We searched the files of the Department of Surgical Pathology at The Rhode Island and Miriam Hospitals from January 1, 2012 through August 31, 2016 for all cases of “Glioblastoma” or “Gliosarcoma”, and retrieved 270 cases. Eligibility for inclusion required a diagnosis of Glioblastoma, IDH-wildtype, meeting the criteria of the 2021 WHO Classification of Tumors of the Central Nervous System [22]. Inclusion criteria also included the availability of clinical follow-up information, and sufficient tissue remaining in a diagnostic tissue block to create a tissue microarray. Sequential cases were reviewed until 100 cases were identified. If the original block used for diagnosis and immunohistochemistry was no longer available, a second appropriate block was chosen. Two neuropathologists (NSL, DCA) agreed on the diagnoses of glioblastoma, IDH-wildtype for all tumor samples in accordance with the most current World Health Organization classification [22].

Clinical data for these 100 patients were collected from review of the medical records, which included demographic information, clinical presentation, tumor location, tumor laterality, surgical procedure, surgical resection status (gross total resection, GTR or subtotal), tumor volume, date of MRI- or CT-documented recurrence and/or progression, date of last follow-up, date of overall survival, and treatment with radiation and/or concomitant temozolomide. Data was collected according to protocols approved by the Institutional Review Board of Rhode Island Hospital.

Pathology data was collected from whole slides, external pathology data, and surgical pathology reports. The data collected included: IDH1-R132H mutation (HistoBioTec clone H09), MGMT status (promoter methylation status of O(6)-methylguanine-DNA methyltransferase, measured by PCR after macrodissection and bisulfite treatment), p53 expression (Dako D07 clone), Ki67 expression (Dako, MIB1), 1p19q status (by FISH, when available), and morphologic findings, including any gemistocytes, spindle-cell features, small cell change, oligodendroglial morphology, and/or giant cell change.

Examination of 4-μm thick slides revealed that 98 cases had sufficient tumor remaining in the block to be included in a paraffin-embedded tissue microarray (TMA). 82 cases had three representative tumor regions only, and 16 cases had three representative tumor regions plus a “normal” or “infiltrating tumor” region. The tissue cores were evaluated by two neuropathologists (NSL, DCA) who supervised all TMA construction steps. Each core sample in the TMA was classified as “tumor”, “normal brain”, or “insufficient” (insufficient tissue for assessment, completely necrotic sample, or missing core on the TMA sections). Cores were not included in the tumor final analysis if the core was missing from the slide, severely damaged, and/or did not have a cellular tumor sample. Eight cases were excluded due to lack of any analyzable tumor in the final TMA blocks, three cases were excluded due to IDH1-mutant status (identified on review to have IDH mutation), and two cases were excluded because they were recurrent tumors that did not meet diagnostic histologic criteria of glioblastoma. Using WHO 2021 criteria, eighty-six (n = 86) cases were ultimately included in the cohort of IDH-wildtype GBM (in three of these cases, focal areas with biphasic spindle-cell morphology, sometimes referred to as gliosarcoma, were present). Seventy-two cases had MGMT promoter methylation status determined at the time of initial diagnosis.

Anti-CD133 [Thermo Cat# PA5-38,014, at 1:600 dilution], anti-SOX2 [(clone 20G5) Thermo Cat# MA1-014, at 1:600 dilution], and anti-NES [(clone 10C2) Thermo Cat# MA1-110, at 1:600 dilution] were obtained commercially and used for immunohistochemistry. Four-μm paraffin sections were cut and incubated at 60 °C for 30 min, and the sections were deparaffinized with xylene and rehydrated in graded alcohols (100%, 95%, 70%) and water. Antigen retrieval was performed in citrate buffer with a pressure cooker and microwave for 10 min. Endogenous peroxidase activity was quenched by incubating slides with Dual Endogenous Enzyme Block (Agilent Technologies, Santa Clara, CA). The sections were incubated with the primary antibodies at respective dilutions for one hour in a humidified chamber at 25C, followed by a 30 min incubation with EnVision Dual Link System-HRP (Agilent Technologies, Santa Clara, CA). Antigen–antibody complexes were visualized with peroxidase-based detection systems using diaminobenzidine (DAB) (Agilent Technologies, Santa Clara, CA) as a substrate.

Since the TMA immunohistochemistry was limited to tissue cores, we also determined the extent of variation of SOX2 expression within larger regions of GBM using immunofluorescence in frozen GBM tumor samples. Using a second set of GBM tumor samples identified in the Tumor Bank of the Rhode Island Hospital (n = 16, samples frozen at − 80 °C), a study approved by Institutional Review Board, we used immunofluorescence to localize and quantify the expression of SOX2 in each of these frozen specimens of glioblastoma, IDH-wildtype (WHO grade IV). For immunofluorescence microscopy, specimens were fixed with 4% paraformaldehyde for 24 h at 4 °C, and cryoprotected by incubation in 5% sucrose for 24 h (4 °C). Fixed cryoprotected tissue was stored at − 80 °C.

