IDH-wild type glioblastomas featuring at least 30% giant cells are characterized by frequent RB1 and NF1 alterations and hypermutation

Giant cell glioblastoma (GC-GBM) is a rare variant of IDH-wt GBM histologically characterized by the presence of numerous multinucleated giant cells and molecularly considered a hybrid between IDH-wt and IDH-mutant GBM. The lack of an objective definition, specifying the percentage of giant cells required for this diagnosis, may account for the absence of a definite molecular profile of this variant. This study aimed to clarify the molecular landscape of GC-GBM, exploring the mutations and copy number variations of 458 cancer-related genes, tumor mutational burden (TMB), and microsatellite instability (MSI) in 39 GBMs dichotomized into having 30–49% (15 cases) or ≥ 50% (24 cases) GCs. The type and prevalence of the genetic alterations in this series was not associated with the GCs content (< 50% or ≥ 50%). Most cases (82% and 51.2%) had impairment in TP53/MDM2 and PTEN/PI3K pathways, but a high proportion also featured TERT promoter mutations (61.5%) and RB1 (25.6%) or NF1 (25.6%) alterations. EGFR amplification was detected in 18% cases in association with a shorter overall survival (P = 0.004). Sixteen (41%) cases had a TMB > 10 mut/Mb, including two (5%) that harbored MSI and one with a POLE mutation. The frequency of RB1 and NF1 alterations and TMB counts were significantly higher compared to 567 IDH wild type (P < 0.0001; P = 0.0003; P < 0.0001) and 26 IDH-mutant (P < 0.0001; P = 0.0227; P < 0.0001) GBMs in the TCGA PanCancer Atlas cohort. These findings demonstrate that the molecular landscape of GBMs with at least 30% giant cells is dominated by the impairment of TP53/MDM2 and PTEN/PI3K pathways, and additionally characterized by frequent RB1 alterations and hypermutation and by EGFR amplification in more aggressive cases. The high frequency of hypermutated cases suggests that GC-GBMs might be candidates for immune check-point inhibitors clinical trials.

by bizarre multinucleated giant cells [1]. It is reported to affect younger subjects and to have a relatively better prognosis compared to conventional IDH-wt GBM [5].
It is still unclear whether GC-GBM represents a distinct entity or only a morphological variant of IDH-wt GBM. Most of our current knowledge on its genetic features comes from few available molecular studies, mainly focusing on the analysis of selected genetic anomalies [6][7][8][9][10][11]. According to these, GC-GBM seems to be a hybrid between IDH-wt and IDH-mutant GBM. Similarly to the former, it has a high prevalence of PTEN mutations (18/58 cases, 31%), but alike the latter, it also shows a high incidence of TP53 mutations (73/83 cases, 88%), low frequency of EGFR amplification (10/89 cases; 11%) and of TERT promoter mutations (21/65, 32%) [6][7][8][9][10][11]. Only one study performed a comprehensive molecular profiling of 10 GC-GBMs by whole exome sequencing [10]. In addition to confirming that GC-GBM has frequent impairment of TP53/MDM2 (5 cases) and PTEN/PI3K (4 cases) pathways, it suggested that this morphological variant may be characterized by mutations in chromatin remodeling genes SETD2 (3 cases) and ATRX (2 cases) and alterations in RB1 (2 cases) [10]. Of note, one of the cases showed elevated tumor mutational burden (TMB) in association with MSH6 somatic mutation [10], which may indicate that this is an additional, though exceptional, feature of this variant.
Based on its heterogeneous DNA-methylation profile, GC-GBM is not currently considered to represent a distinct molecular entity [12]. However, due to the lack of an objective definition, specifying the exact percentage of giant cells required for this diagnosis, the molecular portrait of GC-GBM is hardly definable. In a recent paper, the mutation frequencies of TP53, ATRX, RB1, and NF1 were significantly higher in 17 GBMs featuring > 30% giant cells than in 357 IDH-wt GBMs in the TCGA Pan-Cancer Atlas cohort [6].
In order to clarify the molecular landscape of GC-GBM, in this study we explored the mutations and copy number variation (CNV) of 458 cancer-related genes, microsatellite instability (MSI) and TMB, in 39 GBMs featuring at least 30% multinucleated giant cells and dichotomized into having 30-49% (15 cases) or ≥ 50% (24 cases) GCs.

