The GD3 ganglioside promotes cell growth, plasticity and chemotherapy resistance of human glioblastoma cancer stem cells

Abstract

Background

Glioblastoma is the most aggressive primary brain tumour with no curative treatment and inevitable relapse. Therapeutic resistance is, at least, related to the presence of cancer stem-like cells in these tumours. Here, we aimed to demonstrate that the GD3 ganglioside was a relevant marker and actionable target for glioblastoma cancer stem-like cells.

Methods

To this end, we used commercial glioblastoma cell lines, human glioblastoma samples, organotypic culture and xenografted mouse models to study GD3 antigen expression and consequences of its downregulation through a shRNA strategy targeting the ST8SIA1 mRNA which encodes the key enzyme for GD3 synthesis. We performed mono-dimensional Thin Layer Chromatography to analyse ganglioside composition of the glioblastoma samples and RNA-seq analyses to reveal oncogenic pathways and more specifically transcripts affected by ST8SIA1 silencing. Besides, we evaluated GD3 role in stemness of glioblastoma cancer cell, phenotype, microenvironment interaction, and invasion abilities.

Results

We showed that GD3 is the main ganglioside in glioblastoma and that patient-derived cancer stem-like cell lines strongly expressed GD3. This GD3 + population decreased significantly after cell differentiation. GD3⁺ cells sorted from patient samples had stem-like cell properties: they were plastic, clonogenic, and tumorigenic after orthotopic engraftment. Silencing of ST8SIA1/GD3 was associated with a decrease in sphere size, self-renewal and migratory capacities and increased mouse survival. Moreover, increased temozolomide sensitivity was recorded. Finally, data from RNA-seq showed that silencing ST8SIA1/GD3 decreased oncogenic pathways and more specifically the expression of ADAMTS1 and IL33 transcripts.

Conclusions

Taken together, our results suggest that GD3 ganglioside is essential for glioblastoma cancer stem-like cell properties, opening promising targeted therapeutic development.

Background

Glioblastoma (GBM) is the most frequent primary brain tumour in adult and one of the most devastating cancers. Despite first line treatment including radiotherapy and chemotherapy by temozolomide (TMZ), relapse is inevitable in a median delay of 7 to 10 months and no curative treatment is available [1]. Targeted therapies and immunotherapies that revolutionized systemic cancer treatment have failed in multiple clinical trials for GBM patients [2, 3], revealing the urgent need of specific and adapted therapeutic innovation. The therapeutic failure of GBM is essentially based on the resistance of persistent cancer stem cells (CSC) [4] and their interactions with the tumour microenvironment [5]. CSC carry the same features than normal stem cells. Strictly, CSC fulfil the following criteria: (1) extensive self-renewal/clonogenicity abilities (sphere generation), (2) cancer-initiating ability upon orthotopic implantation, (3) aberrant differentiation properties and (4) karyotypic and genetic alterations [6]. It was first admitted that Prominin-1, also called CD133, was the marker of CSC in GBM [7] but it is now controversial since CD133⁻ cells are also able to initiate a tumour after orthotopic injection in nude mice [8,9,10]. In 2003, Nunes et al. showed that A2B5⁺ cells from the adult normal white matter exhibited stem cell properties [11]. In 2006, we showed that all gliomas contained A2B5⁺ cells [12]. In agreement with other studies, we showed that only A2B5⁺ cells and not A2B5⁻ cells have the potential to generate primary and secondary spheres, and to develop tumours after orthotopic injection in nude mice independently of CD133 [13, 14].

A2B5 monoclonal antibody specifically recognizes sialogangliosides, in particular those from the c-series [15, 16]. Farrer and Quarles demonstrated in 1998 that the principal reactive antigens were GT3 and O-acetyl GT3 [17], whose precursor is GD3. Gangliosides are a subfamily of complex membrane-bound sialic-acid-containing glycosphingolipids [18], abundant in brain in which they are involved in cell interaction thus in cell proliferation, differentiation and migration [19]. Particularly di-sialogangliosides, like GD3, are considered as carbohydrate tumour-associated antigens in neuroectoderm-derived tumours including melanoma, neuroblastoma and GBM [20]. The overexpression of complex gangliosides in cancers is associated with the upregulation of ST8SIA1 (ST8 alpha-N-acetyl-neuraminide alpha-2,8-sialyltransferase 1 / GD3 synthase (GD3S)) gene expression which encodes the key enzyme for complex ganglioside synthesis [21].

Currently, CSC from GBM represent promising therapeutic targets to avoid the inevitable therapeutic resistance of these tumours. Despite A2B5 seems to be one of the optimal GBM CSC marker, its use in clinic is impaired by its IgM conformation and by the heterogeneity of its epitopes. In this context, their precursors, the ST8SIA1 and GD3, represent targets of choice for the development of next-generation anti-cancer therapies. Although the ST8SIA1 and GD3 regulation have been studied in various cancer cell types, their implication in GBM CSC phenotype, regulation and therapeutic resistance remains debated. Our goal was to determine whether the ganglioside GD3 was associated with GBM cell stemness, and how the ST8SIA1 downregulation affected the GBM cell stemness, viability, migration and therapeutic response. First, the ganglioside expression pattern of human GBM and normal brain samples was screened and GD3 appeared as the major ganglioside of human GBM whereas it was not detected in normal brain. We then showed that patient-derived GBM CSC highly expressed GD3 and that this expression drastically decreased after cell differentiation. Thus, GBM CSC lines with various levels of GD3 were analysed in order to decipher relationships between GD3 and stemness phenotype. Moreover, we downregulated GD3 antigen expression by using a shRNA targeting ST8SIA1 mRNA. In vitro properties such as proliferation/self-renewal, clonogenicity, and migration, as well as in vivo tumorigenesis ability were correlated to the expression of ST8SIA1 and GD3 positivity. Interestingly, decreased expression of ST8SIA1, and hence GD3, increased the chemotherapy sensitivity of GBM CSC, as evidenced by increased double-stranded DNA damage and cell death. Finally, RNA-seq analyses revealed that silencing ST8SIA1/GD3 decreased oncogenic pathways and more specifically the expression of ADAMTS1 and IL33.

Methods

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Prof. Emeline Tabouret ([email protected]).

Cell lines

U251 (RRID: CVCL_0021), LN229 (RRID: CVCL_0393) and U87 (RRID: CVCL_0022) cell lines (American Type Culture Collection) were cultured as an adherent monolayer in DMEM (Dubelcco’s Modified Eagle’s Medium) supplemented with 10% heat-inactivated foetal bovine serum (FBS), 50 U/mL penicillin, and 50 µg/mL streptomycin at 37 °C in a humidified atmosphere of 5% CO₂ and 95% air. We also used human primary GBM-initiating cells, mesenchymal-like GBM6 and proneural-like GBM9 [22, 23], previously established on the basis of A2B5 expression from two different human IDHwt GBM samples and characterized in our laboratory [13]. They were maintained in culture as floating spheres in CSC serum-free medium supplemented with EGF (Epidermal Growth Factor) and bFGF (basic Fibroblast Growth Factor) as previously described [13, 24]. All cells were tested monthly for the presence of mycoplasma, using Mycoalert™ Mycoplasma Detection Kit. All reagents used for culture are from ThermoFisher (Courtaboeuf, France).

Human samples

Sixty five GBM IDHwt tumour samples were collected at Assistance Publique-Hôpitaux de Marseille (AP-HM) during neurosurgery resection, prior chemotherapy and radiotherapy, and placed in Hank’s Balanced Salt Solution (HBSS). Samples were obtained from the centre of biological resources of AP-HM (CRB BB-0033-00097) according to a protocol approved by the local institutional review board and ethics committee (2014-A00585–42) and conducted according to national regulations. The study was performed in accordance with the declaration of Helsinki. All the patients provided written informed consent. All samples were processed within the 4–24 h following surgery. A total of 29 primary fresh GBM samples were used for flow cytometry (n = 5) or magnetic separation of GD3 + cells (n = 24, including n = 2 for mouse engraftment). In addition, 22 liquid-nitrogen frozen samples were used for RT-qPCR, 7 for gangliosides extraction and 7 liquid-nitrogen frozen human GBM samples were used for GD3 detection by immunofluorescence.

