probe attachment compatible with neuronal cells vesicles (white arrowhead, magnification ×800).

Radiological features separate into two patterns

Central imaging review identified seven circumscribed lesions (#07–10, #12–13, #16) and nine infiltrative (#01–06, #11, #14–15). In addition, other neuroradiological features were associated with each of these presentations. Circumscribed lesions were mostly well‐circumscribed intra‐axial mass with minimal or no perilesional oedema, presenting a heterogeneous signal on T1 and T2‐weighted images, moderate contrast enhancement and coarse calcifications, and were defined as type A. Whereas, infiltrative lesions were mostly characterised by an infiltrative area with high FLAIR signal intensity, potentially associated with a microcystic component (<1 cm) that may exhibit enhancement and lack of coarse calcification, and defined as type B. These neuroradiological patterns are illustrated in Figure 3.

FIGURE 3.

Radiological features of FGFR2‐fused glial/glioneuronal tumours. Illustration of type A and B neuroradiological patterns. Type A pattern (left panel) was characterised by heterogeneous well‐delineated masses with variable enhancement, and coarse calcification, involving the cortex and subcortical white matter with possible ventricular contact with no or minimal oedema type B pattern (right panel) was characterised by ill‐delineated high FLAIR signal intensity frequently associated with microcysts. No contrast enhancement is observed in these cases.

Tumours exhibiting the first pattern (type A) (#07–10, #12–13, #16), were located in the occipital area in three cases, the insula or parietal in two cases, the posterior temporal lobe in one case and the frontal lobe in one case. Additionally, five out of the seven cases had a deep location with ventricular contact (#07, #09, #12–13, #16).

Tumours exhibiting the second pattern (type B) (#01–06, #14–15) were located equally in the temporal or the parietal lobe, only in superficial cortical and subcortical locations. None were calcified. They all consisted of an area of high FLAIR signal, which was associated with a microcystic (<1 cm) content in 6/8 cases, without contrast enhancement (except for a faint enhancement of the microcystic component in case #14). The high FLAIR signal area had a triangular shape towards the ventricle that might evoke a focal cortical dysplasia in case #01.

Finally, case #11 remained unclassified, because it demonstrated features of both type A and B, with coarse calcifications and subcortical microcysts associated with a high FLAIR signal intensity area.

DNA‐methylation profiling distinguishes two epigenetic clusters of FGFR2‐fused tumours

Six samples were classified by the brain tumour classifier (v12.5, molecularneuropathology.org) in GG MC (n = 5; #01, #02, #04, #14, #15) and PLNTY MC (n = 1; #08bis) with a confident prediction score (≥0.9). Seven samples were classified with a prediction score <0.9 and >0.3 in GG MC (n = 4; #05, #09, #13, #16) and PLNTY MC (n = 3; #08, #10, #12). Two samples (#06 and #11) did not match with any MC (prediction score ≤0.3). By using the v12.8 version of the classifier, the same MCs were recorded although the score might be slightly different between the v12.5 and the v12.8 (see Table S1).

Unsupervised hierarchical clustering and t‐SNE analysis were performed including the present series with eight previously reported low‐grade neuroepithelial tumours FGFR2‐fused (LGNET‐FGFR2) from Gupta et al., and the reference cohort described above [9].

Hierarchical clustering resolved into three clusters (Figure 4). The first cluster (outlier cluster) included two outlier samples (#06 and #11) unclassified by the brain tumour classifier. The second cluster (GG and MVNT cluster, n = 10) included six samples from the present series (#01, #02, #05, #04, #14, #15) and four LGNET‐FGFR2 from Gupta et al., with the following characteristics: Six tumours were located in the temporal lobe and four in the parietal lobe, eight tumours had a neuronal component (either ganglionic for #01, #02, LGNETFGFR2#5, LGNETFGFR2#8; or immature for #04, #05, #15; or both for LGNETFGFR2#3); six tumours had type B neuroradiological pattern; six tumours were matching with GG MC, and three tumours had a calibrated score under 0.9 for the GG MC. The third cluster (PLNTY cluster, n = 11) included seven samples from the present series (#08‐08bis, #09, #10, #12, #13, #16) and four LGNETFGFR2 from Gupta et al. with the following characteristics: Four tumours were located in the temporal lobe (#10 and LGNETFGFR2#2, LGNETFGFR2#4, LGNETFGFR2#7), three tumours in the occipital lobe (#09, #12 and LGNETFGFR2#6), one tumour was insular (#08‐#08bis), one frontal (#16) and one parietal (#13); only one tumour had neoplastic ganglion cells (LGNETFGFR2#6); six cases displayed the type A neuroradiological pattern (#08‐#08bis, #09, #10, #12, #13, #16); one sample (#08) was matching with the PLNTY MC with a calibrated score >0.9, and three samples (#08bis,#10 and #12) had calibrated scored under 0.9 for the PLNTY MC, one sample was matching with the GG MC (LGNETFGFR2#2), and six had a calibrated score <0.9 for the GG MC (#09, #13, #16 and LGNETFGFR2#4, LGNETFGFR2#6, LGNETFGFR2#7) (Figure 4).

