MerlinS13 phosphorylation controls meningioma Wnt signaling and magnetic resonance imaging features

Meningiomas are the most common primary intracranial tumors and are associated with inactivation of the tumor suppressor NF2/Merlin, but one-third of meningiomas retain Merlin expression and typically have favorable clinical outcomes. Biochemical mechanisms underlying Merlin-intact meningioma growth are incompletely understood, and non-invasive biomarkers that predict meningioma outcomes and could be used to guide treatment de-escalation or imaging surveillance of Merlin-intact meningiomas are lacking. Here we integrate single-cell RNA sequencing, proximity-labeling proteomic mass spectrometry, mechanistic and functional approaches, and magnetic resonance imaging (MRI) across meningioma cells, xenografts, and human patients to define biochemical mechanisms and an imaging biomarker that distinguish Merlin-intact meningiomas with favorable clinical outcomes from meningiomas with unfavorable clinical outcomes. We find Merlin drives meningioma Wnt signaling and tumor growth through a feed-forward mechanism that requires Merlin dephosphorylation on serine 13 (S13) to attenuate inhibitory interactions with β-catenin and activate the Wnt pathway. Meningioma MRI analyses of xenografts and human patients show Merlin-intact meningiomas with S13 phosphorylation and favorable clinical outcomes are associated with high apparent diffusion coefficient (ADC) on diffusion-weighted imaging. In sum, our results shed light on Merlin posttranslational modifications that regulate meningioma Wnt signaling and tumor growth in tumors without NF2/Merlin inactivation. To translate these findings to clinical practice, we establish a non-invasive imaging biomarker that could be used to guide treatment de-escalation or imaging surveillance for patients with favorable meningiomas.


Introduction
Meningiomas arising from the meningothelial lining of the central nervous system comprise over 40% of primary intracranial tumors 1,2 , and approximately 1% of humans will develop a meningioma in their lifetime 3 . Meningiomas are treated with surgery and radiotherapy, and systemic therapies remain ineffective or experimental for patients with meningiomas 4 . Bioinformatic investigations have revealed biological drivers and therapeutic vulnerabilities underlying meningiomas with unfavorable outcomes [5][6][7][8][9][10][11][12] , and clinical trials of new therapeutic strategies to treat patients with meningiomas that are resistant to standard interventions are underway 4 . Nevertheless, most meningiomas are benign and many can be safely observed with serial magnetic resonance imaging (MRI) 13 . Imaging surveillance can spare patients from morbidities associated with potentially unnecessary surgical or radiotherapy treatments, but current meningioma classi cation systems rely on histological and molecular analyses of tumor tissue after resection 14 . Thus, there is an unmet need to develop non-invasive imaging biomarkers that predict meningioma outcomes and could be used to guide treatment de-escalation or surveillance for the most common primary intracranial tumor.
Meningiomas are often associated with inactivation of the tumor suppressor NF2/Merlin 15 , but approximately one-third of meningiomas are Merlin-intact and have favorable clinical outcomes [7][8][9]11,16,17 . Merlin is a member of the FERM family of proteins (Four-point-one, Ezrin, Radixin, and Moesin) that link the cytoskeleton to the plasma membrane and are comprised of a FERM domain, an α-helical domain, and a C-terminal domain (CTD). Merlin-intact meningiomas can encode somatic short variants (SSVs) targeting TRAF7, PIK3CA, AKT1, KLF4, or the Hedgehog pathway 16,17 , but some of these variants may be passenger mutations that do not in uence meningioma tumorigenesis [18][19][20] . These data suggest biochemical mechanisms driving Merlin-intact meningiomas are incompletely understood. Here we test the hypothesis that understanding signaling mechanisms in Merlin-intact meningiomas may shed light on new strategies to de ne meningioma biology pre-operatively using non-invasive imaging techniques.

