The extraordinary plasticity of glioma cells allows them to contribute to different cellular compartments in tumor vessels, reinforcing the vascular architecture. It was recently revealed that targeting glioma-derived pericytes, which represent a big percentage of the mural cell population in aggressive tumors, increases the permeability of the vessels and improves the efficiency of chemotherapy. However, the molecular determinants of this transdifferentiation process have not been elucidated. Here we show that mutations in EGFR stimulate the capacity of glioma cells to function as pericytes in a BMX- (bone marrow and X-linked) and SOX9-dependent manner. Subsequent activation of platelet-derived growth factor receptor beta in the vessel walls of EGFR-mutant gliomas stabilized the vasculature and facilitated the recruitment of immune cells. These changes in the tumor microenvironment conferred a growth advantage to the tumors but also rendered them sensitive to pericyte-targeting molecules such as ibrutinib or sunitinib. In the absence of EGFR mutations, high-grade gliomas were enriched in blood vessels, but showed a highly disrupted blood–brain barrier due to the decreased BMX/SOX9 activation and pericyte coverage, which led to poor oxygenation, necrosis, and hypoxia. Overall, these findings identify EGFR mutations as key regulators of the glioma-to-pericyte transdifferentiation, highlighting the intricate relationship between the tumor cells and their vascular and immune milieu. Our results lay the foundations for a vascular-dependent stratification of gliomas and suggest different therapeutic vulnerabilities determined by the genetic status of EGFR.
This study identifies the EGFR-related mechanisms that govern the capacity of glioma cells to transdifferentiate into pericytes, regulating the vascular and immune phenotypes of the tumors.
Diffuse gliomas are histologically classified as low- and intermediate-grade gliomas (lower grade gliomas, LGG; grades 2 and 3) or glioblastomas (GBM; grade 4; ref. 1). The poor prognosis of these tumors has been attributed, in part, to treatment limitations related to the tumor localization, but also to the insufficient knowledge about their physiology.
Florid vascular proliferation and aberrant vasculature are distinctive pathologic hallmarks of glioma progression. Four distinct mechanisms of vascularization have been proposed in gliomas: vascular cooption, angiogenesis, vasculogenesis, and vascular mimicry. The last one is associated with the plasticity of GBM cells, which can transdifferentiate into endothelial cells (EC; ref. 2) and pericytes (3–5). All these mechanisms are interlinked and overlapped in the history of gliomas and give rise to an abnormal vascular network, with dilated and tortuous blood vessels (BV) and poor pericyte coverage, which contributes to the formation of an abnormal blood–brain barrier (BBB; ref. 6).
Whereas the mechanisms that drive glioma neovascularization have been described, the genetic alterations that govern them have not been well established. However, it is known that mutations in the isocitrate dehydrogenase 1/2 (IDH1/2) genes, commonly found in LGG, favor the normalization of the vasculature. IDH-mutant (IDHmut) gliomas have smaller BVs and are associated with less hypoxia compared with their wild-type (wt) counterparts (7). Moreover, we have recently described that the microtubule stabilizer TAU/MAPT is induced in IDHmut gliomas and promotes vascular normalization by opposing EGFR signaling (8). The EGFR gene is amplified in 50% to 60% of IDHwt GBMs and half of these tumors carry the vIII variant (exons 2–7 deletion), which generates a constitutive activation of the receptor's tyrosine kinase. In addition, missense point mutations in the extracellular domain of the receptor are also commonly found in gliomas (9). All these mutations confer a higher oncogenic potential to the tumor cells through the increase in their proliferative capacity. Moreover, EGFR activation in glioma cells promotes STAT3/5 activation and cytokine secretion to modify the microenvironment and thus drives tumorigenesis (10). This adds to the idea that the genotype of glioma driving cells determines their surrounding stroma. Here, we investigated how alterations in EGFR affect the vascular phenotype of gliomas and the implications for the progression of the tumors. Our results show the relevance of the glioma-to-pericyte transdifferentiation, driven by EGFR mutations, in shaping the vascular and the immune phenotype of the tumors. The data highlight the intricate relation between the glioma cells and their surrounding stroma.
Materials and Methods
Normal tissue (NT) was obtained postmortem from nonpathologic brain samples. NT and glioma samples were obtained after subject's written informed consent. The studies were conducted in accordance with the Declaration of Helsinki and with the approval of the Ethical Committee at Hospital 12 de Octubre (Madrid, Spain; CEI-14/023, CEI-18/024, CEI-PI 47–2015). The primary cell lines belong to the Biobank of that Hospital. Fresh tissue samples were digested enzymatically using Accumax (Millipore) and were maintained in stem cell medium; Neurobasal (Invitrogen) supplemented with B27 (1:50) (Invitrogen); GlutaMAX (1:100; Invitrogen); penicillin–streptomycin (1:100; Lonza); 0.4% heparin (Sigma-Aldrich); and 40 ng/mL EGF and 20 ng/mL bFGF2 (PeproTech). The list of primary cell lines and human samples with their genetic information is provided in Supplementary Table S1 and S2, respectively.
Mouse cell lines
Mouse subventricular zone (SVZ) cell lines were obtained by retroviral expression of EGFRwt or EGFRvIII in primary neural stem cell cultures obtained from the SVZ of p16/p19 ko mice as described previously (11) and they were grown in stem cell medium. After infection, the cells were injected into Nude mice. The tumors that grew were dissociated and the lines SVZ-EGFRwt/amp and SVZ-EGFRvIII were established. Both models express GFP and luciferase as a reporter. The GL261 murine glioma cells were maintained in DMEM plus 10% FBS, 2 mmol/L l-glutamine, 0.1% penicillin (100 U/mL), and streptomycin (100 μg/mL). All human and mouse cell lines were tested for Mycoplasma contamination routinely (every month) using a Luminescent Detection Kit (Lonza). The primary cell lines were maintained below passage number 20.
