Involvement of the TGFβ pathway in the regulation of α5β1 integrins by caveolin-1 in human glioblastoma

Authors


Abstract

Caveolin-1 plays a crucial role in the development of cancer and its progression. We previously reported that glioblastoma cells expressing low levels of caveolin-1 exerted a more aggressive phenotype than cells expressing high levels. Such phenotype was due to the induction of α5β1 integrin subsequent to the depletion of caveolin-1. Caveolin-1 was identified as a transcriptional repressor of α5β1 integrin. The current study was designed to identify in vitro, the molecular mechanisms by which caveolin-1 controls α5β1 integrin expression and to determine if a negative correlation between caveolin-1 and α5β1 integrins also exists in biopsies and xenografted human brain tumors. We showed that depletion of caveolin-1 lead to the activation of the TGFβ/TGFβRI/Smad2 pathway which in turn induced the expression of α5β1 integrins. We showed that cells expressing the lowest levels of caveolin-1 but the highest levels of α5β1 integrins and TGFβRI were the most sensitive to a α5β1 integrin antagonist and a TGFβRI inhibitor. Screening human glioma biopsies and human glioblastoma xenografts, we isolated subgroups with either low levels of caveolin-1 but high levels of α5β1 integrin and TGFβRI or high levels of caveolin-1 but low levels of α5β1 integrin and TGFβRI. In conclusion, caveolin-1 controls α5β1 integrin expression through the TGFβ/TGFβRI/Smad2 pathway. The status of caveolin-1/α5β1 integrins/TGFβRI might be a useful marker of the tumor evolution/prognosis as well as a predictor of anti-TGFβ or anti-α5β1 integrin therapies.

Caveolin-1 is the principal structural protein of caveolae. Its ability to control the subcellular distribution, activity and expression of various proteins1–8 confers to caveolin-1 a key role in the regulation of processes altered in cancer. It is therefore not surprising that sporadic mutations or altered expression of caveolin-1 are observed during cell transformation, invasion and metastasis.3, 4

Glioblastoma is the most common intrinsic brain tumor in adults and is nearly uniformly fatal. Despite advances in neurosurgery, radiation and medical oncology, the prognosis for patient with glioblastoma remains poor with a median survival of 9–15 months.9 Although increased expression of caveolin-1 seems to be a norm in glioma,10–15 caveolin-1-positive and caveolin-1-negative cells coexist in many glioblastoma.15 Recently, by manipulating the level of caveolin-1 in glioblastoma cells, we showed that a loss of caveolin-1 shifted cells towards a more aggressive phenotype.16 Looking for the genes involved, we showed that caveolin-1 acted as a transcriptional repressor of α5β1 integrins. Consequently, caveolin-1 and α5β1 integrin levels are inversely correlated and α5β1 integrins were identified as the mediator of the phenotypic changes induced by caveolin-1.16 The loss of caveolin-1 endowed cells with high levels of α5β1 integrins, leading to aggressiveness, and it also confers cells a stronger sensitivity toward a nonpeptidic α5β1 integrin antagonist.16

Nonetheless, the mechanism by which α5β1 integrin expression is regulated by caveolin-1 is still unknown. We reported here that caveolin-1 controls α5β1 integrin expression through the TGFβ/TGFβRI/Smad2 pathway. Data unraveled the possibility to use either anti-α5β1 integrin or anti-TGFβ therapies for glioblastoma treatment. The relevance of our studies was further illustrated on human glioma biopsies and xenografts.

Material and Methods

Cell lines, culture conditions and transfection

Cell lines were cultured in EMEM supplemented with 10% heat-inactivated FCS and 0.6 mg/mL glutamine (Lonza). U87MG and U373 cell lines were from the American Tumor Cell Collection. LN-18, LN-229 and LN-319 were a generous gift from Pr M. Hegi (University of Lausanne, Switzerland).

Chemicals

The compound 34c (2-(S)-2,6-dimethlybenzamido)-3-[4-(3-pyridin-2-ylaminopro oxy)-phenyl] propionic acid) named K34c was synthesized in our laboratory according the procedure described by Heckmann et al.17 U0126 (Cell Signaling) was used at 10 μmol/L for 24 hr. SB431542 (Sigma) was used at 10 μmol/L up to 72 hr. TGFβ1 (Sigma) was used at 1-2 ng/mL up to 72 hr. Activin (MACS Miltenyi Biotec) was used at 1–2 ng/mL up to 72 hr.