Results
To study Merlin signaling mechanisms in meningiomas, CH-157MN human meningioma cells lacking endogenous Merlin 21 were transduced with a doxycycline-inducible NF2 construct and grown as xenografts in mice (Fig. 1a). Merlin rescue with doxycycline in CH-157MN cells did not in uence meningioma histology, growth, or overall survival compared to xenografts in mice without doxycycline (Fig. 1b, c and Extended Data Fig. 1). To validate these ndings, IOMM-Lee human meningioma cells encoding endogenous Merlin 22 were transduced with short-hairpin RNAs (shRNAs) suppressing NF2 (shNF2), or non-targeted control shRNAs (shNTC), and grown as xenografts in mice (Extended Data Fig. 2a). Merlin loss in IOMM-Lee cells did not in uence meningioma histology, growth, or overall survival compared to mice with IOMM-Lee cells expressing shNTC (Extended Data Fig. 2b-e). Single-cell RNA sequencing was performed on 40,765 human meningioma cells from 5 Merlin-de cient CH-157MN xenografts and 7 CH-157MN xenografts with Merlin rescue (Extended Data Fig. 3a). Uniform manifold approximation projection (UMAP) and Louvain clustering de ned 17 meningioma xenograft cell states (Fig. 1d), which were distinguished using differentially expressed genes and phases of the cell cycle (Extended Data Fig. 3b, c and Supplementary Table 1). Comparison of meningioma cell states across xenografts revealed only 1 cluster was enriched in Merlin rescue xenografts compared to Merlin de cient xenografts (Fig. 1e). This cluster was distinguished by expression of the Wnt pathway effector CTNNB1/ β-catenin (Extended Data Fig. 3a). In support of these data, Wnt target genes were also enriched in single cells from Merlin rescue xenografts compared to Merlin-de cient xenografts (Fig. 1f).
Markers of Wnt pathway activation have been identi ed in human meningiomas 23 , but how Wnt signals are transduced through meningioma cells is unknown. Canonical Wnt signaling proceeds through βcatenin, which is degraded by a destruction complex that is active the absence of Wnt stimulation 24 . Wnt stimulation induces PP1A to de-phosphorylate and inactivate the β-catenin destruction complex 25 , allowing β-catenin to localize to the plasma membrane or to the nucleus in complex with the transcriptional co-activator Tcf/Lef.
To identify candidates underlying Merlin regulation of the Wnt pathway, M10G cells were transduced with doxycycline-inducible wildtype (WT) or cancer-associated missense Merlin constructs (L46R, A211D) that encoded C-terminal FLAG and APEX2 tags to enable subcellular localization and proximity labeling proteomic mass spectrometry studies 28 . Immuno uorescence or immunoblots after biochemical subcellular fractionation demonstrated decreased stability and re-localization of Merlin L46R and Merlin A211D compared to Merlin WT that was ampli ed after streptavidin labeling of biotinylated peptides in proximity to Merlin constructs ( Fig. 2a and Extended Data Fig. 5a). Streptavidin pulldown and proximity-labeling proteomic mass spectrometry identi ed β-catenin adjacent to Merlin WT but not Merlin L46R , and also revealed that β-catenin was diminished in proximity to Merlin A211D (Fig. 2b and   Supplementary Table 2). In support of these data, Merlin L46R and Merlin A211D were unable to rescue Wnt signaling in M10G cells with CRISPRi suppression of endogenous NF2 (Fig. 2c). β-catenin was not degraded by loss of NF2 (Extended Data Fig. 5b). Subcellular fractionation of CH-157MN cells after doxycycline-inducible Merlin rescue showed β-catenin was distributed across the plasma membrane, cytoplasm, cytoskeleton, and nucleus with Merlin L46R and Merlin A211D rescue, but was only enriched at the plasma membrane and in the nucleus with Merlin WT rescue (Fig. 2d). β-catenin suppression using siRNAs (siCTNNB1) inhibited meningioma Wnt signaling ( Fig. 2e and Extended Data Fig. 5c), but Merlin was required for maximal Wnt pathway activation in meningioma cells even after β-catenin overexpression ( Fig. 2f and Extended Data Fig. 5d). In the absence of Wnt3a, Merlin overexpression did not activate the Wnt pathway with or without endogenous Merlin, but endogenous Merlin suppression attenuated Wnt signaling (Fig. 2g, h). In the presence of Wnt3a, overexpression of Merlin activated the Wnt pathway regardless of endogenous Merlin status (Fig. 2g, h). In sum, these data suggest Merlin regulates the Wnt pathway through a feed-forward mechanism that in uences Merlin/β-catenin interaction and subcellular localization.
Visual inspection of available crystal structures of Merlin (PDB 4ZRJ) revealed that L46R and A211D cancer-associated missense mutations were located on α-helices embedded in hydrophobic pockets of the Merlin FERM domain, suggesting that charged amino acid substitutions of L46 or A211 may destabilize the secondary structure of the protein (Fig. 3a, b). The Merlin N-terminal domain (NTD) is unique among FERM family members, and structural modeling showed the NTD is a exible, 19-residue α-helix that protrudes from the surface of the protein (Fig. 3a). Overexpression of Moesin, or Merlin rescue using a construct lacking the NTD (Merlin ∆NTD ), were unable to rescue Wnt signaling in Merlin de cient meningioma cells ( Fig. 3c and Extended Data Fig. 6a). Thus, we hypothesized the Merlin NTD may regulate Merlin/β-catenin interaction and Wnt signaling in meningiomas.
