Glioblastomas (GBM) are aggressive brain tumors with extensive intratumoral heterogeneity that contributes to treatment resistance. Spatial characterization of GBMs could provide insights into the role of the brain tumor microenvironment in regulating intratumoral heterogeneity. Here, we performed spatial transcriptomic and single-cell analyses of the mouse and human GBM microenvironment to dissect the impact of distinct anatomical regions of brains on GBM. In a syngeneic GBM mouse model, spatial transcriptomics revealed that numerous extracellular matrix (ECM) molecules, including biglycan, were elevated in areas infiltrated with brain tumor–initiating cells (BTIC). Single-cell RNA sequencing and single-cell assay for transposase-accessible chromatin using sequencing showed that ECM molecules were differentially expressed by GBM cells based on their differentiation and cellular programming phenotypes. Exogeneous biglycan or overexpression of biglycan resulted in a higher proliferation rate of BTICs, which was associated mechanistically with low-density lipoprotein receptor-related protein 6 (LRP6) binding and activation of the Wnt/β-catenin pathway. Biglycan-overexpressing BTICs developed into larger tumors and displayed mesenchymal phenotypes when implanted intracranially in mice. This study points to the spatial heterogeneity of ECM molecules in GBM and suggests that the biglycan–LRP6 axis could be a therapeutic target to curb tumor growth.

Significance:

Characterization of the spatial heterogeneity of glioblastoma identifies regulators of brain tumor–initiating cells and tumor growth that could serve as candidates for therapeutic interventions to improve the prognosis of patients.

Glioblastoma (GBM) is the most common primary malignant brain tumor in adults (1). The current standard of treatment consists of maximal safe resection, followed by radiation and chemotherapy with oral alkylating agent temozolomide (2). Despite these interventions, patients continue to have a median survival of fewer than 21 months (3). A significant factor contributing to the poor prognosis of patients with GBM is the extensive intratumoral heterogeneity in the tumor microenvironment.

Single-cell RNA sequencing (scRNA-seq) studies have identified cell populations displaying multiple cellular programs including: (i) neural progenitor cell—like (NPC-like), (ii) oligodendrocyte progenitor cell-like (OPC-like), (iii) astrocyte cell-like (AC-like), and (iv) mesenchymal-like (MES-like) state within the GBM microenvironment as described by several groups including our own (4–7). Although the cellular origin of these cellular programs is unknown, brain tumor–initiating cells (BTIC) are assumed to be responsible for driving the intratumoral heterogeneity (8). BTIC diversity is distributed along a transcriptional gradient spanning two major cellular programs: developmental and injury response signatures with differing degrees of proliferation and stemness capacity (9). Because of the loss of cell niche information after tissue dissociation in scRNA-seq, the underlying mechanisms driving diversity in BTIC phenotypes remain elusive.

Here, we applied spatial transcriptomics to a rodent model of GBM and used publicly accessible single-cell assay for transposase-accessible chromatin using sequencing (scATAC-seq) along with scRNA-seq data of human GBMs, and confocal protein imaging, to investigate the spatial heterogeneity of brain tumor microenvironment at a single-cell resolution.

Animal experimental models

All mice used in this study were on C57BL/6 background. Charles River provided 6- to 8-week-old female wild-type mice. All experiments were conducted with ethics approval from the Animal Care Committee at the University of Calgary under the regulations of the Canadian Council of Animal Care.

Human tissues

Human GBM tissues for IHC (n = 5; Supplementary Table S1) were obtained in compliance with methods authorized by Indiana University's Institutional Review Board. Fresh tissues were surgically extracted from patients, snap-frozen, and kept at −80°C. The tissues were fixed in formalin and histologically processed for paraffin-embedded blocks. The neuropathologist examined hematoxylin and eosin (H&E)–stained slides. To collect tumor biopsies, all patients gave written informed permission. AMSBIO provided normal human cerebral cortex for this study (catalog number HP-210).

Analysis of extracellular matrix gene expression in GBMs from The Cancer Genome Atlas

Gene expression data for The Cancer Genome Atlas (TCGA)-GBMs was downloaded with the TCGAbiolinks and recount2 R packages. Boxplots associating extracellular matrix (ECM) gene expression with the type of sample were generated with the ggbetweenstats function from the ggstatsplot R package by Indrajeet Patil. For correlation analysis between gene expression, datasets including TCGA GBM RNA-seq and Agilent 4502 were processed via online GBM RNA-seq analysis platform GLIOVIS (http://gliovis.bioinfo.cnio.es/10).

Generation of GBM patient-derived BTICs

Human BTIC lines were generated from resected tissues of patients with GBM and cultured, as previously reported (11–13). Within the University of Calgary BTIC Core, these lines were cultured chronologically, preserved, and verified. Identification of the three human BTIC lines (BT048, BT53M, and BT073) has been reported previously (12). Short tandem repeat analysis was performed for the authentication of cell lines. These lines were routinely tested for Mycoplasma using Venor GeM Mycoplasma Detection Kit (Sigma, MP0025).

Isolation and characterization of mouse BTICs

The mouse BTIC line mBT0309 was used to generate GBM in the C57BL/6 mice. The intracranial implantation of mBT0309 BTICs in mice recapitulated the key characteristics of a human World Health Organization grade 4 GBM (14). This line was generated previously from NPcis mice carrying mutations in Nf1 and Trp53 (14, 15).

Live cell imaging of BTIC growth

For the sphere formation assay, freshly dissociated cells were plated at 10,000 cells per well in 100 μL of BTIC medium into a 96-well flat bottom plate in the presence or absence of 10 μg/mL recombinant human biglycan (R&D Systems, 2667-CM-050). We monitored cell growth using a real-time cell imaging system (IncuCyte live-cell ESSEN BioScience Inc.). Images of the cells were taken within 48 to 72 hours. The resultant number of spheres above the 50-μm diameter cutoff, a convenient parameter to describe growth characteristics, was analyzed as previously described (16, 17). We considered spheres with diameters more than 100 μm as fused spheres and they were counted individually.

Intracranial tumor implantation

After sphere dissociation of firefly luciferase (FL)–expressing BTIC0309, 50,000 viable cells were resuspended in 2 μL of PBS and implanted stereotactically into the right striatum of mice as described previously (11). After implantation, we daily monitored animals to assess weight loss and physical/neurologic abnormalities.

In vivo bioluminescence imaging of mice

C57BL/6 mice were intraperitoneally injected with 150 mg/kg of D-luciferin, potassium salt (Gold Biotechnology, LUCK-1G) in DPBS (no calcium or magnesium) 10 minutes before bioluminescence imaging. Mice were anesthetized with isoflurane (2.5% vaporized in O2) and shaved to reduce signal attenuation caused by black hair. Imaging was conducted using Xenogen IVIS 200 system (Xenogen) (auto exposure time). Total photon flux (photons per second) was measured from a defined area of interest using Living Image software for analysis (Xenogen).

Preparation of libraries for spatial transcriptomics

The Visium Spatial Gene Expression platform (10x Genomics) was used for spatial transcriptomics analysis. Brain tissues were processed according to the manufacturer's instructions. After 40 days of tumor implantation, brain tissues were harvested and snap-frozen in liquid nitrogen. The frozen tissues were then embedded in FSC 22 Frozen Section Media (Leica) for further processing. Using a cryostat (ThermoFisher Scientific), brain tissues were cut coronally into 10-μm sections and placed on the capture area of the Visium Spatial Gene Expression Slide. For this study, 2 brain sections from tumor-implanted mice and 1 brain section from mice without tumor implantation were used. Tissue optimization was carried out using the Visium Spatial Tissue Optimization Slide and Reagents Kit in accordance with the manufacturer's instructions. The optimum tissue permeabilization duration was found to be 10 minutes. High-resolution brightfield H&E images were captured using the EVOS FL Auto Imaging System (ThermoFisher Scientific) with a 10X objective, enabling clear differentiation between tumor and nontumor regions. The tissue permeabilization, cDNA amplification, barcoding, and library construction steps were carried out according to the protocols outlined in the Visium Spatial Gene Expression User Guide (CG000239). The prepared libraries were loaded at 300 pmol/L and sequenced on an Illumina NovaSeq 6000 system, using a NovaSeq 200 Cycle S1 flow cell. Sequencing was performed using the following read protocol: read 1: 28 cycles; i7 index read: 10 cycles; i5 index read: 10 cycles; and read 2: 90 cycles.

Demultiplexing and analysis of spatial transcriptomics

The sequencing depth obtained ranged from approximately 143 × 106 to 183 × 106 reads per library and 45,000 to 100,000 mean reads/spot. The base call (BCL) files were processed with the 10x Genomics Space Ranger software v.1.2, which uses STAR v.2.5.1 for genome alignment, against the mm10 mouse reference dataset. The count files generated for each library were then aggregated with normalization set to ‘Mapped’. The aggregated cloupe file was visualized in 10x Genomics Loupe Browser software 6.0. The tumor area was demarcated on the basis of H&E and high expression of proliferation markers. Differentially expressed gene (DEG) analysis, Uniform Manifold Approximation and Projection (UMAP) plots, violin plots and heat map of DEGs were all performed within Loupe Browser. DEG analysis was performed via the Locally Distinguishing method in the Loupe Browser. The unsupervised K-means clustering method was used to identify clusters of spatial transcriptomics. The K-means value (K = 1–10) was chosen to 10 to show the highest possible amount of heterogeneity in the tumor microenvironment.

scATAC-seq

scATAC-seq analyses were performed on a previously published adult glioblastoma dataset (GSE139136; ref. 7). Copy-number analysis and cycling analysis were performed using Copy-scAT (18). Data analysis was performed using Signac (19). Datasets were pooled and normalized using mean signal across a randomly selected subset of peaks. ChromVAR was used to infer motif activities (20), followed by Nonnegative Matrix Factorization (NMF) analysis to decompose cell states within the tumor. Tightly correlated NMF decomposition-based states were pooled to generate two distinct states, and cells with high scores of either state were termed stem or differentiated, respectively, with cells having an intermediate score being termed intermediate. Gene activity at the BGN locus was computed using the GeneActivity method in Signac. Comparisons between BGN activity in modules were performed using an unpaired t test with Welch's correction.

