Cell lines and patient-derived GSC cultures
Human GBM samples were obtained from patients who provided written informed consent for tissue collection and use in research and publication, under the approval of the Hamilton Health Sciences and McMaster Health Sciences Research Ethics Board (no. 16078). Tumour samples were processed using established protocols60. GBM cells were cultured in Neurocult Complete (NCC) medium, a commercially available serum-free NSC medium (STEMCELL Technologies, 05751), supplemented with human recombinant epidermal growth factor (20 ng ml−1; STEMCELL Technologies, 78006), basic fibroblast growth factor (20 ng ml−1; STEMCELL Technologies, 78006), heparin (2 μg ml−1; STEMCELL Technologies, 07980) and antibiotic–antimycotic solution (1×; Wisent, 450-115-EL). NSCs were cultured and maintained using similar protocols as described previously61. SK-MEL-2 cells, MDA-MB-231 and HEK293T cells were purchased from the American Type Culture Collection (ATCC) and grown in Dulbecco’s Modified Eagle Medium (DMEM) supplemented with 10% FBS and 1% non-essential amino acids (NEAA; Thermo Fisher, 11140050). NHAs were purchased from ATCC and grown in DMEM/F12 medium (Gibco, 11320033) supplemented with 10% FBS, and epidermal growth factor and fibroblast growth factor as above.
Animal studies
All animal experiments were conducted in compliance with the ethical guidelines approved by the Animal Use Protocols (22-12-38) of the McMaster University Central Animal Facility. Mice were housed in pathogen-free, temperature-controlled, 12 h light and dark cycle environment and were fed ad libitum. Intracranial injections were performed on 6–12-week-old NOD/SCID gamma (NSG) or C57BL/6 mice, following previously described methods60. GBM4, GBM8 and MBT06 cell lines (106 cells per mouse) or GL261 cells (105 cells per mouse) were suspended in 10 μl PBS and injected into the right frontal lobe using a Hamilton syringe (Hamilton, 7635-01). A burr hole, 2 mm posterior to the coronal suture and 3 mm lateral to the sagittal suture, was drilled for intracranial access60. At the humane end-point, mice were euthanized, perfused with 10% formalin, and the brains were sectioned into 2 mm slices using a brain matrix for paraffin embedding and haematoxylin and eosin (H&E) staining. Digital images were captured using an Aperio Slide Scanner (Leica Biosystems) and analysed with ImageScope v11.1.2.760 software. Kaplan–Meier survival analyses were based on the time from surgery to end-point. For humanized studies, NOG-EXL mice (Taconic, NOD.Cg-PrkdcscidIl2rgtm1SugTg(SV40/HTLV-IL3,CSF2)10-7Jic/JicTac, 13395-F) were purchased and utilized as outlined above. Intracranial injections followed the previously described protocol for the BT972 cell line (106 cells per mouse). Mice were monitored post-injections, and survival data were analysed using the Kaplan–Meier method.
RNA-seq and differential gene expression ranking
Cells were subjected to RNA extraction, followed by RNA sequencing on the Illumina HiSeq 2500 platform. Four comparative analyses were performed: (1) GBM CD133+ versus CD133−; (2) NSC CD133+ versus CD133−; (3) CD133+ NSC versus CD133+ GBM; and (4) CD133− NSC versus CD133− GBM. Data filtering was performed using a CPM threshold of 3.5. Multidimensional scaling plots indicated clear sample separation across all comparisons. A smear plot analysis confirmed consistent gene expression patterns without significant artifacts at this CPM threshold. DEGs were identified using the quasi-likelihood F-test in edgeR, chosen due to its stringency and appropriateness for datasets with minimal sample sizes (n = 4, with ≥2 per group). Genes were ranked on the basis of P values and fold changes using the formula: ranking score = sign(log(fold change)) × –log10(P value), where the sign(log(fold change)) indicates the direction of expression change (positive for upregulation, negative for downregulation), and –log10(P value) reflects the significance level. Genes were ranked from highest upregulation to highest downregulation, with rankings exported as.rnk files for GSEA.
