Clinical implications of DNA methylation-based integrated classification of histologically defined grade 2 meningiomas

In this single-institutional retrospective study, we screened histologically defined WHO grade 2 meningiomas treated at our institution between 2007 and 2021. Inclusion criteria comprised patients with available archived formalin-fixed paraffin-embedded (FFPE) tumor tissue and clinical as well as radiographic follow-up data of at least one year after surgical resection. All tumors were reassessed by a board certified neuropathologist to confirm a tumor grade 2 according to the 2021 WHO Classification of Tumors of the Central Nervous System [14]. Cases with primary resections and treatment for tumor recurrences were allowed. Patient, tumor, and treatment characteristics were extracted from medical records and patient files. Variables of interest included patient demographics (age, sex), tumor characteristics (grading, histopathological subtype, location, size), treatment modalities (treatment indication, Simpson grade, radiotherapy), local tumor control, and date of death [15]. Gross total resection was defined as Simpson grades I, II, and III, while subtotal resection included all cases with Simpson grade IV and V [15]. The resection status was determined by the surgical notes and, when available, postoperative imaging. The integrated molecular-morphological risk score was calculated as previously described and incorporates WHO grading, copy number variation (CNV), including 1p, 6q, and 14q losses, and the methylation family [10]. Clinical follow-up was calculated from the day of surgery to the last clinical contact. Radiographic follow-up was defined as the period from surgery to the last available imaging with contrast-enhanced magnetic resonance imaging (MRI). Local control was defined as the absence of tumor recurrence or progression such as tumor volume increase, at the initial tumor site. MRI was assessed by a board certified neuroradiologist to evaluate local control and classify local failures. The local control rate was calculated as the proportion of patients who exhibited no evidence of local tumor recurrence or progression during the follow-up period utilizing the Kaplan–Meier estimator. The progression-free survival (PFS) was calculated from the time of surgery to local tumor progression or death of any cause. Overall survival (OS) was defined as the period between surgical resection and death of any cause. Censoring for PFS and OS occurred on the last day of available clinical follow-up, for local control on the last day of available radiographic follow-up.

DNA methylation analysis was performed using the 850 k EPIC Illumina Infinium Methylation Array (Illumina, USA). DNA was extracted from the FFPE tumor samples using the Maxwell RSC FFPE Plus DNA Purification Kit (kit number AS1720) (Promega, USA). After bisulfite conversion using the EpiTect Fast Bisulfite Conversion Kit (Qiagen, Netherlands), the Infinium HD FFPE DNA Restore Kit was used for DNA restoration. The beadarrays were scanned on the iScan system (Illumina, USA). The unprocessed output data (idat files) from the iScan reader were checked for general quality measures according to the manufacturer. Raw data obtained from the analysis platform were processed using the R package minfi, version 1.4.0 [16]. The rgSets of 850 k EPIC array were merged with a 450 k array dataset comprising 148 previously analyzed meningioma samples as a reference cohort [9]. A combined rgSet in 450 k output format was generated using the set of overlap probes on both array types. The combined dataset was then subjected to quality control measures, including filtering and functional normalization procedures, to ensure data integrity and comparability across samples as previously described [17]. After filtering and normalization, the variance in the data set was estimated by applying standard deviation over the β-values of all remaining probes. The top 25,000 most variable probes were kept for dimensionality reduction using the Rtsne package, version 0.15 [18]. A Pearson distance matrix was used as the input object, theta of 0.5, perplexity of 30, and all other default parameters were used. Visualizations were generated with the ggplot2 package version 3.3.5 [19]. DNA methylation-based classification was done via MolecularNeuropathology.org, using the brain tumor classifier v12.5. The meningioma classifier v2.4 was used for exploration of matching agreement. The highest score was used for classification, no cut-offs were applied. All tumors underwent TERT promoter mutation analysis with Sanger sequencing, followed by an assessment of hotspot mutations (C228/C250). Assessment of homozygous deletion of cyclin-dependent kinase inhibitors 2A/B (CDKN2A/B) was based on CNV profiles generated using the conumee R package [20]. Chromosomal arm deletions were assessed using the CNV profile as previously reported [21]. Visualization of methylation families was done utilizing t-distributed stochastic neighbor embedding (t-SNE). Descriptive statistics, including frequencies, ranges, and medians, were used to summarize patient and tumor characteristics. Additionally, multivariable Cox proportional hazards models were utilized to evaluate the association of relevant clinical variables on the local tumor control, PFS, and OS. Variable selection was done a priori and based on the most relevant risk factors known [10, 15]. The proportional hazards assumptions were tested using global tests based on Schoenfeld residuals and visual assessment of log–log plots. The goodness of fit of the created models was assessed with the concordance index (Harrell’s C). Statistical significance was defined as p-values ≤ 0.05. Statistical analysis was performed using STATA MP 17.0 (StataCorp, USA). Figures were created with STATA MP 17.0 and R (R Foundation for Statistical Computing, Austria). The study was approved by the local institutional review board (EA2/059/21).

