Dedifferentiated cells obtained from glioblastoma cell lines are an easy and robust model for mesenchymal glioblastoma stem cells studies

Abstract

Glioblastoma is an aggressive brain tumor with a poor prognosis. Glioblastoma Stem Cells (GSC) are involved in glioblastoma resistance and relapse. Effective glioblastoma treatment must include GSC targeting strategy. Robust and well defined in vitroGSC models are required for new therapies evaluation. In this study, we extensively characterized 4 GSC models obtained by dedifferentiation of commercially available glioblastoma cell lines and compared them to 2 established patient derived GSC lines (Brain Tumor Initiating Cells). Dedifferentiated cells formed gliospheres, typical for GSC, with self-renewal ability. Gene expression and protein analysis revealed an increased expression of several stemness associated markers such as A2B5, integrin α6, Nestin, SOX2 and NANOG. Cells were oriented toward a mesenchymal GSC phenotype as shown by elevated levels of mesenchymal and EMT related markers (CD44, FN1, integrin α5). Dedifferentiated GSC were similar to BTIC in terms of size and heterogeneity. The characterization study also revealed that CXCR4 pathway was activated by dedifferentiation, emphasizing its role as a potential therapeutic target. The expression of resistance-associated markers and the phenotypic diversity of the 4 GSC models obtained by dedifferentiation make them relevant to challenge future GSC targeting therapies.

Introduction

Glioblastoma is the most frequent and aggressive primary brain tumor. Low treatment efficiency and constant relapses are responsible for a poor prognosis, with a median overall survival of 15 months [1]. Microenvironment interactions are involved in disease progression and therapeutic resistance through several mechanisms including glioblastoma stem cells (GSC) development and maintenance [2]. The presence of GSC within the tumor is associated with relapse and resistance to treatment [3]. Indeed, GSC are able to self-renew and can differentiate into glioblastoma cell populations depending on their phenotype [4]. GSC can be mesenchymal or proneural and each phenotype being associated to different levels of aggressivity and therapeutic sensitivity [5]. Mesenchymal GSC are more susceptible to be involved in relapse due to their invasive and resistant profile. In response to radiation and cytotoxic treatments, proneural GSC can shift toward a mesenchymal phenotype through epithelial-mesenchymal transition (EMT) [6-8]. To identify new therapeutic targets, it is crucial to better understand cell and molecular mechanisms involved in GSC development and therapeutic escape.

To study GSC, two main in vitro models can be used: patient derived cells or cells dedifferentiated from glioblatoma cell lines. GSC from both origins typically grow as gliospheres, 3D cell clusters in suspension. The ability to form gliospheres in vitro is a common feature to validate the stemness of the obtained models. Patient derived GSC are usually purified from tumor sample, based on the expression of one or two enrichment markers such as CD133 and CD15 [9,10]. However, CD133 significance as a single GSC marker is controversial. Indeed, proneural GSC are known to express CD133 but mesenchymal GSC are CD133 negative, while they are more resistant and aggressive [11]. Other studies used CD44 and CD133 to isolate mesenchymal and proneural cells from the same sample [12] but some GSC can have a complex phenotype, expressing both markers [11]. Considering GSC heterogeneity, markers-based isolation induces a loss of diversity compared to the initial tumor. Another protocol is based on cultivating tumor samples in a GSC promoting medium [13]. Working with patient derived cells is advantageous because cells are closer to their in vivo characteristics, but requires accessibility to patient samples. To avoid in vitro drift, GSC can be kept as xenograft, being harvested only before experiment [10]. Such models are complex and involve ethical considerations and important financial costs, preventing large scale studies and availability for multiple laboratories. Dedifferentiated gliospheres are easy to produce from commercially available cell lines and have been used as GSC models in several studies [14-16]. An advantage of this model is the possibility to compare GSC and their differentiated counterparts to reveal stemness specific characteristics and pathways involved in dedifferentiation for target identification.

GSC models are usually validated and characterized using some of the most known markers (CD133, SOX2, CD15 (SSEA-1), A2B5, CD44) and cellular characteristics (sphere-forming ability, self-renewal) [12,14,17-19]. GSC enrichment based on markers expression is limited since the cells can express a multitude of markers involved at varying degrees in stemness, growth, invasion and resistance. Before performing therapy evaluation, GSC should be extensively characterized in a more extensive manner to ensure stemness and phenotype.

The objective of this study was to establish robust, accessible and well-characterized GSC models that can be used to identify therapeutic targets. We performed an extensive analysis of 4 GSC models obtained by dedifferentiation of commercially available GBM cell lines and compared them to 2 patient-derived GSC lines. We demonstrate that dedifferentiated GSC show similarities to patient derived stem cells and have a heterogeneous, complex profile, susceptible to challenge therapy evaluation. Based on these models characterization, we provide support for therapeutic targets among CXCR4 pathway.