Analysis of public scRNA-seq datasets

The counts matrix provided for the publicly available datasets used was downloaded and converted into a Seurat object using the package Seurat v3 in R (21). The data were filtered for the following parameters: cells with > 50 genes and the percentage of mitochondrial genes <15%. Data from all 12 libraries were then integrated and normalized with the SCTransform75 wrapper in Seurat using all 17,607 features. A principal component analysis (PCA) reduction was performed, and 17 significant PCA dimensions were accounted for. Clusters were determined using the FindNeighbours and FindClusters functions, and clustering was performed with a resolution of 0.5. Manual annotation of clusters was performed on the basis of the expression of lineage-specific signature genes. To display cell clusters, RunTSNE with PCA reduction was used. DEGs for one cluster (versus all cells in other clusters) were determined through the FindMarkers function. Only DEGs with a statistically significant P value of 0.05 were included for analysis. For each cluster, the DoHeatmap functionality in Seurat was used to plot heat map DEGs of interest. Dot plots depicting the average expression of hallmark genes and genes of interest expressed as a percentage in each of the clusters were created using the DotPlot program. The FeaturePlot and VlnPlot functions were used to create graphs that depict gene expression in cell clusters or sample groups.

Confocal immunofluorescence microscopy

We examined specimens from patients with GBM that were fixed in formalin and embedded in paraffin, as well as frozen sections from mice as described previously (12). The tissues were blocked by incubating them for 1 hour at room temperature using a horse-blocking solution consisting of PBS, 10% horse serum, 1% BSA, 0.1% cold fish skin gelatin, 0.1% Triton X-100, and 0.05% Tween 20. The tissues were then incubated overnight at 4°C with the following antibodies, which were resuspended in an antibody dilution buffer composed of PBS, 1% BSA, 0.1% cold fish skin gelatin, and 0.1% Triton X-100: biglycan (1:500; Novus Biologicals, NBP1–84971), mouse SOX2 (1:200; Thermo Fisher Scientific, 14–9811–82), human SOX2 (1:200; Abcam, ab171380), human Ki67 (1:400; Cell Signaling Technology, 9449), human Ki67 (1:100; Thermo Fisher Scientific, 14–5698–80), mouse Ki67 (1:500; Abcam, ab15580), β-catenin (1:800; Cell Signaling Technology, 37447S), mouse CD44 (1:500; Abcam, ab254530), human CD44 (1:100; Thermo Fisher Scientific, 14–0441–82), low-density lipoprotein receptor-related protein 6 (LRP6; 1:100; Novus Biologicals, MAB1505), and CD31 (1:200; Thermo Fisher Scientific, 14–0311–82). Afterwards, we included corresponding fluorophore-conjugated secondary antibodies (1:500; Jackson ImmunoResearch Laboratories or Thermo Fisher Scientific) alongside 4′,6-diamidino-2-phenylindole (DAPI; 1:1,000). To attach coverslips to the slides, we used Fluoromount-G (SouthernBiotech, 0100–01). We conducted laser confocal immunofluorescence imaging at room temperature using the Leica TCS SP8 laser confocal microscope with a 25/0.5 numerical aperture water objective. Samples were stimulated by 405-, 552-, and 640-nm lasers, and their signals were captured by one low-dark current Hamamatsu photomultiplier tube detector (DAPI) and three high-sensitivity hybrid detectors on the SP8. To obtain images of the samples, we used settings that comprised of 8 bits, bidirectional scanning in a z-stack, 4 to 8× frame averaging, 1 airy unit pinhole, 1 zoom, and 2048 by 2048 pixels x-y resolution. We applied consistent laser, gain, and offset settings to all samples in each trial series to optimize contrast and reduce saturation. Every human and mouse brain slice sample produced three to four fields of view (FOV). We used Leica Application Suite X to capture the images, and picture thresholding and three-dimensional (3D) rendering were executed using Imaris software (v9.9.1 Bitplane).

Cell proliferation analysis click-iT EdU-labeling assay

Cells were treated with 1 μmol/L EdU for 24 hours for the Click-iT EdU proliferation assay. Following cell permeabilization, EdU staining was conducted according to the manufacturer's instructions using the Click-iT EdU Alexa Fluor 488 Flow Cytometry Assay Kit (Thermo Fisher Scientific, C10425). Attune NxT flow cytometry was used for analysis of cells (Thermo Fisher Scientific). FlowJo version 10.7.2 was used to analyze the data (Treestar).

Confocal image analysis

The z-stack confocal images of spinal cords were analyzed with ImageJ (Fiji, National Institutes of Health). Briefly, maximum intensity projections were generated for each channel/marker and converted from 8-bit to RGB files. Positive signals were determined using the Color Threshold for all markers except DAPI. The threshold was used for DAPI. The Watershed algorithm was used to separate connected cells for quantification of DAPI-positive cells. The ‘analyze particles’ function was used to create a mask to quantify the positive signals in each FOV. The same color brightness threshold values, as well as particle analysis size and circularity parameters, were applied consistently across all samples for each experimental set. For representative images shown, maximum intensity projection values of each channel/marker were merged and displayed using pseudo colors. Only the brightness and contrast settings were changed between samples to improve visual presentation. To quantify the number of double-positive cells and the distance between cells and endothelial cells, we used the Imaris software. Specifically, we used the "Surfaces" function to identify the presence of positive immunoreactivity for each marker, and then measured the shortest distance between the surfaces.

Immunofluorescence staining of cells

We seeded BTICs in a 96-well Black/Clear Flat Bottom TC–treated Imaging Microplate (Falcon, 353219) at a density of 10,000 cells per well in BTIC medium. Prior to seeding, we coated the wells with 10 μg/mL Laminin (Millipore, CC095) to facilitate the attachment of BTICs to the plate. After incubation at 37°C, cells were washed with PBS and then fixed in 4% paraformaldehyde for 10 minutes. Cells were permeabilized using 0.25% Triton X-100 for 10 minutes and then blocked with Intercept (PBS) Blocking Buffer (LI-CORE, 927–70001). Cells were incubated with primary antibodies for biglycan (1:500; Novus Biologicals, NBP1–84971), CD44 (1:500; Abcam, ab254530), β-catenin (1:3,000; Cell Signaling Technology, 37447S), and SOX2 (1:200; Abcam, ab171380) diluted in blocking buffer overnight at 4°C. Cells were subsequently incubated with corresponding fluorophore-conjugated secondary antibodies (1:500, Jackson ImmunoResearch Laboratories or Thermo Fisher Scientific) at room temperature for 1 hour and washed, and nuclei were counterstained with DAPI (1:100) and imaged using ImageXpress (Molecular Devices). Multiwavelength cell scoring analysis in the MetaXpress High-Content Image Acquisition and Analysis Software (Molecular Devices) was used to quantify cell numbers. We made adjustments only to the brightness and contrast settings, which were consistently changed between samples to enhance visual presentation.

Lentivirus packaging

Lentivirus production was performed using 293FT cells grown to 90% confluency on five, 15 cm2 cell culture plates, by cotransfecting 112.5 μg pHAGE PGK-GFP-IRES-LUC-W [a gift from Darrell Kotton (Addgene plasmid #46793;http://n2t.net/addgene:46793; RRID:Addgene_46793)], 73 μg psPAX2 and 39.5 μg pMD2.G [gifts from Didier Trono (Addgene, plasmid # 12260; http://n2t.net/addgene:12260 RRID:Addgene_12260)] using CaPO4 precipitation transfection (22). Cell culture supernatant was collected and pooled every 12 to 16 hours, centrifuged at 500 × g for 5 minutes and filtered with a 0.45-μm syringe filter for a total of 3 harvests. The lentivirus-containing supernatant was underlaid with 20% sucrose in PBS and centrifuged at 50,000 × g for 2 hours using a Beckman SW28 ultracentrifuge rotor. The supernatant was discarded, and the lentivirus pellet was resuspended in sterile PBS and frozen at −80°C in 20-μL aliquots. The lentivirus titer determined using the qPCR lentivirus titer kit was ∼108 IU/mL (Applied Biological Materials).

Plasmid construction

The DNA sequence encoding mouse biglycan (NCBI accession NM_001411776) was PCR amplified from mouse whole brain cDNA whereas the human biglycan (NCBI accession BC002416) sequence was obtained from the Center for Cancer Systems Biology Human ORFeome collection (Horizon Discovery Biosciences). Both human and mouse BGN were subcloned into the PB-CMV-MCS-EF1α-Puro piggybac expression vector (System Biosciences) using the 5′ EcoRI and 3′ BamHI restriction sites. The sequences of all constructs were verified by Sanger DNA sequencing.

In vitro limiting dilution assays

For in vitro limiting dilution assays, BTICs were plated at decreasing densities (512, 256, 128, 64, 32, 16, 8, 4, 2, and 1 cell/well) in 96-well plates using flow cytometry sorting. Each cell dose was plated in six technical replicates. For the limiting dilution assays of BTICs with exogenous biglycan, recombinant biglycan was added to the culture every week at 10 μg/mL concentration. Two to 3 weeks later, the presence of tumor spheres in each well were recorded and analyzed using the Extreme Limiting Dilution Analysis (ELDA) software, available at http://bioinf.wehi.edu.au/software/elda/. The sphere-forming frequency was calculated as a percentage [1/(number of cells needed to form a sphere)*100].

Statistical analysis

Data were collated in Microsoft Excel and graph creation and data analysis were performed in GraphPad Prism 9.4.0. To examine statistically significant differences between the means of two or more treatment groups and the control group, one-way ANOVA with Tukey multiple-comparison test was performed. The significance of data with only two groups was tested using a two-tailed, unpaired t test. As stated in the figure legends, asterisks indicate significance: *, P < 0.05; **, P < 0.01; and ***, P < 0.001. Sample size calculation was not performed in this study. Instead, we determined the sample size based on previously published results, the feasibility and cost of the experiment, as well as the availability of sex- and age-matched mice. No inclusion or exclusion criteria were used unless otherwise stated. In violin plots, center lines represent the median and two quartile lines. The Pearson method was used in the correlation analysis between gene expressions. Unless otherwise specified, all data given are indicative of two to three separate studies with comparable findings. Mycoplasma was commonly detected in cell cultures. Blinding was not conducted.

