CD90 Expression Controls Migration and Predicts Dasatinib Response in Glioblastoma

Glioblastoma (GBM) is one of the deadliest human cancers with an incidence of about 3.5/100,000 per year worldwide (1). Despite the aggressive standard of care currently used including surgery, chemo-, and radiotherapy, the prognosis remains very poor with approximately 15 months overall survival (2). The inevitable recurrence of GBM is associated with: (i) resistance to radio- and chemotherapy; (ii) diffuse features due to the invasiveness properties of tumor cells throughout the surrounding brain parenchyma, and (iii) tumor intra- and inter-heterogeneity (1, 3).

CD90 (Thy-1) is a marker for mesenchymal stromal/stem cells (4) that has been earlier described in glioma/GBM specimens (5) and immortalized glioma/GBM cell lines (6–10). In the past years, CD90 has been described as a human GBM stem cell (GSC) marker (11–15). CD90 is also expressed in GBM-associated stromal cells (GASC; ref. 16) and mesenchymal stem cell–like pericytes (17), thereby reflecting GBM cellular heterogeneity. CD90 is a N-glycosylated, glycophosphatidylinositol (GPI)-anchored cell surface protein, originally described in murine thymocytes (18). CD90 is also expressed in many other cell types including endothelial cells, fibroblasts, and neurons (4, 19–21). CD90 has been involved in neurite outgrowth inhibition, T-cell activation and apoptosis, leukocytes and melanoma cell adhesion and migration, tumor suppression in ovarian cancers, and fibroblast proliferation and migration in wound healing and fibrosis (19, 21). Although the exact mechanisms of action of CD90 remain ill-defined, a role in cell–cell/matrix interactions has been proposed (19, 21).

We show that CD90 expression is not only restricted to GBM stem–like cells but is also observed in more differentiated GBM cells (primary adherent lines) and in freshly dissociated GBM specimens. In GBM patients, CD90 is associated with a cell adhesion/migration gene signature and with multifocal/multicentric MRI features. Using in vitro and in vivo approaches, we demonstrate the critical role of CD90 in GBM migration/invasion. We show that CD90 signaling through SRC, FAK, and RhoA promotes cell migration and importantly, that high CD90 expression affects the cell response to the SRC inhibitor dasatinib. We propose a model in which CD90 expression represents a novel stratification tool for selecting patients to be treated with dasatinib. Moreover, in this model, dasatinib could not only impair the adhesion/migration of CD90^high-differentiated tumor cells, but also the proliferation of CD90^high GSCs, thereby increasing its therapeutic potential for CD90^high patients.

All reagents not specified below were purchased from Sigma-Aldrich. Antibodies against human CD31, CD45, CD90, CD200, FAK, phospho-FAK, and VCP were obtained from BD Biosciences; anti-CD90 antibody used for IHC from Novus Biologicals (Bio-Techne); anti-CD133 antibody from Miltenyi Biotec; and anti-IL13Rα2 antibody from Biolegend (Ozyme); anti-phosphotyrosine antibody from Thermo Fisher; anti-SRC and anti-phospho-SRC antibodies from Cell Signaling Technology; and anti-vimentin antibody from Dako.

GBM samples were obtained after informed and written consent from patients admitted to the neurosurgery department at Rennes University Hospital (Rennes, France) for surgical resection in accordance with the local ethical committee and the French regulations. Tumors used in this study were histologically diagnosed as grade IV astrocytoma according to the WHO criteria. For transcriptome analysis, we retrospectively recruited a local cohort of 77 GBM patients (males: n = 54; females: n = 23; median age = 60 years—from 36 to 75 years; recruited between 2004 and 2013) treated with radiotherapy and concurrent/adjuvant temozolomide in accordance with the standard of care (Supplementary Table S1). Tumor samples were snap-frozen immediately after resection. All samples presented at least 70% of tumor cells. The extent of surgery was evaluated with an enhanced MRI performed within 24 hours after the resection. Adherent (RADH) and neurospheres (RNS; enriched in stem cells) GBM primary cell lines were obtained from GBM samples as described in refs. 22, 23. Briefly, fresh tumor tissues were mechanically dissociated using gentleMACS dissociator following the manufacturer's instructions (Miltenyi Biotec). Cells were directly cultured and frozen before further use such as flow cytometry analysis. RADH cells were grown in DMEM (Lonza) supplemented with 10% FBS (Lonza). RNS cells were grown in DMEM/Ham:F12 (Lonza) supplemented with B27 and N2 additives (Invitrogen), EGF (20 ng/mL), and basic fibroblast growth factor (20 ng/mL; PeproTech, Tebu-Bio). All GBM RNS and RADH cells were used between the 5th and 15th passages for the experiments. Human immortalized U251 and U87 GBM cell lines were cultured in DMEM 10% FBS.

