PDGF receptor alpha inhibition induces apoptosis in glioblastoma cancer stem cells refractory to anti-Notch and anti-EGFR treatment

Cancer stem cells from GBM were isolated using defined criteria set up by neurosurgeons as described previously[24, 25]. We can summarize these briefly: lesion removal was achieved with resection margins that included the tumor and the neighboring, apparently normal tissue (between 1-2 cm from the tumor border; larger resections were performed in tumors that grew far from eloquent areas), which were removed entirely en bloc. Neuronavigation and intraoperative ultrasound were used to maximize the extent of intracranial tumor resection. From this bulk we retrieved either core- (c-CSC) or peritumor tissue-derived cancer stem cells (p-CSC). Cytogenetic and molecular analysis showed that the two types of CSC have quite diverse tumorigenic potential and distinct genetic anomalies[24]. Neurospheres of different sizes were obtained from cores of multiple specimen of GBM patients; these continued to propagate in suspension in long-term culture. CSC derived from peritumor tissue of GBM at early passages exhibited a different phenotypic behavior compared to c-CSC: they grew at a slow rate, forming small spheres, most of them attached to the plastic dishes. These latter particular morphological features, in some cases, were gradually lost at late passages in culture (data not shown).

To understand how Notch1 and epidermal growth factor receptor (EGFR) signaling would affect cell growth and survival of GBM CSC, we first assessed the mRNA expression profile in six human cases, consisting of paired samples of c-CSC and p-CSC, for a total number of twelve CSC. RT-PCR experiments for NOTCH1, HES1, EGFR wt and variant EGFRvIII, were performed in triplicate for each sample and the relative expression reported as -∆Ct (Figure 1A-C). Notably, the p-CSC3 and p-CSC4 showed a significant up regulation of NOTCH1 gene compared to relative c-CSC, either at mRNAs level or the protein content of the Notch intracellular domain 1 NICD1, (the active form of Notch1) (Figure 1A, E). We carried out in parallel a custom RT-PCR array in the most studied cases (cases 1-3), which revealed and confirmed the up modulation of Notch signaling components in p-CSC3 versus c-CSC3, expressed as fold change (Fc) and including: NOTCH3 (4.78 Fc), Nicastrin (NCSTN, 3.4 Fc), Presenilin1 (PSEN1, 2.39 FC), Mastermind-like 1-2 (MAML1, MAML2, 2.36 and 3.24 FC respectively), Delta-Like Ligand 1, (DLL1, 7.91 Fc), and Serrate Ligand Jagged2, (JAG2, 3.66 Fc) (Figure 1B, C). The high mRNA levels of HES1, a Notch1 primary target gene, directly correlated to those of Notch1 in p-CSC3 and p-CSC4, suggesting a Notch1 dependent mechanism for Hes1 gene regulation (Figure 1A, B). Conversely, the high levels of HES1 mRNA inversely correlated to Notch1 gene expression in p-CSC2 (Figure 1A, B), suggesting that other signals converged in case-2 for HES1 gene transcription. A custom RT-PCR array for genes encoding Notch signaling components confirmed the reduction of Notch1 activation in p-CSC2 as well as NICD1 protein expression as compared to c-CSC2 (Figure 1D, E). Hes1 protein was detected in all CSC, raising the possibility that further mechanisms may contribute to Hes1 protein stability through the sonic hedgehog pathway as well as post-translational processes[26, 27].

The only mRNA positivity relative to EGFRvIII was found in c-CSC1, which represent around 16% of the specimens studied. EGFR mRNA expression was up modulated in c-CSC1, 3 and 4, down modulated in c-CSC2 compared to relative counterparts, or unmodulated in the rest of the cases (Figure 1C). EGFR protein expression mirrored the levels of transcripts in the same samples. We extended our protein analysis to another RTK, the platelet derived growth factor receptor isoforms α and β (PDGFRα/β). The inherent over expression of PDGFRα in p-CSC pools suggests this as a molecular signature of the GBM peritumor tissues, while PDGFRβ was expressed either in c-CSC or p-CSC of the six cases examined (although we have documented increasingly expression levels in p-CSC pools except for Case 6) (Figure 1E). We also reported the over expression of c-Myc, a direct target of Notch1[10], which was equally express in all CSC.

