Targeting the CSF1/CSF1R axis is a potential treatment strategy for malignant meningiomas

Abstract

Background

Malignant meningiomas are fatal and lack effective therapy. As M2 macrophages are the most prevalent immune cell type in human meningiomas, we hypothesized that normalizing this immunosuppressive population would be an effective treatment strategy.

Methods

We used CIBERSORTx to examine the proportions of 22 immune subsets in human meningiomas. We targeted the colony-stimulating factor 1 (CSF1) or CSF1 receptor (CSF1R) axis, an important regulator of macrophage phenotype, using monoclonal antibodies (mAbs) in a novel immunocompetent murine model (MGS1) for malignant meningioma. RNA sequencing (RNA-seq) was performed to identify changes in gene expression in the tumor microenvironment (TME). Mass cytometry was used to delineate changes in immune subsets after treatment. We measured patients’ plasma CSF1 levels using ELISA and CSF1R expression using multiplex quantitative immunofluorescence in a human meningioma tissue microarray.

Results

Human meningiomas are heavily enriched for immunosuppressive myeloid cells. MGS1 recapitulates the TME of human meningiomas, including an abundance of myeloid cells, a paucity of infiltrating T cells, and low programmed death ligand 1 (PD-L1) expression. Treatment of murine meningiomas with anti-CSF1/CSF1R, but not programmed cell death receptor 1 (PD-1), mAbs abrogate tumor growth. RNA-seq and mass cytometry analyses reveal a myeloid cell reprogramming with limited effect on T cells in the TME. CSF1 plasma levels are significantly elevated in human patients, and CSF1R is highly expressed on CD163⁺ macrophages within the human TME.

Conclusion

Our findings suggest that anti-CSF1/CSF1R antibody treatment may be an effective normalization cancer immunotherapy for malignant meningiomas.

Key Points.

• Meningiomas are heavily enriched for immunosuppressive myeloid cells.

• Anti-CSF1/CSF1R, but not PD-1, mAbs effectively abrogate murine tumor growth.

• CSF1R is highly expressed on CD163⁺ macrophages in human meningiomas.

Importance of the Study.

While most meningiomas are considered benign (grade I), 10% of all cases are of higher grade, including atypical (grade II) and anaplastic (grade III) meningiomas. Patients with these tumors endure long-term morbidities from repeat surgeries. There are currently no known effective medical therapeutic options. In this manuscript, we explored the immunological landscape of human meningiomas and found that these tumors are heavily infiltrated by anti-inflammatory M2 macrophages. In light of this, we attempted to normalize this tumor microenvironment using anti-CSF1/CSF1R antibody regimen in a first-ever immunocompetent Nf2-mutant murine meningioma model. We demonstrated the effectiveness of this strategy and investigated further using RNA sequencing and mass cytometry. Finally, we confirmed that patients with meningiomas indeed have increased plasma CSF1 levels and their tumors, but not normal tissue, have high CSF1R expression.

Meningiomas are the most common primary intracranial tumors with a prevalence of 170 000 cases in the United States.¹ Although most meningiomas are considered benign (grade I), 10% of all cases are of higher grade, including atypical (grade II) and anaplastic (grade III) meningiomas. While surgery is often curative for benign grade I meningiomas, higher-grade tumors tend to recur. In addition, the anatomical location of the tumor may limit the surgical approach, such as those that are deep-seated near eloquent parts of the central nervous system. Atypical or anaplastic meningioma subtypes frequently recur or are not amenable to surgical resection. Anaplastic meningiomas have a median survival of only 1.5 years, underscoring an unmet need for alternative therapies.²

Cancers evade the immune system by a variety of mechanisms that results in an immunosuppressive tumor microenvironment (TME). These mechanisms include upregulation of inhibitory ligands such as B7-H1 (also called PD-L1), infiltration of immunosuppressive tumor-associated macrophages (TAMs) and myeloid-derived suppressive cells, T-regulatory (Treg) cells, prevention of effector lymphocyte infiltration, as well as other molecular pathways yet to be discovered. Immunotherapies based on monoclonal antibodies (mAbs) to PD-1 and B7-H1 (PD-L1) (herein anti-PD therapy) have demonstrated tremendous success in various solid tumors by repairing defective immune responses, leading to normalization of immunity in the TME.³ In addition to the upregulation of the PD pathway, other immune defects may also play the roles in the evasion of tumor immunity. We previously reported significant differences in HLA class II signatures among Nf2-mutant meningiomas and used an advanced RNA deconvolution approach to reveal a skewed M2-macrophage signature within these tumors.⁴ Part of the success of anti-PD therapies in tumors, such as melanoma and non–small cell lung cancer, may be due to the relative abundance of lymphocytes over myeloid cells.^5–7 Tumors that are myeloid dominant, such as pancreatic cancer, may benefit from CSF1 (colony-stimulating factor 1)/CSF1R (CSF1 receptor)-directed therapies.⁸ Although meningiomas can harbor mature, antigen-experienced T and B cells, their quantity are inconsistent and mostly low.⁹ Conversely, meningiomas demonstrate consistently abundant infiltration by macrophages, regardless of grade.¹⁰

