Spatial genomic, biochemical, and cellular mechanisms drive meningioma heterogeneity and evolution

Intratumor heterogeneity underlies cancer evolution and treatment resistance1–5, but targetable mechanisms driving intratumor heterogeneity are poorly understood. Meningiomas are the most common primary intracranial tumors and are resistant to all current medical therapies6,7. High-grade meningiomas cause significant neurological morbidity and mortality and are distinguished from low-grade meningiomas by increased intratumor heterogeneity arising from clonal evolution and divergence8. Here we integrate spatial transcriptomic and spatial protein profiling approaches across high-grade meningiomas to identify genomic, biochemical, and cellular mechanisms linking intratumor heterogeneity to the molecular, temporal, and spatial evolution of cancer. We show divergent intratumor gene and protein expression programs distinguish high-grade meningiomas that are otherwise grouped together by current clinical classification systems. Analyses of matched pairs of primary and recurrent meningiomas reveal spatial expansion of sub-clonal copy number variants underlies treatment resistance. Multiplexed sequential immunofluorescence (seqIF) and spatial deconvolution of meningioma single-cell RNA sequencing show decreased immune infiltration, decreased MAPK signaling, increased PI3K-AKT signaling, and increased cell proliferation drive meningioma recurrence. To translate these findings to clinical practice, we use epigenetic editing and lineage tracing approaches in meningioma organoid models to identify new molecular therapy combinations that target intratumor heterogeneity and block tumor growth. Our results establish a foundation for personalized medical therapy to treat patients with high-grade meningiomas and provide a framework for understanding therapeutic vulnerabilities driving intratumor heterogeneity and tumor evolution.


Experimental design and work ow
To de ne mechanisms underlying meningioma intratumor heterogeneity and evolution, 16 intracranial samples from 10 meningiomas (designated M1-10) that were resected from 9 patients at the University of California San Francisco (UCSF) were assembled for clinical, histological, and molecular analyses ( Fig. 1a and Supplementary Table 1). Preoperative magnetic resonance imaging (MRI) studies and medical records were reviewed to de ne meningioma locations, presentations (primary versus recurrent), treatments, and outcomes. Histological and molecular analyses of all samples were performed using the most recent criteria from the World Health Organization (WHO) Classi cation of Tumors of the Central Nervous System 20 , including targeted next generation DNA sequencing 26 to de ne SSVs and CNVs that are associated with high-grade meningioma classi cation and adverse clinical outcomes 15,16,[27][28][29] (Supplementary Tables 1 and 2). All samples were analyzed using immunohistochemistry (IHC) for cell proliferation (Ki-67), cell cycle regulation (p16), or chromatin markers (H3K27me 3 ), each of which can also be associated with clinical outcomes for patients with meningiomas [30][31][32] . To integrate standard approaches for meningioma classi cation with emerging techniques that de ne biological drivers and therapeutic vulnerabilities in meningiomas, DNA methylation grouping 13 and targeted gene expression pro ling 24 were performed on all samples (Supplementary Tables 1 and 3). These comprehensive clinical, histological, and molecular analyses identi ed meningiomas to study the molecular (M1-3), temporal (M4-7), and spatial evolution (M8-10) of human cancer (Fig. 1b).
Spatial transcriptomic pro ling of 50µm regions from continuous arrays tiled across 6mm cores was performed on all meningiomas using an approach that integrates approximately 10 cells per capture area 33 (Extended Data Fig. 1a). Core selection for each sample was guided by morphological or IHC heterogeneity of whole mount formalin-xed para n-embedded (FFPE) tumor sections. Spatial transcriptomes were aligned with hematoxylin and eosin (H&E) histological images using unique oligonucleotide barcodes corresponding to array positions (Extended Data Fig. 1b). The Harmony bioinformatic pipeline was used for sample integration and batch-correction 34 (Extended Data Fig. 1c), and uniform manifold approximation and projection (UMAP) analysis of 38,718 spatial transcriptomes demonstrated 30 spatial gene expression programs across 16 high-grade meningioma samples (range: 4-15 programs/sample) (Fig. 1c, Table 4). Six spatial gene expression programs that included transcriptomes from all samples were distinguished by enrichment of genes involved in neural development (SIM2, VIT in C1 and C7), angiogenesis (THBS2, HHIP in C3), meningeal homeostasis and neurotransmitter processing (PTGDS, LCNL1 in C5), bone differentiation (MAP1LC3C, ALPL in C9), and differentiation of the neural crest (S100A, S100B in C14), a multipotent embryonic cell population that gives rise to the meninges 35 Region selection for each sample was guided by morphological or IHC heterogeneity of whole mount FFPE tumor sections. Laser microdissection and next generation sequencing was used to quantify binding of 72 antibodies that were conjugated to unique oligonucleotide barcodes from 82 regions (range: 3-12 regions/sample) ( Fig. 1e and Supplementary Table 5). Principal component analysis of spatial protein pro ling data demonstrated divergent biochemical mechanisms within and across highgrade meningiomas (Extended Data Fig. 2d).
Using these clinical, histological, molecular, and spatial data, the study cohort was divided into 3 groups to study genomic, biochemical, and cellular mechanisms underlying intratumor heterogeneity in the context of molecular (Fig. 2), temporal (Fig. 3, 4), and spatial evolution of meningiomas (Fig. 5, 6). Findings were validated using multiplexed seqIF microscopy, spatial deconvolution of meningioma single-cell RNA sequencing, bulk RNA sequencing from 502 meningiomas 13,18 , and CRISPR interference, pharmacology, and live cell imaging in meningioma organoid preclinical models (Fig. 7).
