The dystroglycan receptor maintains glioma stem cells in the vascular niche

Total cellular RNA was isolated from tissue or cell lines using TRIzol reagent (Thermo Scientific). RNA was DNase treated using RQ1 RNase-Free DNase (Promega), then first strand cDNA was synthesised using random hexamers (Random Primer 6, New England BioLabs) SuperScript III Reverse Transcriptase (Thermo Scientific), and dNTPs (Promega). Real-time PCR was performed using a Viia 7 Real-Time PCR System and SYBR-Green PCR Master Mix (both Thermo Scientific). Results were normalised to β-actin (ACTB), with expression represented as copy number per 1000 β-actin. Sequences for gene-specific primer sets were as in Table 1:

Total cellular protein was isolated using 8 M Urea with phosphatase and protease inhibitors. Membrane-and-cytoplasmic protein fractions was isolated using a Mem-PER Plus Membrane Protein Extraction kit (Life Technologies). Equal amounts of protein were loaded, separated by SDS/PAGE and immunoblotted using standard methods. For Western blotting, we used αDG (Millipore IIH6C4, 1:1000), EphA2 (CST 6997, 1:1,000) and EphA3 (Invitrogen, 373,200, 1:1,000), ERK1/2 (CST 4695, 1:1,000), and p-ERK1/2 (CST 4370, 1:1,000). β-Actin was used as a loading control (CST 3700, 1:2,000).

Immunohistochemistry experiments were performed using OCT-embedded frozen specimens from patient brain tumours. 6 μm sections were cut, fixed in a solution of 2:1 acetone:ethanol for 5 min, incubated in 3% hydrogen peroxide (Chem-supply) for 10 min to block endogenous peroxidase activity, and incubated in Background Sniper (Biocare Medical) for 30 min to block non-specific binding, then incubated with the following primary antibodies at 4 °C overnight: Vimentin, clone V9 (DAKO M0725, 1:400), 1F7 (in-house anti-EphA2, 1:75), IIIA4 (in-house anti-EphA3, 1:50), α-dystroglycan/LARGE-glycan, clone VIA4-1 (Merck Millipore 05-298, 1:75), β-dystroglycan (B-4) (Santa Cruz Biotechnology sc-165997, 1:100), CD31 (M-20) (Santa Cruz Biotechnology sc-1506 1:50), GFAP, clone GA-5 (Biocare Medical CM 065, 1:500), CDW49f, clone 4F10 (Merck Millipore BCL458, 1:25), Ki-67 (Abcam ab66155, 1:300), Caspase-3 (Biocare Medical CP 229, 1:80), Phospho-p44/42 MAPK (Erk1/2) (Thr202/Tyr204) (Cell Signalling Technology 4370 s, 1:400), following by incubation with a MACH1 or MACH2 HRP Polymer Detection (Biocare Medical), according to the manufacturers instruction, together with the 3,3′-Diaminobenzidine (DAB) Chromogen Kit (Biocare Medical), for detection. Sections, where then counterstained (Mayer’s Haematoxylin), dehydrated, cleared and coverslipped, using a Leica ST5010 Autostainer XL and Leica CV5030 Fully Automated Glass Coverslipper (both Leica Biosystems).

Expression was analysed using primary antibodies with the following specificities: α-Dystroglycan (EMD Millipore, IIH6C4, 1:100), EphA2 (in-house, 1F7, 1:100), EphA3 (in-house, IIIA4, 1:100), CD133 (Miltenyi Biotec, AC133, 1:20), CD49f (Abcam, GoH3, 1:200), CD29 (Abcam, EP16895, 1:100) and CD15 (Abcam, SP159, 1:200). IgG1 and IgM antibodies (Abcam, 1:100) were used to indicate background staining. Cells were washed twice in PBS and blocked for 20 min in PBS supplemented with 1% BSA (Sigma). After blocking, cells were incubated with primary antibodies for 20 min at room temperature (RT). Washed cells were then incubated for 20 min at 4 °C in the dark with secondary antibodies Anti-Mouse IgM Alexa Fluor 488 (Abcam, 1:2000), Anti-Rabbit Alexa Fluor 647 (Abcam, 1:2000) and/or Anti-Rat Alexa Fluor 647 (Abcam, 1:2000). Before analysis cells were washed in blocking solution to remove any unbound antibody conjugates. Between 20,000 and 50,000 total events were examined using an LSRFortessa (BD) with data analysed using FACSDiva (BD) software. Live cells were gated using propidium iodide (Sigma, 50 μg/ml) staining.