U251 cells were transfected with pLKO.1-puro plasmids containing shRNA constructs #24 (5′-ccggcgaaccaacttcaccagcaaactcgagtttgctggtgaagttggttcgtttttg-3′) and #25 (5′-ccgggctcagagacaaactggtcaactcgagttgaccagtttgtctctgagctttttg-3′) targeting CD90 mRNA and #CTR (5′-ccggcaacaagatgaagagcaccaactcgagttggtgctcttcatcttgttgttttt-3′) targeting nonmammalian mRNA (Sigma-Aldrich) using the Lipofectamine 2000 reagent (Life Technologies) according to the manufacturer's instructions. After one week of culture under the selective antibiotic puromycin used at 10 μg/mL, transfected U251 cells were amplified and then cloned in 96-well plates at 0.1 cell/well. CD90 knocked-down U251 cell lines were expanded and selected for their decreased expression of CD90. U87 cells were transfected with CD90 cDNA (GeneWiz, Sigma-Aldrich) cloned into the pLKO.1-puro plasmid with EcoRI and BamHI enzymes using the Lipofectamine 2000 reagent. CD90-expressing U87 cell lines were obtained as described above with U251 and were selected for their high expression of CD90.

RNS cells were infected with lentiviral particles generated from HEK293T cells using Lenti-X packaging single shot system (Takara, Ozyme) and pLKO.1-puro plasmids containing shRNA constructs #25 targeting CD90 mRNA and #CTR targeting non-mammalian mRNA according to the manufacturer's instructions. After one week of culture under puromycin selection used at 10 μg/mL, RNS cells were amplified and selected for their downregulation of CD90 expression.

Eight-week-old male Balb/c NOD-SCID mice (Janvier) were housed in an animal care unit authorized by the French Ministries of Agriculture and Research (Biosit, Rennes, France - Agreement No. B35-238-40). Parental, transfected U87 cells (5 × 10⁴ cells/mouse) and RNS cells (5× 10⁴ cells/implantation) were orthotopically implanted in immunocompromised mice as described in ref. 24. Mice were daily clinically monitored and sacrificed 28 days after implantation. Mouse brains were collected, fixed in formaldehyde solution 4%, and paraffin embedded for histologic analysis after H&E staining. Tumor burden was compared in the different groups of mice and analyzed using ImageJ software (25). For dasatinib treatment, mice were fed daily with dasatinib (40 mg/kg, Selleckchem) one week after implantation and for 3 weeks.

CD90 mRNA expression data were assessed from the publicly available datasets obtained with immortalized cell lines (GDS4296; ref. 26), glioma cell lines (GDS3885; ref. 27), and glioma specimens (GDS1815; ref. 28), (GDS1975; ref. 29), (GDS4465; ref. 30), and (GDS4470; ref. 31); stored in the Gene Expression Omnibus (GEO) repository of the NCBI website platform http://www.ncbi.nlm.nih.gov/geoprofiles/. Expression levels of CD90 mRNA were expressed as log₂(fluorescence intensities) determined in the different microarray datasets. For transcriptome analysis using a local GBM cohort, total RNA was isolated with the NucleoSpin RNAII Kit (Macherey-Nagel). RNA integrity (RNA integrity number ≥ 8) was confirmed with an Agilent 2100 bioanalyzer (Agilent Technologies). Gene expression profiling was carried out with the Agilent whole human genome 8 × 60K microarray kit (Agilent Technologies). Total RNA was extracted, labeled, and hybridized according to the kit manufacturer's recommendations. Raw intensity data were log₂-transformed and normalized (intra-array and inter-array scaling) using GeneSpring software (Agilent Technologies). Student t tests with a Welch approximation were used to compare expression values between conditions. Adjusted P values were calculated by controlling for the false discovery rate with the Benjamini–Hochberg procedure. Genes were considered significantly differentially expressed if the P value was below 0.05 and the absolute fold-change was greater than 2.

Cells were lysed in ice-cold lysis buffer (30 mmol/L Tris-HCl, pH 7.5, 150 mmol/L NaCl, 1.5% CHAPS). Proteins were resolved by SDS-PAGE (12%, 10%, and 7% PAGE for FAK, phosphotyrosine and SRC proteins respectively) and transferred to nitrocellulose membrane for blotting. The membranes were blocked with 3% BSA in 0.1% Tween 20 in PBS and incubated with the diluted primary antibodies (1/1,000). Antibody binding was detected with the appropriate horseradish peroxidase–conjugated secondary antibodies (1/7,000; anti-rabbit or anti-mouse; Dako) and visualized with ECL (KPL, Eurobio) according to the manufacturer's instructions. Kinase phosphorylation intensities were relative to total corresponding kinase signals using ImageJ.

Human GBM and mouse brain sections were deparaffinized with EZ prep solution (Ventana Medical Systems) at 75°C for 8 minutes. Antigen retrieval was performed using Tris-based buffer solution CC1 at 95°C for 48 minutes and endogen peroxidase was blocked. After rinsing, slides were incubated at 37°C for 60 minutes with diluted primary antibodies against antigens (1/300 and 1/50 dilutions for vimentin and CD90, respectively). Signal enhancement was performed using the DABMap kit (Ventana Medical Systems). Detection kit procedure was optimized on the discovery instrument (Ventana Medical Systems).

