Targeted gene expression profiling predicts meningioma outcomes and radiotherapy responses

Background Surgery is the mainstay of treatment for meningioma, the most common primary intracranial tumor, but improvements in meningioma risk stratification are needed and current indications for postoperative radiotherapy are controversial. Recent studies have proposed prognostic meningioma classification systems using DNA methylation profiling, copy number variants, DNA sequencing, RNA sequencing, histology, or integrated models based on multiple combined features. Targeted gene expression profiling has generated robust biomarkers integrating multiple molecular features for other cancers, but is understudied for meningiomas. Methods Targeted gene expression profiling was performed on 173 meningiomas and an optimized gene expression biomarker (34 genes) and risk score (0 to 1) was developed to predict clinical outcomes. Clinical and analytical validation was performed on independent meningiomas from 12 institutions across 3 continents (N = 1856), including 103 meningiomas from a prospective clinical trial. Gene expression biomarker performance was compared to 9 other classification systems. Results The gene expression biomarker improved discrimination of postoperative meningioma outcomes compared to all other classification systems tested in the independent clinical validation cohort for local recurrence (5-year area under the curve [AUC] 0.81) and overall survival (5-year AUC 0.80). The increase in area under the curve compared to the current standard of care, World Health Organization 2021 grade, was 0.11 for local recurrence (95% confidence interval [CI] 0.07–0.17, P < 0.001). The gene expression biomarker identified meningiomas benefiting from postoperative radiotherapy (hazard ratio 0.54, 95% CI 0.37–0.78, P = 0.0001) and re-classified up to 52.0% meningiomas compared to conventional clinical criteria, suggesting postoperative management could be refined for 29.8% of patients. Conclusions A targeted gene expression biomarker improves discrimination of meningioma outcomes compared to recent classification systems and predicts postoperative radiotherapy responses.


Introduction
Page 6/26 Meningiomas comprise 39.7% of primary intracranial tumors and are the only brain tumors that are more common in women, Black, and elderly patients, who are underrepresented in brain tumor clinical trials 1,2 . Meningioma treatments are largely restricted to surgery and radiotherapy, and systemic therapies remain ineffective or experimental 3,4 . Historically, the World Health Organization (WHO) has graded meningiomas according to histological features such as mitotic count 5 . Most WHO grade 1 meningiomas can be effectively treated with surgery or radiotherapy, but many WHO grade 2 or grade 3 meningiomas (which account for 20-30% of cases 1 ) are resistant to treatment and cause signi cant neurological morbidity and mortality 3 . Moreover, some WHO grade 1 meningiomas develop recurrences that cannot be predicted from histological features, and some WHO grade 2 or grade 3 meningiomas are unexpectedly well controlled with surgery and radiotherapy. In recognition of the controversies surrounding meningioma risk strati cation and treatment, the NRG BN-003 and EORTC 1308 Phase III clinical trials randomize patients with primary WHO grade 2 meningiomas to postoperative surveillance or postoperative radiotherapy after gross total resection 6 . The only multicenter prospective studies of meningioma radiotherapy that have reported data are RTOG 0539 and EORTC 22042, and these Phase II clinical trials provide safety and non-randomized outcome data based on clinical criteria that do not predict radiotherapy responses in many retrospective series [7][8][9][10] . Thus, there are unmet needs for improved risk strati cation and prediction of postoperative radiotherapy responses for patients with meningiomas, the most common primary intracranial tumors.
It is unknown which of these diverse classi cation system(s) may optimize risk strati cation or predict postoperative radiotherapy responses for patients with meningiomas.
Knowledge of the biological pathways underlying breast, prostate, and other cancers has generated robust targeted gene expression biomarkers that are recommended for risk strati cation and prediction of treatment response by the National Comprehensive Cancer Network (NCCN) [25][26][27][28][29][30] . A small pilot study suggested that targeted gene expression pro ling may be useful for meningioma risk strati cation 31 , but an optimized gene expression biomarker, as well as the analytical validity, clinical validity, generalizability, and potential impact of this approach on postoperative meningioma management were unknown.
Here we use known biological pathways underlying meningiomas from bioinformatic studies 11,12,21,22,31,[13][14][15][16][17][18][19][20] to develop a 34-gene expression biomarker predicting clinical outcomes in a single-institution discovery cohort. We perform clinical and analytical validation of the gene expression biomarker using independent meningiomas from a large multicenter retrospective cohort representing North America, Asia, and Europe, and compare the performance of the gene expression biomarker across contemporary meningioma classi cation systems and clinical contexts. We provide investigator-blinded, independent validation of the gene expression biomarker using a multicenter prospective cohort of meningiomas from patients enrolled on RTOG 0539 who represent the social and geographical landscape of the United States. In sum, our results reveal the gene expression biomarker provides additional information for meningioma outcomes compared to recent classi cation systems, including prediction of postoperative radiotherapy responses.