Data availability

Spatial transcriptomic datasets reported in this paper are available to download from the NCBI Sequence Read Archive with BioProject accession numbers PRJNA914489. Previously published scRNA-seq and scATAC-seq data that were reanalyzed in this study were obtained from the European Genome-Phenome Archive (EGAS00001004656) and NCBI GEO (GSE139136), respectively. All other data are available upon request.

Spatial intratumoral heterogeneity in the brains of a GBM mouse model

To elucidate intratumoral heterogeneity in GBM, we performed spatial transcriptomics on brain tissue sections from mice intracranially transplanted with syngeneic mouse BT0309 BTIC line (Fig. 1A). In vivo bioluminescence imaging confirmed intracranial tumor growth 40 days after tumor implantation and before tissue harvest (Fig. 1B). H&E staining revealed the anatomic location of high proliferative regions (Fig. 1C). As a control, tissue from the same area of the brain was collected from normal mice without tumor implantation (Fig. 1C). The quality control (QC) metrics including nCount_spatial (number of transcripts), nFeature-spatial (number of genes), percent_mito (mitochondria content), and percent_hb (hemoglobin content) on spatial transcriptomics data were shown in Supplementary Fig. S1.

Figure 1.

Spatial intratumoral heterogeneity in the brains of a GBM mouse model. A, Schematic showing the spatially resolved transcriptomics experiment workflow. Forty days after intracranial implantation of syngeneic mouse BTICs, brain tissues were subjected to Visium spatial transcriptomics and data analysis. B, Bioluminescence in vivo images of two mice 40 days after tumor implantation. C, Histologic images (H&E) images of brain tissues obtained from two tumor mice and one healthy control without tumor injection. Dashed red lines highlight the high-density tumor areas. D, Cluster overlay images showing 10 spatial clusters on tissues. E, t-SNE plot of 10 spatial clusters. F, Heat map of top DEGs in the 10 spatial clusters. G, t-SNE plots delineating distribution of the 8 ECM genes that were shown to be highly elevated in clusters 8 and 9. H, Spatial distribution of the 8 ECM gene transcripts across tissue sections in tumor 1 and normal brain. Dashed red lines highlight the high-density tumor areas.

Figure 1.

Spatial intratumoral heterogeneity in the brains of a GBM mouse model. A, Schematic showing the spatially resolved transcriptomics experiment workflow. Forty days after intracranial implantation of syngeneic mouse BTICs, brain tissues were subjected to Visium spatial transcriptomics and data analysis. B, Bioluminescence in vivo images of two mice 40 days after tumor implantation. C, Histologic images (H&E) images of brain tissues obtained from two tumor mice and one healthy control without tumor injection. Dashed red lines highlight the high-density tumor areas. D, Cluster overlay images showing 10 spatial clusters on tissues. E, t-SNE plot of 10 spatial clusters. F, Heat map of top DEGs in the 10 spatial clusters. G, t-SNE plots delineating distribution of the 8 ECM genes that were shown to be highly elevated in clusters 8 and 9. H, Spatial distribution of the 8 ECM gene transcripts across tissue sections in tumor 1 and normal brain. Dashed red lines highlight the high-density tumor areas.

Close modal

The K-means unsupervised clustering identified 10 clusters of spatial transcriptomics (Fig. 1D and E). The analysis of differential gene expression revealed the top DEGs between the 10 clusters (Fig. 1F; Supplementary Table S2). Among the genes with differential expression, there were 8 genes encoding ECM proteins including Vcan (encoding versican), Col2a1 (encoding Collagen type II, alpha 1), Col9a1 (encoding collagen type IX, alpha 1), Col9a2 (encoding collagen type IX, alpha 2), Col11a1 (encoding collagen type XI, alpha 1), Col11a2 (encoding collagen type XI, alpha 2), Fmod (encoding fibromodulin), and Bgn (encoding biglycan) that were expressed at higher levels in cluster 8 versus other clusters (Fig. 1F). The expression of these ECM genes in cluster 8 is displayed by t-distributed stochastic neighbor embedding (t-SNE) plots (Fig. 1G). Gene expression of these ECM genes was observed to be higher in the high proliferative tumor regions compared with nonmalignant areas and normal brain (Fig. 1H; Supplementary Fig. S2A and S2B). In addition to cluster 8, there were also high transcript levels of Bgn in cluster 9, which corresponded to areas close to high proliferative regions (Supplementary Fig. S2C).

Differential expression of ECM molecules by cells in human GBMs

Because the Visium spatial transcriptomics platform used here did not enable us to assess transcriptomic expression at the single-cell level, we analyzed publicly available scRNA-seq data from 7 patients with GBM (9) to dissect ECM gene expression on different cell populations. Data of QC, PCA and cluster analysis were shown in Supplementary Fig. S3A and S3B. Unsupervised clustering of all 44,712 cells from the 7 patients with GBM, based on 2,000 variable genes and 15 significant principal components, delineated 18 clusters (Fig. 2A). Lineage cell markers amongst the DEGs were used to identify the clusters. Clusters 0–2, 6, 9, 13, 14, 16, and 17 were identified as microglia and monocyte-derived macrophages based on their expression of the following signature genes: PTPRC (CD45), ITGAM, TMEM119, and CX3CR1 (Fig. 2B; Supplementary Fig. S3C). Cluster 11 was identified as T cells by their expression of PTPRC and CD3E. Clusters 3 and 15 were ascribed as oligodendrocytes based on their expression of MAG and MOG. We identified clusters 4, 5, 7, 8, 10, and 12 as tumor cells through their expression of putative tumor cell marker EGFR. Distinctions between malignant and nonmalignant cell populations were also confirmed by analyzing copy-number variations (CNV) from scRNA-seq profiles using inferCNV pipeline (23, 24). All malignant clusters displayed large-scale CNVs characteristic of GBM, such as gain of chromosome (chr) 7 and deletion of chr10 (Fig. 2C).

Figure 2.

Differential expression of ECM molecules in cells in human GBMs. A, UMAP plot of 44,712 cells from 7 patients with GBM. B, Feature UMAP plot showing the distribution of various genes used to define clusters representing microglia, monocyte-derived macrophages, T cells, oligodendrocytes, and tumor cells in scRNA-seq. C, Inference of CNV analysis based on average expression of 100 genes shows the chromosome 7 gain (red) and chromosome 10 (blue) loss in tumor cells (observations) compared with normal cells (references). D, Dot plot showing the expression of various ECM genes across the 18 cell clusters. The size of the dot corresponds to the percentage of cells expressing the gene in each cluster. The color represents the average gene expression level. E, CytoTRACE UMAP plot shows predicted score of differentiation states of tumor clusters from scRNA-seq data of 7 patients with GBM. F, Phenotype UMAP plot of CytoTRACE analysis. G and H, Violin plots showing expression levels of developmental (G) and injury response (H) signature genes.

Figure 2.

Differential expression of ECM molecules in cells in human GBMs. A, UMAP plot of 44,712 cells from 7 patients with GBM. B, Feature UMAP plot showing the distribution of various genes used to define clusters representing microglia, monocyte-derived macrophages, T cells, oligodendrocytes, and tumor cells in scRNA-seq. C, Inference of CNV analysis based on average expression of 100 genes shows the chromosome 7 gain (red) and chromosome 10 (blue) loss in tumor cells (observations) compared with normal cells (references). D, Dot plot showing the expression of various ECM genes across the 18 cell clusters. The size of the dot corresponds to the percentage of cells expressing the gene in each cluster. The color represents the average gene expression level. E, CytoTRACE UMAP plot shows predicted score of differentiation states of tumor clusters from scRNA-seq data of 7 patients with GBM. F, Phenotype UMAP plot of CytoTRACE analysis. G and H, Violin plots showing expression levels of developmental (G) and injury response (H) signature genes.

Close modal

Visualization of gene expression amongst the 18 clusters delineated that the 8 ECM genes described above were expressed at different levels amongst cells in the tumor microenvironment of human GBMs (Fig. 2D). COL11A2 was expressed at higher levels in a subpopulation of oligodendrocytes (cluster 3) and tumor cells (clusters 4, 5, 7, and 8). COL9A2 transcripts were detected at higher levels in oligodendrocytes (clusters 3 and 15) and tumor cells (clusters 4 and 7). COL11A1, COL9A1, FMOD, and BGN were mostly expressed by tumor cell clusters. Finally, we detected high VCAN expression in a subpopulation of microglia and monocyte-derived macrophages (cluster 6) and tumor cells. COL2A1 was not detected at appreciable levels by scRNA-seq.

To better understand the expression profile of the ECM molecules on different types of tumor cell populations including differentiated tumor cells versus BTICs, we first predicted the differentiation states of tumor cells from scRNA-seq data using CytoTRACE trajectory reconstruction analysis (25). CytoTRACE analysis revealed that clusters 7, 8, 4, and 5 in that order had a transcriptional profile consistent with a less differentiated phenotype (Fig. 2E and F; Supplementary Fig. S4A). Single-cell expression analysis of stemness genes including SOX2, SOX4, SOX11, PROM1, FABP7, NES, MSI1, and FUT4 also showed higher expression levels on clusters 7, 8, 4, and 5 than other tumor clusters (Supplementary Fig. S4B). A previous report has shown that the diversity of BTICs is distributed along a transcriptional gradient that spans two major cellular programs: developmental and injury response (9). We next determined the expression of developmental versus injury response markers on tumor cell clusters with less differentiation phenotypes (clusters 4, 5, 7, 8). Clusters 5 and 7 expressed higher levels of developmental markers such as ASCL1, PTPRZ1, OLIG1, and OLIG2 (Fig. 2G). In contrast, higher expression levels of injury response markers including CD44, S100A11, TXN, and TAGLN2 were observed in clusters 4 and 8 (Fig. 2H). Single-cell expression analysis of the ECM genes showed the expression of COL11A1, COL11A2, COL9A2, FMOD, and VCAN by both developmental and injury response tumor cells (Fig. 2D). COL9A1 was more expressed by developmental compared with injury response BTICs. Interestingly, BGN was either highly expressed or upregulated in BTICs with more injury response than developmental phenotypes (Fig. 2D; Supplementary Fig. S4C). Because the injury response phenotype of BTICs is probably associated with the mesenchymal phenotype of patients with GBM with a worse prognosis than other subtypes (8), we focused on biglycan amongst other upregulated ECM genes in this study.

