Subject selection

We retrieved from the GHU-Paris-Neuro Sainte-Anne database 28 paediatric or adolescents and young adult (AYA) patients operated for a primary spinal tumour with a pathological diagnostic of LGG or glioneuronal tumuor (GNT), non-ependymal, IDH and H3 wildtype, between 2008 and 2020. Written informed consent to participate in this study was provided by the participants’ legal guardian. Data collection was approved under the public health declaration number as folows: DC-2020-3840. Clinical data were retrospectively collected for each case. They included sex, age at diagnosis, tumour location, age at first surgery, extent of surgical resection (gross total resection vs. biopsy or partial resection), follow up (including date of first progression radiologically confirmed, treatments, date of last follow up and survival status).

Radiology data

Initially, post-operative and follow-up magnetic resonance imaging (MRI) were retrieved from Necker Enfants-Malades Hospital database for every patient. Imaging protocols were very heterogeneous as they were realized at different times and in different radiology centers. Spinal studies usually included sagittal SE (Spin Echo) T1-weighted, SE T2-weigthed and SE T1-weighted after gadolinium injection images. Cerebral explorations included T2-weighted, T2-FLAIR (fluid-attenuated inversion recovery), diffusion and T1-weighted with gadolinium injection images. All MRIs were centrally reviewed in consensus by one experienced paediatric neuroradiologist (VDR) and by one radiology resident (WY). Tumour location, extension (height according to the number of adjacent vertebrae), presence of nodular leptomeningeal metastases and leptomeningeal thin enhancement were assessed. Radiological progression was recorded as local recurrence, growth of existing metastases or the appearance of new metastases.

FISH analyses

Fluorescence in situ hybridization (FISH) studies were performed on interphase nuclei according to the standard procedures and the manufacturer’s instructions and previously published methods [18]. The ploidy for the chromosomes 1 and 19q as well as copy number of the BRAF gene was assessed using the following centromeric and locus-specific probes: Vysis LS1 1p36/1q25 and LS1 19p13/19q13 FISH Probe Kit (Abbott Molecular, USA), ZytoLight® SPEC BRAF Dual Color Break Apart (Zytovision, Germany). NTRK2 rearrangement was assessed using the following centromeric and locus-specific probe: ZytoLight® SPEC NTRK2 Dual color Break Apart (Zytovision, Germany). Signals were scored in at least 100 non-overlapping intact interphase nuclei per case. Gene copy number per nucleus was recorded as follows: one copy, two copies, copy number gain (3–7). Copy gain and deletion were considered if they were detected in more than 10% and 30% of nuclei, respectively. Results were recorded using a DM600 imaging fluorescence microscope (Leica Biosystems, Richmond, IL) fitted with appropriate filters, a CCD camera, and digital imaging software from Leica (Cytovision, v7.4). Normal cells (endothelial cells or normal glial or neuronal cells from the adjacent parenchyma) were used as positive internal controls for locus-specific probes 1p36/1q25 and 19p13/19q13. For BRAF and NTRK2 break apart probes, a positive case with confirmed fusion was used.

Targeted array SNP genotyping and RNA sequencing data

Molecular data from DNA analysis or RNA sequencing were retrieved from pathological reports. Most experiments were performed according to previously described methods [31, 42]. Briefly, total DNA was extracted with the use of the QIAampDNA mini-kit® (Qiagen Inc., Courtaboeuf, France) according to manufacturer’s protocols. Briefly, tissues were disrupted in lysis buffer. After removing paraffin, the DNA was purified via sequential centrifugation through membrane spin columns. The purity and quantity of DNA were assessed by measuring the absorbance ratio at 260/280 nm with a NanoDrop® Spectrophotometer (LabTech, Palaiseau, France). A brain tumour gene mutation panel was developed using the MassARRAY iPlex technology and MassARRAY online design tools (Agena Bioscience), including the following mutations: IDH1 R132HLSGC; IDH2 R172KTMGWS; BRAF V600EGAKRD; EGFR A289TSVD, D770ins; FGFR1 K656EQ, N546K; H3F3A K27M, G34VRW; HIST1H3B K27MT; HISTH3C K27M, TERT promoter c228at, c250t. The MassARRAY iPlex procedure involves a three-step process consisting of the initial polymerase chain reaction (PCR), inactivation of unincorporated nucleotides by shrimp alkaline phosphatase and a single-base primer extension. Then, the products are nano-dispensed onto a matrix-loaded silicon chip (SpectroChipII, Agena Bioscience, San Diego, California, USA). Finally, the mutations are detected by MALDI–TOF (matrix-assisted laser desorption-ionization–time of flight) mass spectrometry. Data analysis was performed using MassARRAY Typer Analyzer software 4.0.4.20 (Agena Bioscience, San Diego, California, USA), which facilitates visualization of data patterns as well as the raw spectra.