Study design
A discovery cohort comprised of 173 retrospective meningiomas with well-annotated clinical follow up data from a single institution was used to identify and optimize a 34-gene expression biomarker and risk score (Figures 1A, S1 and Tables 1, S1-S3). The performance of the gene expression biomarker was validated in 3 cohorts. First, the analytical validity of the gene expression biomarker was tested in a retrospective analytical validation cohort comprised of 1219 meningiomas from 8 international institutions, some of which had sparse or absent clinical follow up data ( Figure 1A and Table S4). Meningiomas from the discovery cohort, which had matched RNA sequencing, were also used for analytical validation of orthogonal approaches for gene expression quanti cation (Figure 1A and  Table S4). Second, the clinical validity and performance of the gene expression biomarker in comparison to other meningioma classi cation systems were tested in an independent retrospective clinical validation cohort comprised of 866 meningiomas with well-annotated clinical follow up data from 6 international institutions ( Figure 1A and Tables 1, S5-S9), some of which were also used for analytical validation (Table S4). There was no overlap among meningiomas used to identify and optimize the gene expression biomarker in the discovery cohort (Table S2) and meningiomas used for clinical validation (Tables S4-S9). Concordance index (c-index), log-rank test, Brier error score, time-dependent area under the receiver operant curve (AUC), delta-AUC, the Kaplan Meier method, multivariate analysis (Tables S10, S11), and propensity matching (Table S12) were used to compare gene expression biomarker performance across contemporary molecular and histological classi cation systems and clinical contexts. Third, a prospectively collected cohort of 103 meningiomas from patients enrolled on RTOG 0539 were used for investigator-blinded, independent clinical validation ( Figure 1A and Tables 1, S13, S14). In total, 4898 genomic assays were performed and analyzed across 1856 unique meningiomas to de ne and compare molecular classi cation systems ( Figure 1B). Details on data collection, tissue and nucleic acid processing, genomic assays, pathology review, imaging review, statistical analyses, and code and data availability are reported in the Supplemental Methods.

Targeted gene expression analysis
Targeted gene expression pro ling was performed using a hybridization and barcode-based panel with internal negative and spike-in positive controls 32 (Supplemental Methods). Positive-control normalized gene counts were standardized by normalization to the geometric mean count of 7 meningioma-speci c housekeeping genes (Table S3). Log 2 transformed gene expression values were used for all subsequent analyses. Meningioma related genes of interest (Table S1) were selected based on prognostic or biological signi cance in the literature 11,12,21,22,31,[13][14][15][16][17][18][19][20] (Supplemental Methods), and feature selection was performed using a Lasso regularized Cox regression model with the c-index of local freedom from recurrence (LFFR) in the discovery cohort as the target endpoint (Table S2). An optimized set of 34 genes was identi ed within 1 standard error of the model achieving maximal c-index ( Figure  S1A and Table S3), resulting in a highly discriminatory set of linearly rescaled risk scores between 0 and 1 ( Figures 1C, S2). To further reduce over-tting and to facilitate re-calibration of the model for data derived from frozen or formalin-xed and para n embedded (FFPE) meningiomas, or for data derived from orthogonal approaches for gene expression quanti cation such as RNA sequencing, bootstrap aggregation was used to train 500 ridge-regression sub-models using normalized and log 2 -transformed gene counts as input and discovery cohort risk scores as target variables 33 .
Gene expression risk score cutoffs for Kaplan Meier analyses were determined using a nested procedure ( Figure 1D). An initial cutoff was determined in the discovery cohort using the maximally selected rank statistic. The subsets above and below this threshold were again split by maximally selected rank statistic. The lowest risk score group was considered low risk (LFFR cutoff £0.3760769, overall survival [OS] cutoff £0.4206913), and the highest risk score group was considered high risk (LFFR cutoff >0.5651741, OS cutoff >0.6453035). The intervening risk score groups were combined as intermediate risk (LFFR cutoff (0.3760769, 0.5651741], OS cutoff (0.4206913, 0.6453035]). All model training, calibration, and cutoff determination was performed in the discovery cohort (N=173).