GSEA and enrichment mapping
GSEA was conducted for all four comparisons using the.rnk files and the Human_GOBP_AllPathways_with_GO_iea_December_24_2015_symbol.gmt gene set. A total of 1,000 permutations were performed using a random seed of 349. Comparative GSEA results for the four analyses were compiled in Pathway.xlsx, including normalized enrichment scores and false discovery rate (FDR) q-values. Differences in gene filtering between the comparisons necessitated reanalysis of the RNA-seq data using only protein-coding genes, allowing consistent gene sets across all analyses. Owing to high similarity between: (1) CD133+ versus CD133− in NSC (GSEA2) and GBM (GSEA1); and (2) NSC versus GBM in CD133+ (GSEA3) and CD133− (GSEA4), combined enrichment maps were generated (map A: GSEA1 and GSEA2; map B: GSEA3 and GSEA4). These maps were constructed using a Jaccard coefficient of 0.25 (for edges) and an FDR q-value cutoff of 0.0001 (for nodes). Cytoscape files (.cys) containing both stringent and relaxed conditions (FDR q-value < 0.1, P value < 0.05) are available. To ensure direct comparability across the GSEA analyses,.rnk files were regenerated using all protein-coding genes without CPM filtering. z-scores were calculated for each pathway on the basis of the direction of enrichment and nominal P values. Pathways with significant differences (P < 0.01) were identified, and corresponding enrichment maps were generated.
Interrogation of public databases
Publicly available datasets were incorporated using the GEPIA2 interface to compare gene expression correlations between GBM samples from TCGA and normal brain tissue from the GTEx portal, following established methods62.
Flow cytometry
Cells were dissociated using TrypLE (Thermo Fisher, 12605010), resuspended in PBS containing 2 mM EDTA (Thermo Fisher, AM9260G) and stained with anti-human GPNMB-PE (Invitrogen, HOST5DS, 12-9838-42, 1:20) or anti-mouse GPNMB-eFluor 660 (Invitrogen, CTSREVL, 50-5708-82, 1:20) on ice for 45 min. Following washes, stained cells were analysed on a CytoFLEX flow cytometer (Beckman Coulter), with dead cells excluded using 7-AAD viability dye (1:100 dilution; Beckman Coulter, A07704). Compensation was performed using mouse IgG CompBeads (BD Biosciences, 552843). Cells were gated on unstained controls.
Immunohistochemistry
Formalin-fixed mouse brains were sectioned coronally into 6–7 slices, then paraffin-embedded and mounted onto slides. Slides were deparaffinized in xylene (2 × 5 min) and rehydrated through graded ethanol (100% ethanol, 2 × 3 min; 95% ethanol, 1 × 3 min; 70% ethanol, 1 × 3 min). Antigen retrieval was performed in Tris-based antigen unmasking solution (Vector Laboratories, H-3302-251; 1:100), slides were heated in a microwave, then quenched by incubation in 10% methanol dissolved in H2O2 for 10 min. Slides were blocked for 1 h at room temperature on a shaker using 3% milk, 1% BSA in TBS. Primary antibodies were diluted in blocking buffer and applied overnight at 4 °C (rabbit anti-CD133, D2V8Q, CST, 64326 T, 1:200; goat anti-hGPNMB, R&D AF2550, 3:500). HRP-conjugated secondary antibody was prepared as a 1:1 dilution of secondary reagent (Donkey Anti-Rabbit IgG HRP, ab205722; Donkey Anti-Goat IgG HRP, Abcam, ab214881) and applied for 2 h at room temperature, followed by development with DAB (Vector Laboratories, SK-4100). Slides were counterstained with haematoxylin, blued in ammonia water (20 s) dehydrated through graded ethanol, and cover slipped with xylene-based mounting medium. Slides were cured overnight prior to imaging or whole-slide scanning with a Leica Aperio Scanscope XT. Images were processed and positive cells quantified with QuPath63. Tumour-burdened areas were marked using H&E sections across all coronal sections, and cells were quantified in all regions of interest.
Immunofluorescence
Tissue sections were deparaffinized and rehydrated as with IHC. Antigen retrieval was performed in Tris–EDTA buffer (10 mM, pH 9.0) supplemented with 0.05% Tween-20 by microwaving (3 min on high, then 15 min on low) with slides fully submerged, followed by cooling at room temperature for 30 min. Slides were permeabilized in 0.05% TBS-T (20 min). Tissue was circumscribed with a hydrophobic barrier pen and blocked with CAS-Block (Invitrogen, 00-8120) for 10 min at room temperature in a humidified chamber. Primary antibodies diluted in blocking solution were applied overnight at 4 °C (rabbit anti-IBA1, Abcam EPR16588, 1:1,000; goat anti-mGPNMB, R&D AF2330, 1:500, goat anti-hGPNMB, R&D AF2550, 3:500). Fluorophore-conjugated secondary antibodies (donkey anti-rabbit AF488, Invitrogen, A-21206; donkey anti-goat AF594, Invitrogen, A-11058, chicken anti-goat AF647, Invitrogen, A-21469) diluted 1:500 in blocking solution were applied for 2 h at room temperature. Slides were mounted with DAPI aqueous mounting medium (Abcam, ab104139). Slides were imaged with a BioTek Cytation 5 Cell Imaging Multimode Reader at 20× objective, with brightness and contrast set in comparison to a 2° antibody-only condition. Images were processed and saved using ImageJ64. Whole-brain slices were scanned using a ZEISS Axioscan 7 at 20× objective at the McMaster Centre for Advanced Light Microscopy. Images were processed equally using ZEISS ZEN Lite.
