DNA methylation changes in primary and recurrent glioblastomas
A brief overview of methylation data analysis is depicted in Fig. 1A. Genomic DNA were subjected to Infinium 450K Methylation Bead Array for assessing differentially methylated regions (DMRs) with a Δβ-value cutoff of |0.2| and p ≤ 0.05, revealing 1224 hypermethylated and 526 hypomethylated probes (Supplementary Table 2). We performed a gene-level annotation for these commonly regulated DMRs for probes localizing to 5’-Untranslated Regions, Transcription Start Site (TSS) 200 and TSS1500, and 1st Exon regions. The top 50 hypo/hypermethylated probes with their gene annotation is represented in Supplementary Table 3.
We observed several interesting candidates among our list of differentially regulated genes, categorized according to reported functions in Supplementary Figure S1A. Several genes such as CAV2, HTATIP2, DSC3 for which epigenetic inactivation in different cancers is reported [17–19], came up as hypermethylated genes in recurrent tumors. Genes associated with histological grade such as ZHX2, ASAP1 and genes with reported protumorigenic functions such as ANTRX1/TEM8, FAM46D, AZIN1, CNOT7 were hypomethylated. Several genes encoding solute carrier ion channels or potassium channels (KCNN3, KCNT1, KCNQ1) were dysregulated (hypo-/hypermethylated) in recurrent glioblastomas. This is interesting because involvement of ion channels in gliomagenesis is increasingly being noted , and this data may suggest that there could be unexplored contribution of ion channel dysregulation in tumor progression. The β-values for genes showing most dysregulated hypo- and hypermethylated probes, are shown in Supplementary Figure S1B.
We validated randomly selected hypomethylated genes in our patient cohort, and found them to be upregulated in recurrent glioblastomas (Fig. 1B). These include stem-cell biomarker ANTXR1/TEM8, transcriptional co-repressor CNOT7. We decided to functionally characterize the ANTXR1/TEM8 (Anthrax Toxin Receptor 1/Tumor Endothelial Marker 8) gene as it was hypomethylated at promoter-associated CpG island at TSS200, suggesting that hypomethylation could potentially regulate its transcription, apart from the fact that it is characterized as a stem-cell biomarker in triple negative breast cancer , and is pro-tumorigenic in osteosarcomas, colorectal and gastric cancer [22–24]. However, significance of its expression or role in glioblastoma is not yet reported.
ANTXR1/TEM8 expression in paired sequential glioblastomas.
As the ANTXR1/TEM8 gene was found to be hypomethylated (Supplementary Figure S2A) and transcriptionally upregulated in recurrent glioblastomas, we assessed whether TEM8 protein is overexpressed in paired recurrent glioblastomas compared to primary tumors. To this end, a separate retrospective cohort of 29 paired primary and recurrent glioblastomas was utilized to determine protein expression via immunohistochemistry (Fig. 1C,D). A semi-quantitative labeling index (LI) approach for 2 + staining revealed significant upregulation of protein expression (Wilcoxon paired signed-rank test, p-value 0.001) in recurrent tumors (mean LIrec : 34.5 compared to mean LIpri : 27, Supplementary Figure S2B). Additionally, we utilized a control brain tissue microarray to determine TEM8 protein expression in non-neoplastic brain tissues (Fig. 1E) and there was no detectable staining in control brains.
In glioblastomas, the role or consequence of TEM8 expression is unknown. Analysis of transcript expression in REMBRANDT database (Repository of Molecular Brain Neoplasia Data) revealed TEM8 upregulation in lower-grade gliomas (LGGs) and glioblastomas (Supplementary Fig. 2D). This correlated with upregulation in other datasets such as TCGA, Bredel Brain and Sun Brain (Supplementary Figure S2E). Survival analysis via Gliovis  revealed that higher TEM8 expression conferred poor survival in glioma (Fig. 1F).
TEM8 expression regulates proliferation, invasion, migration, chemoresistance and radiation resistance in glioblastoma cells.