RNA was extracted from two 8-μm-thick formol-fixed paraffin-embedded (FFPE) material sections using the high Pure FFPE RNA Isolation Kit (Roche Diagnostics, Boulogne-Billancourt, France) according to the manufacturer’s instruction. RNA concentrations were measured on a Qubit 4 Fluorometer (Thermo Fisher Scientific, USA) with the Invitrogen Qubit RNA BR Kit (Thermo Fisher Scientific). The percentage of RNA fragments > 200 nt (fragment distribution value; DV200) was evaluated by capillary electrophoresis (Agilent 2100 Bioanalyzer). A DV200 > 30% was required to process the next steps in the analysis. Next-generation sequencing (NGS)-based RNA sequencing was performed using the Illumina TruSight RNA Fusion Panel on a NextSeq550 instrument according to the manufacturer’s instructions (Illumina, San Diego, CA, USA). This targeted RNA sequencing panel covers 507 fusion-associated genes, to assess the most known cancer-related fusions. The TruSight RNA Fusion Panel gene list is available at https://www.illumina.com/content/dam/illumina-marketing/documents/products/gene_lists/gene_list_trusight_rna_fusion_panel.xlsx. A total of 7690 exonic regions are targeted with 21,283 probes. Libraries were prepared according to the Illumina instructions for the TruSight RNA Fusion Panel kit. STAR_v2.6.1a or Bowtie software was used to produce aligned reads in relation to the Homo sapiens reference genome (UCSC hg19) [13]. Manta v1.4.0, TopHat2 and Arriba tools were used for fusion calling [9].

Droplet digital PCR

FGFR1 tyrosine kinase domain (TKD) duplication and hotspot mutations (N546K/K656E) were assessed by previously described droplet digital polymerase chain reaction (ddPCR) [2, 15, 16]. Extracted DNA was quantified using the IDQUANTq kit (ID-Solutions, Grabels, France) with the magnetic induction cycler (Mic) PCR Machine Cycler from Bio Molecular Systems (Göttingen, Germany). After quantification, DNA concentration was adjusted. Eight microliters of DNA comprising 1–5 ng and 14 µl of PCR mix (ready to use) were used for each ddPCR assay. A similar amplification program (50 °C 2 min; 95 °C 10 min; 40 × 95 °C 30 ss–60 °C 1 min; 98 °C 10 min) was used for all targets. The QX200 Droplet Digital PCR System (Bio-Rad, Hercules, California, USA) was used with the AutoDG droplet generator (Bio-Rad). Quantasoft Analysis Pro Software v1.0.596 (Bio-Rad) was used for the qualitative and quantitative analyses. Fractional abundance and copy number variations (CNV) were calculated with the cut-off values and detection thresholds defined by Appay et al. [2]. The cut-off value of positive results for mutant detection were two positive droplets. Detection thresholds were set when the number of positives droplets was strictly above the limit of blank at 95% confidence interval defined for each assay depending on the number of replicates.

DNA methylation profiling data

Nineteen tissue samples (FFPE or freshly frozen if available), for which 500 ng of DNA was extracted were analyzed. DNA was extracted using the QIAamp® DNA Tissue kit or QIAamp® DNA FFPE Tissue Kit (Qiagen, Hilden, Germany) for FFPE samples. DNA from FFPE samples was restored using the Infinium HD FFPE Restore Kit (Illumina, San Diego, California, USA). Bisulfite conversion was performed using the Zymo EZ DNA methylation Kit (Zymo Research, Irvine, California, USA). Standard quality controls confirmed DNA quality/quantity and bisulfite conversion. DNA was then processed using the either Illumina Infinium Methylation EPIC or HumanMethylation450 BeadChip (Illumina, San Diego, California, USA) arrays according to the manufacturer’s instructions. The iScan control software was used to generate raw data files in.idat format, analyzed using GenomeStudio software version v2011 and checked for quality measures according to the manufacturer’s instructions.