Cell dissociation and magnetic separation of GD3 + cells

Within 4 h after obtention, fresh GBM surgical samples were washed, separated by not taking healthy tissue, automatically roughly cut using a McIlwain tissue chopper and enzymatically dissociated with 5 mg/mL of Trypsin (Sigma-Aldrich, Paris, France) and 200 U/mL of DNase (Sigma-Aldrich) for 15 min at 37 °C. The dissociated cell suspension was filtered through 40 μm, centrifugated and incubated with anti-GD3 mouse monoclonal IgG3 antibody (R24 clone, 1/5, ab11779, Abcam) for 20 min at 4 °C, washed and incubated with magnetic microbead-tagged mouse-specific IgG rat antibody (Miltenyi Biotec, Paris, France) for 20 min at 4 °C. Positive magnetic cell separation was performed using MACS LS column following Miltenyi Biotec guidelines.

Lipid extraction and ganglioside isolation

Seven GBM samples and a corticectomy-pooled sample from 4 non-tumoral human tissues were treated to precipitate proteins and solubilize glycosphingolipids [25]. First, tissue samples (0.8–1 g) were homogenized with 3 volumes of water by a Potter-Elvehjem tissue grinder (Thomas Scientific, Swedesboro, USA). Then, 1 volume of the homogenate was mixed with 20 volumes of chloroform (C)/methanol (M)/water (W) (4:8:3) and incubated for 2 h at room temperature (RT). The suspension was centrifuged at 2200 rpm for 2 min. After the supernatant was removed, the pellet was reextracted in 10 volumes of C/M/W (4:8:3). The resulting clarified extracts were combined. Water was added to bring the C/M/W ratio to 4:8:5.6, resulting in 2 phases. The suspension was mixed and centrifuged to separate the phases. The upper phase (gangliosides) was removed, the lower phase was dried, resuspended in C/M/W (4:8:3) and repartitioned by addition of water. The upper phase from the repartition was combined with the original upper phase and dried. The ganglioside residue was resuspended in 200 µl of C/M/W (2:43:55) with sonication and applied to Sep-Pak C-18 cartridge (Waters Associates, Milford MA). The cartridges were washed with C/M/W (2:43:55) and M/W 1:1 then gangliosides were eluted with 2.5 ml of C/M 1:1. The solvent was dried, and the ganglioside sample was resuspended in 50 µl of C/M 1:1.

Mono-dimensional thin layer chromatography

Ganglioside composition was determined by TLC-pattern [26]. Ganglioside samples were applied on TLC aluminium plate (Merck, Darmstadt, Germany) and GM1, GD3, GD1a, GD1b and GT1b (Matreya, Clinisciences, Nanterre, France) dissolved in C/M 1:1 (1 mg/ml) were used as standard gangliosides. The plate was immersed into the chromatographic solvent system composed by C/M/0.25% KCL (60:35:8) and chromatographed in a closed tank at RT. Gangliosides were highlighted using resorcinol (Merck) reagent to detect lipid-bound sialic-acid. The plate was sprayed with the reagent, immediately covered with a glass pre-warmed and maintained for 10 min at 120 °C. Gangliosides reacted as grey spots.

Downregulation of ST8SIA1: shST8SIA1

The ST8SIA1 shRNA cassette [CCCATCTCTTTGCTATGACTA, TRCN0000000000036045] consisted of a 21 bp target gene-specific sequence, a 6 bp loop, and another 21 bp reverse complementary sequence, under human U6 promoter in a pLKO.1 vector. In addition, a non-target shRNA control (CTRL) was used as negative control (Mission^® RNA, Sigma-Aldrich). The preparation of lentiviral particles was performed with a standard protocol using HEK293T cells co-transfected with pLKO.1 together with psPAX2 packaging plasmid (Addgene #12260) and pMD2.G envelope vector (Addgene #12259). Lentivirus-containing supernatants were filtered through 0.45 μm cellulose acetate filters, aliquoted, and stored at − 80 ◦C until use. For lentiviral infection of 80%-confluent U251, LN229 and U87, 5 µg/mL polybrene (Sigma-Aldrich) was mixed with lentiviral particles to enhance efficacy. For GBM6 and GBM9, infection was performed on poly-DL-ornithin-coated plates. For establishing stable clones, the shRNA-modified cells were selected with puromycin (ThermoFisher) at 0.5, 1, 0.8, 1 and 1.4 ug/ml for U251, LN229, U87, GBM6 and GBM9 respectively starting 48 h after infection and during at least 5 passages. Of all the clones isolated, 1 shCTRL and 2 shST8SIA1-puromycin-selected clones were selected on the basis of ST8SIA1 expression level monitored by Real-Time quantitative PCR (RT-qPCR).

Flow cytometry

Flow cytometry was performed using MACS Quant 10 (Miltenyi Biotec) on dissociated cells to quantify the number of GD3⁺ cells (mouse IgG3 antibody, R24 clone, Abcam and Cy3 AffiniPure Goat Anti-Mouse IgG, Fcγ subclass 3 specific, Interchim, Montluçon, France). Determination of proliferation we performed using PE-Vio^®615, REAfinity-Ki67 antibody (Miltenyi Biotec). For DNA fragmentation analyses cells were seeded in previously poly-DL-ornithin-coated 6-well plate (150 000 per well). After 24 h, cells were treated for 72 h with 100µM and 200µM of TMZ (Sigma-Aldrich) in DMSO. The labelling process with propidium iodide was carried out as described previously [27]. All data were processed using FlowJo software (Tree Star, Inc., Ashland, USA).

Immunofluorescence

Adherent cell lines were grown as monolayers on glass coverslips in 24-well plates in DMEM with 10% FBS. For CSC lines, immunofluorescence was performed directly on 10-day-old floating spheres or on cells cultured on glass coverslips coated with 10 µg/mL poly-DL-ornithin (Sigma-Aldrich) in serum-free medium supplemented with EGF and bFGF. Primary antibodies anti-GD3 (R24 clone, 1/200, Abcam), anti-SOX2 (rabbit monoclonal IgG, clone SP76, 0.03 µg/mL, kindly provided by the Timone hospital - Marseille), anti-NESTIN (mouse IgG, clone 10C2, 1/1000, Abcam) and anti-γ-H2AX (rabbit IgG, clone 20E3, 1/400, Cell Signaling) were applied and incubated for 16 h at 4 °C. Cy3 Goat Anti-Mouse IgG (for GD3, Interchim), Alexa fluor 568 goat anti-rabbit IgG (for SOX2 and γ-H2AX, ThermoFisher), Alexa fluor 488 goat anti-mouse IgG (for NESTIN, ThermoFisher) secondary antibodies were applied at 2 µg/mL for 1 h at RT together with Hoechst 33,342. All the images were obtained using a Zeiss AXIO-Observer Z1 microscope (Carl Zeiss SAS, Rueil Malmaison, France) and were used to count the positive cells.

Clonal Dilution assay

To assess the clonal frequency of secondary spheres obtained from GD3 + cells isolated from GBM as well as GBM6-shST8SIA1 and GBM9-shST8SIA1 cells, single cells were plated in limiting dilution conditions (< 1 cell/well) in 96 well plates. On day 7, the number of spheres was quantified and expressed as a percentage of the total number of cells seeded. At the same time, the diameter of spheres was recorded as previously described [13].