Small-interfering RNA and plasmids

SiRNA (Dharmacon Research) was transfected at 100 nmol/L for 72 hr as recommended. When double/triple transfections (siRNAcav1, siRNASmad2, siRNAERK1, siRNAERK2) were performed, siRNA was co-transfected at 50 nmol/L for 72 hr. Efficiency of protein silencing was determined by western blotting and densitometry analysis. ShRNA (OpenBiosystems) were transfected at 2.5 μg and stable cell lines were selected with puromycine. Other constructs were created as previously described,16 transfected at 1 μg and stable cell lines were selected with G418.

Real-time quantitative PCR

RNA was extracted as previously described.16 mRNA expression of the human ITGA5, CAV1 and TGFBRI were evaluated by relative quantitative RT-qPCR analysis using the ABI7000 SYBRGreen PCR detector with the primers (Invitrogen) described in Supporting Information Data 1. Target cDNA expression was quantified using the comparative ΔΔCt method.18

Western blot

Cells were lysed 30 minutes on ice in lysis buffer (1% Triton-X100, NaF 100 mmol/L, NaPPi 10 mmol/L, Na3VO4 1 mmol/L in PBS, supplemented with Complete anti-protease cocktail; Roche). Protein (10 μg) was separated by SDS-PAGE (BioRad) and transferred to PVDF membranes (Amersham). For phosphorylation detection, cells were directly lysed in Laemmli buffer (BioRad) supplemented by β-mercaptoethanol (1:20). Blots were probed with anti-α5, anti-β1 and anti-GAPDH (Chemicon), anti-caveolin-1 and anti-TGFβRI (Santa Cruz Biotechnologies), anti-phospho-ERK-1/2, anti-ERK-1/2, anti-phospho-Smad2, anti-Smad2/3 (Cell Signalling), followed by the HRP-conjugated antibodies (1:10000, Promega). Proteins were visualized with enhanced chemiluminescence.

Clonogenic assay

Clonogenic assays were performed using stable shRNA-transfected cell lines. Cell lines are maintained in culture with the antibiotic of selection to prevent the loss of the plasmid and therefore the re-appearance of caveolin-1. Therefore, the transfection efficiency should not influence results. After attachment (250 cells/well, 6 well plate), cells were treated 72 hr with solvent (DMSO), SB431542 (5 or 10 μmol/L), K34c (10 or 30 μmol/L) or a combination of both in 2%-FCS containing medium. Medium was refreshed and cells were allowed to regrow in fresh medium containing 10% FCS for 6 days. Clonogenic survival was determined as previously described.16 Data were expressed as surviving fraction (SF). SF was determined as follows: PE (plating efficiency) = number of surviving cells/number of cells plated; SF = PE of experimental group (K34c or SB431542 or both drugs)/PE of control group (solvent). PE = 0.4 ± 0.1 and 0.6 ± 0.1 for shRNActrl and siRNAcav-1, respectively.

Quantification of human TGFβ1

Cells were plated at 3 × 105 cells/well in 60-mm Petri dishes and grown in serum-free medium for 24 and 72 hr. Supernatant were collected and activated TGFβ1 was measured using the Quantikine® human TGFβ1 immunoassay according to manufacturer's instructions (R&D Systems). Results are expressed as ng of TGFβ1 produced per ml of supernatant.

Human tissue samples

This study was conducted on a total of 108 adult brain tumors collected retrospectively from archival material stored at the Centre de Ressources Biologiques et Tumorothèque (Hôpitaux Universitaires de Strasbourg). All patient characteristics are summarized in the Supporting Information Data 2. Each sample was histologically analyzed by a pathologist to specify the tumor grade and the percentage of tumor cells. Control tissues were obtained from epileptic surgery. The study was conducted in accordance with the Declaration of Helsinski and each patient was entered in the study after his/her consent. RNA was purified with RNeasy Mini Kit (Qiagen). RNAs were transcribed into cDNA as described previously.16 Real-time quantitative PCR was performed using the ABI7000 SYBRGreen PCR detector with the primers described in Supporting Information Data 1.

