Inhibiting Integrin β8 to Differentiate and Radiosensitize Glioblastoma-Initiating Cells

Glioblastoma (GB, grade IV glioma) is the most frequent and lethal malignant primary adult brain tumor. Surgical resection followed by radio/chemotherapy (30 × 2 Gy and Temozolomide) are the current standard of care. However, prognosis remains extremely poor (median survival of ∼15 months; ref. 1). Recurrences may be notably explained by the strong cellular heterogeneity of these tumors which contain GB-initiating cells (GIC; ref. 2). GICs are characterized by their unlimited self-renew ability, their stem markers overexpression (CD133, Nestin, Olig2, Sox2, Nanog, A2B5, ITGA6…), their multipotent aptitude to differentiate into neural lineages, their localization in perivascular niches, their invasion capacity, their strong radio/chemoresistance, and their higher tumorigenic potential (2). Consequently, GIC targeting represents a strong therapeutic interest in GB.

Among the most considered targets, integrins are the subject of an increasing number of studies. These transmembrane receptors are composed of two noncovalently associated α and β subunits and are key components of the dialogue between cells and their microenvironment. They regulate cell adhesion to extracellular matrix (ECM) or cell surface proteins. They also serve as transmembrane links between extracellular contacts and intracellular cytoskeleton via a bidirectional signaling (inside-out/outside-in; ref. 3). Integrins can then activate pathways leading to gene transcription to sustain proliferation, survival, differentiation, and migration (3, 4). In GB, several integrins have been highlighted in GB progression. αvβ3, αvβ5, and β1 are involved in invasion, angiogenesis, and radioresistance (for review, see ref. 5). α3 (6), α6 (7), α7 (8), and β8 (9) were also reported to be overexpressed in GIC and, particularly for α6 and α7, to be associated with GIC stemness (5). Thus, GIC might express a specific integrin pattern compared with other tumor cells, and some of these integrins may be considered as specific GIC targets. Considering that clinical targeting of αvβ3/αvβ5 integrins in GB patients led to some disappointing results in a recent phase III trial (10), there is an urgent need to identify new targets specific to this GIC population.

β8 integrin (β8) has drawn our attention as a new integrin that may have specific functions in GIC. β8 heterodimerizes with αv and is able to link arginine–glycine–aspartic acid motifs such as in vitronectin, laminin, and latent-TGFβ1/3. β8, expressed in restrictive cell types including neural stem, neuroepithelial, embryonic, and mesangial renal cells (11, 12), displays a major role in self-renewal and survival of neural progenitor cells (12), and, then, could have a similar function in GIC. Indeed, β8, highly expressed in this GIC subpopulation (9), was preferentially found in vivo in GB perivascular and perinecrotic areas (13), some preferential GIC-enriched niches (14). Furthermore, β8 overexpression was also associated with glioma tumor grade (9). Although few studies were conducted in glioma, β8 appears to have a proinvasive role in GB cells (15), whereas it could negatively regulate angiogenesis in transformed human astrocytes (16). In addition, it was recently shown that β8^HIGH-GB cells display several stem characteristics (17). Nevertheless, the effects of β8 targeting on GIC stemness, viability, migration, radioresistance, and tumorigenicity have never been explored.

We hypothesized that β8 could sustain GIC stemness and radioresistance and represent a new GIC-targetable marker. We demonstrated that β8 is overexpressed in GIC and associated with bad prognosis in GB patient cohorts. We highlighted in patient GIC primocultures that β8 is associated in vitro and in vivo with characteristics and features unique to GIC, including self-renewal ability, stemness status, markers expression, viability, adhesion/migration, radioresistance, and tumorigenesis. We also showed that selective β8 inhibition can impair GIC properties in vitro and suppress orthotopic tumor growth in vivo. Together, these results identify β8 as a new membrane GIC marker and a potential therapeutic radiosensitizing target.

The study was conducted on newly diagnosed GB tumor samples isolated from 14 patients. Ten samples were used for IHC studies, and 4 additional GB samples were used to establish 4 primary GIC cell lines (SRA5, SRB1, SRC3, and SB7). For quantitative real-time RT-PCR (qPCR) experiments, additional GIC patient cell lines were also used (A0, C1, D1, G, I, and K). These samples were all obtained after written-informed consent from patients admitted to the Neurosurgery Department at Toulouse University Hospital and processed in accordance with the Institution's Human research Ethics Committee. Tumors used in this study were histologically diagnosed as grade IV astrocytoma according to the World Health Organization criteria.