Sequence analysis of the Merlin NTD showed a phosphorylation site on serine 13 (S13) within consensus motifs for a kinase (PKC) and a phosphatase (PP1A), both of which are core components of the Wnt pathway that were also identi ed in proximity to Merlin FLAG−APEX2 constructs in meningioma cells (Supplementary Table 2). Proteomic proximity-labeling mass spectrometry of unphosphorylatable (S13A) or phospho-mimetic (S13D) Merlin constructs revealed β-catenin adjacent to Merlin S13D but not Merlin S13A (Fig. 2a, b). Immunoprecipitation and immunoblots validated β-catenin interaction with Merlin S13D but not Merlin S13A (Fig. 3d). Rescue of Merlin S13D but not Merlin S13A sequestered β-catenin at the plasma membrane in meningioma cells (Fig. 2d), and Merlin S13A but not Merlin S13D rescued Wnt signaling in meningioma cells lacking NF2 (Fig. 3e). Moreover, Merlin S13D rescue attenuated meningioma cell proliferation in vitro (Fig. 3f), inhibited meningioma xenograft growth in vivo ( Fig. 3g and Extended Data Fig. 6b, c), and prolonged overall survival compared to Merlin WT or Merlin S13A rescue (Fig. 3h). Immunoblots of meningioma cell lysates after shRNA suppression of PKC ± phosphatase inhibition revealed a Merlin doublet that was eliminated with suppression of PKCα or PKCγ, or with overexpression of Merlin S13A (Extended Data Fig. 6d). A novel phospho-speci c antibody recognizing Merlin pS13 (Extended Data Fig. 6e) showed small-interfering RNAs (siRNAs) suppressing PP1A increased Merlin pS13 immunoblot intensity but siRNAs suppressing PKCα and PKCγ decreased Merlin pS13 immunoblot intensity compared to non-targeted control siRNAs (siNTC) in meningioma cells (Fig. 3i, j and Extended Data Fig. 6f). Moreover, Merlin-dependent Wnt signaling was attenuated by siRNAs suppressing PP1A and activated by siRNAs suppressing PKCγ in meningioma cells compared to siNTC (Fig. 3k).
Surgery is the mainstay of meningioma treatment and is often essential to relieve neurological symptoms from tumor mass effect 29 . Nevertheless, many meningiomas are diagnosed incidentally or with minimal presenting symptoms, and the majority of incidentally-diagnosed meningiomas will not grow on longterm imaging surveillance 13 . These clinical observations underscore the unmet need for non-invasive, clinically-tractable biomarkers that predict meningioma outcomes and could be used to guide treatment de-escalation or imaging surveillance. Qualitative MRI features such as peritumoral edema, tumor calci cation, tumor location, adjacent bone destruction, or irregular tumor margins can be associated with higher-grade meningiomas on preoperative imaging studies 30-32 . Although MRI can easily diagnose meningiomas, qualitative approaches are not reliable for distinguishing meningioma outcomes 33-36 . Quantitative apparent diffusion coe cient (ADC) hypointensity on diffusion-weighted MRI is prognostic for unfavorable meningioma outcomes 37 , and may be associated with meningioma Wnt signaling 27 , but biochemical mechanisms underlying meningioma imaging features are unknown.
To study associations between meningioma ADC and tumor biology, 100 preoperative MRIs from meningiomas with available DNA methylation pro ling and targeted exome sequencing of the NF2 locus were retrospectively reviewed and imaging features were analyzed in the context of clinical follow-up data 7 . Meningioma ADC was dichotomized at the mean (Fig. 4a), and ADC high meningiomas had favorable clinical outcomes compared to ADC low meningiomas (5-year local freedom from recurrence 90.3% versus 48.7%, p < 0.0001, log-rank test) (Fig. 4b). DNA methylation pro ling controlled for artifacts from chromosome copy number variants (CNVs) reveals meningiomas are comprised of Merlin-intact, Immune-enriched, and Hypermitotic DNA methylation groups 7 , which are concordant with groups and subgroups of meningiomas derived from RNA sequencing or DNA methylation pro ling integrated with RNA sequencing, CNVs, and SSVs 8, 11,12 . Analysis of DNA methylation groups across ADC high versus ADC low meningiomas showed the majority of ADC high meningiomas were Merlin-intact (56% versus 16%, p < 0.0001, chi-squared test) (Fig. 4c). To determine if Wnt signaling underlies meningioma ADC, βcatenin was suppressed in meningioma cells using shRNAs (Extended Data Fig. 7a), which inhibited Wnt signaling (Extended Data Fig. 7b), attenuated cell proliferation and tumor growth ( Fig. 4d and Extended Data Fig. 7c), and prolonged overall survival compared to meningioma cells and xenografts expressing shNTC (Fig. 4e). MRI of Merlin-de cient meningioma xenografts showed Merlin WT or Merlin S13A rescue did not alter ADC, but ADC was increased with Merlin S13D rescue or Merlin WT rescue with concurrent suppression of β-catenin (Fig. 4f). Thus, meningioma ADC is inversely correlated with Wnt pathway activation, and ADC high meningiomas have favorable clinical outcomes in human patients and preclinical xenograft models.