Reproduction of molecular classi cation systems in validation cohort meningiomas
Assignment of validation cohort meningiomas to DNA methylation groups 22 or DNA methylation subgroups 21 (WCC, AC, CHGL, HNV, STM, DRR), DNA methylation families 24 or integrated score 17 (SLNM, FS), or gene expression types 19 (JCB, ASH, AH, TK, AJP) was performed independently by investigators who developed each of these classi cation systems. Integrated grade 16 was assigned using CNVs derived from DNA methylation pro les and histological features under supervision of investigators who developed this classi cation system (SS, WLB). DNA methylation probe risk scores were estimated by training a Lasso regularized Cox regression model with LFFR as the endpoint in the discovery cohort using β-values of 283 unfavorable CpG loci 23 . The resulting continuous risk score was converted into low, intermediate, and high risk groups using the same nested procedure described for the gene expression risk score above. All meningioma classi cation system assignments were performed by investigators who were blinded to clinical outcomes and other molecular characteristics of the meningiomas included in this study ( Figure 1A).

Gene expression biomarker development and optimization
Targeted gene expression pro ling of 173 meningiomas in the discovery cohort (Tables 1, S1, S2) resulted in a 34-gene expression biomarker and continuous risk score between 0 and 1 that was converted into discrete low, intermediate, and high risk groups for Kaplan Meier analyses (Figs. 1C-E, S1 and Table S3). The gene expression biomarker was well distributed across intracranial meningioma locations and recurring somatic short variants, and was prognostic for LFFR and OS ( Figures S2, S3). The gene expression biomarker model, risk score, and cutoffs were locked and applied without alteration to multicenter retrospective and prospective validation cohorts from 12 institutions (Tables 1, S4-S9).

Gene expression biomarker analytical validation
Analytical validity, including reproducibility over time and across laboratories, frozen and FFPE meningiomas, and different approaches for gene expression quanti cation, was established using the multicenter analytical validation cohort (N = 1219 meningiomas, 8 institutions, Figs. 1A, S4 and Table  S4). Test-retest conditions, different centers, and paired frozen/FFPE meningiomas generated concordant barcode hybridization gene expression risk scores ( Figures S4A, S4B) that were tractable and discriminatory for meningioma outcomes when RNA sequencing or microarray approaches were used to assess the 34-gene signature ( Figures S4C-S4G).
Comparison across meningioma classi cation systems based on molecular 18,19, 21-24 , molecular and histological 16,17 , or WHO criteria 5,11 using pairwise model combinations 34 revealed the gene expression biomarker provided additional prognostic information for LFFR and OS in combination with each of the 9 other systems tested (Figs. 3A, S6A). No other meningioma classi cation system provided additional prognostic information for LFFR in combination with the gene expression biomarker (Figs. 3A, S6B, S6C), and only WHO 2021 grade provided additional prognostic information for OS (Fig. 3A). The gene expression biomarker achieved the lowest Brier error score over time for LFFR across meningioma classi cation systems, and had an error score that was comparable to WHO 2021 grade and integrated grade over time for OS (Fig. 3B). The gene expression biomarker achieved the highest 5-year AUC for LFFR (0.81) and OS (0.80) across meningioma classi cation systems, with a delta-AUC for LFFR of + 0.07 (95% CI 0.02-0.12, P < 0.001) compared to the next best performing system (integrated grade), and a delta-AUC for LFFR of + 0.11 (95% CI 0.07-0.17, P < 0.001) compared to the current standard of care (WHO 2021 grade) (Fig. 3C). To translate these ndings into clinical practice, nomograms were generated for prediction of 5-year LFFR or OS based on meningioma gene expression risk score, setting (primary or recurrent), extent of resection, and WHO grade (Figs. 4, S7).