Generation of gene-knockout constructs
sgRNAs targeting AAVS1 (GGGGCCACTAGGGACAGGAT) and human GPNMB (KO-A: AATGATGGTACAGACCTCCG, KO-B: AGGAATCCTACTCAGCTCCA) mouse Gpnmb (GAAAGUCUCUGCGGGGUCCU, AAAGGGCCUGGCCCAUCAUU and UCACGCUUGGCAGCCUGGAG) were obtained from the TKOv3 library65 and cloned into lentiCRISPRv2 constructs (Addgene #52961). Successful ligations were packaged independently into lentiviruses using second-generation packaging constructs. In brief, HEK293T cells were seeded at 1.5 × 107 cells per T75 flask and incubated overnight in high-glucose DMEM medium with 2 mM L-glutamine and 1 mM sodium pyruvate (Thermo Fisher,11995065), supplemented with 1% non-essential amino acid solution (Thermo Fisher, 11140050) and 10% fetal bovine serum (Gibco, 12483020). The following day, the HEK293T medium was replaced with viral harvesting medium consisting of HEK medium supplemented with 10 mM HEPES (Thermo Fisher, 15630080) and 1 mM sodium butyrate (Sigma Aldrich, 303410). Transfection was conducted with a mixture of pMD2.G (VSV-g; 2.6 μg; Addgene #12259), pRSV-REV (REV; 2.6 μg; Addgene #12253), pMDLG/RLE (gag/pol; 5.3 μg; Addgene #12251), transfer plasmid (10.6 μg), polyethylenimine (63.5 μg; Sigma Aldrich, 408719) in 1.3 ml of Opti-MEM. After incubating for 15 min at room temperature, PEI/DNA mixture was added dropwise to T75 flasks. Viral supernatants were collected 48 h after transfection and then concentrated using ultracentrifugation (20,000 rpm for 2 h at 4 °C) before being snap-frozen and stored at −80 °C.
Cell proliferation assays
Single cells were plated in 96-well plates at 1,000 cells per well in 200 μl of medium and incubated at 37 °C, 5% CO2 for 5 days. Presto Blue (20 μl; Thermo Fisher, A13262) was added 4 h prior to fluorescence readout using a FLUOstar Omega Microplate Reader with 544 nm excitation and 590 nm emission wavelengths. Proliferation was calculated by subtracting the background fluorescence from medium-only control wells.
Western blotting
Cell pellets were lysed in RIPA buffer (50 mM Tris-HCl pH 8.0, 150 mM NaCl, 1% NP-40, 0.5% sodium deoxycholate, 0.1% SDS) supplemented with HALT protease/phosphatase inhibitor cocktail (Thermo Fisher, 78440) for 30 min at 4 °C. Lysates were clarified by centrifugation (14,000g, 15 min, 4 °C), and protein concentration was determined by Bradford assay (Bio-Rad). Equal amounts of protein were resolved on 4–12% Bis-Tris gradient gels (Novex, Invitrogen) and transferred to PVDF membranes (Immobilon-P, Millipore) at 200 mA for 2 h. Membranes were blocked in 5% BSA for 1 h at room temperature and incubated with primary antibodies overnight at 4 °C with gentle agitation. Following TBS-T washes, membranes were incubated with HRP-conjugated secondary antibodies diluted in 5% BSA for 1 h at room temperature. Blots were developed using Clarity Western ECL substrate (Bio-Rad, 1705061), imaged on a ChemiDoc MP imaging system (Bio-Rad), and quantified using Image Lab software.
Primary antibodies were used as follows: human GPNMB (Cell Signaling Technology, E4D7P, 38313; 1:1,000), mouse GPNMB (Abcam, EPR18226-147, ab188222; 1:1,000), β-actin (Cell Signaling Technology, 13E5, 4970; 1:5,000) and GAPDH (Cell Signaling Technology, 14C10, 2118; 1:1,000). Secondary antibodies were goat anti-mouse IgG (H+L)–HRP (Bio-Rad, 1706516; 1:3,000) and goat anti-rabbit IgG (H+L)-HRP (Bio-Rad, 1706515; 1:10,000).