Cases
Thirty-nine formalin-fixed paraffin-embedded (FFPE) surgically resected and treatment naïve GBMs, featuring at least 30% multinucleated giant (i.e. having from few to more than 20 nuclei and a minimum diameter of 20 µm), bizarre (i.e. with atypical, hyperchromatic nuclei, and with evident nucleoli at times), with positive GFAP staining or not, were included in this study.
Taking as a reference the method proposed by Cantero et al. [6], the percentage of multinucleated giant cells was manually quantified by counting at least 1000 neoplastic cells in 10-20 random fields at 200 × magnification.
All cases were independently revised by three pathologists (VB, MM, CG), who assessed the percentage of giant cells. In case of disagreement, the cases were reviewed using a multi-headed microscope. The paraffin block with the highest number of GCs was selected for the subsequent molecular and immunohistochemical analyses.
Data on the overall survival (OS) were retrieved using clinical records.

Mutational and copy number variation status of cancer-related genes
Tumor mutational burden, mutations and copy number variations of 409 cancer-related genes were assessed using the targeted next generation sequencing (NGS) panel Oncomine Tumor Mutational Load (TML) (Ther-moFisher), which covers 1.65 Mb of genomic space.
The results were confirmed using the SureSelectXT HS CD Glasgow Cancer Core assay (Agilent) in 29 GC-GBMs (cases 42GL-71GL).
DNA was obtained from 10 FFPE consecutive 4-μm sections using the QIAamp DNA FFPE Tissue Kit (Qiagen) and qualified as reported elsewhere [13].
Sequencing was performed on Ion Torrent platform using 20 ng of DNA for each multiplex PCR amplification and subsequent library construction. The quality of libraries was evaluated using the Agilent 2100 Bioanalyzer on-chip electrophoresis (Agilent Technologies). Libraries were clonally amplified by emulsion PCR with Ion OneTouch OT2 System (Thermofisher) and sequencing was run on Ion Proton (Thermofisher) loaded with Ion PI Chip v3.
Torrent Suite Software v.5.10 (Termofisher) was used for data analysis, including alignment to the hg19 human reference genome and variant calling. Filtered variants were annotated using a custom pipeline based on vcflib (https:// github. com/ ekg/ vcflib), SnpSift [14], Variant Effect Predictor (VEP) [15] and NCBI RefSeq database. Additionally, alignments were visually verified with the Integrative Genomics Viewer (IGV) v2.9 [16] to confirm Page 3 of 15 Barresi et al. Acta Neuropathologica Communications (2021) 9:200 the presence of identified mutations. Germline mutations were assigned based on Sun et al. [17]. CNV was evaluated using OncoCNV v6.8 [18], comparing the BAM files obtained from tumor samples with those obtained from blood samples of four healthy males. The software includes a multi-factor normalization and annotation technique enabling the detection of large copy number changes from amplicon sequencing data and permits to visualize the output per chromosome.

Confirmation of mutational and copy number variation status of 125 cancer-related genes and further exploration of 49 genes
Twenty-nine cases (42GL-71GL) were additionally analyzed using the SureSelectXT HS CD Glasgow Cancer Core assay (www. agile nt. com), hereinafter referred as CORE [19] (details in Additional file 2). This spans 1.85 Mb of the genome and interrogates 174 genes (49 of which are not included in the TML panel) for somatic mutations, copy number alterations and structural rearrangements.
Sequencing libraries were prepared by targeted capture using the SureSelect kit (Agilent Technologies) according to the manufacturer instructions as previously described [20]. Genomic DNA was enzymatically fragmented with the SureSelect Enzymatic Fragmentation Kit (Agilent Technologies). Quality and quantity of pre-capture libraries was assessed using the Qubit BR dsDNA assay (ThermoFisher). Hybridization-capture and purification of the libraries was performed using 100 ng from each pre-capture library to prepare 16-library pools (1.6 µg of total pooled DNA). Captured library pools were enriched by PCR, purified, and quantified using the Qubit dsDNA HS assay. Quality of the library pools was verified with the Agilent 4200 Tape Station and High Sensitivity D1000 ScreenTape (Agilent Technologies). Sequencing was performed on a NextSeq 500 (Illumina) loaded with 2 captured library pools, using a high-output flow cell and 2 × 75 bp paired-end sequencing.
CORE panel analysis was performed as previously described [20]. Briefly, demultiplexing was performed on the BaseSpace Sequence Hub (https:// bases pace. illum ina. com). Paired-end reads were aligned to the human reference genome (version hg38/GRCh38) using BWA and saved in the BAM file format [21]. BAM files were sorted, subjected to PCR duplicate removal, and indexed using biobam-bam2 v2.0.146 [22]. Coverage statistics were produced using samtools [23]. Single nucleotide variants were called using Shearwater [24]. Small (< 200 bp) insertions and deletions were called using Pindel [25]. Small nucleotide variants were further annotated using a custom pipeline based on vcflib (https:// github. com/ ekg/ vcflib; last access 11/30/2020), SnpSift [14], the Variant Effect Predictor (VEP) software [15], and the NCBI Ref-Seq transcripts database (www. ncbi. nlm. nih. gov/ refseq/). Annotated variants were filtered keeping only missense, nonsense, frameshift, or splice site variants. All candidate mutations were manually reviewed using Integrative Genomics Viewer (IGV), version 2.9 [16] to exclude sequencing artefacts. Gene copy number alterations were detected using the geneCN software (https:// github. com/ wwcrc/ geneCN). Whole-chromosome or chromosomearm alterations were assessed by measuring the ratio of normalized, GC-adjusted coverage of tumor samples' alignments to the mean, normalized, GC-adjusted coverage of 20 non-neoplastic samples for all targeted regions of a chromosome arm. Targeted regions included both targeted genes and a set of "backbone" regions probing each chromosome at 1 megabase intervals. Each large alteration was further confirmed by checking the copy number status of targeted genes included in the large alteration itself as reported by the geneCN software.