It is unclear whether higher-grade meningiomas will respond to anti-PD therapy.^11–13 To date, there have been limited in vivo studies assessing immunotherapeutics for meningiomas due to the lack of mouse meningioma cell lines that mimic the biology of human disease. To normalize the local immune dysfunction within the TME of meningiomas, a greater understanding of the immune landscape within these tumors is needed. In this study, we interrogated the immunological landscape of malignant meningiomas using in silico RNA deconvolution approaches in a large meningioma transcriptomic database. We identified and confirmed an excessive abundance of M2 macrophages compared to M1 macrophages, in addition to a paucity of T-cell infiltration in these tumors.^10,14 Next, we utilized an established MGS1 cell line (NF2 and Cdkn2ab loss) that overexpresses PDGF-B (platelet-derived growth factor subunit B) and mimics the TME of human meningiomas to study immune infiltration in this novel immunocompetent model.¹⁵ Blockade of CSF1 and CSF1R, targets chosen for their known properties in macrophage differentiation, in this model results in myeloid cell reprogramming within the TME and tumor suppression.

Methods

Study Approval

Institutional review board approval (HIC Protocol Number 9406007680) for genetic and related immunological studies, along with written consent from all study subjects, was obtained at Yale University. All mouse protocols were in accordance with NIH guidelines and were approved by the Institutional Animal Care and Use Committee of Yale University School of Medicine.

Estimation of Immune Cell Fractions in Human Meningiomas

Using CIBERSORTx (available at https://cibersortx.stanford.edu), we applied the gene signature file LM22 that defines 22 immune cell subtypes to our Yale human meningioma database with whole-exome sequencing and gene expression analyses as previously described.^4,16–18 The number of permutations was set to 1000. We utilized Bulk-mode batch correction and 1000 permutations for the analysis in relative mode. The filter criterion of each sample is set as the CIBERSORTx calculation of P < .05 (statistical significance of the deconvolution result across all cell subsets to ensure adequate goodness of fit).

Mice and Cell Lines

FVB mice aged 6-8 weeks old were purchased from Jackson Laboratories. MGS1 cell line was obtained from Peyre et al.¹⁵ The cell line was tested to be mycoplasma-free. These cells were shown to display consistent imbalances at the Cdkn2a locus with the original primary tumors.¹⁵ MGS1 was maintained in DMEM with 10% FBS (fetal bovine serum), 3.3 nM of EGF (epidermal growth factor), and 870 nM of insulin. All cells were grown at 37°C in a humidified atmosphere (5% CO₂). Flow cytometry to detect CSF1R expression on MGS1 was done using PE-conjugated anti-mouse CD115 (CSF1R) antibody (Biolegend, clone AFS98).

Mouse Model and Tumor Studies

A heterotopic, subcutaneous flank model was used. Briefly, 1 × 10⁶ MGS1 cells in 100 µl PBS (phosphate-buffered saline) were injected subcutaneously into the right flank. Tumor growth was monitored by an electronic caliper regularly and presented as tumor volume calculated as (length × width × width)/2 in millimeters. When indicated, mice were treated intraperitoneally with anti-CSF1 (clone 5A1), anti-CSF1R antibody (clone AFS98), and the respective isotype antibodies (BioXcell, USA). Anti-PD-1 (clone G4) and anti-4-1BB (clone 2A) antibodies were produced in-house. CD8⁺ T cells were depleted by pretreatment of the mice via intraperitoneal injections of 200 µg of anti-CD8 mAb (clone 2.43, BioXCell) 3 days prior to tumor inoculation, followed by every week maintenance injections. On the indicated days, tumors were excised, mechanically dissociated with the gentleMACS Dissociator (Miltenyi Biotec), and digested using the mouse tumor dissociation kit (Miltenyi Biotec, Germany). All heterotopic experiments measuring tumor growth were repeated. In addition, an orthotopic model was performed once by subdural (underneath the skull) injection of 0.25 × 10⁶ MGS1 cells in 25 µl PBS to the right and anterior to the bregma. The treatment schedule was twice weekly intraperitoneal injection of 200 µg of antibody starting on day +1 of tumor inoculation for a total of 7 weeks. Survival of the mice was monitored daily.

Gene Expression in Tumor Tissue Using RNA Sequencing (RNA-seq)

Tumor tissues were mechanically disaggregated, and mRNA was isolated using mirVana™ miRNA Isolation Kit, with phenol (Thermo Fisher Scientific, USA). RNA-seq was then performed at the Yale Center for Genome Analysis (YCGA).