Material and methods

Cell culture

Glioblastoma cell lines U-87MG, U-118MG, U-251MG and T98-G were provided by ATCC. Adherent cells were maintained in DMEM high glucose (Sigma-Aldrich^®) supplemented with 10% FBS (Biowest) and 1% penicillin/streptomycin solution (Gibco^®) at 37°C, 5% CO₂. Cells were passaged using trypsin 0.25% (Sigma-Aldrich^®) every week and kept in culture for 6 consecutive weeks maximum. Two GSC lines, BTIC 25m and 12m, were obtained from Dr H. Artee Luchman and from Pr. Samuel Weiss (Hotchkiss Brain Institute). BTIC were cultivated using NeuroCult™ NS-A Proliferation Kit Human (Stemcell^®). Gliospheres were dissociated using accutase (Sigma-Aldrich^®) and diluted every week, for less than 6 consecutive weeks.

Neuroblastoma cell lines (SH-SY5Y and NGP) as well as cervix adenocarcinoma cell line (HeLa) were purchased from ATCC. Normal human astrocytes (NHA) were acquired from Lonza Bioscience^®. Cells were cultivated in similar conditions as glioblastoma cell lines.

Dedifferentiation protocol

Adherent cells from U-87MG, U-118MG, U-251MG and T98-G cell lines were harvested at 70-80% confluence, washed with DPBS (Gibco^®) and resuspended in dedifferentiation medium and incubated at 37°C, 5% CO₂. Dedifferentiation medium, adapted from [20], was composed of DMEM/F-12 (Gibco^®) supplemented with 1% N-2 (Gibco^®), 1% B-27 (Gibco^®), 20 ng/mL hEGF (human Epidermal Growth Factor, Miltenyi^®), 20 ng/mL bFGF-2 (basic Fibroblast Growth Factor, Miltenyi^®) and 1% penicillin/streptomycin. After 24 h, cells were mechanically resuspended and fresh medium was added. Gliospheres were harvested when they reached their maximum size before impact on cell viability (to avoid necrotic core): at 48 h for U-251MG and T98-G cell lines, at 96 h for U-118MG and U-87MG cell lines. A medium addition was performed at 72 h for 96 h incubation time.

Morphological characterization

Gliospheres pictures were recorded using an inverted microscope (Zeiss^®, Primovert) when they reached their maximum growth before viability decreases. Gliosphere areas were measured with ImageJ software thanks to the help of J. Chaigneau (image analysis engineer). Briefly, images were smoothed, a threshold was set to remove background before analysis by particle detection ImageJ plugin. Particle area was converted from pixel² to µm² using picture scale. Such analysis was performed on pictures from 3 independent experiments, giving areas measurement for at least 100 gliospheres.

Self-renewal assessment

Gliospheres were dissociated using accutase and decreasing cell quantity (400 to 1) was seeded in 96-well plates containing 200 µL dedifferentiation medium. Each condition was replicated in 10 wells. To quantify gliosphere forming yield, gliosphere number in 5 wells containing 100 cells each was recorded. Data were gathered and analyzed from 3 independent experiments.

RT-qPCR

The expression of 31 GSC-associated genes was measured by RT-qPCR to identify key markers for further studies at the protein level (see Table S1 for markers information). Adherent cells and gliospheres were harvested and washed with cold DPBS before storage as dry pellet at -80°C. RNA extraction and RT-qPCR were carried out by J. Cayon and L. Bonneau PACeM platform (Plateforme d’Analyse Cellulaire et Moléculaire, SFR ICAT 4208, Angers, France). Primer sequences are property of PACeM platform and are available on request. Total mRNAs were extracted with RNeasy Micro-kit (Qiagen) according to the manufacturer’s instructions. Quantification was performed using NanoDrop2000 (ThermoFisher Scientific^®). cDNAs were generated using the Reverse Transcriptase SuperScript II^® kit (Invitrogen^®) and purified using QIAquick^® PCR purification kit (Qiagen^®) according to the manufacturer’s instructions. Specific primers were designed on Primer Blast website. qPCR was performed using LightCycler 480 II (Roche) and SYBR Green as fluorescent probe. Cq were converted to relative mRNA expression according to Vandesompele method [21], with two reference genes: GAPDH and HPRT1 and the most expressed gene as internal calibrator. To easily compare genes expression between adherent and dedifferentiated cells, fold expression was calculated following Livak method [22] with mean reference genes Ct and adherent cells as control conditions for each cell line. Fold expression <0.5 indicates a decreased expression in gliospheres while a fold expression >2 shows overexpression. Heatmaps representing fold expression and relative expression were created with the R package ComplexHeatmap version 2.12.1 [23,24].