FIGURE 4.

Unsupervised hierarchical clustering of DNA‐methylation data. Unsupervised hierarchical clustering of 23 FGFR2‐fused glial/glioneuronal tumours (15 samples of the present series and eight samples (LGNET‐FGFR2) from Gupta et al. [9]) based on the 10,000 most variably methylated probes (only 2000 CpG are displayed on the heat map for easier reading). Samples with calibrated scores (CS) >0.9 were considered as matching with the methylation classes (MC) proposed by the v12.5 version of the DKFZ classifier, CS ≤0.9 were considered not matching. DNT, dysembryoplastic neuroepithelial tumour; GG, ganglioglioma; HEMIS, hemispheric; LGNET, low‐grade neuroepithelial tumour; MC, methylation class; MVNT, multinodular and vacuolating neuronal tumour; NA, not available; NOS, not otherwise specified; PA, pilocytic astrocytoma; PLNTY, polymorphous low‐grade neuroepithelia tumour of the young; SEGA, subependymal giant cell astrocytoma; t‐SNE, t‐distributed stochastic neighbour embedding.

By t‐SNE reduction dimension analysis, although the separation appeared less clear‐cut, we observed two clusters closely reproducing the results of hierarchical clustering. Two outliers were observed: sample #06 between GG and hemispheric pilocytic astrocytoma MC and sample #11 clustered with reactive tumour microenvironment MC (Figure 5A). DBSCAN clustering of t‐SNE results for GG and FGFR2 samples was performed and indicated relatively similar clusters although four reference GG samples were split off as a separate group (Figure 5B,C).

Radiological pattern type A was significantly associated with the PLNTY hierarchical cluster (p = 0.0001) and t‐SNE cluster (p = 0.0002). Ventricular contact tended to be associated with PLNTY hierarchical cluster (p = 0.059). There was no association between t‐SNE or hierarchical clusters and tumour location.

CNV data obtained from the methylation arrays did not show gains of chromosome 7/losses of chromosome 10, EGFR amplification or CDKN2A/B deletion. The FGFR2 fusion transcript partner was not correlated with the clinical, pathological or neuroradiological features. Details of all clinical, histopathological, molecular and neuroradiological data available are reported in Table S1.

DISCUSSION

The multi‐omic approaches in neuropathology, especially in the field of paediatric neuro‐oncology, have strongly contributed to the introduction of new tumour types in the latest edition of the WHO classification of CNS tumours [6]. However, it appears that one single molecular alteration is rarely sufficient to diagnose one tumour type, especially in the field of paediatric LGG and glioneuronal tumours. Similarly, although DNA‐methylation profiling has provided invaluable help towards a better classification of CNS tumours as well as the emergence of new tumour types [19, 23, 24], it appears that the MC (as defined by the DKFZ classifier) of some tumours may change depending on the updated version of this classifier, and this may become more challenging with the emergence of new classifiers. As an example, some FGFR3::TACC3 tumours classified as glioblastoma IDH‐wild type by version 11b6 were reclassified as GG with version 12.5 of the DKFZ classifier [5, 25]. Although the idea that one molecular alteration equals one tumour type is appealing, as in the case of IDH‐mutated gliomas, things seem more complex in the case of FGFR alterations and MAP kinase pathway alterations in general. Nevertheless, a better characterisation of tumours that share a single common molecular alteration is particularly attractive in the era of targeted therapies [26]. In this study, we focused on FGFR2‐fused tumours to better characterise their clinical, neuroradiological, histopathological and epigenetic spectrum. FGFR2 fusions have only recently been described in neuroepithelial tumours [27]. Although FGFR2 fusions have been associated with the epigenetic class of PLNTY [9], results of the present study showed that this fusion can be observed in other neuroepithelial tumours that fulfilled histologically the diagnosis of GG, MVNT or remained unclassified. This suggests that FGFR2‐fused tumours might be observed in several tumour types among LGNET that can be very difficult to distinguish on histopathological level [8, 9, 10, 11, 12]. Of interest, until now FGFR2‐fusions have always been encountered in benign tumour types that fulfilled LGNET criteria. This is in contrast with FGFR1 or FGFR3 fusions that can be observed in a broader spectrum of tumour types, some of which are malignant [5, 28, 29].