DNA constructs and lentiviral/retroviral production
Retroviral vectors used were pBabe-EGFRwt (Addgene #11011) and MSCV-XZ066-GFP-EGFRvIII (Addgene #20737). pLV-Hygro-Luciferase (VectorBuilder #VB150916–10098). Lentiviral vectors to express shRNAs were shCD248 (Sigma #SHCLNG-NM_020404: TRCN00000437821) and shSOX9 (Addgene #40644).
In vitro treatments
Cells were treated with ibrutinib (MedChemExpress, 31976) 5 μmol/L, DMSO (control), MG132 (Millipore) 10 μmol/L, or different concentrations of dacomitinib (Pfizer, PF-299804).
In vivo assays
Animal experiments were reviewed and approved by the Research Ethics and Animal Welfare Committee at our institution (Instituto de Salud Carlos III, Madrid, Spain; PROEX 244/14 and 02/16), in agreement with the European Union and national directives. Intracranial or subcutaneoous transplantations into Nude mice were established as described previously (12). Mice were treated with ibrutinib at 12 mg/kg/day through intraperitoneal injection (i.p.), sunitinib (Selleckchem, S1042) at 6 mg/kg/day (i.p.; twice a week), or dacomitinib (Pfizer, PF-299804) at 15 mg/kg/day (i.p.). For drug preparation, ibrutinib was dissolved in 4% DMSO + 10% Hydroxypropyl-β-Cyclodextrin (HP-β-CD), sunitinib was dissolved in 4% DMSO + 10% Polysorbate and dacomitinib was dissolved in 20 mmol/L sodium lactate (pH = 4; 1.5 mg/mL). Control animals were treated with these solvents. When applicable, tumor growth was monitored in an IVIS equipment (Perkin Elmer) after intraperitoneal injection of d-luciferin (75 mg/kg; PerkinElmer). For MRI analyses, animals were previously anesthetized using 2% isofluorane (Isobavet, Schering-Plough) and they were intraperitoneally injected with 0.1 mL of gadolinium (Gd)-DOTA (Dotarem, Guerbet). Images were acquired on a 4.7 T Biospec BMT 47/40 spectrometer (Bruker), equipped with a 6 cm actively shielded gradient system, capable of 450 mT/m gradient strength (Universidad Complutense CAI facility).
SVZ-EGFRwt/amp and SVZ-EGFRvIII tumors were surgically excised from nude mice and the tissue was dissociated of enzymatic digestion at room temperature with Accumax by 30 minutes. Samples then were filtered through 70-μm strainers and collected in staining medium (PBS containing 2% BSA). Live tumors cells were discriminated from dead cells using propidium iodide and GFP+ cells were isolated with BD FACS Cell Sorter. Cells were collected into 2 mL staining medium and were recovered by centrifugation for further analysis.
Chicken chorioallantoic membrane assay
For in vivo evaluation of the angiogenic inductive potential of SVZ-derived conditioned media (CM), a chicken chorioallantoic membrane (CAM) assay was performed as reported previously (13). Photographs were taken with a digital camera HD IC80 (Leica) and the number of vessels within a 6-mm radius of the scaffold were counted to determine the angiogenic score, using ImageJ software (NIH, Bethesda, MD). In each photograph, the diameter of 220 vessels was measured using ImageJ software (NIH, Bethesda, MD).
Immunofluorescence and IHC staining and quantification
Immunofluorescence (IF) and IHC staining was performed as described previously (8) using primary and secondary antibodies described in Supplementary Table S3. For quantification, slides were scanned at ×63 or ×40 magnification. The number of BrdU-positive cells per field was counted with Fiji-ImageJ software and normalized with the total number of cells. To quantify the IgG extravasation Fiji-ImageJ software was used. The signal from the endomucin channel was subtracted from the IgG channel. To quantify delocalized αSMA-positive mesenchymal cells, the signal from the endomucin channel was subtracted from the αSMA channel. For the quantification of the vasculature, we counted the number of dilated vessels (those with a visible lumen of ≥5 μm diameter) per field and the CD34 staining with the Fiji-ImageJ software, as described previously (5, 14). A similar approach was used to count the number of immune cells (CD45+ or CD68+ cells per field). Density measurements of BV density, necrotic area, and pericytes coverage, were performed with ImageJ software (http://rsb.info.nih.gov/ij). Furthermore, in case of the necrotic area, we used a score to grade the intensity of the quantified necrosis. To calculate vasculature per random field areas was measured in the intratumoral regions of tumor sections.
Flow cytometry analysis of human tumors
Tumors suspensions were obtained after mechanical and enzymatic disaggregation (Accumax; Merck Millipore; 15 minutes, room temperature and filtered through 70-μm nylon mesh cell strainer; Thermo Fisher Scientific). Erythrocytes were lysed with Quicklysis buffer (Cytognos) and cells were incubated with hFcR Blocking (Miltenyi Biotec), previous to antibody (Supplementary Table S3) incubation (20 minutes at 4°C in PBS 1% FBS). Viable cells were labelled with a Fixable Viability Stain (Becton Dickinson; 20 minutes, room temperature). The analysis was conducted in a Macsquant10 flow Cytometry (Miltenyi Biotec). The gating strategy and the definition of the lymphoid and the myeloid subsets was already described in ref. 14.