Expression of biglycan in BTICs within the human GBM microenvironment

The spatial transcriptomics data from mouse GBMs revealed higher expression of stemness markers such as Sox2, Sox4, Olig2, and Fabp7 in areas with high expression of the ECM transcripts (Supplementary Fig. S5A). There were higher expression levels of the stemness markers in malignant tissues compared with the normal brain (Supplementary Fig. S5B). Furthermore, expression levels of the stemness genes were upregulated in the spatial transcriptomic clusters 8 and 10 (Supplementary Fig. S5C), suggesting these spatial clusters that were associated with higher Bgn expression exhibited stemness signatures. The trajectory reconstruction analysis of BGN expression revealed that there were higher expression levels of BGN in tumor cells with lower differentiation phenotypes (Fig. 3A).

Figure 3.

Expression of biglycan in BTICs within the human GBM microenvironment. A, Force-directed layout plots of CytoTRACE analysis show high expression of BGN in tumor cells with less differentiation states. B, Identification of three cell types in GBM based on differentiation and stemness modules in scATAC-seq. C,BGN activity in differentiation versus stemness modules. D, Bar graph comparing BGN activity in three cell types identified in scATAC-seq. E and F, Representative confocal staining assessing the proximity of biglycan and SOX2+ cells in human GBM tissues from two patients (101220 and 101711). G, Representative confocal images of staining for biglycan and SOX2 in mouse GBM tissues. H, 3D reconstruction of images of SOX2 and biglycan in GBM tissues of the mouse model. I, Representative confocal images of outside tumor regions of brain mice implanted with tumor cells. The tissues were labeled with DAPI, SOX2, and biglycan. J, Representative confocal images of normal mouse brain (without tumor implantation). The tissues were labeled with DAPI, SOX2 and biglycan.

Figure 3.

Expression of biglycan in BTICs within the human GBM microenvironment. A, Force-directed layout plots of CytoTRACE analysis show high expression of BGN in tumor cells with less differentiation states. B, Identification of three cell types in GBM based on differentiation and stemness modules in scATAC-seq. C,BGN activity in differentiation versus stemness modules. D, Bar graph comparing BGN activity in three cell types identified in scATAC-seq. E and F, Representative confocal staining assessing the proximity of biglycan and SOX2+ cells in human GBM tissues from two patients (101220 and 101711). G, Representative confocal images of staining for biglycan and SOX2 in mouse GBM tissues. H, 3D reconstruction of images of SOX2 and biglycan in GBM tissues of the mouse model. I, Representative confocal images of outside tumor regions of brain mice implanted with tumor cells. The tissues were labeled with DAPI, SOX2, and biglycan. J, Representative confocal images of normal mouse brain (without tumor implantation). The tissues were labeled with DAPI, SOX2 and biglycan.

Close modal

To understand the chromatin accessibility of BGN gene in different phenotypes of tumor cells, we examined a previously published dataset of single-cell chromatin accessibility in adult GBM (7). NMF decomposition of motif activity in the scATAC-seq data identified three cell types based on the stemness versus differentiation module: differentiated, intermediate, and stem cells (Fig. 3B). BGN gene activity was shown to be significantly higher in stem cells compared with intermediate and differentiated cells (Fig. 3C and D), confirming the higher expression levels of BGN in BTICs than in other types of tumor cells.

Next, we evaluated biglycan protein expression in BTICs in human GBM specimens. While there is no uniform single marker for BTICs, SOX2 is widely used to identify BTICs within the GBM microenvironment (26, 27). Immunofluorescence confocal microscopy of human GBM specimens showed expression of biglycan by SOX2-positive cells (Fig. 3E and F; Supplementary Fig. S6A and S6B). Three-dimensional reconstruction of double labeling corroborated the presence of biglycan on SOX2-positive cells within the tumor microenvironment (Supplementary Fig. S6C and S6D). As controls, lower expression levels of biglycan and SOX2 were found in normal brains (Supplementary Fig. S6E–G). We were unable to detect staining signals after incubating human GBM tissues with secondary antibodies (Supplementary Fig. S6H), ruling out secondary antibody reactivity with target markers. We also assessed biglycan expression on SOX2-positive cells in the mouse GBM model. Similarly, there was immunoreactivity of biglycan on SOX2-positive cells in the tumor microenvironment (Fig. 3G), which was corroborated by 3D reconstruction of double labeling (Fig. 3H). In contrast to tumor area, there was lower expressions of biglycan and SOX2 in nonmalignant areas of mice implanted with tumor (Fig. 3I) and normal mouse brain (Fig. 3J). In contrast to the RNA expression study, our analysis did not reveal protein expression of biglycan exclusively on BTICs. This was due to the presence of biglycan-expressing SOX2-negative cells in the tumor microenvironment.

Because perivascular areas are one of known cancer stem cell niches in brain tumors, we interrogated biglycan expression on BTICs in relation to CD31-positive endothelial cells. While biglycan expression was shown to be high on endothelial cells (Supplementary Fig. S7A and S7B), we did not find a significant accumulation of biglycan-positive BTICs close to perivascular niches (Supplementary Fig. S7C–S7E).

Biglycan is associated with the injury response phenotype of BTICs

Because BTICs with the injury response phenotype are likely related to the mesenchymal phenotype of patients with GBM, we first examined mRNA expression of BGN in GBM subtypes using The Cancer Genome Atlas (TCGA) database. We observed significant upregulation of BGN in the mesenchymal subtype compared with classical and proneural subtypes (Fig. 4A). Examining the TCGA GBM Agilent-4502A and TCGA GBM RNA-seq databases confirmed significant positive correlations between BGN and CD44 expression (r = 0.34, P value = 0.001; r = 0.45, P value = 0.001, respectively; Fig. 4B). We also leveraged public scRNA-seq data from 21 adult GBM and 7 pediatric high-grade gliomas resections encompassing 24,131 total cells (4). As previously described, most tumor cells in these 28 specimens clustered into cellular states according to multi-gene expression signatures (4). These categories were previously labeled as: (i) NPC-like, (ii) OPC-like, (iii) AC-like, and (iv) MES-like states. We observed that there was a high expression of BGN in MES-like cells (Fig. 4C).

Figure 4.

Biglycan is associated with the injury response phenotype of BTICs. A, Analysis of BGN transcript expression in different GBM subtypes via The TCGA database. B, Correlation between BGN and CD44 expression in the TCGA-GBM Agilent-4502A (top) and TCGA-GBM RNA-seq (bottom) database. C, Expression of BGN in four different GBM cellular states including AC-like, MES-like, OPC-like, and NPC-like signatures (meta-modules) using relative expression of each multigene signature (relative meta-module score) previously described (4). D and E, Representative confocal images of human GBM specimens from two patients labeled with DAPI, SOX2, CD44 and biglycan. F, Graph comparing the percentage of CD44-positive cells between SOX2+ biglycan+ versus SOX2+ biglycan cells in four patients with GBM. G, Representative widefield microscopy images of patient-derived BTICs labeled with DAPI, biglycan, and CD44. H, Bar graph comparing expression of CD44 between biglycan-positive and -negative cells. Data are representative of two to three separate experiments. Significance indicated as: *, P < 0.05; ***, P < 0.001; two-tailed, unpaired t test. Data are represented as mean ± SEM.

Figure 4.

Biglycan is associated with the injury response phenotype of BTICs. A, Analysis of BGN transcript expression in different GBM subtypes via The TCGA database. B, Correlation between BGN and CD44 expression in the TCGA-GBM Agilent-4502A (top) and TCGA-GBM RNA-seq (bottom) database. C, Expression of BGN in four different GBM cellular states including AC-like, MES-like, OPC-like, and NPC-like signatures (meta-modules) using relative expression of each multigene signature (relative meta-module score) previously described (4). D and E, Representative confocal images of human GBM specimens from two patients labeled with DAPI, SOX2, CD44 and biglycan. F, Graph comparing the percentage of CD44-positive cells between SOX2+ biglycan+ versus SOX2+ biglycan cells in four patients with GBM. G, Representative widefield microscopy images of patient-derived BTICs labeled with DAPI, biglycan, and CD44. H, Bar graph comparing expression of CD44 between biglycan-positive and -negative cells. Data are representative of two to three separate experiments. Significance indicated as: *, P < 0.05; ***, P < 0.001; two-tailed, unpaired t test. Data are represented as mean ± SEM.

Close modal

To elucidate the correlation between biglycan expression and the injury response phenotype of BTICs at the level of protein, we subjected human tissue specimens from four patients to immunofluorescence staining for CD44, SOX2, and biglycan. Confocal imaging of tissues revealed significantly higher expression of CD44 on biglycan-positive than negative BTICs (Fig. 4DF). To further corroborate the higher expression of CD44 on biglycan-positive BTICs, we cultured patient-derived BTICs and assessed the expression of CD44 and biglycan on BTICs (Fig. 4G). Image analysis showed that there was a higher expression of CD44 on biglycan-positive than biglycan-negative cells in two BTIC lines (Fig. 4H).

Biglycan is expressed in tumor areas with more proliferating BTICs

The spatial transcriptomics data demonstrated high expression levels of proliferation markers such as Mki67, Ccnd1, Ccnd3, and Cdk4 in cluster 8 and at a lesser amount in clusters 9 and 10, as seen in the violin plots in Fig. 5A. In addition, the spatial distribution of gene transcripts across the tissue section showed high expression of the proliferating markers in tumor areas with high levels of Bgn transcripts (Fig. 5B). Similarly, the scATAC-seq uncovered that cycling cells had significantly higher BGN activity than non-cycling cells (Fig. 5C). Correlation analysis through the TCGA GBM Agilent-4502A and TCGA GBM RNA-seq databases confirmed significant positive correlations between BGN and MKI67 expression (r = 0.19, P value = 0.001; r = 0.29, P value = 0.001 respectively; Fig. 5D). To validate the higher proliferative levels of biglycan expressing BTICs in the human GBM microenvironment, we stained tumor specimens from four patients for biglycan and the proliferative marker Ki67 as well as SOX2. Confocal microscopy and quantification of images affirmed significantly higher expression of Ki67 on biglycan-positive than biglycan-negative BTICs (Fig. 5EG). Together, these data suggest that biglycan may be involved in promoting the proliferation of BTICs in GBMs.