Ten patient GIC primocultures were established from patient surgical GB samples and were fully characterized either in our previous work (18) or in this study (Supplementary Fig. S1). Briefly, GICs were cultured as GIC-enriched neurospheres (NS) in stem cell medium (SCM), compared with their differentiated counterparts (GB-differentiated cells, GDC), kept in FCS-enriched differentiation medium for at least 15 days, as described (18).

The related protocol, previously described (18), is detailed in the Supplementary Procedures.

For siRNA transfection, NS from SRB1 and SRC3 patient cell lines were dissociated and seeded (500,000 cells/well) in 6-well plates before being transfected using Lipofectamine RNAiMAX (Invitrogen) with 3 μL of siRNA [-CTR or -B8-5 or -B8-6 (10 μmol/L; Qiagen)]. Sequences of siRNA were respectively as follows: AATTCTCCGAACGTGTCACGT, AACGTCTATGTCAAATCGACA, and CAGCCTGTTTGCAGTGGTCGA (5′-3′). Cells were harvested for analysis 5 days after transfection. For shRNA transfection, NS from SRB1, SRC3, or SB7 patient cell lines were dissociated and seeded (750,000 cells/well) in 6-well plates before being transfected using Fugene HD (Promega) with 3 μg of shRNA (shRNA-CTR or -B8-1 or -B8-4; Qiagen). For transient transfections, GFP-coupled sequences were used. Cells were then observed by fluorescence microscopy (Nikon), photographed (Camera Coolsnap HQ), and used for experiments. For stable expression, shRNA sequences were coupled with a Neomycin (NEO) resistance gene. Transfected cells were then continuously selected 5 days after transfection with Neomycin (2,000 μg/mL, Merck). Sequences of shRNA-CTR, shB8-1, or shB8-4 were respectively (5′-3′) as follows: GGAATCTCATTCGATGCATAC, CCAAGCTACTTGAGAATATTT, and TCTCGCTCTTGATAGCAAATT.

The related protocols, previously described (18), are detailed in the Supplementary Procedures. ITGαvβ8 antibody was a gift from Stephen L. Nishimura, University of California, San Francisco, CA (19).

For adhesion assay, NS cells from SRB1- or SRC3-stable patient cell lines were dissociated, and GICs (50,000 cells in 100 μL/well) were seeded on laminin-coated wells (10 μg/mL; Sigma) in 96-well plates for 6 hours at 37°C. Three random fields per well from triplicate wells were pictured under a 10X objective. Cells were manually delineated. The cell surface (A), perimeter (P), and circularity [C = 4π (A/P²)] of at least 30 cells per experiment were calculated using Image J software. Cells were classified in two groups: rounded cells with 0.85 < C < 1; and polarized cells with 0 < C ≤ 0.85. Limits for C were determined by visual identification of a subset of rounded cells as previously described (20). Results are presented as the percentage of polarized cells. For example, if 60 cells have a circularity below or equal to 0.85, and 40 cells have a circularity above 0.85, the percentage of polarized cells is 60%.

For migration assay, transwells (12 wells, BD Biosciences) were precoated in their undersurface with laminin (10 μg/mL). The upper chamber medium consisted of DMEM-F12 only (without EGF/FGF), and the lower chamber medium consisted of SCM. NS from SRB1- or SRC3-stable patient cell lines were dissociated, and GICs (40,000 cells in 500 μL/well) were seeded in Transwells incubated for 24 hours at 37°C. Nonmigrated cells were removed by wiping the membrane upper side with cotton swab. Transmigrated cells were fixed with PFA 4% (ChemCruz) and stained with hematoxylin (Merck). The number of transmigrated cells was counted in at least 3 random fields from duplicate transwells under 10X magnification.

SubG1 and Annexin V/propidium iodide (AV/PI) staining, as well as clonogenic irradiation assays, were performed as previously described (18) and detailed in the Supplementary Procedures. Postmitotic cell death was analyzed after cell permeabilization (PBS-0.15% Triton, 5 minutes, room temperature) and DAPI staining (0.2 μg/mL; Sigma). The number of aberrant nuclei observed with DAPI staining was counted in at least 3 random fields using fluorescence microscopy (Axio-vert). For those three cell death assays, dissociated GICs were treated or not with a 5 Gy irradiation and harvested for analysis 5 days after irradiation (Gamma-cell Exactor 40, Nordion). Finally, Caspase-8 and Caspase-3/7 activity assays were also performed. GICs were dissociated and seeded (20,000 cells in 100 μL/well) on white 96-well plates. Resuspended caspase substrate (Caspase-Glo8 or Caspases-Glo3/7; Promega) was added in each well (100 μL, for 1 hour at 37°C). Caspase activities were quantified by luminometry using a plate reader (FLUOstar Optima). Results show relative caspase activities measured after subtracting the background from all values.

