Meningioma classification with multi-omics data predicts tumor behavior

Currently, the WHO classification of CNS tumors includes the use of two molecular features, for which sufficient evidence of the relationship with patient outcomes exists, in the grading criteria for meningiomas: TERT promoter mutation, and CDKN2A and/or CDKN2B homozygous deletion [1,2,3]. The presence of either one of these features satisfy criteria for a grade 3 meningioma diagnosis. Otherwise, grading is based on histologic criteria, including mitotic activity and brain invasion. However, this is not optimal for risk-stratification in all cases. For example, an important clinical problem relates to risk stratification of grade 2 meningiomas for which treatment decisions depend on extent of surgical resection [4], and an unmet need is standardization of recurrence risk and adjuvant therapy in these patients. To meet this need, evidence has been accumulating in support of epigenetic analysis as a valuable tool in meningioma prognostication [5, 6]. In fact, in clinical practice, some pathologists utilize immunohistochemistry for H3K27 trimethylation to assist in assessing tumor aggressiveness, as loss of trimethylation has been associated with increased risk of recurrence [7]; however, epigenetic assessment is not yet incorporated in the WHO guidelines.

Well-known molecular features that correlate with meningioma aggressiveness are specific copy number variations (CNVs) [5]. Increased risk of recurrence is seen with increasing numbers of CNVs, including losses of 1p, 6q, 14, 10 and 22 [8, 9]. Mitotic index and CNVs have been incorporated with clinical factors to create a nomogram for recurrence risk, yielding an “Integrated Grade” [10]. The Integrated Grade reclassified 32% of the meningiomas in the cohort into higher or lower grades compared to the grades assigned to the tumors by the 2016 WHO based solely on histologic criteria. More recently, genome-wide methylation classification has been shown to be a powerful method for determining meningioma behavior [10,11,12,13,14]. The methylation data builds on earlier studies separating meningiomas into more and less aggressive groups based on subsets of hyper- and hypomethylated loci [15,16,17,18].

Multiple -omics technologies have been combined to better understand meningioma behavior and biology, and several studies have revealed similar findings (Fig. 1, details in Table 1). In addition to showing that losses of 1p, 6q, 14, 18 and 22q exist in higher-risk molecular groups [13], cluster-of-cluster assignments of DNA methylation, transcriptomic and proteomic data demonstrated that these diverse types of molecular information all contributed to classifying meningiomas, with the combination revealing four stable molecular groups (immunogenic, benign NF2 wild-type, hypermetabolic, proliferative). The molecular groups were also associated with recurrent oncogenic variants and inactivating mutations in tumor suppressors, and the most aggressive molecular groups demonstrated increased genomic instability. A similar study [11] identified three subgroups of meningiomas (immune-rich, NF2/Merlin-intact, hypermitotic), again with distinct clinical outcomes and biology that correlated with response to therapies. CDKN2A/2B homozygous deletion, indicative of grade 3 or anaplastic meningioma, was detected with increasing frequency in the hypermitotic (aggressive) meningiomas [11]. Another study [19] found two predominant methylation clusters, one with dismal outcomes (8/11 patients with poor outcomes), and the other with better outcomes. A larger and more comprehensive study used a 6-class methylation classifier scheme [20] and proposed an integrative model of histologic grade, specific CNVs, and methylation class to create a meningioma grading system [12]. By their model, a subset of WHO grade 1 meningiomas are reclassified as “intermediate” grade, subsets of WHO grade 2 meningiomas are reclassified as “low” and “high” grade, respectively, and rare WHO grade 3 cases are reclassified as “low” or “intermediate” grade. These discrepancies presumably reflect the inadequacies of current WHO grading, with the integrated model classifications better reflecting prognosis, and therefore better guiding patient care, although use of the system and resulting outcomes data are necessary to verify its effectiveness. An advantage of the proposed integrated model is that for most cases, meningiomas could be fully characterized into a risk group solely with the use of methylation profiling, suggesting feasibility in the clinic.