Discussion
Here we show Merlin drives meningioma Wnt signaling and tumor growth through a feed-forward mechanism that requires Merlin S13 dephosphorylation to attenuate inhibitory interactions with β-catenin and activate the Wnt pathway (Fig. 4g). Integrating data from meningioma xenografts and patients, our results establish meningioma ADC as a potential non-invasive imaging biomarker of Wnt signaling in Merlin-intact meningiomas with S13 phosphorylation and favorable clinical outcomes (Fig. 4g). These data shed new light on how meningiomas can grow despite expressing the canonical tumor suppressor NF2/Merlin. Nevertheless, Merlin-intact meningiomas tend to be benign and have the most favorable outcomes across molecular groups of human meningiomas 7,8,11,12 . Many meningiomas can be safely observed with serial imaging surveillance 13 , but differentiating benign from aggressive meningiomas without subjecting patients to the morbidities associated with potentially unnecessary surgical treatments has been a barrier to improving clinical paradigms for patients with the most common primary intracranial tumor. To address this unmet need for patients with meningiomas, our data elucidate a non-invasive, clinically-tractable MRI biomarker that sheds light on meningioma biology and could be used to guide treatment de-escalation or imaging surveillance.
Our nding that Merlin post-translational modi cation can promote oncogenic Wnt signaling is unexpected considering the myriad tumor suppressor functions associated with NF2. CNVs deleting NF2 on chromosome 22q are early events underlying Immune-enriched or Hypermitotic meningioma tumorigenesis 7,27 , but Merlin-intact meningiomas encode SSVs targeting TRAF7, AKT1, or KLF4 that may be tumor-initiating 16,17 . Our data suggest Merlin S13 phosphorylation status and Wnt signaling may modify Merlin-intact meningioma growth. Proximity-labeling proteomic mass spectrometry coupled with mechanistic and functional approaches demonstrate PKC and PP1A are important for this signaling mechanism, but other (1) kinases or phosphatases, (2) Merlin domains, or (3) Wnt pathway members may also contribute to meningioma Wnt signaling. In support of that hypothesis, we show (1) suppression of PP1A or PKC isoforms partially regulates Merlin S13 phosphorylation (Fig. 3i, j), (2) epistatic Merlin S13D or Merlin ∆NTD rescue partially restores Wnt signaling in meningioma cells (Fig. 3c, e), and (3) the non-canonical Wnt pathway regulators AMOT, AMOTL1, AMOTL2, DSG2, and DLG1 were identi ed in proximity to Merlin alongside β-catenin in meningioma cells [38][39][40] (Supplementary Table 2). Indeed, the Wnt pathway can be activated by multiple mechanisms in meningiomas with unfavorable clinical outcomes 23 , and not all Merlin-intact meningiomas can be controlled with existing therapies of surgery or radiotherapy 7,12 . Thus, our results also shed light on potential targets for future molecular therapies that could be used to treat meningiomas which are resistant to standard interventions.
To our knowledge, the Merlin S13 phosphorylation site that sequesters β-catenin to the plasma membrane and inhibits the Wnt pathway has not been previously reported. The NTD in Merlin is unique compared to other FERM family members but is evolutionarily conserved across Merlin orthologs in higher eukaryotes. Previous studies show Merlin serine 518 (S518) phosphorylation regulates tumor suppressor functions by controlling protein conformation [41][42][43][44] , but cellular mechanisms that are in uenced by Merlin S518 phosphorylation are incompletely understood and how tertiary conformational changes affect Merlin functions are unclear. Like other FERM family members, the open/closed conformation of Merlin requires interactions between the CTD and the FERM domain 43,45,46 , which are not in uenced by S13 phosphorylation on the exible NTD (Fig. 3a). Consequently, our data reveal novel structural/functional insights for an important and extensively studied tumor suppressor.