Nude mice were used in accordance to a protocol (APAFlS#7660-2016110818123504 v2) reviewed and approved by the Institutional Animal Care and Use Committee of Région Midi-Pyrénées (France). This protocol, previously described (18), is detailed in the Supplementary Procedures.

IHC was performed on paraffin-embedded sections (5 μm) of excised brains of xenografted mice (single staining) or GB patients samples (double staining). The sections were incubated at 57°C for 20 minutes followed by rehydration in xylene and ethanol series. Antigen retrieval was performed at 95°C for 20 minutes using DAKO buffer pH9 (EnVision) and washed with PBS.

For single staining, slides were blocked with Peroxidase Blocking Reagent (EnVision; 5 minutes), washed with PBS, incubated 30 minutes with primary antibodies diluted in antibody diluent (EnVision), and then washed with 0.3% Tween in PBS. The primary antibodies used were rabbit anti-ITGβ8 (Abcam; Ab80673), rabbit anti-Sox2 (Abcam; Ab92494) and mouse anti-Nestin (Millipore; MAB5326). Slides were then incubated with FLEX/HRP (EnVision; 20 minutes), washed (0.3% Tween in PBS), and revealed with DAB (Diaminobenzidine peroxidase buffer) mixed with FLEX/DAB chromogen (EnVision; 5 minutes). Sections were then washed, counterstained with Hemalun of Mayer (Merck), dehydrated, and mounted with Eukitt solution before being viewed on a Nikon microscope. The numbers of positive and total cells were counted in at least 3 random fields under 63X magnification. Results are presented as the average percentage of positive and negative cells for each marker (ITGβ8 or Sox2) in each condition (shCTR or shB8-1 or shB8-4; 4 mice/group).

For double staining, IHC was performed on tumor patients slides (n = 10) according to the manufacturer's protocol [EnVision G2 Doublestain System, Rabbit/Mouse (DAB+/Permanent Red)], with rabbit anti-ITGβ8 (Abcam; Ab80673) marked with Permanent Red and mouse anti-Sox2 (Abcam Ab171380) marked with DAB.

We used The Cancer Genome Atlas (TCGA) dataset to analyze gene expression in newly diagnosed primary GB patients (21, 22). For our analyses, a dataset (processed data: version 2015-02-24, downloaded from https://genome-cancer.ucsc.edu) was used based on RNA-seq data (Illumina HiSeq 2000 RNA Sequencing platform, University of North Carolina TCGA genome characterization center, n = 149). We first assessed overall survival (OS) using this TCGA dataset among a subgroup of newly diagnosed primary GB patients treated by standard chemo/radiotherapy (n = 59). For these analyses, gene expression levels were dichotomized based on a high expression cutoff (within the 75% quartile). We also used the Repository for Molecular Brain Neoplasia Data (REMBRANDT) database (n = 178 GB with available survival data) to validate the prognostic ability of ITGB8. Survival rates were estimated by the Kaplan–Meier method with 95% confidence intervals. Univariate analyses were performed using the Log-rank test. We then performed correlation analyses between ITGB8 expression and several classic stem and differentiation genes in the TCGA dataset. Multiple comparisons between molecular subgroups were performed using the Mann–Whitney U test. The Benjamini–Hochberg procedure was applied for multiple comparisons. Two-sided P values of less than 0.05 were considered statistically significant. Statistical analyses were performed using STATA 12.0 software.

All the results depicted in this study are presented as mean ± SEM of at least three independent experiments in the same cell line and were reproduced in at least 1 or 2 other cell lines. Significant differences (^∗, P < 0.05; ^∗∗, P < 0.01; and ^∗∗∗, P < 0.001) were evaluated with the unpaired Student t test. The Log-rank analysis of Kaplan–Meier survival curves was used to evaluate the tumorigenesis of injected cells, with P < 0.05 considered as significantly different (^∗∗, P < 0.01 and ^∗∗∗, P < 0.001).