Classification of genetic variants
Following the five-tier classification system recommended by the joint consensus of the American College of Medical Genetics and Genomics and the Association for Molecular Pathology (ACMG/AMP) [26], variants were classified: Benign (class 1); Likely Benign (class 2); Variant of Un-certain Significance (VUS -class 3); Likely Pathogenic (class 4); Pathogenic (class 5). Variants' classification was retrieved from the ClinVar database when available (https:// www. ncbi. nlm. nih. gov/ clinv ar/) and accepted when the record complied with the following requisites: reviewed by expert panel according to the ACMG/AMP guidelines and/or reported by multiple submitters with evaluation criteria according to the ACMG/AMP guidelines and no conflicts. When a consistent classification was unavailable or when the variant was not present in the ClinVar database, variants were evaluated in-house, according to the ACMG/AMP guidelines using also the following databases and software to gather and integrate all relevant information: My Cancer Genome (https:// www. mycan cerge nome. org), Intogen [27] (https:// www. intog en. org/) and QIAGEN Clinical Insight (QCI) software (https://variants. qiagenbioinformatics.eu/qci/).

TERT promoter mutational analysis
TERT was amplified by PCR and both strands were sequenced using the ABI PRISM 3500 Genetic Analyzer (Applied Biosystems) as previously described [28]. The primers used were: TERT-F GTC CTG CCC CTT CAC CTT and TERT-R GCA CCT CGC GGT AGTGG. Tumor mutational burden TMB and mutational spectrum were evaluated using the Oncomine TML 5.10 plugin available on IonReporter software (Thermofisher). Default Modified parameters were used to exclude sequencing artefacts. In detail, a threshold of at least 60 reads and 10% allelic frequency was used for variant calling. TMB was expressed as the number of mutations per Mb (muts/ Mb), where mutations included nonsynonymous missense and nonsense single nucleotide variants (SNVs) detected per Mb of exonic sequences.

Comparison with GBMs IDH-wt and IDH-mutant in The Cancer Genome Atlas database
In order to compare the clinical and genetic findings in this cohort of GBMs with giant cells with those in IDHwt and IDH-mutant GBMs, we accessed The Cancer Genome Atlas (TCGA) databases for GBMs (cbioportal.org) and retrieved data from the series of "Glioblastoma Multiforme (TCGA PanCancer Atlas)".

Statistical analysis
We used Chi-squared and Mann-Whitney tests to analyze the correlation between the percentage of giant cells or TMB and the various genetic alterations, and to assess the statistical difference in the patients age, frequency of genetic alterations or in TMB between the present 39 GBMs with giant cells and IDH-wt or IDHmutant GBMs in TCGA PanCancer Atlas.
Overall survival (OS) of the patients was assessed by the Kaplan-Meier method, using the date of surgery as the entry data and the length of survival until the patient's death as the endpoint. Patients who died of GBM-independent diseases were censored. Mantel-Cox log-rank test was applied to assess the strength of association between OS and each variable. Successively, a multivariate analysis (Cox regression model) was utilized to determine the independent effect of the variables on OS.
Mantel-Cox log-rank test was also carried out to analyze the difference in the OS of patients win this cohort with and those with IDH-wt or IDH-mutant GBM in TCGA PanCancer Atlas.
A P-value < 0.05 was considered as significant. All analyses were performed using MedCalc for Windows version 15.6 (MedCalc Software, Ostend, Belgium) and R v.3.2.1.