K-means clustering analysis

In order to pool patients into homogenous groups with respect to their levels of CAV1, TGFBRI and ITGA5, k-means clustering procedure, which aims to partition the points into K groups such that the sum of squares from point to the assigned cluster centers is minimized, has been performed.19 Because it could clinically make sense, and strengthened by a local maximum of the Calinski criterion, we choose to divide the data in three 3-D-clusters.

Xenografts and immunohistochemistry

Protein expression analysis was performed using immunochemical methods on 18 tumor samples obtained from 18 glioma xenografts. Xenografts were generated in female athymic NMRI-nude mice as previously described20 and all animal experiments were carried out in accordance with the French Animal Scientific Procedures Act. When the xenografts diameter reached up to 1 cm, tumors were excised and samples were fixed in 4% formaldehyde (pH 7.4) for 16 hr, processed and embedded in paraffin. Five micrometer thick adjacent tissue sections were deparaffinized and rehydrated, and endogenous peroxidase was inhibited with 6% H2O2/distilled H2O. For antigen retrieval, slides were boiled in 0.01 mol/L (pH 6.0) sodium citrate buffer for 10 minutes. After blocking, primary anti-caveolin polyclonal antibody (1:400, N20, Santa Cruz), anti-α5 polyclonal antibody (1:200, H104, Santa Cruz) and anti-β1 polyclonal antibody (1/200, AB52971, Abcam) diluted in 0.1 mol/L PBS, 0.3% BSA, 0.1% sodium azide, 0.06% n-ethyl-maleimide, 20% glycerol were incubated at 4°C overnight. Primary antibodies were detected with Histofine Simple Stain Rabbit Max PO (Nicheiri). Bound peroxidase was identified using the Novared TR system (Abcys). Nuclear counterstaining was performed with 1/2-diluted Harris hematoxylin. Image acquisition was carried out with a microscope (Axiophot II, Zeiss) equipped with a cooled AxioCam HRc CCD camera (Zeiss) controlled by the Axiovision 4.4 digital image processing software.

Statistical analysis

Data are represented as mean ± SEM. In all cases, n refers to the number of independent experiments. Statistical analyses were done with the Student's t-test and ANOVA where p < 0.05 was considered significant.

Results

Caveolin-1 modulates the TGFβ/TGFβRI/Smad2 pathway

Caveolin-1 was identified as a negative regulator of α5β1 integrins16 and integrins are known targets of TGFβ.21–25 We determined if caveolin-1 was able to affect the TGFβ pathway in glioblastoma cells. In U87MG cells, depletion of caveolin-1 was associated with an increased secretion of TGFβ1, an increased expression of TGFβRI and an increased phosphorylation of Smad2 (shRNAcav-1 versus shRNActrl, Fig. 1a and 1b). Opposite effects were observed when caveolin-1 was overexpressed (pEGFPcav-1 versus pEGFPctrl, Fig. 1a and 1b). Data suggest that caveolin-1 controls the secretion of TGFβ, the expression of TGFβRI and the activation status of Smad2 in U87MG cells.

Figure 1.

Caveolin-1 modulates the TGFβ1/TGFβRI/Smad2 pathway. (a) Secretion of TGFβ1 in shRNActrl-, shRNAcav1-, pEGFPctrl- or pEGFPcav1-transfected U87MG was measured by immunoassay. Histograms show ng of TGFβ1 secreted per mL of supernatant (n = 4–6, *p < 0.05, **p < 0.01 and ***p < 0.001). (b) Expressions of α5β1 integrin, TGFβRI, phospho-Smad2, Smad2 and caveolin-1 were detected by western blot in shRNActrl-, shRNAcav1-, pEGFPctrl- or pEGFPcav1-transfected U87MG. Histograms show the fold increase in the protein expression or phosphorylation normalized to GAPDH levels (n = 3–5, *p < 0.05, **p < 0.01).

