Genomic markers of recurrence risk in atypical meningioma following gross total resection

Abstract Background Meningiomas are the most common primary central nervous system (CNS) tumor in adults and CNS World Health Organization grade 2 (atypical) meningiomas show an intermediate risk of recurrence/progression. Molecular parameters are needed to better inform management following gross total resection (GTR). Methods We performed comprehensive genomic analysis of tumor tissue from 63 patients who underwent radiologically confirmed GTR of a primary grade 2 meningioma, including a CLIA-certified target next-generation sequencing panel (n = 61), chromosomal microarray (n = 63), genome-wide methylation profiling (n = 62), H3K27me3 immunohistochemistry (n = 62), and RNA-sequencing (n = 19). Genomic features were correlated with long-term clinical outcomes (median follow-up: 10 years) using Cox proportional hazards regression modeling and published molecular prognostic signatures were evaluated. Results The presence of specific copy number variants (CNVs), including -1p, -10q, -7p, and -4p, was the strongest predictor of decreased recurrence-free survival (RFS) within our cohort (P < .05). NF2 mutations were frequent (51%) but did not show a significant association with RFS. DNA methylation-based classification assigned tumors to DKFZ Heidelberg benign (52%) or intermediate (47%) meningioma subclasses and was not associated with RFS. H3K27 trimethylation (H3K27me3) was unequivocally lost in 4 tumors, insufficient for RFS analysis. Application of published integrated histologic/molecular grading systems did not improve prediction of recurrence risk over the presence of -1p or -10q alone. Conclusions CNVs are strong predictors of RFS in grade 2 meningiomas following GTR. Our study supports incorporation of CNV profiling into clinical evaluation to better guide postoperative patient management, which can be readily implemented using existing, clinically validated technologies.

Meningiomas are common central nervous system (CNS) tumors, representing approximately 45% of primary CNS tumors in patients over age 40. 1 Within the 2021 World Health Organization (WHO) Classification of CNS tumors, meningiomas are considered to comprise a single entity, assigned a CNS WHO grade 1 (approximately 80%), grade 2 (approximately 20%), or grade 3 (approximately 2%). 2 Grading of meningiomas remains largely based on morphology, with 2021 WHO Classification incorporating TERT promoter mutation and CDKN2A/B homozygous deletion as molecular criteria to support a CNS WHO grade 3 designation. 2 Recent studies uniformly affirm the prognostic significance of WHO grade, but histology alone is inadequate to fully predict recurrence risk. This is particularly true for WHO grade 2 tumors, which have an intermediate recurrence risk and unpredictable clinical course following gross total resection (GTR).
Recent large-scale genomic studies have aimed to identify molecular subgroups of meningioma and molecular markers that better predict recurrence risk. Numerous meningioma classification systems have been proposed based on genomic driver alterations, 3,4 copy number, 5 RNA expression, 6,7 DNA methylation signature, [8][9][10] integrated multi-omic signatures, 8,10,11 or have incorporated both histologic and molecular features. 12,13 When applied across large cohorts of meningioma patients, these molecular signatures may better predict recurrence risk than WHO grade alone. However, most prior studies included meningiomas of all histologic grades and have not carefully controlled for extent of resection. As optimal treatment of patients following GTR of CNS WHO grade 2 meningiomas remains controversial, improved molecular stratification may be of particular benefit in these patients.
In this study, we performed detailed histologic and genomic analyses of a retrospective cohort of patients with radiographically confirmed GTR of histologically CNS WHO grade 2 meningioma, including chromosomal microarray, targeted next-generation sequencing (NGS), DNA methylation profiling, and H3K27me3 immunohistochemistry (IHC), to evaluate for molecular features associated with risk of tumor recurrence.

