3D Brain Vascular Niche Model Captures Glioblastoma Infiltration, Dormancy, and Gene Signatures

Abstract

Glioblastoma (GBM) is a lethal brain cancer with no effective treatment; understanding how GBM cells respond to tumor microenvironment remains challenging as conventional cell cultures lack proper cytoarchitecture while in vivo animal models present complexity all at once. Developing a culture system to bridge the gap is thus crucial. Here, a multicellular approach is employed using human glia and vascular cells to optimize a 3D brain vascular niche model that enabled not only long-term culture of patient derived GBM cells but also recapitulation of key features of GBM heterogeneity, in particular invasion behavior and vascular association. Comparative transcriptomics of identical patient derived GBM cells in 3D and in vivo xenotransplants models revealed that glia-vascular contact induced genes concerning neural/glia development, synaptic regulation, as well as immune suppression. This gene signature displayed region specific enrichment in the leading edge and microvascular proliferation zones in human GBM and predicted poor prognosis. Gene variance analysis also uncovered histone demethylation and xylosyltransferase activity as main themes for gene adaption of GBM cells in vivo. Furthermore, the 3D model also demonstrated the capacity to provide a quiescence and a protective niche against chemotherapy.

Keywords: 3D vascular model; brain vascular niche; glia‐vascular unit; glioblastoma; human brain microvasculature; tumor quiescence.

Figure 1

Optimization of a 3D brain vascular niche model. a) Fluorescence live cell imaging of brain endothelial cells (bEC, tdTomato⁺), cultured alone in 3D gel or together with astrocytes (AC), pericytes (PC), or both at 14 days of culture. The bEC were seeded at a density of 6 × 10⁶ cells/ml and AC and PC at 3 × 10⁶ /ml. Note that bEC cultured alone or with AC only tended to form vacuoles that were not interconnected (arrowheads), while co‐culture with PC, and more so with both AC and PC, led to the formation of interconnected vessels (arrows). b) Average vessel length and average vessel diameter quantified in 3D cultures across different seeding concentrations of pericytes (PC) and astrocytes (AC). Data represent mean ± SD. n = 4–6 per condition. Statistical analysis was performed using one‐way ANOVA with Tukey's post hoc test. * P ≤ 0.05, ** P ≤ 0.01, *** P ≤ 0.001, **** P ≤ 0.0001. c) Representative images of vascular development of bEC (tdTomato⁺) in 3D gel when cultured with increasing seeding densities of AC and PC. d) Quantifications of total vessel length and vascular area in cultures with different seeding densities of AC and PC. n = 4–12 for each condition. e) Time course of formation of interconnected vascular development of bECs (arrows) from culture day 7 to day 21, with AC and PC seeded at 3 × 10⁶ cells/ml each. f) Co‐immunofluorescence imaging of day 21 culture shows juxtaposition of PC (NG2⁺) with bEC (tdTomato⁺). Note AC (GFAP⁺) extending endfeet (arrow) toward vasculature. DAPI for nuclei staining. g) Orthogonal view from different planes (x/y, x/z or y/z) of the confocal microscope images of the 3D vascular niche showing lumen formation (arrowhead) and astrocyte endfeet (arrows) contacting the vasculature. Pericytes (PC, NG2⁺), brain endothelial cells (bEC, tdTomato⁺), and astrocytes (AC, GFAP⁺) are shown.