Cells were washed in PBS 2% FBS and incubated with saturating concentrations of human immunoglobulins and fluorescence-labeled primary antibodies for 30 minutes at 4°C. Cells were then washed with PBS 2% FBS and analyzed by flow cytometry using a FACSCanto II flow cytometer (BD Biosciences). The population of interest was gated according to its FSC/SSC criteria. In most of the experiments, the dead cell population was excluded using 7-amino-actinomycin D (7AAD) staining (BD Biosciences). Data were analyzed with the FACSDiva (BD Biosciences) or the FlowJo software (Tree Star Inc.) and the results were expressed as specific fluorescence intensity given by the ratio of geometric mean of test/geometric mean of the isotype control. For intracellular staining of phospho-SRC, cells were previously fixed with methanol overnight at −20°C and then permeabilized using the eBioscience fixation/permeabilization kit (Thermo Fisher). Cells were then incubated with anti-phospho-SRC antibody (Cell Signaling Technology) for 30 minutes at 4°C. After washing with permeabilization solution (eBiosciences), cells were finally stained with secondary anti-rabbit antibody (Millipore) and directly analyzed by flow cytometry.

RNS cells were cultured in 96-well plates at 25,000 cells/well in the presence of gradual amount of dasatinib (from 0 to 100 μmol/L; Selleckchem). After 5 days of culture, 20 μL of the WST1 reagent (Roche) was added in each well. After 4 hours at 37°C, optical densities (OD) were analyzed by spectrophotometry at 450 and 595 nm. Specific ODs were given by the difference between the OD observed at 450 nm and the OD observed at 595 nm. The cell viability was calculated by the ratio of the specific OD observed with cells incubated in the presence of different concentration of dasatinib and the specific OD observed with cells cultured in medium alone.

Ninety-six-well plates were coated with a filtered solution of 400 μg/mL rat tail collagen I in PBS (Sigma). Parental, controls, and transfected U251 and U87 cells (25,000 cells) were plated for time points 0, 30 minutes, 1, 2, and 4 hours. Medium and unattached cells were aspirated. Wells were washed with PBS and attached cells were stained with the WST1 reagent as described above. The percentage of cell attachment was calculated by the ratio of the specific OD observed with tested cells and the specific OD observed with total cells used at the beginning of the experiment. For RNS cells, neurospheres were obtained after culture and placed into slides in DMEM 10% FCS for 24 hours. Neurosphere adhesion was evaluated by counting the percentage of adherent neurospheres. These assays were performed in the absence or in the presence of 1 μmol/L dasatinib.

Spheroid formation experiments were performed by incubating 5,000 parental, controls, and transfected U251 and U87 cells per well in a 96-well plate previously coated with 75 μL of 1.5% agar gel. Images were taken after 72 hours and spheroid density was estimated using ImageJ by calculating the spheres' size. High spheres' size corresponds to the lower cell–cell adhesion. These assays were performed in the absence or in the presence of 1 μmol/L dasatinib.

Parental, controls, and transfected U251 and U87 cell lines were washed in DMEM, placed in Boyden chambers (10⁵ cells/chamber in DMEM) that were placed in DMEM 20% FBS, and incubated at 37°C for 24 hours. After 24 hours, Boyden chambers were washed in PBS and cells were fixed in PBS 0.5% paraformaldehyde. For RNS cell migration, Boyden chambers were previously coated with collagen (400 μg/mL) overnight. Nonmigrated cells inside the chambers were removed and cells were then stained with Giemsa (RAL Diagnostics). After washes in PBS, pictures of five different fields were taken. Migration was given by the mean of number of migrated cells observed per field. For inhibition with chemical drugs, cells were preincubated 15 minutes with SRC family kinases inhibitors PP2 (Sigma-Aldrich; 10 μmol/L) and dasatinib (1 μmol/L); ROCK inhibitor Y27632 (Selleckchem) and with FAK inhibitor Y15 (Sigma-Aldrich; 1 μmol/L). Kinases inhibitors were kept during the time of migration assay. For migration of CD90^high RNS cell lines in the presence of dasatinib, values of migration were corrected by a factor of 1.15 due to the fact that viability of these cells were affected by this inhibitor (10%–20% of cell death).