In silico analysis
The Cancer Genome Atlas (TCGA) GBM, LGG, and GBM+LGG dataset was accessed via cBioPortal (https://www.cbioportal.org/), UCSC xena-browser (https://xenabrowser.net), and Gliovis (http://gliovis.bioinfo.cnio.es) for extraction of the data: overall survival, gene's expression level and the distribution of the different genetic alterations. Kaplan–Meier survival curves were done upon stratification based into low and high groups using expression values from each gene. Significance of differences in survival between groups was calculated using the log-rank test. For the functionality studies we have used “David Gene ontolog” analysis. First, we selected a cluster of 365 genes coexpressed with SOX9, using the highest value of the Spearman correlation. Then, “David gene ontology” analysis associates the expression of thesw genes with the biological processes involved. The hypoxic-related genes signature included hypoxia and HIF1α pathways genes. Ivy Gap date set analysis (http://glioblastoma.alleninstitute.org/) was used to analyze gene signature enrichment between the different anatomic structures identified in the tumor.
GraphPad Prism 5 software was used for data presentation and statistical analysis. For bar graphs, the level of significance was determined by a two-tailed unpaired Student t test. When comparing three or more groups a one-way ANOVA test was used, followed by a t test analysis by pairs of experimental groups. For the comparison of the expression of a gene signature between two experimental groups, we performed a paired t test analysis. For that, we paired the expression of the same gene within an experimental group in two different situations. The statistical value associated with the set of expressed genes was obtained by comparing both situations. For Kaplan–Meier survival curves, the level of significance was determined by the two-tailed log-rank test. P values < 0.05 were considered significant (*, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001; n.s., nonsignificant). All quantitative data presented are the mean ± SEM. Precise experimental details (number of animals or cells and experimental replicates) are provided in the figure legends.
The authors confirm that the data supporting the findings of this study are available within the article and its Supplementary Material. Additional methodologic details, including a list of primers used for qRT-PCR analysis (Supplementary Table S4), are provided in the Supplementary Material. Cell lines are available upon request.
Stratification of gliomas by the genetic status of IDH and EGFR distinguishes between different vascular phenotypes
To begin the analysis of the vasculature of gliomas, we validated that the transcription of three recognized angiogenic markers (VEGFA, ANGPT2, and IGFBP2), showing a strong inverse correlation with the survival of patients with glioma (Supplementary Fig. S1A–S1C), even if we consider LGG (Supplementary Fig. S1D–S1F) and GBM (Supplementary Fig. S1G–S1I) separately. Moreover, the progressive increase in their expression paralleled the evolution of the glioma disease (Supplementary Fig. S1J–S1L). We then measured the frequency of mutations (Supplementary Fig. S1M–S1O) and copy-number amplifications (CNA; Supplementary Fig. S1P–S1S) in the groups with high or low quantities of the three genes. As expected, the frequency of IDH1 mutations was much higher in the gliomas that contain less angiogenic-related mRNAs (15). Among the rest of the genes analyzed, we found a consistent enrichment in mutations and CNA of EGFR and EGFR-related molecules, like ELDR (EGFR long noncoding downstream RNA) or EGFR-AS1, in the most angiogenic tumors.
Tumors with alterations in IDH1/2 and EGFR account for almost 90% of all gliomas in TCGA cohort, and they show a mutually exclusive pattern, with only a small percentage of IDHmut gliomas harboring EGFR gains (Fig. 1A). To relate the vascular features with these genetic alterations, we classified our own cohort of patient's samples in three groups, IDHmut gliomas, IDHwt gliomas without EGFR mutations (herein called EGFRwt/amp) and IDHwt/EGFRmut gliomas, and we measured the amount of the angiogenic-related mRNAs. The vasculature of IDHmut tumors was close to normal and there was no overexpression of any of the three angiogenic and vascular markers in these tumors compared with normal brain tissue (Fig. 1B–D). Among gliomas, we noticed that there was a gradual increase in the amount of the three mRNAs from the IDHmut to the IDHwt/EGFRmut gliomas, being the IDHwt/EGFRwt/amp group in the middle (Fig. 1B–D). This result was consistent with the in silico data (Supplementary Fig. S2A–S2C) and with the IHC (Fig. 1E; Supplementary Fig. S2F) and the transcriptional (Fig. 1F and G; Supplementary Fig. S2D, S2E, S2G, S2H) analysis of CD34 and CD31 [also known as platelet endothelial cell adhesion molecule (PECAM-1)], which label ECs, and αSMA, and CD248, which label pericytes. The expression of these markers showed a gradual increase in the three glioma subgroups. Moreover, we did not find much differences in terms of vascularization between normal tissue (NT) and IDHmut tumors. The parameters that changed the most among gliomas were the vascular density (Fig. 1H) and the number of dilated BVs (Fig. 1I), typical of malignant tumors (16).
Our results suggest important differences between gliomas that overexpress wild-type or mutant EGFR, probably related to the increased phosphorylation of the receptor (Supplementary Fig. S2I), and the increase in the expression of several angiogenesis-related signatures (Supplementary Fig. S2J–S2M) observed in the second group. We narrowed down the set of differentially expressed vascular genes to define a “vascular signature” and we confirmed its higher expression in EGFRmut compared with EGFRamp glioma patients (Fig. 1J). Notably, this vascular signature showed a strong correlation with the overall survival of patients with glioma (Fig. 1K).
EGFRwt/amp and EGFRvIII cells have different vascular capacities
The results presented so far suggest important differences in the vascular architecture of gliomas depending on the presence of EGFR mutations. To test the effect of wt or mut EGFR cells in the surrounding vessels, in the absence of any masking effect of concomitant mutations in other genes, we generated two mouse glioma models by transforming SVZ progenitors from p16/p19 ko mice with retrovirus carrying the wt or the vIII isoform of the receptor (Supplementary Fig. S3A and S3B). These cells grew in vitro (Supplementary Fig. S3C) and they were both very sensitive to dacomitinib (Supplementary Fig. S3D), an inhibitor of the receptor tyrosine kinase activity (12). The two mouse cell lines formed subcutaneous tumors, although SVZ-EGFRvIII cells grew faster (Supplementary Fig. S3E). Moreover, they were both very sensitive to dacomitinib in vivo (Supplementary Fig. S3F and S3G).