Declarations Data availability
Single-cell RNA sequencing data (n=12 human meningioma xenograft samples) reported in this manuscript have been deposited in the NCBI Gene Expression Omnibus under the accession GSE224347.

Author contributions statement
All authors made substantial contributions to the conception or design of the study; the acquisition, analysis, or interpretation of data; or drafting or revising the manuscript. All authors approved the manuscript. All authors agree to be personally accountable for individual contributions and to ensure that questions related to the accuracy or integrity of any part of the work are appropriately investigated,

Methods
This study complied with all relevant ethical regulations and was approved by the UCSF Institutional Review Board (IRB #10-01141 and #18-24633). As part of routine clinical practice at both institutions, all patients who were included in this study signed a waiver of informed consent to contribute data and tissue to research.

Cloning
Plasmids encoding genes of interest were purchased from Addgene, or when unavailable, PCR ampli ed from cDNA. PCR products were cut using sticky-end restriction enzymes and ligated into plasmids with T4 ligase (NEB, Cat# M0202L). Ligated plasmids were transformed into Top10 or Stable II E.coli, colonies were isolated and expanded, and plasmid DNAs were sent for Sanger sequencing to con rm genes of interest.

Immunohistochemistry and light microscopy
Depara nization and rehydration of 5 µm FFPE meningioma xenograft sections and hematoxylin and eosin staining were performed using standard procedures. Immunostaining was also performed on 5 µm FFPE meningioma xenograft sections using an automated Ventana Discovery Ultra staining system. Immunohistochemistry was performed using rabbit monoclonal Ki-67 (Ventana, clone 30-9, Cat# 790-4286, 1:6) with incubation for 16 min. Histologic and immunohistochemical features were evaluated using light microscopy on an BX43 microscope with standard objectives (Olympus). Images were obtained and analyzed using the Olympus cellSens Standard Imaging Software package.