Immunoblotting for TEM8 on glioblastoma tissue lysates revealed a predominant ~ 63 kDa band, suggesting that this is an important isoform expressed in glioblastomas (Supplementary Figure S2F). To understand the functional consequences of TEM8 expression in glioblastomas, we overexpressed a 3X Flag-hTEM8/pcDNA3.1 construct in U87-MG, A172 and LN229 cell lines (Supplementary Fig S2G). We observed increased proliferation in TEM8 overexpressing (OE) cells compared to vector controls (VC) as seen by Trypan-blue based viability assays (Fig. 2A) and BromodeoxyUridine uptake for 4 hours (Supplementary Figure S3A). We assessed differences in cell cycle progression by stimulating cells with 10% FBS for 12 hours after prolonged (72h) serum-starvation induced synchronization in U87 cells. At 12 hours of growth-stimulation, we observed 8.83% cells in S phase in OE cells compared to 4.58% in VC cells (Fig. 2B, Supplementary Figure S3C), suggesting that TEM8 OE cells had basally increased S phase cells in the absence of growth factors, which progressed faster on additional serum-stimulation, contributing to the proliferation advantage. We also observed increased invasion (Fig. 2C), migration (Fig. 2D, Supplementary Figure S3B) in OE cells; while gelatin zymography using conditioned media revealed expression of active MMP-2 (Fig. 2E) suggesting a possible role in the increased invasion. Using the MTT assay we explored if TEM8 enhanced chemo-resistance in cells towards four drugs viz. Temozolomide, Cisplatin, Etoposide, and 5-FluoroUracil (Fig. 2F). Relative 50% Inhibitory Concentration (IC50) calculations by non-linear regression curve fitting revealed increased IC50 values for all the above drugs in TEM8 OE cells except in the case of 5-FluoroUracil, which showed a marginal increase (Supplementary Figure S3E). Additionally, TEM8 OE cells generated bigger and about 7 fold increase in neurospheres than compared to control cells (Fig. 2G, Supplementary Figure S3D)
When TEM8 was stably knocked-down in U251 glioma cells with two shRNAs, pLKO.1-D6 and D9, we observed a growth lag in TEM8 knock down cells (Fig. 3A,B) compared to scrambled-shRNA transfected cells (SCR). We used a similar approach of 72h serum-starvation induced synchronization and release with 10% FBS-media. After 72h (i.e. at 0h timepoint) we observed more cells (76%) being arrested in G1 phase in D6 and D9 compared to 61% cells in SCR controls (Supplementary Figure S4A,B). To ascertain if parental U251 cells showed similarly low levels of G1-synchronization as SCR cells, we performed the same experiment with U251-parental cells which revealed that ~ 66% cells were G1 arrested (Supplementary Figure S4C). This suggested that TEM8 knockdown led to more efficient G1-arrest in D6 and D9 cells on prolonged serum-starvation. Further, at 12h of growth-stimulation, we observed 26.3% cells in G2/M phase in SCR, compared to 15 and16% in D6 and D9 cells respectively. The experiment was repeated thrice and is quantified in Fig. 3C.
We also demonstrate reduced invasion (Fig. 3D) and migration (Fig. 3F) in D6 and D9 cells when compared to SCR, and gelatin zymography with conditioned media from these cells revealed reduced MMP2 levels in knockdown cells compared to SCR controls (Supplementary Figure S4D). To assess whether TEM8 knockdown led to reduced chemo- and radioresistance, we performed MTT and clonogenic assays respectively. In response to Temozolomide and Cisplatin, we found relative IC50 concentrations were reduced in D6 and D9 cells compared to SCR (Fig. 3E, Supplementary Figure S4E). Additionally, clonogenic assays after γ-radiation exposure (0–10 Gray) revealed that TEM8 knockdown-D6 and D9 cells had attenuated radioresistance compared to SCR cells (Fig. 3G), as evidenced by lesser area-under-curve (SCR : 2.67, D6 : 1.98, D9 : 2.5) and surviving fraction (also see Supplementary Figure S4F).
The TEM8 gene regulates β-catenin signaling in glioblastoma cells.