The flow cytometry procedure was repeated for GBM cell lines, as described above. Samples were then run on an ImageStream^x. Briefly, 5,000 events were collected for each sample and single color controls used to create a compensation matrix to correct for spectral overlap. All data were then analyzed using IDEA software (Amnis Corporation, Seattle, WA, USA).

Single cells were suspended in StemPro NSC SFM (Invitrogen) without growth factors and seeded in 6-well plates (5 × 10⁴ per well) in triplicate and treated with α-Dystroglycan inhibiting antibody (EMD Millipore, IIH6 C4, 1:100, 50 µg/ml) or an IgM control (Mo1, in-house, 1:100, 50 µg/ml). After 7 days, adherent and neurosphere cultures were collected and dissociated using accutase, proliferation was calculated by direct cell counting using a haemocytometer.

Single GBM cells (1 × 10³) were plated onto individual wells (96-well format) for 16 h prior to experimentation. Cells were treated with α-Dystroglycan inhibiting antibody (EMD Millipore, IIH6C4, 1:100, 50 µg/ml) or an IgM control (Mo1, 1:100, 50 µg/ml). The Incucyte system (Essen Instruments) was employed to quantitate cell adhesion, as a surrogate read out for differentiation, in real time.

Apoptosis and viability was assessed using the ApoTox-Glo Triplex assay (Promega) as described [4]. Cleaved caspase3/7 activity and cell viability was assessed 48 h post IIH6 treatment, 7.5 mg/mL cisplatin treatment was used as a positive control. Results are shown as caspase3/7 activity normalised to cell viability relative to IgM control (n = 3).

Target cells were plated on Matrigel-coated (Corning, 1:200) plates in StemPro NSC SFM (Invitrogen) at 50% confluency 24 h prior to viral infection. The next day, plating media was removed and polybrene (Santa Cruz, 5 μg/ml) added, followed by the addition of the Lentiviral Particles α/β-Dystroglycan shRNA (Santa Cruz, sc-43488-V, 20 μl/ml) and control shRNA (Santa Cruz, sc-108080, 20 μl/ml). 24 h following transduction, fresh media was added and cells underwent puromycin (Gibco, 0.3 μl/ml) selection.

In-vivo experiments—sample size or replicate number (designated as “n”) for each experiment are indicated in the figure. Survival plots were generated using Graphpad software. A Log-rank (Mantel-Cox) test was used to determine significance between experimental groups. P-values are as indicated, *p ≤ 0.05 was considered to be statistically significant. In-vitro experiments—a Student’s t-test determined the probability of difference; p < 0.05 was considered significant; all statistical tests were 2-sided. The correlation coefficient was determined using a Spearman rank.

αDG and βDG are transcribed from the Dystrophin-Associated Glycoprotein 1 (DAG1) gene. DG is translated as a single pre-pro-polypeptide. The receptor is autocatalytically cleaved into alpha and beta chains that remain non-covalently bound. The mucin-like domain of αDG becomes heavily glycosylated en-route to the surface membrane. Dystroglycan binds proteins of the extracellular matrix such as laminin by virtue of matriglycan (xylose and glucuronic acid polysaccharide). Receptor glycosylation is essential for proper function, without which αDG is unable to bind laminin [19, 42]. To determine if DAG1 was important in the context of brain cancer we interrogated both the Rembrandt and TCGA databases to correlate DAG1 gene expression with survival. In the context of GBM specifically and also glioma, patient tumours with elevated DAG1 led to a significantly shorter survival time (Fig. 1a and Online Resource 1a). The Rembrandt database was also used to assess DAG1 gene expression in GBM as well as other forms of malignant brain cancer versus normal brain tissue. This revealed that DAG1 tended to correlate with tumour grade, as expression was highest in GBM compared to oligodendroglioma and astrocytoma cases, while all tumour types were elevated above normal brain tissue (Online Resource 1b). DAG1 expression in GBM was further stratified into molecular subtype [8, 59, 60]. DAG1 expression was highest in classical (CL) subtype GBM while approximately equivalent in other subtypes, mesenchymal (MES), proneural (PN) (Online Resource 1c). To assess the relative mRNA levels of DAG1 in GBM we performed QPCR on 28 GBM tumour specimens from our in-house tumour bank. We compared expression to other receptors (EPHA2, EPHA3 and ITGA6) previously identified as having a role in GSC maintenance. DAG1 levels were equivalent or higher to these receptors in all cases assessed (Fig. 1b).