RNA-seq Quality Control

Total RNA quality is determined by estimating the A260/A280 and A260/A230 ratios by nanodrop. RNA integrity is determined by running an Agilent Bioanalyzer gel, which measures the ratio of the ribosomal peaks.

RNA-seq Library Prep

mRNA is purified from approximately 200 ng of total RNA with oligo-dT beads and sheared by incubation at 94C in the presence of Mg (KAPA mRNA HyperPrep Kit). Following first-strand synthesis with random primers, second-strand synthesis and A-tailing are performed with dUTP for generating strand-specific sequencing libraries. Adapter ligation with 3′-dTMP overhangs is ligated to library insert fragments. Library amplification amplifies fragments carrying the appropriate adapter sequences at both ends. Strands marked with dUTP are not amplified. Indexed libraries that meet appropriate cutoffs for both are quantified by qRT-PCR using a commercially available kit (KAPA Biosystems) and insert size distribution determined with the LabChip GX or Agilent Bioanalyzer. Samples with a yield of ≥0.5 ng/µl are used for sequencing.

Flow Cell Preparation and Sequencing

Sample concentrations are normalized to 1.2 nM and loaded onto an Illumina NovaSeq flow cell at a concentration that yields 25 million passing filter clusters per sample. Samples are sequenced using 100-bp paired-end sequencing on an Illumina NovaSeq according to Illumina protocols. The 10-bp dual index is read during additional sequencing reads that automatically follow the completion of read 1. Data generated during sequencing runs are simultaneously transferred to the YCGA high-performance computing cluster. A positive control (prepared bacteriophage Phi X library) provided by Illumina is spiked into every lane at a concentration of 0.3% to monitor sequencing quality in real time.

Data Analysis and Storage

Signal intensities are converted to individual base calls during a run using the system’s Real Time Analysis (RTA) software. Base calls are transferred from the machine’s dedicated personal computer to the Yale high-performance computing cluster via a 1 Gigabit network mount for downstream analysis.

RNA-seq Analysis

We first trimmed the low-quality bases and removed adapter sequences using the cutadapt. RNA-seq reads were aligned to the reference mouse (mm10) transcriptome using the STAR mapping method. PICARD was used for quality assessment of reads and alignment rate. HTSEQ was conducted for quantification of the transcript counts. Differential gene expression analysis was performed using DESeq2. Ensembl and gene symbol ID’s mapping in mouse data were downloaded from BioMart in R. Pathway and Ontology enrichment analysis were performed using Metascape for mouse differentially expressed genes with a significance cutoff of FDR value <0.05 and Log2FC absolute value of >1.

Mass Cytometry and Analysis

Mouse tumor tissues were digested as described above. Upon filtration in a 70-µm cell strainer (BD Falcon), cells were first incubated with anti-mouse CD16/32 mAb for 10 minutes at room temperature to block Fc receptors and subsequently stained with the metal-labeled mAb cocktail against cell surface molecules. After the treatment with the Fixation/Permeabilization Buffer (Fluidigm), cells were further incubated with mAb cocktails against intracellular proteins. Antibodies used in the mass cytometry analysis were either purchased from Fluidigm or conjugated in-house (Supplementary Table S1). Cell samples were diluted in ddH2O containing bead standards to approximately 10⁶ cells per ml and then analyzed by a mass cytometer (CyTOF, Fluidigm) equilibrated with ddH2O.

All mass cytometry files were normalized, and manually gated in Cytobank software by DNA, event length, live/dead discrimination, CD45 and 4 bead channels to exclude dead, debris, doublets, and nonimmune cells and beads. As tumor specimens varied by size, degrees of immune cell infiltration were reported as percentage of total cells and all subpopulations were gated from thereon. Phenograph clustering analysis in R cytofkit package was used to automatically identify underlying immune subsets.¹⁹ Heatmaps were generated on the basis of the normalized mean value for each marker in clusters. Cell frequency in each cluster was calculated as the assigned cell events divided by the total CD45⁺ cell events in the same sample.

Human Plasma CSF1 Quantification

Blood samples were obtained from healthy controls and patients undergoing meningioma resection via venipuncture or blood sample from arterial lines, respectively, in collection tubes containing EDTA. The plasma was isolated by centrifugation of the blood samples and frozen at −80°C until analysis. CSF1 levels were measured using the Human M-CSF Quantikine ELISA Kit (R&D Systems, Minneapolis, MN, USA). The CSF1 levels of healthy controls and different grades of meningiomas were compared using the Kruskal-Wallis test.

Multiplexed Quantitative Immunofluorescence (QIF) by Automated Quantitative Analysis (AQUA) Method

Sections from the tissue microarray (TMA) block were stained for CD163 and CSF1R proteins. MUC1 protein was used to localize the tumor cells and DAPI (4′-6-diamidino-2-phenylindole) was used to stain the nuclei.