Generation of CAR-T cells
CAR constructs were packaged into lentiviral vectors as described above. Viral supernatants were collected 48 h post-transfection, concentrated by ultracentrifugation at 20,000 rpm for 2 h at 4 °C, reconstituted in 50 μl of ImmunoCult-XF T Cell Expansion Medium (StemCell, 10981), and stored at −80 °C. Peripheral blood mononuclear cells (PBMCs) were isolated from consenting healthy donors and cultured in ImmunoCult-XF medium, activated with ImmunoCult Human CD3/CD28/CD2 T Cell Activator (25 μl ml−1; StemCell, 10970) in 96-well roundbottom plates. Sixteen hours following activation, T cells were transduced with 15 μl of CAR lentivirus. T cells were expanded in ImmunoCult-XF medium with 2.5 μg ml−1 recombinant human IL-2 (StemCell, 78036) in 24-well plates for 14 days post-activation before plating assays. Transduction efficiency was determined using flow cytometry for GFP expression 7 days post-transduction.
CAR-T cytotoxicity assay
Ffluc-expressing target cells were plated at 1 × 104 cells per well in 96-well plates containing 100 μl of NCC medium with D-firefly luciferin potassium salt (Revvity, 122799; 75μg ml−1). Effector T cells were added at various effector-to-target ratios and incubated at 37 °C for up to 48 h. Spontaneous lysis controls contained target cells without CAR-T cells, and maximal lysis controls were treated with 1% NP-40 (Thermo Fisher, 98379). Bioluminescence was measured periodically with a luminometer (FLUOstar Omega Fluorescence 566 Microplate Reader; BMG LABTECH) as relative luminescence units (RLU). Per cent viability was normalized to target cells cultured without effectors, and subtracted from 100% to attain percentage specific lysis.
CAR-T activation assays
Effector cells were co-cultured with 2.5 × 105 target cells at a 1:1 ratio for 24 h in a 24-well plate. T cells were stained with anti-human CD3 (PE-Cy7-conjugated; BD Biosciences, 563423, 1:20) and analysed for activation markers CD25 (BV421-conjugated anti-human CD25 antibody; clone M-A251; BioLegend, 356113, 1:20) and CD69 (APC-conjugated mouse anti-human CD69 antibody; BD Biosciences, 555533, 1:20) by flow cytometry as described above.
CAR-T cytokine release assays
Effector cells were co-cultured with 2.5 × 105 target cells at a 1:1 ratio for 24 h. Supernatants were collected and stored at −80 °C. The DuoSet ELISA human IFNγ (R&D Systems, DY285B) and human TNF (R&D Systems, DY210) kits were used for quantification of cytokines, according to the manufacturer’s descriptions.
Generation of murinized anti-GPNMB CAR-T cells
Mouse T cell activation and gamma retroviral production has been described previously66. In brief, single-cell suspensions were generated from freshly isolated mouse spleens. Spleens were mechanically dissociated and filtered through a 70-µm mesh and red bloods cells lysed (BioLegend, 420301). Cells (3 × 106 per ml) were stimulated with 0.1 µg ml−1 each of anti-mouse CD3 (BioLegend, 100339) and CD28 (BioLegend, 102115) in T cell medium (RPMI supplemented with 10% FBS, L-glutamine, HEPES, non-essential amino acids, sodium pyruvate, penicillin–streptomycin, β-mercaptoethanol and mouse IL-2 (BioLegend, 575402)) for 5–6 days, expanding T cells every 2 days.
Gamma-retroviruses were generated on Plat-E cells transfected with pCL-Eco (Addgene #12371) and retrovirus transfer vector pRV2011 containing Thy1.1 transduction marker and a 2nd generation CD28/CD3z mouse CAR with scFVs targeting either human HER2 or mGpnmb and a MYC tag. Viral supernatants were concentrated 36 h after transfection with 100-kDa centrifugal filters (Amicon) and added to activated T cells 24 h after T cell activation. Transduction efficiency was measured by flow cytometry using Thy1.1–APC-Fire (BioLegend, 202543) and AF647–MYC (BioLegend, 626809).
CAR-T in vivo trials
Ffluc-expressing GBM4 or GBM8 cells were intracranially implanted in NSG mice as described above. For GBM–macrophage co-inoculation, mice were injected with 1 × 105 GBM8-ffluc and 1 × 105 IL-4/IL-10/TGFβ-conditioned U937 cells (described below) in NSG mice. GL261-ffluc were intracranially implanted into C57BL/6 mice. Engraftment was confirmed by positive IVIS signal 2–3 days post-surgery, at which point mice were randomized to control or treatment arms. Mice received intracranial doses of 106 UTD or CAR-T cells in 5 μl of PBS through the same burr hole at the indicated time points. Mice were monitored weekly by IVIS until the last control mouse succumbed to disease.