Figure 5.

Expression of biglycan in tumor areas with more proliferating cells. A, Violin plots showing the expression levels of cell cycle genes in the 10 spatial clusters. B, Spatial distribution of cell cycle gene transcripts in tissue sections from tumor and normal brains. Dashed red lines highlight the high-density tumor areas. C, A bar graph comparing BGN activity in non-cycling versus cycling cells in human GBM tissues as determined by scATAC-seq analysis. D, Correlation between BGN and MKI67 expression in the TCGA-GBM Agilent-4502A (left) and TCGA-GBM RNA-seq (right) database. E and F, Representative confocal images of human GBM specimens from two patients labeled with DAPI, SOX2, Ki67, and biglycan. SOX2-positive cells expressing biglycan and Ki67 are indicated by orange arrowheads. G, Bar graph comparing expression of Ki67 between biglycan-positive and -negative cells in four patients with GBM. Significance indicated as: **, P < 0.01; two-tailed, unpaired t test. Data are represented as mean ± SEM.

Figure 5.

Expression of biglycan in tumor areas with more proliferating cells. A, Violin plots showing the expression levels of cell cycle genes in the 10 spatial clusters. B, Spatial distribution of cell cycle gene transcripts in tissue sections from tumor and normal brains. Dashed red lines highlight the high-density tumor areas. C, A bar graph comparing BGN activity in non-cycling versus cycling cells in human GBM tissues as determined by scATAC-seq analysis. D, Correlation between BGN and MKI67 expression in the TCGA-GBM Agilent-4502A (left) and TCGA-GBM RNA-seq (right) database. E and F, Representative confocal images of human GBM specimens from two patients labeled with DAPI, SOX2, Ki67, and biglycan. SOX2-positive cells expressing biglycan and Ki67 are indicated by orange arrowheads. G, Bar graph comparing expression of Ki67 between biglycan-positive and -negative cells in four patients with GBM. Significance indicated as: **, P < 0.01; two-tailed, unpaired t test. Data are represented as mean ± SEM.

Close modal

Biglycan promotes proliferation of patient-derived BTICs in culture

We next sought to understand the effects of biglycan on the growth of BTICs isolated from patients with human GBM. After generating BTICs from surgically resected tumor tissues, we treated cells with recombinant biglycan as shown in Fig. 6A. We found that exogenous biglycan significantly increased sphere number in culture (Fig. 6B and C). In addition, after 72 hours of treatment with recombinant biglycan, the ATP proliferation test revealed increased cell growth in BTICs (Supplementary Fig. S8A). Flow cytometry EdU (ethynyl deoxyuridine) proliferation assay also showed that biglycan significantly elevated BTIC proliferation (Fig. 6D and E; Supplementary Fig. S8B). Moreover, measuring the sphere-forming frequency of cells in the limited dilution assay indicated that recombinant biglycan significantly elevated sphere-forming capacity of BTICs (Fig. 6F).

Figure 6.

Biglycan promotes proliferation of patient-derived BTICs in culture. A, Schematic showing the procedure of BTIC generation from patients with GBM in the presence of EGF, FGF, and heparin solution (HS) and treatment with recombinant biglycan. B, Representative brightfield microscopy images of 48- to 72-hour outcomes of tumor spheres. C, Quantification of tumor spheres of two human BTIC lines. Four FOVs from four cell culture wells were used to quantify the number of spheres with diameters between 50 and 100 μm. D and E, Representative flow cytometry histograms (D) and bar plots (E) of the proliferation of human BTIC lines measured by EdU proliferation assay after 24 hours of treatment. F, Sphere forming frequency of BTICs treated with recombinant biglycan compared with control. G, Representative widefield microscopy images of patient-derived BTIC lines (BT048 and BT073) overexpressing BGN compared with control. Cells were labeled with DAPI and biglycan. H, Bar graph comparing the percentage of cells expressing biglycan between biglycan-overexpressing cells compared with control. I, Bar graph comparing the percentage of spheres in BGN-OE BTICs versus control. J, Sphere forming frequency of BTICs overexpressing BGN compared with control. Data are representative of two to three separate experiments. Means were compared between groups by unpaired (two-tailed) t test. *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001. Data are presented as the mean ± SEM (error bars). (A, Created with BioRender.com.)

Figure 6.

Biglycan promotes proliferation of patient-derived BTICs in culture. A, Schematic showing the procedure of BTIC generation from patients with GBM in the presence of EGF, FGF, and heparin solution (HS) and treatment with recombinant biglycan. B, Representative brightfield microscopy images of 48- to 72-hour outcomes of tumor spheres. C, Quantification of tumor spheres of two human BTIC lines. Four FOVs from four cell culture wells were used to quantify the number of spheres with diameters between 50 and 100 μm. D and E, Representative flow cytometry histograms (D) and bar plots (E) of the proliferation of human BTIC lines measured by EdU proliferation assay after 24 hours of treatment. F, Sphere forming frequency of BTICs treated with recombinant biglycan compared with control. G, Representative widefield microscopy images of patient-derived BTIC lines (BT048 and BT073) overexpressing BGN compared with control. Cells were labeled with DAPI and biglycan. H, Bar graph comparing the percentage of cells expressing biglycan between biglycan-overexpressing cells compared with control. I, Bar graph comparing the percentage of spheres in BGN-OE BTICs versus control. J, Sphere forming frequency of BTICs overexpressing BGN compared with control. Data are representative of two to three separate experiments. Means were compared between groups by unpaired (two-tailed) t test. *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001. Data are presented as the mean ± SEM (error bars). (A, Created with BioRender.com.)

Close modal

To affirm the proliferative effects of biglycan, we also overexpressed BGN in patient-derived BTICs. Immunofluorescence cell staining confirmed a significant upregulation of biglycan protein in BGN-overexpressing (OE) cells compared with control (Fig. 6G and H). Quantifying number of spheres showed that there was a significantly higher percentage of spheres in BGN-OE BTICs than in control BTICs (Fig. 6I). ATP proliferation assay also revealed higher cell number in BGN-OE BTICs compared with control (Supplementary Fig. S8C). Moreover, measuring sphere-forming capacity of cells in the limited dilution assay indicated that BGN-OE BTICs produced significantly more spheres than controls (Fig. 6J). Together, these data showed that biglycan promoted proliferation of BTICs in culture generated from patients with GBM.

Biglycan activates Wnt/β-catenin through interaction with LRP6 on BTICs

According to the existing literature, biglycan can bind to several receptors, such as Toll-like receptor 2 and 4, P2X purinoceptor (P2RX) 2 and 4, and low-density lipoprotein receptor-related protein 6 (LRP6; refs. 28–30). We assessed the mRNA expression of these receptors in two human BTIC lines and found higher expression of LRP6 than other receptors (Fig. 7A). In addition, scRNA-seq analysis of 7 patients with GBM showed that compared with other receptors, there were higher expression levels of LRP6 on clusters 4, 5, 7, 8, 10, and 12 that correspond to tumor cells (Fig. 7B). Spatial transcriptomics of mouse GBM confirmed anatomical expression of Lrp6 in tumor area corresponding to cluster 9 (Fig. 7C and D). The higher expression of Lrp6 transcripts within the high proliferative regions than in nonmalignant areas and normal brain was corroborated by regional expression analysis (Fig. 7E and F). Based on the role of biglycan in BTIC proliferation, we analyzed the TCGA GBM Agilent-4502A and TCGA GBM RNA-seq datasets and found a positive association between MKI67 and LRP6 (r = 0.34 and r = 0.39, respectively; Supplementary Fig. S9A). Co-labeling of human GBM tissues confirmed expression of LRP6 on SOX2-positive cells which were in proximity to biglycan (Fig. 7G). Together, these findings point to LRP6 as a possible biglycan receptor on BTICs and that blocking this receptor may reduce the biglycan proliferative effects. To test this, we treated BGN-OE BTICs with monoclonal antibodies against LRP6. After 48 hours, BGN-OE cells treated with blocking antibodies had significantly reduced proliferation levels compared with cells incubated with isotype control antibodies (Fig. 7H).

Figure 7.

Biglycan activates the Wnt/β-catenin through interaction with LRP6 on BTICs. A, Expression levels of different biglycan receptors in two human BTIC lines determined via bulk RNA-seq. FPKM, fragments per kilobase of exon per million mapped fragments. B, Dot plot showing the expression of various biglycan receptors across 18 single-cell cell clusters of 7 patients with GBM. The size of the dot corresponds to the percentage of cells expressing the gene in each cluster. The color represents the average gene expression level. C, Spatial distribution of Lrp6 transcripts across tissue section in tumor and normal mouse brains. Dashed red lines highlight the high-density tumor areas. D, Violin plot displaying expression levels of Lrp6 gene transcripts amongst the 10 spatial clusters. E, Regional expression analysis of Lrp6 in high-density tumor areas, nonmalignant regions, and normal brain. The high-density tumor areas were defined on the basis of the H&E staining. F, Violin plot displaying expression levels of Lrp6 in three regional clusters. G, Representative confocal microscopy images of human GBM specimens. Tissue was labeled with DAPI, SOX2, biglycan, and LRP6. H, ATP proliferation assay of human BTICs overexpressing BGN after 48- to 72-hour treatment with LRP6 blocking antibody. I, Violin plot displaying expression levels of Ctnnb1 transcripts (encoding β-catenin) in 10 spatial clusters. J, Spatial distribution of Ctnnb1 transcripts across tissue sections in tumor and normal mouse brains. Dashed red lines highlight the high-density tumor areas. K, Violin plot displaying expression levels of Ctnnb1 in three regional clusters of spatial transcriptomics. L, Representative widefield microscopy images of patient-derived BTICs overexpressing BGN and control cells. Cells were labeled with DAPI, and antibodies against biglycan and β-catenin. M, Bar graph comparing the percentage of cells expressing biglycan and β-catenin. Data are representative of two to three separate experiments. Means were compared with respective control by unpaired (two-tailed) t test when comparing two groups. For more than two groups, one-way analysis of variance (ANOVA) with Tukey post hoc was used. **, P < 0.01; ***, P < 0.001; ****, P < 0.0001. Data are presented as the mean ± SEM (error bars).