Eighty-nine treatment-naïve GBM patients (males: n = 59; females: n = 30; median age = 59 years - from 14 to 89 years) from the Cancer Genome Atlas (TCGA) cohort were analyzed for CD90 expression from transcriptome data and corresponding pretreatment MRI data. The images were downloaded from the NCI's The Cancer Imaging Archive (TCIA; http://cancerimagingarchive.net/). Preoperative qualitative and semiquantitative imaging variables were provided by the Visually Accessible Rembrandt Images (VASARI) feature set (32). Details of the imaging variables and acquisition were published previously (33, 34). Medians of the CD90 mRNA expression level between each VASARI feature were compared using the Mann–Whitney test. Kaplan–Meier analysis was used to estimate the survival difference between different imaging features. For animal experiments, untreated mice bearing parental U87 (n = 1) or U87 CD90 cells (n = 4), and dasatinib-treated mice bearing U87 CD90 cells (n = 4) were analyzed by MRI as described previously (35). Briefly, MRI acquisitions were performed on a horizontal 4.7 T Bruker Biospec 47/40 magnet interfaced to an AVANCE console (Bruker BioSpin) and a workstation running the ParaVision 5.1 software. A linear-volume radio frequency coil (72 mm inner diameter) was used as transmitter and a home-made surface radio frequency coil (15 mm diameter) was used as receiver. The anesthesia was induced with 3% of isoflurane in 0.5 L/minute of air. The mice were positioned in a MRI-compatible stereotactic holder to maintain the head fixed using ear bars a nose bar. Brain lesion was assessed using T2-weighted images obtained using a rapid acquisition with relaxation enhancement (RARE) as described in (36) with the following acquisition parameters: TR = 3500 ms, TE = 32.1 ms, RARE factor=8. Nine to eleven contiguous slices of 750 μm were acquired to cover the whole brain.

Values represent the mean ± SD of different experiments. Student t test was applied using a two-tailed distribution of two conditions of unequal or equal variances on groups of data obtained in experiments. ANOVA method was used for comparing multiple conditions. These analyses were performed with GraphPad Prism software. The significance level was P < 0.05.

CD90 has been initially described in GBM specimens (5) and immortalized glioma/GBM cell lines (6–10), but is currently considered as a human GSC marker (11–15). CD90 is also found in GASCs (16) and mesenchymal stem cell–like pericytes (17). However, we observed expression of CD90 in human adherent primary GBM cells, which do not display stem-like characteristics. To clarify this discrepancy, CD90 mRNA expression was analyzed in primary GBM stem-like cells (RNS, n = 12) that were previously characterized by us in ref. 23; and in adherent cells (RADH, n = 5) using transcriptome data (22, 23). Both CD133^high and CD133^low RNS cells, RADH cells, and GBM samples (n = 4) expressed high levels of CD90 mRNA (Fig. 1A). Interestingly, CD90 transcripts were also observed in the NCI brain-derived cell lines panel (26) (Supplementary Fig. S1A), in both adherent and GSC lines (ref. 27; Supplementary Fig. S1B), and finally in high-grade gliomas and GBM tumor specimens (refs. 28–31; Supplementary Fig. S1B and S1C).

CD90 protein expression was studied using flow cytometry in primary RNS and RADH cells and in U251 and U87 cell lines (Fig. 1B; Supplementary Fig. S2A). All the primary cells [i.e., CD133^high RNS (n = 6), CD133^low RNS (n = 6), RADH (n = 11)] and U251 cells expressed CD90 at high levels with a stronger expression observed on RADH and U251 cells (Fig. 1B; Supplementary Fig. S2A). In contrast, U87 cells did not express CD90 (Supplementary Fig. S2A). Interestingly, correlation between CD90 mRNA and protein expression was observed using RNS cell lines (Supplementary Fig. S2D). The analysis of dissociated GBM samples (n = 36) revealed that CD90 was expressed in most GBM specimens tested (34/36; Fig. 1C; Supplementary Fig. S2B and S2C) with various intensities (∼3 logs variation of specific fluorescence intensity; Fig. 1C). CD90 expression was confirmed using IHC on high (GBM#179) and intermediate (GBM#217) CD90-expressing specimens with a clear staining in both tumor cells and on blood vessels (Fig. 1D). A staining restricted to the vessels was only observed in CD90^low tumors (GBM#233; Fig. 1D). CD90 expression was also observed in xenografts obtained after injection with GBM primary cell lines, although inconsistent CD90 immunostaining was observed (Supplementary Fig. S2E). To further confirm that CD90 was expressed selectively in tumor cells, double staining with antibodies against CD90 and tumor (IL13Rα2; ref. 37), neural (CD200), endothelial (CD31), or stem (CD133) markers were performed in 8 GBM specimens and their peritumoral counterparts using flow cytometry (Fig. 1E; Supplementary Fig. S2F and S2G). Most of the CD90-positive cells from GBM were also positive for IL13Rα2 (69% in average). In contrast, most of the CD90-positive cells found in peritumoral tissues coexpressed CD200 (74% in average), a neural marker. A small fraction of CD90⁺ cells also expressed either CD31 or CD133 only in GBM samples indicating that endothelial and GBM stem cells within the tumor also expressed CD90 (respectively, 15 and 6% in average). Other brain cell types such as astrocytes and pericytes could also express CD90, but were not identified using our experimental settings. These results indicate that CD90 expression varies between GBM patients is not restricted to CD133^high stem-like GBM cells but is also found in CD133^low stem-like GBM cells as well as in more differentiated tumor cells.