As we hypothesized a specific integrin expression pattern in GIC, we screened at the RNA level the expression of a panel of integrins in 10 patient GIC-enriched NS cell lines, fully characterized either in our previous work (18) or in this study (Supplementary Fig. S1A). All stem markers (Olig2, Sox2, A2B5) appeared overexpressed in all GIC compared with the corresponding GDCs, which contrarily overexpressed differentiation markers (neuronal TUJ1, astrocytic GFAP; Supplementary Fig. S1B–S1D). We then analyzed mRNA and protein expressions of main integrins in GIC and GDC (Supplementary Fig. S2). Briefly, ITGα6 was overexpressed in GIC, as previously shown (7). ITGβ3 displayed a similar pattern, but at varying degrees according to the cell line. Contrarily, ITGβ5 and ITGβ1 appeared to be restricted to the GDC cell surface, as well as ITGβ4 in GDC total protein extracts. ITGαv protein was, for its part, equally expressed in both GIC and GDC. However, we noticed that ITGβ8 mRNAs were upregulated in all GICs compared with their differentiated progeny (Fig. 1A) and compared with adult healthy neural tissues, neural stem cells, or related GDC (Fig. 1B). Moreover, we confirmed in 4 patient cell lines (SRA5, SRB1, SRC3, and SB7) that β8 protein was also overexpressed in GIC compared with their GDC counterparts, as well as in whole cell lysates by Western blot (Fig. 1C) and at the membrane cell surface by FACS analysis of both β8 and αvβ8 (Fig. 1D). Based on these results, we focused our study on ITGβ8, which appeared, as ITGα6 (7), to be selectively expressed at the GIC membrane and barely present in GDC.

As β8 seemed to be selectively expressed in GIC, we performed statistical correlation analysis between ITGB8 expression and stem (NG2, GLI1, MSI1, NANOG, NESTIN, NOTCH1, SOX2, SHH, PROM1, ITGA6, OLIG2, and CD44) and differentiated (GFAP, MAP2, TUJ1, MAL, OMG) markers in TCGA database (n = 146 GB patients). We observed a significant positive correlation of ITGB8 with all the stem markers tested, except PROM1 and OLIG2 (Supplementary Table S1; Supplementary Fig. S3A). In contrast, β8 negatively correlated with neuronal markers (MAP2, TUBB3, and HRNBP3/NeuN) and oligodendrocytic markers (MAL and OMG; Supplementary Table S1; Supplementary Fig. S3B). To further study the link between β8 and stem markers, we analyzed the expression of ITGβ8 and Sox2 by IHC double-staining in GB patients cohort (n = 10) for both ITGβ8 and Sox2 (Fig. 1E; Supplementary Fig. S4A). Patients that did not express Sox2 (2–5 and 9) also failed to express β8 (Fig. 1E; Supplementary Fig. S4B). Contrarily, patients that were positive for Sox2 (1, 6–8, and 10) also expressed β8 at varying degrees. Importantly, β8-stained cells appeared to be positive for Sox2 as well (Fig. 1E). Altogether, these results showed that ITGβ8 is selectively expressed by GIC in vitro and that ITGβ8-positive cells in GB patients are also Sox2-positive.

We subsequently designed shRNA and siRNA strategies to selectively downregulate β8 and assess its functions in GIC. Two GFP-coupled β8-targeting shRNA sequences (shB8-1/-4) and a negative control (shCTR) were used to transiently transfect NS-dissociated cells for 7 days. We observed in SRB1 and SRC3 GIC cultures that shB8s potently inhibited ITGβ8 expression at mRNA and protein levels, particularly at the cell membrane, compared with shCTR condition (Fig. 2A–C). For stable inhibition, SRB1 and SRC3 GIC were transfected with the same shRNA sequences coupled to a neomycin-resistance gene and then G418-selected to establish stable cell lines. These stable NS cell lines displayed a significant inhibition for ITGβ8 mRNA and protein expression (Fig. 2D and E). A similar effect was observed after GIC transfection for 5 days with two β8-targeting siRNAs (siB8-5/-6) compared with a control siRNA (siCTR; Supplementary Fig. S5A–S5C). Finally, as integrins are mainly known as key players in adhesion and migration processes, both largely activated in GIC (5, 23), we consequently performed GIC adhesion assays on laminin, a known β8 substrate (24), to validate the functional consequences of ITGβ8 targeting. We demonstrated that stable shβ8 transfection decreased the cell adhesion and spreading rates, with a reduction of both cell area and polarization compared with shCTR (Fig. 2F). We also assessed GIC migration by seeding stable shB8 or shCTR cells (SRB1/SRC3) on laminin-coated transwells for 24 hours. ITGβ8 knockdown reduced migration rate by 30% to 50% compared with shCTR (Fig. 2G). Collectively, these results show that our interference RNA strategies to inhibit ITGβ8 can potently block both protein expression and fundamental functions of this integrin in GIC.