Figure 2

3D GAPE model recapitulates GBM invasion patterns as observed in vivo. a) Schematics of 3D GBM model fabrication with GBM cells seeded either in dispersed or spheroid fashion, together with astrocytes, and brain‐derived pericytes and endothelial cells (EC). Formation of a brain vascular network within the 3D structure with close interaction with GBM cells occurred over 3 weeks of culture. b) Examples of SD2‐patient derived GBM stem cells (expressing GFP) in various 2D or 3D cultures or in intracranial in vivo transplant in SCID mice. For the 3D gel model, an increased amount of matrix was used to sustain 3D gel model for over 2 weeks of culture. c) Immunofluorescence images of patient derived GBM stem cells (GFP⁺) seeded in different 3D culture conditions in dispersed or spheroid fashion, as compared to in vivo intracranial transplants (human nuclear antigen⁺ in tumor edge, or GFP⁺ in bulk+edge). Both SD2 and SD3 exhibited similar growth patterns in the 3D G condition, but distinctive invasion patterns in the 3D GAPE model, with SD3 showing stronger vascular association, resembling in vivo invasion at tumor margin. Vasculature was visualized by tdTomato⁺ ECs in 3D GAPE model or by staining for PECAM1 for in vivo model. d) Quantification shows a higher fraction of SD3 GSCs associated with vasculature in the dispersed 3D GAPE model (day 21) compared to SD2 GSC. Unpaired two‐tailed Student's t test, n = 4 independent cultures for each condition. Data represent mean ± SD. e) Quantifications show higher total vessel length and vessel area for dispersed 3D GAPE model with SD3 GSCs compared to 3D GAPE with SD2 GSC (day 21). Unpaired two‐tailed Student's t test, n = 16 independent cultures for each condition. Bar graphs represent mean ± SD. f) Quantifications show differences in sizes of aggregates of SD2 and SD3 GSC in dispersed 3D GAPE model, but not in dispersed 3D G model. One way ANOVA with Tukey's post hoc test. n = 2 independent cultures for 3D G, n = 4 independent cultures for 3D GAPE. Bar graphs represent mean ± SD. g) Fluorescence images show growth pattern of human GBM cell line U87MG (GFP⁺) seeded as dispersed or spheroid cells in different 3D models or in in vivo intracranial transplant. Note that U87MG expanded as bulk mass without invasive pattern both in 3D and in vivo models. h) Orthogonal views from different planes (x/y, x/z, or y/z) of confocal images of the 3D GAPE model showing lumen formation (arrowhead) and SD3 GSCs (arrows) positioned near the vasculature.

Figure 3

Comparative transcriptomics reveal shared differentially expressed genes in 3D GAPE and in vivo featuring neural/synaptic regulation and immunosuppression. a) Schematic of RNA sequencing analyses of GBM cells grown in different in vitro conditions (2D, 3D G, 3D GAPE) or in vivo intracranial transplants. GBM cells from 3D GAPE or in vivo transplants were isolated by FACS for GFP⁺ cells or HLA⁺ cells, respectively. in vivo model schematics created with BioRender. b) Venn diagram illustrating overlap of differentially expressed genes (DEGs, cutoff: P<0.05 and |log₂fold change| >2) in GBM cells from different conditions relative to 2D. c) Heatmap of expression of genes in top GOs enriched in upregulated DEGs shared by 3D GAPE and in vivo relative to 2D, across all four conditions (2D, 3D G, 3D GAPE, in vivo). d) Network plot of enriched GO of shared upregulated DEGs of 3D GAPE and in vivo conditions relative to 2D. Circular nodes depict GO Biological Process, while hexagonal nodes indicate GO Cellular Component. e) GO Enrichment Analysis of shared DEGs of 3D GAPE and in vivo relative to 2D, separated into up‐ and down‐regulated DEGs. f) Enrichr analysis of shared DEGs, separated into up and downregulated genes in 3D GAPE and in vivo relative to 2D. g) Summary depiction of gene adaptation of GBM cells in 3D GAPE and in vivo featuring neural/synaptic regulation and immunosuppression.