Immunoprecipitation
After protein lysis and quanti cation, protein content was normalized across samples and incubated with pre-washed HA (Sigma-Aldrich, Cat# A2095, 50ml per IP) or FLAG (Sigma-Aldrich, Cat# M8823, 50ml per IP) antibody-bound beads. The sample/bead slurry was left to rotate at 4°C for 4 hours before washing 4 times in lysis buffer supplemented with protease and phosphatase inhibitors as described above. Bound proteins were eluted from beads by boiling in 2x Laemmli buffer and separated by gel electrophoresis as described above.
Luciferase assay Cells were transfected with pRL-TK TOP-Flash Tcf/Lef luciferase reporter and with or without additional genes of interest, as indicated, for 48 hours. 24 hours before experimentation, cells were treated with or without 200ng/ml Wnt3a (R&D Systems, Cat# 5036-WN). Luciferase activity was detected using the Dual Luciferase Kit (Promega, Cat# G4100) and a GLO-Max Promega plate reader.
Relative gene expression was calculated using the DDCt method against a control gene, GAPDH. QPCR primers used were NF2_F (5'-TTGCGAGATGAAGTGGAAAGG-3 Proximity-labeling proteomic mass spectrometry M10G cells stably expressing pLV.APEX2 constructs encoding wildtype or variant Merlin constructs were seeded onto 5 x 15cm plates. Cells were treated with 0.1mg/ml doxycycline to induce Merlin expression 24 hours before APEX labelling. For labelling, 0.5mM biotin-phenol (Berry and Associates, Cat# BT1015) was added to each plate for 30 minutes at 37°C, and 1mM H 2 O 2 (Sigma Aldrich, Cat# H1009) was added to cell media on ice for 30 seconds to initiate the reaction. Media was replaced with quenching media (10mM Sodium Ascorbate, 1mM Azide, 5mM Trolox) for 2 washes. For mass spectrometry, cells were scraped and pelleted for biotin/streptavidin precipitation as previously described 7 .

Mouse xenografts
ForshRNA induction, cells were pre-treated with 3mM IPTG (Sigma Aldrich, Cat# I6758) 10-days prior to injection. Xenograft experiments were performed by injecting 3 million human meningiomas cells, either CH157-MN or IOMM-Lee, into the ank of 4-6-week-old NU/NU female mice (Envidigo). To induce plasmid or shRNA expression, mice were treated with or without 200mg/ml doxycycline (Sigma-Aldrich, Cat# D9891) and with or without 10M IPTG in cage water that was changed every 2-3 days. Kaplan-Meyer curves were created by recording deaths at the protocol end of 50% ulcerated tumor or tumor >2000 mm 3 . Tumors were processed for single-cell dissociation and single-cell RNA sequencing, QPCR, or immunoblotting.

Xenograft lysis
Tumors were dissected into small chunks and allocated for either RNA extraction for cDNA synthesis and QPCR, or protein extraction and immunoblotting. Tumor chunks were placed in a 2ml Eppendorf tube with a 7mm metal bead and 350ml of RLT lysis buffer for RNA extraction, or 300ml RIPA buffer for protein.
Tubes were shaken in a TissueLyser for 2 minutes at 30mhz, lysate was cleared by spinning in a desktop centrifuge at full speed for 5 minutes at 4°C and used for downstream analysis.

Single-cell dissociation of solid tumors
Tumors were freshly harvested from mice and sliced into small pieces using two #10 scalpels before enzymatic dissociation in 0.1mg/ml Colagenase Type 7 (Worthington Biochemicals, Cat# LS005332) at 37°C for 30 minutes, followed by 0.25% trypsin (Thermo Fisher Scienti c, Cat# 25200114) digest at 37°C for 10 minutes. Red blood cells were removed in 1X RBC lysis buffer (Invitrogen, Cat# 00-4300-54) at room temperature for 10 minutes. Sequential ltering of cells through 70mm and 40mm lters and cell counting using a Life Technologies Countess II was performed to generate a single-cell suspension.
Single-cell RNA sequencing and analysis Single cells were processed through a 10X Genomics Chromium Controller and libraries were created using the Chromium Single Cell 3' Library & Gel Bead Kit v3 (10X Genomics, Cat# 1000121) according to the manufacturers protocol with an intended yield of 8,000 single cells. Libraries were sequences on a NovaSeq S4 6000 at the UCSF Center for Advanced Technology.
Demultiplexing, identi cation of empty droplets, removal of duplicates, and alignment to the human or mouse reference genomes was performed using the CellRanger v6.1.2 pipeline. Count matrices were selected using cells with more than 50 unique genes, less than 8000 unique genes, and with less than 20% of genes assigned as mitochondrial. Data were processed using Seurat in RStudio using SCTransform. Dimensionality reduction was performed using principal component analysis and uniform manifold approximations and projections (UMAP) were performed on the reduced dimensionality data using the minimum distance of 0.2 and a Louvain clustering resolution of 0.8. Differentially expressed genes were identi ed with Wilcoxon Rank Sum test in Seurat.