The self-renewal ability in limiting dilution condition, a major stem characteristic, was then assessed by evaluating the NS-forming capacity of SRB1/SRC3 GIC at low cell density after β8 targeting. After transient transfection, GFP⁺/ITGβ8⁻ population isolated from shB8 GIC displayed a dramatically lower ability to form NS compared with the GFP⁺/ITGβ8⁺ population sorted from shCTR GIC (Fig. 3A). Moreover, stable GIC transfection with shB8-1/-4 induced a similar decrease of the NS generation ability (Fig. 3B and C), with, in some cases, the appearance of an adherent phenotype potentially compatible with a more differentiated state (Fig. 3B). This major alteration of the in vitro GIC self-renewal ability after ITGβ8 inhibition can be mediated by several processes, notably by an impairment of their stem phenotype. Indeed, we observed in SRB1/SRC3 GIC a noticeable decrease (∼5% to 15%) of the GIC subpopulation within the NS structures in response to ITGβ8 inhibition after either transient or stable transfection (Fig. 4A and B). Moreover, the transient transfection of shB8s induced a significant decrease of Olig2 and Nestin stem markers and an increase of the differentiation markers TUJ1 (neuronal) and MAL (oligodendrocytic) at the mRNA level (Fig. 4C). By Western blot, we confirmed the decrease of these stem markers in favor of an increase of TUJ1/OMG differentiation proteins after ITGβ8 inhibition (Fig. 4D). These variations in stem (Sox2, Nestin, and ITGA6) and differentiation (GFAP, OMG, MAL, and O4) markers were also found in GIC stably depleted for β8 at mRNA and protein levels (Fig. 4E and F). In addition, siRNA-mediated β8 inhibition (siB8-5/-6) was also associated with a strong downregulation of stem markers (Sox2, Nestin, NG2, and Olig2) and an upregulation of differentiation markers (TUJ1, MAL, or OMG) at the mRNA and protein levels, in 5-day–transfected SRB1/SRC3 GIC (Supplementary Fig. S5D and S5E). Altogether, these results suggest that ITGβ8 knockdown induces a differentiation process in GIC and an alteration of their stemness characteristics.

As an impairment of GIC self-renewal ability could also be linked to viability alteration, we analyzed subG1 population after GIC transient transfection with β8-shRNAs. We demonstrated that this subpopulation was significantly increased compared with shCTR GIC cells (Fig. 5A). We hypothesized that this viability loss may be associated with the induction of a proapoptotic mechanism. We observed by qPCR that GIC transient transfection with shB8s induced a decrease of Bcl2 and Survivin, two antiapoptotic genes, and an increase of Bax/Bcl2 ratio (Fig. 5B). These transcriptional data were confirmed by Western blot as we noticed an increase of effector Caspases-3/7 and PARP cleavages, an overexpression of the proapoptotic protein Bax, and a downregulation of the antiapoptotic protein Bcl2 (Fig. 5C). To fully describe this apoptotic process, we measured caspase activities and found a significant increase in Caspases-3/7 activity in SRB1 and SRC3 GIC cell lines (Fig. 5D). Interestingly, we similarly observed a reduction of Bcl2 and Survivin mRNAs and an increase in Caspases-3/7 activity in cells transfected with ITGβ8-targeting siRNAs (Supplementary Fig. S5F and S5G). Of note, we also noticed with siβ8 a significant activation of initiator Caspase-8. Altogether, these results suggest that ITGβ8 knockdown induces in GIC, in addition to their differentiation, the induction of an apoptotic process.