CSC undergoing treatment with γ-secretase inhibitor-X (GSI-X), a Notch signaling inhibitor, either alone or along with an EGFR signaling inhibitor AG1478, were evaluated for any changes in cellular response by MTS assay. Here, the experiments were conducted only on cases 1, 2 and 3 (Figure 2). c-CSC2 and c-CSC3 were significantly sensitized with 2 μM of GSI-X after 3 days. Conversely, p-CSC2 showed an increase of cell proliferation, while p-CSC3 were not affected even at higher concentration of GSI-X (Additional file1: Figure S1). In Case 1, both c-CSC1 and p-CSC1 showed resistance to the treatment after 3 days (Figure 3A); instead the treatment on the 5th day dropped proliferation of c-CSC1 to 70 ± 4.6% (Figure 2A). The treatment with 10 μM of AG1478 resulted in a significant sensitization of all six CSC as detected by MTS assay (Figure 2B).

No close correlation between sensitization to AG1478 with the levels or the mutational state of EGFR gene was found (Figure 1C, E). In fact, cell proliferation decreased to 28.3 ± 6.1% and 30,9 ± 8.3% in c-CSC1 and p-CSC1, respectively (Figure 2B), although c-CSC1 would overexpress both full-length and the variant III of EGFR (Figure 1E). In c-CSC2 and c-CSC3 cell proliferation dropped to 44,8 ± 6.4% and 49 ± 2.1% respectively against to 59.2 ± 4.1% of p-CSC2 and 65.2 ± 5.5% of p-CSC3. The treatment with the two inhibitors caused a synergistic anti-proliferative effect in c-CSC2 and c-CSC3, bringing cell proliferation down to 19 ± 8.3% and 34 ± 3.8% respectively (Figure 2B), while growth of p-CSC2 and p-CSC3 remained unmodified at 57 ± 2.9% and 70 ± 5.6%, respectively. This finding argued for a stronger anti-Notch1 and EGFR resistance of p-CSC vs. c-CSC pools, at least in these 2 cases.

To substantiate the effects observed by MTS assay, we carried out Western blots analysis for the main signal molecules relevant for proliferation and cell survival (Figure 3). The concentration of the inhibitors used in our treatment was effective in inhibiting Notch1 cleavage and EGFR phosphorylation on Try1068. Notably, only c-CSC1 showed a protein doublet of 180 and 140 KDa (corresponding to the full-length and EGFRvIII respectively); the rest of cases displayed the full lenght isoform (Figure 3). We evaluated the activation and modulation of the main downstream effectors of EGFR signaling, such as Akt1, Erk1/2 and Stat3, after two days of treatment. No significant changes of phosphorylation state of p-Akt1 and p-Erk1/2 were found in the three cases examined after 2 days of treatment with GSI-X. p-CSC3 displayed resistance even to higher concentration of GSI-X (Additional file1: Figure S1). Conversely, p-Erk1/2 was suppressed in case 1, and significantly down modulated in case 2 and 3, except for p-CSC3, following AG1478 treatment. The combination of the two agents suppressed p-Erk1/2 expression in case 2 and 3, but it was still preserved in p-CSC3 and similarly results were seen for p-Akt-1 activation. Stat3 phosphorylation on Y705 has been shown to be directly mediated by EGFR or indirectly by proteins of JAK family. EGFR signaling inhibition caused a decrease of phosphorylation of p-Stat3 in case 2 and c-CSC3, conversely in p-CSC1 and p-CSC3 the levels of p-Stat3 were unmodulated, suggesting a mechanism of resistance to EGFR TKI treatment in these cases. PDGFRβ was more expressed in p-CSC than in c-CSC pools and unmodulated by either GSI-X or AG1478 alone, except for CSC2 that showed PDGFRβ down modulation by AG1478 treatment (Figure 3A-C).

The main biological responses of cancer stem cells described in the literature following treatment with Notch inhibitors are: arrested cells and induction of apoptosis. Flow cytometry analyses showed a moderate apoptosis/necrosis induced by GSI-X, which resulted 4.7% and 12% in c-CSC2 and c-CSC3 respectively respect to 0.3% and 5% in DMSO-treated cells (Figure4B, C). Single AG1478 treatment was more effective than GSI-X, as c-CSC2 and c-CSC3 were both shifted in G1 phase (61.8% and 82.1% respectively as compared to 54.2% and 74.5% in DMSO-treated cells) and induced to apoptosis/necrosis (9% and 10%, respectively).