Orthotopic implantation of the SVZ models generated gliomas with a high penetrance. Survival rates were clearly reduced after the injection of vIII-expressing cells in comparison with SVZ-EGFRwt/amp–bearing mice (Fig. 2A), which correlates with the worse prognosis of EGFRmut gliomas in comparison with those harboring the wt receptor (amplified or not; Fig. 2B). The IHC and the IF analysis of the tumors revealed a higher compact and proliferative growth after EGFRvIII expression (Fig. 2C and D). The vascular density was not significantly different between the two models (Fig. 2E and F) but we observed a strong increase in the size of the vessels in EGFRvIII gliomas (Fig. 2E and G). To obtain an independent confirmation of this observation, we used the CAM assay, which has been widely used to study angiogenesis (17). Bio-cellulose scaffolds were embedded in VEGFA (as a control) or CM from SVZ-EGFRwt/amp or SVZ-EGFRvIII cells, and they were layered on top of growing CAMs (Supplementary Fig. S3H). Although the CM from both type of cells demonstrated a proangiogeneic capacity (Supplementary Fig. S3I), the diameters of the vessels formed were larger in the presence of CM from SVZ-EGFRvIII cells compared with their wild-type counterparts (Supplementary Fig. S3J). Moreover, we observed that the vascular signature was strongly upregulated in SVZ-EGFRvIII compared with SVZ-EGFRwt/amp tumors (Fig. 2H and I). Notably, some of the top genes in this comparative analysis were linked to pericytic differentiation (Cd248) and function (Timp1; Fig. 2I). Overall, these results confirm the proangiogenic function of EGFR signaling in gliomas. Moreover, they suggest that the expression of different isoforms of the receptor have a distinct effect on the vascular microenvironment.
EGFRwt/amp expression is associated with a hypoxic phenotype
To our surprise, the MRI analysis of the grafted mice revealed a cumulative increase in the contrast enhancement of SVZ-EGFRwt/amp compared with SVZ-EGFRvIII–injected brains (Fig. 3A and B). We also observed a stronger extravasation of Evans Blue (Supplementary Fig. S4A), IgG (Fig. 3C), and erythrocytes (TER119+ cells; Fig. 3D) in SVZ-EGFRwt/amp compared with mutant tumors. These observations suggest that the integrity of the BBB is severely compromised in the former. The disruption of the BBB has been associated with impaired blood perfusion and the formation of hypoxic regions in tumors (18). Accordingly, HIF1α expression (Fig. 3E) and the hypoxia-related signature (Fig. 3F) were higher in SVZ-EGFRwt/amp compared with SVZ-EGFRvIII tumors. This signature was defined using the IvyGAP (Ivy GBM Atlas Project) data-set analysis (Supplementary Fig. S4B), selecting the most relevant genes included in hypoxia and hypoxia-inducible factor 1α (HIF1α) pathways signatures that were upregulated in the perinecrotic and pseudopalisading-cell-necrosis tumor zones (Supplementary Fig. S4C). Notably, this signature was increased in mesenchymal, compared with classic or proneural gliomas (Supplementary Fig. S4D; ref. 19). Mesenchymal gliomas are characterized by a high frequency of EGFR amplifications (but not mutations) and show a higher overall fraction of necrotic area and a stronger expression of hypoxia-regulated genes compared with the other subtypes (19, 20). In contrast, tumors harboring EGFR mutations tend to accumulate in the classical subgroup, characterized by a highly proliferative phenotype (19, 21).
We then performed an IF analysis in different orthotopic patient-derived-xenografts (PDX) that express EGFRwt/amp or EGFRmut (deletions and/or point mutations). We observed the increase in the permeability of IgG in the prior compared with the latter (Fig. 3G and H). This effect was paralleled by an upregulation of the hypoxia-related signature in EGFRwt/amp PDXs (Fig. 3I). Notably, the extravasation of IgG was also increased in EGFRwt/amp compared with EGFRmut patients' tumors (Fig. 3J and K). Besides, we found a strong increase in the expression of carbonic anhydrase IX (CAIX), one of the main HIF1α targets, in IDHwt/EGFRwt/amp gliomas compared with tumors with IDH or EGFR mutations (Fig. 3L). IDHmut gliomas contain a more “normalized” vasculature without vessel leakage (8) and a reduced extent of necrosis (22). The histologic analysis of the tumor's sections confirmed these observations (Fig. 3M). Moreover, it showed an accumulation of necrotic areas in EGFRwt/amp compared with EGFRmut gliomas (Fig. 3M), as well as a strong correlation between the amount of IgG extravasation and the necrotic area score (Supplementary Fig. S4E). These results suggest that the angiogenic signals induced by amplification or overexpression of wt EGFR result in a dense vascular network, but with a severely compromised BBB, a different phenotype to the one observed in mutant EGFR tumors. This vascular fragility seems to be associated with a less efficient fueling of tumor proliferation and with the appearance of areas of necrosis and hypoxia.