Caveolin-1 modulates α5β1 integrins through the TGFβ/TGFβRI/Smad2 axis

The involvement of the TGFβ/Smad2 pathway in the regulation of α5β1 integrins by caveolin-1 was investigated. U87MG cells were exposed to TGFβ1 at concentrations similar to those achieved during caveolin-1 depletion (see Fig. 1a). Under such conditions, TGFβ1 increased α5β1 integrin protein expression (Fig. 2a), whereas no alteration of α5β1 integrin expression was observed in response to activin, another member of the TGFβ superfamily (Fig. 2a). Blocking TGFβRI with SB431542, a TGFβRI inhibitor, prevented the induction of α5β1 integrin observed after depletion of caveolin-1 both at the protein (Fig. 2b) and mRNA level (Fig. 2c). The activity of SB431542 was confirmed by the abolition of Smad2 phosphorylation observed after caveolin-1 depletion (Fig. 2b). Finally, blocking the activation of Smad2 with siRNA prevented the induction of α5 integrin subunit observed after depletion of caveolin-1 (Fig. 2d). Key results were confirmed in another glioblastoma cell line, U373. Blocking TGFβRI using SB431542 or blocking specifically the activation of Smad2 with siRNA prevented the induction of α5β1 integrin observed after depletion of caveolin-1 (Fig. 2e). Altogether, data confirm that the depletion of caveolin-1 lead to the secretion of TGFβ that induces the activation of the TGFβRI and subsequent Smad2 phosphorylation resulting in the induction of α5β1 integrin in glioblastoma cells.

Figure 2.

Caveolin-1 modulates α5β1 integrins through the TGFβ1/TGFβRI/Smad2 pathway. (a) α5 integrin, β1 integrin and caveolin-1 were detected by western blot in U87MG treated 72 hr with solvent (DMSO), TGFβ1 (1 and 2 ng/mL) or activin (1 and 2 ng/mL). Histograms show the fold increase in the protein expression normalized to GAPDH levels (n = 4, *p < 0.05, **p < 0.01, ***p < 0.001). (b) Expressions of α5 integrin, β1 integrin, phospho-Smad2, Smad2 and caveolin-1 were detected by western blot in shRNActrl- and shRNAcav1-transfected U87MG treated 72 hr with solvent (DMSO) or SB431542 (10 μmol/L). Histograms show the fold increase in the protein expression normalized to GAPDH levels (n = 4, *p < 0.05, **p < 0.01). (c) Histograms represent the fold increase in ITGA5 and ITGB1 mRNA in shRNActrl- and shRNAcav1-transfected U87MG treated 72 hr with solvent (DMSO) or SB431542 (10 μmol/L). (d) Expressions of α5 integrin, caveolin-1 and Smad2 were detected by western blot in siRNActrl- and siRNAcav1-transfected U87MG cotransfected with siRNActrl and siRNASmad2. Histograms show the fold increase in the protein expression normalized to GAPDH levels (n = 4–6, *p < 0.05). (e) Left panel, Expressions of α5 integrin, β1 integrin, caveolin-1 and phospho-Smad2 were detected by western blot in siRNActrl- and siRNAcav1-transfected U373 treated 72 hr with solvent (DMSO) or SB431542 (10 μmol/L). Right panel, Expression of α5 integrin, caveolin-1 and Smad2 were detected by western blot in siRNActrl- and siRNAcav1-transfected U373 cotransfected with siRNActrl and siRNASmad2.

Involvement of the ERK pathway in Smad2 activity

Funaba et al. reported a direct phosphorylation of Smad2 by ERK.26 Blocking ERK with U0126 prevented the induction of α5β1 integrin as well as the phosphorylation of Smad2 observed after depletion of caveolin-1 (Fig. 3a). The activity of U0126 was confirmed by the abolition of ERK phosphorylation observed after caveolin-1 depletion (Fig. 3a), while TGFβRI expression remained unaffected (not shown). The involvement of ERK in the regulation of α5β1 integrin expression was confirmed using siRNA directed against ERK1, ER2 and both ERK1/2. As observed using U0126, the silencing of ERK1/2 prevented the induction of α5β1 integrin observed after the depletion of caveolin-1 (Fig. 3b). Blocking TGFβRI using SB431542 did not prevent the phosphorylation of ERK observed after depletion of caveolin-1 (Fig. 3c). Data suggest that ERK is indeed involved in the regulation of α5β1 integrin expression by caveolin-1 through the activation of Smad2. As ERK remained unaffected by the blockade of TGFβRI, the regulation of Smad2 by ERK seems independent of TGFβ/TGFβRI axis. These data uncover a secondary pathway by which Smad2 regulates α5β1 integrin.