Figure 7.

Biglycan activates the Wnt/β-catenin through interaction with LRP6 on BTICs. A, Expression levels of different biglycan receptors in two human BTIC lines determined via bulk RNA-seq. FPKM, fragments per kilobase of exon per million mapped fragments. B, Dot plot showing the expression of various biglycan receptors across 18 single-cell cell clusters of 7 patients with GBM. The size of the dot corresponds to the percentage of cells expressing the gene in each cluster. The color represents the average gene expression level. C, Spatial distribution of Lrp6 transcripts across tissue section in tumor and normal mouse brains. Dashed red lines highlight the high-density tumor areas. D, Violin plot displaying expression levels of Lrp6 gene transcripts amongst the 10 spatial clusters. E, Regional expression analysis of Lrp6 in high-density tumor areas, nonmalignant regions, and normal brain. The high-density tumor areas were defined on the basis of the H&E staining. F, Violin plot displaying expression levels of Lrp6 in three regional clusters. G, Representative confocal microscopy images of human GBM specimens. Tissue was labeled with DAPI, SOX2, biglycan, and LRP6. H, ATP proliferation assay of human BTICs overexpressing BGN after 48- to 72-hour treatment with LRP6 blocking antibody. I, Violin plot displaying expression levels of Ctnnb1 transcripts (encoding β-catenin) in 10 spatial clusters. J, Spatial distribution of Ctnnb1 transcripts across tissue sections in tumor and normal mouse brains. Dashed red lines highlight the high-density tumor areas. K, Violin plot displaying expression levels of Ctnnb1 in three regional clusters of spatial transcriptomics. L, Representative widefield microscopy images of patient-derived BTICs overexpressing BGN and control cells. Cells were labeled with DAPI, and antibodies against biglycan and β-catenin. M, Bar graph comparing the percentage of cells expressing biglycan and β-catenin. Data are representative of two to three separate experiments. Means were compared with respective control by unpaired (two-tailed) t test when comparing two groups. For more than two groups, one-way analysis of variance (ANOVA) with Tukey post hoc was used. **, P < 0.01; ***, P < 0.001; ****, P < 0.0001. Data are presented as the mean ± SEM (error bars).

Close modal

Previous studies demonstrated LRP6 as an upstream driver of the Wnt/β-catenin signaling pathway (31). Spatial transcriptomic analysis of the mouse tumor microenvironment showed high expression of Ctnnb1 gene in clusters 8 and 9 corresponding to high proliferative regions of tumor (Fig. 7I and J). Notably, there was high Bgn expression in clusters 8 and 9 of spatial transcriptomics (Supplementary Fig. S2C). The higher expression of Ctnnb1 transcripts within the high proliferative regions than nonmalignant areas and normal brain was affirmed by regional expression analysis (Fig. 7K). However, because of the limited resolution of spatial transcriptomics, we analyzed the scRNA-seq data of 7 patients with GBM for CTNNB1 and observed that CTNNB1 expression is not limited to cancer cells (Supplementary Fig. S9B). To compare CTNNB1 expression in differentiated tumor cells versus BTICs, we did CytoTRACE trajectory reconstruction analysis on the scRNA-seq data. Consistent with BGN expression, we found high expression of CTNNB1 in GBM cells with more stemness profiles (Supplementary Fig. S9C and S9D). To validate the activation of the Wnt pathway in BTICs with mesenchymal characteristics, we analyzed the expression of other Wnt signaling genes and observed that mesenchymal BTICs exhibited higher upregulation than developmental BTICs (Supplementary Fig. S9E). Also, the CytoTRACE analysis showed high expression levels of Wnt signaling genes in cells with more stemness profiles (Supplementary Fig. S9F). These observations suggested that activation of Wnt/β-catenin may be downstream of the LRP6–biglycan axis. To test this, we quantified β-catenin expression in BGN-OE cells compared with control BTICs in culture (Fig. 7L). There were significantly higher expression levels of β-catenin in BGN-OE BTICs than in the control (Fig. 7M).

Biglycan accelerates in vivo tumor growth in the GBM animal model

To address in vivo benefits of biglycan on tumor growth, we implanted syngeneic mouse BT0309 with stable upregulated Bgn into the striatum of immunocompetent C57BL/6 mice. By day 45, we found that mice with upregulated Bgn BTICs had significantly higher tumor burden compared with control mice as observed by in vivo bioluminescence imaging (Fig. 8A and B). Staining for Ki67 as an index of cell proliferation showed significantly higher immunoreactivity in mice implanted with Bgn-OE cells compared with control (Fig. 8C and D). However, we could not observe a significant difference in the percentage of SOX2-positive cells between mice implanted with Bgn-OE and control BTICs (Fig. 8E). To determine if in vivo biglycan overexpression led to activation of the Wnt/β-catenin pathway, we stained tumor tissues for β-catenin. Forty-five days after tumor implantation, there were higher expression levels of β-catenin in the tumor of mice implanted with Bgn-OE BTICs than control (Fig. 8F and G). Staining for CD44 as an index of mesenchymal state also showed higher immunoreactivity in mice implanted with Bgn-OE compared with control BTICs (Fig. 8H and I).

Figure 8.

Biglycan accelerates in vivo tumor growth in the GBM animal model. A, Representative images of in vivo bioluminescence monitoring of tumor growth in mice implanted with Bgn-OE mouse BTICs compared with that in mice grafted with control cells. B, Graph comparing total Flux signal in mice implanted with Bgn-OE BTICs versus control. C, Representative confocal microscopy images of mouse GBM tissues implanted with Bgn-OE versus control BTICs. Tissue was labeled with DAPI, and antibodies against SOX2 and Ki67. D, Graph comparing the percentage of Ki67-positive cells between mice grafted with Bgn-OE versus control BTICs. E, Graph comparing the percentage of SOX2-positive cells between mice grafted with Bgn-OE versus control BTICs. F, Representative confocal microscopy images of mouse GBM tissues implanted with Bgn-OE versus control BTICs. Tissue was labeled with DAPI, SOX2, and β-catenin. G, Graph comparing the percentage of β-catenin positive cells between mice grafted with Bgn-OE versus control BTICs. H, Representative confocal microscopy images of mouse GBM tissues implanted with Bgn-OE versus control BTICs. Tissues were labeled with DAPI, SOX2, and CD44. I, Graph comparing the percentage of CD44-positive cells between mice implanted with Bgn-OE versus control BTICs. Data are representative of two to three separate experiments. Means were compared between groups by unpaired (two-tailed) t test. **, P < 0.01; ***, P < 0.001. Data are presented as the mean ± SEM (error bars).

Figure 8.

Biglycan accelerates in vivo tumor growth in the GBM animal model. A, Representative images of in vivo bioluminescence monitoring of tumor growth in mice implanted with Bgn-OE mouse BTICs compared with that in mice grafted with control cells. B, Graph comparing total Flux signal in mice implanted with Bgn-OE BTICs versus control. C, Representative confocal microscopy images of mouse GBM tissues implanted with Bgn-OE versus control BTICs. Tissue was labeled with DAPI, and antibodies against SOX2 and Ki67. D, Graph comparing the percentage of Ki67-positive cells between mice grafted with Bgn-OE versus control BTICs. E, Graph comparing the percentage of SOX2-positive cells between mice grafted with Bgn-OE versus control BTICs. F, Representative confocal microscopy images of mouse GBM tissues implanted with Bgn-OE versus control BTICs. Tissue was labeled with DAPI, SOX2, and β-catenin. G, Graph comparing the percentage of β-catenin positive cells between mice grafted with Bgn-OE versus control BTICs. H, Representative confocal microscopy images of mouse GBM tissues implanted with Bgn-OE versus control BTICs. Tissues were labeled with DAPI, SOX2, and CD44. I, Graph comparing the percentage of CD44-positive cells between mice implanted with Bgn-OE versus control BTICs. Data are representative of two to three separate experiments. Means were compared between groups by unpaired (two-tailed) t test. **, P < 0.01; ***, P < 0.001. Data are presented as the mean ± SEM (error bars).

Close modal

We also assessed the association between BGN and LRP6 with patient survival in different public databases. While we could not observe significant differences in the TCGA database, in the Chinese Glioma Genome Atlas (CGGA) database with a higher sample size, above median BGN expression was associated with worse overall survival in patients with brain tumors (Supplementary Fig. S10).

In this study, spatial transcriptomics was used to analyze a model of GBM, which identified 10 distinct cell clusters in the brains of mice. These clusters exhibited different transcriptomic profiles and were distributed in specific anatomical locations of the tissues, including the highly proliferative tumor core, the tumor margin, and nonmalignant regions. The intratumoral heterogeneity identified in our study may not be comparable to that reported in scRNA-seq studies (4, 23). This could be due to the limited resolution of the Visium transcriptomics that obscure some levels of heterogeneity at the single-cell level. Therefore, future studies employing an elevated version of spatial transcriptomics are required to uncover regional heterogeneity at higher resolution. However, spatial transcriptomics allowed us to characterize different clusters that were spatially distributed in tumor-bearing mice. Notably, unbiased spatial comparisons are not possible with scRNA-seq because tissue dissociation is required.

Among upregulated genes in the tumor area, we found an increase in 8 ECM compounds, including different types of collagens, (Col2a1, Col9a1, Col9a2, Col11a1, Col11a2). In line with our findings, GBM tissues have higher levels of collagen expression than normal brain tissues (32–34). Furthermore, higher collagen gene expression was associated with a poor prognosis in patients with GBM (35). However, the roles of the various collagens found in our study in promoting intratumoral heterogeneity of BTICs have yet to be thoroughly studied. FMOD was another ECM molecule that was increased in our spatial and single-cell transcriptomic analysis of GBMs. Interestingly, FMOD through interaction with collagen promotes glioma cell migration and invasion (36). Upregulation of genes encoding CSPGs such as versican was also observed in the GBM microenvironment compared with normal tissues. Other studies, like ours, have shown higher levels of CSPGs in GBM tissues than in normal brains (37, 38).