To summarize the staining protocol, TMA sections, formalin-fixed paraffin-embedded (FFPE) slides, were deparaffinized, rehydrated, and then underwent; then, were undergone antigen retrieval in 97°C EDTA solution, pH 8, for 20 minutes. Endogenous peroxidase activity was blocked using 30% hydrogen peroxide in methanol solution. Nonspecific antigens were blocked with a Tris-buffered saline-based solution containing bovine serum albumin (0.3%) and Tween (0.05%). Both the blocking steps were performed at room temperature and for 30 minutes each. Slides were then incubated overnight at 4°C temperature with the cocktails of primary antibodies as below; CD163 (1:7500; Leica, Mouse IgG1, clone 10D6) and CSF1 receptor (1:1000; Cell Signaling Technology, Rabbit IgG, clone E4T8Z). The staining was then completed the next day following a sequential multiplexed immunofluorescence protocol as previously published.²⁰ Tumor cells were stained using antibody against MUC1 (1:300, Cell Signaling Technology, Mouse IgG1, clone VU4H5). Secondary antibody and fluorescent reagents used for staining of CSF-R were Anti-Rabbit Envision (K4009, Dako) with TSA Cyanine 5 Reagent (Akoya Biosciences). Rat Anti-Mouse IgG1 Secondary Antibody (eBioscience) and Goat Anti-Mouse IgG3 heavy chain (Abcam) were used as secondary antibodies for CD163 accordingly. The sections were scanned on PM-2000 systems (Navigate BioPharma, Carlsbad, CA, USA) and the acquired fluorescence images were analyzed for measuring the protein expression of all the markers using the AQUA method as described in more details previously.²¹ Visual cutoff point was determined by examining the QIF score for which there was membrane staining as determined by one of the authors (V.Y.).

Statistical Analysis

Prism 7.0 software (GraphPad, USA) was used for statistical analysis. Kruskal-Wallis test was used to compare the results from RNA deconvolution of immune compositions. Mann-Whitney test was used to compare the CSF1 plasma levels. Simple linear regression was used to assess the relationship between CSF1R and cell marker expressions. Otherwise, data were analyzed by 2-tailed Student’s t test for the mice experiments. Values were considered statistically significant if P < .05. The error bars in the figures represent the standard error of the mean (SEM).

Results

RNA Deconvolution Reveals a Myeloid-Enriched Immune Landscape in Human Meningiomas

To investigate the immunological landscape of human meningiomas, we used the Yale Brain Tumor Database to examine 201 cases of meningiomas and 22 cases of normal meninges that had undergone whole-exome sequencing with RNA expression analysis.^17,18 CIBERSORTx was used to estimate the fraction of 22 immune cell types by in silico RNA deconvolution. Overall, normal meninges are enriched with monocytes and M2 macrophages in comparison with other tumor-infiltrating hematopoietic cells (Figure 1A). These cells far outnumbered M1 macrophages. Interestingly, several subsets of human meningiomas, such as those with SMARCB1 (±NF2) mutation or NF2 and/or chromosome 12 loss, have more abundant monocytes whereas the majority of meningioma had increased M2 macrophage (Figure 1A). Resting memory CD4⁺ T cells were more abundant than CD8⁺ T cells within meningiomas (Figure 1A). We observed that meningiomas with NF2/chr22 loss (P < .001) or PI3K (±TRAF7) mutations (P < .0001) had more abundant M2 macrophages when compared to normal meninges (Figure 1B, C). Except in meningiomas with KLF4 (±TRAF7), PI3K (±TRAF7), or TRAF7 mutations, the overall abundance of CD8⁺ T cells is increased when compared to normal meninges (Figure 1D). Therefore, the majority of human meningioma has elevated infiltration of M2 macrophages compared to normal meninges.

Fig. 1.

In silico dissection of immunological landscape in human meningiomas. (A) Heatmap of the relative proportion of different immune populations in human meningiomas. Relative proportion of monocytes (B), M2 macrophages (C), and CD8 T cells (D) in human meningiomas among various genetic subtypes. *P < .05, **P < .01, ***P < .001, ****P < .0001.