Cases
The clinical-pathological features of the 39 GBMs are summarized in Additional file 3: Table 1.
Male to female ratio was 2:1 (26 male and 13 female patients) and median age was 63 years (mean age: 57.6 years, range 15-84). Fifteen patients were < 55 years, while 24 were ≥ 55 years. All tumors were localized in the brain lobes, except for 3 cases that were in the third ventricle. All patients had surgery, followed by chemotherapy with temozolomide and radiotherapy.

Mutational status of 458 genes
The alterations in 409-cancer related genes, detected in the 39 GBMs using TML and CORE panels, and those in additional 49 genes using the CORE panel in a subset of 29 cases (42GL-71GL) are summarized in Fig. 2, detailed in Additional file 3: Table 2 and described below according to the altered pathway. Genes' mutations and CNV were not significantly different according to the percentage of GCs. Regarding the 125 genes in common between the two panels, the CORE confirmed the presence of the alterations identified using the TML panel in the subgroup of 29 cases (cases 42GL-71GL).

MMR genes
Nine (23%) GBMs had sequence alterations in MMR genes. Three had somatic mutations of MSH2, three featured somatic mutations of MSH6 and one had a somatic mutation of MLH1. One additional case had concurrent somatic mutation of MLH1 and germinal mutation of MSH2 and another had a germinal mutation of MSH6.

Other genes
GBMs featured mutations in other genes, among which NF1 was the most frequently mutated (10/39; 25.6%). Of note, one case (48GL) had POLE mutation.

Correlation of MMR immunohistochemistry, MSI status and MMR gene mutations and TMB
Of the 8 cases with MMR protein losses, only 2 featured MSI, while 5 with concordant losses and the case with loss of MSH6 only had stable microsatellites ( Fig. 1; Additional file 3: Table 3). Of the 2 cases with MSI, one had MMR gene mutations (29GL), while the other case (30GL) had no MMR gene mutations. Of the 37 cases with stable microsatellites, 8 showed MMR gene mutations. These included 2 with retained MMR proteins, 1 with concordant loss of MLH1/PMS2, four with concordant loss of MSH2/MSH6 and 1 with loss of MSH6 (Additional file 3: Table 3).
Of the 16 hypermutated cases, 2 had MSI and matched loss of MSH2/MSH6 proteins or MLH1/PMS2, 5 had stable microsatellites and the matched loss of MSH2/MSH6 (4 cases) or of MLH1/PMS2 (1 case), 1 had stable microsatellites and the isolated loss of MSH6 protein and 9 had stable microsatellites and no MMR loss.

Survival analysis
Information on the OS was available for all patients. At the last follow-up time, 15 patients were alive and 24 had died of GBM. OS ranged between 4 and 27 months for died patients, while follow-up time ranged between 2 and 72 months for alive patients (Additional file 3: Table 1).
Multivariate analysis, including age of the patients, EGFR amplification and hypermutation as covariates, showed that all three were independent prognostic variables ( Table 1).

Comparison of the present GBM series with the TCGA PanCancer Atlas GBM series
To clarify whether GBMs featuring > 30% GCs are a distinct group, we compared their clinical features, TMB and genes mutations/CNV with those of 567 IDH-wt and 26 IDH-mutant GBMs in TCGA PanCancer Atlas series.
The age of the patients in the present series was significantly higher than that of the patients with IDH-mutant GBMs (P = 0.0001), but not different from that of the patients with IDH-wt GBM (P = 0.440) ( Table 2).
In the TCGA series, 551 IDH-wt and 24 IDH-mutant GBMs were profiled for CNV; 371 IDH-wt and all 26 IDH-mutant GBMs were profiled for gene mutations.