Our results show that ITGβ8 can support in GIC several major in vitro mechanisms, such as prosurvival processes, stemness maintenance, and proadhesive/migratory properties. We hypothesized that β8 inhibition in clinic could contribute to sensitize GIC, strongly chemo/radioresistant (2, 25), to conventional therapies. Thus, we studied the combined effects of ITGβ8-stable inhibition and ionizing radiations (IR) on GIC. We treated shB8-stable GIC clones (SRB1/SRC3) with IR (5 Gy) and observed, 4 days after irradiation, that ITGβ8 targeting significantly increased the percentage of subG1 (Fig. 6A) and apoptotic AV-positive cells (Fig. 6B), compared with shCTR cells. In addition, the postmitotic cell death, a common death mechanism observed in cancer cells after irradiation and characterized by the presence of giant multinucleated cells and micronuclei (26), appeared also significantly increased in irradiated shB8s (Fig. 6C). To fully demonstrate the radiosensitizing effect of ITGβ8 downregulation in GIC, we performed clonogenic assays at low cell density and measured the NS generation potential of shCTR or shB8s stably-transfected GIC (SRB1/SRC3) in response to increasing IR doses (0–10 Gy). We observed that in response to IR, ITGβ8 silencing led to a significant decrease in the NS-surviving fraction compared with shCTR condition (Fig. 6D). Altogether, ITGβ8 inhibition appears to radiosensitize GIC for cell death induction.

To further explore β8 functions in GB cells stemness in vivo, we analyzed the tumorigenic potential of ITGβ8-downregulated GIC in orthotopically xenografted nude mice. All the three mice groups (shCTR/shB8-1/shB8-4) experienced, at different time points, some neurologic signs and displayed, after autopsy, Nestin-positive–invasive brain tumor areas (Fig. 7A). However, we noticed a significant survival advantage for mice xenografted with shB8-4 SRB1 GIC compared with shCTR, and a similar tendency with shB8-1 (Fig. 7B). We then analyzed by IHC both β8 and Sox2 expression patterns in tumor areas of xenografted mice and noticed a marked decrease of both ITGB8- and Sox2-positive cells (from 60%–70% to 10%–20%) in shB8s subgroups compared with shCTR (Fig. 7C). As β8 inhibition appeared to decrease the GIC tumorigenicity, we hypothesized that low ITGβ8 expression could confer to GB patients a survival advantage after standard chemoradiotherapy. Using TCGA, we determined patient OS according to ITGB8 expression on a specific group of newly diagnosed primary GB solely treated with standard chemoradiotherapy (n = 59). In this particular population, β8 overexpression was significantly associated with OS decrease (P = 0.033), from 21.3 (β8^LOW) to 14.5 months (β8^HIGH; Fig. 7D). REMBRANDT GB dataset (primary GB with survival data, n = 177) confirmed (Fig. 7E) that ITGB8 overexpression is associated with a reduced OS, as previously reported in a larger REMBRANDT cohort of low- and high-grade gliomas (n = 339; ref. 15). Interestingly, we showed in the REMBRANDT GB cohort that ITGB8 is significantly overexpressed in GB compared with normal brain tissues and with lower-grade astrocytomas and oligodendrogliomas, both associated to a better prognosis than GB (Fig. 7F; ref. 27). Eventually, ITGB8 expression is associated in vivo to a reduced survival in GIC-xenografted mice, and its overexpression in GB patients, positively correlated with a large panel of stem markers (Fig. 1E), is linked to a worse prognosis in two independent GB cohorts.

Despite intense efforts to understand GB-malignant processes and identify new druggable targets, it must be noted that prognostic and therapeutic protocols have not changed since more than a decade. To improve GB patient's outcome, current researches mostly focus on GIC biology to specifically target these tumorigenic stem-like cells and overpass their resistance mechanisms. In line with this, our study identifies β8 integrin as a new GIC functional marker and highlights for the first time its potential therapeutic interest for GB radiosensitization.

We first demonstrated in GB patient primocultures that β8 is strongly overexpressed at the GIC surface. We also highlighted a stronger β8 expression in GIC compared with their differentiated counterparts, as already suggested in GB cells (9, 17), but also in colorectal cancer cells (28). In addition, we analyzed TCGA database and found that β8 expression in GB samples is correlated positively with the expression of numerous stem markers and negatively with neural differentiation markers. We also performed IHC costaining of the stem-factor Sox2 and β8 in several patient GB biopsies and demonstrated that β8-positive cells were also positive for Sox2 (Fig. 1). Altogether, these elements are in favor of a more selective β8 expression in GIC. Interestingly, GICs were proved in vivo to reside in perivascular niches (14), and several works described indeed a positive correlation of β8-stained cells with perivascular areas in primary GB patient's samples (13, 17). Besides, we previously showed that ITGβ8 is significantly overexpressed in the Classical molecular GB subtype, compared with Mesenchymal and Proneural subtypes in the Verhaak's molecular classification (5). Actually, this Classical subtype was characterized through single-cell molecular analysis by a notable stemness signature (29) and was shown to overexpress ITGα6 and ITGα7, two other GIC-associated integrins (7, 8). Finally, it was recently shown in GB patient tumor bulks that β8^HIGH-sorted cells were defined by a superior self-renewal ability in vitro and an increased tumorigenicity in vivo, two major GIC properties (17). It could be concluded from our results and these observations that ITGβ8 may be used as a new biomarker to detect and, possibly, enrich GIC from the tumor mass.