Reverse transcription and RT-qPCR analysis

Total RNA extraction was performed using RNeasy^® Mini kit (Qiagen, Hilden, Germany). Before use, RNA samples were treated with 1U ribonuclease-free deoxyribonuclease (Roche Applied Science, Meylan, France) at 37°C for 20 minutes. One µg of total RNA was used for reverse transcription using random hexamers and Superscript II reverse transcriptase as recommended by the manufacturer (ThermoFisher). Ribosomal 18S and glyceraldehyde-3-phosphate-dehydrogenase (GAPDH) were used as reference genes. Forward and reverse primer 5’-3’ sequences are as follows: 18 S (NR_003286): CTACCACATCCAAGGAAGGCA, TTTTTCGTCACTACCTCCCCG; GAPDH (NM_002046): CAAATTCCATGGCACCGTC, 5’-CCCACTTGATTTTGGAGGGA; ST8SIA1 (NM_003034): TTCAACTTACTCTCTCTTCCCACA, TCTTCTTCAGAATCCCACCATT; ADAMTS1 (NM_006988): ACAGTTCCACCTCCTGCGGC, TCTCCGCTTTCCCAGTCGGC; IL33 (NM_033439): GTGACGGTGTTGATGGTAAGAT, AGCTCCACAGAGTGTTCCTTG. The PCR reactions using SYBR Green reagent and the ABI 7500 Real-Time PCR system (ThermoFisher). The PCR conditions were 10 min at 95 °C followed by 45 cycles of 15 s at 95 °C and 30 s at 65 °C (for ST8SIA1 and GAPDH), 30 s at 67 °C for 18 S, 30 s at 66 °C for ADAMTS1 or 30 s at 55 °C. For each gene, PCR reactions were performed in triplicate for each sample. In addition, human total RNA (Agilent Technologies, Les Ulis, France) was submitted to the same RT-qPCR reactions. The relative expression ratio of the target gene transcript and reference genes transcripts was calculated as ddCt as previously described [28] and expressed as percentage.

Colony formation assay

The colony formation assay was performed by seeding 600 cells per well in 6-well culture dishes. One day after seeding, cells were treated with TMZ at 0 (DMSO as control), 5, 15 and 25 mM and then incubated for 7 days. The colonies were stained with 0.5% crystal violet in 25% ethanol. Colonies larger than 90 μm were counted using ImageJ software (FIJI, National Institutes of Health, USA). As it was not possible to count colonies for U87, stained cells were dissolved with 1% SDS and quantified at 562 nm (Elx800 microplate reader, Bio-Tek, les Ulis France). Results are expressed as percentage of control.

Viability

For two-dimensional (2D) experiments, cell viability was measured using the colorimetric MTT assay (3-(4,5-dimethylthiazol-2yl)-diphenyl tetrazolium bromide, Sigma-Aldrich). U251, LN229 and U87 were seeded in 96-cell plates (5000 cells/well) in 200 µL of media. After 24 h, 20 µL of MTT reagent were added to each well and plates were incubated for 3h30 at 37 °C. The reduced formazan was dissolved in DMSO, and absorbance was measured at 562 nm. Cell viability was expressed as a percentage as compared to control cells.

Intracranial cell transplantation into nude mice

All experimental procedures and animal care were carried out in accordance with the guidelines of the French Government, reviewed and approved by the Regional Institutional Committee for Ethics on Animal Experiments under the authorization number 33388-2021100616089494.

Athymic female nude mice (Envigo, Gannat, France) were anesthetized, and 100,000 cells in 2 µl of sterile Hank’s Balanced Salt Solution were stereotactically injected into the corpus callosum (1 mm anterior to bregma, − 1 mm lateral right, and − 2 mm deep from the cortex surface) as previously described [14]. After surgery, mice were observed until fully recovered. They were weighed 3 times per week throughout the experiment and were euthanized as soon as they started to develop clinical symptoms. Their brains were immediately extracted, fixed in formalin, embedded in paraffin and sectioned for analysis. The mouse tumour sizes were evaluated following the RANO criteria [29]: the product of the maximal perpendicular diameters was recorded and compared.

Mouse organotypic brain slices

For spheroid formation, 5000 cells/well from GBM9-shST8SIA1, GBM6-shST8SIA1 and GBM 304 lines were stained with PKH red fluorescent dye (Merck) according to the standard protocol and seeded in appropriate medium with 20% methylcellulose on U-bottom 96-well plates for 48 h before their use for organotypic slice experiments. The experimental conditions used for organotypic brain slices culture are fully detailed in Baeza-Kallee et al. [30]. One day after cutting of slices and their culture on insert, a spheroid of a diameter of approximately 250 μm from each cell line was harvested using a micropipette and implanted on top of a brain slice. The co-culture was maintained at 37 °C, 5% CO2, 95% humidity. At day 14 post spheroid implantation, slices were formalin-fixed, embedded in paraffin. Between 10 and 15 slices were implanted for each cell line studied. The Zeiss AXIO-observer Z1 microscope (Carl Zeiss) was used for images acquisition.

Histology and immunohistochemical analyses

Paraffin-embedded brains and organotypic slices were sectioned at 4 μm thickness and processed for standard histology (haematoxylin-eosin staining). Immunohistochemistry was performed on adjacent paraffin sections with monoclonal anti-Ki67 (clone MIB-1 #GA626, Dako, les Ulis, France) antibody using the avidin–biotin–peroxidase method (Vectastain Elite ABC kit, #PK-6100, Vector Laboratories, les Ulis, France) according to the manufacturer’s instructions. Sections without primary antibody served as a negative control. Counterstaining was performed with Mayer haematoxylin.

In vitro migration assay

A total of 100,000 cells was seeded in serum-free DMEM on the filter membranes of 8 μm transwell migration chambers (Corning, Amsterdam, Netherlands). DMEM 10% FBS was added to the wells under the filter and cells were allowed to migrate for 24 h. Transmigrated cells were fixed in 4% paraformaldehyde, stained with 0,1% crystal violet, and membranes were photographed using a Zeiss AXIO-observer Z1 microscope (Carl Zeiss). Crystal violet was solubilized in 1% SDS and absorbance was measured at 550 nm (Elx800 microplate reader).

Bulk RNA-seq

RNA isolation, library preparation, and RNA-seq were performed by BGI genomics (Hong Kong, China). Total RNA was extracted from tissues lysed by TissueLyser II (Qiagen) using phenol-chloroform extraction method with Trizol (Invitrogen). The quality of the total RNA was controlled with concentration determination by using the Bioanalyzer 2100 (Agilent). All RNAs used in the present study had RNA Integrity Numbers (RINs) higher than 9.0.  One µg of total RNA was used to prepare mRNA library using MGIEasy RNA Library Prep kit (MGI, China). Libraries were qualified on Bioanalyzer 2100 (Agilent) and quantified by qPCR; single strand circle DNA (ssCir DNA) were formatted as the final library. The libraries of 9 samples were then amplified to make DNA nanoball (DNB) and sequenced on DNBSEQ G400 sequencer (MGI) by pair-end 150, at least 40 M clean reads were obtained for each library.

Differential gene expression analyses

The raw image data obtained by sequencing is converted into raw sequence data (raw data or raw reads) through base calling and stored in the fastq (referred to as fq for short) file format, which contains the reads sequence and sequencing quality information. This project uses the filtering software SOAPnuke (Version v1.5.2) developed by BGI independently for filtering. The specific steps are as follows: (1) Remove the reads containing the adaptor (adaptor pollution); (2) Remove reads whose N content is greater than 5%; (3) Remove low-quality reads. HISAT2 (Version: v2.0.4) was used to genome alignment with Reference Genome (Version: GCF_000001405.39_GRCh38.p13). Bowtie2 [31] was used to map the clean reads to the reference gene sequence (transcriptome), and then RSEM is used to calculate the gene expression level of each sample. The DEseq2 method is used to perform differential gene, DEG, which is an abbreviation for differentially expressed genes, and refers to the gene’s detection of different samples with different expression, the parameter for DEseq2:|log2FC| >1.585, Q-value (Adjusted P-value) < 0.05.