Dual treatment raised apoptosis/necrosis to 10.2% in c-CSC2 and 27% in c-CSC3. These results were verified by induction of Caspase-3 and PARP-1 cleavage, reduction of both Survivin and Cyclin D1 as reported by Western blots (Figure 4B, C). AG1478 determined the accumulation of p-CSC2 in G1 phase (74,9%), while the dual treatment raised this to 78,7% compared to 61.3% in DMSO-treated cells. The slowing of cell cycle progression was associated with loss of Cyclin D1 and increase of p27^Kip, but we monitored very low levels of Caspase-3 and PARP-1 cleavage fragments, and no effect on Survivin protein levels. Conversely, p-CSC3 were fully resistant to treatment (Figure 4C). EGFR inhibition induced cell cycle arrest of both c-CSC1 and p-CSC1 which amounted to 84.3% and 87.9% respectively versus 66.2% and 58.1% in DMSO-treated cells. This finding correlated with high induction of p27^Kip1 (Figure 4A). Cyclin D1 was not detected by Western blots, although mRNAs were monitored by custom RT-PCR array (data not shown).

In order to assess the invasion capabilities of CSC and the effects of the inhibitors either alone or in combination a matrigel invasion assay was performed. To invade the cells need an active process that involves cell motility and extracellular matrix proteolysis, and for this reason we monitored the expression of MMP-9, matrix metallopeptidase-9, which belongs to MMP family involved in tumor progression, including invasion and cancer metastasis[28]. Cell invasion ability was assessed in cases 1, 2 and 3 in absence or presence of growth factors (GF). In absence of GF, both p-CSC1 and p-CSC2 hold higher invasive ability compared to relative counterparts; conversely, c-CSC3 are much more invasive than p-CSC3. Following addition of GF, all CSC moved significantly through the plastic insert of the trans-wells dishes (Figure 5A-C). GSI-X treatment did not affect the invasive ability of any CSC, except for c-CSC3, although other authors report that genetic Notch1 knockdown resulted in reduced cell migration invasion in a glioma cell line by inhibition of β-catenin and NF-κB signaling[29].

It has been reported that EGFR inhibition efficiently blocked EGF-induced activation of MMP9 and cancer invasiveness[30]. In our work AG1478 treatment induced differential effects, in part explained by the RTK profile. AG1478 inhibited cell invasion and MMP9 protein expression in c-CSC1 and p-CSC1, whereas the combination therapy resulted in suppression of MMP-9 (Figure 5A). AG1478 alone did not affect the invasive ability of c-CSC2 and p-CSC2, but the dual treatment influenced significantly both pools of cells and directly correlated with a stronger reduction of MMP-9 in c-CSC2 than in p-CSC2 (Figure 5B). Similar outcomes were obtained in c-CSC3, and the influence of this combination treatment was superior to monotherapy with AG1478. Conversely, p-CSC3 had a modest invasion ability compared to c-CSC3, probably due to intrinsic reduced expression of EGFR (Figure 5C).

The high PDGFRα protein expression reported in all p-CSC of the six cases examined, prompted us to test the pharmacological inhibition of PDGFRα to evaluate its impact on cell growth and survival of CSC. In this study, EGFR signaling inhibition have been shown to be more effective therapeutically than GSI-X, suggesting that the combination with Crenolanib (CR, a selective inhibitor of PDGFRα), would enhance the growth inhibitory effect especially in p-CSC, which were more resistant to treatments. As reported in Figure 6A, B, we observed a significant decrease of cell proliferation in both c-CSC1 and p-CSC1 (38 ± 9.5% and 56 ± 4.4% respectively) with CR (10 μM) alone after 1 day of treatment by MTS assay, but when used in combination with AG1478 (10 μM) the average values declined to 24 ± 9.6% and 34 ± 9.9% respectively. In addition, Western blots analysis clearly showed an induction of Caspase-3 and PARP-1 cleavage fragments, the effect of which was maximized by combining the two inhibitors (Figure 7A). Of note, is the de-repression of PDGFRα in AG1478-treated CSC1 as well as in p-CSC2 (Figure 7B), this is probably due to a compensatory activation mechanism to evade EGFR TKI. In addition, we detected a downmodulation of PDGFRα expression following Crenolanib treatment, which was a common feature in all cases examined (Figure 7A-C). We applied the same treatments to CSC2 and CSC3 (Figure 6C-F). The MTS assay monitored a significant reduction of cell proliferation to 65 ± 2.6% and 47 ± 5% in c-CSC2 and c-CSC3 respectively, while in p-CSC2 and p-CSC3 cell growth declined to 65 ± 4.8% and 51 ± 7.3% respectively with CR alone. Dual treatment was more effective than treatment alone in both CSC2 and CSC3, as reported in Figure 6C-F. Of note, as early as the first day of treatment the two agents exerted a synergistic antiproliferative effect in CSC2 and CSC3 with induction of Caspase-3 and PARP-1 cleavage and reduction of p-Akt1 in CSC3 or Bcl2 in CSC2 (Figure 7B, C).