To better characterize the role of CD90 in GBM, gene expression profiling was performed on an in-house cohort of 77 GBM specimens (Supplementary Table S1) and analyzed with regards to CD90 expression. To this end, two groups of 16 GBM patients were defined according to CD90 expression in the microarray data with CD90^low patients exhibiting a CD90 expression value lower than the 20th percentile of the CD90 expression distribution, and CD90^high patients with a CD90 expression value higher than the 80th percentile of the CD90 expression distribution (Fig. 2A). Differential gene expression profiling revealed that CD90^high tumors also exhibited a cell adhesion/migration gene signature comprising a highly connected gene network (Fig. 2B; Supplementary Tables S2–S4). Interestingly, genes found in the adhesion/migration signature, such as collagens and PDGFRB, were also overexpressed in CD90^high RNS cell lines (Supplementary Fig. S3A). The expression of CD90 mRNA was also highly expressed in mesenchymal and classical GBM subtypes previously described by Verhaark and colleagues (Fig. 2C; ref. 38). In addition, CD90^high tumors exhibited characteristic features of epithelial-to-mesenchymal transition (EMT) as indicated by the upregulation of αSMA, collagens (COL1A1, COL1A2 and COL5A2), metalloproteinases (MMP-2 and -9; Supplementary Fig. S3B) and the transcription factors FOXC, GSC, SNAI2, TWIST-1, and -2 (Supplementary Fig. S3C). Furthermore, other mesenchymal markers such as cadherins were also overexpressed in CD90^high tumors (Supplementary Fig. S3D), whereas epithelial markers were expressed at similar levels in both CD90^low and CD90^high groups (Supplementary Fig. S3E). These results show that CD90 expression is linked to a cell adhesion/migration and EMT-linked gene signatures in GBM patients. Our data were then correlated to those relative to images from patients' tumors within the TCGA GBM cohort (32). Indeed, the Visually Accessible Rembrandt Images (VASARI) feature set was applied in the imaging assessments and analysis of 89 GBM patients from the TCGA GBM cohort (33, 34) and tested for associations with CD90 expression. Among all VASARI imaging features, CD90 mRNA expression level was significantly different in those patients with nonenhancing tumor crossing midline versus those not crossing midline, and in patients with multifocal/multicentric versus focal tumors (Fig. 2D and E). Interestingly these imaging features were linked previously to an invasive profile revealed in patients with shorter survival and specific tumor gene signature associated with mitochondrial dysfunction (34). These data demonstrate that CD90 expression in tumor cells is associated also with a more invasive tumor phenotype.

To study the role of CD90 in GBM cells, CD90 was either silenced in CD90-positive U251 cells or reexpressed in CD90-negative U87 cells. Silencing efficiency was verified using both flow cytometry (Supplementary Fig. S4A and S4B) and Western blot analysis (Supplementary Fig. S4C). Cell viability and proliferation were not affected by CD90 modulation when compared with parental or mock-transfected cell lines over 5 days (Supplementary Fig. S4D and S4E). Cell–cell adhesion properties were tested in a spheroid formation assay using both U251 and U87 cells. CD90^high U251 cells formed spheroids less compact than those formed by CD90^low U251 cells and CD90-negative U87 cells (Supplementary Fig. S4F and S4G) suggesting that high CD90 expression could alter cell–cell adhesion. In contrast, cell adhesion to collagen was reduced when CD90 expression was decreased in U251 cells (Supplementary Fig. S4H), whereas no significant effect was observed with U87 cells, which present a strong collagen–cell interaction unrelated to CD90 expression (Supplementary Fig. S4H). Overall, these data indicate that CD90 limits cell–cell adhesion and increases cell–matrix adhesion ability of GBM cells. Cell migration was then evaluated using both wound-healing- (Supplementary Fig. S5A and S5B) and Boyden chamber- (Fig. 3A and B) based migration assays. In both cases, decreased CD90 expression in U251 cells reduced migration and reexpression of CD90 in U87 increased migration. Similarly, CD90^high GBM primary lines exhibited stronger migration indexes than their CD90^low counterparts (Fig. 3C and D). Furthermore, shRNA-mediated silencing of CD90 in CD90^high primary GBM lines (Supplementary Fig. S5C and S5D) dramatically reduced cell migration (Fig. 3E; Supplementary Fig. S5E). Overall, these results show that CD90 expression is involved in GBM cell adhesion/migration properties.