Importantly, receptor function correlates closely with glycosylation status rather than gene expression. To determine the level of αDG glycosylation we used a monoclonal antibody (mAb) (IIH6), previously developed by Campbell and colleagues, specific to glycan moieties on αDG with known receptor blocking function [20]. We have developed a GBM patient-derived cell line resource (Q-Cell) [14, 57] in which GBM lines are maintained as glioma neural stem cell (GNS) cultures [52]. This approach promotes a more de-differentiated stem cell-like phenotype in culture. We assessed a panel (n = 20) of subtype classified (n = 12) and unclassified (n = 8) early passage GBM cultures generated in-house. Flow cytometric analysis revealed the majority of the lines expressed high levels of glycosylated αDG irrespective of GBM molecular subtype (Fig. 1c and Online Resource 1d). A panel of four IDH1 WT GBM lines were selected for further analysis, JK2-(MES), RN1-(MES), SJH1-(PN) and WK1-(CL), αDG protein expression was confirmed by western blot (WB) (Fig. 1d). We next performed cell fractionation studies followed by WB on the panel of GBM lines, data indicated the majority of the protein was present in the cell membrane (Fig. 1e).

A previous report in breast cancer had shown that EphA3 could regulate αDG [63]. Given these findings we assessed the co-expression of glycosylated αDG and EphA3 protein levels by flow cytometry. Mean channel fluorescence (mcf) analysis revealed a high correlation (r = 0.899) between glycosylated αDG and EphA3 in the primary GBM GNS lines tested (n = 10) (Fig. 1f).

To examine the spatial localisation and tumour specificity of glycosylated αDG (hereafter stated as αDG for simplicity), we first examined a variant of GBM, gliosarcoma (GS), which undergoes a partial process of de-differentiation with secondary loss of the glial differentiation marker GFAP and acquisition of mesenchymal characteristics, a process similar to epithelial–mesenchymal transition (EMT) [45]. GS is characterised histologically by a biphasic pattern of gliomatous glial-like (GFAP⁺) and sarcomatous mesenchymal-like (MES-like) (vimentin⁺) tumour elements (morphology shown in Fig. 2a) [27, 45]. Common genetic alterations are found between GBM and GS, and both have a poor prognosis. A known clinical feature of GS is that the mesenchymal tumour compartment is highly migratory and infiltrates ECM protein-rich brain regions such as the leptomeninges [6]. As previously shown by others, the mesenchymal elements stain positive for vimentin [27]. We performed immunohistochemistry (IHC) for vimentin, and found a clear separation between glial-rich and predominantly mesenchymal-rich regions (Fig. 2b). We next performed IHC on sequential GS tumour tissue sections to determine the localisation of αDG, βDG, EphA2 and EphA3. A representative example shows that αDG and EphA3 are expressed predominantly in regions, where mesenchymal tumour elements are present and expression is low or absent on the more differentiated GFAP⁺ glial tumour element (Fig. 2c). This discrete pattern was not evident for EphA2 and βDG which showed comparable expression in both glial and mesenchymal tumour elements. The observed biphasic tumour tissue pattern in GS also appeared, although less dramatically discrete, in GBM specimens (Online Resource 2a). These regions were smaller and less prominent than in GS tumours, and also showed heterogeneous expression of GFAP, vimentin, EphA2, EphA3 αDG, CD49f and CD31, indicating that this tumour tissue patterning might also occur in GBM (Online Resource 2b). In addition, immunofluorescent (IF) staining revealed in GBM patient specimens (n = 3) that αDG staining was strongest in perivascular regions immediately surrounding tumour blood vessels (CD31⁺) (Fig. 2d and Online Resource 2c).