Fixed and cryoprotected GBM tissues were sectioned at 5 µm on positive-coated slides and rehydrated for 15 min in phosphate-buffered saline, pH 7.4 (PBS). The sections were then post-fixed in 4% paraformaldehyde for 15 min and rinsed in PBS. Non-specific background was blocked by incubation in blocking solution (5% goat serum, 5% donkey serum, 0.3% Triton X-100 in PBS, pH7.4) for 1 h at room temperature. The sections were incubated with anti-SOX2 rat monoclonal antibody (ThermoFisher #14–9811-82, 1:200 dilution) overnight at (+ 4 °C) followed by donkey anti-rat IgG Alexa 488 staining (ThermoFisher #A-21208, 1:1,000 dilution). Slides were mounted with ProLong Glass Antifade Mountant with NucBlue Stain (ThermoFisher #P36981) and cured overnight at room temperature. Immunofluorescence slides were examined with confocal and conventional fluorescence microscopy, using GFP and DAPI filters. For each case, three fields were photographed at 40 × and 120x (60 × 2) magnification, and the images were used for quantitation. The total number of nuclei was identified with NucBlue staining, and total number of cells counted. Individual cells were identified as negative (NucBlue staining only), or positive (SOX2-expressing; Alexa green 488) determined by the presence of a strong nuclear expression of SOX2.

Semi-quantitative assessment of the immunohistochemistry results was performed for CD133, SOX2, and NES. For CD133, cytoplasmic and membranous circumferential staining was considered positive. The staining intensity was graded as none (0), weak (1 +), moderate (2 +), or strong (3 +) and was multiplied by the percentage of positive cells, ranging from 0 to 300 (3 × 100%) as the raw score. Four categories were defined as follows from the raw scores: 0 = 0, 1–100 = 1, 101–200 = 2, and 201–300 = 3. For final analysis, CD133 was divided into two categories: scores of 0 or 1 was considered low expression, and scores of 2 or 3 was considered high expression.

SOX2 nuclear staining was considered positive. The staining intensity was graded as none (0), weak (1 +), moderate (2 +), or strong (3 +) and was multiplied by the percentage of positive cells to create a raw score. Four categories were defined as follows from the raw scores: 0 = 0, 1–100 = 1, 101–200 = 2, and 201–300 = 3. Scores of 0 and 1 were considered low SOX2 expression, and scores of 2 and 3 were considered high SOX2 expression.

NES cytoplasmic staining was considered positive. NES, which is expressed in both tumor cells and in neovascularization, was assessed for whether the expression was in tumor cells or in blood vessels in areas of neovascularization. Tumor cell expression intensity was graded as none (0), weak (1 +), moderate (2 +), or strong (3 +) and was multiplied by the percentage of positive cells to create a raw score. Four categories were defined as follows from the raw scores: 0 = 0, 1–100 = 1, 101–200 = 2, and 201–300 = 3. Blood vessel expression of NES was recorded as positive or negative, but was not used in scoring of glioma stem cell expression.

Analysis of bulk tumor RNA-sequencing was performed on 169 GBM tumor samples and 5 normal control samples obtained from The Cancer Genome Atlas Project (TCGA). HTSeq-FPKM files, clinical, and metadata were downloaded using the GDC Data Transfer Tool and GDC data portal from the National Cancer Institute per the data manifest. A log2(FPKM + 1) transformation was then performed on all FPKM values and normal sample controls were removed. GBM tumor samples were then separated based on the expression of individual genes at the median: separating high SOX2 expressing tumors (above the median) from low SOX2 expressing tumors (below the median), and high PROM1 expressing tumors (above the median) from low PROM1 expressing tumors (below the median). Survival analysis (Log-rank (Mantel-Cox) test) was then performed, where tumors were classified based on high or low expression of PROM1 or SOX2 alone and in combination with MGMT methylation status. MGMT methylation status was mapped to samples by matching the submitter IDs to case IDs where data was available from Brennan et al. [11]. For two-gene correlations, the Betastasis two-gene scatterplot tool (Affymetrix HT HG U133A) and the glioblastoma Rembrandt (GEO GSE108476) dataset were used, with log-2 (FPKM + 1) transformation.

SPSS v22 for MacOS (SPSS, Chicago, IL, USA) was used for all statistical analysis. P < 0.05 was considered statistically significant. Overall and progression free survival were estimated by Kaplan–Meier analysis and were compared using the two-sided log-rank test. All tests were two-sided.