Flow cytometry

For stem cell phenotype evaluation, adherent cells and gliospheres were stained using a panel of antibodies against 10 known GSC markers (CD44, CD133, A2B5, CD90, integrin α-5, integrin α-6, CXCR4, CD74, CD15 and CD36) for flow cytometry analysis (see Table S1 for markers information). Antibodies and corresponding isotype controls were purchased from Miltenyi^® (see Table S2 for complete panel references). Adherent cells and gliospheres were dissociated using accutase, washed with corresponding complete medium and 250 K cells were resuspended in 50 µL DPBS containing 0.5% BSA and 2 mM EDTA (FC buffer). Cells were stained 15 min at 4°C with antibodies at 1:50 dilution, except for anti-CD74 antibody, at 1:11 dilution. Cells were washed twice with FC buffer. Ten thousand events were recorded using CytoFLEX^® flow cytometer equipped with CytExpert^® software (material from PACeM platform). Compensation matrix was automatically established with single-stained samples, manually corrected, and was then applied on registered data. Cytometry measurements were processed with FlowJo^® V10 software. Positive population gating used isotype controls as references, giving the percentage of positive cells in each sample. Mean fluorescence intensity (MFI) was calculated for each sample and was normalized on corresponding isotype control MFI. The MFI ratio obtained enables comparison between samples.

Immunocytochemistry

To complete GSC marker analysis, the expression of 3 proteins was evaluated in gliospheres and adherent cells by immunocytochemistry (see Table S1 for markers information). Olig2 transcription factor and cytoskeleton components (Nestin and GFAP) were analyzed. Adherent cells were seeded on glass coverslips during 48 h at 37°C, 5% CO₂. Gliospheres were seeded on CellTak^® (Corning^®) coated coverslip (3.5 µg/cm²) during 45 min at 37°C, 5% CO₂. Cells were washed three times with DPBS 1% BSA between each protocol step. Fixation was performed in 4% paraformaldehyde (Euromedex^®) during 15 min for gliospheres and 10 min for adherent cells. Samples were incubated 1 h in DPBS 1% BSA containing 0.1% Triton X-100 before overnight incubation at 4°C in primary antibody solutions (see Table S3 for antibodies information). Triton X-100 concentration was increased to 0.3% for intranuclear staining (Olig2). Primary antibodies were revealed using AF488 or AF568 anti-rabbit IgG antibodies (ThermoFisher Scientific^®) at 1:250 for 2 h. Finally, nuclei were stained with 3 µM DAPI (Sigma-Aldrich^®) for 15 min. Coverslips were mounted on SuperFrost^® slides (ThermoFisher Scientific^®) using ProLong Gold antifade solution (ThermoFisher Scientific^®). Slides were observed with a confocal microscope (Leica TCS SP8 from Leica Biosystems^®) with the help of R. Perrot from SCIAM platform (Service Commun d’Imageries et d’Analyses Microscopiques, SFR ICAT 4208, Angers, France).

Statistical analysis

Statistical analyses were performed with GraphPad^® Prism^® 8.3 software (GraphPad^® Software, LLC). All experiments, except immunocytochemistry, were repeated at least 3 independent times. Means comparisons were performed with Mann-Whitney test, differences were considered significant at P<0.05.

Results

Dedifferentiated gliospheres are morphologically similar to patient derived GSC

We first dedifferentiated 4 GBM cell lines (U-87MG, U-251MG, U-118MG and T-98G) using a standardized and accessible protocol to obtain GSC and compared their morphology to 2 patient-derived GSC lines (BTIC 12m and 25m) as standards. As for cancer and neural stem cells, glioblastoma stem cells cultivated in vitro are known to form 3D cell clusters in suspension named “gliospheres” (GS). As illustrated by Figure 1 (top), when cultivated in dedifferentiation medium, adherent GBM cell lines rapidly grown as gliospheres, closely alike BTIC. This morphological change was specific to glioblastoma cells since it was not observed in normal astrocytes (NHA) nor neuroblastoma (SH-SY5Y and NGP) or cervix adenocarcinoma (HeLa) cell lines in the same culture conditions (Figure S1). Their morphology varied across cell lines, some gliospheres being partly adherent and smaller (T-98G, U-118MG, BTIC 12m) with others being completely floating (U-87MG, U-251MG, BTIC 25m) and sometimes forming gliosphere aggregates (U-251MG). To perform quantified comparison between cell lines, gliosphere areas were measured from pictures taken with an inverted microscope. Areas displayed in Figure 1 illustrate size heterogeneity within the same cell line, which corroborates picture observations. Despite variations between cell lines distribution, areas were similar. Particularly, U-251MG, U-87MG and BTIC 25m had similar medians while U-118MG were closer to BTIC 12m, as previously noted on pictures. However, important differences in behavior could be noted for T98-G and U-251MG. T98-G gliospheres were smaller, had a shorter distribution and their size did not increase after 48 h. U-251MG gliospheres grew rapidly, reaching large sizes in only 48 h of dedifferentiation. Moreover, the cells tended to stick to each other, forming gliosphere aggregates. All these results suggest that gliosphere morphology is cell line dependent, emphasizing the need for multiple models to represent GSC heterogeneity.

