Molecular crosstalk between tumour and brain parenchyma instructs histopathological features in glioblastoma

Abstract

The histopathological and molecular heterogeneity of glioblastomas represents a major obstacle for effective therapies. Glioblastomas do not develop autonomously, but evolve in a unique environment that adapts to the growing tumour mass and contributes to the malignancy of these neoplasms. Here, we show that patient-derived glioblastoma xenografts generated in the mouse brain from organotypic spheroids reproducibly give rise to three different histological phenotypes: (i) a highly invasive phenotype with an apparent normal brain vasculature, (ii) a highly angiogenic phenotype displaying microvascular proliferation and necrosis and (iii) an intermediate phenotype combining features of invasion and vessel abnormalities. These phenotypic differences were visible during early phases of tumour development suggesting an early instructive role of tumour cells on the brain parenchyma. Conversely, we found that tumour-instructed stromal cells differentially influenced tumour cell proliferation and migration in vitro, indicating a reciprocal crosstalk between neoplastic and non-neoplastic cells. We did not detect any transdifferentiation of tumour cells into endothelial cells. Cell type-specific transcriptomic analysis of tumour and endothelial cells revealed a strong phenotype-specific molecular conversion between the two cell types, suggesting co-evolution of tumour and endothelial cells. Integrative bioinformatic analysis confirmed the reciprocal crosstalk between tumour and microenvironment and suggested a key role for TGFβ1 and extracellular matrix proteins as major interaction modules that shape glioblastoma progression. These data provide novel insight into tumour-host interactions and identify novel stroma-specific targets that may play a role in combinatorial treatment strategies against glioblastoma.

Keywords: angiogenesis; endothelial cells; glioblastoma; patient-derived xenograft; tumour microenvironment.

Figure 1. Phenotypic intertumoural heterogeneity in patient-derived glioblastoma xenografts

Patient-derived tumour spheroids were implanted intracranially in NOD/Scid mice. Based on histological features, xenografts derived from different patients were classified into three groups: highly invasive, intermediate and angiogenic phenotypes. A. Hematoxylin/Eosin staining showing representative histology of the three phenotypic groups: invasive (P8), intermediate (P3) and angiogenic (P13). Tumour centre, contralateral hemisphere and blood vessel morphology are presented in higher magnification. Black arrows indicate areas of necrosis with pseudopalisading cells in the angiogenic tumour. Blue arrow highlights rare enlarged vessels in intermediate tumours. White arrows point to vessels with glomeruloid structures typical for angiogenic xenografts. Scale bars represent respectively 1 mm (black) and 100 μm (white). B. Presence of tumour cells in the tumour centre and varying levels of invasion to contralateral hemispheres in representative phenotypes were confirmed with human-specific nestin immunohistochemistry. C. T2 weighted MR images confirmed the presence of oedema in angiogenic and intermediate tumours (bright field at the tumour side). T1 weighted MRI scans with contrast agent showed the absence of contrast enhancement in invasive tumours, while intermediate and angiogenic tumours showed low and high contrast enhancement respectively (n = 4) (examples shown for P8 invasive, P3 intermediate and P13 angiogenic tumours). See Suppl. Table 1 and 2 for more information on patient material and PDX models.

Figure 2. Time course of glioblastoma development and neuropathological analysis in patient-derived xenografts

A. Tumours representing the three phenotypes were analysed at post-implantation day 25, 35 and at day of sacrifice (‘End’). Hematoxylin-Eosin (magnification) and human nestin-specific stainings (inserts) reveal clear histological differences between the phenotypes at all stages of tumour development. Black arrows represent areas of necrosis with pseudopalisading cells in angiogenic xenografts and white arrows show vessels with glomeruloid structures typically seen in angiogenic xenografts already at early time points. Black and white scale bars represent 1mm and 100 μm respectively (P8 invasive, P3 intermediate and P13 angiogenic, n = 2 per time point). B. Kaplan-Meier survival curves of xenotransplanted mice (generation 2) displaying invasive phenotypes (blue for P8, T101, T185, T233, T239, T251), intermediate phenotypes (green for P3, T16, T238, T341, T434) and angiogenic phenotype (red for P13). See Suppl. Table 2 for details on PDX models. C. Neuropathological analysis for necrosis (Yes/No), vascular proliferation (Yes/No), invasion (high/low invasion score) and cell proliferation (percentage of Ki67 positive cells) was performed for invasive (P8, T101, T185, T233, T239, T251), intermediate (P3, T16, T238, T341, T434, NCH421k) and angiogenic phenotypes (P13, NCH644); 30 mice in total. D. Kaplan-Meier survival curves of xenotransplanted mice based on the presence of neuropathological features; p values were calculated with the Wilcoxon signed-rank test; *p-value < 0.05, **p-value < 0.01, ***p-value < 0.001.

Figure 3. Vascular morphology in glioblastoma phenotypes

A. Blood vessels from normal and tumour containing brains were visualized by mouse-specific anti-CD31 immunohistochemistry (Scale bars 100μm). Marked differences in blood vessel morphology were observed in the three tumour types (examples shown for P8 invasive, P3 intermediate and P13 angiogenic). B. Quantification of mCD31 staining to determine average vessel size and total vessel area. Results are displayed as mean +/− SEM. Xenografts used for quantification: (1) Invasive: T101, T185, T239, T251, P8; (2) Intermediate: P3, T16, T238, T434, T341, NCH421k; (3) Angiogenic: P13, NCH644 (n = 3 for each patient-derived xenograft; *p value < 0.05, **p value < 0.01, ***p value < 0.001). C. Functional flow cytometric analysis of CD31 positive cells in microenvironment (green, left) and tumour (black, right) compartments (examples shown for P3). Mouse CD31 positive ECs (blue) displayed the side population (SP) phenotype typical of brain endothelium. Rare human tumour cells with very low CD31 positivity were never in the SP. See Suppl. Figure 3 for more details on gating strategy.