Murine MGS1 Model of Human Malignant Meningioma

To date, there have been no preclinical meningioma models to evaluate the mechanisms and efficacy of anti-PD therapy. We utilized the MGS1 murine malignant meningioma cell line to establish an immune-competent preclinical model for immunotherapeutic testing.¹⁵ This cell line was generated via Nf2 and Cdkn2ab inactivation that is common in high-grade human meningiomas.²² Syngeneic FVB mice were heterotopically injected in the subcutaneous space overlying the flank with MGS1 cells to allow for tumor volume measurements (Supplementary Figure 1). Using immunohistochemistry, we found MGS1 tumors were highly infiltrated with myeloid cells (CD11b⁺, F4/80⁺, or CD68⁺) with very little CD3⁺ T-cell infiltration (Figure 2A–D). There was faint staining of PD-L1 in the tumor tissues (Figure 2E, F), despite the upregulation of PD-L1 expression on the surface of MGS1 cells in response to IFN-γ (interferon gamma) stimulation (Figure 2G). This suggests that MGS1 tumors did not upregulate PD-L1 expression in vivo due to the paucity of T-cell infiltration. The minimal amount of T-cell infiltration and abundant myeloid cell infiltration of murine meningiomas similarly reflects the immune landscape within the human meningioma TME as suggested by our RNA deconvolution results. Overall, the immune profiling of murine MGS1 meningioma TME shows a similar phenotype of human meningioma and demonstrates a tumor immunity in the microenvironment (TIME) type 1 (PD-L1⁻/TIL⁻) or type 4 (PD-L1⁺/TIL⁻), which predicts poor response to anti-PD therapy.²³

Fig. 2.

MGS1 as an immunocompetent murine model of malignant meningiomas. (A–E) Immunohistochemistry demonstrating immune infiltration in MGS1 tumors. (F) PD-L1-positive control using human placenta. (G) In vitro stimulation of MGS1 with IFN-γ. ***P < .001. (H) Treatment schema of MGS1 model treated with anti-4-1BB and anti-PD-1 antibodies. (I) Tumor growth after anti-4-1BB and anti-PD-1 antibody treatments. Data are mean ± SEM (n = 5 mice per group).

We tested whether this murine model would respond to established T-cell–focused immunotherapies. We henceforth targeted 2 common immunotherapeutic pathways in this murine model. We administrated agonistic anti-4-1BB mAb (clone 2A) as an immune-enhancer and antagonist anti-PD-1 mAb (clone G4) (Figure 2H).^24,25 These antibodies have previously been reported as effective immunotherapies in several murine tumor models.^24,26 However, mice bearing MGS1 tumors did not respond to either single-agent or the combination of agonistic 4-1BB and antagonist PD-1 therapy (Figure 2I), suggesting the T-cell–based immune-boosting therapies may be ineffective for meningiomas.

Targeting the CSF1/CSF1R Axis Reduces MGS1 Tumor Growth

The immune landscape of both human and murine meningiomas is predominantly comprised of monocytes/M2 macrophages, implicating that the abundance of these cells is crucial to meningioma growth and progression.²⁷ CSF1 and CSF1R have been shown to be key mediators in monocyte recruitment and are associated with M2-phenotype differentiation.²⁸ We hypothesized that blockade of CSF1 and CSF1R interaction may be an effective treatment strategy for meningiomas. Treatment with neutralizing antibodies targeting CSF1 (clone 5A1) or CSF1R (clone AFS98) significantly reduced MGS1 tumor growth when compared to isotype control (Figure 3A–D). Delaying treatment until tumors were established for more than 1 month with anti-CSF1R mAb also reduced the growth of murine meningiomas (Figure 3E, F). We did not detect any CSF1R expression in the MGS1 cell line in vitro, which made the possibility of direct antibody-mediate cytotoxicity less likely (Supplementary Figure 2). The addition of anti-PD-1 therapy did not yield additional benefits (Supplementary Figure 3). To test whether the treatment effect of CSF1/CSF1R blockade is immunologically mediated by CD8⁺ T cells, we depleted CD8⁺ T cells prior to tumor inoculation, which was confirmed via flow cytometry (Supplementary Figure 4). The therapeutic effect of CSF1R-blockade was only partially abrogated by CD8⁺ cell depletion (Figure 3G), indicating that immune cells other than CD8⁺ T cells may also contribute to anti-tumor immunity. The efficacy of this treatment was confirmed in an orthotopic manner, in which anti-CSF1R blockade, as opposed to anti-PD1 blockade, significantly increased the survival of mice injected intracranially with MGS1 (Figure 3H).

Fig. 3.

Treatment of MGS1 with anti-CSF1/CSF1R antibodies. MGS1 tumors were treated with anti-CSF1 (A and B) and anti-CSF1R (C and D) antibodies. (E and F) Treatment of MGS1 tumors with anti-CSF1R antibody after 30 days of tumor growth. (G) Treatment of MGS1 tumors with anti-CSF1R antibody with CD8 depletion. (H) Orthotopic injection of MGS1 and treatment with anti-CSF1R antibodies. Data are mean ± SEM. *P < .05, ***P < .001, ****P < .0001.