Figure 1.

Dedifferentiated gliospheres are similar to BTIC for appearance, size and heterogeneity. Microscope observations before/after dedifferentiation of T98-G (A, a), U-251MG (B, b), U-118MG (C, c), U-87MG (D, d), BTIC 25m (e) and 12m (f). Capital letters are used for adherent controls (AC) and lowercase letter for gliospheres (GS). Pictures were taken when gliospheres reached their maximum viable growth size. Scale is 10 µm. Corresponding gliospheres sizes are shown below. Areas were measured using ImageJ software for 100 gliospheres obtained from 3 independent experiments. Medians are represented in red.

Dedifferentiated gliospheres have GSC-like self-renewal abilities

GSC are characterized by their ability to self-renew after dissociation. GSC should maintain themselves in a GSC promoting medium containing growth factors. As gliospheres are composed of GSC and more differentiated cells, dissociation enables to separate cell populations and observe cell types. All our dedifferentiated cell lines were able to form gliospheres after dissociation, at different yields (Figure 2A). U-87MG and U-118MG cells generated gliospheres with similar size when compared to original gliospheres, showing high growth capacity (Figure 2B). However, T98-G and U-251MG gliospheres were much smaller (Figure 2B), with a low growth rate. Such results may suggest a quiescent GSC profile. Particularly, in T-98G and U-251MG cell lines, adherent cells were observed, indicating heterogeneity in the original gliospheres. Overall, dedifferentiated U-118MG and U-87MG seem closer to BTIC 25m in terms of yield and growth while T-98G and U-251MG resemble BTIC 12m.

Figure 2.

Dedifferentiated cells self-renew and reform gliospheres after dissociation. A. Yield in gliospheres calculated from their number in wells seeded with 100 cells. Mean represented with standard deviation from 15 measurements across 3 independent experiments. B. Examples of spheres formed in wells with 100 cells. Scale is 20 µm.

Dedifferentiation conditions induce a gene expression shift toward stem-cell, EMT and CXCR4 pathways

To characterize our models, we compared 31 genes expression levels between dedifferentiated gliospheres (GS), corresponding adherent counterparts (AC) and BTIC. Genes were chosen according to a literature review on GSC and glioblastoma prognostic markers (See Table S1 for complete markers information). Fifteen genes are associated with stem-cell maintenance and/or commonly used as GSC markers. Among them, 4 are known to be specific for the proneural (PN) phenotype (such as PROM1 or CD133) and 3 for the mesenchymal (MES) phenotype (such as CD44). 8 genes associated with Epithelial-Mesenchymal Transition (EMT) were also included in the study as they can be expressed in GSC. These genes can be associated with therapeutic resistance through PN to MES phenotype conversion (such as SLUG), sometimes even being GBM prognostic markers (such as CDH2). Associated with EMT, CXCR4-related pathway has been identified as a key actor of GSC maintenance and therapeutic resistance [25]. The 4 receptors CXCR4, CXCR2, CXCR7 and CD74 and their 2 ligands MIF and CXCL12 (also known as SDF-1) were included in our study. Finally, VEGFA and VEGFC were studied for their implication in angiogenesis and EMT.

The Figure 3 shows the global mRNA expression of the analyzed genes among all samples according to Vandesompele analysis method (see Figure S2 for complete heatmap including adherent cells). Such method allows comparison between genes and between cell lines. Results are reported on a heatmap with blue to red colors indicating low to high expression level. Different mRNA expression patterns could be seen between GSC models, even between BTIC 25m and 12m. A common tendency was nevertheless noticed for some well-expressed genes (FN1, EZH2, CD44, BMI1, VIM, CDH2) as well as for low/not expressed ones (POU5F1 (OCT3/4), NANOG, FUT4 (SSEA-1), CD36, PROM1 (CD133), CDH1, CXCR2). Some genes appeared major, reaching high expression levels, but only for some GSC models (SPP1, NES, SOX2, SLUG, S100A4, ZEB1, MIF). All the 4 generated models expressed a wide variety of GSC associated genes and related pathways, underlining their GSC profile. According to their global expression profiles, the models seemed to be oriented toward a mesenchymal GSC phenotype: low expression of PN markers (PROM1 (CD133), CD36) while MES-specific and EMT-related genes were generally highly expressed (CD44, BMI1, VIM, CDH2).