Glioma-derived pericytes stabilize the vasculature in EGFRmut tumors
SVZ-EGFRvIII tumors showed a highly compact growth, whereas in SVZ-EGFRwt/amp tumors cells appeared detached from each other and from the tumor vessels (Fig. 4A). It has been shown that pericytic functions can be performed by the highly plastic glioma stem cells, which acquire mesenchymal and mural cell features (3) in a process regulated by EGFR signaling (8). In agreement with these notions, some tumor cells expressed αSMA in both SVZ glioma models (Fig. 4B). Furthermore, pericyte-related markers were expressed in GFP+ cells, sorted out after tumor dissociation, independently of the genetic status of EGFR (Fig. 4C). However, we noticed an increased transcription of αSma and Cd248 in SVZ-EGFRvIII tumor cells, compared with their wild-type counterparts (Fig. 4D). Notably, among the vascular signature, several pericyte-related genes were upregulated in cultured SVZ-EGFRvIII cells (Supplementary Fig. S5A), suggesting that they express these markers even in the absence of the microenvironment-derived signals, and that this process is exacerbated in the presence of the mutant isoforms of the receptor. In agreement with these results, there was a significant increase in the expression of human αSMA and CD248 (Fig. 4E) and in the ratio of tumor pericytes to mouse ECs (Supplementary Fig. S5B) in EGFRmut (including tumors expressing EGFR deletions or point mutations) compared with EGFRwt/amp PDXs. In contrast, the transcription of host pericyte genes in all the PDXs tested was similar to that of normal mouse brain (Fig. 4E; Supplementary Fig. S5B). Notably, human endothelial transcripts (CD31, END) were not overexpressed in the PDXs, independently of the genetic status of EGFR (Supplementary Fig. S5C). These results suggest that in the presence of EGFR mutations glioma cells have a higher capacity to differentiate into pericytes but not to ECs. Indeed, we found that up to 80% of the αSMA+ cells in a patient's tumor express the vIII mutation (Supplementary Fig. S5D). In agreement with these data, when we costained tumor sections from EGFR-mutant (vIII isoform or point mutations) PDXs with an antibody that recognizes the human but not the mouse GAPDH protein, we observed that 70% to 80% of the mural cells were indeed derived from the tumor cells (Fig. 4F). Moreover, we found an increased pericyte coverage in mouse (Fig. 4G) and human (Fig. 4H) glioma transplants. In contrast, we noticed that the amount of delocalized αSMA+ mesenchymal cells (those that were not in close contact with ECs) was higher in wild-type compared with mutant EGFR tumors (Supplementary Fig. S5E and S5F).
To further study the capacity of mutant EGFR to stimulate the transdifferentiation of glioma cells we introduced EGFRvIII in GBM1 cells (EGFRamp). We observed a strong increase in the amount of αSMA and CD248 protein in vitro, similar to the expression observed in GBM3 cells (EGFRamp/EGFRvIII; Fig. 4I). Overexpression of the mutant receptor increased the aggressiveness of the GBM1 tumors (Fig. 4J) and induced a higher transcription of human-specific pericyte genes (Fig. 4K). Notably, downregulation of CD248, a master regulator of pericyte differentiation in malignant solid tumors (8, 23), decreased the aggressiveness of GBM1-EGFRvIII tumors (Fig. 4J), with a concomitant decrease in the expression of other pericyte makers (Fig. 4K). When we analyzed the tumors formed, we observed no changes in the leakiness of IgG (Supplementary Fig. S5G) or in the pericyte coverage (Supplementary Fig. S5H) after CD248 downregulation, whereas there was a reduction in the number of dilated BVs (Supplementary Fig. S5I). Taken together, these results confirm that EGFR mutations promote the growth of gliomas, at least in part, through the increase in the plasticity of the tumor cells, which can work as pericytes and reinforce the stability of the vessels. Moreover, they suggest that reducing this transdifferentiation capacity of the glioma cells might normalize the vasculature of the tumors.
EGFR mutations modulate the vascular properties of glioma cells in a BMX- and SOX9-dependent way
EGFR signaling activates transcription factors that drive tumor growth. We performed an in silico analysis to find those factors overexpressed in EGFRmut gliomas but with a similar expression in wild-type or amplified EGFR gliomas. The best hit was the Sex-determining region Y (SRY)-box 9 (SOX9) gene (Supplementary Fig. S6A), which has been previously linked to EGFR (24, 25). We found a strong upregulation of the SOX9 protein (Fig. 5A) and mRNA (Fig. 5B) in those PDXs harboring EGFR mutations. Similarly, SOX9 was accumulated in SVZ-EGFRvIII allografts, but not in SVZ-EGFRwt/amp or in GL261 (commonly used mouse glioma model) tumors (Supplementary Fig. S6B). Furthermore, in the presence of a proteosomal inhibitor the amount of SOX9 was increased in EGFRwt/amp cells, but did not change in EGFRmut cells (Fig. 5C), suggesting the stabilization of the protein in the latter.
SOX9 participates in a variety of functions during development, although it has also been implicated in the regulation of cancer stem cells (26). Moreover, we found a positive association of SOX9 with angiogenic processes in gliomas (Fig. 5D). Downregulation of this protein in GBM3 (EGFRamp/EGFRvIII; Fig. 5E and F) or in GBM1-EGFRvIII (Supplementary Fig. S6C and S6D) cells impaired their orthotopic growth. Notably, the expression of CD248 and other pericyte-related genes was decreased in GBM3-shSOX9 compared with control tumors (Fig. 5F; Supplementary Fig. S6E), suggesting that SOX9 mediates the induction of pericyte properties in EGFRmut tumors. In agreement with that, we found a positive correlation between the transcription of SOX9 and different pericytic-related genes in gliomas (Supplementary Fig. S6F–S6I). Notably, we observed a decrease in the pericyte coverage (Fig. 5G), concomitant with the upregulation of the IgG extravasation (Fig. 5H) and the hypoxia-related signature expression (Fig. 5I) in SOX9-interfered tumors. These observations reinforce the idea that blockade of the transdifferentiation capacity of EGFRmut GBM cells favors the fragility of the tumor vessels and the subsequent induction of hypoxia, similar to what occurs in tumors overexpressing wild-type EGFR.