Figure 4

Shared GBM gene signatures of 3D GAPE and in vivo are expressed at distinct zones of human GBM and predict poor survival. a,b) Gene‐concept network depiction of 5 top enriched GOs (ranked by p‐value) of the upregulated DEGs in 3D GAPE and in vivo relative to 2D (n = 409). The layout displays the relationships between significantly enriched GO terms (colored nodes with predicted biological functions) and their associated genes. c,d) Heatmap and quantification of the expression of 43 (of 50) top upregulated DEGs shared by 3D GAPE and in vivo relative to 2D in different zones of human GBM patient samples (Ivy GAP database). Note significantly higher expression of the shared DEGs in leading edge and infiltrating tumor zones, as well as microvascular proliferation zone. Mean expression scores for the 43 top shared DEGs across GBM tumor zones, normalized to the Cellular tumor (CT) zone. One‐way ANOVA, followed by Tukey's post hoc test. n = 270 specimens (n = 19‐111 for each zone) in the Ivy GAP database. Data represent mean ± SD. Note that only 43 of the top 50 genes were represented in Ivy GAP database. e) Top 50 shared upregulated DEGs of 3D GAPE and in vivo were applied for cluster analysis of the TCGA GBM Biodiscovery portal. f) Prognostic index based on expression of 50 upregulated shared DEGs from 3D GAPE and in vivo in human GBM patients, shown for GBM in total and for individual transcriptional subtypes.

Figure 5

Variable genes highlight histone demethylation for epigenetic adaptation as a main theme shared by GBM cells in 3D brain vascular niche model and in vivo. a) Principal component analysis (PCA) of RNA‐seq samples show separation of in vivo samples from in vitro samples for both SD2 and SD3, but 3D GAPE samples being closest to in vivo samples on PC1 axis than other culture conditions. b) Spearman correlation distances of cluster gene expression profiles between different in vitro conditions and in vivo samples (SD2 and SD3 combined), based on the top 1000 most variable genes. Distances were calculated using hierarchical clustering. Violin plots show median, quartiles, and minimum and maximum values. One‐way ANOVA, followed by Tukey's post hoc test. c) Left, unsupervised clustering of expression of top 1000 most variable genes across SD2 and SD3 combined samples in different conditions. Right, enriched GOs of top 1000 variable genes, separated into up and downregulated genes. d) Gene ontology enrichment analysis of top 50 most closely correlated genes of GBM cells in 3D GAPE and in vivo conditions highlight epigenetic adaptations by histone demethylation as the main biological theme. e) Top variable GO ontology terms across samples, based on geneset databases of brain regions and neural stem cells, as well as genesets associated with tumor propagation and GBM stemness.

Figure 6

The 3D brain vascular model supports GBM quiescence niche. a) Schematic of doxycycline (Dox)‐induced expression of histone 2B(H2B)‐GFP and progressive dilution of H2B‐GFP label in proliferative cells during ‐Dox chase phase, while quiescent cells retain H2B‐GFP label. b) Experimental timeline of H2B‐GFP induction (+Dox) and subsequent ‐Dox chase of GBM cell culture. c) Fluorescence images of GBM cells in different culture conditions on day 3 and day 23. Note that day 3 was the last day of Dox pulse and day 23 was after 21 days of ‐Dox chase. d) Quantification of GFP^high fluorescent areas across culture conditions and culture periods. Note that GBM cells in 3D GAPE maintained a significantly higher amount GFP^high quiescent cells than other culture conditions. One‐way ANOVA, followed by Tukey's post hoc test. n = 3 independent cultures for each condition. Bar graphs represent mean ± SD. e) Two representative immunofluorescence images show aggregation of H2B‐GFP^high quiescent GBM cells in close proximity to vasculature at day 14. f) Top left, timeline of H2B‐GFP induction (+Dox) and treatment with chemodrug temozolomide (±TMZ) of 3D cultures from day 7 to day 21. Fluorescent images and quantification show significantly more H2B‐GFP label‐retaining GBM cells in 3D GAPE than in 3D G cultures at day 21, which survived even after TMZ treatment. One‐way ANOVA, followed by Tukey's post hoc test. n = 4 independent cultures for each condition. Bar graphs represent mean ± SD.