CSF1 Blockade of Meningiomas Results in Global Transcriptomic Changes

To further explore the mechanisms of anti-CSF1 therapy, we isolated total RNA from MGS1 tumors treated with anti-CSF1 mAb at 3 weeks’ post-tumor inoculation (2 weeks’ posttreatment) and performed RNA-seq. Pathway enrichment analysis revealed differentially expressed genes and cellular programs in tumors treated with anti-CSF1 mAb (Figure 4A). Interestingly, CSF1 mAb treatment resulted in significant downregulation of myeloid-associated genes, including CSF1R, CCR1, C1qa, and CX3CR1 in comparison with the control (Supplementary Figures 5 and6). In addition, CSF1 neutralization resulted in the upregulation of gene sets involved in negative regulation of cell migration/proliferation and extracellular matrix organization. (Figure 4B, C) (Supplementary Figures 5 and 6). These data suggest that CSF1 neutralization results in profound transcriptional changes within the meningioma TME leading to active immune responses as well as changes in tissue architecture.

Fig. 4.

RNA-seq transcriptomic analysis of anti-CSF1 monoclonal antibody treatment in vivo. (A) Heatmap demonstrating genes that are significantly differentially expressed (FDR < 0.05 and absolute value of Log2 fold-change > 1). Control (green) and treatment (purple) cluster together with genes clustering into 2 major subsets: upregulated genes (red) and downregulated genes (red). (B) Bubble plot of significantly differentially expressed genes with term enrichment using Metascape, with “Member” terms from downregulated genes (blue) and terms in upregulated genes (red). Size of bubble corresponds to the number of genes enriched in results from a term and the ratio (x-axis) represents the number of genes enriched in RNA-seq results divided by the number of genes in a given term and color corresponds to the “Summary” term of each “Member” term bubble. Log-q value represents the adjusted P value for each term. (C) Pathway “Summary” terms enriched using Metascape of significantly downregulated (left/blue) and “Summary” terms describing upregulated genes (right/red).

CSF1/CSF1R Blockade Selectively Alters Myeloid Cell Phenotypes

We sought to examine changes in the immune microenvironment in MGS1 tumors following CSF1/CSF1R-blockade using mass cytometry (CyTOF), a single cell proteome tool. In our study, 36 mAbs were included for the analysis with ~1/2 of them specific for cell lineage markers and others for functional status. Despite effective anti-tumor immunity, CSF1/CSF1R-blockade did not significantly affect the numbers of lymphocytic populations infiltrating the tumor although a general trend showed elevation of these lymphoid cells (Figure 5A). In contrast, CSF1/CSF1R-blockade resulted in a significant decrease in CD11b⁺ F4/80⁺ cells infiltrating tumors (Figure 5B). We further analyzed the changes of myeloid cell subsets within the TME in response to CSF1/CSF1R-blockade and found that a Ly6C^lowMHCII^low subset decreased significantly whereas other subsets are not (Figure 5C).

Fig. 5.

Mass cytometry of tumor-infiltrating leukocytes. (A) Lymphocytic populations. (B) Depletion of CD11b⁺F4/80⁺ TAMs. (C) Different populations of Ly6c and MHC class II myeloid cells in the CD11b compartment. (D) t-distributed stochastic neighbor embedding (t-SNE) plot of tumor-infiltrating leukocytes overlaid with color-coded clusters. (E) Heatmap displaying normalized marker expression of each immune cluster. (F) Frequency of clusters of indicated immune cell subsets. Data are mean ± SEM (n = 5 mice per group). *P < .05, **P < .01, ***P < .001.

We then utilized an unsupervised clustering method to further interrogate changes in the infiltrating immune cells in the TME with anti-CSF1R therapy.²⁹ Based on a set of lineage markers and the functional status of these myeloid cells, the exploratory phenograph clustering analysis revealed 22 distinct cell subsets or clusters, with 11 myeloid and 6 lymphocytic clusters (Figure 5D, E). Among myeloid cell clusters, there were multiple clusters that were F4-80⁺, but not all experienced changes after the anti-CSF1R mAb treatment (Figure 5E). There were significant decreases in Clusters #1-3 (CD11b⁺ F4-80⁺) after treatment with anti-CSF1R mAb (Figure 5F). These clusters are characterized by a high expression of CD44, CCR6, CD80, suggesting that these are macrophages with a prominent TME homing capacity. There was a significant decrease in cluster #6, which is another F4-80⁺ TAM cluster similar to Cluster #3, but with low expression of CTLA-4, we observed an increase in Cluster #9, which comprised less than 1% of CD45⁺ cells, that is high in CD8a, CD127, GITR, Ly-6G, CD11c, and PD-L1, suggestive of a pro-inflammatory dendritic cell population. Our data suggest that anti-CSF1R mAb treatment selectively modulates subsets of macrophages and dendritic cells to improve immune function.

Among the lymphocytic clusters, we noted an increasing trend in Cluster #12, which is a CD8⁺ T-cell population (Figure 5F). This T-cell cluster demonstrates increases in activation markers, such as GITR, PD-1, CD69, FAS, and CTLA-4. There was a significant increase in Cluster #14, which is a Treg population that is characterized by the expressions of CD25 and FoxP3 (Figure 5F). We observed a significant increase in Cluster #15, which is CD3⁻CD161⁺, suggestive of natural killer (NK) cells. Lastly, there was an increasing trend of cluster #17, which is CD3⁺ but negative for both CD4⁻ and CD8⁻, suggesting a population of double-negative T cells that had infiltrated the meningiomas.