Given the overlapping expression profile of EphA3 and αDG in GS and GBM tissue we wanted to assess the membrane localisation and potential interaction between these receptors. IF analysis showed a significant overlap in the membrane of EphA2, EphA3 and αDG in the four primary GBM cell lines tested (Fig. 3a). To assess if these proteins could be complexed together we performed immunoprecipitation (IP) of EphA3 in these lines. Results show co-immunoprecipitation of EphA3 with EphA2 and αDG (Fig. 3b). To further confirm an association between these receptors we performed shRNA mediated EphA3 knockdown (KD) in SJH1 and JK2 cell lines. Following EphA3 KD a commensurate drop in αDG levels was observed (Online Resource 3a). These data suggest that αDG and EphA3 are likely associated in the cell membrane.

We next used Amnis image stream technology, a flow cytometric approach which captures single-cell multi-colour fluorescent images. This approach was used to assess cell-surface co-expression of αDG with known GSC markers. Initially, this was performed on dissociated GBM patient tumour tissue (n = 1) obtained directly at time of surgery. Results show strong overlapping expression of αDG with EphA2, EphA3, CD49f (integrin α6), CD133 and CD15 (Fig. 3c). We repeated this analysis using both Amnis and standard flow cytometry using the panel of GBM cell lines (n = 4). Results demonstrated a similar pattern of co-expression of αDG especially with EphA2, EphA3 and CD49f, this was less pronounced for CD15, CD29 and CD133 (Fig. 3d and Online Resource 3b,c). To assess stemness characteristics in the αDG high versus low expressing population we sorted WK1 GBM cells using FACS and performed a neurosphere assay. Results show that neurosphere formation was significantly elevated in the αDG^high compared to the αDG^low population. In addition, we observed cell changes in the αDG^low population; low expressing cells that failed to form neurospheres displayed an attached differentiated cell morphology (Fig. 3e). Post αDG^high versus αDG^low sort we recombined the two cell populations and assessed αDG expression overtime. Interestingly, 14 days post-sort both low and high expressing populations were maintained, suggesting that these two unique populations can co-exist in culture for extended periods (Online Resource 3d). αDG expression and stemness characteristics should be lost following GSC differentiation. To test this we induced differentiation in our neurosphere cultures and 72 h post-differentiation assessed αDG expression by flow cytometry and GSC marker expression by QPCR. Results show a marked reduction in αDG protein expression following differentiation (Fig. 3f), this was accompanied by significant gene downregulation of stemness markers and upregulation of the differentiation markers GFAP, βIII-tubulin (TUBB3) and OLIG2 (Fig. 3g and Online Resource 3e).

To assess the contribution of αDG glycosylation to the progression of GBM and maintenance of a GSC phenotype, we employed an αDG mAb (IIH6) which specifically binds and blocks the ability of glycan moieties on αDG to bind laminin [20]. Following incubation of neurospheres with the IIH6 antibody, we observed a robust and rapid loss of sphere formation compared to cultures incubated with an isotype control mAb in which sphere integrity was maintained. The response was reliably observed in all primary cultures tested (Fig. 4a). The observed differentiation response was dose dependent, while no response was observed with three independent IgM control antibodies (Online Resource 4a). Neurosphere loss is suggestive of differentiation and loss of proliferative ability. Differentiation was confirmed by acquisition of neuronal (βIII-tubulin) and glial (GFAP) lineage markers (Fig. 4b and Online Resource 4b). This differentiation response was quantitated using IncuCyte technology in real time and showed pronounced cell-attachment following IIH6 treatment (Fig. 4c). The observed differentiation response was accompanied with a significant reduction in proliferation (p < 0.01) (Fig. 4d). We observed negligible cell death induction in the short term (48 h) following antibody treatment (Fig. 4e and Online Resource 4c). Interestingly, the differentiation effects were reversible if antibody was removed within seven days of treatment. If additional IIH6 mAb was added at seven days, the majority of rechallenged tumour cells showed pronounced cell morphology changes and ultimately underwent cell death within two weeks (Fig. 4f). These data highlight the ability of GBM cells to recover during early differentiation but also susceptibility to prolonged terminal differentiation induced by the IIH6 antibody.