A recent study has discovered that the bone marrow and X-linked (BMX) nonreceptor tyrosine kinase is highly expressed in glioma-derived pericytes but not in normal mural cells in the brain (27). However, the upstream signals have not been described yet. To test whether EGFR signaling could be connected to BMX activation in glioma cells, we performed an in vitro analysis in response to EGF. Stimulation with the ligand induced EGFR activation and signaling in SVZ-EGFRvIII (Fig. 5J), GBM7 (EGFRV774M; Supplementary Fig. S6J), and SVZ-EGFRwt/amp (Supplementary Fig. S6K) cells. However, it only stimulated BMX phosphorylation in the presence of the mutations (Fig. 5J; Supplementary Fig. S6J and S6K). Moreover, we observed an accumulation of SOX9 protein at short times after EGF stimulation in EGFRvIII (Fig. 5J) and EGFRV774M (Supplementary Fig. S6J) mutant cells, but not in EGFRwt/amp cells (Supplementary Fig. S6K). Ibrutinib, a dual BMX/BTK (Bruton tyrosine kinase) inhibitor that targets neoplastic pericytes (27), blocked BMX stimulation and SOX9 accumulation in response to EGF in SVZ-EGFRvIII (Fig. 5J) and EGFRV774M (Supplementary Fig. S6J) mutant cells. EGFR and AKT phosphorylation were impaired in the presence of ibrutinib in both types of SVZ cells (Fig. 5J; Supplementary Fig. S6K). However, ibrutinib impaired tumor growth of SVZ-EGFRvIII cells (Fig. 5K), but not their wild-type counterparts (Fig. 5L). The reduced tumor burden was parallel to a reduced phosphorylation of EGFRvIII and BMX, as well as a reduction in the expression of SOX9 and pericyte markers in the treated tumors (Supplementary Fig. S6L). These results suggest that EGFRmut-BMX signaling induces the accumulation of SOX9 protein and the subsequent formation of tumor-derived pericytes, increasing the aggressiveness of EGFRmut gliomas but rendering these tumors sensitive to ibrutinib.
PDGFRβ activation in the pericytes of EGFR-mutant gliomas is responsible for the entrance of immune cells into the tumors
To further characterize the vascular phenotype of wild-type and mutant EGFR gliomas, we compared the expression of several important signaling pathways in angiogenic and vascular development in the SVZ models. We observed a strong increase in the levels of phospho-PDGFRβ (platelet derived growth factor receptor beta), with no differences in the amount of total PDGFRβ, phospho-AKT, or β-catenin in the tumors expressing EGFRvIII (Supplementary Fig. S7A). This increase was detected as well in several EGFRmut PDXs (Supplementary Fig. S7B) and human tumors (Fig. 6A) compared with the wild-type tumors. In both type of samples, we found a strong correlation between the phosphorylation of EGFR and PDGFRβ (Fig. 6B; Supplementary Fig. S7B). Activation of the latter in response to the secretion of PDGF ligands by endothelial cells is crucial for the proper integration of pericytes in the vessel walls (28). In agreement with that, we detected specific phospho-PDGFRβ labeling adjacent to the endothelium in several EGFRmut PDXs (Supplementary Fig. S7C). Notably, PDGFRβ activation was not observed in SVZ-EGFRvIII cells in vitro or in GL261 tumors (Supplementary Fig. S7A), which show many vascular abnormalities, even though they do not carry any EGFR alterations (6).
The results presented so far indicates that the presence of EGFR mutations in gliomas lead to activation of PDGFRβ in the pericytes that cover the tumor vessels. Emerging evidence suggest that these mural cells are essential players in the regulation of blood cells extravasation (29). Moreover, the activation of PDGFRβ in pericytes has been shown to promote a latent immunologic phenotype in these cells, which results in the recruitment of large numbers of leukocytes into the brain (30). In agreement with that, we found a strong positive correlation between the levels of phosphorylation of this receptor in human samples and the percentage of leukocytes (Fig. 6C), myeloid (Fig. 6D), and lymphoid (Fig. 6E) cells detected by flow cytometry after tumor dissociation. The presence of these cells in gliomas also correlated with the levels of phospho-EGFR (Supplementary Fig. S7D–S7F). Altogether, these data suggest that tumors expressing EGFR mutations, through the modulation of the vessel wall, have a stronger immune infiltrate. Indeed, the IHC analysis showed a significant accumulation of CD45 (leukocytes; Fig. 6F and G) and CD68 (myeloid cells; Fig. 6F and H) positive cells in IDHwt/EGFRmut tumors, compared with the other groups of gliomas. Besides, we detected increased amounts of leukocytes (Fig. 6I), myeloid (Fig. 6J), and lymphoid (Fig. 6K) cells in these tumors when measured by flow cytometry.
To confirm the relevance of PDGFRβ activation in EGFRmut gliomas, we exposed SVZ-EGFRvIII tumors to sunitinib, which reduced the amount of phosphorylated receptor in the tumors (Supplementary Fig. S7G). The drug impaired glioma growth and increased the overall survival of the injected mice (Fig. 6L; Supplementary Fig. S7H). In agreement with our previous results, we measured a reduced number of dilated BVs in the sunitinib-treated tumors (Supplementary Fig. S7I), concomitant with an increased IgG extravasation (Fig. 6M) and an upregulation of the hypoxia-related signature (Supplementary Fig. S7J). In addition, sunitinib reduced the presence of immune cells in EGFRvIII allografts (Fig. 6N). Altogether, these results confirm that the activation of PDGFRβ, upon recruitment of the pericytes to the vessels, is important for the permanence and the fully functional differentiation of these cells, stabilizing the vasculature in EGFRmut gliomas. Moreover, the data highlights that tumor-derived-pericytes drives tumor growth through the regulation of both, the vascular and the immune compartments of gliomas.