TCGA

Data from 491 samples, including 481 IDHwt GBM patients and 10 non-tumour brain samples included in the TCGA GBM cohort (Plateform HG-U133A) were analysed using the GlioVis dataset [32]. TCGA dataset patient characteristics are described in Additional Table 1.

Table 1 Downregulated genes in GBM9 shST8SIA1 as compared to GBM9 ShCTRL

Statistical analyses

Categorical variables are presented as numbers and percentages, and the quantitative results as a median with minimum and maximum range or a mean with standard error as an index of dispersion. Comparisons were made with Student’s t-test, Chi2 or Fisher tests for qualitative data and T-test or Mann–Whitney and Kruskal-Wallis for quantitative data, as appropriate. Data are expressed as mean ± SEM. Overall survival (OS) was defined to be the time from the date of surgery to death, censored at the date of last contact. The Kaplan–Meier method was used to estimate survival distributions. Log-rank tests were used for univariate comparisons. Analyses were performed with the GraphPad Prism 5.0 statistical software. All statistical tests were 2-sided and the threshold for significance was p < 0.05. Significances: *** p < 0.001, ** p < 0.01, * p < 0.05. Tendency: ¤ 0.05 < p < 0.06.

Results

GD3 is the major ganglioside of IDHwt GBM and is highly expressed by GBM CSC

In order to screen for GD3 (Fig. 1A) expression in human IDHwt GBM samples, the expression was first analysed by TLC in 7 samples. Interestingly, GD3 was systematically expressed and appeared to be the major ganglioside in most samples. Importantly, with this technique, the GD3 ganglioside was not detected in the normal cortex from 4 independent cortectomy samples (Fig. 1B). By flow cytometry GD3 was found at the membrane of 17.60% +/- 2.25% cells (n = 5 GBM samples, Fig. 1C). By immunofluorescence, GD3 was detected in the same proportion, at the membrane of small clusters of GBM cells (Fig. 1D). In addition, ST8SIA1 gene expression was analysed by RT-qPCR in 22 frozen patient GBM samples. ST8SIA1 was more expressed in GBM (232.5% +/- 30.95%) as compared with normal human brain standard (CTRL, set at 100%) (Fig. 1E). To go a step further, ST8SIA1 gene expression was assessed in 481 GBM and 10 non-tumour brain from the public TCGA database through the prism of MGMT promoter methylation and the Verhaak molecular classification [22] (Fig. 1F, Additional Fig. 1 and Additional Table 1). ST8SIA1 expression was higher in GBM as compared to non-tumour brain samples, and ST8SIA1 was very rarely amplified or mutated in GBM samples. ST8SIA1 expression was not correlated to patient overall survival. No correlation was found between the level of ST8SIA1 gene expression and the molecular features of GBM samples, suggesting that ST8SIA1 high expression was a constant feature of GBM whatever their MGMT promoter methylation status or molecular profile, classical, mesenchymal or proneural.

Fig. 1

GD3 is highly expressed in GBM and by GBM cancer stem cells. (A) Schematic diagram of the biosynthesis of GT3 from GD3 and GM3. Graphic notations are labeled: glucose; galactose; N-acetylneuraminic acid. (B) Demonstration of GD3 expression after total gangliosides purification from 7 individual GBM samples and separation by Thin Layer Chromatography. Normal cortex pool from 4 patients is shown as reference. Positions of GM1, GD3, GD1a, GD1b and GT1b standards are reported. (C) GD3 detection by flow cytometry on 5 freshly dissociated GBM samples. Results are presented as % of expressing cells. (D) Immunofluorescence staining of GD3 (red) on cryosections of 3 GBM samples. Hoechst staining of the cell nuclei (blue) is also shown. Scale bar = 50 μm. (E)ST8SIA1 mRNA quantification by RT-qPCR in 22 human GBM samples. Results are expressed as % from standard RNA. (F)ST8SIA1 gene expression profiles according to MGMT promoter methylation status (left graph) and molecular classification: classical, mesenchymal, proneural (right graph) in 481 patient samples from TCGA database. * p < 0.05. (G) Detection of GD3 (red) expression by immunofluorescence on 14-days old floating spheres of GBM6 and GBM9 cell lines grown in CSC medium. Hoechst staining of the cell nuclei (blue) is also shown. Scale bar = 50 μm. (H) GBM6 cell phenotype after 14 days in CSC medium versus 7 days in 10% FBS. Scale bar = 50 μm. (I) GBM9 cell phenotype after 14 days in CSC medium versus 7 days in 10% FBS. Scale bar = 50 μm. (J) GD3 and ST8SIA1 expression in GBM6 after 14 days in CSC medium or 7 days in 10% FBS medium. At least 4 independent experiments were performed. * p < 0.05. (K) GD3 and ST8SIA1 expression in GBM9 after 14 days in CSC medium or 7 days in 10% FBS medium. At least 4 independent experiments were performed. * p < 0.05

To evaluate the GD3 and ST8SIA1 expression in the stem-like cell population, we used the GBM6 and GBM9 CSC lines, that we previously established [24]. Detection of GD3 expression by immunofluorescence on 14-day-old floating spheres showed that GD3 was highly expressed in both models, strongly suggesting that GD3 was a stem cell marker (Fig. 1G). In continuity, the expression of GD3 and ST8SIA1 was evaluated in these CSC lines in standard condition (CSC medium) versus 10% FBS. As expected, in CSC medium, the cells grew as floating spheres (Figs. 1H left and 1I left). Conversely, in the presence of 10% FBS, the cells adhered to the support and adopted a phenotype of differentiated cells (Figs. 1H right and 1I right). Flow cytometry showed that GD3 expression was significantly decreased in the presence of FBS (* p < 0.05), concomitantly to the CSC differentiation (Figs. 1J left and 1K left). In accordance, in the presence of FBS, we observed a drastic and significant reduction of ST8SIA1 expression, in comparison to standard CSC medium (Figs. 1J right and 1K right). Overall, these results suggested that GD3 was one of the major gangliosides in IDHwt GBM and was associated with the stemness state of GBM CSC.

The GD3⁺ cells isolated from patient GBM samples exhibit cancer stem cells properties in vitro

We have developed a reproducible protocol for sorting GBM cells based on GD3 expression. A total of 24 tumour samples were processed. The purity of GD3 sorting was analysed by cytometry, allowing us to define a GD3⁺ population, composed by 72.1% +/- 8.8% of GD3⁺ cells, versus a GD3⁻ population composed by 11.7% +/- 3.82% of GD3⁺ cells (* p < 0.05; Additional Fig. 2A). After magnetic cell-sorting and culture in CSC serum-free medium, primary spheres were recorded in the GD3⁺ fraction (Fig. 2A) in 13/24 GBM cases within 2 weeks on average (from 5 days to 25 days). These spheres were abundant, slightly irregular in shape and their volume increased continuously. As a comparison, rare and small irregular spheres were observed in the GD3⁻ fraction for only 2 samples. In subsphere-forming assay, we observed that GD3⁺ secondary spheres demonstrated long-term proliferation and self-renewal. Interestingly, 5 cases could be expanded for several passages (> 5 passages) indicating a greater self-renewal capacity in vitro, including the cell lines GBM 301 and GBM 304. Both produced primary spheres as early as 6 and 10 days respectively and gave rise to secondary spheres (Fig. 2A) for more than 30 passages with a high clonogenicity rate of around 80% for the 2 cell lines (Fig. 2B). Moreover, we recorded the size of the secondary spheres at 7 days after passage and reported in Fig. 2C the results for 3 passages. Although the average sizes varied from one passage to the other, there was a consistency in proliferation capacities over passages. As expected, detection of GD3 expression by immunofluorescence and cytometry on 14-day-old floating secondary spheres showed that GD3 was highly expressed in both GBM 301 and GBM 304. After dissociation, more than 91.83% +/- 1.81% GBM 301 (n = 4) cells and 59.30% +/- 3.66% GBM 304 (n = 4) cells were GD3⁺ (Fig. 2D and E). Similarly, ST8SIA1 mRNA was highly expressed in both cell lines (Fig. 2F).