Figure 3.

Involvment of ERK pathway in Smad2 activity. (a) Expressions of α5 integrin, β1 integrin, phospho-ERK, ERK, phospho-Smad2, Smad2 and caveolin-1 were detected by western blot in shRNActrl- and shRNAcav1-transfected U87MG treated 24 hr with solvent (DMSO) or U0126 (10 μmol/L). Histograms show the fold increase in the protein expression or phosphorylation normalized to GAPDH levels (n = 3–6, *p < 0.05, **p < 0.01, ***p < 0.001). (b) Expressions of α5 integrin, β1 integrin, ERK and caveolin-1 were detected by western blot in siRNActrl- and siRNAcav1-transfected U87MG cotransfected with siRNActrl, siRNAERK1, siRNAERK2 and siRNAERK1/2. Histograms show the fold increase in the protein expression normalized to GAPDH levels (n = 6, *p < 0.05). (c) Expressions of phospho-ERK, ERK and caveolin-1 were detected by western blot in shRNActrl- and shRNAcav1-transfected U87MG treated 24 hr with solvent (DMSO) or SB431542 (10 μmol/L). Histograms show the fold increase in the protein expression or phosphorylation normalized to GAPDH levels (n = 4, **p < 0.01).

Antagonizing α5β1 integrins and inhibiting the TGFβ pathway significantly reduce the clonogenic survival of human glioblastoma

We next investigated the effect of the α5β1 integrin antagonist K34c and the TGFβRI inhibitor SB431542 on the clonogenic survival of U87MG. The study was performed in U87MG expressing low levels of caveolin-1 to induce a more aggressive phenotype due to the high content of α5β1 integrins16 and activate the TGFβ/TGFβRI/SMAD2 axis. K34c (Fig. 4a) or SB431542 (Fig. 4b) dose-dependently reduced the surviving fraction of caveolin-1-depleted cells without affecting control cells. The combination of both inhibitors did not induce further reduction of the surviving fraction (Fig. 4c). These data confirm our previous study16 suggesting that tumors expressing high levels of α5β1 integrin thus, highly aggressive, might be more responsive to α5β1 integrin antagonists and/or anti-TGFβ therapies.

Figure 4.

Antagonizing α5β1 integrins and inhibiting the TGFβ pathway significantly reduce the clonogenic survival of human glioblastoma cells. shRNActrl- and shRNAcav1-transfected U87MG were treated 72 hr with (a) solvent (DMSO) or K34c (10 or 30 μmol/L); (b) solvent (DMSO) or SB431542 (5 or 10 μmol/L); (c) solvent (DMSO) or a combination of K34c (10 μmol/L) and SB431542 (5 μmol/L). Histograms represent the surviving fraction of shRNAcav1- versus shRNActrl-transfected cells in presence of the solvent or drugs. Plating efficiency are 0.4 ± 0.1 and 0.6 ± 0.1 for shRNActrl and shRNAcav1, respectively (n = 5–8, *p < 0.05, **p < 0.01, ***p < 0.001).

Caveolin-1 is inversely correlated to α5β1 integrins and the TGFβ pathway in glioblastoma cell lines as well as in human brain tumor biopsies

Western blot analysis in three additional glioblastoma cell lines revealed a significant negative correlation between the α5 integrin subunit and caveolin-1 (Fig. 5a). Although not significant, data clearly evidenced a trend toward a negative correlation between phospho-Smad2/TGFβRI and caveolin-1. TGFβRI is positively correlated to phospho-Smad2 (not shown).

Figure 5.

Caveolin-1 is inversely correlated to α5β1 integrins and the TGFβ pathway in several glioblastoma cell lines as well as in human brain biopsies. (a) Expressions of α5 integrin, β1 integrin, caveolin-1, phospho-Smad2, Smad2 and TGFβRI were detected by western blot in LN-18, LN-229, LN-319, U87MG and U373 and normalized to GAPDH levels. Plots show the reverse correlation between indicated proteins (n = 3, *p < 0.05). (b) Left panel, K-means clustering analysis of ITGA5, CAV1 and TGFβRI. Black dots = cluster I, green dots = cluster II and red dots = cluster III. Right panel represents the cluster centers or the average fold increase into brackets. (c) Tumor origin, number and percentage of samples in each clusters.