Furthermore, several studies measured β8 protein expression in tumor samples obtained from large patient cohorts and reported a rather specific expression of β8 in tumor cells, much higher than in tumor endothelial cells and healthy brain tissues (30, 31). In this context, β8 expression was shown to positively correlate with the glioma grade, with a higher expression in grade IV GB (9, 30). To go further, using TCGA and REMBRANDT databases, we demonstrated that β8 is upregulated in GB compared with healthy tissues (5) and that a high β8 level is associated with a worse prognosis in GB patients (Fig. 7). Consequently, it would be of great interest in clinic to take advantage of this new GIC biomarker as a potential prognostic marker. Of note, this new prognostic biomarker may be used in combination with other integrins previously described as prognostic factors in GB, such as α5 (32), β1 (33), and αvβ3 (34).

Besides the potential of ITGβ8 as a new biomarker to detect GIC in tumor mass and to assess GB aggressiveness, we hypothesized that this particular integrin could be an attractive target to specifically impair GIC functions. To this end, we demonstrated that β8 silencing by interference RNA in GIC resulted in vitro in a strong decrease of their NS formation ability (Fig. 3), linked to their self-renewal property, and of cell adhesion and migration capacities (Fig. 2), some key features sustaining GIC invasion and tumorigenesis (35). These effects were correlated with a global increase of GIC differentiation pattern in response to β8 downregulation (Fig. 4). In this respect, stable β8 targeting significantly delayed the growth of GIC-derived orthotopic xenografts and improved nude mice survival, with a major decrease of Sox2 staining in β8-negative tumors (Fig. 7). Interestingly, the β8 role in stem markers maintenance and in self-renewal was previously evoked in murine NPC (12) and in sorted-β8^HIGH GB cells isolated from tumors (17). In this study, the β8^LOW population was also shown to be less tumorigenic in orthotopically xenografted nude mice, confirming our in vivo results. Besides that, GICs also display high capacities for migration/invasion (23, 36), and we demonstrated here that β8 supports GIC-adhesive and -migratory properties. Interestingly, β8 was previously described to sustain proinvasive ability in GB cells and to increase the migration of transformed human astrocytes via TGFβ activation (15, 16). Moreover, β8 was shown to promote motility of prostatic tumor cells (37) and to favor a MMP-9/-2–dependent proadhesive/prometastatic process in human lung cancer cells (38). Altogether, we showed that β8 inhibition in GIC could mediate the differentiation of these tumor cells and the loss of their stem-associated properties. Although the precise mechanisms of these alterations need to be deciphered, therapeutic strategies should take advantage of this differentiation process to lower the high resistance potential of this cell subpopulation (2) and promote GIC cell death.

In that way, we demonstrated here for the first time that, additionally to its role in GIC stemness maintenance, β8 inhibition can induce a significant increase of caspase-dependent GIC apoptosis, with activation of both initiator and effector caspases (Fig. 5). Interestingly, this survival impediment appeared to be potentiated after irradiation, with a notable increase of postmitotic cell death in β8-downregulated GIC (Fig. 6). These results confirm the interest of blocking β8-mediated stemness to favor GIC cell death and hamper their therapeutic resistance. Regarding cancer cell survival impairment, β8 inhibition was shown to induce a potent cell-cycle blockade in lung cancer cells (38) and to decrease hepatic cancer cell survival and proliferation via the inhibition of Survivin and Bcl2, two antiapoptotic proteins, and via the impairment of the TGFβ/Smad pathway (39). Focusing on apoptosis, β8 was notably associated to protection of Oligodendrocytic Progenitor Cells from H₂O₂-induced cell death in response to high levels of Osteopontin (OPN), a secreted ECM phosphoprotein (40). Because OPN, a candidate ligand for αvβ8 (41), was described as a major actor for GIC stemness, GB cell dedifferentiation and tumorigenesis (42–44), and GB cell radioresistance (45), it would be of interest to decipher the OPN role in the β8-induced molecular mechanisms to sustain GIC survival and stemness. Another thing to consider is the role of Survivin in β8-induced GIC stemness, viability, and resistance, because we and other demonstrated that this protein, markedly overexpressed in GIC, is involved in GB cell dedifferentiation, radioresistance, and faster recurrence (18). Finally, a particular focus is required on the TGFβ role in β8-mediated GIC survival and resistance. Indeed, this paracrine and autocrine pathway, notably activated via αvβ8-binding to latent TGFβ complex, was shown to sustain GIC radioresistance (46) as well as to promote proliferation, invasion, angiogenesis, stemness, resistance, and tumor initiation/progression in GB (47). Manifestly, signaling pathways associated to β8 effects on GIC stemness, survival, and resistance remain to be explored, particularly the identification of ligands which may interact with β8 to transduce signals and modulate GIC biological processes and functions.

