Intratumor and informatic heterogeneity influence meningioma molecular classification

CDKN2A inactivation or TERT promoter mutation are new criteria for World Health Organization (WHO) grade 3 meningiomas [9]. These infrequent alterations can occur in cases that do not otherwise qualify for WHO grade 3 based on histologic criteria [5, 16, 20, 25]. Indeed, most recurring short somatic variants in meningiomas are found in tumors with favorable outcomes [27, 28], but DNA methylation profiling [3, 14, 15, 21, 26], RNA sequencing [17, 26], enhancer analysis [19], copy number variants (CNVs) [4, 10], or integrated molecular strategies [13] can identify broad groups of meningiomas with aggressive biology. Meningioma DNA methylation grouping controlled for artifacts from CNVs further reveals biological drivers and therapeutic vulnerabilities across Merlin-intact, Immune-enriched, or Hyper-mitotic meningioma DNA methylation groups [3]. In sum, these studies suggest molecular classification may improve meningioma stratification or identify draggable dependencies.

Regionally-distinct molecular or histologic features exist within individual meningiomas, suggesting intratumor heterogeneity may confound meningioma molecular classification [6, 7, 11, 18, 22, 23]. Stereotactically collected, regionally-distinct meningioma samples showing regional molecular alterations are not associated with differences in WHO histologic grade [11], and 92% of regionally-distinct stereotactic samples classify in concordance with the consensus DNA methylation group of each meningioma [3]. Nevertheless, meningiomas with mismatched intratumor DNA methylation groups are valuable resources for interrogating the robustness of meningioma molecular classification systems. To test this hypothesis, a capture-based targeted DNA sequencing assay interrogating all published recurrent short somatic variants in meningiomas was performed on 10 regionally distinct samples from 4 previously reported meningiomas displaying intra-tumor differences in DNA methylation groups [3, 8, 11]. Clonal NF2 alterations were identified across all stereotactic samples from different DNA methylation groups in 2 WHO grade 1 meningiomas (M8 and M9) and 1 WHO grade 3 meningioma (M7) (Fig. 1a). A TERT promoter mutation was identified in only 1 of 2 regionally distinct samples with equivalent histologic characteristics but different DNA methylation groups in a WHO grade 2 meningioma (M2). A histone methyltransferase KMT2C alteration was identified in 2 of 3 regionally distinct samples with different DNA methylation groups in a WHO grade 1 meningioma (M9). These data suggest regional sampling can influence meningioma molecular classification.

Fig. 1.

Intratumor and informatic heterogeneity influence meningioma molecular classification. a 3D stereotactic meningioma sampling maps reconstructed from preoperative magnetic resonance imaging. b Meningioma sample purity across DNA methylation groups. c Intratumor phylogenies based on clonal ordering of CNVs derived from meningioma DNA methylation profiling. d CNV phylogenies based on meningioma RNA sequencing. e CNVs derived from DNA methylation profiling and allelic frequency derived from targeted DNA sequencing of meningioma M8E. f CNVs derived from DNA methylation profiling of meningioma MSC6. g Uniform manifold approximation and projection of MSC6 single-cell RNA sequencing transcriptomes

Tumor purity of Immune-enriched meningiomas can be reduced compared to meningiomas from other DNA methylation groups, and single-cell RNA sequencing reveals non-meningioma cell types comprise an average 30 ± 8% of cells in meningiomas [3]. To study the impact of cell type heterogeneity on meningioma DNA methylation groups, tumor purity analysis was performed on 26 stereotactic samples from the 4 meningiomas. The PAMES bioinformatic pipeline, which has been validated across 14 tumor types [1], revealed the purity of Immune-enriched meningioma regional samples was lower than Merlin-intact or Hyper-mitotic samples from M2 and M7, but also showed tumor purity across all 3 DNA methylation groups from M8 or M9 regional samples was comparable (Fig. 1b). Thus, reduced tumor purity is sufficient but not necessary to influence meningioma DNA methylation classification.

CNVs are prognostic for meningioma recurrence [4, 10], but CNVs across regionally-distinct meningioma samples are heterogeneous [11]. To determine if intratumor CNV variability influences meningioma molecular classification, CNVs were identified in 14 regional samples from 2 meningiomas using the SeSAMe bioinformatic pipeline to identify CNVs from DNA methylation profiles, or the CaSpER bioinformatic pipeline to identify CNVs from RNA sequencing [24, 30]. Phylogenetic trees were constructed for each tumor based on the CNV distribution across stereotactic samples (Fig. 1d, c) [29]. There were notable differences in CNV architecture depending on whether DNA methylation profiles (Fig. 1c) or RNA sequencing (Fig. 1d) was used to define CNVs. Whole exome and targeted DNA sequencing validate 99.12% of meningioma CNVs identified using DNA methylation profiles and SeSAME [3], but RNA sequencing identified only 40 of 102 CNVs (39%) in the 14 regional samples (Fig. 1d). The allelic frequency provided by targeted DNA sequencing, which enables zygosity assessment, revealed near whole genome haploidization in 5 stereotactic samples (Fig. 1e), but DNA methylation profiling using either SeSAMe or conumee [11] incorrectly identified diploid chromosomes in these samples as being amplified (chromosomes 5, 6, 12, 15, 18, 19, and 20) (Supplementary Fig. 1). Thus, meningioma CNV analysis using DNA methylation profiling or DNA sequencing outperforms RNA sequencing, but DNA sequencing may be required when ß methylation values cannot be normalized across a euploid genome.

CNVs in meningiomas can have diminished amplitude of sequencing reads or ß methylation values (Fig. 1f), suggesting sub-clonal populations of tumor cells may influence meningioma molecular classification. To test this hypothesis, single-cell CNVs were identified using CONICSmat in 18,880 cells from single-cell RNA sequencing of a previously reported meningioma [3, 12]. The distribution of CNVs was partially overlapping across cells in this tumor (Fig. 1g), supporting the hypothesis that sub-clonal cell populations can influence meningioma molecular classification.

In conclusion, intratumor heterogeneity and limitations of current bioinformatic pipelines influence meningioma molecular classification, and how these schemes may change over the course of tumor evolution is incompletely understood. Some somatic variants or CNVs, such as those inactivating NF2, are early, stable events underlying meningioma evolution [11]. We re-analyzed DNA methylation profiles from paired primary/recurrent samples and found 3 of 4 pairs classified as hyper-mitotic meningiomas at presentation and at recurrence [3]. One meningioma classified as Hypermitotic at presentation and Immune-enriched at recurrence, further suggesting reduced tumor purity may influence meningioma DNA methylation groups (Fig. 1b). The stability of TERT promoter mutations, other somatic variants, or other CNVs over time is unknown, but with the increasing availability of prospective, molecularly-stratified trials for meningioma patients [2], samples to answer these questions are on the horizon.

Data availability

DNA methylation profiling, RNA sequencing, single-cell RNA sequencing, or DNA sequencing data for all previously reported meningiomas that were reanalyzed in this study have been deposited in the NCBI Gene Expression Omnibus under accession numbers GSE151067, GSE151921, or GSE183656.