High-grade meningiomas are distinguished by divergent intratumor gene and protein expression programs The WHO de nes meningioma grades according to histological features, such as mitotic count and brain invasion, and rare molecular features such as CDKN2A/B homozygous deletions or hotspot TERT promoter mutations that are su cient for diagnosis of WHO grade 3 meningioma 20 . Most WHO grade 1 meningiomas can be effectively treated with surgery or radiotherapy, but many WHO grade 2 or grade 3 (high-grade) meningiomas, which account for 20-30% of cases 7,9 , are resistant to treatment and cause signi cant neurological morbidity and mortality 6 . Morphological features can in uence meningioma WHO grading, and rhabdoid morphology associated with inactivating BAP1 mutation is also associated with WHO grade 3 meningioma 38 . Thus, current clinical classi cation systems group meningiomas with different driver mutations into the same high-grade group, which may not provide an optimal framework for clinical trials. To determine if high-grade meningiomas were associated with convergent or divergent intratumor gene or protein expression programs, spatial genomic and biochemical mechanisms were studied across meningiomas with BAP1 inactivation (M1), CDKN2A/B homozygous deletion (M2), or TERT promoter mutation (M3) (Extended Data Fig. 3a, b and Supplementary Table 1).
Spatial gene expression programs in M2 and M3 also demonstrated heterogeneous ontologies that correlated with morphological features (Fig. 2f-I and Extended Data Fig. 3b). Differential expression analysis of spatial transcriptomes in M2 revealed a connective tissue and hemorrhagic cluster (COL3A1, COL4A4, HBA1, HBA2 in C2), a brain parenchyma cluster (NNAT, SYN2 in C6), and 4 other clusters comprised of variably cellular tumor that were distinguished by enrichment of in ammatory and immune genes (IRF1, CD55, IL18, LYZ, LY6D) (Fig. 2f, h, I and Extended Data Fig. 3b). C4 was comprised of braininvasive meningioma with enrichment of oncogenes (MN1) and tissue invasion genes (TAC3). C5 from M2 and C3 from M3 showed decreased expression of in ammatory and immune genes but enrichment of MT2A, which is implicated in cell stress, homeostasis, and differentiation 40,41 . Other cell stress genes and DNA damage response genes were enriched in C3 from M3 (HSP1A, NR4A1, ANKRD1), and the 5 other spatial gene expression programs in M3 were distinguished by differential expression of ion transport, cell stress, and immune genes that were not differentially expressed in M1 or M2 (SLC9A3, LTK, DEPP1, HSP1A, NOTCH3, FOS, SERPINE1, MT1X) (Fig. 2i).
Spatial sub-clonal copy number variants, signaling mechanisms, and cell types underlie high-grade meningioma recurrence Surgery is the mainstay of meningioma treatment, but postoperative radiotherapy is recommended to reduce the risk of high-grade meningioma recurrence 9,42 . Nevertheless, local recurrence of high-grade meningioma is common 21 , and recurrence is the leading cause of death in patients with meningiomas that are resistant to standard interventions 43 . Mechanisms underlying meningioma resistance to treatment are poorly understood. To address this limitation in our understanding of meningioma biology, spatial genomic, biochemical, and cellular mechanisms were studied in the context of histological and molecular classi cation systems across matched pairs of primary (M4-7) and recurrent (M4'-7') highgrade meningiomas that were treated with radiotherapy between primary and recurrent resections (Fig. 3a, Extended Data Fig. 4a and Supplementary Table 1).
Histological analysis showed higher WHO grades and increased immunostaining for Ki-67 in paired recurrent versus primary meningiomas ( Fig. 1b and Extended Data Fig. 4a). Bulk molecular approaches demonstrated higher gene expression risk scores, increased CNV burden, and aggressive driver mutations such as TERT promoter mutation or CDKN2A/B homozygous deletion in paired recurrent versus primary meningiomas ( Fig. 1b and Supplementary Table 2, 3). Spatial gene expression programs were divergent in paired primary and recurrent meningiomas despite sample integration and batch-correction with Harmony ( Fig. 3b and Extended Data Fig. 4b-d). Incorporation of CNVs can improve prognostic models for meningioma outcomes 15,16 , but the spatial architecture and evolution of meningioma CNVs over time is incompletely understood. To determine if spatial expansion of sub-clonal CNVs underlies high-grade meningioma recurrence, inferCNV 44,45 was used to deconvolve paired primary and recurrent meningioma spatial transcriptomes (Extended Data Fig. 5a). Spatial projection validated CNVs that were identi ed in the recurrent but not in the primary meningioma from paired samples using targeted next generation DNA sequencing (Fig. 1b, 3c and Supplementary Table 2). Spatial projection also identi