Sustained ERK activation, as distinct from transient signalling, induces translocation of ERKs to the nucleus resulting in pro-differentiation transcriptional changes in neuronal cells [38]. Based upon our previous work and also the studies of others, it appears this mechanism is also observed in a number of cancers including GBM [13, 16]. We, therefore, assessed ERK activation following IIH6 treatment and found that pERK activation was sustained 24 h post-treatment (Fig. 5a). Analysis of pERK cellular localisation, following IIH6 treatment, showed that activated ERK was predominantly localised the nucleus. The observed response was equivalent to a positive differentiation control (2% FBS), while in control-treated cells, the majority of pERK was retained in the cytoplasm (Fig. 5b). To examine if this same mechanism was operative in patient tissue and not an artefact of culture we assessed sequential (#1–4) GS tumour sections by IHC for αDG, EphA3, GFAP, pERK and Ki67. Results confirmed co-expression of αDG and EphA3 in the mesenchymal fraction which was discrete from the more differentiated GFAP⁺ glial fraction. pERK IHC staining revealed clear co-expression with the GFAP⁺ glial-predominant tumour regions which also showed less proliferation as shown by very low level Ki67 staining (Fig. 5c). This data confirmed that sustained ERK activation occurs in high-grade glioma tissue and is restricted to the more differentiated, GFAP⁺, less mitotically active tumour regions.

A study in breast cancer found that tumour cell populations differ in expression of integrin α6 cytoplasmic domain splice variants; α6A is expressed on epithelial tumour cells while α6B on mesenchymal tumour cells [26]. Data show that α6B is the critical variant driving EMT and cancer stem cell function and promotes tumour initiation. Laminin-integrin interactions can activate ERK but only in cells expressing predominantly α6A with an intact cytoplasmic tail, not α6B [61]. While DG when bound to laminin, can suppress integrin a6-mediated ERK activation [23]. Integrin α6 has previously been demonstrated to be expressed and functional in the GSC niche and is typically active in the process of niche formation and basement membrane attachment [11, 34]. We assessed α6A and α6B mRNA expression via QPCR in 28 GBM specimens. Results show that α6B and the mesenchymal markers N-cadherin and Slug are predominantly expressed, while α6A and the epithelial marker E-cadherin are essentially absent (Fig. 5d). We also investigated the relative mRNA levels of α6A and α6B in 6 primary GBM GNS cell lines and found an average 12.1 fold increase in α6B compared to α6A. This correlated with an average 210 fold increase in the mesenchymal marker N-cadherin compared to the E-cadherin (Online Resource 5a). In addition, α6B correlated highly with mesenchymal and GSC markers in the primary GNS lines tested (n = 6) (Fig. 5e). The oncogene BMI-1 is a stem cell factor and polycomb group family member that functions as a repressor of the splicing factor ESRP-1. Mercurio and colleagues show, in breast cancer, that α6B is positively controlled by autocrine VEGF-A signalling that culminates in transcriptional repression of the RNA-splicing factor ESRP-1 [26]. We, therefore, assessed BMI-1, VEGF-A and ESRP-1 mRNA levels by QPCR in GNS cell lines (n = 6) (data not shown) and GBM tissue (n = 28), this revealed high levels of BMI-1 and VEGF-A, while ESRP-1 levels were undetected (Fig. 5f). This suggests that a similar mechanism to that observed in breast cancer could be operative in GBM. To determine if integrin α6 was responsible for driving sustained ERK activation following IIH6 treatment we performed an integrin α6 shRNA mediated KD compared to control shRNA. These cells were then treated with the IIH6 blocking mAb, KD of integrin α6 did not reverse the differentiation response. A similar response was observed when cells were pre-treated with the integrin α6 blocking mAb GoH3 (Online Resource 5b, c). The GoH3 blocking antibody had little effect when used as a single agent on multiple GNS cell lines and did not appear to induce a differentiation response (data not shown). GoH3 treatment alone appeared to induce very low level ERK activation (Fig. 5g). This was expected given the majority of integrin α6 present is the α6B splice variant with a truncated intracellular tail with reduced ERK signalling capacity. When IIH6 was combined with GoH3 we observed a somewhat more rapid differentiation response, this coincided with an increased level of sustained ERK activation than IIH6 alone 24 h post-treatment (Fig. 5g), and a greater reduction in cell proliferation (Fig. 5h). Indicating that the sustained ERK activation, observed following IIH6 treatment, was not driven through integrin α6.