Fig. 2

GD3⁺ cells isolated from patient GBM samples exhibit cancer stem cells properties in vitro and in vivo. (A) Phase contrast images of GBM 301 and GBM 304 spheres after 14 days in CSC medium at 3 different passages (p). Scale bar = 100 μm. (B) Percentage of clonogenicity at 7 days after seeding of GBM 301 and GBM 304 cell lines. (C) Sphere sizes measured at day 7 at different passages for GBM 301 and GBM 304 cell lines. (D) Detection of GD3 (red) expression by immunofluorescence on 14-days old floating spheres of GBM 301 and GBM 304 cell lines. Hoechst staining of the cell nuclei (blue) is also shown. Scale bar = 50 μm. (E) GD3 expression by flow cytometry of GBM 301 and GBM 304 cell lines. Four independent experiments for each cell line were performed. Results are presented as % of expressing cells. (F)ST8SIA1 gene expression on 3 different passages of GBM 301 and GBM 304. Results are expressed as % from standard RNA. (G) Ki67 proliferation marker expression shown by flow cytometry in GBM 301 and GBM 304 cell lines. Four independent experiments were performed. (H) Immunofluorescence expression of GD3 as well as SOX2 and NESTIN stem cell markers on GBM 301 (left) and GBM 304 (right) cell lines grown as adherent monolayer on poly-DL-ornithin. Scale bar = 50 μm. (I) Plasticity test: flow cytometry and immunofluorescence expression of GD3 on GBM 304 cell lines grown in CSC medium, then in 10% FBS and then again in CSC medium. Scale bar = 50 μm. (J) Tumour detection in nude mice after orthotopic injection of GD3⁺ cells (top) and GD3⁻ cells (below). Tumour size evaluation in mm² after GD3⁺ or GD3⁻ cell orthotopic injection in nude mice is also shown (p = 0.2). Tumour sizes were calculated following the RANO criteria using the product of maximal perpendicular diameters of tumours. Scale bar = 500 μm. (K) Representative coronal brain slices 14 days after implantation of a GBM 304 spheroid. Scale bar = 500 μm. Haematoxylin / eosin (H&E) and Ki67 staining are shown. Magnification of the implantation area are also presented (right). Scale bar = 100 μm

Finally, we further analysed their proliferation rate and showed that Ki67 antigen was expressed in more than 80% of the cells suggesting that GBM 301 and GBM 304 were in active cell cycle status (Fig. 2G). To analyse their stemness properties, we first evaluated the markers SOX2 and NESTIN expression by immunohistochemistry in both cell lines and observed a positive staining in the vast majority of cells. Indeed, counting showed that 77.18% (GBM 301) and 86% (GBM 304) of cells were positive for SOX2, while 82.55% (GBM 301) and 97.91% (GBM 304) were positive for NESTIN (Fig. 2H). Then, we carried out a phenotypic plasticity test to analyse the response of GBM 304 to variations in culture conditions with focus on the sphere formation ability and the GD3 expression. From day 0, dissociated cells were cultured into stem cell medium. At day 14, spheres highly expressed GD3 (Fig. 2I left). After dissociation, the cells were put into DMEM 10% FBS for 7 days. Under these conditions, GBM 304 cells were adherent and exhibited a fully differentiated phenotype associated to a dramatic GD3 expression decrease, both by flow cytometry and immunofluorescence (Fig. 2I middle). After a further passage in stem cell medium, the cells were able to form spheres again and showed an increase in GD3 expression (Fig. 2I right).

Overall, these results showed that the GD3⁺ cell lines that have been established displayed a remarkable capacity for adaptation to environment and for plasticity, as they re-acquire stemness properties after passing transiently through a differentiated state.

The GD3⁺ cells isolated from patient GBM samples exhibit cancer stem cells properties ex vivo and in vivo

To evaluate the tumorigenicity of GD3⁺ cells, we sorted GD3 cells from 2 fresh GBM patient samples, allowing us to orthotopically engraft 100 000 cells per mouse. In total, GD3⁺ and GD3⁻ cells were injected in 6 mice, respectively (3 mice per condition and per patient). Two animals (one per patient) were not analysable due to early medical complication unrelated to tumour cell injection and tumour growth, leading to the exclusion of 4 mice (1 couple per patient). At final, 4 GD3⁺ and 4 GD3⁻ animals were analysed and compared by couple that were euthanized at the same time (one GD3⁺ mouse and one GD3⁻ mouse from the same series) at 188, 207, 237 and 351 days after graft. Representative sections for both groups of mice are shown. Brains extracted were examined macroscopically and microscopically for the presence of a tumour using standard histology and Ki67 staining. In the GD3⁺ group, 4/4 mice developed tumours (Fig. 2J left), heterogeneous in size (mean tumour size 22.45 mm² +/- 11.15), and highly positive for Ki67 (not shown). We noticed that in 3 of them, tumour cells were observed at distance in the contralateral hemisphere. In contrast, in the GD3⁻ group, 3/4 mice presented with a very small tumour without any extension at distance. The last one had no brain tumour. In this group, mean tumour size was 3.157 mm² +/- 1.189 (Fig. 2J right).

To go further, and to assess the ability of GBM 304 to grow and infiltrate the brain microenvironment, we used an integrated ex vivo organotypic slice model. Brain slices were used to graft GBM 304 spheroids. After 14 days, the slices were prepared for microscopic analysis. Significant engraftment and growth in the slice, often with a large zone of infiltration were observed as well as a high Ki67 positivity (Fig. 2K).

Taken together, and although we have limited data, these results highlight the tumorigenic ability of GD3⁺ GBM human cells.

ST8SIA1 Silencing decreases GD3 expression, CSC proliferation, clonogenicity and migration

Using GBM6 and GBM9 CSC lines, we generated stable clones using the shRNA method to downregulate the ST8SIA1 gene expression. For each CSC line, 1 shCTRL and 2 shST8SIA1 (shS1 / shS2) were analysed. The silencing of ST8SIA1 was confirmed by RT-qPCR (Fig. 3A) in correlation with the loss of GD3 expression (Fig. 3B). This downregulation was associated with a decrease in sphere size over time between shCTRL and shST8SIA1 (Fig. 3C) in both GBM6 and GBM9 (* p < 0.05; ¤ p = 0.057). Moreover, the downregulation of ST8SIA1 was associated with a significant reduction of cell clonogenicity ability from 84% (shCTRL) to 60% (** p < 0.01) and 44% (* p < 0.05) for GBM6 shS1 and shS2, respectively. Regarding GBM9 clones, the clonogenicity ability dropped from 83% in shCTRL to 44% (* p < 0.05) and 43% (* p < 0.05) for GBM9 shS1 and shS2, respectively (Fig. 3D). Moreover, transwell assays revealed that ST8SIA1 downregulation drastically impaired migration ability compared to shCTRL cells as shown on representative phase contrast images (Fig. 3E top). When compared to GBM6 shCTRL (100%), quantification of cell migration after solubilization was 36.67% +/- 14.23% (* p < 0.05) for GBM6 shS1 and 39.11% +/- 18.12% (* p < 0.05) for GBM6 shS2 (Fig. 3E below). The same impairment is measured for GBM9.