Parental and CD90-expressing U87 cells were tested for their tumorigenicity in an orthotopic xenograft mouse model. Most of the mice bearing parental U87 cells developed a clear encapsulated tumor mass 28 days postinjection that was evaluated using both MRI (Fig. 3F; n = 1) and H&E staining of brain sections (Fig. 3G; n = 5 out of 7). Tumor formation was not detected in mice injected with U87 CD90 cells using MRI (Fig. 3F; n = 4). However, H&E staining revealed the presence of tumors with an irregular/invasive shape in mice injected with U87 CD90 cells (n = 5 out of 7), whereas mice injected with parental U87 cells displayed encapsulated tumors with regular edges (Fig. 3G). Moreover, serial H&E sections revealed the presence of the U87 CD90 tumor mass at a distance from the injection site, indicating its invasive feature (Fig. 3H and I). Furthermore, CD90^low and CD90^high expressing RNS cells (n = 2 and 4, respectively) were used in an orthotopic xenograft mouse model. Clinical signs appeared between 76 and 140 days postimplantation, depending on the cell line but independent on CD90 expression. Massive tumor infiltration within the brain parenchyma was observed with CD90^high RNS cells contrasting with a more limited invasion observed with CD90^low RNS cells (Fig. 3J and K). These data are consistent with the results obtained in patients from the TCGA cohort and demonstrate that CD90 expression in tumor cells is associated with a more invasive tumor phenotype.

To investigate CD90-dependent signaling, U251 shCD90 and U87 CD90 transfectants were analyzed for total phospho-tyrosine containing proteins using Western blot analysis in comparison with parental and U251 shCTR and U87 CTR cells, respectively (Fig. 4A). Phospho-tyrosine signal increase was observed in U87 CD90^high cells compared with the parental cells that was confirmed in U251 cells with an observed decrease of phospho-tyrosine–containing proteins in U251 shCD90 (Fig. 4A). FAK and SRC kinases were previously described to transduce CD90 signals (39, 40) and total SRC and FAK did not vary in CD90^high and CD90^low cell lines (Fig. 4A). However, increase in SRC and FAK tyrosine phosphorylation was observed in U87 CD90^high cells (Fig. 4A and B). In contrast, a decrease in SRC and FAK phosphorylation was observed with U251 silenced for CD90. The expression levels of other SFK family members including FYN, LCK, and YES1 was high in primary GBM cell lines (Supplementary Fig. S6A) and GBM patients' specimens (Supplementary Fig. S6B), with a significant increase of FYN mRNA expression in CD90^high patients. However, no difference in SRC and FAK expression at the protein level was observed in CD90^low and CD90^high primary cells (Supplementary Fig. S6C and S6D). Finally, increased phosphorylation of SRC and FAK was observed in CD90^high primary GBM cells compared with CD90^low primary cells (Fig. 4C). Remarkably, FAK tyrosine phosphorylation was stronger in RADH lines compared with RNS lines, whereas no difference linked to culture conditions was observed regarding SRC tyrosine phosphorylation. To further document SRC activation in CD90^low and CD90^high primary GBM cell lines and in GBM patients' tumors, the SRC gene signature established by Bild and colleagues (41) was used to score SRC activation. Interestingly, SRC activation was predominantly associated with high CD90 expression in primary GBM cell lines (Fig. 4D). However, using the same approach, high SRC activation was found in GBM tumors independent of CD90 expression (Supplementary Fig. S6E). This was possibly due to the high cellular heterogeneity and to the complex stroma of those tumors. To further investigate this observation in GBM tumors, freshly dissociated GBM specimens were tested for the presence of CD90 and phospho-SRC using flow cytometry. High levels of SRC phosphorylation correlated with high CD90 expression in GBM specimens (Fig. 4E and F). Importantly, pretreatment of GBM cells with dasatinib, an SRC inhibitor, blunted SRC phosphorylation (Supplementary Fig. S6F). These data indicated that high CD90 expression correlates with the activation of SRC signaling in GBM tumor cells.

To confirm the impact of CD90-dependent signaling on cell migration in U251 and U87 cells, the chemical SRC inhibitors PP2 and dasatinib, the FAK inhibitor Y15 and the ROCK inhibitor Y27632 were tested in Boyden chamber migration assays using parental U251 and U87 CD90 cells (Fig. 5A). These inhibitors had no effect on cell viability during the time of the assay (Supplementary Fig. S7A). The SRC family kinases inhibitors PP2 and dasatinib dramatically reduced U251 and U87 CD90 migration. Migration inhibition was also observed upon treatment with the ROCK inhibitor Y27632 (53% and 76% reduction using U251 and U87 CD90 cells, respectively) but to a lesser extent. In contrast, the FAK inhibitor Y15 had no or limited effect on the migration of U251 and U87 CD90 cells. Interestingly, dasatinib also dramatically reduced the migration of primary CD90^high GBM lines (Fig. 5B and C). These results indicate that CD90-mediated migration mainly depends on the SRC family kinases, and to a lesser extent of ROCK and FAK. Inhibitors of downstream signals such as Rac1, MEK1, and JNK also affected the migration of CD90^high cells (Supplementary Fig. S7B).