Structural modeling
To obtain a full-length 3D representation of the Merlin human protein, a structural model was generated using the Robetta server (https://robetta.bakerlab.org/) and the solved structure of human Merlin-FERM as a template (PDB 4ZRJ). The model was inspected using pymol (v2.x) visualization software to rationalize the role of disease-associated mutations. Robetta provides a fully automated modeling procedure exploiting both ab initio and comparative models of protein domains. Comparative models were built from structures detected and aligned by HHSEARCH, SPARKS, and Raptor. Loop regions were assembled from fragments and optimized to t the aligned template structures.

Human magnetic resonance imaging
All patients underwent MRI examinations on a 3T Discovery MR750 scanner (GE Healthcare) using an eight-channel phased-array head coil prior to surgical resection. The imaging protocol included anatomical T2-weighted Fluid Attenuated Inversion Recovery (FLAIR) and Fast Spin Echo (FSE) images, along with 3D T1-weighted Inversion Recovery-Spoiled Gradient Recalled echo (IR-SPGR) imaging preand post-injection of a gadolinium-based contrast agent. Diffusion-weighted imaging or DTI was obtained in the axial plane with b=1000s/mm 2 and either 6 gradient directions and 4 excitations or 24 gradient directions and 1 excitation or b=2000s/mm 2 and 55 gradient directions (echo time/repetition time=108/1000ms, voxel size=1.7-2.0x1.7-2.0x2.0-3.0mm). To calculate ADC maps, a pipeline using the FMRIB's Diffusion Toolkit was applied to the diffusion-weighted imaging and DTI data as previously Multi-slice diffusion tensor imaging (DTI) was acquired using a single-element diffusion-weighted echoplanar imaging sequence with the following parameters: echo time/repetition time=30/2500ms, 8 signal averages, 3 diffusion directions, two b-values per direction (b-500 and 1000s/mm 2 ), in-plane resolution of 0.333x0.333mm with a partial Fourier factor of 1.5 in the phase-encoding direction, and the same eld of view, slice thickness, and slice number as the T2-weighted images. With respiratory gating, the total imaging time was 2 minutes and 20 seconds. To generate apparent diffusion coe cient (ADC) maps, Horos (Lesser General Public License, v3.0) imaging software was used to manually place polygonal regions-of-interest (ROIs) over the solid portion of the meningioma xenograft tumors and these volume measurements were averaged for each group. Areas of cystic change or hemorrhage as denoted by T2weighted images were excluded.

Statistics
No statistical methods were used to predetermine sample sizes, but our cohort sizes 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.
Bioinformatic analyses were performed blind to molecular characteristics. The clinical samples used in this study were non-randomized with no intervention, and all samples were interrogated equally. Thus, controlling for covariates among clinical samples is not relevant. Unless speci ed otherwise, lines represent means, and error bars represent standard error of the means. Results were compared using Student's t tests, Chi-squared tests, and log-rank tests, which are indicated in the text, methods, and gure legends alongside approaches used to adjust for multiple comparisons. Statistical signi cance is shown by *p£0.05, **p£0.01, or ***p£0.0001.  Merlin regulates b-catenin localization in meningioma cells. a, Immuno uorescence for FLAG (Merlin) or streptavidin in M10G meningioma cells expressing Merlin FLAG-APEX2 constructs after proximity-labelling.