Fig. 3

The silencing of ST8SIA1 decreases GBM stem-like cell growth, proliferation, clonogenicity and migration. (A)ST8SIA1 mRNA quantification by RT-qPCR at different passages of GBM6 shST8SIA1 (shCTRL, shS1, shS2) and GBM9 shST8SIA1 (shCTRL, shS1, shS2) cell lines. Results are expressed as % from shCTRL. (B) Detection of GD3 (red) expression by immunofluorescence in GBM6 shST8SIA1 (shS1, shS2), GBM9 shST8SIA1 (shS1, shS2) and shCTRL cell lines. Hoechst staining of the cell nuclei (blue) is also shown. Scale bar = 50 μm. (C) Sphere sizes measured in µm at 3, 6, 9 and 14 days (D) after seeding at different passages in GBM6 shST8SIA1 (shS1, shS2) (top) and GBM9 shST8SIA1 (shS1, shS2) (below). Results for GBM6 shCTRL and GBM9 shCTRL are also shown. * p < 0.05; ¤ p = 0.057 (D) Percentage of clonogenicity at 7 days after seeding of GBM6 shST8SIA1 (shS1, shS2) and GBM9 shST8SIA1 (shS1, shS2) cell lines and comparison with shCTRL cell lines. (E) 24 h transwell migration assay for GBM6 shST8SIA1 (shS1, shS2) and GBM9 shST8SIA1 (shS1, shS2) cell lines as compared to respective shCTRL. Top: representative migration pictures after crystal violet staining. Scale bar = 50 μm. Below: cell migration quantification expressed as % from shCTRL. Four independent experiments were performed. * p < 0.05. (F) Ki67 staining of coronal brain slices cultured 14 days after spheroids implantation of GBM6 shST8SIA1 (shS1, shS2) and GBM9 shST8SIA1 (shS1, shS2) as compared to shCTRL. Scale bar = 500 μm

Finally, we took advantage of the ex vivo organotypic slice model to evaluate the impact of ST8SIA1 silencing on the engraftment, growth and infiltration capacity of CSC GBM in the brain microenvironment. Pictures showed that GBM6 shCTRL were able to engraft, proliferate and infiltrate the mouse brain slice whereas both GBM6 shS1 and shS2 did not grow on the slice. For the GBM9 engineered cells, we draw the same conclusions, with the difference that GBM9 shS1 and GBM shS2 were able to settle lightly in the brain slice (Fig. 3F). Taken together, these results showed that ST8SIA1 expression downregulation impaired CSC GBM proliferation, clonogenicity, migration and brain microenvironment interactions.

The Silencing of ST8SIA1 decreases proliferation and viability of adherent GBM cells and increases mouse survival

Reduced CSC proliferation and growth hampered our ability to assess the in vivo impact of ST8SIA1 inhibition. We therefore sought classical adherent GBM cell lines that significantly expressed the ST8SIA1 gene in order to silence it (Fig. 4A). We downregulated ST8SIA1 by shRNA in U251, LN229 and U87 cell lines, and confirmed, after stabilization of the modified cell lines, by RT-qPCR (Fig. 4B) and GD3 immunofluorescence (Fig. 4C) that the ST8SIA1 gene and GD3 expressions were efficiently downregulated in three classical GBM cell lines. First, we confirmed the impact of ST8SIA1 downregulation on colony formation ability (Fig. 4D) and cell viability (Fig. 4E). Then, we performed a series of stereotactic orthotopic injections of U87-shCTRL and U87-shST8SIA1 (shS1 cell line) in corpus callosum of four nude mice for each cell line. We observed a significant improvement of median survival in the U87-shST8SIA1, in comparison to the shCTRL group (* p = 0.04). In the shST8SIA1 group, no mouse died during the study while in the shCTRL group, 3 mice died at 19, 19 and 48 days after injection (Fig. 4F). The brains of these mice were analysed by standard histology and Ki67 staining and shown a significant tumour proliferation in each case of shCTRL group (Fig. 4G).

Fig. 4

The silencing of ST8SIA1 decreases proliferation and viability of adherent GBM cells and increases mouse survival. (A)ST8SIA1 mRNA quantification by RT-qPCR in U251, LN229 and U87 cell lines. Results are expressed as % from standard RNA. At least 3 independent experiments were performed. (B)ST8SIA1 mRNA quantification by RT-qPCR in U251-shST8SIA1 (shS1, shS2), LN229-shST8SIA1 (shS1, shS2) and U87-shST8SIA1 (shS1, shS2) cell lines. Results are expressed as % from shCTRL. At least 4 independent experiments were performed. * p < 0.05 and ** p < 0.01. (C) Detection of GD3 (red) expression by immunofluorescence on U251-, LN229- and U87-shST8SIA1 (shS1, shS2) and shCTRL cell lines. Hoechst staining of the cell nuclei (blue) is also shown. Scale bar = 50 μm. (D) Colony assays of U251-, LN229- and U87-shST8SIA1 and shCTRL cell lines. Representative pictures of colony assays (left); quantification of colonies (right). (E) Relative cell viability of U251-, LN229- and U87-shST8SIA1 cell lines after 24 h in culture. Results are expressed as % from shCTRL. (F) Mouse survival plot of nude mice injected with U87-shS1 (n = 4) or U87-shCTRL (n = 4). * p = 0.04. (G) In vivo tumour growth of U87-shCTRL versus U87-shST8SIA1. Scale bar = 500 μm

The Silencing of ST8SIA1 induces the modifications of major oncogenic pathway and decreases the expression of ADAMTS1 and IL-33

To understand the involvement of ST8SIA1 and GD3 in GBM CSC behaviour, we performed a pan-transcriptomic analysis of replicates of GBM9 shCTRL, shS1 and shS2 cell lines cultured for 7 days in CSC medium. Differentially expressed genes were identified as those with a fold change of Log2 > [1.585] which corresponds to a ratio > [3] and a Q-value < 0.05. Four lists of genes were established as follows: 2 lists for GBM9 shS1 and 2 lists for GBM9 shS2 corresponding to the downregulated and upregulated genes (Additional Tables 2, 3, 4 and 5).

Table 2 Upregulated genes in GBM9 shST8SIA1 as compared to GBM9 ShCTRL

A total of 78 genes were found differentially expressed in GBM9 shS1 as compared to GBM9 shCTRL, 36 being downregulated and 42 being upregulated. A total of 496 genes were differentially expressed in GBM9 shS2 as compared to GBM9 shCTRL, 279 being downregulated and 217 being upregulated (Fig. 5A and B and Additional Tables 2, 3, 4 and 5). In total, silencing of ST8SIA1 in the GBM9 CSC line induces changes in the expression of 533 genes (Venn diagrams: Figure 5A and B, left). Two heatmaps showing the downregulated and the upregulated genes differentially expressed upon ST8SIA1 silencing by shRNA were generated (Fig. 5A and B, middle). We analysed all the genes of interest through KEGG functional enrichment analyses: most of them were involved in major biological processes, including signal transduction, signalling molecules and interaction, oncogenesis, the sensory system, development and regeneration, and finally the immune system (Fig. 5A and B, right).