Patient Cohort
We previously evaluated the impact of radiation therapy across a cohort of 69 patients who underwent radiographically confirmed GTR of WHO grade 2 meningiomas at our institution between 1988 and 2011. 14 All tumors were verified to meet histologic criteria for a CNS WHO 2 designation according to the 2021 WHO Classification of CNS Tumors. Tumor tissue from 63 patients of this initial series were available for further clinical and genomic analysis. Medical records were reviewed (through March 15, 2022) for additional follow-up since publication of the original series in 2017. 14 All aspects of the study were approved by the Institutional Review Board and biospecimens subcommittee.

Chromosomal Microarray Analysis
DNA was extracted from formalin-fixed paraffin embedded (FFPE) tissue sections using the QIAamp DNA FFPE Tissue Kit (Qiagen) as described previously. 15 Copy number changes were assessed using a molecular inversion probe array (MIP) (OncoScan CNV Plus Array, Thermo-Fisher Scientific), which covers common copy number changes across 900 cancer genes. Briefly, isolated DNA was hybridized to MIPs; gap filling was performed utilizing the bound FFPE DNA as a template to circulize MIPs; FFPE DNA was removed using an exonuclease; and the circularized MIPs were linearized, amplified, digested, and hybridized to oligonucleotide arrays. The arrays were scanned on a GeneChip Scanner (Affymetrix), and the images processed with OncoScan Console Software (Affymetrix). CNVs and copy neutral loss of heterozygosity were identified by manual annotation of the data using Chromosome Analysis Suite (Affymetrix). Arm level alterations were defined as greater than 50% of the arm gained or lost relative to the copy neutral state and a deviation of >0.3 from baseline.

Importance of the Study
Meningiomas are the most common primary central nervous system (CNS) tumors in adults. Although the majority are low grade (CNS World Health Organization [WHO] grade 1) with favorable prognosis, atypical (CNS WHO grade 2) meningiomas show an intermediate risk of recurrence and the role of adjuvant radiation therapy following gross total resection (GTR) remains controversial. Molecular parameters may help better inform management following GTR. Our study demonstrates that specific copy number variants (CNVs), including -1p, -10q, -7p, and -4p, are a strong predictor of decreased recurrence-free survival (RFS) (P < .05). In contrast, DNA methylation-based classification was not associated with RFS, nor was the NF2 mutation status. Our study supports incorporation of CNV profiling into clinical evaluation of atypical meningiomas to better guide postoperative patient management and can be readily implemented using existing, clinically validated technologies.

Targeted NGS
To characterize known CNS somatic mutations, a custom GeneRead DNAseq Targeted Neuro-Oncology Panel (Qiagen) (Supplementary Table S1) was used to amplify selected targets and the amplicons were purified using Beckman Coulter's AMPure XP kit. Libraries were prepared using Illumina's TruSeq kit and sequenced on a HiSeq 2500 (Illumina) using the rapid run mode. Variants with allele frequency ≥ 10% from regions with ≥200X coverage were manually annotated as pathogenic or as a variant of unknown significance.

DNA Methylation Profiling
CpG methylation analysis was performed using Infinium 850k Methylation array (Illumina) according to the manufacturer's recommendations. Briefly, 200 ng of DNA extracted from each sample was utilized to restore the degraded FFPE DNA to a state that is analyzable by the Infinium HD FFPE whole genome amplification workflow as described in the Infinium FFPE Restoration guide (Illumina). DNA was denatured using 8 µL of NaOH 0.1N for 10 min at room temperature. A 1 h reaction at 37°C was then performed with Primer Pre-Restore (PPR) and Amplification Mix Restore (AMR) reagents supplied by the kit manufacturer in which DNA repair is accomplished. Restored DNA was cleaned using a ZR-96 DNA Clean and Concentrator-5 kit (Zymo Research) following the manufacturer's protocol. DNA was concentrated between 150 ng/uL of concentration. Between 100 and 200 ng of prepared DNA was used as input for the hybridization on the Illumina 850K Epic BeadChip array and processed according to kit instructions. Data were quality assessed using Illumina Genome Studio software (Illumina), and idat files were uploaded to the DKFZ Heidelberg Classifier v12.5 (www.molecularneuropathology.org). 16