It also appeared that the DNA‐methylation analysis, of FGFR2‐fused tumours from the present series with those previously published by Gupta et al. [9], failed to identify clearly separated methylation clusters associated with distinct tumour types, as defined on the histopathological level. Although unsupervised DNA‐methylation hierarchical clustering and t‐SNE suggested that FGFR2‐fused tumours resolved into two main clusters: a PLNTY cluster that included mainly histological PLNTY and another cluster including glioneuronal and neuronal tumours (histologically resembling GG and MVNT); dbSCAN analysis did not confirm this hypothesis. Indeed, we observed that the dbSCAN appeared to split off four reference GG samples as a separate group, and therefore, we could not fully exclude that spatial clustering is effectively statistical noise. Besides, we do not obverse a specific MC for cases exhibiting MVNT histological features in line with the literature that has not yet reported a specific MVNT MC [10, 11, 30]. Unfortunately, it was not possible to enrich our t‐SNE with cases exhibiting BRAF V600E or MAP 2 K1 mutation and classified at histology either as PLNTY, GG or MVNT. The epigenetic profiles of PLNTY and GG are highly similar and are thus by extension likely to cluster closely together [8]. Therefore, there is relatively little true separation of FGFR2 fusion‐positive tumours, both at the histopathological and at the DNA‐methylation level, suggesting that FGFR2 fusions are observed in a spectrum of benign LGNET tumours, extending the GG epigenetic cluster.

Of particular interest, we observed that FGFR2‐fused tumours exhibited two main neuroradiological patterns based on their circumscribed or infiltrative presentation with additional features on MRI. Type A pattern was defined as a well‐circumscribed intra‐axial mass, frequently with ventricular contact, with minimal or no perilesional oedema, presenting a heterogeneous signal on T1 and T2‐weighted images, moderate contrast enhancement and coarse calcifications. Type B was characterised as an infiltrative area with high FLAIR signal intensity, potentially associated with a microcystic component (<1 cm) that may exhibit enhancement and lack of coarse calcification. These two patterns tended to be associated with distinct histopathological diagnoses. In our series, pattern A was recorded in 4/5 histologically defined PLNTY, but also in three histopathologically unclassifiable tumours lacking a neuronal component. Interestingly, pattern A was consistent with previously described radiological features of PLNTY [31, 32, 33, 34]. Of interest, pattern B characterised tumours exhibiting a neuronal component: the four histological MVNT, and the three histological GG but also an unclassifiable case devoid of a neuronal component on pathological examination (of interest, this case clustered on t‐SNE within the group of GG). Moreover, pattern B appeared to share similarities with BRAF‐altered GG described by Stone et al. in terms of temporal localisation and lack of contrast enhancement [35]. The rather different neuroradiological patterns could be explained by the FGFR2 differential expression profile in oligodendrocyte development. In mice, FGFR2 is thought to be expressed by multipotent neuroepithelial cells in the ventricular zone, then postnatally downregulated and re‐expressed in differentiated oligodendrocytes of the white matter in adulthood [36]. Consequently, it could be hypothesised that FGFR2‐fused tumours have the same lineage origin but a different developmental cell fate, explaining the epigenetic vicinity and histologic similarity but also the radiological phenotypic difference. Larger sample sizes are needed to improve resolving power in order to separate robustly and confidently closely related groups. These results question the current diagnostic criteria of some tumour types such as PLNTY, or MVNT and suggest the addition of other criteria (such as age, location and radiological features) in the diagnostic work‐up of these tumour types.

A recent review of the literature of 67 cases of PLNTY (mainly defined histopathologically) revealed heterogeneity of molecular alterations associated with the diagnosis of PLNTY given that most published cases exhibited BRAF V600E mutation (57.1%), whereas 28.5% demonstrated FGFR2 fusion and 7.1% FGFR3::TACC3 fusion, but also NTRK fusions have been reported in this tumour type [34, 37]. Among FGFR2 fusions, the most frequently reported fusion partners were INA, CTNNA3 and SHTN1 (previously KIAA1598) [34]. Here, we report two new partners: HDX and FLT3. Of importance, only a limited number of PLNTY was published with DNA‐methylation data supporting this diagnosis (n = 26) [4, 7, 8, 9], with a majority of FGFR2 fusions or BRAF mutations. However, the lack of PLNTYs defined by DNA‐methylation profiling represents a limitation in the interpretation of literature data for a newly described and rare tumour type whose diagnostic criteria lack specificity. The initial report on PLNTY described heterogeneous histological findings and an absence of a neuronal component [8]. The phenotypic heterogeneity is taken further by Gupta et al. who report the presence of a neuronal component in 3/9 samples [9]. Moreover, the WHO classification does not specify how to integrate the presence of ganglionic neurons into PLNTYs [6]. Here, FISH analysis suggested the presence of FGFR2 rearrangement in the neuronal component of tumours, supporting the existence of a tumoural neuronal component in some FGFR2‐fused glioneuronal tumours. Only two FGFR3‐fused cases with DNA‐methylation profiling have been reported so far, including one with a malignant transformation [8, 38]. Of interest, we learned from the above‐cited meta‐analysis that PLNTY exhibiting BRAF V600E mutation had a better prognosis than FGFR2‐fused PLNTY and none of these cases displayed recurrence [34]. In line with this meta‐analysis, we observed that 5/15 cases of the present series with available follow‐up exhibited tumour recurrence. In contrast to FGFR3::TACC3‐fused LGG and glioneuronal tumours that might demonstrate TERT promoter mutation [5], TERT promoter was always wild type when studied (10/10 cases) in the FGFR2‐fused tumours included in the present series.

