Ex-vivo drug screening of surgically resected glioma stem cells to replace murine avatars and provide personalise cancer therapy for glioblastoma patients

Materials

Maintenance of GSC culture conditions

Preparation of GSC ‘stem’ culture media:

Here we describe the procedure used to prepare 500 mL of GSC media used to maintain stem-like cell populations. Reagents used are listed in Table 1.

Preparation of GSC ‘bulk’ culture media:

Here we describe the procedure used to prepare 500 mL of differentiating ‘bulk’ media, the added serum induces differentiation of GSC cultures.

Protocol for coating plasticware with basement membrane:

Here we describe the procedures used to coat plasticware with basement membrane extract (BME) (Cultrex), this allows GSC populations to grow as adherent monolayers. Concentrations and volumes used are listed in Table 2. To avoid any batch-to-batch variation of this product, sufficient volumes were purchased to cover the entirety of the project.

Protocol for coating 384-well drug plates with basement membrane:

Here we describe the procedures used to coat pre-treated 384 drug plates with basement membrane extract (BME; Cultrex™).

Note – BME 1:20 dilution used for drug plate coating to account for dilution when added the wells pre-coated with 5 μl drug.

Drug plate printing

Here describes the protocol used for the 384-well drug plate printing. Drug compounds can be purchased as either pre diluted liquids or solids and diluted in DMSO (or water for platinum-based compounds such as cisplatin) to stock working concentrations of 5-10 mM and stored in -80°C. All compounds used in this study are listed in Table 3 with their concentrations.

Figure 1. Master 96-well plate layouts.

Initially, compounds were diluted to ××10 the top concentration to be used and added to their corresponding wells on the ×10 master plate. Drugs from this plate were then serially diluted down 1:2 into ×5, ×2.5 and ×1.25 master plates. Vehicle control (DMSO) and positive controls (Staurosporine and aphidicoline) were added to the plate independently at single concentrations.

Note – Row A and column 1 are left empty as these will correspond to the outer wells of the final 384 drug plate, which are left empty due to plate evaporation effects.

Patient recruitment and sample transfer

Fresh treatment-naïve glioblastomas were collected from patients who provided informed consent undergoing surgery at Sheffield Teaching Hospitals NHS Foundation Trust (Ethical approval: Yorkshire & The Humber – Leeds East REC (11-YH-0319/STH15598). Fresh glioblastoma tissue surplus to histological requirements resected by the operating surgeon was collected intraoperatively whilst surgery was ongoing, pseudonymised/assigned a unique sample identifier and then rapidly transferred to the ex vivo laboratory within <30 minutes within a dry sterile specimen pot (NHS) at room temperature. Tumour samples were taken from both male and female patients across a range of ages within the parameters associated with the prevalence for this disease and the local gender/age consenting at Sheffield’s Royal Hallamshire Hospital and anonymised to the ex vivo screening team.

GliExP sample processing and culture

Here we describe the protocol used for the dissociation and seeding of tissue specimens onto pre-coated 384-well drug plates.

Tip – If the tissue homogenate is too large to be picked up with Pasteur pipette, repeat mechanical dissociation. Be careful when using pipette as tissue can suck up into bulb.

Tip – It is possible to derive smaller samples with 1-2 ~0.5 mL pellets, however samples screened in this study are usually in excess, therefore we find it important to dissociate the majority of tissue to give accurate representation of cells in entire the tumour.

Note – Viability was typically between 20-40% for these methods.

Note – End users should test different incubation periods to optimise appropriate incubation times for their specific needs/cell populations.

Here we describe the protocol used for the culturing of GliExP samples for relevant follow up experiments:

Note – for the purpose of this paper unless the ‘p’ prefix is used, the GSCs are derived from matched dissociated tissue. Furthermore, although well-established in our laboratory, to confirm that our GSC growth media appropriately maintained GSC niches within the 384-well format, we compared known GSC marker expression in matched GSC CX18 cell line populations grown in bulk and stem enrichment media and that GSC plasticity was maintained (Figures 2 and 3 respectively).

Figure 2. RT-PCR analysis of mRNA expression of stem cell markers nestin, SOX2 and CD133 within two model growth conditions ‘stem’ and ‘bulk’.