RNA Sequencing
Gene expression analysis was performed using a previously described method. 17 Briefly, RNA was extracted from the frozen tumor tissue material using the AllPrep DNA/RNA FFPE kit (Qiagen) according to the manufacturer's recommendations. RNA was converted to cDNA and cDNA concertation and size distribution was measured using Agilent Fragment Analyzer and Qubit fluorometry (Invitrogen) following the manufacturer's recommendations. Approximately 300 ng of cDNA was utilized to prepare paired-end sequencing libraries using TruSeq RNA Exome kit (Illumina) according to the manufacturer's recommendations. Concentration and size of the prepared libraries were measured as previously described. Prepared libraries were sequenced at 6 samples per lane following Illumina's standard protocol using the Illumina cBot and HiSeq 3000/4000 PE Cluster Kit. The flow cells were sequenced as 101 × 2 paired end reads on an Illumina HiSeq 4000 using the HiSeq 3000/4000 sequencing kit and HD 3.4.0.38 collection software. Base-calling was performed using Illumina's RTA version 2.7.7.
Raw sequencing data was processed using MAPRSeq pipeline to extract gene-wise read count data as previously published. 18 Briefly, FastQC software was utilized to perform quality assessment of the read data of each sample. Sequenced reads were aligned to the hg38 reference using the STAR aligner. 19 HTSeq software was utilized to extract gene-wise read count data based on Ensembl version 78 transcriptome. 20 Analysis of raw counts and figure generation for RNAseq data was completed in R. 21 Raw counts were normalized and differential expression between progressors and non-progressors was calculated using DESeq2. 22 Differential expression was visualized using a volcano plot generated using Enhanced Volcano. 23 Uniform Manifold Approximation and Projection (UMAP) plots were creating using the UMAP (https://CRAN.R-project. org/package=umap) and ggplot2 packages. 24 Correlation between patient transcriptomes was measured by calculating the Pearson's correlation between normalized count profiles. The correlation matrix was visualized using ComplexHeatmap. 25 Genes that had a |log 2 (Fold Change)| > 0.5 and an adjusted P-value < .05 as calculated by DESeq2 analysis were input into WebGestalt for overrepresentation analysis in Reactome Pathways. 26,27 The most significant pathways as measured by false discovery rate (FDR) were plotted as the −log 10 (FDR).

H3K27me3 IHC
IHC was performed on FFPE tissue sections using antibodies directed against H3K27me3 (1:50 titer, clone C36B11, Cell Signaling Technology). Staining was performed on a Benchmark Ultra Immunostainer (Roche Tissue Diagnostics) and detected with OptiView DAB (Roche) after pretreatment with CC1 (Roche). H3K27me3 expression was scored as (1) retained, (2) partial loss, or (3) complete loss based on independent evaluation by two neuropathologists (A.R. and R.A.V.) and consensus review of discordant cases.

Statistical Analysis
Patient characteristics were summarized with frequencies and percentages or medians and ranges, as appropriate. Recurrence-free survival (RFS) was defined as the time from original diagnosis to earliest radiographic evidence of tumor recurrence (defined as the presence of new nodular enhancement in or adjacent to the resection bed after radiographically confirmed GTR), and patients were censored at date of last follow-up or death if they did not have a recurrence. Associations with RFS were assessed with likelihood ratio tests from Cox proportional hazards regression models, and hazard ratios together with 95% confidence intervals were reported. RFS at 5 and 10 years were estimated with the Kaplan-Meier method. The concordance index (c-index) was calculated as a measure of how well each predictor discriminates between those with versus without recurrence. Due to the low number of events (recurrences), no multivariable analysis was performed and no adjustment for multiple testing was performed. P-values less than .05 were considered statistically significant. Analyses were conducted using SAS version 9.4 (SAS Institute Inc.) or R. 21