Comparison of classifications made by the various algorithms on a common cohort may be helpful to harmonize these multi-omic classification systems and delineate precise recommendations for clinical implementation. Some work has been performed on this front, with encouraging results [21].

Fig. 1
figure 1

Multiple molecular studies of meningiomas have divided tumors into similar groups, demonstrating that characterizing these tumors by methylation profile, CNVs and gene variants is a replicable method of determining meningioma type, indicative of clinical outcome. Four studies are indicated at the top of the diagram, with the meningioma groups found by the studies depicted vertically from best to worst outcomes. The similar molecular features of the groups are highlighted across studies, with gene mutations in light grey, and CNVs in dark grey

Full size image

The correlation of mutational profile with both tumor location and prognostic methylation status deserves comment. Non-NF2-mutated tumors in general are enriched in the clinically less aggressive methylation classes. Specifically, mutations in NF2 tend to be mutually exclusive with mutations in TRAF7, AKT1, KLF4 or SMO. Tumors with mutations in TRAF7/AKT1 and SMO are preferentially associated with anterior fossa, median middle fossa, or anterior calvarial locations. Tumors with TRAF7 and KLF4 mutations tend to be located in anterior and medial fossa and may show secretory meningioma histopathology [22]. PIK3CA variants are also seen in lower risk meningiomas, but may be present in rare higher histologic grade meningiomas that are clinically aggressive [23,24,25,26]. Mutations in POLR2A indicate a lower risk subset of genomically stable meningioma, with low grade meningothelial histology, and frequent tuberculum sellae origin [27]. Further, NF2-altered meningiomas may have low- or high-risk behavior [11,12,13, 20]. Overall, while meningiomas can be classified based on genetic mutations to an extent [28], clinical utility of sequencing analysis may be best interpreted in the context of additional -omics technologies, including methylation profiling and DNA copy number aberrations.

Three histologic classes of meningioma deserve mention based on specific correlations of histopathology with genetics. Clear cell meningiomas have intermediate risk (equivalent to WHO grade 2) [29]. In fact, as studied separately, clear cell meningiomas fall in a distinct methylation class, perhaps due to the fact that they almost always have mutations in SMARCE1 and/or loss of the encoded protein, which is a subunit of the SWI/SNF chromatin remodeling complex [29,30,31]. Rhabdoid meningiomas [32, 33] are currently considered grade 3 tumors, regardless of the presence or absence of high-grade histologic features [34]. However, not all meningiomas with rhabdoid features have the equivalent of a grade 3 outcome [35]; specifically those tumors with loss of BAP1 tend to show worse outcomes [36]. Finally, papillary meningiomas, a rare type of grade 3 meningioma, show frequent inactivation (truncating mutation or homozygous deletion) of the PBAF complex gene PBRM1 [37].

Table 1 Comparison of multiple studies that have employed methylation profiling and/or CNV analysis to characterize meningiomas
Full size table

Transcriptomics data divides meningiomas into biologic groups and supports CNV- and methylation-based meningioma classification

Meningioma gene expression data has been generated and analyzed to determine molecularly different groups and their clinical correlations. One study identified three groups via transcriptomic analysis, with distinct outcomes [38]. A second study used consensus clustering of gene expression data across all grades to identify four subtypes of meningioma, which also had different DNA methylation patterns [40]. Despite the differences in approach, their groups are again reminiscent of the divisions found in other studies, demonstrating both the strengths and limitations of histologic grading. Mutations in genes such as TRAF7, AKT1 and KLF4 were seen in “subtype 1” meningioma, NF2 mutations with chromosome 22q loss only detected in “subtype 2,” and almost all meningiomas containing chromosome 1p loss with NF2 mutation or chromosome 22q loss falling in “subtype 3.” A scoring system based on up- and down-regulated genes predicted recurrence-free survival, but not overall survival in validation cohorts [40].