Elevated Plasma CSF1 Levels and Intratumoral CSF1R Expression in Human Meningiomas

We collected plasma from healthy subjects and patients with meningiomas at the time of their surgeries. Patients (n = 29) with meningiomas have significantly higher levels of CSF1 concentration in their plasma as compared to healthy controls (n = 5) (Figure 6A). Furthermore, there was a trend toward higher CSF1 levels in the atypical/anaplastic group (n = 11) compared to benign group of meningiomas (n = 18) (Figure 6B), suggesting that CSF1 may be more highly expressed in malignant meningiomas.

Fig. 6.

Patient plasma CSF1 levels and CSF1R expression in human meningiomas. Plasma CSF1 concentrations in (A) normal humans and patients and (B) among different grades of human meningiomas. (C) QIF score distribution of CSF1R expression in tissue microarray of human meningiomas. (D) Representative images of formalin-fixed paraffin-embedded tissue sections with positive CSF1R staining with tumor mask (MUC1), nuclei staining (DAPI, blue), CSF1R (red). Scale bars, 100 µm. (E) Correlation of CSF1R QIF signal in CD163 mask and DAPI. (F) Representative figure showing colocalization of CSF1R expression on CD163⁺ cells. *P < .05.

Next, we performed QIF to evaluate the expression of CSF1R in human meningiomas after constructing a TMA with human meningioma samples (n = 74). In our TMA, CSF1R is found to be significantly upregulated above the cutoff level in all but 2 cases of human meningioma (97%) and is minimal in normal meninges (Supplementary Figure 7). Expression of CSF1R appears to colocalize well with CD163⁺ TAM cells but not with meningioma cells (MUC1⁺) (Figure 6D–F). There is a strong correlation between CSF1R expression in DAPI⁺ cells and CSF1R expression in CD163⁺ TAMs, suggesting that most of the CSF1R signal originates from CD163⁺ TAMs (Figure 6E). Our results thus support that CSF1/CSF1R pathway is frequently upregulated in human meningioma and could be potential targets for the therapy.

Discussion

In this report, we show that the majority of immune infiltrates in meningiomas are M2 macrophages, which are known to provide stromal support for tumor growth and immunosuppression in cancer and scant T-cell infiltration.^30,31 Using a novel immunocompetent preclinical model, we found that malignant meningiomas did not respond to anti-PD therapy. We showed that the selective depletion of TAM populations using anti-CSF1/CSF1R mAb successfully abrogated the growth of murine meningiomas. More importantly, our results indicated that TAMs in meningioma play an integral in the growth of these tumors.

The development of effective therapies in human meningiomas has been hindered by the lack of suitable, immune-competent preclinical models. Peyre et al. developed a mouse meningioma cell line (MGS1) mimicking grade II/III meningiomas by neonatal arachnoidal cell inactivation Nf2 and Cdkn2ab in FVB mice, both genes known to concomitantly result in malignant meningiomas in humans.¹⁵ MGS1 cells harbor 4q loss that corresponded to 1p loss in humans and 19q loss to 14q loss in humans, both aberrations being characteristic of grade-II and -III human meningiomas.¹⁵ This presented a unique preclinical model for testing immunotherapies for meningiomas. MGS1 tumors did not respond well to anti-PD and anti-4-1BB therapies, both thought to mainly act on effector T cells, as could be predicted due to these tumors being devoid of T-cell infiltration which represents type 1 or type 4 tumors based on the TIME classification.²³ Human cancers that lack T-cell infiltration do not respond well to anti-PD therapy.²³ Furthermore, it is unclear whether there are underlying genetic or transcriptomic regulators for PD-L1 expression in meningiomas, if present, as is the case found in gliomas and other tumors.^32,33 Reports on PD-L1 expressions in human meningioma are also highly variable.^34–37 We previously found that meningiomas are heavily infiltrated by CD68⁺ macrophage populations with and without CD163 expression, which may be the source of PD-L1 expression instead of meningioma cells, and this was consistent between benign and atypical/anaplastic meningiomas.^10,36 We hypothesized that targeting the M2 macrophages that comprise up to 40% of the immune composition in these tumors may present an approach to normalize the immune response. CSF1 is responsible for proliferation, chemotaxis, and reversible monocytic differentiation and its receptor CSF1R is important in the maintenance of TAMs,^38,39 Our studies demonstrate that CSF1 as well as CSF1R are broadly upregulated in human meningioma (Figure 6). In this context, we decided to target this pathway in meningiomas.