Gene expression biomarker prediction of radiotherapy responses
To incorporate the gene expression biomarker into a clinical framework consistent with contemporary NCCN and European Association of Neuro-Oncology (EANO) guidelines 4,35 , meningiomas treated with surgical monotherapy in the multicenter retrospective clinical validation cohort were strati ed by extent of resection and gene expression risk score, resulting in a range of clinical subgroups spanning the spectrum of recurrence risk from 5-year LFFR of 96.1% for gene expression low risk meningiomas with GTR, to 9.8% for gene expression high risk meningiomas with STR (Fig. 5A). Based on these combined biomarker/surgical strata, favorable and unfavorable meningiomas were distinguished using (1) gene expression low risk with any resection, or gene expression intermediate risk with GTR (favorable), versus (2) gene expression intermediate risk with STR, or gene expression high risk with any resection (unfavorable) (Fig. 5A).
In clinical practice, meningiomas with unfavorable histological features or STR are often treated with postoperative radiotherapy based on retrospective data 4,6,35 . NRG BN-003 and EORTC 1308 represent important prospective studies of radiotherapy for meningioma, but these trials were initiated before the development of biomarkers for risk strati cation, and they do not incorporate biomarkers potentially elucidating postoperative radiotherapy responses, as de ned by a reduced risk of recurrence. In the multicenter retrospective clinical validation cohort, the gene expression biomarker remained prognostic for primary meningioma outcomes among patients receiving fractionated postoperative radiotherapy ( Figure S8A), and among patients with primary WHO grade 2 meningiomas with GTR who may have been eligible for NRG BN-003 and EORTC 1308 ( Figure S8B). However, in the absence of biomarker strati cation, primary WHO grade 2 meningiomas with GTR did not bene t from postoperative radiotherapy in the multicenter retrospective clinical validation cohort ( Figure S8C). Thus, to determine if the gene expression biomarker could predict meningioma radiotherapy responses, primary WHO grade 2 meningiomas were strati ed based on favorable versus unfavorable biomarker/surgical criteria (Fig. 5A), revealing that unfavorable primary WHO grade 2 meningiomas bene tted from postoperative radiotherapy (HR 0.33, 95% CI 0.14-0.76, P = 0.009) but favorable primary WHO grade 2 meningiomas did not (P = 0.88) (Fig. 5B). Applying the same biomarker/surgical strata across all WHO grades in the multicenter retrospective clinical validation cohort with propensity matching based on gene expression risk score, extent of resection, and WHO grade revealed that unfavorable meningiomas bene tted from postoperative radiotherapy (HR 0.54, 95% CI 0.37-0.78, P = 0.0001) but favorable meningiomas did not (P = 0.42) (Fig. 5C and Table S12). potentially re ne postoperative management, meningiomas in the multicenter retrospective clinical validation cohort were assigned to RTOG 0539 clinical risk groups and compared across assignments to gene expression biomarker risk groups. The gene expression biomarker improved discrimination of meningioma outcomes across clinical groups used for postoperative radiotherapy strati cation in RTOG 0539 ( Figure S8D) and re-classi ed 52.0% (Table S15)  Investigator-blinded, independent validation of the gene expression biomarker was performed using meningiomas and clinical data that were prospectively collected from patients enrolled on RTOG 0539 itself (N = 103, Tables 1, S13). In comparison to clinical risk groups used to allocate patients to postoperative radiotherapy or postoperative surveillance on this study, the gene expression biomarker reclassi ed 39.8% of meningiomas from RTOG 0539 (Fig. 5D, Table S15), including downstaging 30.3% of intermediate clinical risk patients who received postoperative radiotherapy (Fig. 5D, Table S15). The gene expression biomarker was prognostic for progression free survival (PFS) and OS in patients from RTOG 0539 (Figs. 5D and 5E) and was well calibrated with 5-year PFS of 92.0%, 76.5%, and 38.6% for low, intermediate, and high risk groups, respectively. Moreover, the gene expression biomarker remained independently prognostic on multivariate analysis incorporating meningioma setting (primary or recurrent), extent of resection, and WHO grade using data from RTOG 0539 (Table S14).