Expression levels of all three GSC markers mRNA were higher within passage matched GSC populations compared to their bulk equivalent cells. Data represents the 25^th and 75^th percent quartiles (box) median (middle line). The lower and upper whiskers show the smallest and largest values within 1.5x of the interquartile range below the respective 25^th and 75^th percentile. Data shown is derived from two independent biological repeats from separate passages of the same patient derived cell line, with each repeat including three technical replicates for each marker condition. Statistical significance was calculated using the Mann-Whitney-U test (**P<0.002).

Figure 3. Western blot detecting the expression of stem cell markers nestin and Sox2 within undifferentiated (primary stem), differentiated (bulk) and de-differentiated (secondary stem) cultured CX18 GSCs.

Serum differentiated CX18 GSCs show the ability to revert to a de-differentiated state re-expressing stem markers SOX2 and Nestin, when bulk cells are cultured back in stem growth conditions, thus demonstrating and confirming GSC plasticity.

Immunofluorescent staining

Staining of drug plates

Here we describe the protocol used for the immunofluorescent staining of 384-well drug plates for microscopy imaging to evaluate treatment response. These fixing, washing and staining steps have been optimised to reduce the number of tumour cells being washed away.

Tip – At this stage it is possible to add another 30μL of PBS and store the plate at 4°C for up to 48hrs or proceed directly with staining steps.

Tip – where possible implement conjugated antibodies to reduce washing steps and retain cell population on plates.

Tip – If not proceeding with microscopy immediately, plates can be wrapped in foil and stored in the fridge at 4°C. It is recommended to image the plates within a week of staining to avoid degradation of antibodies.

Optimisation of GSC markers by immunofluorescence

Glioblastoma stem and bulk cells from early passage primary tumour samples (Rominiyi et al., under revision) were seeded at a density of 3×10⁵ cells per well onto BME-coated coverslips inside a six-well tissue culture dish. Cells were then incubated at 37°C in 5% CO₂ for 48 hours without treatment. The cells were then fixed and permeabilised using cold 4% PFA and 0.6% Triton X-100 (500 μL) for 10 minutes, followed by three PBS washes (5 minutes each). Cells were then blocked for 1 hour at room temperature in PBS containing BSA (3%). Conjugated and primary antibodies used were diluted in PBS containing Tween (0.05%) and bovine serum albumin (3%), 500μl of required antibody was added to each well at appropriate concentration (Table 4) and incubated overnight at 4°C in the dark. The following day the primary/conjugated antibodies were then aspirated before 3 additional washes with PBS. Next 500 μL of DAPI and/or secondary antibody was added for one hour at room temp on a gentle shaking platform in the dark to prevent photobleaching, the antibodies were aspirated and washed with PBS (5 minutes each). Following the final wash, coverslips were mounted onto labelled microscope slides using ProLong Gold Antifade Mountant (Thermo Scientific). Slides were then stored in the dark to drug overnight, before being stored in the fridge at 4°C. Antibodies for GSC markers were CD133, Nestin and SOX2, with Vimentin used as a marker of differentiation and DAPI used to identify cell nuclei.

Microscopy imaging

Drug plate imaging

Here describes the protocol used for the microscopy imaging of sample drug plates, all plates in this study were imaged using a Cell Discoverer 7 microscope using Zen Blue 3.1 software.

Coverslip imaging

Imaging of microscope slides was performed on an LSM-980 confocal microscope all images were taken using the Photometrics sCMOS Prime BSI imaging camera using 40x objective. Images were captured using the Zeiss Blue (3.0) software package with all settings (laser power, digital gain, digital offset) kept constant between experimental conditions and repeated measures.

Image analysis

All image analysis was performed using Zeiss Zen Blue 3.5 analysis software. Using the processing tab, an image subset from the initial image file was extracted selecting approximately 30 individual tiles, which included a selection of wells from positive, Staurosporine and negative DMSO control wells along with any wells containing significant debris or artifacts. This subset was then selected to train the software using the Zen Intellesis machine learning module, which allows segmentation of images based on the user training distinguishing between object and background. Objects (DAPI positive nuclei) were manually labelled using the labelling options tab with the brush option, background including artifacts were labelled the same way. The setting dropdown box was set to 50 features, as this enables Zen Intellesis machine learning module to full access the computer GPU, reducing the time taken to fully analyse the images (Figure 4). The supervised learning had to be done independently for each sample as every sample is very heterogeneous, therefore the model could not be transferred between different GBM samples. Once the Intellesis training model was completed a sample specific image analysis program was then created.