Discussion
The 2016 WHO classification defines GC-GBM as a variant of IDH-wt GBM characterized histologically by numerous multinucleated giant cells and molecularly by a high frequency of TP53 mutations and rare EGFR amplification [1].
In this study on 39 GBMs featuring a percentage of giant cells ranging between 30 and 90%, the alterations found in 458 cancer-related genes analyzed with NGS were not associated with the giant cell content (30% -50% or > 50%). As expected, no cases had IDH1/2 mutations and a high percentage (82%) featured alterations of TP53/MDM2 pathway. However, a consistent proportion (69.2%) of GC-GBMs also harbored alterations in RB1/CDKN2A/CDK4 pathway, with 25.6% cases having impairment of RB1, 33.3% displaying CDKN2A homozygous deletion and 10% showing CDK4 amplification. EGFR amplification was found in 18% cases and was significantly correlated to a worse prognosis. Other frequent alterations were detected in NF1 (25.6%), chromatin remodeling genes (25.6%) (including 12.8% mutations in ATRX and 7.6% in SETD2), and MMR genes (23%). The comparison with GBMs in the TCGA PanCancer Atlas cohort revealed that the rates of TP53 and ATRX mutations, PTEN alterations, EGFR amplification and CDKN2A/B homozygous deletion in the present series were intermediate between those found in IDH-wt and IDH-mutant GBMs. In contrast, the frequency of RB1 or NF1 (25.6%) alterations was significantly higher than in both TCGA groups (14% vs 3.8%, for RB1; 12.1 vs 3.8% for NF1), suggesting that this is a distinctive feature of GBMs enriched in GCs. In accordance, 2/10 (20%) GC-GBMs analyzed in a previous study by whole exome sequencing had RB1 mutations [10], and 8 (47%) and 6 (35%) of 17 GBMs with > 30% giant cells had RB1 and NF1 mutations in another [6]. One of the present GBMs had a pathogenic POLE mutation, similarly to other reported cases of GBMs enriched in giant cells [6,28,31,32], which suggests that also POLE mutations may be part of the molecular portrait of GC-GBM.
Therefore, our findings confirm and expand the concept that GC enriched GBM is a peculiar entity, distinct from either IDH-wt or IDH-mutant GBM. In most cases (32 cases; 82%), it is driven by the alteration of P53 function due to either TP53 gene mutations (29 cases, 74.4%) or amplification of its principal cellular antagonist, the MDM2 gene (3 cases, 7.7%). However, it is also enriched in alterations of RB1/CDKN2A/CDK4 pathway and mutations in NF1, POLE, and chromatin remodeling genes.
A major issue in the diagnosis of GC-GBM is represented by the lack of a cut-off of giant cells required for this diagnosis. Only one previous study specified the percentage of giant cells in the cases analyzed [6]. In agreement with our results, it reported that the mutation frequencies of RB1 and NF1 were significantly higher in 17 GBMs with > 30% giant cells than in TCGA IDH-wt GBMs [6]. Moreover, the extrapolated mutation frequencies of RB1 and NF1 were significantly higher in the 17 GBMs with > 30% giant cells (8/17, 47.1%; 5/17, 29.4%) than in the 18 GBMs with < 30% giant cells (2/18, 11%; 2/18, 11%) [6].
Of note, the rate of MMR genes mutations in the present GBMs (9/39 cases; 23%) was significantly higher than in the IDH-wt GBMs of TCGA PanCancer Atlas (1.6%) or in other cohorts of conventional GBMs of adults (3%) and children (6.6%) [33,34]. However, in only one case MMR genes mutations were coupled with MSI, similarly to that found in TCGA PanCancer Atlas, where all 7 MMR-mutated GBMs lacked MSI. In our series, another GBM had MSI but lacked MMR mutations. Both GBMs with MSI had the loss of the matched MMR MSH2/MSH6 protein partners. Nevertheless, MMR losses were found in 6 additional cases with stable microsatellites. The absence of MSI in cases with the immunohistochemical loss of MMR proteins was previously reported in other cohorts of gliomas or in meningiomas [35,36] and suggests caution in the use of immunohistochemistry for MMR proteins as a surrogate of MSI.
MMR deficiency and hypermutation are currently considered as biomarkers predictive of the response to immune checkpoint inhibition [37]. Indeed, it is reported that tumors with MMR deficiency have 10 to 100 times more somatic mutations than MMR-proficient tumors and this hypermutation state could lead to a high neoantigen load and consequent activation of the immune system and tumor destruction [38].
In agreement, one of the present hypermutated GBMs had POLE mutation and 8 had mutations in MMR genes. Of these latter, only one had MSI, suggesting that mechanisms different from defective MMR system may lead to a hypermutational status in gliomas and that the recent proposal to use MMR immunohistochemistry to identify hypermutated cases for immunotherapy should be considered with caution [43].
This study is the first to address the question on whether genetic alterations may have prognostic relevance in GC enriched GBMs. Similar to IDH-mutated or conventional IDH-wt GBMs [44], the presence of EGFR amplification was associated with significantly shorter patients' survival. In contrast to that reported in gliomas treated with temozolomide [36], hypermutation was an independent predictor of longer overall survival. Of note, the OS length overlapped that of patients with IDH-wt GBM in TCGA PanCancer Atlas series, which might suggest that GC variant does not harbor a better prognosis than conventional IDH-wt GBMs. However, the subgroup of patients younger than 55 years had an OS length similar to patients with IDH-mutant GBM and significantly longer than patients with IDH-wt GBMs.