Fig. 5

(A) downregulated genes in GBM9 shS1 (green) and shS2 (orange) cell lines: Venn diagram (left), expression Heat Map and KEGG functional pathway classifications (right) of all downregulated genes in GBM9 shST8SIA1. (B) upregulated genes in GBM9 shS1 (green) and shS2 (orange) cell lines: Venn diagram (left), expression Heat Map and KEGG functional pathway classifications (right) of all upregulated genes in GBM9 shST8SIA1. (C)IL33 gene expression profiles in 491 patient GBM samples from TCGA database as compared to non-tumoral brain (top). IL33 expression profiles according to molecular classification are also shown (middle). Overall survival curve (months) of patients with low or high IL33 gene expression in GBM (bottom). * p = 0.019. (D) ADAMTS1 gene expression profiles in 491 patient GBM samples from TCGA database as compared to non-tumoral brain (top). ADAMTS1 expression profiles according to molecular classification are also shown (middle). Overall survival curve (months) of patients with low or high ADAMTS1 gene expression in GBM (bottom). * p = 0.011. (E)IL33 (top) and ADAMTS1 (below) mRNA quantification by RT-qPCR in 5 independent samples of GBM9 shCTRL, GBM9 shS1 and GBM9 shS2, ** p < 0.01

We focused on the common genes differentially expressed between GBM shS1 and GBM9 sh2 to identify those co-regulated with ST8SIA1 and analysed their expressions in the public TCGA database (Tables 1 and 2). Among these 41 common genes (Fig. 5A and B, left), only 2 genes presented with a coherent variation between shCTRL and shST8SIA1, and between GBM and normal brain (after exclusion of ST8SIA3, Tables 1 and 2) and presented with a significant unfavourable prognostic value for GBM patients in the TCGA database. Indeed, IL-33 and ADAMTS1 were consistently overexpressed in GBM samples as compared to non-tumoral brain tissue (*** p < 0.001 and * p < 0.05 respectively, Fig. 5C and D, top). Moreover, ADAMTS1 expression was significantly lower in the proneural molecular subgroup than in the mesenchymal and classical subgroups (*** p < 0.001, Fig. 5D, middle). Regarding their prognostic impact, if IL-33 and ADAMTS1 expression did not impact mesenchymal or classical patient survivals, their expressions allowed to discriminate different prognostic groups in the proneural molecular subgroup. Patients with a higher IL-33 expression had a worse outcome (* p = 0.019, Fig. 5C, bottom). Similarly, patients with the lowest ADAMTS1 expression level had a better prognosis while the patients with the highest expression presented with the worse outcome (* p = 0.011, Fig. 5D, bottom).

Finally, the RT-qPCR method was chosen to validate the RNA-seq results in both GBM9-shST8SIA1 and GBM6-shST8SIA1. Five independent samples from each cell line were used to quantify the mRNA levels of the two selected genes. As expected from the RNA-seq results and as shown in Fig. 5E, the expression levels of IL-33 and ADAMTS1 were significantly decreased in the GBM9-shS1 (** p = 0.008) and GBM9-shS2 (** p = 0.008) lines as compared with the GBM9-shCTRL. Similar results were found for GBM6-shST8SIA1 cells as compared with GBM6-shCTRL (Additional Fig. 3).

Overall, our results show that the downregulation of ST8SIA1 is associated with the reduction of oncogenic pathways and the poor prognostic factors IL-33 and ADAMTS1.

To complete these analyses, we performed a literature review of putative co-targets of IL33 and ADAMTS1 [33,34,35,36,37,38,39,40]. In the TCGA database, eight and seven genes were differentially expressed between GBM and non-tumour brain samples, were associated with patient overall survival and were functionally associated with IL33 and ADAMTS1 at the protein level (STRING analyses), respectively (Additional Fig. 4). For instance, IL33 network included actors of MAPK and PI3K signalling and interactions with immune system (ICAM1 and ANXA1), while ADAMTS1 network included genes and proteins involved in cell adhesion, migration and extracellular matrix remodelling such as L1CAM, VCAN or TIMP3.

The Silencing of ST8SIA1 sensitizes GBM CSC to TMZ

Since GBM CSCs are highly chemoresistant, it is crucial to determine conditions that make these cells chemosensitive. Therefore, we evaluated the potential effect of ST8SIA1 silencing on TMZ sensitivity. γ-H2AX, the phosphorylated form of histone H2AX functions as a sensor of DNA damages. We first assessed damages accumulation using immunofluorescence expression of γ-H2AX. We showed that reduced ST8SIA1 expression led to a significant increase in DNA damage accumulation under TMZ, in a dose-dependent manner, both in GBM CSC lines (Fig. 6A and B) and in adherent GBM cells (Additional Fig. 5B, C and D) (* p < 0.05; ¤ p = 0.057). We then showed that ST8SIA1 reduction was associated with an increasing CSC death in both GBM6- and GBM9 shST8SIA1 versus shCTRL (* p < 0.05, Fig. 6C). For GBM6, at 100 mM TMZ, cell death was 35.7% +/- 3.9% in shCTRL, 59.3% +/- 6.1% in shS1 and 50.3% +/- 4.7% in shS2 (* p < 0.05; ¤ p = 0.052); Fig. 6C left). At 200 µM TMZ, cell death was 45.4% +/- 2.7% in shCTRL, 64.2% +/- 5.5% in shS1 and 58.0 +/- 4.8% in shS2. For GBM9, similar differences between shCTRL and shST8SIA1 were measured (* p < 0.05; ¤ p = 0.057). Figure 6C right). Interestingly, colony analysis and viability assays of U251-shCTRL and U251-shST8SIA1 (shS2) cell lines showed that ST8SIA1 silencing restored TMZ sensitivity alter cell viability (** p < 0.01, Additional Fig. 5).

Fig. 6

The silencing of ST8SIA1 is synergistic and sensitize GBM CSC to TMZ. (A) Detection of γ-H2AX (red) expression by immunofluorescence in GBM6 shST8SIA1 and shCTRL cell lines (left). Hoechst staining of the cell nuclei (blue) is also shown. Scale bar = 50 μm. The graph (right) displays quantification of positive nuclei for γ-H2AX (n = 6). * p < 0.05; ¤ p = 0.057. (B) Detection of γ-H2AX (red) expression by immunofluorescence in GBM9 shST8SIA1 and shCTRL cell lines (left). Hoechst staining of the cell nuclei (blue) is also shown. Scale bar = 50 μm. The graph (right) displays quantification of positive nuclei for γ-H2AX (n = 5). * p < 0.05. (C) DNA fragmentation of 100µM and 200µM TMZ treated cells determined by flow cytometry of propidium iodide-stained nuclei. GBM 6- and GBM9 shST8SIA1 or shCTRL cell lines were treated for 72 h (n = 6 for GBM 6 and n = 4 for GBM 9). * p < 0.05; ¤ p = 0.057

Taken together, these results showed that ST8SIA1 silencing sensitizes GBM cells to chemotherapy by increasing DNA double-stand breaks and cell death.

Discussion

In the present study, we showed that GD3 was the main ganglioside of GBM and was associated with the stemness phenotype, plasticity and tumorigenicity of GBM CSC. Moreover, the silencing of the ST8SIA1/GD3S decreased the aggressiveness of CSC GBM, reversed their chemoresistance and increased the mouse survival. Finally, we observed that this silencing impacted GBM CSC regulation and maintenance by decreasing oncogenic pathways and by reducing tumorigenic factor expression.

GBM therapeutic resistance is a major challenge in oncology and the development of innovative and efficient therapies for patient is crucial. This systematic chemo and radio-resistance is at least led by the presence of CSC that were reported to be critical contributors to tumour therapeutic resilience [4]. One of the most causes of the radio-chemoresistance of GBM CSC is the innate activation of DNA repair mechanisms [41], that we were able to counteract by inhibiting the ST8SIA1 expression. Because the CSC are one of the main actors of GBM therapeutic resistance, the identification of a dedicated biomarker of GBM CSC, allowing their targeting, would be a major improvement for patient management. Until now, GBM CSC markers included, in part, CD133, CD15, CD90, integrin α-6 or CD171/L1CAM [4]. But none was sufficiently robust and useful for therapeutic development. Antibody-based CD133 showed promising results in preliminary preclinical evaluation [42] but currently failed to reach clinical trial evaluation. Similarly, other therapeutic approaches, such as BMI1 or EZH1 inhibition [43] or pyrimidine synthesis disruption [44] were currently not translated to clinical trials. Finally, our team reported that A2B5 antibody has the ability to recognize GBM CSC with a very high specificity [13]. Unfortunately, its use in clinical practice is impaired notably by its IgM conformation.