Additional approaches, employed by several groups, have focused on outcomes, and characterized the difference between poor and better outcomes within their cohort. For example, Schmidt et al. determined gene expression profiles defining aggressive versus less aggressive meningiomas, which revealed low PTTG1 and high LEPR expression associated with aggressive behavior, both of which were confirmed by immunohistochemical staining [41]. Additional studies have confirmed the potential utility of gene expression profiling to identify prognostic classes [42], and a gene expression profile to predict recurrence from a cohort including all histologic grades of meningiomas has been proposed [43]. In addition, transcriptomic analyses were performed in the multi-omics/methylation studies described above, and the results support the proteomics and methylation profiling data [11, 13, 21]. For example, meningiomas in proposed hypermitotic groups were enriched for cell cycle regulation pathways and proliferation-associated transcription factor networks and protein complexes [11, 13]. Meningiomas in the immunogenic group had enrichment of immune regulation and signaling pathways [13], and immune-enriched meningiomas were enriched for meningeal lymphatic genes [11]. Comparing NF2-mutant meningiomas with the methylation class of clear cell meningiomas demonstrated decreased NF2 expression in the former, and relative decreased SMARCE1 and increased EZH2 expression in the latter [29]. While gene expression profiling is less directly translatable than methylation profiling for clinical diagnostics, it holds promise in the potential identification of markers that could be developed into immunohistochemical tests for clinical use.

Radiation-induced meningiomas

Analyses have been undertaken to understand the differences and similarities between sporadic meningiomas and radiation-induced meningiomas (RIMs), generally post therapeutic radiation for a tumor in childhood [44]. Over 20 years ago, studies of 7 cases [45] and of 25 cases [46] found CNV patterns that characterized these tumors.

More recent studies have revealed that in general, RIMs are more genomically unstable than sporadic meningiomas [47]. However, many molecular similarities are seen between RIMs and sporadic meningiomas. Loss of chromosome 1p, known to be associated with higher-grade meningiomas, is seen in the vast majority of RIMs, in line with the 90% seen by Sahm et al. [48], with, in fact, dual loss of 1p and 22q present in some tumors. RIMs also demonstrate NF2 changes, with several types of alterations seen [48]. One study revealed NF2 inactivation due to structural rearrangements, with 12 of 31 RIMs demonstrated NF2 intronic rearrangements [49].

Molecular characterization of pediatric meningiomas highlights differences with adult meningiomas

Pediatric meningiomas are relatively uncommon [50], and WHO grading has historically been less predictive of clinical outcomes than for adult meningiomas, with an increased proportion showing higher grade by WHO criteria [51,52,53]. In contrast to adults, higher proportions of tumors in children are seen in male versus female patients, and more have a spinal or intraventricular location compared to adult meningiomas [52,53,54,55,56]. These differences have spurred investigation into their molecular features to better understand pediatric meningiomas.

Many pediatric meningiomas are seen in neurofibromatosis type 2 syndrome, and this has been understood for more than 20 years [52, 54, 56, 57]. One cohort of pediatric meningiomas found that close to half of their cases had deleterious variants in NF2, almost always with concomitant loss of chromosome 22 [58]. Rare cases with TERT promoter, SMARCB1, BRAF, FUBP1, SMAD2 or GATA3 pathogenic or likely pathogenic variants were identified, but no mutations were seen in TRAF7, SMO, KLF4, AKT1 and PIK3CA, in contrast to adult meningiomas (Fig. 2). Of note, the TERT promoter mutation was found in a histologic grade 3 case. Regarding CNVs, aside from loss of chromosome 22 (68% of cases), chromosome 1 partial losses were seen frequently (29% of cases), and partial losses in 9, 10 and 14 were occasionally seen [57, 58]. Given the differences between adult and pediatric meningiomas, and the high number of pediatric tumors considered grade 2, it has been suggested than alternative grading schemes may provide a better indicator of recurrence-free survival for pediatric meningioma [58].