As described above, EphA3 is co-localised with dystroglycan and integrin α6. Following EphA3 IP we observed co-immunoprecipitation of EGFR in GBM cells, suggesting that EGFR is also present as a part of this complex and may be involved in regulating ERK activation (Online Resource 5d). To test if EGFR was responsible for driving sustained ERK activation following antibody treatment, neurospheres were pre-treated with either Gefitinib (2 µM) or Erlotinib (2 µM) prior to IIH6 treatment. EGFR inhibition did not prevent the induction of differentiation (Online Resource 5e). Taken together, this data suggests that αDG when bound to laminin, forms a complex with EphA2, EphA3, integrin α6B, EGFR and likely other receptors to mediate a de-differentiated GSC mesenchymal-like phenotype through tight regulation of ERK signalling.

To further define DG function in GBM we performed lentiviral DAG1 shRNA mediated KD, using a combination of five target-specific 19–25 nucleotide sequences against DAG1, four GBM neurospheres lines were tested. Following effective constitutive KD in JK2, RN1 and WK1 cells (Fig. 6a), we observed a significant (p < 0.05) reduction in proliferation (Fig. 6b) and neurosphere loss typical of differentiating GBM cells (Fig. 6c). In the case of SJH1, we observed a florid response to DAG1 KD; all cells underwent cell death within 7–14 days following KD (Online Resource 6a). Given this rapid response, SJH1 cells were excluded from further analysis. In-vitro DAG1 KD also reduced expression of EphA2, EphA3 and CD49f (integrin α6) while markedly increasing the expression of the glial differentiation marker GFAP, indicating KD was inducing glial lineage differentiation (data shown for JK2) (Online Resource 6b). To explore if DAG1 KD reduced the tumourigenic potential of GBM cells in-vivo we performed orthotopic engraftment of control shRNA versus DAG1 KD cells into the right striatum of NOD/SCID mice. Animals were euthanised when signs of illness or disease burden were observed. The two highest DAG1 expressing lines (WK1-(CL), JK2-(MES)) were selected. Kaplan Meier survival analysis for WK1 showed DAG1 KD significantly increased overall survival p = 0.0266 (n = 4), median survival for control was 94 days versus DAG1 KD 109 days (Fig. 6d). qPCR analysis of post-mortem tumour tissue confirmed a significant (p < 0.01) reduction in DAG1 mRNA levels compared to control shRNA engrafted animals (Online Resource 6c). Positive tumour formation was confirmed by haematoxylin and eosin (H&E) staining in all control and DAG1 KD engrafted animals (Online Resource 6d). Kaplan Meier survival analysis for JK2 showed DAG1 KD significantly increased overall survival p = 0.0043 (n = 7), with no animals reaching survival endpoint, whereas median survival for control group was 195 days (Fig. 6d). All DAG1 KD animals developed spontaneous thymic lymphoma, which is a common occurrence in this strain, and were euthanised as per our ethical guidelines [53]. Positive tumour formation was confirmed by H&E staining in all control shRNA engrafted animals, while no clinical signs of intracranial tumour formation were observed in DAG1 KD animals or intracranial tumour detected by H&E staining post-mortem (Fig. 6e).