Figure 4. Example of how the Zen Blue software images were obtained using the Cell Discoverer 7 10x magnification.

Primary glioblastoma positively DAPI stained nuclei were identified as objects by the algorithm for scoring purposes (yellow outlines).

RStudio analysis

Plate layout templates were designed which contained compound name, concentration and well number it corresponded to. Using RStudio software version 4.2.2 the output CSV files that were produced by Zen Blue were initially concatenated using the summarise function by dplyr to obtain a DAPI frequency count per well. This file was then merged with the plate layout template by well number, so that each well corresponded to DAPI frequency, drug and concentration. Any data visualisation drug response plots were produced using the ggplot2 package. In order to obtain cell metrics such as LD₅₀ values and AUC, a package called GRmetrics was used. GRmetrics files for each sample was then imported back into RStudio to generate a heatmap representing AUC values. This was performed using the pheatmap package.

Quantitative PCR

For RNA extraction, Bulk/stem cells were seeded into Cultrex coated 6-well plates and incubated until 70% confluent before harvesting. Qiagen RNeasy Mini Kits were used to extract RNA. Initially, media was aspirated from cultured cells and cells were washed twice with PBS. PBS was then aspirated and 350 μL RLT buffer was added to each well and cells were then dislodged using a cell scraper and transferred to labelled QIAshredder columns before being centrifuged at 8000 RCF for 2 minutes. The purple columns were discarded and 350 μl of 70% ethanol was added to the supernatant, this solution was then transferred to a RNeasy Minispin column and centrifuged for 15 seconds. The flow through was discarded and 700 μl RW1 buffer was added to each column and centrifuged for another 15 seconds at 800 RCF. This process was then repeated again twice but with 500 μl RPE buffer, the first instance centrifuged for 15 seconds and the second for 2 minutes, discarding the flow through for each step. Excess ethanol was then removed by centrifuging the column for 1 minute, before adding 50 μl of RNase free water and centrifuging again for another 1 minute to elute the RNA. Total RNA was then quantified for each sample using Nanodrop 200 spectrophotometer, which quantifies absorbance at wavelength at 260 nm to establish the concentration of RNA in each sample, and the ratio of sample absorbance at 260/280 was used to confirm purity (absence of contaminants such as protein or phenol, which absorb strongly at/near 280 nm) with ratios of ~2 deemed to represent ‘pure’ RNA. After quantification RNA samples were either used immediately or stored at -80°C. RNA samples were reverse transcribed using High-Capacity RNA-to-DNA Kit (Applied Biosystems). The RT reaction was prepared on ice in order to obtain a 30 μl sample of cDNA for downstream qPCR. Volumes of sample equivalent to 1μg of RNA were added to reagents in the kit after 15 μl buffer and 3 μl of enzyme and made up to 30 μL using RNase-free water. Samples were then reverse transcribed using a Biorad thermal cycler T100 using cycling parameters 37°C for 1 hour, 95°C 5 minutes to convert RNA into cDNA. Each resulting cDNA sample was analysed in triplicate for each individual probe within a 384-well PCR plate with GAPDH used as a control ‘housekeeping’ gene for each sample. Each reaction consisted of: 2 μL cDNA, 5 μL TaqMan Universal PCR Mastermix, 2.5 μL ddH₂O and 0.5 μL of probes (provided in Table 5). The plate was sealed using optical adhesive film (MicroAmp) and loaded onto a 7900HT Fast Real-Time PCR System to perform quantitative PCR. The system was set to report 18S FAM with repeats of 40 cycles. Cycling conditions involved an initial 95°C hold for 10 minutes followed by 40 cycles of 15 second 95°C denature step and finally 60°C for 1 minute in order to anneal and extend. Double delta Ct (2-ΔΔCt) analysis was used to determine relative gene expression using an average Ct value from the triplicate runs. For example, to detect differences in expression between primary, patient derived glioblastoma stem cells grown in stem conditions and bulk differentiated cells: ΔCt = Mean Control (GAPDH) Probe Ct – Mean Gene Probe Ct, and ΔCt Expression = 2-ΔCt. Then, to calculate fold-change expression in GSCs relative to bulk cells: ΔΔCt = ΔCt Expression (Stem)/ΔCt Expression (Bulk).