Discussion
Here we use targeted gene expression pro ling to develop and validate a polygenic biomarker that provides additional information for meningioma outcomes compared to recent classi cation systems, including prediction of postoperative radiotherapy responses. The gene expression biomarker we report is independently prognostic across all clinical, histological, and molecular contexts tested 5 Supervised bioinformatic models incorporating clinical endpoints have re ned risk strati cation for meningioma local recurrence 16,17,24,36 . The gene expression biomarker reported here provides additional prognostic information for local recurrence and overall survival when combined with all unsupervised or supervised meningioma molecular classi cation systems tested. These ndings are concordant with pancancer analyses examining gene expression, CNV, DNA methylation, protein expression, and DNA sequencing data in 10,884 patients, which suggest gene expression encodes the greatest prognostic information across cancer types 28 . In support of these data, targeted gene expression biomarkers and continuous risk scores have proven successful for multiple cancers [25][26][27][37][38][39] , particularly for breast cancer where polygenic biomarkers are standard of care 25,30 . Previous efforts to reduce meningioma molecular classi cation systems to immunohistochemical stains have thus far not been reproducible 40 . More broadly, qualitative or semi-quantitative protein expression is unlikely to capture the quantitative signal of a gene expression continuous risk score, especially when incorporating non-protein coding genes, as is the case for the biomarker we report (Fig. 1E, Table S3).
Current indications for postoperative radiotherapy for patients with meningiomas are controversial, particularly for patients with primary WHO grade 2 meningiomas who are randomized to postoperative surveillance or postoperative radiotherapy on NRG BN-003 and EORTC 1308 after GTR 3,6 . Con icting retrospective series have variably reported a bene t 9, 41-48 or no bene t from radiotherapy in this setting 47,[49][50][51][52][53][54][55][56] , which has fueled debate and inspired these international Phase III clinical trials of radiotherapy for patients with meningiomas. The gene expression biomarker reported here improves risk strati cation for primary WHO grade 2 meningiomas and may identify favorable WHO grade 2 meningiomas where postoperative radiotherapy could be safely omitted in favor of close surveillance. The gene expression biomarker also identi es primary WHO grade 1 meningiomas with elevated risk of recurrence ( Figure S8E) 59 . The data we present using meningiomas from RTOG 0539 demonstrate the gene expression biomarker was prognostic for overall survival both before and after adjusting for WHO grade on multivariate analysis, and that outcomes remained well-calibrated in this prospective, investigator-blinded validation cohort. For patients with meningiomas, prospective trials such as these will be critical to distinguish conventionally higher risk cases that may safely undergo postoperative surveillance ( Figures S8F, S8G), elucidate which biomarker(s) could be used for strati cation ( Figures   S8H, S8I), and determine whether the timing of postoperative radiotherapy or other interventions improves overall survival ( Figure S8J). As clinical trials develop, we do not anticipate targeted gene expression pro ling will obviate longstanding and robust meningioma classi cation systems, such as WHO grade 11 , or more recent classi cation systems that are tractable across multiple brain tumor types, such DNA methylation pro ling which elucidates biological drivers and vulnerabilities to molecular therapy for meningiomas 21,22,60 . Rather, if incorporated alongside other meningioma classi cation systems and clinical factors such as extent of resection that are already in widespread use, the gene expression biomarker reported here may offer additional bene t to patients with the most common primary intracranial tumor 1 , particularly in terms of postoperative radiotherapy response.
This study should be interpreted in the context of its limitations. First, clinical data in the discovery and multicenter validation cohorts were obtained retrospectively, suggesting our results are susceptible to biases inherent to retrospective research. To address this limitation, we provide additional investigatorblinded, independent validation using meningiomas and clinical data that were prospectively collected from patients enrolled on RTOG 0539. Second, pathology and radiology reviews were performed independently at each institution for meningiomas in the retrospective discovery and validation cohorts. Nevertheless, inter-observer concordance for meningioma WHO grade and imaging characteristics are high 61-63 , and any heterogeneity in clinical review across independent cohorts may better represent the heterogeneity intrinsic to routine clinical practice than might be anticipated from central review. To further address this limitation, the meningiomas from RTOG 0539 that were included in this study underwent central pathology and radiology review 7-9, 63 .

Conclusions
Targeted gene expression pro ling of meningiomas identi es, optimizes, and validates a biomarker predicting local recurrence, overall survival, and radiotherapy bene t, re-classifying up to 52.0% of meningiomas compared to conventional clinical criteria and potentially re ning postoperative management for 29.8% of patients.

Discovery
Retrospective clinical The discovery cohort was comprised of frozen meningiomas from a single institution (UCSF, Table S2). The non-overlapping retrospective clinical validation cohort was comprised of frozen (N=572) and FFPE meningiomas (N=294) from 6 institutions: consecutive meningiomas from The University of Hong Kong (Table S5), and nonconsecutive meningiomas from Northwestern University (Table S6), UCSF (Table S7), Baylor College of Medicine (Table S8), Heidelberg University and Medical University of Vienna (Table S9). The non-overlapping prospective clinical validation cohort was comprised of FFPE meningiomas from RTOG 0539 (Table S13) Study design and gene expression biomarker characteristics.
Panel A shows the study design and numbers of meningiomas used for gene expression biomarker development, analytical validation ( Figure S4, A simpli ed color scheme shows genes associated with higher risk in red and genes associated with lower risk in blue in the rst 2 principal components.

Figure 2
The gene expression biomarker improves discrimination of meningioma outcomes.  The gene expression biomarker predicts meningioma radiotherapy responses.
Panel A shows Kaplan Meier curves for local freedom from recurrence (LFFR) for meningiomas in the multicenter retrospective clinical validation cohort that were treated with surgical monotherapy, strati ed by extent of resection and the gene expression risk score.  Table S15) of retrospective validation cohort meningiomas, and 39.8% (N=41 , Table S15) of RTOG 0539 meningiomas. Reclassi ed meningiomas were better strati ed by gene expression risk ( Figure S8D).

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.