Antibodies targeting CSF1 and CSF1R demonstrated striking responses in suppressing MGS1 tumor growth in immune-competent syngeneic mice. The therapeutic effect of anti-CSF1R mAbs may only be partially mediated through CD8⁺ T cells as CD8⁺ depletion did not reverse all the therapeutic effects, suggesting other non–T-cell-dependent therapeutic effects with CSF1R blockade in these tumors. In our model, RNA-seq of MGS1 tumors after anti-CSF1 mAb treatment suggests that there are many other global changes in stromal-related gene sets, some of which may be the end result of the treatment as the tumor growth is stunted. Mass cytometry revealed a specific decrease in MHCII^low TAMs, which tend to express M2-associated genes such as Arg1 (arginase-1), Cd163, Stab1 (stabilin-1), and Mrc1 (MMR), relative to MHCII^high TAMs in the monocytic compartment.⁴⁰ The change in MHCII^low to MHCII^high TAM ratio is consistent with a report of M2 to M1 phenotype conversion after anti-CSF1/CSF1R treatment.⁸ Further in-depth unbiased, hierarchical clustering revealed significant decreases in multiple populations of F4-80⁺ macrophages, and surprisingly, a significant increase in pro-inflammatory dendritic cells. The change into a more inflammatory environment resulted in increase in NK cell infiltration, a modest increase in CD8⁺ T cells, and likely compensatory increase in Treg, which has been reported previously.⁴¹ Therefore, anti-CSF1/CSF1R treatment may mainly modulate myeloid cells to improve immune/inflammatory responses in the TME in this model.

Interestingly, the addition of anti-PD-1 therapy did not have additional benefit in this murine meningioma model, unlike other murine cancers.^8,42 Combined together with the initial observation that anti-PD therapy alone had no therapeutic effects in this model (Figure 2) suggests that the PD-1/PD-L1 axis is not a major mechanism of immune resistance in this meningioma model. Since human meningiomas and our preclinical murine meningiomas share a similar TME raises the question whether ongoing anti-PD therapy clinical trials will demonstrate any therapeutic efficacy in human meningiomas. In contrast, our patient data demonstrate a significant increase in plasma CSF1 level, even in patients with grade 1 meningiomas, suggesting that the CSF1/CSF1R axis plays a role in human meningiomas. In addition, CSF1R is avidly expressed in these human tumors. The lack of efficacy of anti-CSF1R therapies in many solid tumors thus far may be confounded by tumor-type specifics of TAM, and it is unclear whether certain patients or tumor types are more likely to respond to CSF1R inhibition.⁴³ It is also unclear which type of TME would be most suited or resistant for this therapy.^23,44 Unlike melanoma and non–small cell lung cancer, where there is relative abundance of lymphocytes over myeloid cells, meningiomas may respond to anti-CSF1/CSF1R-directed therapies as the therapeutic effects may not be act via a CD8⁺ T-cell–mediated mechanism.^5–7

It is unclear whether CSF1 is released by the human tumors or whether CSF1 can originate from other immune cells.^42,45 In fact, data suggest that anti-CSF1 blockade could suppress CSF1-negative tumors, suggesting the important role of stromal support for tumor growth.⁴⁶ In addition to CSF1, IL-34 could also work via CSF1R as a receptor.⁴⁷ Nonetheless, neutralization of CSF1 in our murine model demonstrated similar treatment efficacy as targeting CSF1R, suggesting IL-34 plays a minimal role in this model. Whether or not IL-34 participates in the immune suppression in human meningioma will need to be tested in future studies. The changes in global genetic expression in our RNA-seq experiment may be indirect results of anti-CSF1 treatment and the results could vary due to the timing of sequencing analysis relative to the initiation of treatment, but the results suggest many nonimmune-related gene sets at play. Paradoxically, the downregulation in gene sets involved in “inflammatory response” may represent decreased chemotactic recruitment of monocytes, a function, if present, that may paradoxically aid in immunosuppression.⁴⁸ These findings implicate other underlying myeloid functions outside of the directed immunosuppression of cytotoxic T cells and the underlying mechanisms including immune- and nonimmune-mediated remain to be explored.

In summary, we utilized a rational approach to dissecting the immunological landscape of human meningiomas and devised a rational strategy by focusing on the abundance of M2 macrophages in these tumors by targeting the CSF1/CSF1R axis. Our data using a preclinical immunocompetent murine model suggests that targeting the CSF1/CSF1R axis may be efficacious in human meningiomas due to an abundance of myeloid cells and low T-cell infiltration. The current data provide a strong rationale for future human clinical trials targeting CSF1/CSF1R in human malignant meningiomas for which there are currently no available therapies.

Funding

This study is partially supported by the Neurosurgery Research and Education Foundation Tumor Research Fellowship (J.Y.), the National Institutes of Health grant P30CA046934, and an endowment from United Technologies Corporation.