ed clonal CNVs from recurrent meningiomas in sub-clonal spatial transcriptomes from paired primary tumors that were below the limit of detection using bulk molecular approaches (Fig. 3c). In support of these data, spatial transcriptomes demonstrated decreased expression of MAPK genes (RAB7, MAPK11, PLCE1) or epigenetic regulators (SMARCA2) that were lost through copy number deletions in paired recurrent versus primary meningiomas (Fig. 3d). Interestingly, an intracranial meningioma overlying the frontal cortex (M8) also demonstrated divergent histological, SSV, CNV, and spatial transcriptomic architecture compared to patient-matched primary (M7) and recurrent (M7') meningiomas overlying the occipital cortex (Fig. 1b To determine if the diverse genomic mechanisms underlying high-grade meningioma recurrence were associated with convergent or divergent biochemical or cellular phenotypes, spatial protein pro ling ( Fig. 4a and Extended Data Fig. 7a-d) was performed alongside multiplexed sequential immuno uorescence (seqIF) to stain and image whole mount sections of primary (M4-7) and recurrent (M4'-7') meningiomas (Fig. 4b, c, Extended Data Fig. 7e and Supplementary Table 7). Principal component analysis of spatial protein pro ling data demonstrated divergent biochemical mechanisms in primary versus recurrent tumors (Extended Data Fig. 7a, b), but inspection of individual proteins revealed conserved trends underlying high-grade meningioma recurrence (Extended Data Fig. 7c). Proteins associated with cell proliferation (Ki-67) and PI3K-AKT signaling (PLCG1) were enriched in recurrent meningiomas, whereas proteins associated with MAPK signaling (pan-Ras), immune signaling (CD45, VISTA, CD14), and PI3K-AKT inhibition (INPP4B) were suppressed in recurrent meningiomas ( Fig. 4a and Extended Data Fig. 7d). In support of these ndings, multiplexed seqIF showed Ki-67 was enriched in recurrent versus primary meningioma cells that were marked by SSTR2A 46 (Fig. 4c). Primary meningiomas were enriched in pan-Ras, INPP4B, macrophages (CD68, CD163) that were concentrated in the perivascular niche (CD31), and VISTA, an inhibitor of T cell activation (Fig. 4c). Meningiomas have poor responses to immune checkpoint inhibitors that target T cells 47,48 , and T cells marked by CD4 or CD8 were rare in either primary or recurrent meningiomas (Extended Data Fig. 7e). To validate these ndings, meningioma cell types were deconvolved from spatial transcriptomes using single-cell RNA sequencing of 57,114 cells from 8 meningioma samples representing all DNA methylation groups 13 . Spatial deconvolution of single-cell types showed CD163 macrophages, differentiated meningioma cells, SSTR2A meningioma cells, and non-cycling G1 phase meningioma cells were decreased in recurrent versus primary meningiomas (Fig. 4d). Cycling G2M phase and S phase meningioma cells were enriched in recurrent versus primary meningiomas (Fig. 4d).
Regionally distinct sub-clonal spatial transcriptomes, signaling mechanisms, and cell types underlie histological heterogeneity in high-grade meningiomas High-grade meningiomas can arise de novo, progress from lower grade meningioma at the time of recurrence 50-52 , or may show predominantly low-grade histology with only focal evidence of high-grade treansformation 53 . Thus, regionally distinct histological or genomic intratumor heterogeneity can in uence meningioma classi cation 8, 19 , but the identity and spatial relationships among mechanisms driving intratumor heterogeneity in high-grade meningiomas are unknown. To address this limitation in our understanding of meningioma biology, spatial genomic and cellular mechanisms were studied across high-grade meningiomas demonstrating regionally distinct intratumor heterogeneity (M9-10) ( Fig. 5a- Histological analyses of M9 revealed a well-demarcated area of increased cellularity, increased immunostaining for Ki-67, and increased mitotic count that was su cient for diagnosis of WHO grade 3 meningioma in 1 of 2 regionally distinct cores (Fig. 5a). Both cores from M9 were otherwise comprised of WHO grade 2 histology, lower immunostaining for Ki-67, Hypermitotic meningioma DNA methylation grouping, and high gene expression risk scores but showed divergent SSVs inactivating epigenetic regulators (ARID1A, ASXL1) and divergent CNVs deleting chromosomes 4 and 14q that were only identi ed in the core with WHO grade 3 histology (Fig. 1b). Histological analyses of M10 revealed WHO grade 3 meningioma with mosaic immunostaining for p16 that inversely correlated with immunostaining for Ki-67 in 2 regionally distinct cores ( Fig. 5b and Extended Data Fig. 8b). Both cores from M10 classi ed in the Hypermitotic meningioma DNA methylation group but showed divergent gene expression risk scores and divergent CNVs amplifying chromosome 1q or deleting chromosomes 4q, 9p, and 10q ( Fig. 1b).