To evaluate the CD90-dependent effects of dasatinib in vivo, U87 CD90 cells were used in an orthotopic xenograft mouse model. One week postimplantation, mice were treated with dasatinib (40 mg/kg) for 20 days. Mice were then analyzed using MRI 28 days postinjection and were sacrificed. As observed previously (Fig. 3F and G), U87 CD90–injected mice did not reveal any MRI-detectable tumors (Fig. 5D, n = 4), whereas an irregular tumor mass was observed after H&E staining (Fig. 5E, n = 6 out of 7). In contrast, small but clear encapsulated tumors with regular edges were observed using both MRI (Fig. 5D, n = 3 out of 4) and H&E staining (Fig. 5E, n = 4 out of 7) in mice injected with U87 CD90 and treated with dasatinib. Again, serial H&E sections revealed a deep infiltration of U87 CD90 tumor cells within the parenchyma (up to 400 μm away from the injection site), whereas very limited invasion was observed after dasatinib treatment (Fig. 5F and G). These data demonstrate that dasatinib attenuates CD90-dependent tumor migration/invasion properties in vivo, thus leading to smaller and less invasive tumors.

To further investigate the effect of dasatinib on CD90-mediated GBM cell–matrix interaction, we first tested U87 CD90 and U251 cells both displaying strong collagen adherence properties. Dasatinib dramatically inhibited U87 CD90 and CD90^low U251 cells cell attachment to collagen (Fig. 6A). We confirmed the effect of dasatinib on cell–matrix adhesion using primary RNS lines (enriched in stem cells) that were cultured under neurosphere conditions but had the unique property to partially adhere to the substrate, as described previously (42). Remarkably, CD90^high cells lost the ability to adhere to plastic upon dasatinib treatment, whereas neurospheres derived from CD90^low cells remained adherent (Fig. 6B and C). These results indicate that dasatinib modulates CD90-dependent cell/matrix adhesion properties of GBM cells.

As CD90 was considered as a GSC marker, we finally evaluated the impact of dasatinib on the viability of primary CD90^low and CD90^high RNS cells. Interestingly, CD90^high RNS cells were more sensitive to dasatinib than CD90^low RNS cells (IC₅₀ in average 0.3 μmol/L vs. 46.5 μmol/L respectively; Fig. 6D). Furthermore, the downregulation of CD90 in CD90^high primary lines significantly reduced their sensitivity to dasatinib (Fig. 6E). These data indicate that dasatinib affects the viability of CD90^high RNS cells enriched in GSCs.

In this study, we show that CD90 is expressed on both stem and differentiated GBM tumor cells, and we demonstrate that CD90 expression controls tumor cell migration/adhesion mainly through SRC signaling. In addition, we show that CD90 expression regulates tumor invasive characteristics in a mouse model and in human tumors. CD90 also regulates cell–cell/matrix adhesion properties of GBM cells. Finally, we provide evidence that dasatinib dramatically reduces CD90-mediated invasiveness of U87 CD90 cells in an orthotopic xenograft mouse model and that CD90 expression impacts on dasatinib sensitivity in patient-derived GSC lines. Collectively, this study unveils the importance of CD90 in GBM migration/invasion and points toward CD90 expression as a predictor of dasatinib response in GBM patients.

CD90 was described as a candidate marker for cancer stem cells from primary high-grade gliomas (11–15, 43). More recently, CD90-positive cells were also associated with blood vessels in human GBM tissues and characterized as immature mesenchymal stem cell–like pericytes (17). However, we observed that CD90 expression was high in human adherent primary GBM cells (22). High CD90 mRNA amounts were reported in established (26, 27) and primary (27) GBM cell lines as well as in tumor specimens (5, 28–31). To sort out this apparent discrepancy, we used flow cytometry and IHC approaches in GBM samples, and we found that CD90 was highly expressed in endothelial cells within the tumor and on neurons present in the brain parenchyma as described previously (4, 19–21). We also observed CD90 expression in GBM-derived stem and in more differentiated tumor cells using both cell lines and human tumor specimens.

One of the important features of GBM is the diffuse invasion of tumor cells within the surrounding brain parenchyma (44), thus rendering complete surgical resection impossible (45, 46). We showed that U87- or RNS-derived CD90^high tumors displayed a more invasive phenotype in an orthotopic mouse model compared with their CD90^low counterparts. This observation correlated with patients' data as CD90^high tumors also presented invasive imaging features. Interestingly these invasive features were previously linked to other gene signatures including mitochondrial dysfunction and remodelling of epithelial adherens junctions; and displayed shorter patients' survival (34). The CD90-dependent invasive characteristics of GBM cells were then correlated to gene expression and signaling data. Interestingly, CD90^high GBMs were characterized by an adhesion/migration gene signature and were enriched in GBM mesenchymal type, exhibiting elevated expression of mesenchymal markers such as αSMA, COL1A1, COL1A2, and MMP-2 and -9. Overexpression of these specific genes could be related to the increased invasiveness observed in CD90^high tumors and linked CD90-dependent signaling through p100, CD45, the SRC family kinases (SFK) LYN and FYN, and small G proteins (18–21). CD90 also regulates actin and tubulin cytoskeleton reorganization, focal disassembly, leading to modulation of cell migration (19, 20). In this study, we demonstrate that CD90 controls GBM cell migration/invasion mainly through SRC signaling. SRC and c-YES kinases have been recently involved in migration of glioma stem cells (47). The relevance of SRC signaling was confirmed by the identification of a SRC gene signature (41) in CD90^high GBM primary cell lines compared with their CD90^low counterparts. Furthermore, increased SRC phosphorylation was observed in freshly dissociated CD90^high GBM tumors. Interestingly SRC phosphorylation has previously been observed in GBM cells of the center and the borders of the tumor site as well as in invasive tumor cells within the brain parenchyma (48, 49). We cannot presently completely rule out that other SFK family members might also be involved. However, we did not observe any c-YES activation in U251 and U87 cells modified for CD90 expression. Only a significant increase of FYN mRNA expression was also observed in CD90^high patients.