Real-time PCR analysis in 108 human brain biopsies showed that ITGA5, CAV1 and TGFBRI levels increased in parallel with tumor aggressiveness (Supporting Information Data 3). Using K-means clustering analysis, tumors could be separated into three subgroups according to the levels of ITGA5, CAV1 and TGFBRI. Cluster I assembled tumors with low levels of the three genes (black dots, Fig. 5b and 5c), whereas tumors in cluster II (green dots, Fig. 5b) and cluster III (red dots, Fig. 5b) expressed higher levels of the three genes. In a 3D graphic, tumors in cluster II localized in an area of low CAV1 expression but high ITGA5 and TGBRI expression and cluster III in an area of high CAV1 expression but low ITGA5 and TGBRI expression (Fig. 5b). Interestingly, cluster II becomes enriched during tumor progression and half of the grade IV gliomas are distributed in clusters II and III (Fig. 5c). Altogether, our results corroborate the in vitro data that suggested that caveolin-1 negatively regulates α5β1 integrin through the TGFβ pathway as confirmed by the evidence of accessory subgroups of tumors expressing either low levels of caveolin-1 but high levels of α5β1 integrin and TGFβRI (cluster II) or high levels of caveolin-1 but low levels of α5β1 integrin and TGFβRI (cluster III).

Xenografted tumors exert opposite pattern of caveolin-1 and α5 integrin staining

Finally, we wanted to determine if transcriptomic data obtained from biopsies could be validated at the protein level in human xenografted glioblastoma biopsies. On 18 tumors analyzed, seven had similar caveolin-1 and α5 integrin staining intensity (representative slide 1 in Fig. 6), three were intensively stained for α5 integrin but poorly or negatively stained for caveolin-1 (representative slide 2 in Fig. 6) and eight were intensively stained for caveolin-1 but poorly or negatively stained for α5 integrin (representative slide 3 in Fig. 6). Although β1 integrin subunit staining was observed in all tumors, the staining intensity remains inversely correlated to caveolin-1 (slides 1–3 in Fig. 6). As observed for α5 integrin, on 18 tumors analyzed, seven had similar caveolin-1 and β1 integrin staining intensity, three were stained for β1 integrin but not for caveolin-1 and eight were intensively stained for caveolin-1 but poorly stained for β1 integrin (see the table in Fig. 6).

Figure 6.

Xenografted tumors exert opposite pattern of caveolin-1 and α5β1 integrin staining. Left panel: staining pattern of caveolin-1 and α5 integrin. Right panel: immunostaining of xenografed tumors displaying (1) similar caveolin-1 and α51 integrin intensity, (2) intensive staining for α51 integrin but weak or negative staining for caveolin-1 and (3) intensive staining for caveolin-1 but weak or negative staining for α51 integrin. Magnification bars in red = 50 μm.

Discussion

We previously reported that caveolin-1 acts as a repressor of α5β1 integrin expression in glioblastoma cell lines. The loss of caveolin-1 frees α5β1 integrin from repression which endow glioma cells with an aggressive phenotype as cells become more proliferative and invasive.16 Looking for mechanisms by which caveolin-1 might regulate α5β1 integrin, we found that caveolin-1 represses the TGFβ pathway in high-grade glioma cells. The TGFβ pathway has been shown oncogenic in glioma since TGFβ is able to confer survival advantage to tumors27 and the knock-down of TGFβR abolished TGFβ-induced glioblastoma invasiveness in vitro.28 Also, high phospho-Smad2 level is a poor prognosis marker in glioma.29 TGFβ1 was reported to enhance the expression of several integrin.21–25 In accordance, we showed that exogenous TGFβ1 leads to the induction of α5β1 integrin. Moreover, blocking TGFβRI and Smad2 totally abolished the induction of α5β1 integrin observed after caveolin-1 depletion. To our knowledge, data reported here are the first to show the involvement of the TGFβ pathway in the regulation of α5β1 integrin expression by caveolin-1 in glioblastoma cells. Additionally, the MEK/ERK inhibitor and the depletion of ERK using siRNA were able to block both the phosphorylation of Smad2 and the induction of α5β1 integrin following caveolin-1 downregulation. As the activation of ERK resulting from caveolin-1 depletion remained unaffected by TGFβRI inhibitors, Smad2 seems to be able to regulate α5β1 integrins independently of the TGFβRI. In accordance, Funaba et al. reported a direct phosphorylation of Smad2 by ERK resulting in the stabilization of Smad2 and an increase of its transcriptional activity.26 Therefore, α5β1 integrin expression is not regulated by a single Smad2-dependent pathway but two, one being ERK-dependent and the other being TGFβ-dependent.