Spatial gene expression programs were analyzed across regionally distinct high-grade meningioma cores after sample integration and batch-correction with Harmony ( Fig. 5c-f). Clusters C3, C6, and C9 in M9 correlated with WHO grade 3 histology (Fig. 5a, e) and differential expression analysis of spatial transcriptomes revealed shared enrichment of embryonic transcription factors (SOX11, ELF3) but divergent expression of meningeal homeostasis (PTGDS in C3) or immune genes (CXCL8 in C6, HLA-DPA1, IGHG1 in C9) in WHO grade 3 regions ( Fig. 5g-i). Clusters C2, C8, and C10 in M9 correlated with WHO grade 2 histology that was immediately adjacent to the WHO grade 3 region and lacked embryonic transcription factor expression but was enriched in meningeal homeostasis (PTGDS in C2 and C10) or immune genes (HLA-DPA1 in C8). Clusters C1, C4, C5, and C7 in M9 correlated with WHO grade 2 histology that was distant from the WHO grade 3 region and was enriched in tissue differentiation (FIBIN in C1 and C5, ACTA2 in C4) and innate immune genes (IFI27, IFIT3 in C7). M10 clusters C4, C5, and C6 correlated with reduced immunostaining for p16 (Fig. 5b, f), and differential expression analysis of spatial transcriptomes revealed shared enrichment of cell signaling and proliferation genes (GPC1, CRABP1) but divergent expression of immune genes in these regions (IGHG1, IGKC, CLEC3B in C6) ( Fig. 5j-l). Cluster C8 correlated with intermediate immunostaining for p16 and demonstrated divergent cell signaling and proliferation genes (MET, EGFL6), supporting the hypothesis that regionally distinct mechanisms activating the cell cycle can exist in individual meningiomas (Fig. 2d, e). The remainder of M10 showed diffusely positive immunostaining for p16 and was enriched in senescence and cell cycle regulation genes (MX2 in C7, CDKN2B in C9 and C10). Multiplexed seqIF showed that Ki-67 was enriched in the WHO grade 3 region of M9 and in the region of M10 with reduced immunostaining for p16 ( Fig. 6ac). Moreover, M9 and M10 showed regionally distinct expression of pan-Ras, INPP4B, CD68, CD163, VISTA, and the pericyte marker CD31. Spatial deconvolution of meningioma single-cell types 13 validated regionally distinct changes in CD163 macrophages, pericytes, endothelia, SSTR2A meningioma cells, extracellular matrix (ECM) remodeling meningioma cells, and G1/G2M/S phase meningioma cells in M9 and M10 (Fig. 6d). Thus, in support of the genomic, biochemical, and cellular phenotypes underlying temporal evolution of high-grade meningiomas (Fig. 3, 4 and Extended Data Fig. 4-7), regionally distinct cell proliferation, cell signaling, and immune mechanisms underlie spatial evolution of high-grade meningiomas.
A preclinical platform for testing personalized medical therapies to overcome intratumor heterogeneity in highgrade meningiomas Sub-clonal evolution underlies tumor recurrence and treatment resistance 1-5 , but preclinical models of intratumor heterogeneity or tumor evolution in response to treatment are scarce. To develop reagents to study high-grade meningioma heterogeneity and evolution in response to treatment, patient-derived highgrade M10G meningioma cells stably expressing CRISPRi machinery (M10G dCas9 − KRAB ) 8, 13 were transduced with sgRNAs suppressing the cell cycle inhibitors CDKN2A (sgCDKN2A) or CDKN2B (sgCDKN2B), the epigenetic regulator ARID1A (sgARID1A), or non-targeted control sgRNAs (sgNTC) (Extended Data Fig. 9a). RNA sequencing of triplicate M10G dCas9 − KRAB cultures with differential expression and ontology analyses revealed CDKN2A/B suppression inhibited developmental and metabolic gene expression programs, whereas ARID1A suppression induced metabolic and mitotic gene expression programs ( Fig. 7a and Supplementary Table 9). These data suggest drivers of high-grade meningioma intratumor heterogeneity, such as CDKN2A/B homozygous deletion or SSVs inactivating epigenetic regulators like ARID1A (Fig. 1b), may be associated with non-overlapping therapeutic vulnerabilities. In support of this hypothesis, preclinical experiments demonstrate meningiomas with loss of cell cycle regulators are susceptible to CDK4/6 inhibitors such as abemaciclib 13 , and meningiomas with loss of epigenetic regulators may be susceptible to histone deacetylase inhibitors such as vorinostat 17 .
To identify pharmacologic strategies inhibiting intratumor heterogeneity in high-grade meningiomas, M10G dCas9 − KRAB cells transduced with sgCDKN2A/B, sgARID1A, or sgNTC were reciprocally labeled with red or green uorescence proteins to track pharmacologic responses and assembled into 3D organoid cocultures for live cell microscopy. Abemaciclib blocked the growth of M10G dCas9 − KRAB cells with CDKN2A/B suppression but did not block the growth of M10G dCas9 − KRAB cells with ARID1A suppression or sgNTCs (Fig. 7b, c). To identify therapeutic vulnerabilities underlying meningiomas with loss of epigenetic regulators, spatial protein pro ling was analyzed across 21 regions with or without SSVs inactivating ARID1A from M10 (Fig. 1e). These data revealed regionally distinct potential vulnerabilities to small molecule inhibitors of the DNA damage response (niraparib), EGFR signaling (erlotinib), MEK/ERK signaling (selumetinib), MET signaling (capmatinib), or PI3K-AKT signaling (copanlisib) (Fig. 7d).
Vorinostat, niraparib, erlotinib, selumetinib, and copanlisib blocked the growth of M10G dCas9 − KRAB cells expressing sgNTC, and selumetinib and copanlisib blocked the growth of cells with ARID1A suppression (Fig. 7e). To determine if combination molecular therapy could overcome intratumor heterogeneity in high-grade meningiomas, 3D organoid co-cultures of M10G dCas9 − KRAB cells expressing sgCDKN2A and sgNTC (Fig. 7f), or sgCDKN2A and sgARID1A (Fig. 7g), were treated with abemaciclib and selumetinib, or abemaciclib and copanlisib. Combination molecular therapy blocked the growth of meningioma cells with loss of CDKN2A and loss of ARID1A in both co-culture conditions and attenuated the growth of meningioma cells expressing sgNTC (Fig. 7h, i). Thus, high-grade meningiomas with loss of cell cycle and/or epigenetic regulators are susceptible to combination molecular therapy blocking CDK4/6, MEK/ERK signaling, and PI3K-AKT signaling.