In the past few years, intensive research programs have identified new therapeutic agents that target glioma migration/invasion (45, 46). For instance inhibition of metalloproteinases (50–52), blockade of integrins (53–55), targeting of cytoskeleton reorganization (56), and inhibition of signaling molecules such as FAK (57–59) and SFK (48, 60–62) showed promising effects on GBM invasiveness in vitro and GBM progression in mouse models (45, 46). Some of these molecules have also been used in recent GBM clinical trials. As such marimastat, a MMP inhibitor, showed encouraging effects on recurrent GBM patients (63) but failed to improve patient survival in a phase III clinical trial (64). Cilengitide, a αvβ3 and αvβ5-integrins antagonist, combined with temozolomide showed limited effects on GBM patients (65). Dasatinib showed promising effect on inhibiting bevacizumab-induced glioma cell invasion at a preclinical stage (62), but failed to improve bevacizumab-treated recurrent GBM patients in a phase II trial (66). Interestingly, SFK family kinases including SRC, FYN, and c-YES are involved in glioma proliferation and motility in vitro (67). LYN and c-YES have opposite effects on survival in a glioma orthotopic xenograft model. Here we show that dasatinib affects the viability of CD90^high RNS cells and blocks CD90-dependent GBM migration in vivo. We propose a model that defines the rational for using dasatinib in CD90^high GBM patients by targeting both GSC proliferation and GBM cells migration/invasion (Fig. 6F). Our results strongly emphasize the need of readdressing dasatinib response in GBM patients following a CD90-based stratification This type of approach could for instance be completely applicable for plurilobar GBM tumors (i.e., that can neither be resected nor irradiated, as observed in the multifocal/multicentric group from the TCGA cohort).

In conclusion, our data point toward CD90 as a marker of tumor invasion and might also be considered as a GBM stratification tool for clinical trials testing new therapeutic agents that target SRC-dependent GBM migration/invasion. Our results also pave the way for novel therapeutic approaches targeting CD90 and its downstream signaling to be applied to GBM patients.

J. Mosser is a consultant/advisory board member for Sophia Genetics. No potential conflicts of interest were disclosed by the other authors.

Conception and design: T. Avril, V. Quillien

Development of methodology: T. Avril, R. Pineau, F. Jouan

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): T. Avril, A. Etcheverry, R. Pineau, J. Obacz, G. Jegou, F. Jouan, P.-J. Le Reste, M. Hatami, R.R. Colen, B.L. Carlson, J.N. Sarkaria, E. Vauléon

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): T. Avril, A. Etcheverry, R. Pineau, J. Obacz, G. Jegou, F. Jouan, M. Hatami, R.R. Colen, P.A. Decker, D.C. Chiforeanu, E. Chevet, V. Quillien

Writing, review, and/or revision of the manuscript: T. Avril, A. Etcheverry, J. Obacz, P.-J. Le Reste, R.R. Colen, P.A. Decker, J.N. Sarkaria, E. Vauléon, D.C. Chiforeanu, A. Clavreul, J. Mosser, E. Chevet, V. Quillien

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): T. Avril, R.R. Colen, E. Chevet

Study supervision: T. Avril, B.L. Carlson, E. Chevet

Other (tissue samples): P.-J. Le Reste

The authors thank Caroline Gouat and Ester Porée for their excellent technical assistance; Laurent Riffaud, Claire Haegelen, and the medical staff of the Neurosurgery department at the CHU Pontchaillou (Rennes) for their contribution; Alain Fautrel and Pascale Bellaud from the Biosit histopathology H2P2 platform (Université de Rennes 1, France) for IHC analyses on GBM specimens and tumor xenografts; the Biosit ARCHE animal facility (Université de Rennes 1) for animal housing; and Pierre-Antoine Eliat from the Biosit PRISM platform (Université de Rennes 1) for mouse MRI analyses.

This work was supported by grants from la Ligue Contre le Cancer Comité d'Ille-et-Villaine, d'Indre-et-Loire et du Morbihan (to T. Avril); Région Bretagne AAP CRITT santé 2013 and Aidez la recherche! from the Centre Eugène Marquis (to V. Quillien); and la Ligue Contre le Cancer Comité des Landes (LARGE project), l'Institut National du Cancer (INCa_5869, INCa_7981, PLBIO: 2015-111), and EU H2020 MSCA ITN-675448 (TRAINERS; to E. Chevet).

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.