Patient Characteristics
We evaluated a retrospective cohort of 63 patients who underwent radiographically confirmed GTR of an atypical (CNS WHO grade 2) meningioma at our institution between 1988 and 2011 ( Table 1). The impact of radiation therapy on recurrence and survival across this cohort has been reported previously. 14 Median age at diagnosis was 59.8 years (range 27.4-90.6) and the cohort included 41 (65%) female and 22 (35%) male patients. Tumors were located to the convexities (n = 45, 71%), falx/tentorium (n = 8), skull base (n = 6), sagittal sinus (n = 1), orbit (n = 1), occipital (n = 1), or infratentorial (n = 1) regions. As reported previously, 8 of 62 patients (13%) underwent fractionated radiation therapy as part of their initial treatment plan following GTR (treatment unknown for 1 patient), with a median dose of 5400 cGy (range 5051-6120) over median of 30 fractions (range 28-30). 14 Clinical follow-up was available for all patients, with a median follow-up of 10 years (range 0.8-24.1). Tumors recurred in 17 patients, with a 5-year and 10-year RFS of 84% and 71%, respectively. There was no significant association between patient sex, tumor location, or treatment with radiation therapy and RFS. Overall, the characteristics of our patient cohort are typical of meningioma.

Identification of Molecular Driver Alterations by Targeted NGS
Targeted NGS was performed in 61 tumors using a Clinical Laboratory Improvement Amendments (CLIA)-certified NGS panel covering sequence variants in 50 genes associated with CNS tumors (Figure 1). Of note, this panel includes NF2, SMARCB1, AKT1, and SMO, but does not include TRAF7 or KLF4. Within our cohort, pathogenic alterations of NF2 were frequent (n = 31; 51%). A smaller subset of tumors harbored pathogenic alterations in SMARCB1 (n = 5; 8%), most in conjunction with NF2 mutation (4 of 5). Rare tumors showed mutations of AKT1 (n = 2), SMO (n = 1), SUFU (n = 1), TP53 (n = 1), and PTEN (n = 1). A single tumor harbored a TERT promoter mutation, an alteration associated with poor prognosis and consistent a CNS WHO grade 3 designation within the updated 2021 WHO Classification of CNS tumors. 2 This patient remains alive, with multiple tumor recurrences 15 years after the initial diagnosis.
Neither NF2 nor SMARCB1 mutation showed statistically significant association with RFS (Table 2; Figure 2A). However, NF2-mutant tumors showed an apparent trend toward decreased RFS (5-year and 10-year RFS: 75% and 66% vs 92% and 77%, respectively; P = .08). None of the 5 tumors harboring SMARCB1 mutation recurred, although no statistically significant association with RFS (P = .089) was identified in this limited subset. Overall, within our series, mutation status was not a strong predictor of progression-free survival (PFS).

Neuro-Oncology Advances
patient harbored a segmental loss of one copy of DMD on Xp21.2, which has been associated with shorter survival in meningioma. 29 This patient is alive without recurrence after 20 years of follow-up. No tumors harbored CDKN2A/B homozygous deletion.

Association of Methylation Subgroups With Recurrence Risk
Genome-wide methylation profiling was performed in 62 cases and a methylation class assigned using the DKFZ methylation classifier (version 12.5). In 61 cases, the highest matching methylation class was meningioma (Supplementary Table S2A), while a single tumor grouped with medulloblastoma with a low confidence score (0.07). Using a score cutoff of ≥0.9 or ≥0.85, a definitive match was obtained in 28 (47%) and 33 cases (53%) respectively. The DKFZ classier groups meningiomas into three benign subclasses (1-3), two intermediate subclasses (A-B) Table S2A).
There were no significant differences in RFS based on methylation subtype (Table 2 and Supplementary Table  S2B; Figure 3A and B). RFS was not significantly different for benign and intermediate subtypes, as defined by the highest matching score (5-year and 10-year RFS: 87% and 75% vs 80% and 69%, respectively; P = .96). Similarly, no differences in PFS were observed when limiting analysis to tumors with high confidence scores of ≥0.85 (P = .56) or ≥0.9 (P = .68). Overall, methylation subtype was not significantly associated with RFS within this series.