Immunoblotting

For immunoblotting glioblastoma (bulk/stem) cells were seeded into Cultrex-coated 6-well plates and incubated at 37°C in 5% CO₂ for 48 hours without any treatments. Before media removal, cells were washed twice in ice-cold PBS. Cells were then lysed with the addition of 100 μL lysis buffer (20 mM Tris-HCl pH 7.5, 150 mM NaCl, 1% Triton X-100, 1 mM DTT and 1 mM EDTA supplemented with 50 U/μL benzonase (Novagen), protease and phosphatase inhibitors (Sigma). Cells were immediately harvested using a cell scraper and transferred to labelled Eppendorf tubes on ice. Each sample was then vortexed for 10 seconds before being placed on ice for 15 minutes. Samples were vortexed again following a final 15-minute incubation on ice and one final vortex. Samples were then centrifuged at 14,000 × g for 15 min at 4°C. Gel electrophoresis was performed using NuPage system (Invitrogen). Samples were resolved on 4-12% Bis-Tris gels in MOPS buffer, transferred to Protran nitrocellulose membranes (0.1 μm pore size), which were then probed for proteins of interest using antibodies diluted in 5% BSA (details of antibodies shown in Table 6 below).

Statistical analyses

Shapiro-Wilk normality tests were performed to assess the datasets prior to statistical analysis to decide the correct tests to perform. In order to confirm differences between two groups Mann-Whiney U-test was performed on non-parametric datasets or Students t-test on parametric datasets, p-values less than 0.05 would determine whether the differences in results were significantly different. All normality tests and statistical analysis was performed using R package stats version 4.2.2. As the current study is not comparing ex vivo drug responses to any clinical indications, normal power calculation or simulation studies were not used to determine the sample size at this time.

Optimisation of plating order and SoC therapy responses to support a high-throughput drug screening platform for primary GSCs

Glioblastoma cultures grow as neurospheres unless a BME is provided, giving structural support for the cells, thus providing a dense layer of extracellular matrix (ECM) to allow them grow as adherent monolayers. For the purpose of high throughput screening, monolayer growth is attractive as it will facilitate a rapid examination of cell-drug interactions using automated high-content microscopy. However, the requirement to pre-coat plates with a basement membrane that can only be stored for a couple of weeks once used, introduces a level of complexity compared with standard 2D established cell line-based monolayer drug screening. To keep the process as streamlined as possible, drug plates would ideally be pre-printed, stored at -80°C, then on the day of surgery the plates thawed and coated in ECM ready for GSC sample seeding directly from the clinic. Using the glioblastoma SoC drug Temozolomide (TMZ) we therefore determined if the order by which the ECM, compounds and cells affected drug efficacy (Figure 5), using primary GSCs derived from a patient tumour. The LD₅₀ values for TMZ in Conditions 1 and 2 were consistent within both the four-day and eight-day incubation periods, highlighting that when TMZ was seeded underneath the ECM it still has the same efficacy as if it was seeded above. However, within both the four-day and eight-day incubation Condition 2 showed much higher variation for percentage inhibition within repeats, suggesting this method of drugging may lead to discrepancies between repeated measures. Additionally, Condition 3 showed an increased LD₅₀ compared to the other 2 conditions, which may be a consequence of higher cell adherence and cell cycling, by allowing the cells to adhere to basement membrane prior to drugging. Despite Condition 3 being the most representative model of patient tumour treatment, for logistical reasons described above around the critical need for ex vivo screening of clinical samples with short notice requiring the use of pre-printed drug plates, we therefore proceeded to adopt condition 1 for future GliExP development.

Figure 5. Optimisation of plating conditions for high-throughput GSC drug screening.

A: Diagrammatic representation of the three different drug, cell and ECM alternations tested to ensure drug efficacy within the screening platform. B: TMZ dose response curves for CX18 GSCs following 4-day drug incubation using the different plating conditions. C: Same as (B) but following an 8-day TMZ incubation/growth period. For both datasets, mean data derived from 3 independent biological repeat experiments is shown with their respective SDs. Condition 1 average LD₅₀ = 18.95 ± 3.45 μM, condition 2; 16.28 ± 6.22 μM, and condition 3; 56.60 ± 50.15 μM for 4 days TMZ incubation/growth. For 8 days, Condition 1 average LD₅₀ = 9.61 ± 4.9 μM, condition 2; 6.07 ± 0.87 μM, and condition; 23.04 ± 4.04 μM.