To confirm the pivotal role of PDGFRα in cell proliferation, we downregulated his expression by short hairpin RNA mediated-knockdown in p-CSC2. After a brief period of antibiotic selection, either control infect cells (pLKO.1) or two cell clones expressing shPDGFRα were plated for cell growth assessment. shPDGFRα-cells showed a significant reduction of cell growth rate compared to pLKO.1 cells, which was consistently maintained up to the seventh day in culture (Figure 7D). In order to evaluate p-CSC response to PDGFAA, c-CSC2 and p-CSC2 were undergone to growth factors (EGF and bFGF) deprivation for 2 days, and then exposed to PDGFAA (40 ng/ml) for short periods of time (Figure 7E). PDGFAA stimulation induced early activation of p-Erk1/2 and p-Akt1 in p-CSC2 within 5 minutes, which declined to basal levels later. Conversely, c-CSC2 displayed a faint response to PDGFAA.

Procedures for collection of adult human GBM CSC were approved by the Ethical Committee of the Catholic University of Rome as reported previously[24]. Informed consent was obtained and all patients were fully aware of the aims and scope of this work. The ethical principles of the declaration of Helsinki were strictly followed.

We have used the same clinical materials reported in our previous papers[24, 25]. In brief, the CSC cells were retrieved from adult patients affected by GBM and undergoing craniotomy at the Institute of Neurosurgery, Catholic University-School of Medicine of Rome, Italy. Dissociated cells were cultured in the presence of human recombinant EGF (20 ng/ml; PeproTech, Rocky Hill, NJ), human recombinant bFGF (10 ng/ml; PeproTech), in DMEM/F12 (1:1) serum-free medium (Invitrogen, Carlsband, CA) containing L glutamine 2 mM, glucose 0.6%, putrescine 9.6 ug/ml, progesterone 0.025 mg/ml, sodium selenite 5.2 ng/ml, insulin 0.025 mg/ml, apo-transferrin sodium salt 0.1 mg/ml, sodium bicarbonate 3 mM, Hepes 5 mM, BSA 4 mg/ml, heparin 4 ug/ml (all purchased by Sigma-Aldrich). Floating neurospheres were dissociated with Accutase at 37°C (Merck-Millipore). In some cases, neurospheres were passaged up to passage P60 and the experiments were performed between P14 and P60. Cell starvation was planned for 2 days in Stem Medium w/o EGF and bFGF. Subsequently, PDGF-AA was added (40 ng/ml, Peprotech) for different time points (5′, 10′, 30′, 120′ and 24 hours). Cells treatments were performed with GSI-X (also named L-685,458, Calbiochem), AG1478 (Calbiochem) and Crenolanib (CP-8685596, Selleckchem).

The experiments on RNA interference were performed using Mission Lentivirus-based shRNA for PDGFRα (NM_006206-Sigma-Adrich). We amplified three DNA clones (clones numbers: TRCN0000195132, TRCN0000196272, TRCN0000196928) and pLKO.1-puro, as a control for infection. We selected puromycin resistant p-CSC2 expressing three different shRNA sequences, but only two were able to donwmodulate PDGFRα (TRCN0000195132 and TRCN0000196928). Human embryonic kidney (HEK)-293 T cells in log-phase growth were transiently transfected using standard LipofectAmine reagent (Invitrogen), with lentivirus-based shRNA (and with pLKO.1 as control) plus helper plasmids (Invitrogen Packging Plasmids, Carlsbard, CA). Media containing the virions were harvested two days after cell transfection and transferred directly onto p-CSC2. Lentiviral infection was performed in the presence of polybrene solution at 8 μg/ml (Sigma-Aldrich) and the antibiotic puromycin (Euroclone) was added to the cells at 1 μg/ml for a week to select CSC expressing shRNA sequences.