Few reports suggests that ANTXR1 can interact with LRP6 [26, 27], and engaging ANTXR1 with the C5 fragment of Collagen VI or with Anthrax toxin component Protective Antigen, lead to induction of Wnt target genes such as Axin2 or Zeb1 in triple-negative breast cancer or endothelial cells [21, 28]. We therefore explored if TEM8 activation led to enhanced β-catenin signaling in glioblastoma cells. In U87 TEM8 OE cells, we observed induction of Wnt target genes Zeb1, Axin2, Nanog, Twist1 and α-SMA, although β-catenin or LRP6 transcript levels were unchanged (Fig. 4A). In three glioma cell lines U87, LN229 and A172, we found induction of Zeb1, Twist, Vimentin, Oct4 and α-SMA, all of which are induced by β-catenin (Fig. 4B). To determine if this association held true in patients, we assessed the co-expression of TEM8 with β-catenin target genes in glioblastoma patients from the Chinese Glioma Genome Atlas (CGGA) dataset in Gliovis. The expression of Zeb1, Axin1, Axin2, CyclinD1, c-myc and Vimentin were all positively correlated (Pearson’s correlation co-efficient r2: 0.33–0.77) to ANTXR1/TEM8 expression (Supplementary Figure S5A).
We observed no induction of canonical (Wnt1,Wnt2) or non-canonical (Wnt4,Wnt11) Wnt ligands (Supplementary Figure S5B) in OE cells, suggesting that β-catenin induction by TEM8 is likely Wnt-ligand independent. We further confirmed enhanced nuclear β-catenin accumulation in U87 and LN229 OE cells via nuclear-cytoplasmic fractionation (Fig. 4C). Next, we utilized a dual-luciferase assay to gauge nuclear β-catenin activity in overexpressing and knockdown cells. β-catenin responsive 7X TCF containing luciferase reporter construct, Super8X pTOPFlash, and its negative control with 7X mutated TCF, pFOPFlash, was transfected into these cells. We observed enhanced luciferase induction in TEM8 OE cells (Fig. 4D) and reduced induction in D6 and D9 knockdown cells compared to SCR cells (Fig. 4E). Additionally, we found reduced levels of β-catenin target proteins such as Zeb1, Twist1, α-SMA, Vimentin in U251-D6 and -D9 knockdown cells (Fig. 4F).
TEM8 regulation of β-catenin is via Src/PI3K/AKT/GSK3β cascade in glioblastoma cells.
As TEM8 could upregulate β-catenin and its effector genes, and as per our data Wnt ligands may not be involved, we explored the signaling pathway responsible for β-catenin translocation into the nucleus. We therefore determined the phosphorylation of protein kinases such as GSK3β, whose inactivating phosphorylation at Ser9 leads to β-catenin stabilization. Expectedly, p-GSK3β Ser 9 levels in TEM8 OE cells was upregulated (Fig. 5A) in three cell lines : U87, LN229 and A172, concomitant with upregulated β-catenin levels. GSK3β is regulated by varied upstream kinases, such as PI3 Kinase induced AKT [29, 30], Integrin-linked kinase [31, 32] or focal adhesion kinase  in glioblastomas. We observed elevated levels of phospho-AKT S473 and T308 in TEM8 OE cells. Additionally, we observed elevated phospho-ILK (S246) and phospho-FAK (Y397) levels in TEM8 OE cells.
In TEM8 knockdown cells, we observed a concomitant decrease in p-AKT (S473, T308), p-FAK (Y397) and p-GSK3β (S9) levels (Fig. 5B). Interestingly, the total levels of FAK were reduced in knockdown cells, suggesting destabilization of focal adhesions and reduction of focal adhesion-mediated survival signals in these cells.
Since multiple kinases were induced, we utilized a panel of small-molecule inhibitors to identify the exact signaling cascade involved in β-catenin stabilization in the presence of TEM8 : LY294002 (PI3K inhibitor), CPD-22 (ILK inhibitor), FAK inhibitor 1 (324877), RGD peptide (GRDGNP), PP2 (Src inhibitor) and PP3, a non-functional analog of PP2 was used as a negative control. We observed that phosphorylation on AKT (both T308 and S473) and GSK3β (S9) was abrogated in presence of PI3 kinase inhibitor LY294002 (Fig. 5C), suggesting PI3 kinase activation was responsible for induction of p-AKT in TEM8 OE cells. Additionally, we observed that PP2 (Src kinase family inhibitor) abrogated p-AKT and p-GSK3β induction as well, suggesting that Src-dependent activation of PI3 Kinase/AKT was responsible for inactivating phosphorylation of GSK3β and stabilization of β-catenin in TEM8 OE cells. We therefore surmise that a Src/PI3K/AKT/GSK3β/β-catenin pathway is activated in TEM8 expressing glioma cells (Fig. 5D) which leads to enhanced proliferation, invasion, migration, chemo- and radioresistance.