Fig. 2
figure 2

Comparison of pediatric, spinal and adult intracranial meningiomas highlights similarities and differences among the tumor types. Except for spinal meningioma methylation group 1 clustering with the adult intracranial meningioma MC benign group with TRAKL genotype mutations (TRAF7, AKT1 and KLF4), and proximity of the few spinal meningiomas with increased CNVs to the malignant methylation group of adult intracranial meningiomas, the methylation groups are different for pediatric, spinal and adult intracranial meningiomas, despite overlap in genotypes, such as similarities across all compartments with NF loss-of-function (NF2 LOF) alterations

Full size image

Adding to this body of work, another study found that their pediatric cohort also predominantly showed NF2 alterations [55]. The most frequent CNV was chromosome 22 loss; losses of chromosomes 1, 18 and 14 were also found, to a lesser degree. Importantly, DNA methylation profiling suggests that pediatric meningiomas are epigenetically distinct from adult meningiomas. Despite the similarities in NF2 molecular changes, a majority of the pediatric meningiomas cluster separately from adult meningiomas and three methylation groups have been suggested. The first group consisted almost entirely of clear cell meningiomas with SMARCE1 mutations, with frequent losses on chromosomes 22 and 19. Whether these clear cell meningiomas cluster with the adult clear cell meningiomas described above [29], and whether the SMARCE1 variants were germline or somatic in these specific cases remain open questions. The remaining meningiomas could be divided into two subgroups. The first subgroup (“NF2-driven”) consisted mainly of atypical meningiomas with loss of chromosome 22, and most patients carried the clinical diagnosis of NF2. As mentioned, although these young patients carried NF2 alterations, their tumors clustered separately from adult NF2-altered meningiomas. The balance of the tumors clustering in the final subgroup were a mixture, including rhabdoid meningiomas, and tumors with loss of chromosome 11, but not including NF2 patients [55]. The group also performed DNA sequencing, and found, similar to Toland et al., that the mutations seen in adult meningiomas in TRAF7, SMO, KLF4, AKT1 and PIK3CA, and even the TERT promoter (unlike Toland et al.), were not found in their pediatric cohort. Aside from the NF2 alterations and the SMARCE1 changes, only variants of uncertain significance in a few genes were seen. Conclusions regarding any prognostic significance of the molecular findings were not made. An additional and important difference between pediatric and adult meningiomas is the presence to YAP1 fusions in pediatric tumors [59], where 9 out of 102 cases of pediatric meningioma were shown to have YAP1 fusions. The most frequent fusion partner was MAML2, although other partners (LMO1, PYGO1) were also identified. A second study extended these findings, identifying YAP1-FAM118B fusions in 2 cases [60]. A third study [61] identified an additional case with a YAP1-MAML2 fusion and showed on the CNV plot that such a fusion can be suggested by copy number breakpoints on chromosome 11. It has been speculated that fusions involving YAP1 may be a phenocopy to NF2 inactivation, since NF2 alterations have not been identified in the fusion-positive cases.

Similarities and differences between spinal and intracranial meningiomas

Recently, spinal meningiomas have been studied separately, as they have not been well-represented in prior cohorts. Hua et al. performed targeted sequencing on WHO grade 1 spinal meningiomas and found two mutation types [62]. One group with AKT1 p.E17K mutations were slightly older (median age 71 years) and had tumors most often occurring in a ventral/ventrolateral cervical location. The identification of the AKT1 mutation in spinal meningiomas was in line with a prior molecular study [63]. AKT1-mutant tumors in the spine in one study appeared to be genetically distinct from intracranial AKT1-mutant meningiomas, which frequently harbor co-occurring mutations in TRAF7 [64], a combination only exceptionally seen in the spinal meningiomas. In that study, a second group (median age 65.5 years) frequently harbored NF2 mutations, were predominantly females, and more commonly showed thoracic and dorsal/dorsolateral tumor location [62].