Figure 3.

Transcriptional expression supports a mesenchymal GSC profile in gliospheres associated with high expression of some EMT, angiogenesis and CXCR associated genes. RT-qPCR analysis of gene expression in U-251MG, U-118MG, T98-G and U-87MG dedifferentiated gliospheres and in BTIC 25m and 12m. Relative mRNA expression was calculated and normalized according to Vandesompele method with two control genes (GAPDH and HPRT1) and an internal calibrator. Expression levels are colored from blue to red with increasing values. Genes are sorted by mean expression level in each group. PN: Proneural, MES: Mesenchymal, EMT: Epithelial-Mesenchymal Transition.

To estimate the transcriptional impact of the dedifferentiation, fold expression levels in gliospheres vs adherent cells were analyzed (Figure 4, see Figure S2 for global mRNA expression). Despite some heterogeneity between cell lines, GSC-associated gene expression was clearly increased after the dedifferentiation process. Particularly, we observed an enhancement of SOX2, FUT4 (SSEA-1), POU5F1 (OCT3/4) and NANOG, transcriptional factors essential for stem cell phenotype acquisition or maintenance. Mesenchymal orientation was not only kept but emphasized by dedifferentiation as shown by the increase in CD44 and EMT-related gene expression. However, even if CD133 remained weakly expressed, its expression was conserved or increased in 3 dedifferentiated models, further supporting a GSC phenotype acquisition. Interestingly, several genes of the CXCR4-associated pathways were more expressed after dedifferentiation, especially in U-87MG and T-98G, bringing new evidence for the importance of this pathway in GSC phenotype. VEGF pathway, moderately expressed and stimulated by dedifferentiation conditions in some cell lines could be also identified as a marker for EMT and aggressiveness of the obtained GSC.

Figure 4.

Dedifferentiated GSC overexpress stem-cell transcriptional markers, EMT-related genes and CXCR4 pathway. RT-qPCR analysis of gene expression in U-251MG, U-118MG, T98-G and U-87MG was used to calculate fold expression in dedifferentiated gliospheres with adherent cells as reference. A fold expression >2 (red) means an overexpression in GS whereas a fold expression <0.5 (green) means a decreased expression in GS. Grey color stands for conditions with undetectable expression in gliospheres and in adherent cells. N: Proneural, MES: Mesenchymal, EMT: Epithelial-Mesenchymal Transition.

All these data support the acquisition of a GSC phenotype after dedifferentiation, with a gene expression profile comparable to the one of patient-derived stem cells such as BTICs. The obtained cells appear to belong to a more resistant phenotype, mainly mesenchymal, as shown by the high expression of numerous EMT-related genes as well as CXCR4 and VEGF pathways.

Flow cytometry highlights important GSC markers in dedifferentiated gliospheres

To confirm GSC markers at the protein level, the expression of 10 surface markers (A2B5, Integrin α5 and 6, CD133, CD44, CD90, CD36, CD15 (SSEA-1), CXCR4 and CD74) was analyzed by flow cytometry in BTIC, dedifferentiated gliospheres and corresponding adherent cells (See Table S1 for complete markers information).

A2B5, Integrin α6, CD15, CD90 and CD36 are GSC markers, the first three being used for GSC enrichment. Interestingly, A2B5 expression was negative in adherent cells and was detectable only after dedifferentiation in U-118MG, U-251MG and U-87MG cells (Figure 5-1A, 5-1B). Emphasizing once more GSC variability, this marker was not expressed in either T98-G gliospheres and in BTIC 25m while expressed in BTIC 12m. Similarly, integrin α-6 (or CD49f, coded by ITGA6), was highly expressed in all studied models (except in U-87MG). Its signal intensity was increased by dedifferentiation in U-251MG and T98-G (Figure 5-1C). These results support stemness characteristics acquisition induced by dedifferentiation. CD90 was not ubiquitous and did not seem impacted by dedifferentiation. Indeed, CD90 was highly expressed (100% positive cells, high MFI) in U-118MG, U-251MG and T-98G, with little change in GS (Figure S3). In BTIC, CD90 was only expressed in 57% of 12m cells. Our data is consistent with the literature as CD90 is not necessary for stem cell maintenance [26-28]. Finally, CD36 and CD15 were not revealed in any of our sample (data not shown), despite previous studies on U-251MG and U-87MG gliospheres showing CD36 expression in both cell lines and low CD15 expression in U-251MG [18].

Figure 5.