Macrophage culturing and polarization
U937 cells were a gift from A. Rullo and cultured in RPMI medium (Gibco, 11875093) supplemented with 10% Fetal Bovine Serum (Gibco, 10082147). U937 monocytes were primed with 60 ng ml−1 PMA (Millipore Sigma, P1585) for 72 h, then treated with, 100 ng ml−1 IFNγ (STEMCELL Technologies, 78020), 20 ng ml−1 IL-4 (STEMCELL, 78045.1), 20 ng ml−1 IL-10 (STEMCELL, 78024), 20 ng ml−1 TGFβ (Gibco, 100-21) or conditioned medium collected and filtered from GBM8 cells, for 72 h as indicated. CAR-T cell cytotoxicity assays were performed as described above, with cells incubated in equal parts basal RPMI and Immunocult-XF medium (STEMCELL Technologies, 100-0956).
GSC–macrophage–T cell triple co-cultures
For triple co-cultures, iRFP-expressing GBM8 cells were co-cultured with U937 cells and UTD- or GPNMB CAR-T cells at a 1:1:1 ratio in a 24-well plate, in equal parts NCC, basal RPMI, and Immunocult-XF medium. Cells were collected after 24 h and analysed by flow cytometry as described above with mouse anti-CD45 (BV421-conjugated, BioLegend, 368521) and anti-CD3 (PE/Cy7-conjugated, BD Pharmingen, 563423) antibodies. Tumour cells were gated on iRFP positivity. T Cells were gated on CD45/CD3 dual positivity. U937 cells were gated on CD45 positivity and CD3 negativity. Live cells were quantified by 7-AAD dye negativity and were normalized to the UTD control.
scRNA-seq analysis
Patient samples and mouse brain tissue were immediately snap-frozen in liquid nitrogen after collection and stored at −80 °C for sci-RNA-seq3-based single-nucleus RNA-seq processing. Sci-RNA-seq3 libraries were generated as previously described using three-level combinatorial indexing67. After barcodes and UMIs were extracted from the read1 of FASTQ files, rawFASTQ files were aligned to the hg19genome using STAR aligner (STAR v2.5.2b) or with the mouse genome (mm10) and GencodevM12 gene annotations. For bioinformatic analyses, including preprocessing, clustering and annotation were performed as described17.
Automated multiplexed sequential immunofluorescence imaging
Automated hyperplex immunofluorescence staining and imaging was performed on formalin-fixed, paraffin-embedded or frozen sections using the COMET platform (Lunaphore Technologies). The multiplex panel included antibodies described in Supplementary Table 2. The 22-plex protocol was generated using the COMET Control Software, and reagents were loaded onto the COMET device to perform the seqIF protocol. All antibodies were validated using conventional IHC and/or immunofluorescence staining in conjunction with corresponding fluorophores and DAPI counterstain (Thermo Fisher Scientific). 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 fluorophore 555 (Thermo Fisher Scientific) and Alexa fluorophore 647 (Thermo Fisher Scientific) were used at 1/200 and 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 workflow was performed parallelized on a maximum of four simultaneous slides, with automated iterative cycles of two targets with primary and secondary antibodies’ staining at a TME, followed by imaging, and elution of the primary and secondary antibodies. No sample manipulation is required during the entire workflow. All reagents were diluted in Multistaining Buffer (BU06, Lunaphore Technologies). Elution step lasted 2 min for each cycle and was performed with Elution Buffer (BU07-L, Lunaphore Technologies) at 37 °C. Quenching step lasted for 30 sec and was performed with Quenching Buffer (BU08-L, Lunaphore Technologies). Incubation TME was set at 4 min for all primary antibodies and at 2 min for secondary antibodies. The imaging step was performed with Imaging Buffer (BU09, Lunaphore Technologies) with an integrated epifluorescence microscope at 20× magnification. Image registration was performed immediately after concluding the immunofluorescence staining and imaging procedures by COMET Control Software. Each seqIF protocol resulted in a multi-layer OME-TIFF file in which the imaging outputs from each cycle are stitched and aligned. COMET OME-TIFF files contain DAPI image, intrinsic tissue autofluorescence in TRITC and Cy5 channels, and a single fluorescent layer per marker. Markers were subsequently pseudocoloured for visualization of multiplexed antibodies using the HORIZON Viewer software. Spatial bioinformatic analysis and quantification were then conducted using the learning algorithm and phenoplex feature of the Visiopharm Software.
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