Discussion
Here we integrate spatial transcriptomics, spatial protein pro ling, multiplexed seqIF, and spatial deconvolution of single-cell RNA sequencing across high-grade meningiomas to identify genomic, biochemical, and cellular mechanisms linking intratumor heterogeneity to the molecular, temporal, and spatial evolution of human cancer. Our results reveal divergent intratumor gene and protein expression programs distinguish high-grade meningiomas that are otherwise grouped together by the World Health Organization Classi cation of Central Nervous System Tumors 20 , one of the systems that is currently used to determine patient eligibility on clinical trials 6 . Analyses of matched pairs of primary and recurrent meningiomas reveal spatial expansion of sub-clonal copy number variants, decreased immune cell in ltration, decreased MAPK signaling, increased PI3K-AKT signaling, and increased cell proliferation underlie treatment resistance and tumor recurrence. We nd regionally distinct high-grade meningioma samples displaying histological and molecular heterogeneity are associated with spatial gene expression programs that correlate with intratumor heterogeneity and cell proliferation. To translate these ndings to clinical practice, we use epigenetic editing and lineage tracing approaches in human meningioma organoid models to identify new combinations of FDA-approved molecular therapies that target intratumor heterogeneity and block meningioma growth. In sum, our results establish a foundation for personalized medical therapy to treat patients with high-grade meningiomas and provide a framework for understanding mechanisms and therapeutic vulnerabilities driving intratumor heterogeneity and tumor evolution.
The human meningiomas in this study that were analyzed using bulk genomic, spatial transcriptomic, spatial protein pro ling, multiplexed seqIF, and single-cell RNA sequencing deconvolution approaches were clinical FFPE samples, as opposed to fresh, frozen, or curated research specimens that are used for many exploratory investigations. Thus, the biological ndings in this study may be generalizable to routine clinical practice. In support of this hypothesis, we show mechanisms underlying meningioma intratumor heterogeneity and evolution from our discovery cohort are conserved across a validation cohort comprised of 504 meningiomas from independent, international institutions.
Clinical trials of molecular therapy that are based on molecular inclusion criteria are underway for patients with meningiomas 6 . We identify divergent temporal evolution in recurrent versus primary meningiomas, suggesting that molecular analyses guiding clinical decision-making should be performed on recurrent tumor tissue rather than archival samples from prior resections. Our results also indicate that regionally distinct spatial evolution represents a barrier to accurate tumor classi cation and should be considered during histological or molecular analyses of meningiomas. Beyond classi cation, our preclinical model for testing personalized medical therapies to overcome intratumor heterogeneity may address the limitations molecular, temporal, or spatial evolution place on improving treatments for patients. Indeed, we show this system can enable medium-throughput screening of novel pharmacological strategies to treat tumors that are resistant to standard interventions. This system also suggests that meningioma cell growth patterns can be in uenced by cell heterogeneity in the tumor microenvironment (Fig. 7b, f, g), and phenotypes such as these may hint at additional response or resistance mechanisms. To that end, clinical trials of abemaciclib (NCT02523014) or selumetinib (NCT03095248) as monotherapy for meningiomas are ongoing, but our data suggest that combination molecular therapy may be necessary to reverse the longstanding trend of non-positive clinical trials for patients with meningiomas 22,23,47,48 .

Declarations
Data availability DNA sequencing and spatial transcriptomic data that support the ndings of this study have been deposited in the Sequence Read Archive (https://www.ncbi.nlm.nih.gov/sra) under BioProject ID PRJNA950017. DNA methylation, and RNA sequencing data that support the ndings of this study have been deposited in the NCBI Gene Expression Omnibus (https://www.ncbi.nlm.nih.gov/geo/) under accession numbers GSE228316 (DNA methylation) and GSE228433 (RNA sequencing). The publicly available GRCh38 (hg38) and CRCh37.p13 (hg19) were used in this study. Source data are provided with this paper.

Inclusion and ethics
This study complied with all relevant ethical regulations and was approved by the UCSF Institutional Review Board (13-12587, 17-22324, 17-23196 and 18-24633). As part of routine clinical practice at UCSF, all patients who were included in this study signed a waiver of informed consent to contribute deidenti ed data to research projects.

Meningiomas, clinical data, histology, and light microscopy
The study cohort consisted of 16 samples from 10 clinically aggressive meningiomas that were resected from 9 patients at UCSF from 2009 to 2021. Patient demographics, treatments, and clinical outcomes were recorded from the electronic medical record (Supplementary Table 1). Magnetic resonance imaging studies were reviewed to de ne meningioma locations and clinical outcomes. Detailed pathologic examination of the entire cohort was performed by a board-certi ed neuropathologist (C-H.G.L) to assess for histological or molecular heterogeneity. Histological and molecular grading were assigned using the 2021 WHO Classi cation of Central Nervous System Tumors 20 . For bulk sequencing analyses, meningioma tissue was isolated from formalin-xed, para n-embedded (FFPE) blocks using biopsy punches (Integra Miltex Instruments, cat# 33-31-P/25). Genomic DNA was extracted from macrodissected FFPE tumor tissue using the QIAamp DNA (Qiagen, cat# 56404) and the QIAamp RNeasy FFPE Tissue Kits (Qiagen, cat# 73504) at UCSF. For spatial pro ling assays, 6 mm cores were punched from FFPE blocks using biopsy punches, and serial sections were mounted onto glass slides for spatial transcriptomic, protein pro ling, H&E histology, or immunohistochemistry. Clinically validated immunohistochemistry for Ki-67 (DAKO, 1:50 dilution, MIB1 clone, cat# M7240), H3K27me3 (Cell Signaling, 1:50 dilution, C36B11 clone, cat# 9733, and p16 (MTM Labs, undiluted, E6H4 clone, cat# 9511) were performed at UCSF on core mounts with appropriate controls using a Leica Bond III platform and imaged using light microscopy on an BX43 microscope with standard objectives (Olympus). Images were obtained and analyzed using the Olympus cellSens Standard Imaging Software package (v1.16).