Change in membrane markers expression supports acquisition of GSC with a mainly mesenchymal phenotype in dedifferentiated gliospheres models. Flow cytometry analysis of GSC markers (A2B5 and integrin α-6), mesenchymal/proneural markers and CXCR4 in U-87MG, U-118MG, U-251MG, T98-G and BTIC 25m and 12m. Adherent cells (AC) were used as control for dedifferentiated gliospheres (GS). A2B5 staining results are shown as typical flow cytometry histograms (1A) and corresponding positive population percentage (1B). Integrin α6 analysis is depicted as mean fluorescence ratio with corresponding positive cells percentage in the table below (1C). Percentage of positive cells are represented for CD133 (2A) and CXCR4 (3A). Mean fluorescence ratios are shown for CD44 (2B) and CXCR4 (3B). Mean fluorescence ratio is calculated as mean fluorescence intensity (MFI) of stained sample on corresponding isotype control MFI. Statistical significance was calculated using Mann-Whitney test on 4 independent experiments, except for U-87MG (n=6) and U-118MG (n=5). *p-value <0.05, **p-value <0.001.

CD44 and CD133 markers analysis was used to confirm the GSC phenotype. In line with RT-qPCR data, CD44 was expressed in all studied cells and its expression was significantly increased after dedifferentiation in U-118MG, U-251MG and U-87MG (Figure 5-2B). However, despite its low transcriptional expression, CD133 was detectable in all studied cell lines except BTIC 12m (Figure 5-2A). Low CD133 expression in BTIC is consistent with previous results [29]. Dedifferentiation induced a decrease in CD133 population in U-251MG, T-98G and U-87MG suggesting a proneural phenotype loss. Moreover, integrin α5, an EMT marker, was highly expressed in all cell lines, including BTIC at a lower level and slightly increased with dedifferentiation (Figure S3). Altogether, these results suggest that dedifferentiation directs cells toward a more aggressive mesenchymal profile, supporting CD44 and EMT-associated marker expression instead of CD133.

As the CXCR4 pathway seemed activated in our RT-qPCR data, we analyzed the expression of CXCR4 and CD74 receptors by flow cytometry. However, despite high transcriptional expression of its main ligand MIF, CD74 was expressed only in 30% of BTIC 25m and U-87MG gliospheres, with a slight increase compared to adherent cells (22%) (Figure S3). CXCR4 protein expression was higher in BTIC and was increased after dedifferentiation in the 4 cell lines, supporting both the stemness of our models and the significance of CXCR4 pathway in GSC (Figure 5-3A and 5-3B). Interestingly, in U-118MG gliospheres, the CXCR4 positive population was higher in A2B5 positive cells, showing an association between these markers (25% vs 52%) (Figure S4).

Immunocytochemistry reveals the expression of Nestin and GFAP in GSC

To confirm the dedifferentiation state of gliospheres, immunocytochemistry was used to characterize the expression of cytoskeleton components being GSC (Nestin) or glial differentiation (GFAP) markers. GFAP was present in U-251MG, U-118MG and U-87MG and BTIC (Figure 6). GFAP expression could be surprising, as it was previously considered as a mature astrocytic marker and thus, a BTIC differentiation marker [30]. However, it can also be expressed in neural stem cells and in glial cells at various differentiation state [31]. GFAP expression, albeit controversial, supports the glial profile of the 6 GSC models. The GSC marker Nestin was expressed in all studied cell lines, in a particularly strong manner in gliospheres (Figure 6). Its expression appeared increased by the dedifferentiation process, emphasizing Nestin importance in gliosphere structure while supporting acquisition of GSC characteristics.

Figure 6.

Expression of stem cells and glial marker proteins in adherent cells and in gliospheres. Typical fluorescent confocal photographs of cells immunostained to reveal Olig2 (AF488, green), Nestin (AF568, red), GFAP (AF568, red) and nuclei (DAPI, blue). Scale is 20 µm. Pictures represents BTIC 25m (A-C) and 12m (D-F), U-251MG (G-I), U-118MG (J-L), T98-G (M-O) and U-87MG (P-R), Capital letters stands for adherent cells (AC) and lowercase letters for gliospheres (GS).

Finally, the presence of Olig2 transcription factor was screened. Olig2 signal was very weak except in BTIC 12m (Figure 6), consistently with the RT-qPCR results.

Discussion

Pharmacological research on glioblastoma requires glioblastoma stem cell models for the identification and evaluation of targeted treatments. Each model has its own limitations, such as accessibility, lack of heterogeneity, low reproducibility, drifting due to long term culture or differences from in vivocharacteristics. Among available models, dedifferentiation of glioblastoma cells is frequently used as an easy way to obtain GSC, but these models are often poorly characterized. In this study, we characterize GSC obtained from 4 commercially available GBM cell lines with the same dedifferentiation protocol and compared to 2 patient-derived GSC lines as reference. We gathered existing data on GSC markers to perform an extensive morphological, transcriptional and protein expression characterization. This analysis validated our models while defining their phenotypes and providing support for potential common therapeutic target.