The problem of differential diagnosis between PLNTY and GG remains, particularly in the case of tumours with an FGFR2 fusion. Although the data presented here is difficult to interpret in comparison with previous studies, notably because different reference cohorts are used to study these very rare tumours, here our data suggest an epigenetic and phenotypic heterogeneity of FGFR2‐fused tumours and therefore a cautious diagnostic approach based on the integration of radiological features to the diagnostic work‐up. This study shows the importance of enriching these rare tumour cohorts with new cases and of obtaining as much data as possible to better characterise them.

AUTHOR CONTRIBUTIONS

Dominique Figarella‐Branger and Alice Métais supervised the study and wrote the manuscript. Alice Métais, Jeremy Garcia, Myriam Edjlali and Volodia Dangouloff‐Ros designed the figures. Arnault Tauziède‐Espariat, David Meyronet, Emmanuelle Uro‐Coste, Claude‐Alain Maurage, Marie Csanyi, Valérie Rigau, Audrey Rousseau, Guillaume Chotard, Johan Pallud, Lelio Guida and Thomas Blauwblomme provided tumour samples (FFPE) and/or related available imaging and clinical data. Myriam Edjlali, Volodia Dangouloff‐Ros, Quentin Vannod‐Michel, Morgan Ollivier, Margaux Roques, Corentin Provost and Jean‐Philippe Cottier performed the central review of neuroradiology. The Biopathology RENOCLIP‐LOC network contributed to the pathological central review of a few tumour samples and performed some molecular studies. Carole Colin and Dominique Figarella‐Branger monitored the data in the RENOCLIP‐LOC e‐CRF. Dominique Figarella‐Branger and Alice Métais made the selection of the FFPE material and performed the pathological review. Jeremy Garcia and Romain Appay processed the data from the DNA‐methylation analysis. Jocelyne Hamelin and Gaelle Pierron performed RNA sequencing analysis. All authors read and approved the final manuscript.

CONFLICT OF INTEREST STATEMENT

The authors declare that they have no competing interests.

PEER REVIEW

The peer review history for this article is available at https://www.webofscience.com/api/gateway/wos/peer-review/10.1111/nan.13014.

ETHICS APPROVAL AND CONSENT TO PARTICIPATE

Written informed consent to participate in this study was provided by the participant or their legal guardian/next of kin. This study was reviewed and approved by the Aix‐Marseille University ethics committee (2019‐25‐04‐003).

Supporting information

Table S1. Supporting Information

Figure S1: Illustration of Chromogranin A and MAP 2 expression profile in ganglioglioma and MVNT tumours

The left side panel illustrates two cases (#01 and #06) of ganglioglioma tumours displaying a neoplastic neuronal component co‐expressing Chromogranin A and MAP 2, whereas, on the right side, an MVNT tumour (#04) is illustrated and shows MAP 2 expression without Chromogranin A expression. MVNT: multinodular and vacuolating neuronal tumour.

Métais A, Dangouloff‐Ros V, Garcia J, et al. Phenotypic and epigenetic heterogeneity in FGFR2‐fused glial and glioneuronal tumours. Neuropathol Appl Neurobiol. 2024;50(6):e13014. doi: 10.1111/nan.13014

DATA AVAILABILITY STATEMENT

The data that supports the findings of this study are available in the Supporting informationof this article, epigenetic data are openly available in ArrayExpress at https://urldefense.com/v3/__https:/www.ebi.ac.uk/biostudies/arrayexpress/studies/E‐MTAB‐13847__;!!JQ5agg!Yq7JeqLshD_vFt3UijYHLzZ5AbSjge4mjxFDrkIsMXL5nkKSe7id2RujxXtRTuKD2a6mE_Ir2GU2s0ecO3TzeclSKoM$, reference number E‐MTAB‐13847.