To further confirm that our ex vivo screening platform was able to give robust and accurate readouts of TMZ sensitivity, we tested its ability to predict MGMT promoter methylation status; a well-established clinical feature of around 40-50% of all glioblastomas that epigenetically silences the expression of the key O6-methylguanine detoxifying enzyme MGMT.³⁷

TMZ sensitivity was therefore assessed in an MGMT+ (unmethylated) GSC (Ox5) and a clinically determined MGMT methylated GSC (CX18). Note that CX18 has been deem as “equivocal” methylation status based on 10.7% exhibiting methylation across all CpG islands analysed (clinical data not shown), and therefore could be considered on the weaker side of MGMT methylation and subsequent TMZ sensitivity. However, GliExP was able to distinguish significant TMZ sensitivity between OX5 and CX18 GSC which correlated with their MGMT methylation status (Figure 6A). Importantly, when analysis of further primary GSCs was carried out before clinical MGMT status was known, the resulting TMZ sensitivity allowed us to predict with 100% accuracy the MGMT status (Figure 2B).

Figure 6. Discrimination of MGMT promoter methylation status within primary GSCs using temozolomide sensitivity within GliExP.

A: Comparing the temozolomide (TMZ) LD₅₀ values of OX5 (MGMT+/unmethylated) and CX18 (MGMT-/methylated) GSCs with differential MGMT methylation as defined by clinical methylation sensitive pyrosequencing of the MGMT gene promoter in parental tumour tissue. MGMT methylated CX18 GSCs exhibited a significant increased sensitivity to TMZ. Data shown represent median and the 25th and 75th percent quartiles (box) median (middle line). The lower and upper whiskers show the smallest and largest values within 1.5x of the interquartile range below the respective 25th and 75th percentiles. Median LD₅₀ are 46.13 ± 60.5 μM (standard deviation) compared to MGMT unmethylated GSCs; LD₅₀ = 385.72 ± 102.5 μM (standard deviation). Data shown is derived from three independent biological repeats from two different patient derived GSC models from separate passages (each containing four technical replicates); CX18 and OX5. B: Heatmap showing AUC responses to TMZ alone or in combination with 2Gy IR for 6 different tumours assessed by GliExP. Also shown is their MGMT methylation status which was subsequently determined clinically. GliExP was able to accurately predict the MGMT status for all these tumours based solely on TMZ sensitivity prior to MGMT status being known.

In addition to TMZ, the current post-surgical SoC regimen for glioblastomas involves a course of radiotherapy (typically 60Gy over six weeks in 2Gy fractions).¹⁶ We are fortunate, and in the minority of research laboratories, in that we have access to an experimental ionising radiation (IR) source within our research facility (~2Gy/min ¹³⁷Cs irradiator; CIS Bio International IBL437c). However, many places that may wish to adopt an ex vivo drug screening platform as an alternative to animal-based preclinical models may not have such facilities. We therefore assessed the potential use of the radiomimetic agent bleomycin within GliExP as this can simply be added in a similar manner to small molecule compounds by comparing a dose range against known ionising radiation doses within our experimental Cs¹³⁷ facility (Figure 7). For future use and for the benefit of groups that do not have access to a radiation source, we calculated that a concentration of around 6.85μM Bleomycin was equivalent to a 2Gy dose of IR (Figure 7).

Figure 7. Dose response curves for CX18 GSCs treated with either IR or Bleomycin.

A: Cell frequency measured using DAPI positive cells. B: Calculated percentage growth inhibition derived from the same data. The response readout (DAPI count) was collected using automated IF microscopy, the counts were normalized using positive and negative controls on each dose plate to provide the response measure (relative inhibition %). Bleomycin LD₅₀ = 6.85 μM which is equivalent to LD₅₀ IR – 2Gy. Data shown is the mean derived from 3 independent biological repeats with respective SDs.

As the current SoC for glioblastomas involves a course of radiotherapy alongside TMZ for patients deemed well enough to receive this, we combined IR with TMZ within GliExP. Encouragingly, we were able to determine the combined cytotoxic effects of IR on top of concomitant TMZ across all primary GSC tested to date (Figure 8).

Figure 8. Assessment of combination IR with TMZ treatments in GliExP.

A: Box plot comparing the median and interquartile AUC values calculated by GliExP for the indicated GSCs treated with TMZ ± 2Gy IR. Samples shown in green represent methylated MGMT status while red represents unmethylated. B: Line graphs for the same data showing % growth inhibition. All data shown is derived from 3 independent biological repeats (with 4 technical replicates from the same patient derived cell line) with their respective SDs.