Dedifferentiation protocol led to rapid gliosphere formation in the 4 studied cell lines. Morphological study confirmed that dedifferentiated gliospheres looked like BTIC, with similar size range. Gliosphere areas were highly variable within a given sample, cells being able to form large aggregates, particularly in U-251MG cell line. We hypothesize that the formation of large gliospheres could be associated with higher aggressiveness. Indeed, in the core of huge gliospheres, cells would lack nutrients and oxygen. Hypoxia has been demonstrated to reinforce cancer cell aggressiveness leading to chemotherapy and radiotherapy resistance [32]. Gliosphere size may also be linked to proliferation capacity. More proliferative cells would form larger gliospheres and would be more susceptible to produce aggressive tumors. Given these hypotheses, we believe that morphometric characteristics should be considered for new therapy evaluation.

A first validation of dedifferentiated GS stemness was their ability to self-renew after dissociation. Separating GS constituting cells also reveals their heterogeneity, with GSC-like cells coexisting with more differentiated cells that became adherent when separated from the GS structure. The 4 dedifferentiated GSC models, as well as BTIC, displayed differences in terms of proliferation characteristics. As anticipated during morphological studies, “second generation” T98-G GS were rare and composed of few cells, with a low increase in size over time. A similar profile could be observed for U-251MG, which was contradictory with the large gliospheres formed directly after dedifferentiation. We hypothesize that such spheroids are rather caused by individual GS aggregation than by high proliferation induced by dedifferentiation. Both cell lines may have a quiescent profile, which may be interesting for further pharmacological studies as quiescent cells can resist to conventional therapies [33]. On the contrary, U-118MG and U-87MG reform gliospheres of similar sizes and shapes after dissociation, even with low initial cell quantities, which supports their rapid proliferation characteristics. Such models would be more aggressive and may escape treatment by rapid resistance acquisition or by population renewal. Such heterogeneity in our models is interesting to study how to overcome resistance in GSC and to look for new therapeutic targets that could restore conventional therapy sensitivity.

By screening of 31 genes involved in GSC identification, maintenance or in glioblastoma therapeutic resistance, we confirmed similar characteristics between dedifferentiated cells and GSC and suggested relevant markers for protein studies. Indeed, each model expresses several genes associated with stemness and dedifferentiation increased expression of important stem cell genes, such as SOX2, NANOG, POU5F1 (OCT3/4), ITGA6 and SSEA1. At the protein level, the increase in Nestin, a GSC marker, supports cell stemness in dedifferentiated gliospheres. Flow cytometry pointed out the presence of two other important GSC markers: A2B5 and integrin α-6 (ITGA6). Strikingly, expression of these proteins were enhanced by dedifferentiation which supports GSC characteristics acquisition with this protocol. A2B5 has been associated with GSC survival, self-renewal and proliferation [34] as well as integrin α6 [35], which is also linked to radiotherapy resistance in mesenchymal GSC [36]. Altogether, these markers’ enhancement suggests a resistant GSC profile that remains to be determined.

Mesenchymal GSC are the most challenging therapeutic target as they survive conventional therapies [8,37]. The studied cell lines presented a mesenchymal oriented profile, in line with the literature on BTIC 12m and 25m [29]. Such characteristics were emphasized by dedifferentiation that increased markers associated with EMT or with mesenchymal GSC, such as CD44, integrin α-5 (ITGA5) or FN1. CD44 is particularly important in our models as it was highly expressed and increased after dedifferentiation, both at the transcriptional and protein levels. CD44 is a mesenchymal GSC marker, a receptor for extracellular matrix components, known to be involved in invasiveness and therapeutic resistance [38]. CD44 is a promising target, its inhibition in GSC leading to reduced migration and invasion with an increased survival in mice [39]. It was also proved that CD44 promotes GSC and aggressive growth when associated with its ligand, osteopontin. Our RT-qPCR data showed a moderate to high expression of SPP1 gene (coding for osteopontin) in 4/6 GSC models. Particularly, both CD44 and SPP1 were increased in dedifferentiated U-251MG which may explain their rapid growth. Conversely, proneural phenotype associated markers were weakly expressed. Despite an apparent increase in RT-qPCR, Olig2 was very weakly detected by immunocytochemistry. Similarly, CD133 mRNA expression was low but mainly conserved in dedifferentiated cells and the percentage of positive cells decreased in flow cytometry. It is possible that few dedifferentiated cells expressed CD133, implying heterogeneity within gliospheres. Considering the high CD44 expression, some GSCs expressed both CD133 and CD44, having an intermediate phenotype, between mesenchymal and proneural, which combines both profile complexity. Such phenotype have been associated with large, invasive tumors, presenting rapid growth [11]. CD133 is the most known proneural GSC marker and has been used for a long time as a GSC enrichment marker. Indeed, it appeared essential for GSC maintenance [40] but there is controversy about its use as the sole GSC marker [11]. In our study, weak CD133 expression and GSC diversity support the use of multi-marker and multi-method analysis as the key stone for any GSC study.