In agreement with other authors (10, 31, 32), we have established a positive correlation between the presence of EGFR genetic alterations and the expression of angiogenic molecules and the appearance of vascular abnormalities. However, our data suggest for the first time that wild-type or mutant EGFR induce the formation of two distinct vascular phenotypes, which in turn differ from the more normalized vasculature of mutant IDH1/2 gliomas. Some authors have found a small increase in the expression of angiogenic and vasculogenic molecules in IDHmut tumors compared with normal brain (14). Moreover, it has been suggested that these tumors express Wnt signals, which allow them to coopt preexisting vessels (33). Therefore, a certain increase in angiogenesis, together with vascular cooption could fuel tumor growth in IDHmut gliomas. Among IDHwt gliomas, we have found two vascular phenotypes, both highly different from the normal vasculature: the first one is characterized by the instability of the BBB and the induction of hypoxic signals and necrotic areas in the tumors, whereas the second phenotype, induced by EGFR mutations, contains more robust and enlarged tumor vessels that nurture a very compact and hyperproliferative tumor tissue. In agreement with this, the neuroimaging analysis showed an accumulation of Gd over time in SVZ-EGFRwt/amp tumors. These results could seem contradictory with the literature as it has been suggested that EGFRvIII gliomas have a higher perfusion values in dynamic contrast-enhanced (DCE) analyses in the patients (34), as well as in animals models (31, 35). This measure is related to the degree of vessel permeability, but it is also influenced by the blood flow and the vessel area, which are enlarged in SVZ-EGFRvIII compared with SVZ-EGFRwt/amp tumors. On the basis of our data, EGFRwt/amp tumors would show more contrast enhancement at longer times after Gd injection, due to the presence of leakier tumor vessels. Although long neuroimaging studies are nonviable in clinical practice, the multiparametric analysis of the DCE data showed no major differences in the Gd enhancement of EGFRvIII gliomas, which were otherwise characterized by increased cell density and blood flow, compared with other tumors (36). A similar study has revealed the existence of two different subtypes of IDHwt GBMs: a glycolytic phenotype with predominant neovasculature and a necrotic/hypoxic dominated phenotype. However, no correlation was stablished with the genetic status of EGFR (37). These kind of studies or other approaches to measure tumor vessel caliber and/or structure in gliomas (16) would certainly help to characterize the contribution of the two different vascular ecosystems in IDHwt GBMs and how they evolve in response to therapy.
Pericytes play an essential role maintaining the structure and function of the vessel wall and the BBB. These mural cells were thought to originate from mesenchymal progenitors that are recruited from the bone-marrow in response to hypoxia (38, 39). However, it has been recently shown that the majority of vascular pericytes in GBMs derive from tumor cells (3). A similar epithelial-to-pericyte transition has been proposed in other cancers, promoting vascular integrity and tumor growth (40). Moreover, tumor cells seem to use a pericyte-like spreading for brain metastatic colonization (41, 42). Nevertheless, to date it was unknown if different genetic alterations could modulate this transdiferentation process. Results from this study and our recently published data (8) suggest that it is governed by the EGFR/BMX signaling pathway. Moreover, we have shown here that tumor cells expressing mutant isoforms of EGFR have a higher capacity to differentiate into functional pericytes and overexpress molecules such as CXCR4, CX45, and TIMP1, which could be responsible for the improved pericyte recruitment toward endothelial cells and the increase pericyte coverage observed in EGFRmut gliomas. In contrast, in the absence of EGFR mutations, mesenchymal cells appeared frequently detached from the vessels, being hard to determine the specific nature of these cells, which could resemble tumor-associated fibroblasts (43). Even in the perivascular space, a wide variety of fibroblasts, smooth muscle cells, and pericytes have been characterized in different brain locations (both in normal and in pathologic conditions; refs. 44, 45). However, none of these studies has determined how the different genetic backgrounds of gliomas modulate the presence and/or the function of these highly plastic cells.
Although the transdifferentiation capacity of EGFR mut cells confers a higher aggressiveness to the tumors, they also increase their sensitivity to molecules that target pericytic function, such as ibrutinib. This BMX inhibitor specifically affects tumor-derived pericytes (27) and cancer stem cells (46), preserving normal brain cells, which could limit the toxicity of the drug. Besides, this compound inhibits mutant EGFR activity in gliomas (47), as well as in other cancers (48). Ibrutinib is being tested on unselected GBM patients in combination with radio- and chemotherapy (NCT03535350). We propose that its activity could be higher in tumors harboring EGFR mutations so future retrospective studies should be carried on to validate this idea.
Downstream of EGFRmut/BMX signaling, we have found an enrichment of SOX9 expression and protein stability. The oncogenic function of SOX9 has been proposed in different cancers (26), including gliomas (49, 50), where it induces proliferation and cell survival, partly through BMI1 upregulation (51). Our data indicate that SOX9 also regulates the vascular properties of gliomas by inducing the cellular plasticity of tumor cells, which could be linked to its well-known effects in cancer stem cells (26). Notably, SOX9 has been linked to the upregulation of PDGFRα expression during development (52), and regulates genes involved in the extracellular matrix such as collagen, aggrecan, and Timp1, which are involved in the maintenance of the BV structure (53). In addition, BMI1 has been linked to the regulation of the epithelial-to-mesenchymal transition (54), allowing us to propose that SOX9, either acting directly or through the modulation of BMI1, may act as a master regulator of the vascularization processes in gliomas, especially for those tumors carrying EGFR mutations.