RNA-Sequencing Analysis
RNA sequencing (RNA-seq) was performed in a subset of 19 cases with available frozen tumor tissue, including 5 tumors (26%) that recurred. Unsupervised analysis (hierarchical clustering and UMAP analysis) did not separate tumors by recurrence or NF2 mutation status (Supplementary Figure S2). A set of 203 genes was identified as differentially expressed in tumors which subsequently recurred (defined by adjusted P-value < .05), which were not found to show clear enrichment for pathways known to be relevant to tumor recurrence/progression (Supplementary Table S3; Figure S3). While multiple independent studies have identified FOXM1 upregulation in aggressive and proliferative meningiomas, 3,30-32 no significant difference in expression of FOXM1 or its gene targets was identified in recurrent vs. non-recurrent tumors in our cohort. None of the genes within a recently published 36-gene meningioma prognostic signature 33 were differentially expressed within our cohort. Similarly, neither PTTG1, LEPR, nor a set of 8 additional genes previously identified as prognostically significant in meningiomas were differentially expressed. 6 These negative findings likely reflect the limited number of cases available for analysis but also suggest that gene expression may not be a strong predictor of recurrence in atypical meningioma.

Association of H3K27me3 Expression With RFS
Loss of H3K27 trimethylation has been shown to correlate with aggressive behavior in meningiomas and can be evaluated using H3K27me3 IHC. [34][35][36] Across our cohort, H3K27me3 was evaluated in 62 tumors, with one case excluded due to lack of expression in endothelial nuclei serving as internal controls. Four tumors (7%) showed unequivocal loss of expression of H3K27me3 in tumors cells with strong internal control staining of vessels (Supplementary Figure S4A), and 36 tumors (58%) showed unequivocal retained expression across tumor cells (Supplementary Figure S4B). In 22 tumors (35%), H3K27me3 staining was heterogeneous and difficult to interpret, with loss observed in subsets of tumor cells in areas with strong internal control staining (Supplementary Figure S4C and D). Although the 4 tumors with unequivocal loss of H3k27me3 expression had a high recurrence rate (5-year and 10-year RFS of 38%), the small number of tumors with this finding was insufficient for further statistical analysis. There was no significant difference in RFS between tumors with retained or partial loss of H3K27me3 expression (Table 2). Overall, loss of H3K27me3 expression is rare in atypical meningioma.
Calculating the c-index for each predictor, loss -10q (c = 0.72) and -1p (c = 0.70) showed stronger concordance with recurrence risk than the integrated models of either  Table 3).