Among identified ECM molecules, scRNA-seq studies showed considerable expression of biglycan on BTICs. However, protein expression analyses of tissue specimens revealed expression of biglycan on other cell types in addition to BTICs. This is consistent with prior research that found biglycan expression on endothelial and neuronal cells in the brain tumor microenvironment (39, 40). When we focused on BTIC populations, we found that biglycan was mostly expressed by injury response compared with developmental phenotypes. The TCGA and scRNA-seq analyses affirmed that the mesenchymal subtype of GBM had significantly higher transcript levels of BGN than the other subtypes.

Distinct regions have been identified within the tumor microenvironment that support the maintenance and growth of cancer stem cells. One of the well-known niches in the brain tumor microenvironment is the perivascular areas (41). Other studies have shown that different ECM molecules such as laminins and their receptors are key structural components of the perivascular niches (42–44). While in our study we noticed an elevated expression of biglycan on endothelial cells, there was no significant accumulation of biglycan-positive BTICs around perivascular niches. Therefore, investigating the expression of biglycan in other BTIC niches such as peri-necrotic niches and microglia/macrophage niches may be of high interest in future studies.

Biglycan was found to bind with LRP6, hence activating the Wnt/β-catenin pathway and cell growth. β-catenin is a key downstream effector in the Wnt signaling pathway and its expression is used to determine activity of the Wnt signaling pathway (45, 46). In addition to tumor cells, β-catenin expression and the Wnt signaling pathway play roles in other cell types such as immune cells, which is consistent with our scRNA-seq analysis (47).

The Wnt/β-catenin pathway is associated with the epithelial–mesenchymal transition (48). Because the mesenchymal state is linked with a poor prognosis, the biglycan–LRP6 axis may be a new therapeutic target in GBMs. While the lack of survival data from mice implanted with Bgn-OE BTICs is a limitation of our study, we analyzed patient survival data from public databases. We found brain tumor patients with higher BGN expression had worse survival than other individuals in the CGGA database. However, we could not observe any difference in the TCGA database. This might be because of the higher sample size in CGGA, which includes different types of brain tumors. Furthermore, because LRP6 is expressed by numerous cell types within the tumor microenvironment and its expression may have distinct impacts on different cell types, analyzing bulk RNA-seq databases may disguise the effects of LRP6 expressed on tumor cells on patient survival.

Currently, identifying BTICs poses a challenge as there are no universally accepted markers that specifically identify them. In this study for the identification of BTICs in the scRNA-seq analysis, we employed CytoTRACE trajectory reconstruction analysis, which uses gene counts and expression to predict the differentiation state of cells from scRNA-seq data (25). On the basis of this analysis, we identified four cell clusters with lower differentiation phenotypes. Because there is a phenotypic transition between cancer stem cells and differentiated cells, it would be difficult to define a line separating BTICs from differentiated GBM cells. However, we found that BGN expression was mostly accumulated in cells with higher stemness properties. We acknowledge that the identification of the BTIC population in this study was based on their transcriptomic profiles and their absolute identification required functional studies.

In summary, spatial transcriptomic and scRNA-seq analyses of the mouse and human GBM microenvironment reveal unique transcriptional alterations in the expression of ECM genes associated with the heterogeneity of BTICs. Exploiting these datasets highlights biglycan, which is produced by injury response BTICs and contributes to their proliferation through interaction with LRP6. New investigations into other gene signatures identified in this study may provide new candidates for abrogating tumor progression.

P. Bose reports other support from OncoHelix Inc. outside the submitted work. M. Gallo received grants from CIHR and NSERC, and as Canada Research Chair during the conduct of the study. No disclosures were reported by the other authors.

R. Mirzaei: Conceptualization, data curation, software, formal analysis, validation, visualization, methodology, writing–original draft, writing–review and editing. C. D'Mello: Software, formal analysis, methodology, writing–review and editing. M. Liu: Formal analysis, investigation, writing–review and editing. A. Nikolic: Software, investigation, writing–review and editing. M. Kumar: Software, writing–review and editing. F. Visser: Methodology, writing–review and editing. P. Bose: Software, writing–review and editing. M. Gallo: Software, supervision, writing–review and editing. V.W. Yong: Conceptualization, resources, supervision, funding acquisition, validation, project administration, writing–review and editing.

The authors thank the generosity of Anita C. Bellail and Chunhai Hao from Indiana University School of Medicine who kindly provided the human GBM specimens. They acknowledge the Hotchkiss Brain Institute Advanced Microscopy Platform (HBI AMP) and the Cumming School of Medicine for support and use of Leica TCS SP8, ImageXpress Micro XLS High-Content Analysis System and image analysis platforms. The authors are grateful to D. Elliott at the HBI AMP for his significant assistance with the Imaris image analysis. They thank the Snyder Institute's Live Cell Imaging Resource laboratory at the University of Calgary for support and use of IncuCyte Live-Cell Analysis System. The authors also thank the BTIC Core and S. Weiss, G. Cairncross, A. Luchman, and R. Hassam for isolating BTIC lines from patient-resected specimens as well as the flow cytometry core and the UCDNA sequencing at the University of Calgary. R. Mirzaei was supported by a fellowship from the University of Calgary's Eyes High program. This study was supported by grants from the Canadian Institutes of Health Research and the Canadian Cancer Society.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.

Note: Supplementary data for this article are available at Cancer Research Online (http://cancerres.aacrjournals.org/).