GBM CSC seeded as single cells (1 × 10⁶/dish) were left for 1 day in proliferation medium before treating them for 1-2 days with the inhibitors, either singly or in combination. Afterwards, cells were collected and washed with PBS plus proteases inhibitors before protein extraction in 100-200 μl of lysis buffer (1% NP-40, 0.01% SDS, 20 mM Tris–HCl pH 7.4, 300 mM NaCl, 1 mM EDTA, 1 mM Na₃VO₄ and protease inhibitors cocktail - from Sigma–Aldrich). Then, cells were sonicated with two pulses of 5 sec at 50% of amplitude (Sonics and Materials, Newtown, CT). Equal amounts (30 μg/lane) of total protein extracts, determined by Bio-Rad protein Assay (Bio-Rad, Munchen, Germany), were loaded on NuPAGEBis-Tris gels (Invitrogen), and transferred on Hybond-P Extra membrane (Amersham Biosciences, GE Healthcare Life Science-Buckinghamshire, UK). Filters were immunoblotted using the following primary antibodies: rabbit anti-NICD1, rabbit anti-HES1, rabbit anti-DLL1, rabbit anti-EGFR, rabbit anti-pY1068-EGFR, rabbit anti-PDGFRα, rabbit anti-PDGFRβ, rabbit anti-T202/Y204-ERK1/2 and anti-ERK1/2, rabbit anti-Y705-STAT3 and rabbit anti-STAT3 and anti-Caspase3, (all purchased from Cell Signaling, MA-USA), mouse anti-S473-AKT1 and rabbit anti-AKT1 (Calbiochem), rabbit anti-BCL2 (Millipore), rabbit anti-survivin and anti-MMP9 (Abcam), rabbit anti-Cyclin D1 and anti-p27 (Santa Cruz-USA), mouse anti-β-actin and anti-GAPDH (SIGMA). After three washing with TBS-Tween buffer, immuno-reactive proteins were detected using rabbit anti-mouse, donkey-anti-rabbit and donkey anti-goat horseradish peroxidase-conjugated secondary antibodies directed to the appropriate primary antibodies (Jackson Immunoresearch Laboratories, West Grove, PA). The proteins were then visualized using the chemiluminescence system (Millipore). Gels and Images acquisition was done by HP Photosmart Essential Ver. 1.12 and Adobe Photoshop CS5 respectively. The densitometric analysis of protein bands normalized against to β-actin protein levels were performed from three independent experiments using the ImageJ software (NIH, USA).

For cell proliferation assay, neurospheres were dissociated into single cells and 1 × 10⁴ cells/well were plated in triplicate on 60 mm plates. Cells were harvested and counted at different time points (3, 5, 7 days in vitro) in growth medium, considering the starting time as the day after plating. For pharmacological studies we used the CellTiter 96 Aqueous One Solution Reagent (Promega), a cell proliferation colorimetric assay containing a novel tetrazolium compound MTS. Briefly, 2 × 10⁴ cells/well were plated in triplicate for each group on 12-well plates and the drugs or DMSO-vehicle were added the day after plating. Before harvesting cells, they were incubated with 100 μl/ml of MTS at 37°C for approximately 1 hour. The metabolically active cells reduced MTS into a soluble formazan product, whose absorbance was measured at 490 nm in a plate reader (Bio-Rad). These experiments were performed three times and each time in triplicate.The absorbance values of the collected samples were subtracted from the background absorbance of medium-only control and expressed as % of control and calculated as mean average ± SD (n = 3).

For an invasion assay, 1 × 10⁵ cells were resuspended in 0.3 ml of Stem Medium with inhibitors or DMSO-vehicle and placed in triplicate into the top chamber of matrigel-coated transwell insert (Millipore). The bottom wells contained 0.4 ml of Stem Medium with EGF and bFGF, or without these media as controls for the experiment. After 72 h, cells on the top surface of the filter were removed with a cottonswab. Thereafter, the filters were fixed and stained with crystal violet and subsequently washed to collect the staining solution. OD values, proportional to the number of cells, were measured on a plate reader with a 550 nm filter (Bio-Rad). These experiments were performed three times and each time in triplicate. The absorbance values were calculated as mean ± SD (n = 3). These experiments were repeated three times and each time in triplicate.

Cells were fixed with 0.5 ml of cold Methanol/Acetone Solution (1:5) and left at 4°C for at least 1 hour, then centrifuged at 950 RPM for 5 minutes. The cell pellet was resuspended in 100 μl of PBS, 375 ul of Rnase A (100 μg/ml) and 25 μl of Propidium Iodide (1 mg/ml). Cells were incubated at room temperature for 15 minutes and left in the dark at 4°C until FACS analysis. Samples were acquired on FacsCalibur (BD Becton Dickinson) at 488 nm. The acquired FACS data were analyzed by ModFit LT software (Verity Software House, Inc.) to determine the percentage of cells sub G1, G1, S and G2/M phases. These experiments were performed twice for each sample.

Statistical analysis was performed with Prism5 (GraphPad) and Microsoft Office Excel 2007. All data shown are representative of results obtained from experiments conducted two or three times as specified in the specific sections. The results were analyzed by either One-way Anova and Newman-Keuls post-tests or Two-way ANOVA and Bonferroni’s post tests. Data are expressed as mean ± standard deviation (SD) and P values ≤0.05 (*), ≤0.01 (**), ≤0.001 (***) were considered statistically significant. To evaluate the synergistic effect of the combination of two drugs on cell growth we referred to Lama et al.[50].