Figures
Scale bar, 10 μm. b, Heatmap of b-catenin biotinylation peptide intensity from proximity-labeling proteomic mass spectrometry as in a (n=3/condition). c, TOP-Flash Tcf/Lef luciferase reporter assay in M10G dCas9-KRAB meningioma cells expressing sgRNAs suppressing NF2 (sgNF2) with or without rescue of Merlin FLAG-APEX2 wildtype or cancer-associated missense constructs. d, Immunoblots for FLAG (Merlin) or b-catenin after biochemical fractionation of CH-157MN meningioma cells expressing Merlin FLAG-APEX2 rescue constructs. Immunoblots for calreticulin, a-tubulin, vimentin, Rb, or histone H3 mark membrane (M), cytoplasmic (Cp), cytoskeletal (Cs), nuclear (N), or chromatin fractions (Ch), respectively. e, TOP-Flash Tcf/Lef luciferase reporter assay in M10G meningioma cells expressing non-targeted control siRNAs (siNTC) or siRNAs suppressing b-catenin (siCTNNB1) with or without 24-hours of Wnt3a treatment (100ng/ul). f-h, TOP-Flash Tcf/Lef luciferase reporter assays in M10G dCas9-KRAB meningioma cells expressing non-targeted control sgRNAs (sgNTC) or sgNF2 with or without 24-hours of Wnt3a treatment (100ng/μl) in the presence or absence of b-catenin or Merlin overexpression or rescue. f shows b-catenin overexpression fails to hyperactivate the Wnt pathway in the absence of Merlin. g and h show Wnt stimulation is necessary for Merlin overexpression to hyperactivate the Wnt pathway. Lines represent means, and error bar represent standard error of the means. **P≤0.01, ***P≤0.0001 (Student's t-test, one sided). Cyan shows the NTD, which is not conserved in other FERM family members, containing a consensus PP1A/PKC phosphorylation motif and phosphorylation site (S13) in orange. b, the 4RZJ X-ray structure of Merlin shows L46 and A211 in hydrophobic pockets that may be destabilized by charged cancerassociated missense mutations. c, TOP-Flash Tcf/Lef luciferase reporter assay in M10G dCas9-KRAB meningioma cells expressing sgRNAs suppressing NF2 (sgNF2) with or without rescue of Merlin constructs or overexpression of the FERM family member Moesin. d, Immunoblots for FLAG (Merlin), bcatenin, or GAPDH before versus after FLAG immunoprecipitation from M10G meningioma cells with or without doxycycline-inducible Merlin FLAG overexpression (1μg/ml). e, TOP-Flash Tcf/Lef luciferase reporter assay in M10G dCas9-KRAB meningioma cells expressing sgNF2 with or without rescue of Merlin wildtype or S13 unphosphorylatable (S13A) or phospho-mimetic (S13D) constructs. f, CH-157MN meningioma cell MTT assays for cell proliferation with or without doxycycline-inducible Merlin rescue (100ng/ml). g, CH-157MN xenograft measurements in NU/NU mice with or without doxycycline-inducible Merlin rescue (20μg/ml) as in f. h, Kaplan-Meier survival curve for CH-157MN xenograft overall survival in NU/NU mice as in g (log-rank test). i, j, Immunoblots for HA (Merlin) or Merlin with phosphorylated S13 (Merlin pS13 ) after Merlin immunoprecipitation from M10G cells with or without Merlin overexpression and concurrent expression of non-targeted control siRNAs (siNTC) or siRNAs suppressing PP1A or PKC isoforms. GAPDH immunoblots are shown from immunoprecipitation inputs as a loading control. k, TOP-Flash Tcf/Lef luciferase reporter assay in M10G dCas9-KRAB meningioma cells expressing non-targeted control sgRNAs (sgNTC) or sgNF2 with concurrent siNTC or siRNA suppression of PP1A or PKC isoforms.

Figure 4
High MRI apparent diffusion coe cient distinguishes Merlin-intact meningiomas with favorable clinical outcomes. a, Example brain magnetic resonance imaging (MRI) T1 post-contrast and apparent diffusion