Radioresistance of mesenchymal glioblastoma initiating cells correlates with patient outcome and is associated with activation of inflammatory program

Glioblastoma (GBM) still remains an incurable disease being radiotherapy (RT) the mainstay treatment. Glioblastoma intra-tumoral heterogeneity and Glioblastoma-Initiating Cells (GICs) challenge the design of effective therapies. We investigated GICs and non-GICs response to RT in a paired in-vitro model and addressed molecular programs activated in GICs after RT. Established GICs heterogeneously expressed several GICs markers and displayed a mesenchymal signature. Upon fractionated RT, GICs reported higher radioresistance compared to non-GICs and showed lower α- and β-values, according to the Linear Quadratic Model interpretation of the survival curves. Moreover, a significant correlation was observed between GICs radiosensitivity and patient disease-free survival. Transcriptome analysis of GICs after acquisition of a radioresistant phenotype reported significant activation of Proneural-to-Mesenchymal transition (PMT) and pro-inflammatory pathways, being STAT3 and IL6 the major players. Our findings support a leading role of mesenchymal GICs in defining patient response to RT and provide the grounds for targeted therapies based on the blockade of inflammatory pathways to overcome GBM radioresistance.


Supplementary Figure 2: Characterization of stem functional features of established GICs cultures
(A) Representative picture of Soft Agar Assay carried out with PG88-DGC and PG88-GICs. Cell seeded for each well is reported: 1000 and 2000 cell/well in PG88s; 1000, 2000, 5000 and 20000 cell/well in PG88. DGC cells had to be seeded 10-times more concentrated than neurospheres to appreciate real colony growth, and still the colonies generated by neurosphere cultures were considerably bigger.
(B) Self-renewal capability of neurosphere cultures. Data are plotted as percentage mean of cells capable of growing as spheres bigger than 100 μm in diameter after 14 days (n = 3 independent experiment).
(D) Hematoxylin/eosin stain and immunohistochemistry for GFAP, CD44 and Vimentin. Tumor area generated in nude mouse brain following injection with PG82-GICs or PG90-GICs and the corresponding parental tumors were analyzed. Scale bar, 100 µm. Expression levels of the different proteins analyzed in mice and original tumors were found to be consistent.

Supplementary Figure 3: Comparative detection of CSCs markers expression in established cultures
(A) Real-time PCR analysis of CSCs markers L1CAM, ITGA6 and CD44. R-fold was obtained from data processed according to the ΔΔCt method, using corresponding DGC expression level as internal control (n = 3 with unpaired t-test, ** P < .01, *** P < .001).
(C) A representative picture of immunofluorescence analysis of selected GICs markers. Nuclei are counterstained with Draq5. Confocal images were taken with same settings between primary GICs and monolayer cultures in order to allow for effective comparison of fluorescence intensity. Scale bar, 50 μm.

Supplementary Figure 4: GSEA gene sets positively and negatively enriched after the first IR cycle
List of gene sets significantly enriched after IR in PG35-DGC and PG35-GICs reported following GSEA nomenclature (FDR< .05). Gene sets were obtained by interrogating three different databases (Reactome, KEGG and BioCarta). FDR, false discovery rate; NES, normalized enrichment score.

Supplementary Figure 5: GSEA gene sets positively enriched after the radioresistant switch
List of gene sets significantly enriched after double-IR in PG35-GICs-R reported following GSEA nomenclature (FDR< .05). Gene sets were obtained by interrogating three different databases (Reactome, KEGG and BioCarta). FDR, false discovery rate; NES, normalized enrichment score. (B) Box-plot diagram showing the expression of the indicated genes across the different molecular subtypes of GBM. Data were obtained from the TCGA array platform (AgilentG4502A_07_2) extracted from the TCGA cancer browser (* P < .05, ** P < .01, *** P < .001 with ANOVA in comparison to expression in Mes subtype). Among gene composing the panel PTGS2, LIF, ICAM1, IL6, CTGF and NNMT reported a significantly higher expression in Mes subtype.
(C) Kaplan-Meier curves showing progression-free survival (left) and overall survival (right) of GBM patients with either no expression alteration or upregulation of PTGS2, LIF, ICAM1, IL6, CTGF and NNMT. According to cBioPortal default settings, median PFS was 4.07 months for patients with high expression of the gene set, whereas it was 7.62 months for those with low expression (P < .001). In addition, median OS was 11.83 and 14.62 months for patients with high and low expression, respectively (P < .05).

Table S1. List of genes differentially expressed in cluster 3 versus cluster 2
Genes significantly upregulated or downregulated following double-IR in PG35-GICs-R showing FDR< .05. Indicated nominal p-value was calculated with two-tailed unpaired t-test.

Isolation of GICs from parental tumors
Samples were first rinsed with Hank's Balanced Salt Solution (HBSS), then mechanically dissociated and enzymatically digested with 20 U/ml Papain (Worthington) stabilized with 8.25 M L-Cysteine (Sigma) and 3.42 M EDTA (Panreac) for 30 min at 37ºC with constant agitation, and then cultured in FBS-free media [1]. Neurosphere primary cell cultures were subsequently split 1:10 every 5-7 days by mechanical dissociation and extensively maintained for at least twenty passages. All cultures were maintained at 37°C in a humidified atmosphere with 5% carbon dioxide and were tested for mycoplasma routinely. All experiments were performed before passage 20.

Methylation-specific PCR
Genomic DNA was isolated from frozen tumor using the Qiamp DNA mini kit (Quiagen). DNA methylation status of CpG islands of the enzyme O6-methylguanine methyltransferase (MGMT) promoter was determined by methylation-specific PCR (MSP) as previously described [5].

Aberrant genomic copy number detection
Copy numbers of EGFR, TP53 and PTEN were detected through Multiplex Ligation-dependent Probe Amplification (SALSA MLPA Kit P105, MLPA®, Mrc-Holland, The Netherlands). Analysis was carried out on patient tumor tissues (GBT), and primary DGC and GICs cultures. DNA was extracted with Qiamp DNA mini kit (Quiagen), and MLPA was executed following the manufacturer's instructions. MLPA profiles of primary cultures and tumor samples were compared to genomic DNA obtained from healthy donors. Peak areas were analyzed with Coffalyser software (MRC Holland, The Netherlands) considering as normal variation values those falling between 0.8 and 1.2. Ratios below 0.8 were considered as locus deletions. Ratios above 1.2 were considered as amplifications.

Self-renewal assay
For the self-renewal assay, neurospheres were mechanically dissociated and seeded at extremely low cell density (1 cell per well) in a 96-well flat-bottomed plate. After 24 hours, plates were visually scanned with inverted light microscope to select wells containing a single cell. Colony formation was recorded 14 days later and the percentage of growth as neurosphere was determined. Only neurospheres exceeding 100 µm in diameter were counted (ProgRes CapturePro).

Intracranial tumor assay
All mouse experiments were approved by and performed according to the guidelines of the IDIBELL Animal Care Committee in agreement with the European Union and national directives. 1x10 5 GICs were inoculated approximately into the corpus striatum of the right brain hemisphere of 7-week-old male athymic mice (Harlan). After 7 weeks mice were euthanatized, and brains were collected, formalin fixed and paraffin embedded. Immunohistochemistry was performed on 4 μm slices using the following primary antibodies: anti-GFAP (clone 6F2; Dako), anti-vimentin (clone 3B4; Dako) and anti-CD44 (clone 156-3C11; Thermo). Samples were counterstained with hematoxylin. Original patient tumor samples underwent the same procedures.