A total of 123 patients treated between 2007 and 2021 were screened (Fig. 1). Sixty-nine patients were diagnosed between 2007 (4th WHO CNS tumor classification) and 2016, the remaining 54 between 2016 and 2021 (4th edition update WHO CNS tumors classification) [22, 23]. After reassessment of tumor grading according to the WHO 2021 classification, 23 cases did not fulfill current criteria. The final study set thus comprised 100 cases (Fig. 1). The exclusion of 23 cases was primarily based due to changes of the cutoff of mitotic rate in WHO 2021 (22/23, 95.6%). One tumor was found to have a TERT promoter mutation as well as a homozygous deletion of CDKN2A/B. The median radiographic and clinical follow-up for the final cohort were 54.4 (range: 12.3–155.2) and 59.8 months (range: 12.4–155.2), respectively. The majority of analyzed tumors were atypical meningiomas (94/100, 94.0%) and resected as part of the primary management at first diagnosis (97/100, 97.0%). The median age at surgery was 59.1 years (range: 20.9–86.9). Most tumors were located around the convexity (66/100, 66.0%) and at the skull base (32/100, 32.0%), with only two cases (2.0%) located in the posterior horn of the lateral ventricles. Gross total and subtotal resection were achieved in 78/100 (78.0%) and 14/100 (14.0%) patients, respectively. Fifteen patients of 100 (15/100, 15.0%) received adjuvant radiotherapy. Seven of these patients (7/15, 46.7%) had a subtotal resection resulting in postoperative treatment. The remaining eight treatment decisions (8/15, 53.3%) were based on individual preferences.

A total of 38 recurrences (38.0%) and 17 deaths (17.0%) were observed. The local control rates of the entire cohort after 2-, 4-, and 6-years were 84.3%, 68.5%, and 50.8%, with a median local control time of 77.2 months (95% confidence interval: 50.9—not available). The most common methylation family according to the brain tumor classifier version 12.5 was benign (51/100, 51.0%), followed by intermediate (43/100, 43.0%) and malignant (6/100, 6.0%) (Fig. 2). The patient and tumor characteristics as well as follow-up information are summarized in Table 1. Assessment of CNV revealed a total of 60 1p (60.0%), 30 6q (30.0%), and 35 14q (35.0%) losses (example case in Additional file 1: Fig. S1). Calculation of the integrated molecular-morphological classification led mostly to intermediate risk tumors (54/100, 54.0%), followed by low (31/100, 31.0%) and high risk (15/100, 15.0%) meningiomas (Fig. 3). The number of 1p losses increased from low risk (0/31, 0.0%), intermediate risk (45/54, 83.3%) to high risk tumors (15/15, 100.0%). The local control rates between risk groups varied markedly and are shown in Fig. 4. The median times of local control for the intermediate and high risk groups were 50.9 and 54.7 months, respectively. The median local control time for the low risk group was not reached during the available follow-up period. Comparison of 1p status and integrated risk groups (low vs. intermediate/high) demonstrated nearly identical local control rates among both corresponding subgroups (Additional file 1: Fig. S2).

The multivariable Cox regression analyses confirmed the integrated risk group assignment as the predominant factor influencing the local control (hazard ratio intermediate risk group: 9.91, hazard ratio high risk group: 7.29, p = 0.002, concordance index: 0.73, Table 2). A gross total resection decreased the risk of local tumor progression (hazard ratio: 0.19, p = 0.001, Table 2, Additional file 1: Fig. S3). The use of adjuvant radiotherapy was not formally associated with an improved local control (hazard ratio: 0.40, p = 0.09, Table 2, Additional file 1: Fig. S4).

The PFS rates of the full cohort after 2-, 4-, and 6-years were 83.6%, 68.1%, and 48.5%, with a median PFS time of 65.5 months (95% confidence interval: 54.7–122.5) (Additional file 1: Fig. S5). Comparable to the local control times, the PFS rates notably varied between the integrated risk groups (Additional file 1: Fig. S6). The median PFS time for the intermediate and high risk groups were 57.1 and 54.7 months, respectively. The median PFS for the low risk group was not reached. The integrated risk groups also showed a significant association with PFS (hazard ratio intermediate risk group: 4.38, hazard ratio high risk group: 4.75, p = 0.006, concordance index: 0.72, Table 2). In addition, gross total resection led to a decreased risk of progression (hazard ratio: 0.36, p = 0.01, Table 2). The multivariable analysis for OS revealed a gross total resection to be associated with a decreased risk of death (hazard ratio: 0.11, p = 0.004, concordance index: 0.88, full data not shown), while older age increased the risk (hazard ratio: 1.14, p < 0.001) (Additional file 1: Fig. S7). The proportional hazards assumptions were fulfilled for all variables and investigated endpoints.

Of the 15 patients receiving adjuvant radiotherapy, four (4/15, 26.6%) had a low risk tumor according to the integrated molecular-morphological risk classifier. None of these four cases with radiotherapy suffered from local tumor progression during the available follow-up (Additional file 1: Fig. S8). The 85 patients without radiotherapy comprised 13 high risk tumors (13/85, 15.2%) and 45 intermediate risk cases (45/85, 52.9%) (Additional file 1: Figs. S9 and 10). No clear benefit can be derived from the use of postoperative radiotherapy in intermediate tumors. In the high risk subgroup, patients without adjuvant radiotherapy had a poor local control with a median time of less than three years. Both patients with postoperative treatments, however, remained free of local tumor progression. Assuming at least a moderate local control benefit from adjuvant radiotherapy for intermediate and high risk tumors, a total of 58 out of 85 patients (68.2%) without postoperative radiotherapy would have been potential candidates for radiation therapy. Conversely, 26.6% of patients (4/15) with a low risk group tumor could have been potentially spared immediate adjuvant radiotherapy due to the low risk of local tumor progression, assuming a marginal benefit of radiation therapy in low risk cases (Additional file 1: Fig. S8). In summary, besides the resection status, the information of the integrated molecular-morphological risk classifier could have informed the treatment decision making process in 62 cases (62/100, 62.0%).