Figure 4. Tumour-instructed stromal cells affect tumour cell behaviour in vitro

A. Sorted eGFP-negative tumour cells were plated in agar coated plates to form stroma-free spheroids (T). To obtain stroma-containing spheroids (T+S), sorted eGFP-negative tumour cells were premixed with sorted eGFP-positive stromal cells. B. Effect of stromal cells from different phenotypes on tumour spheroid growth. Stroma-containing spheroids were prepared with 500 stromal cells of the same tumour (P8T+P8S; P3T+ P3S; 25% of stromal cells) or with 500 stromal cells of another phenotype (P8T+ P3S; P3T+ P8S, 25% of stromal cells) pre-mixed with 1500 tumour cells. Two stroma-free control spheroids were considered: ‘100%’ control (2000 tumour cells) and ‘75%’ control (1500 tumour cells). Spheroids were cultured for 14 days and the spheroid area was determined at day 14 and 7. Results are presented as the percentage of the control spheroid (100%) of respective tumour cells. Images (insert below graphs) show representative spheroids after 14 days of culture. Scale bar represents 100μm; *p value < 0.05, **p value < 0.01, ***p value < 0.001 C. Tumour cells with or without stromal cells were plated in a Boyden chamber pre-coated with Matrigel and incubated for 96h; *p value < 0.05, **p value < 0.01).

Figure 5. Tumour cell-specific gene expression profile of the three phenotypes

A. Sorted tumour cells of invasive (P8), intermediate (P3) and angiogenic (P13) glioma xenografts were used for transcriptomic analysis (n = 3). Heatmap representing gene expression levels of differentially expressed genes between angiogenic, intermediate and invasive tumours (FDR < 0.01, abs(FC) > = 2). See Suppl. Table 4 for detailed gene lists. B. Revigo summary of main GO terms (DAVID^® database) up-regulated in angiogenic versus invasive tumour cells (FDR < 0.01, FC > = 2). Colour scale represents the log10 p-value; see Suppl. Table 5 for complete list of GO terms. C. Western Blot analysis showing increased levels of HIF1α and HIF2α proteins in angiogenic (Ang) and intermediate (Interm) tumours. Representative images were cropped from the same blots. D. Immunohistochemistry for MCT4 lactate transporter shows increased expression in the perinecrotic niches of angiogenic tumours. E. Elisa-based quantification revealed a gradual increase of VEGF in xenografts from invasive to angiogenic phenotype (mean +/− SEM; n = 3).

Figure 6. Phenotypic adaptation of endothelial cells in angiogenic tumours

A. Sorted ECs (eGFP⁺CD31⁺) of normal brain (NB), invasive (P8), intermediate (P3) and angiogenic (P13) glioblastoma xenografts were used for gene expression analysis. Cluster analysis of gene expression profiles indicates that xenograft-derived ECs are more closely related to each other and differ from normal brain ECs. B. Revigo summary of main GO terms (DAVID^® database) up-regulated in ECs within angiogenic tumours compared to normal brain (FDR < 0.01, any fold change). Color scale represents the log10 p-value. See Suppl. Table 7 for complete list of GO terms. C. Venn diagram showing the comparison of up-regulated DEGs (see Suppl. Table 8 for detailed list of DEGs). D.-E.. Immunohistochemistry confirming increased expression of Angiopoietin 2 (Angpt2), D and Thrombospondin 1 (Thbs1), E in ECs of the intermediate and angiogenic tumours (D: scale bar 30μm; E: scale bar 20μm). Note that increased expression is also visible in the tumour cells of the angiogenic phenotype. F. Integrative protein-protein interaction analysis using genes up-regulated in tumour cells and ECs of the angiogenic phenotype (selected tumour list: cell membrane and extracellular matrix proteins, P13vsP8 cells, FDR < 0.01, FC > = 2; selected EC list: cell membrane and extracellular matrix proteins P13vsNB, FDR < 0.01, FC > 1;). The network shows direct protein-to-protein interactions between tumour-specific modules (‘grey’) and EC-specific modules (‘green’), as well as indirect interactions via putative protein partners (‘yellow’). Genes expressed by both modules are indicated in the ‘red’ module. Number of genes per module is displayed in brackets. G. Selected network of protein-protein interactions for THBS1. THBS1 and its first neighbours are displayed in a circular layout grouped by category (green: expressed in ECs; grey: expressed in tumour; red: expressed in both cell types). Only the direct interactions between molecules up-regulated in tumor and ECs are shown. (THBS1: thrombospondin-1, FBN1: fibrillin-1, ECM1: extracellular matrix protein-1, COL1A1: collagen-1a1, COL1BA1: collagen-1ba1, ITGB3: integrin-b3, TGFB1: transforming growth factor B1, ITGB1: integrin-b1, MMP2: matrix metalloproteinase-2, TNFRSF11B: tumour necrosis factor receptor superfamily, member 11b, IGFBP5: insulin-like growth factor binding protein 5, VEGFA: vascular endothelial growth factor 1, TFP1 : transferrin pseudogene 1, BGN : biglycan, DCN : decorin, JAG1 : jagged 1, ITGA4 : integrin-a4, FN1 : finbronectin-1, COL4A1 : collagen-4a1).