DNA methylation pro ling and analysis
Genomic DNA underwent bisul te conversion using the EZ DNA Methylation kit (Zymo Research, cat# D5004), followed by ampli cation, fragmentation, and hybridization to In nium EPIC 850k Human DNA Methylation BeadChips (Illumina, cat# 20020530) according to manufacturer's instructions at the Molecular Genomics Core at the University of Southern California (Los Angeles, CA). Bioinformatic analysis was performed in R (v3.6.1). Meningioma DNA methylation data were preprocessed using the SeSAMe pipeline (Bioconductor v3.10) as previously described 13,54 . In brief, probes were ltered and analyzed using normal-exponential out-of-band background correction, nonlinear dye bias correction, pvalue with out-of-band array hybridization masking, and β value calculation (β=methylated/[methylated+unmethylated]). Meningioma samples were assigned to Merlin-intact, Immune-enriched, or Hypermitotic DNA methylation groups using a support vector machine classi er, as previously described 13 .
Targeted DNA sequencing and analysis Targeted DNA sequencing was performed using the UCSF500 NGS panel, as previously described 26 . In brief, this capture-based next-generation DNA sequencing assay targets all coding exons of 479 cancerrelated genes, select introns, and upstream regulatory regions of 47 genes to enable detection of structural variants such as gene fusions and DNA segments at regular intervals along each chromosome to enable genome-wide copy number and zygosity analyses, with a total sequencing footprint of 2.8 Mb (Supplementary Table 2 Spatial transcriptome sequencing and analysis Spatial transcriptomic pro ling was performed on FFPE sections using the 10x Genomics Visium Spatial assay (v1, cat# 1000336). 6 mm cores were mounted within capture areas on Visium glass slides, depara nized, stained with H&E, and imaged at the Gladstone Institutes Histology Core (San Francisco, CA). Libraries were prepared according to manufacturer instructions at the Gladstone Institutes Genomics Core (San Francisco, CA). Libraries were sequenced on an Illumina NovaSeq 6000 instrument at the UCSF Center for Advanced Technology. Sequencing was performed with the recommended protocol (read 1: 28 cycles, i7 index read: 10 cycles, i5 index read: 10 cycles, and read 2: 91 cycles). FASTQ sequencing les and histology images were processed using the 10x SpaceRanger pipeline and the Visium Human Transcriptome Probe Set v1.0 GRCh38-2020-A. Data were visualized using the 10x Loupe Browser software (v6.3.0). Principal component analysis (PCA) was run on the normalized ltered feature-barcode matrix to reduce the number of feature (e.g. gene) dimensions. Uniform manifold approximation and projection (UMAP) analysis was used to visualize spatial transcriptomes in a 2D space. Graph-based clustering was performed to cluster spatial transcriptomes with related expression pro les together based on their concordance in PCA space. Differential expression analyses were performed using mean gene expression in each cluster, log2 fold-change of gene mean expression in a cluster relative to all other spatial transcriptomes, and a p-value denoting gene expression signi cance in each cluster relative to spatial transcriptomes in other clusters. P-values in each cluster were adjusted for false discovery rate to account for the number of genes being tested. Heatmaps of spatial transcriptomic data were generated in the Loupe Browser, which considers the top N genes for each cluster, sorted by log2 fold-change (by default N = 120/X, where X is the total number of spatial transcriptome clusters). Heatmaps were generated using hierarchical clustering with euclidean distance and average linkage.
Spaceranger generated ltered feature matrices were imported into a Seurat object (v4.3.0, arguments min.cells=3, min.features=100) using R (v4.2.1) and RStudio (v2022.07.2 Build 576) (Supplementary Table 4). The individual count matrices were normalized by nFeature_RNA count (subset=nFeature_RNA>1500 and nFeature_RNA<9500) and integrated with Harmony (v0.1.1). Optimal cluster resolution was determined using Clustree (v0.5.0, analyzing resolutions 5, 2, 1, 0.9, 0.8, 0.7, 0.6, 0.6, 0.5, 0.4, 0.3, 0.1, 0.0), and subsequent principal component (npcs=30) and UMAP (dims=1:30, min.dist=0.2) analyses were performed. UMAP projections and cluster distributions were visualized in the Loupe browser after combining spatial transcriptomic data from individual capture areas using the 10x Spaceranger aggr pipeline (v2.0.0). CNV analysis from spatial transcriptomes was performed using inferCNV (v1.14.0) and spatialinferCNV (v0.1.0). Capture areas of interest were combined with an additional capture area containing a geographic population of non-neoplastic cells, using the 10x Spaceranger aggr pipeline and Harmony, as described above. The cluster distribution was visually assessed in the Loupe browser to identify the cluster containing non-neoplastic tissue such as brain or endothelial. All cluster annotations were exported into a csv le and imported into R, along with the aggregate ltered feature matrix. The count matrix, annotated clusters, and a gene order le were input into inferCNV (arguments: cutoff=0.1, cluster_by_groups=TRUE, HMM = TRUE, denoise=TRUE) to generate a six-state CNV probability model for each spatial transcriptomic cluster. Deconvolution of meningioma cell types from single-cell RNA sequencing was performed using SCDC (v 0.0.0.9000). To do so, each spatial transcriptome was treated as a pseudobulked RNA sequencing dataset and leveraged against known cell types from a reference single-cell RNA sequencing dataset comprised of 57,114 cells from 8 human meningioma samples representing all DNA methylation groups 13 . Spatial and single-cell transcriptomic data were separately processed for quality control using QC ltering, normalization, dimensionality reduction, and clustering. Single-cell transcriptomic data were subsampled to 1000 cells per cell type, and the top differentially expressed genes were selected for each cell type. Using this expression set, spatial transcriptomes were deconvolved to yield a matrix with predicted proportions of cell type for each spatial transcriptome, which were visualized using SpatialFeatureplot (Seurat v3).