Our results suggest that once attached to the endothelial cells, there is an activation of PDGFRβ in the glioma-derived pericytes. This is a key signal for their fully differentiation and integration into the vessels, and provides the proliferative and survival cues needed to maintain the vascular structure (28). Our results confirm the higher functionality of these tumor vessels in EGFR-mutant tumors, as they are able to support the proliferative capacity of glioma cells and favor the extravasation of blood cells into the tumor stroma. Although pericytes were classically associated with structural functions, an accumulating body of evidence suggests that these cells also display immune properties, including antigen-presenting activities and expression of cytokines and adhesion molecules to control the immune cell trafficking across vessel walls (29, 55). Particularly, activation of PDGFRβ in pericytes has been shown to promote their immunomodulatory properties, which results in the expression of recruitment molecules in the adjacent endothelium and the entrance of large numbers of blood cells into the brain (30). These results are in agreement with recent reports that strongly suggest that changes in the tumor vasculature affect the degree of extravasation of immune cells, as well as their pro- or antitumor profiles (56–58). Moreover, increasing evidences support the synergism between antiangiogenic and immune therapies in other cancers (59), as well as in gliomas (60). Although no major breakthroughs have been achieved using sunitinib in patients with GBM (61), our results suggest that there could be a synergistic effect with immune modulators. Moreover, we propose that patients harboring EGFR mutations could have a differential sensitivity, not only to PDGFRβ inhibitors but also to strategies targeting the immune compartment. In any case, future experiments with different immunocompetent models will certainly help to study the vascular-immune connection in gliomas.
In the absence of EGFR mutations, we have found an induction of HIF1α protein and function. A direct transcriptional activation of HIF1α by EGFR has been proposed in lung cancer, especially in the presence of mutations (62), but little is known about this interaction in gliomas. Although we cannot discard a direct regulation of HIF1α expression by EGFR activation, we believe that this transcription factor is induced and/or stabilized in the tumor context in response to the environmental conditions (such as acidosis or nutrients/oxygen deprivation) that occur after the disruption of the BBB in EGFRwt/amp gliomas. Notably, there seems to be a reciprocal relation as tumor hypoxia upregulates EGFR expression in lung cancers (62) and gliomas (63), promoting its activation in the absence of mutations. This could serve as a prosurvival signal for hypoxic cancer cells. Moreover, hypoxia has been proposed to induce resistance to EGFR inhibitors in different cancers (64), suggesting that combinatorial approaches targeting both pathways could be a promising strategy for EGFRwt/amp gliomas.
In summary, we propose the existence of two distinct microenvironmental phenotypes in IDHwt GBM, orchestrated by the genetic status of EGFR and the downstream modulation of BMX-SOX9 activity, which induces the transdifferentiation of tumor cells into pericytes. This model could have diagnostic as well as great predictive value, as the different subtypes could have a distinct sensitivity to anti-angiogenic or immunomodulatory strategies. Moreover, the function of the tumor-derived-pericytes could limit the entrance of cytotoxic therapies through the BBB (27, 65). Overall, our results place the angiogenic properties of gliomas at the top of the glioma hierarchy and suggest that future combinatorial therapeutic approaches should combine agents targeting the glioma vasculature with conventional therapies, molecularly directed drugs or immunotherapies. Moreover, these strategies should be tailor-designed for each specific glioma subtype.
J.M. Sepúlveda-Sánchez reports grants from Ministerio de Economía y Competitividad de España during the conduct of the study. The authors declare that they have no competing interests. P. Sánchez-Gómez reports grants from Ministerio de Ciencia, Innovación y Universidades during the conduct of the study. No disclosures were reported by the other authors.
B. Segura-Collar: Formal analysis, investigation, visualization, writing–review and editing. M. Garranzo-Asensio: Investigation, writing–review and editing. B. Herranz: Investigation, writing–review and editing. E. Hernández-SanMiguel: Investigation, writing–review and editing. T. Cejalvo: Investigation, writing–review and editing. B.S. Casas: Investigation, writing–review and editing. A. Matheu: Resources, writing–review and editing. Á. Pérez-Núñez: Resources, writing–review and editing. J.M. Sepúlveda-Sánchez: Resources, funding acquisition, writing–review and editing. A. Hernández-Laín: Resources, formal analysis, writing–review and editing. V. Palma: Resources, investigation, visualization, writing–review and editing. R. Gargini: Supervision, funding acquisition, investigation, visualization, writing–original draft, writing–review and editing. P. Sánchez-Gómez: Conceptualization, supervision, funding acquisition, writing–original draft, writing–review and editing.
This work was supported by FONDECYT grant (1140697 to V. Palma), CONICYT Fellowship (to B.S. Casas), by Ministerio de Economía y Competitividad and FEDER funds (PI13/01258 to A. Hernández-Laín; PI17/01621 to J.M. Sepúlveda-Sánchez; and PI16/01580 and DTS18/00181 to A. Matheu), by Young Employment Initiative (Comunidad de Madrid) to M. Garranzo-Asensio, by “Asociación Española contra el Cancer” (AECC) grants (INVES192GARG to R. Gargini; GCTRA16015SEDA to J.M. Sepúlveda-Sánchez); and by Ministerio de Ciencia, Innovación y Universidades and FEDER funds (RTI2018-093596 to P. Sánchez-Gómez). The authors would like to acknowledge Manuel Serrano for kindly donating the p16/p19 ko mice and Jacqueline Gutiérrez and Rafael Hortigüela for their technical support. The graphical abstract was created with BioRender.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.