1.
Louis
DN
,
Perry
A
,
Wesseling
P
,
Brat
DJ
,
Cree
IA
,
Figarella-Branger
D
, et al
.
The 2021 WHO classification of tumors of the central nervous system: a summary
.
Neuro Oncol
2021
;
23
:
1231
51
.
2.
Stupp
R
,
Mason
WP
,
van den Bent
MJ
,
Weller
M
,
Fisher
B
,
Taphoorn
MJ
, et al
.
Radiotherapy plus concomitant and adjuvant temozolomide for glioblastoma
.
N Engl J Med
2005
;
352
:
987
96
.
3.
Stupp
R
,
Taillibert
S
,
Kanner
A
,
Read
W
,
Steinberg
D
,
Lhermitte
B
, et al
.
Effect of tumor-treating fields plus maintenance temozolomide vs maintenance temozolomide alone on survival in patients with glioblastoma: a randomized clinical trial
.
JAMA
2017
;
318
:
2306
16
.
4.
Neftel
C
,
Laffy
J
,
Filbin
MG
,
Hara
T
,
Shore
ME
,
Rahme
GJ
, et al
.
An integrative model of cellular states, plasticity, and genetics for glioblastoma
.
Cell
2019
;
178
:
835
49
.
5.
Gangoso
E
,
Southgate
B
,
Bradley
L
,
Rus
S
,
Galvez-Cancino
F
,
McGivern
N
, et al
.
Glioblastomas acquire myeloid-affiliated transcriptional programs via epigenetic immunoediting to elicit immune evasion
.
Cell
2021
;
184
:
2454
70
.
6.
Venkataramani
V
,
Yang
Y
,
Schubert
MC
,
Reyhan
E
,
Tetzlaff
SK
,
Wißmann
N
, et al
.
Glioblastoma hijacks neuronal mechanisms for brain invasion
.
Cell
2022
;
185
:
2899
917
.
7.
Guilhamon
P
,
Chesnelong
C
,
Kushida
MM
,
Nikolic
A
,
Singhal
D
,
MacLeod
G
, et al
.
Single-cell chromatin accessibility profiling of glioblastoma identifies an invasive cancer stem cell population associated with lower survival
.
Elife
2021
;
10
:
e64090
.
8.
Mirzaei
R
,
Yong
VW
.
Microglia-T cell conversations in brain cancer progression
.
Trends Mol Med
2022
;
28
:
951
63
.
9.
Richards
LM
,
Whitley
OKN
,
MacLeod
G
,
Cavalli
FMG
,
Coutinho
FJ
,
Jaramillo
JE
, et al
.
Gradient of developmental and injury response transcriptional states defines functional vulnerabilities underpinning glioblastoma heterogeneity
.
Nat Cancer
2021
;
2
:
157
73
.
10.
Bowman
RL
,
Wang
Q
,
Carro
A
,
Verhaak
RG
,
Squatrito
M
.
GlioVis data portal for visualization and analysis of brain tumor expression datasets
.
Neuro Oncol
2017
;
19
:
139
41
.
11.
Kelly
JJP
,
Stechishin
O
,
Chojnacki
A
,
Lun
X
,
Sun
B
,
Senger
DL
, et al
.
Proliferation of human glioblastoma stem cells occurs independently of exogenous mitogens
.
Stem Cells
2009
;
27
:
1722
33
.
12.
Mirzaei
R
,
Gordon
A
,
Zemp
FJ
,
Kumar
M
,
Sarkar
S
,
Luchman
HA
, et al
.
PD-1 independent of PD-L1 ligation promotes glioblastoma growth through the NF-κB pathway
.
Sci Adv
2021
;
7
:
eabh2148
.
13.
Dzikowski
L
,
Mirzaei
R
,
Sarkar
S
,
Kumar
M
,
Bose
P
,
Bellail
A
, et al
.
Fibrinogen in the glioblastoma microenvironment contributes to the invasiveness of brain tumor–initiating cells
.
Brain Pathol
2021
;
31
:
e12947
.
14.
Pisklakova
A
,
McKenzie
B
,
Zemp
F
,
Lun
X
,
Kenchappa
RS
,
Etame
AB
, et al
.
M011L-deficient oncolytic myxoma virus induces apoptosis in brain tumor–initiating cells and enhances survival in a novel immunocompetent mouse model of glioblastoma
.
Neuro-oncol
2016
;
18
:
1088
98
.
15.
Reilly
KM
,
Loisel
DA
,
Bronson
RT
,
McLaughlin
ME
,
Jacks
T
.
Nf1;Trp53 mutant mice develop glioblastoma with evidence of strain-specific effects
.
Nat Genet
2000
;
26
:
109
13
.
16.
Rawji
KS
,
Young
AMH
,
Ghosh
T
,
Michaels
NJ
,
Mirzaei
R
,
Kappen
J
, et al
.
Niacin-mediated rejuvenation of macrophage/microglia enhances remyelination of the aging central nervous system
.
Acta Neuropathol
2020
;
139
:
893
909
.
17.
Mirzaei
R
,
Sarkar
S
,
Dzikowski
L
,
Rawji
KS
,
Khan
L
,
Faissner
A
, et al
.
Brain tumor–initiating cells export tenascin-C associated with exosomes to suppress T-cell activity
.
Oncoimmunology
2018
;
7
:
e1478647
.
18.
Nikolic
A
,
Singhal
D
,
Ellestad
K
,
Johnston
M
,
Shen
Y
,
Gillmor
A
, et al
.
Copy-scAT: deconvoluting single-cell chromatin accessibility of genetic subclones in cancer
.
Sci Adv
2021
;
7
:
eabg6045
.
19.
Stuart
T
,
Srivastava
A
,
Madad
S
,
Lareau
CA
,
Satija
R
.
Single-cell chromatin state analysis with Signac
.
Nat Methods
2021
;
18
:
1333
41
.
20.
Schep
AN
,
Wu
B
,
Buenrostro
JD
,
Greenleaf
WJ
.
chromVAR: inferring transcription-factor-associated accessibility from single-cell epigenomic data
.
Nat Methods
2017
;
14
:
975
8
.
21.
Butler
A
,
Hoffman
P
,
Smibert
P
,
Papalexi
E
,
Satija
R
.
Integrating single-cell transcriptomic data across different conditions, technologies, and species
.
Nat Biotechnol
2018
;
36
:
411
20
.
22.
Wilson
AA
,
Murphy
GJ
,
Hamakawa
H
,
Kwok
LW
,
Srinivasan
S
,
Hovav
AH
, et al
.
Amelioration of emphysema in mice through lentiviral transduction of long-lived pulmonary alveolar macrophages
.
J Clin Invest
2010
;
120
:
379
89
.
23.
Patel
AP
,
Tirosh
I
,
Trombetta
JJ
,
Shalek
AK
,
Gillespie
SM
,
Wakimoto
H
, et al
.
Single-cell RNA-seq highlights intratumoral heterogeneity in primary glioblastoma
.
Science
2014
;
344
:
1396
401
.
24.
Tirosh
I
,
Izar
B
,
Prakadan
SM
,
Wadsworth
MH
,
Treacy
D
,
Trombetta
JJ
, et al
.
Dissecting the multicellular ecosystem of metastatic melanoma by single-cell RNA-seq
.
Science
2016
;
352
:
189
96
.
25.
Gulati
GS
,
Sikandar
SS
,
Wesche
DJ
,
Manjunath
A
,
Bharadwaj
A
,
Berger
MJ
, et al
.
Single-cell transcriptional diversity is a hallmark of developmental potential
.
Science
2020
;
367
:
405
11
.
26.
Zhu
Z
,
Mesci
P
,
Bernatchez
JA
,
Gimple
RC
,
Wang
X
,
Schafer
ST
, et al
.
Zika virus targets glioblastoma stem cells through a SOX2-integrin α
.
Cell Stem Cell
2020
;
26
:
187
204
.
27.
Bohm
AK
,
DePetro
J
,
Binding
CE
,
Gerber
A
,
Chahley
N
,
Berger
ND
, et al
.
In vitro modeling of glioblastoma initiation using PDGF-AA and p53-null neural progenitors
.
Neuro Oncol
2020
;
22
:
1150
61
.
28.
Babelova
A
,
Moreth
K
,
Tsalastra-Greul
W
,
Zeng-Brouwers
J
,
Eickelberg
O
,
Young
MF
, et al
.
Biglycan, a danger signal that activates the NLRP3 inflammasome via Toll-like and P2X receptors
.
J Biol Chem
2009
;
284
:
24035
48
.
29.
Schaefer
L
,
Babelova
A
,
Kiss
E
,
Hausser
HJ
,
Baliova
M
,
Krzyzankova
M
, et al
.
The matrix component biglycan is proinflammatory and signals through Toll-like receptors 4 and 2 in macrophages
.
J Clin Invest
2005
;
115
:
2223
33
.
30.
Berendsen
AD
,
Fisher
LW
,
Kilts
TM
,
Owens
RT
,
Robey
PG
,
Gutkind
JS
, et al
.
Modulation of canonical Wnt signaling by the extracellular matrix component biglycan
.
Proc Natl Acad Sci USA
2011
;
108
:
17022
7
.
31.
Cselenyi
CS
,
Jernigan
KK
,
Tahinci
E
,
Thorne
CA
,
Lee
LA
,
Lee
E
.
LRP6 transduces a canonical Wnt signal independently of Axin degradation by inhibiting GSK3's phosphorylation of beta-catenin
.
Proc Natl Acad Sci USA
2008
;
105
:
8032
7
.
32.
Tsai
HF
,
Chang
YC
,
Li
CH
,
Chan
MH
,
Chen
CL
,
Tsai
WC
, et al
.
Type V collagen alpha 1 chain promotes the malignancy of glioblastoma through PPRC1-ESM1 axis activation and extracellular matrix remodeling
.
Cell Death Discov
2021
;
7
:
313
.
33.
Mammoto
T
,
Jiang
A
,
Jiang
E
,
Panigrahy
D
,
Kieran
MW
,
Mammoto
A
.
Role of collagen matrix in tumor angiogenesis and glioblastoma multiforme progression
.
Am J Pathol
2013
;
183
:
1293
305
.
34.
Huijbers
IJ
,
Iravani
M
,
Popov
S
,
Robertson
D
,
Al-Sarraj
S
,
Jones
C
, et al
.
A role for fibrillar collagen deposition and the collagen internalization receptor endo180 in glioma invasion
.
PLoS One
2010
;
5
:
e9808
.
35.
Yin
W
,
Zhu
H
,
Tan
J
,
Xin
Z
,
Zhou
Q
,
Cao
Y
, et al
.
Identification of collagen genes related to immune infiltration and epithelial–mesenchymal transition in glioma
.
Cancer Cell Int
2021
;
21
:
276
.
36.
Mondal
B
,
Patil
V
,
Shwetha
SD
,
Sravani
K
,
Hegde
AS
,
Arivazhagan
A
, et al
.
Integrative functional genomic analysis identifies epigenetically regulated fibromodulin as an essential gene for glioma cell migration
.
Oncogene
2017
;
36
:
71
83
.
37.
Pibuel
MA
,
Poodts
D
,
Díaz
M
,
Hajos
SE
,
Lompardía
SL
.
The scrambled story between hyaluronan and glioblastoma
.
J Biol Chem
2021
;
296
:
100549
.
38.
Chu
YH
,
Liao
WC
,
Ho
YJ
,
Huang
CH
,
Tseng
TJ
,
Liu
CH
.
Targeting chondroitin sulfate reduces invasiveness of glioma cells by suppressing CD44 and integrin β1 expression
.
Cells
2021
;
10
:
3594
.
39.
Ghochani
Y
,
Muthukrishnan
SD
,
Sohrabi
A
,
Kawaguchi
R
,
Condro
MC
,
Bastola
S
, et al
.
A molecular interactome of the glioblastoma perivascular niche reveals integrin binding sialoprotein as a mediator of tumor cell migration
.
Cell Rep
2022
;
41
:
111511
.
40.
Chen
S
,
Guo
D
,
Lei
B
,
Bi
J
,
Yang
H
.
Biglycan protects human neuroblastoma cells from nitric oxide-induced death by inhibiting AMPK-mTOR mediated autophagy and intracellular ROS level
.
Biotechnol Lett
2020
;
42
:
657
68
.
41.
Calabrese
C
,
Poppleton
H
,
Kocak
M
,
Hogg
TL
,
Fuller
C
,
Hamner
B
, et al
.
A perivascular niche for brain tumor stem cells
.
Cancer Cell
2007
;
11
:
69
82
.
42.
Lathia
JD
,
Li
M
,
Hall
PE
,
Gallagher
J
,
Hale
JS
,
Wu
Q
, et al
.
Laminin alpha 2 enables glioblastoma stem cell growth
.
Ann Neurol
2012
;
72
:
766
78
.
43.
Lathia
JD
,
Gallagher
J
,
Heddleston
JM
,
Wang
J
,
Eyler
CE
,
Macswords
J
, et al
.
Integrin alpha 6 regulates glioblastoma stem cells
.
Cell Stem Cell
2010
;
6
:
421
32
.
44.
Haas
TL
,
Sciuto
MR
,
Brunetto
L
,
Valvo
C
,
Signore
M
,
Fiori
ME
, et al
.
Integrin α7 is a functional marker and potential therapeutic target in glioblastoma
.
Cell Stem Cell
2017
;
21
:
35
50
.
45.
Conlon
TM
,
John-Schuster
G
,
Heide
D
,
Pfister
D
,
Lehmann
M
,
Hu
Y
, et al
.
Inhibition of LTβR signaling activates WNT-induced regeneration in lung
.
Nature
2020
;
588
:
151
6
.
46.
Tan
SH
,
Swathi
Y
,
Tan
S
,
Goh
J
,
Seishima
R
,
Murakami
K
, et al
.
AQP5 enriches for stem cells and cancer origins in the distal stomach
.
Nature
2020
;
578
:
437
43
.
47.
Pukrop
T
,
Klemm
F
,
Hagemann
T
,
Gradl
D
,
Schulz
M
,
Siemes
S
, et al
.
Wnt 5a signaling is critical for macrophage-induced invasion of breast cancer cell lines
.
Proc Natl Acad Sci USA
2006
;
103
:
5454
9
.
48.
Gorka
J
,
Marona
P
,
Kwapisz
O
,
Waligórska
A
,
Pospiech
E
,
Dobrucki
JW
, et al
.
MCPIP1 inhibits Wnt/β-catenin signaling pathway activity and modulates epithelial–mesenchymal transition during clear cell renal cell carcinoma progression by targeting miRNAs
.
Oncogene
2021
;
40
:
6720
35
.

Supplementary data