Ex vivo Dynamics of Human Glioblastoma Cells in a Microvasculature-on-a-Chip System Correlates with Tumor Heterogeneity and Subtypes

Abstract

The perivascular niche (PVN) plays an essential role in brain tumor stem-like cell (BTSC) fate control, tumor invasion, and therapeutic resistance. Here, a microvasculature-on-a-chip system as a PVN model is used to evaluate the ex vivo dynamics of BTSCs from ten glioblastoma patients. BTSCs are found to preferentially localize in the perivascular zone, where they exhibit either the lowest motility, as in quiescent cells, or the highest motility, as in the invasive phenotype, with migration over long distance. These results indicate that PVN is a niche for BTSCs, while the microvascular tracks may serve as a path for tumor cell migration. The degree of colocalization between tumor cells and microvessels varies significantly across patients. To validate these results, single-cell transcriptome sequencing (10 patients and 21 750 single cells in total) is performed to identify tumor cell subtypes. The colocalization coefficient is found to positively correlate with proneural (stem-like) or mesenchymal (invasive) but not classical (proliferative) tumor cells. Furthermore, a gene signature profile including PDGFRA correlates strongly with the "homing" of tumor cells to the PVN. These findings demonstrate that the model can recapitulate in vivo tumor cell dynamics and heterogeneity, representing a new route to study patient-specific tumor cell functions.

Keywords: brain tumor dynamics; ex vivo assays; microvasculature; organ‐on‐a‐chip.

Figure 1

Growth of BTSC‐incorporated microvasculature‐on‐a‐chip. a) Microfluidic device (250 µm in height) containing a cell/gel loading microchamber (8000 µm × 1300 µm) flanked by two medium flow channels (500 µm in width). An array of triangular microposts separate the gel chamber and the medium flow channel, allowing for loading and confining the hydrogel precursor to the midchamber only. Scale bar: 1000 µm. b) Representative time course images of the microvessel formation over a period of 6 d. Scale bar: 15 µm. Green: GFP‐HUVEC. c) Whole chip scan showing microvasculature formation (96 h postcell in fibrin) and loading of single BTSCs (GS5). Green: GFP‐HUVECs. Red: BTSCs. Scale bar: 1000 µm. d) Comsol Finite Element Simulation of flow velocity magnitude (mm s⁻¹). The finite element model was constructed using the experimental whole‐chip microvessel network in (c). e) Immunostaining of VE‐cadherin and vWF to examine the formation of adherens junction and the function of tissue‐engineered endothelial vessels at day 3. Scale Bar: 20 µm. f) Cross‐sectional confocal image showing two adjacent microvessels and collagen IV deposition. Green: GFP‐HUVECs. Red: collagen IV. g) Infusion of 70 kDa fluorescent dextran to measure impermeability of the lumen and examine microvessel opening to the media channel (anastomosis). Green: GFP‐HUVECs. Red: Dextran. Scale Bar: 110 µm. h) Flowing fluorescent microbeads (red) (10um) through a microvessel network (green) that were grown for 3 d. Scale Bar: 100 µm. i) An SEM image of the microvessel. Scale Bar: 5 µm.

Figure 2

Quantification of tumor cell localization relative to microvessels. a) Phase and fluorescent images of microvessels with BTSC GS5. BTSCs were incubated with the Dil membrane dye for 40 min prior to coculture with GFP‐HUVECs. BSTCs localize more preferentially to the branching points of the microvessel network. Scale bar: 80 µm. b) Fluorescent images of GBM cell line RFP‐U87 cells loaded in a microvessel network. RFP‐U87 cells randomly distributed in the gel space and the microvessels constantly remodeled. Scale bar: 80 µm. c) Quantitative analysis of colocalization of GS5 versus U87 cells with the microvessels. d) Confocal images to examine colocalization of GS5 cells and microvessels. Red: GS5 cells. Green: HUVECs. Scale Bar: 100 µm. e) SEM image (false color) showing a BTSC on the microvessel. Scale:5 µm.

Figure 3

Tumor cell migration in the microvasculature‐on‐a‐chip. a) SEM image (false color) of a typical microvessel network (green) with GS5 cells (red). Scale: 50 µm. b) Fluorescence images of a representative region at different time points to track tumor cell migration. c) Migration trajectory of 13 tumor cells in the region shown in (b) for a day 3 microchip and measured over a period of 20 h. Axes unit: pixel. d) Average motility (µm h⁻¹) of single tumor cells from three groups defined by the initial location of tumor cells relative to microvessels. On‐Vessel (10.29 ± 7.23 µm, n = 28); Proximity (4.60 ± 5.11 µm h⁻¹, n = 16); distant (5.94± 5.29 µm h⁻¹, n = 13). ANOVA test (p = 0.0133, Kruskal‐Wallis Test). e) Absolute displacement of single tumor cells from three groups. On‐Vessel (70.71 ± 75.96 µm, n = 28); proximity (19.32 ± 38.42 µm, n = 16); distant (46.28± 59.5 µm, n = 13). ANOVA test (p = 0.0042, Kruskal‐Wallis Test).

Figure 4

All patient samples: tumor cell localization relative to microvessels. a) Patient information table. All GBM samples are IDH wild‐type. M, MGMT promoter methylated; UM, MGMT promoter not methylated; NT, not tested. b) Immunostaining of nestin and Sox2. Scale: 30 µm. c) Whole chip scan of GBM6 in the microvasculature chip at Day 4. Scale: 700 µm. Red: GBM6 cells. Green: HUVECs. d) Colocalization coefficient of tumor cells and microvessels measured for all patient samples. e) Representative images of patient cells in the microvasculature chip. Red: GBM cells. Green: HUVECs. Scale: 10 µm.

Figure 5

Single‐cell RNA‐seq correlates on‐chip colocalization to transcriptional signatures/subtypes. a) tSNE plot of single cell RNA‐seq data from all patient samples. b) Single‐cell pseudotime lineage trajectory obtained by semisupervised clustering of subtype‐specific gene panel using Monocle. c) Representative images of tumor cells in microvasculature and the matched Monocle plots for top three and bottom three colocalization coefficient samples (Day 4–7). Scale bar: 200 µm. d) The linear regression model showing the percentage of GBM subtype in each sample in correlation with the colocalization R value. e) The multivariate linear model showing the predictor variables (average gene expression of PDGFRA, C1GALT1, THY1, and MKI67) as a combined transcriptional signature that correlates with colocalization. f) Relative gene expression of angiogenesis (PDGFRA, SCG2), proliferation (MKI67), and housekeeping marker (RPLP0) in single cells from all patient samples.