Discussion
Recent studies have significantly expanded our understanding of the molecular drivers of meningioma [37][38][39] and identified genomic and epigenic features associated with histologic grade, recurrence, and progression. 4,10,12,13,35,40 However, molecular risk-stratification of meningiomas has yet to be fully incorporated into clinical practice. The 2021 WHO Classification of CNS Tumors has integrated select Vaubel et al.: Genomic markers of atypical meningioma recurrence Neuro-Oncology Advances molecular parameters into the grading of meningiomas, with TERT promoter mutation or CDKN2A/B homozygous deletion now sufficient for a CNS WHO grade 3 (anaplastic) designation. 2 These alterations are present in only a small number of meningiomas and, therefore, the vast majority of meningiomas are graded solely by histology. Recent comprehensive studies continue to confirm the significance of histologic grading. 10,12,13 However, given the variable clinical course, there remains a lack of consensus on optimal treatment of gross-totally resected atypical (WHO grade 2) meningiomas, which is currently being evaluated in an ongoing phase III clinical trial (NCT03180268/ NRG-BN003). This group of patients is most likely to benefit significantly from improved prediction of recurrence risk.
In this study, we performed detailed molecular characterization, including targeted NGS, chromosomal microarray, RNA-seq, and DNA methylation profiling, across a cohort of patients who underwent GTR, confirmed by postoperative imaging studies, of a histologically grade 2 meningioma and correlated molecular alterations with long-term clinical outcomes (median follow-up: 10 years). Within our series, the presence of specific CNVs was the strongest predictor of RFS ( Table 2). Loss of 1p was frequent (n = 30; 48%) and significantly associated with decreased RFS ( Figure 2B). Chromosome 1p loss has a well-established correlation with histologic grade 41,42 and multiple recent studies have shown a strong association with PFS across all histologic grades of meningioma. 12,13 Losses -4p, -7p,  13 . RFS, recurrence-free survival.
-10q, and -18 were also found to be strongly associated with decreased RFS in our series (Table 2; Figure 2C-F), and these CNVs have been identified to be predicative in prior studies, although somewhat less consistently than -1p. 12,13 Losses of 6q and 14 also occur in higher-grade meningiomas and have previously been associated with unfavorable PFS in some series. 12,13 Although frequent in our cohort, -6q and -14 were not associated with RFS (Table  2). Overall, our results confirm a strong association of specific CNVs with meningioma recurrence risk.
Within our series, NF2 was the most frequent genomic driver mutation (n = 31; 51%), and a smaller subset of tumors harbored alterations of SMARCB1 (n = 5, most in conjunction with NF2), AKT1 (n = 2), SMO (n = 1), and SUFU (n = 1). At the time of sequencing, our clinically validated targeted NGS panel did not include several well-known meningioma driver genes (TRAF7, KLF4, or POLR2A), which is a limitation to this study. TRAF7, KLF4, AKT1, and SMO alterations are mutually exclusive with NF2 mutation, are typically found in meningiomas localized to the skull base, and are associated with a more favorable prognosis. 4,43 In contrast, NF2 mutations are enriched in meningiomas of the convexities/falx and in higher-grade tumors. While we found NF2-mutant tumors to show a trend toward decreased RFS (Table 2; Figure  2A), this did not reach statistical significance (P = .08), which likely reflects the overall size and limited number of recurrences in our series. None of the 5 meningiomas harboring SMARCB1 mutation recurred, which is in keeping with the limited published outcome data available in prior studies. 4 Only a single tumor harbored a TERT promoter mutation, consistent with the frequency previously reported in atypical meningiomas. 44 These findings suggest that CNVs are stronger predictors of RFS than underlying genomic driver mutations.
Recent large-scale efforts have focused on defining meningioma subgroups based on DNA methylation profiling, with multiple independent studies describing groups with favorable, intermediate, and poor prognoses. 8-10 A comprehensive methylation-based classification system has been developed by the DKFZ Heidelberg group based on an analysis of 497 meningiomas, which can be readily applied to individual samples using the available online classification tool. 16 Across meningiomas of all histologic grades, methylation-based classification showed stronger association with PFS than histology alone. 13 Applying the DKFZ classifier to our cohort, approximately 50% of tumors were classified as benign and 50% as intermediate meningioma methylation subtypes, while no cases classified as malignant (Supplementary Table S2). Somewhat surprisingly, there was no significant difference in RFS for tumors classified as benign or intermediate subtypes, whether defined by the highest score or limiting to tumors above a threshold score of ≥0.85 (Table 2; Figure 3A and B). This may reflect, in part, the size of our series. However, it likely also reflects the well-established impact of extent of resection on overall meningioma recurrence risk, which may be abrogated in our cohort as all patients underwent radiologically confirmed GTR. Within our series, RFS of patients with an intermediate methylation subtype meningioma defined by the highest score was high relative to previously published series which included patients with both gross-and subtotal resections (5-year and 10-year RFS: 80% and 69% vs ~50% and ~30%, respectively). 10,13 Therefore, methylation class may be a stronger predictor of RFS following subtotal (vs gross total) resection, which cannot be evaluated in our cohort. It is also important to note that a significant proportion (>30%) of meningiomas within our study as well as the previously published series 10,13 were not definitively classified based on score Abbreviations: RFS, recurrence-free survival; CI, confidence interval; CNS, central nervous system; WHO, World Health Organization.