Affiliation predictions to CNS tumour classes were obtained from a DNA methylation-based classification of CNS tumours from DKFZ (Deutsches Krebsforschungszentrum—German Cancer Research Center) based on a random forest algorithm available on the web platform www.molecularneuropathology.org. Version v12.5 of the algorithm was used for the present study. The output of this classifier is a score (calibrated score, CS) indicating the resemblance to the reference CNS tumour class in the algorithm. We choose a dimension reduction technique for data visualization: the t-SNE algorithm (t-distributed stochastic neighbor embedding). This non-linear method allows the visualization of data in the form of scatter plots and is well suited for the analysis of raw methylation data. Distinct samples from the same tumour type will usually lead to compact clusters. However, there is no distance threshold that can serve to determine if one sample of interest belongs to one particular cluster; we thus consider that a sample belongs to one class of reference if it overlaps the corresponding cluster or fell in the close vicinity. This method is frequently used in cancer research and to study DNA methylation profiling data in CNS tumours. It was used in the original paper on the classification of central nervous system tumours based on DNA methylation profiles by Capper et al. [7]. Parameters used in this study are the same as those from the DKFZ.Data from EPIC and 450k methylation array were analysed with R language (v4.0.4). The minfi package was used to load idat file and preprocessed with function preprocess.illumina with dye bias correction and background correction. We removed probes located on sex chromosomes, not uniquely mapped to the human reference genome (hg19), probes containing single nucleotide polymorphisms and probes that are not present in both EPIC and 450k methylation array. A batch effect correction was done with removebatchEffect function from limma package, to remove difference between FFPE and frozen samples. The probes were sorted by standard deviation with 10,000 most variable probes were kept for the clustering analysis. These probes were used to calculate the 1-variance weighted Pearson correlation between samples. The distance matrix was used as input for t-distributed stochastic neighbor embedding (t-SNE) from Rtsne package, with the following non-default parameters: theta = 0, pca = F, max_iter = 2500 and perplexity = 20. Visualization was done using ggplot2 packages.

CNV analysis of KIAA1549:BRAF fusion and 1p deletion

KIAA1549:BRAF and 1p deletion were searched by visual inspection of CNV profiles generated by the molecularneuropathology.org platform as described in Stichel et al. [43]. Visual inspection indicated a deletion of 1p if a complete loss of chromosome 1p arm was present. A gain of 7q34 region was indicative of the BRAF duplication and KIAA1549:BRAF fusion.

Histopathological analysis, immunohistochemistry and integrated diagnostic

Samples were stained with hematoxylin-phloxin-saffron (HPS) according to standard protocols. Original slides from all tissue samples were centrally reviewed (ATE, AM) on a Nikon Eclipse E600 (Nikon, Japan) light microscope with Nikon Plan Fluor objectives. Due to often limited examinable surface, mitotic activity was monitored on five high-power fields (HPF, 40 ×/0.75) corresponding to 1.6 mm2. The following primary antibodies were used: Glial Fibrillary Acidic Protein (GFAP) (1:200, clone 6F2, Dako, Glostrup, Denmark), Olig2 (1:3000, clone C-17, Santa Cruz Biotechnology, Dallas, USA), CD34 (1:40, clone QBEnd10, Dako, Glostrup, Denmark), Chromogranin A (1:200, clone LK2 H10, Diagnostic Biosystem, Pleasanton, USA), Neurofilament Protein (1:100, clone 2F 11, Dako, Glostrup, Denmark), Synaptophysin (1:150, clone DAK-SYNAP, Dako, Glostrup, Denmark), Ki-67 (1:200, clone MIB-1, Dako, Glostrup, Denmark). Integrated diagnoses were performed according to the current WHO classification [48].

Statistics

Quantitative variables are expressed by median and compared using Mann–Whitney tests. Qualitative variables are expressed by proportions and percentages and compared using Fischer’s exact test. Those analyses were performed in GraphPad Prism version 7.0a. The Reverse Kaplan–Meier method was used to determine the median follow up [37]. Time to treatment initiation (TTI) was calculated as the duration of time between diagnosis and the initiation of first treatment. Progression-free survival, termed PFS-R, was defined as the time between first treatment start and first radiologically confirmed progression on magnetic resonance imaging (MRI). Overall survival (OS) was calculated as the time from the date of diagnosis to the date of death from any cause or the date of last follow-up. Patient were considered disease-free (DF) if gross total resection was achieved and without any signs of disease recurrence at last follow-up. The appearance of new leptomeningeal contrast enhancement and/or of distant metastases, and/or the growth of the main tumour site, as assessed by MRI, were considered as disease progression (DP). Censored variables were analyzed using the Kaplan–Meier method and comparison were assessed using the log-rank test performed in R version 4.1.2 [33]. All statistical significance was considered at a 5% alpha level. Values between brackets [] are 95% confidence intervals unless stated otherwise.