Spatial protein pro ling and analysis
Spatial protein pro ling was performed on FFPE sections using the NanoString Digital Spatial Pro ler at the UCSF Laboratory for Cell Analysis Genome Core (San Francisco, CA). Meningioma sections were labeled with DAPI and a multiplexed cocktail of 78 oligo-conjugated antibodies (Supplementary Table 5) using human protein panel modules generated at NanoString Technologies (Seattle, WA). H&E stained whole slide images were overlayed on uorescent DAPI projections and 200μm regions of interest were annotated based on histological and morphological heterogeneity by a board-certi ed neuropathologist (C-H.G.L). Oligonucleotides were released from regions of interest using ultraviolet cleavage, aspirated tags were hybridized to optical barcodes, and processed using the NanoString nCounter Analysis System.
Barcodes were rst normalized with internal spike-in controls and then normalized against housekeeping genes. Principal components analysis was performed using the prcomp function in R (v3.6.1) using default settings.
Multiplexed sequential immuno uorescence (seqIF) and microscopy Automated multiplexed seqIF staining and imaging was performed on FFPE sections at Northwestern University using the COMET platform (Lunaphore Technologies). The multiplexed panel was comprised of 29 antibodies (Supplementary Table 7). The 29-plex protocol was generated using the COMET Control Software, and reagents were loaded onto the COME device to perform seqIF. All antibodies were validated using conventional IHC and/or IF staining in conjunction with corresponding uorophores and 4',6diamidino-2-pheynlindole counterstain (DAPI, ThermoFisher Scienti c, cat# 62248). For optimal concentration and best signal-to-noise ratio, all antibodies were tested at 3 different dilutions, starting with the manufacturer-recommended dilution (MRD), MRD/2, and MRD/4. Secondary Alexa uorophore 555 (ThermoFisher Scienti c, cat# A32727) and Alexa uorophore 647 (ThermoFisher Scienti c, cat# A32733) were used at 1/200 or 1/400 dilutions, respectively. The optimizations and full runs of the multiplexed panel were executed using the seqIF technology integrated in the Lunaphore COMET platform (characterization 2 and 3 protocols, and seqIF protocols, respectively). The seqIF work ow was parallelized on a maximum of 4 slides, with automated cycles of iterative staining of 2 antibodies at a time, followed by imaging, and elution of the primary and secondary antibodies, with no sample manipulation during the entire work ow. All reagents were diluted in Multistaining Buffer (Lunaphore Technologies, cat# BU06). The elution step lasted 2min for each cycle and was performed with Elution Buffer (Lunaphore Technologies, cat# BU07-L) at 37°C. Quenching lasted for 30sec and was performed with Quenching Buffer (Lunaphore Technologies, cat# BU08-L). Imaging was performed with Imaging Buffer (Lunaphore Technologies, cat# BU09) with exposure times set at 4min for all primary antibodies, except P16 antibody at 8min, and secondary antibodies at 2min. Imaging was performed with an integrated epi uorescent microscope at 20x magni cation. Image registration was performed immediately after concluding the staining and imaging procedures by COMET Control Software. Each seqIF protocol resulted in a multi-stack OME-TIFF le where the imaging outputs from each cycle were stitched and aligned. COMET OME-TIFF les contain a DAPI image, intrinsic tissue auto uorescence in TRITC and Cy5 channels, and a single uorescent layer per marker. Markers were subsequently pseudocolored for visualization of multiplexed antibodies. Cell culture RNA sequencing and analysis RNA was extracted from triplicate M10G cultures (sgNTC, sgCDKN2A, sgCDKN2B, sgARID1A) using the RNeasy Plus Mini Kit (Qiagen, cat#74134). 1ug of RNA from each condition was shipped to Medgenome (Foster City, CA) for bulk RNA sequencing (Supplementary Table 9). Quality control was performed using FASTQC (v0.11.9) and the results were aggregated using MultiQC (v1.12). Adapter sequences and bases with quality scores <30 at the 3' and 5' ends of the reads were trimmed using Cutadapt (v3.7). Trimmed treads that were less than 20 bases in length were discarded. Processed reads were mapped to the reference genome GRCh38 using HISAT2 (v2.2.0) with default parameters. FeatureCounts (v2.0.0) was used to extract gene expression counts. The resulting count matrix was used to perform differential gene expression analysis with DESeq2 (v1.36.0).