Proof-of-concept therapeutic and preclinical compound screening using GliExP

Given the encouraging development and optimisation data around GliExP (see Figures 4-8), we proceeded to carry out an initial proof-of-concept drug screening. This consisted of creating an initial pre-printed drug plate consisting of just over 30 therapeutic and pre-clinical compounds. The dose ranges for the initial batch of pre-printed GliExP plates we based on existing literature around LD₅₀ values across a range of cell lineages, including data from our own laboratories and consisted of four concentrations + a vehicle control for each compound. Using this batch of plates, we proceeded to screen GSCs derived from 18 individual patients taken straight from surgery, including a couple of multi-region samples to assess potential differential drug response associated with spatial heterogeneity, a known driver of therapeutic resistance in the clinic (Figure 5).

This initial set of encouraging data demonstrates that GliExP is able identify potential differential drug sensitivity for a given individual tumour using freshly derived GSC populations from surgically resected tumour within a week of surgery taking place. The limited number of multi-region samples on this initial screening plate also suggests that GliExP is able to determine differential drug sensitivity within a given tumour, further highlighting how spatial heterogeneity can lead to therapeutic resistance. It is important to note that the perceived universal inhibitory effects of the EGFR inhibitor are likely an artifact of the EGF-conditioned media used to enrich/grow GSCs. Encouragingly, we observe that some GSCs are sensitive to a number of compounds with related targets, such as GBM21 being sensitive to both WEE1 and Aurora kinase inhibitors, and GBM19 being sensitive to both PARP1 and PARG inhibitors (Figure 9). However, moving forward, longer incubation times and extended concentration ranges would be beneficial for certain compounds to allow a deeper assessment of inhibitory effects and calculation of more accurate LD₅₀ values (see Discussion section for further details).

Figure 9. GliExP heatmap depicting raw AUC drug response for primary GBMs dissociated from 18 individual tumours within a week of surgery.

Scale shown represents cell death as 0 (blue) and cell survival as 1 (red). Larger samples such as GBM10 (A, B and C) and GBM20 (A and B; indicated in bold) were sectioned into multiple regions and screened independently in order to highlight the intertumoral heterogeneity of the disease. AUC values were calculated from 2 technical replicates following 96hr drug incubations using R package GR metrics and the heatmap was constructed using R package pheatmap as detailed in the methods section. Note, due to the fact that this was performed on limited fresh clinical material, biological repeats were not performed.

Finally, in order to further develop GliExP to be able to identify potential GSC-specific compound cytotoxicity within the cell milieu, we have carried out initial staining optimisation of the GSC markers Sox2, Nestin and CD133 using direct fluorescent conjugated antibodies. To determine GSC specificity of the staining, we cultured CX18 cells as either tumour bulk or GSC-enriched populations (see methods section) prior to staining (Figure 10). Across multiple optimisation experiments, we found that the most robust of these GSC markers within GliExP were Sox2 and Nestin (Figure 10A and 10B), and that these correlate with mRNA expression levels (Figure 2). It is interesting that in our hands, CD133 was not as robust a GSC marker as one would have predicted based on mRNA expression data (Figure 2). However, there is evidence within previous studies that CD133-negative cells can be labelled incorrectly due to specific antibodies only targeting the AC133 specific epitope, which is located in one of the extracellular domains of membrane-bound CD133 (Barrantes-Freer et al., 2015). We therefore would need to test additional CD133 antibodies within our platform to ascertain if this is indeed the case. However, the robust ability of both Sox2 and Nestin to distinguish GSC-enriched populations gives us confidence that these could be implemented in future iterations of GliExP.

Figure 10. Initial optimisation of GSC marker expression within GliExP to determine potential GSC-specific drug efficacy.

A: Left panel; violin plots showing mean fluorescence intensity of Sox2 expression (pixels/cell) in both bulk and stem enriched CX18 GSCs. Right panel; example images. B: and C: Same as in A but showing data for Nestin and CD133 respectively. Data shown are means derived from three independent biological repeat experiments from separate passages of matched ‘bulk’ and ‘stem’ derived CX18 cells with their respective SDs. Each biological repeat experient contained analysis of 100 individual cells. Statistical analysis was performed using Man-Whitney U test to highlight significant differences GSC marker staining intensity between bulk and stem populations (p < 0.001).