Validation and next-generation update of a DNA methylation-based recurrence predictor for meningioma: A multicenter prospective study

Abstract

Background: We previously developed a DNA methylation-based risk predictor for meningioma, which has been used locally in a prospective fashion since its original publication. As a follow-up, we validate this model using a large prospective cohort and introduce a streamlined next-generation predictor compatible with newer methylation arrays.

Methods: Genome-wide methylation profiles were generated with the Illumina EPICArray. The performance of our next-generation predictor was compared with our original model and standard-of-care 2021 WHO grade using time-dependent receiver operating characteristic curves. An nomogram was generated by incorporating our methylation predictor with WHO grade and the extent of resection.

Results: A total of 1347 meningioma cases were utilized in the study, including 469 prospective cases from 3 institutions and an external cohort of 100 WHO grade 2 cases for model validation. Both the original and next-generation models significantly outperform the 2021 WHO grade in predicting early postoperative recurrence. Dichotomizing patients into grade-specific risk subgroups was predictive of outcomes within both WHO grades 1 and 2 tumors (P < .05), whereas all WHO grade 3 tumors were considered high-risk. Multivariable Cox regression demonstrated the benefit of adjuvant radiotherapy (RT) in high-risk cases specifically, reinforcing its informative role in clinical decision-making. Finally, our next-generation predictor contains nearly 10-fold fewer features than the original model, allowing for targeted arrays.

Conclusions: This next-generation DNA methylation-based meningioma outcome predictor significantly outperforms the 2021 WHO grading in predicting time to recurrence. We make this available as a point-and-click tool that will improve prognostication, inform patient selection for RT, and allow for molecularly stratified clinical trials.

Keywords: DNA methylation; meningioma; neuro-oncology; outcome prediction; prognosis.

Figure 1.

Overview of study cohort and model performance. (A) CONSORT diagram of the retrospective model training cohort (N = 778), the full prospective cohort (N = 469), and the external grade 2 validation cohort (N = 100). (B) and (C) PFS outcomes of the prospective cohort (B) and the external grade 2 cohort (C) stratified by WHO grade. (D) and (E) AUC ROC analysis for the next-generation methylation-based risk predictor versus the previously published first-generation predictor, and standard-of-care WHO grade in predicting PFS at 5 years when all models were tested in the full prospective cohort (D) and the external grade 2 cohort (E). Notably, only cases profiled using the 850k array are included in this comparison, since the original methylation predictor is not compatible with the updated EPICv2 array.

Figure 2.

Representative MRI images of 2 sphenoid wing meningiomas in 2 different patients preoperatively, postoperatively following gross total resection in both cases, and at last radiographic follow-up prior to tumor recurrence in Case 1 and interval stability in Case 2. Both cases were clinically graded as WHO grade 2, with mitoses exceeding 4 per 10 high-powered fields, MIB1 of 10-15%, and nearly identical other histopathological features including the presence of hypercellularity, necrosis, sheeting, and prominent nucleoli without brain invasion. The estimated 5-year risk of recurrence for Case 1 was 0.995, whereas for Case 2 the probabilistic risk of recurrence was 0.467 based on its DNA methylation profile alone, agnostic to any clinical features. Notably, when combined with Simpson grade and WHO grade, our molecular nomogram estimates a 4% probability of 5-year recurrence-free survival in Case 1 (therefore a recurrence probability of 96%) and 49% probability of 5-year recurrence free survival in case 2 (therefore a 5-year recurrence probability of 51%), further suggesting that Case 1 is particularly high risk compared with Case 2.

Figure 3.

DNA methylation profiling identifies high-risk meningiomas among cases that would traditionally be considered “low risk.” (A) PFS outcomes among WHO grade 1 meningiomas stratified by extent of resection (EOR). This demonstrates that DNA methylation identifies high-risk cases regardless of EOR, even among completely resected WHO grade 1 tumors which are traditionally considered to be cured. (B) PFS outcomes among completely resected WHO grade 2 cases, a cohort associated with significant clinical equipoise, in both the prospective cohort and external grade 2 cohort.

Figure 4.

Response to adjuvant RT in methylation-defined risk groups. (A-E) Results of stratified multivariable Cox regression in the low-risk (A) and high-risk (E) methylation group before propensity score matching. Love plot demonstrating covariate balance before and after propensity-score matching of baseline clinical covariates in the low-risk (B) and high-risk (F) methylation groups. PFS analysis of adjuvant RT vs observation after propensity score matching (PSM) in the low-risk (C) and high-risk (G) methylation groups. Results of stratified multivariable Cox regression in the low-risk (D) and high-risk (H) methylation group after propensity score matching.