Spinal meningiomas have also been studied by DNA methylation profiling [65]. T-distributed Stochastic Neighbor Embedding demonstrated that a majority of the spinal meningiomas clustered in two groups, separate from intracranial meningiomas; in addition, several tumors were found to be separate from the two clusters, close to the intracranial meningiomas [65]. The two clusters were assessed using the framework of six tumor types previously generated for intracranial meningiomas [20]. The mutational profiles of the spinal meningiomas were consistent with the findings of Hua et al. [62]. A majority of tumors in cluster 1 matched to the less aggressive intracranial methylation class whose tumors had mutations in AKT1, SMO, KLF4 and TRAF7. Many tumors in cluster 1 demonstrated AKT1 mutations, but no other mutations seen in intracranial meningiomas. Although only a few tumors in cluster 2 matched to a defined class, a majority showed NF2 mutations and hemizygous chromosome 22q loss. In contrast, cluster 1 included only one NF2-mutated tumor and three tumors with 22q loss [65]. The several spinal meningiomas that did not cluster with the other cases had an increased number of CNVs. Although clear cell meningiomas have been found not uncommonly in the spine [29, 66], only one case in this cohort was a clear cell meningioma. Whether this case would cluster with the intracranial clear cell meningiomas [29] is unclear. The clinical relevance of the different molecular groups in studies of spinal meningiomas requires further study; thus far, outcome data documents an occasional recurrence, but systemic patient outcomes remain to be discerned [62, 65].

Conclusions

Although meningiomas are frequently considered “benign,” the morbidity and the therapeutic challenges encountered in some cases bring this designation into question. As described here, multiple genomic and epigenomic studies have established the existence of distinct biologic groups of meningiomas, which have clinical correlates. Several of these studies have established multi-omic classification schemes that yield robust and coherent groups; harmonization of these schemes may serve to further clarify optimal biological grouping strategies. Improvements in this realm are particularly important for grade 2 meningiomas, as patients with these tumors display a wide variety of clinical courses and circumstances in which the decision between adjuvant therapy versus surveillance can be difficult. Biomarker assessment is also important for the identification of the subset of histologic grade 1 meningiomas that display poor-prognostic markers, for example those that show an unfavorable methylation class, specific CNVs (e.g. loss of 1p, 6q or 14q) or those that have TERT promoter mutations. From a clinical perspective, grading and marker assessment needs to be performed in the context of the factors such as extent of surgical resection, an important predictor of recurrence.

Patient and provider access to the DNA methylation, SNV and CNV profiling necessary for molecular classification for meningiomas can be a challenge. However, access to molecular testing is increasing, with more institutions developing the capacity and infrastructure for molecular testing. In addition, creation of central reference labs and/or utilization of existing reference would allow increased access, although test re-imbursement issues remain a challenge. Data-driven studies that include molecular analyses and reviews of such studies are essential to promote changes in diagnostic practices, which require time for development.

Finally, the molecular characterization of meningiomas may open the door to precision therapy approaches, with the hope of improved patient outcomes. Treatment development and clinical trials are underway to attempt to attack the various molecular targets that have been found in meningiomas. For example, treatment of an AKT1-mutated meningioma by an AKT1 inhibitor has demonstrated long-term control of disease, a dual TORC1-mTORC2 inhibitor is in clinical trials for treating meningiomas with loss of Merlin function in NF2-dependent meningiomas, and a phase II clinical trial with a SMO receptor antagonist is in process for patients with SMO-mutated meningiomas [65]. Multi-omic studies have begun to unravel the biological characteristics of meningiomas and further work in this area will likely yield insights into new therapeutic strategies, which are urgently needed.

The molecular underpinnings of meningioma are only just beginning to be understood. The current WHO classification has begun to incorporate molecular characteristics of meningiomas into their diagnostic criteria and grading. Given the clinical implications of the multi-omics classification schemes, the combination of mutational profiles with DNA copy number changes and methylation subclasses may serve, along with traditional histopathology, as components of a meningioma grading or classification scheme in the future. However, further studies, especially prospective clinical trials, are needed to investigate the value and use of mutational profiles, CNV patterns or methylation classes as relevant clinical markers in meningioma.