Growing evidences point out ITGβ8 as a specific GIC marker and highlight its role in GIC stemness, survival, and resistance. We therefore propose this integrin as a new potential therapeutic target in GB, because β8 appears to act by various mechanisms to alter GIC functions, aggressiveness, and radioresistance. Considering the fact that (i) the clinical targeting of αvβ3/αvβ5 by Cilengitide recently failed in GB patients (10) and that (ii) the GIC subpopulation, proved to greatly support the quick recurrence of GB after treatment, was described to express low and variable expression levels of αvβ3/αvβ5 [our present results and (ref. 9)], it is worth hypothesizing that the targeting of β8 in this specific GIC subpopulation could, in combination with radio/chemotherapy, significantly impede the development of these aggressive brain tumors. Significant effort is then required to develop specific β8 inhibitors, already tested at the preclinical level for tumor immunotherapy (48), and to set up new combined therapeutic strategies for GB outcome improvement.

No potential conflicts of interest were disclosed.

Conception and design: L. Malric, S. Monferran, P. Dahan, S. Boyrie, E. Cohen-Jonathan Moyal, C. Toulas, A. Lemarié

Development of methodology: L. Malric, C. Delmas, F. Arnauduc, P. Dahan, S. Boyrie, A. Lemarié

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L. Malric, S. Monferran, C. Delmas, F. Arnauduc, P. Dahan, V. Lubrano, D. Ferreira Da Mota, S. Evrard, A. Lemarié

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): L. Malric, S. Monferran, C. Delmas, F. Arnauduc, P. Dahan, S. Boyrie, J. Gilhodes, T. Filleron, A. Lemarié

Writing, review, and/or revision of the manuscript: L. Malric, S. Monferran, F. Arnauduc, S. Boyrie, V. Lubrano, J. Gilhodes, T. Filleron, A. Kowalski-Chauvel, E. Cohen-Jonathan Moyal, C. Toulas, A. Lemarié

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): L. Malric, C. Delmas, P. Deshors, D. Ferreira Da Mota, A. Siegfried, S. Evrard, A. Lemarié

Study supervision: L. Malric, A. Kowalski-Chauvel, E. Cohen-Jonathan Moyal, C. Toulas, A. Lemarié

We would like to thank Emmanuelle Uro-Coste and Monique Courtade-Saïdi at the IUCT-oncopole Anatomopathology service, Jean-José Maoret and Fréderic Martins at the Toulouse GeT Platform, Manon Farcé (INSERM U1037, Technology Cluster) for her technical assistance on the FACS sorting, and Julie Sesen and Nicolas Skuli for their scientific inputs. We also want to acknowledge the Institut Claudius Regaud and GRICR, RITC Foundation (Stemri), INSERM, the Canceropôle Grand Sud-Ouest (A. Lemarié), the Association pour la Recherche sur les Tumeurs Cérébrales (A. Lemarié and L. Malric), and the THE program from Plan Cancer 2016 (MoGlimaging). L. Malric and S. Boyrie were supported by a PhD fellowship from Plan Cancer/ITMO and P. Dahan and P. Deshors by a government PhD fellowship.

This work was supported by Recherche Innovation Thérapeutique Cancérologie (RITC), the THE program (INSERM, INCa, Plan cancer), the Groupe de Recherche de l'Institut Claudius Regaud (GRICR), the Canceropôle Grand Sud-Ouest (GSO), and the Association pour la Recherche sur les Tumeurs Cérébrales (ARTC).

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.