Antagonizing the biological effect of TGFβ has become a new experimental strategy to treat glioblastoma. Anti-TGFβR therapies have shown promise in preclinical and early clinical studies30–32 and seems to improve survival rates when compared with conventional chemotherapy.33 Considering α5β1 integrins, several publications including ours highlight their potential role as therapeutic targets in several cancers including glioma.16, 34–38 We showed that antagonizing TGFβRI or α5β1 integrins more efficiently blocked the survival of glioblastoma cells expressing low levels of caveolin-1 but high levels of α5β1 integrins and TGFβRI. Similar results were already obtained using another α5β1 integrin antagonist, SJ749.16 Therefore, anti-TGFβ/TGFβR/Smad or anti-α5β1 integrin therapies would be more beneficial for patients presenting the most aggressive subtype of glioblastoma. However, combining both drugs confers no advantage probably because both treatments target a common mediator, α5β1 integrin. Nonetheless, we propose the anti-α5β1 integrin therapy as an alternative to the anti-TGFβ therapy already in clinical trial. Profiling glioma using our markers might predict the response to anti-TGFβ/TGFβR/Smad or anti-α5β1 integrin therapies and might be taken into account in future clinical trials.

Finally, as caveolin-1 positive and negative tumor cells coexist in glioblastoma,15 we postulate that those cells might be respectively negative and positive for α5β1 integrin/TGFβ as observed in vitro. Although significant reverse correlations could be established using various human glioblastoma cell lines, none was found per se when human brain tumor biopsies were analyzed. Recently, two subgroups of glioblastoma showing strong or weak patterns of TGFβ activation were reported39 and strong activation of TGFβ was associated to high expression of many molecules involved in integrin signaling including the β1 integrin.39 Accordingly, we were able to discriminate patients in three subgroups of tumors among which a subgroup with high levels of α5 integrin/TGFβRI associated to weak expression of caveolin-1 (cluster II) and a subgroup that expresses low levels of α5β1 integrin/TGFβ but strong expression of caveolin-1 (cluster III). These observations strengthen the repressive effect of caveolin-1 on α5β1 integrin/TGFβ pathway that we uncovered in vitro but also suggest that integrin signaling is under the transcriptional control of TGFβ in human glioblastoma cells. Interestingly, the number of biopsies in cluster II and III increased with the tumor grade. Results suggest that far from being homogenous, high grade glioma express heterogeneous molecular profiles with strong or weak expression of α5β1 integrin/TGFβ pathway associated with weak or strong expression of caveolin-1. Combined with our in vitro data showing that strong expression of α5β1 integrin endowed cells with an aggressive phenotype and data in the literature showing that the overexpression of α5β1 integrin and TGFβ associated with Smad2 activation is related to tumor stage and associated with a poor prognosis in glioma,29, 40–42 the molecular profiling of glioma using three interconnected markers might predict the tumor prognosis and its evolution.

In conclusion, caveolin-1 controls α5β1 integrin expression through the TGFβ/TGFβRI/Smad2 pathway. The status of caveolin-1/α5β1 integrins/TGFβRI might be a useful marker of the tumor evolution/prognosis as well as a predictor of anti-TGFβ or anti-α5β1 integrin therapies. Our data open new aspects in the understanding of the caveolin-1/α5β1 integrin/TGFβ partnership in fundamental cell biology and its contribution to glioma biology.

Acknowledgements

We thank Madeleine Jaillet and Michèle Legrain for technical assistance.

Ancillary