Heterogeneity is typical in glioblastoma and particularly in GSC. Due to their plasticity, GSC can easily reshape intracellular pathways to adjust to their environment. Our dedifferentiated and patient derived GSC showed common phenotypic characteristics and stem cell transcriptional patterns, while having their specificities. The 3D GSC organization can be responsible for intra-gliosphere heterogeneity, leading to different cell phenotypes depending on their location. Indeed, hypoxia is known to support GSC phenotype [41] and to increase the expression of several GSC markers such as Sox2 [42] and CD133 [11], but also CD44 under severe hypoxia [39]. Consequently, dedifferentiation process, yet applied to a uniform cell line, may have several intracellular consequences or leads to cell specificities within gliospheres. In vivo, therapies must be effective on high intratumoral cell heterogeneity. If all the cells composing gliospheres may not be GSC, some are definitely stem-like cells. Such diversity in differentiation is mimicking in vivo conditions and gliosphere treatment may induce clonal selection or phenotype drifting. For this reason, future pharmacological studies must include post-treatment analysis of key marker expression such as CD44, CD133, CXCR4, A2B5 and integrin α6. Even if in vitro GSC models lack microenvironment effect, our heterogeneous 3D cultures are suitable to study the effect of inhibitory molecules on complex GSC. Further validation of these GSC models would require in vivo tumorigenicity studies. Tumor growth can be assessed after subcutaneous [14] or intracranial injection [29]. Animal survival and tumor characteristics, such as invasiveness, could be associated with the marker expression profile of GSC models. Moreover, serial transplantation of dedifferentiated GSC would validate their cancer stem cell phenotype, as the number of cells required to induce a tumor can be correlated with the frequency of CSC (Cancer Stem Cells) in a sample [43].

Our gliosphere characterization supports the choice of CXCR4-associated ligand and receptors as therapeutic targets. Indeed, many genes involved in this signaling pathway were more expressed after dedifferentiation. Particularly, MIF and its main receptor, CD74 were highly present at the transcriptional level, but CD74 positive population was rare on flow cytometry. Considering high MIF mRNA expression, its recognition by our GSC may involve secondary receptors such as CXCR4 or CD44. Indeed, CD44, highly expressed in our models, is also able to form complex heteromers for MIF recognition with CXCR4, CXCR7 and CXCR2 [44]. MIF inhibition in GSC decreases stemness, increases radiotherapy sensitivity and induces apoptosis [45]. To confirm MIF as a potential target in our GSC models, future studies must include MIF analysis. Among MIF receptors, CXCR4 appears to have a key role in glioblastoma, which is confirmed in our study. CXCR4 is necessary for GSC phenotype and is associated with EMT, treatment resistance and angiogenesis [25]. Given its importance in tumoral aggressiveness, several inhibitors have been tested on glioblastoma cells. CXCR4 inhibition in GSC decreases self-renewal ability and induces differentiation [46]. Our study provides support to CXCR4 importance in GSC as CXCR4 expression was increased after dedifferentiation in all studied models and was highly expressed in BTIC. Consequently, MIF or CXCR4 inhibitors appear relevant in future GSC targeting strategies.

Conclusion

In summary, we demonstrate that dedifferentiation of 4 commercially available GBM cell lines is an easy way to obtain glioblastoma stem cells. Dedifferentiated cells grow as gliospheres in culture with characteristics closely resembling 2 patient-derived GSC (BTIC) such as self-renewal and stemness marker expression. Analysis of more than 30 GSC and GBM markers revealed a mesenchymal, probably therapy-resistant, GSC phenotype and supports CXCR4 therapeutic targeting in GSC with these 6 studied GSC models. Models have common phenotypic characteristics, but each model has its own specificities in terms of marker expression, gliosphere morphology and growth which may lead to different therapeutic response. To conclude, dedifferentiated cells are robust GSC models and the variability of the 4 models presented in this study would be interesting to evaluate new therapies.

Disclosure of conflict of interest

None.

Abbreviations

ATCC

American Type Culture Collection

BTIC

Brain Tumor Initiating Cells

AC

Adherent cells

CSC

Cancer Stem Cells

EMT

Epithelial-Mesenchymal Transition

FC

Flow Cytometry

GS

Gliospheres

GSC

Glioblastoma Stem Cell

GBM

Glioblastoma

NSC

Neural Stem Cell

PN

Proneural

MES

Mesenchymal

RT-qPCR

Reverse Transcription-quantitative Polymerase Chain Reaction