In this context, we evaluated the expression and the implication of GD3, which is the main precursor of the A2B5 epitopes, in different patient samples and patient-derived models. Derived from the enzymatic action of ST8SIA1/GD3S on the mono-sialylated ganglioside GM3, GD3 is the leading ganglioside of the b-series. Previous publications reported that GD3 expression is finely regulated during brain development but its expression drops drastically before birth, preventing major toxicity of its therapeutic inhibition [45, 46]. In the oncology field, GD3 was reported to be highly expressed by the neuroectodermal derived tumours such as melanoma and neuroblastoma [47, 48]. The GD3 expression in glioma was also previously reported to be correlated to the tumour grade and aggressiveness [49,50,51,52]. Previous preclinical studies showed the implication of GD3 and the GD3S in glioma cell growth and proliferation, cell cycle, invasion, migration, and microenvironment regulation [53]. The stemness implication of GD3 was also suggested in the major publication from Yeh et al., published in 2016. After generating neurospheres from commercial GBM cell lines, the authors showed that the positive cells for GD3 displayed functional characteristics of GBM CSC. Moreover, these cells were capable of subcutaneous tumour initiation in mouse models, suggesting that GD3 was a major determinant of GBM tumorigenicity. In our study, we demonstrate the implication of GD3 in GBM stemness by using fresh patient tumour samples and patient-derived models allowing us to generate GD3⁺ cell lines. This validation on patient tissue and patient-derived models is crucial before transferring the results from bench to bed, taking into account the well-known limitations of commercial cell lines. Moreover, we showed that GD3⁺ cells had plasticity capacities and were directly tumorigenic after orthotopic brain engraftment. Cancer stem cell plasticity is the ability to dynamically switch between CSC and non-CSC states and plays an important role in therapeutic resistance and tumour relapse [54]. Importantly, we also showed that the inhibition of GD3S reversed the chemoresistance of GBM CSC. The use of increasing doses of TMZ was associated with a dose-dependent increase of DNA damage accumulation in the shST8SIA1 CSC leading to an increasing cell death rate, while the shCTRL cells remained insensitive to chemotherapy DNA damages. These results highlight that decreasing ST8SIA1 and GD3 expression may lead to restore the chemosensitivity of GBM CSC through their incapacity to repair DNA damages, as we showed by the increasing expression of γ-H2AX under treatment.

Finally, we observed that the downregulation of ST8SIA1 was associated with specific pathway alterations, including the downregulation of oncogenic pathways and ADAMTS1 and IL-33 expressions. ADAMTS1 is a secreted protease that has the ability to modify the extracellular matrix during physiological and pathological processes [55]. Interestingly, we showed that ADAMTS1 was upregulated in GBM (TCGA database), suggesting that the downregulation of ST8SIA1 may reverse this overexpression and turn the tumour cell behaviour towards a non-tumoral brain phenotype. ADAMTS1 was already reported as a pejorative prognostic factor in breast [56] and lung [55] cancers, pancreatic adenocarcinoma [57], renal cell carcinoma [58] and gliomas [59]. In the neuro-oncology field, the literature related to ADAMTS1 implication remains limited. ADAMTS1 was suspected to participate to the cell interactions and signalling events leading to GBM heterogeneity and proliferation [60, 61]. Interestingly, ADAMTS1 was associated with GBM stem-like features in two preclinical studies. In the first one [62], the U87 and the U251 cell lines KO for ADAMTS1 and culture in CSC medium were unable to form secondary spheres suggesting a loss of self-renewal capacity after ADAMTS1 downregulation. In another one [63], the authors suggested that ADAMTS1 may affect the invasive phenotype of GBM CSC by up-regulating NOTCH1-SOX2 signalling pathway and then promoting glioma growth. The co-expression of ADAMTS1 with matrix proteins such as L1CAM, VCAN or TIMP3 strongly suggest that ADAMTS1 may actively participate in remodeling the extracellular matrix environment to promote tumour progression and metastasis [56]. In parallel, we observed that the downregulation of ST8SIA1 led to the downregulation of IL-33 expression. Again, because we showed that IL-33 was overexpressed in GBM, its down-regulation through the decrease of ST8SIA1 may reverse the cell phenotype toward a normal behaviour. IL-33 was shown to be correlated with a bad prognosis in several types of cancer [64], to be involved in interactions with immune system [39] and interestingly, IL-33 was previously reported to be implicated in DNA damaging agent sensibilization, tumour microenvironment modulation and stemness phenotype of gliomas [65,66,67,68]. Notably, Chung et al. showed that glioma cells overexpressing IL-33 were resistant to TMZ because of the increasing expression of DNA repair genes [65]. These results are in line with our last observation showing the reversion of TMZ resistance by decreasing ST8SIA1 expression. Regarding stemness plasticity, Lin et al. observed that IL-33 activated epithelial to mesenchymal transition and stemness of glioma cells through JNK signalling pathway [67]. Based on these preliminary literature reports, we could hypothesize that the inhibition of ST8SIA1 in GBM CSC leads to ADAMTS1 and IL-33 downregulation, allowing stemness inhibition, tumour cell differentiation and chemotherapy sensitivity.

Taken together our results open promising perspectives in therapeutic development for GBM patients: GD3 represents a relevant tumour antigen able to unmask GBM CSC for treatment approaches while its enzyme, ST8SIA1/GD3S, appears to be also a promising therapeutic target. Therapeutic development targeting GD3 and ST8SIA1 started few years ago, mainly in the melanoma field. Different therapeutic approaches were developed, alone or in combination, including direct inhibition with monoclonal antibodies [69,70,71] or by stimulation of immune response with vaccines [71,72,73]. Anti-GD3 immunization has been developed mainly in malignant melanoma or other solid tumours [74]. The main immunotherapies developed against GD3 were an anti-idiotypic antibody, Bec2 (xenogeneic protein)-keyhole limpet hemocyanin-adjuvant (KLH). Monoclonal antibodies represent another approach to target GD3. Two main antibodies, KW-2871 (ecromeximab) and R24 were evaluated in clinical trials. The main toxicities were skin reaction and thrombocytopenia [75]. Clinical responses remained limited but the use of R24 was associated with the induction of an infiltration of immune cells (CD4 and NK) in the tumour microenvironment following the GD3 inhibition [76]. Currently, no clinical development targeting GD3 or GD3S is ongoing in neuro-oncology but promising strategies could include bi-specific antibody or CAR T-cells therapy [76, 77].

Our study has some limitations, including the number of patients samples used for the translational analyses and the low number of mice used for the evaluation of GD3 tumorigenicity, associated with the generation of small tumours in the GD3⁻ animal models. But it is important to remind that the GD3 cell sorting was imperfect, leading to the presence of 10–15% of GD3⁺ cells in the GD3⁻ cell population, potentially explaining the tumour foci found in some GD3⁻ animals. Moreover, working with fresh patient samples remains a challenge and requires a perfect coordination between hospital, clinicians, neurosurgery and laboratory, leading in our study to the analysis of more than 60 patient samples. Despite they did not represent a wide range of patients and did not perfectly represent the heterogeneity of GBM, they opened interesting perspectives in the role of GD3 in gliomagenesis. Finally, we only validated the RNA-seq results in independent tumoral cell lines and TCGA database, while mechanistical explorations would be very interesting for future studies.

Conclusions

In conclusion, we showed that GD3 is the main ganglioside of GBM and is an interesting GBM CSC biomarker. Moreover, the downregulation of its enzyme is associated with reduced tumour aggressiveness and reverses the chemo-resistance of GBM CSC. The mechanisms of ST8SIA1 implication in GBM cell stemness and aggressiveness should now be confirmed. These results open promising clinical perspectives for therapeutic development.