Gene Set Enrichment Analysis (GSEA, v4.3.2) was performed to determine whether differentially expressed in M10G cultures belonged to common biological pathways. Gene rank scores were calculated using the formula: sign(log 2 fold-change) × −log10(p-value). Pathways were de ned using the gene set le Human_GOBP_AllPathways_no_GO_iea_December_01_2022_symbol.gmt, which is maintained by the Bader laboratory. Gene set size was limited to range between 15 and 500, and positive and negative enrichment les were generated using 2000 permutations. The EnrichmentMap App (v3.3.4) in Cytoscape (v3.7.2) was used to visualize the results of pathway analysis. Nodes with FDR q value < 0.05 and p-value < 0.05, and nodes sharing gene overlaps with Jaccard + Overlap Combined (JOC) threshold of 0.375 were connected by blue lines (edges) to generate network maps. Clusters of related pathways were identi ed and annotated using the AutoAnnotate app (v1. 3.5) in Cytoscape that uses a Markov Cluster algorithm to connect pathways by shared keywords in the description of each pathway. The resulting groups of pathways were designated as the consensus pathways in a circle.

Statistics
All experiments were performed with independent biological replicates and repeated, and statistics were derived from biological replicates. Biological replicates are indicated in each gure panel or gure legend.
No statistical methods were used to predetermine sample sizes, but sample sizes in this study are similar or larger to those reported in previous publications. Data distribution was assumed to be normal, but this was not formally tested. Investigators were blinded to conditions during clinical data collection and analysis of mechanistic or functional studies. Bioinformatic analyses were performed blind to clinical features, outcomes or molecular characteristics. The clinical samples used in this study were retrospective and nonrandomized with no intervention, and all samples were interrogated equally. Thus, controlling for covariates among clinical samples is not relevant. Cells and animals were randomized to experimental conditions. No clinical, molecular, or cellular data points were excluded from the analyses.
Unless speci ed otherwise, lines represent means, and error bars represent standard error of the means. Results were compared using Student's t-tests, which are indicated in gure legends alongside approaches used to adjust for multiple comparisons. In general, statistical signi cance is shown by asterisks (*p£0.05, **p£0.01, ***p£0.0001), but exact p-values are provided in the gure legends when possible.

Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this article. Figure 1 Experimental design and work ow. a, 16 high-grade meningioma samples from 10 meningiomas that were resected from 9 patients were analyzed using comprehensive histologic, immunohistochemical, and bulk and spatial bioinformatic techniques, including spatial transcriptomics, spatial protein pro ling, multiplexed sequential immuno uorescence microscopy, and spatial deconvolution of meningioma single-cell RNA sequencing. Results were validated using RNA sequencing from 502 meningiomas, and CRISPR interference, pharmacology, and live cell imaging of meningioma organoid preclinical models.

Figures
Scale bars, 1mm for meningiomas and 100μm for meningioma organoids. b, Oncoprint comprised of the clinical, histologic, genetic, epigenetic, and gene expression features of the meningioma samples in this study. c, Uniform manifold approximation and projection (UMAP) of 38,718 meningioma spatial transcriptomes after Harmony batch correction shaded by sample of origin. d, UMAP of meningioma spatial transcriptomes after Harmony batch correction shaded by unsupervised hierarchical clusters. e, Heatmap of meningioma spatial protein pro ling comprised of 72 proteins from 82 regions revealing signi cant inter-and intratumor heterogeneity.    Regionally distinct sub-clonal spatial transcriptomes underlie histological heterogeneity in high-grade meningioma. a, Ki-67 immunohistochemistry (IHC) of regionally distinct samples from M9 demonstrating heterogeneous histological (WHO grade 2 or 3), mutational (ARID1A, ASXL1), and cytogenetic (chromosome 4, 14q) features (Fig. 1b). b, p16 IHC of regionally distinct samples from M10 demonstrating heterogeneous histological (p16, Ki-67) and cytogenetic (chromosome 1q, 4q, 9p, 10q) features (Fig. 1b). c, UMAP analysis of M9 spatial transcriptomes after Harmony batch correction shaded by region of origin (left) or unsupervised hierarchical clusters (right). Scale bar, 1mm. d, UMAP analysis of M10 spatial transcriptomes after Harmony batch correction shaded by region of origin (left) or unsupervised hierarchical clusters (right). Scale bar, 1mm. e, Spatial distribution of unsupervised hierarchical spatial transcriptome clusters from M9 after Harmony batch correction. Scale bar, 1mm. f, Spatial distribution of unsupervised hierarchical spatial transcriptome clusters from M10 after Harmony batch correction. Scale bar, 1mm. g, Distribution of unsupervised hierarchical spatial transcriptome clusters from M9 after Harmony batch correction. Spatial transcriptome clusters correlating with WHO grade 3 histology are annotated. h, Top 89 differentially expressed genes across unsupervised hierarchical spatial transcriptome clusters from M9. I, Spatial distribution of differentially expressed genes associated with histological variability across regionally distinct samples from M9. Scale bar, 1mm. j, Distribution of unsupervised hierarchical spatial transcriptome clusters from M10 after Harmony batch correction. k, Top 110 differentially expressed genes across unsupervised hierarchical spatial transcriptome clusters from M10. l, Spatial distribution of differentially expressed genes associated with histological variability across regionally distinct samples from M10. Scale bar, 1mm.  A preclinical platform for testing personalized medical therapies to overcome intratumor heterogeneity in high-grade meningiomas. a, Network of gene circuits distinguishing M10G dCas9-KRAB meningioma cells expressing sgNTC (n=3), sgCDKN2A (n=3), sgCDKN2B (n=3), or sgARID1A (n=3) using RNA sequencing.