Methylation-dependent Tissue Factor Suppression Contributes to the Reduced Malignancy of IDH1-mutant Gliomas

Approximately 20%–30% of diffusely infiltrative gliomas contain point mutations in the metabolic enzyme isocitrate dehydrogenase 1 or, less commonly, IDH2 (hereafter collectively referred to as “IDH1^mut”; ref. 1). These mutations unmask latent oxidoreductase activities, causing α-ketoglutarate to be converted into D-2-hydroxyglutarate (D2HG; ref. 2). The D2HG product of IDH1^mut competitively inhibits many α-ketoglutarate–dependent DNA and histone demethylases, resulting in genomic hypermethylation and altered transcription of many genes (3, 4). Although IDH1^mut has been linked to suppression of cellular differentiation and predisposition toward oncogenic transformation, it is also associated with lower World Health Organization (WHO) grade and longer survival in patients with glioma (5–7). As further indication of their attenuated malignancy, IDH1^mut gliomas are much more difficult to grow in culture or as patient-derived xenografts (PDX), relative to IDH1^wt gliomas (8, 9). The basis for the lesser malignancy of IDH1^mut tumors is far from fully understood.

Previously, we reported that the absence of intratumoral microthrombi is the best histologic predictor of IDH1^mut in gliomas, independent of WHO grade and patient age (10). We also found that IDH1^mut gliomas are far less likely to cause systemic venous thromboemboli (VTE) than IDH1^wt gliomas, again independent of WHO grade and patient age. Further investigation into the mechanistic underpinnings for these clinically based observations revealed that F3, the gene encoding tissue factor (TF), is one of the most hypermethylated and downregulated genes in IDH1^mut gliomas (10). TF is a highly conserved glycoprotein that binds with activated coagulation factor VII (FVIIa) to initiate the clotting cascade, and its production and release by cancers into the systemic circulation have been repeatedly linked to VTE in patients (11). Circulating TF is lower in patients with IDH1^mut gliomas than IDH1^wt gliomas, and elevated plasma TF positively correlates with increased risk of glioma-induced VTE (10).

In addition to its procoagulant activity, the TF–FVIIa complex can also activate transmembrane G protein–coupled protease-activated receptors (PAR), especially PAR2 (12). TF-PAR2 activity is important for wound healing at sites of vascular injury and hemostasis. In cancer, however, TF-PAR2 initiates signaling cascades that promote growth, angiogenesis, invasion, and metastasis (13). An unbiased screen for metastatic enhancer elements found that TF is a major contributor to tumor metastasis (14). Less is known about the activity of TF in gliomas, although it has been shown to increase malignant phenotypes when ectopically expressed in glioblastoma (GBM) cell lines (15–17). Our prior work showed that, in patients with IDH1^wt GBM, high intratumoral TF protein expression correlated with an approximately 50% reduction in median survival (10). Therefore, we hypothesized that TF enhances the malignancy of IDH1^wt gliomas, and that its suppression is an important component of the less aggressive IDH1^mut glioma phenotype.

Three patient-derived glioma cell types with endogenous IDH1^mut were used: GBM164 from Mayo Clinic; anaplastic oligodendroglioma BT142 from ATCC (18); anaplastic astrocytoma TB09 from Dr. Hai Yan at Duke University (Durham, NC; ref. 19). All three cell types express R132H IDH1 (Fig. 1C); to ensure retention of functional IDH1^mut, cells were regularly subjected to Sanger sequencing analysis, and their ability to produce D2HG was verified by liquid chromatography–mass spectrometry (LC-MS; Supplementary Fig. S1B and S1C). BT142 cell line lost the IDH1^wt allele, and therefore no longer produces high levels of D2HG (Supplementary Fig. S1C). GBM164 was maintained as slow-growing subcutaneous flank PDX, with short intervals of in vitro cultures supplemented with 110 mg/mL pyruvate. TB09 was cultured in 1:1 ratio of DMEM (Corning) and NeuroCult NS-A (StemCell Technologies) with 10% FBS, 1% penicillin/streptomycin, 2 μg/mL heparin sulfate, 10 ng/mL fibroblast growth factor, 4.5 g/L sodium pyruvate, and 20 ng/mL EGF. IDH1^wt gliomas (GBM6, GBM12, GBM43) were maintained as PDX or cultured briefly for in vitro testing and have been described elsewhere (20, 21). IDH1^wt gliomas were cultured in DMEM supplemented with 10% FBS and 1% penicillin/streptomycin. The unique identity of all six cell lines was authenticated using short tandem repeat analysis (Supplementary Table S1), and cells were periodically verified to be Mycoplasma-free with a testing kit (Sigma Aldrich).

F3-specific shRNA sequences were designed and cloned into BLOCK-iT U6 RNAi Entry Vector (Invitrogen; Supplementary Table S2). F3 shRNA or empty vector control cassettes were then recombined into a BLOCK-iT Viral RNAi vector (Invitrogen), and used to generate lentiviral particles for stable constitutive shRNA delivery.

F3 cDNA was cloned into pcDNA4/TO/myc-His Vector (Invitrogen) under control of CMV promoter. F3 cDNA cassettes were then transfected into cells using Lipofectamine 2000 (Thermo Fisher Scientific). For generation of stable transfectants, cells were selected with zeocin.

Cell invasiveness was assessed using polycarbonate membrane transwell inserts with 8.0-μm pores (Corning). Cells were serum starved for 24 hours prior to Matrigel chemoinvasion assay. Transwell membranes were coated with a thin layer Matrigel basement membrane matrix (Corning). Culture medium containing 10% FBS was used in the bottom chamber as chemoattractant. A total of 1 × 10⁵ cells were added to the top chamber in serum-free medium and allowed to migrate for 12 hours. Membranes were fixed with cold methanol, stained with Vectashield mounting medium containing DAPI (Vector Laboratories), and a cotton swab was used to remove nonmigrated cells from the top chamber. Migrated cells were counted in 3–5 fields per membrane and performed in triplicates.

Cellular anchorage-independent growth was measured by preparing two layers of agar (BD Biosciences). The bottom layer of agar was prepared at 1%, autoclaved, and diluted 1:1 with 2× cell culture medium; 1.5 mL was added to each well of a 6-well tissue culture plate and allowed to solidify. The top layer of agar was prepared at a 0.6% agar concentration, and mixed with 5,000 cells per well. Plates were cultured at 37°C in 5% CO₂ and fed twice a week. After 3–5 weeks, the colonies were stained with nitroblue tetrazolium chloride (Sigma Aldrich), imaged using Bio-Rad ChemiDoc MP Imaging System, and quantified with ImageJ.

The proteome profiler human phospho-kinase Array Kit (R&D Systems) was used according to the manufacturer's protocol. Briefly, membranes were incubated with blocking buffer for 1 hour, then incubated with cell lysates at a protein concentration of 400 μg/membrane overnight at 4°C. After incubation, membranes were washed, incubated with detection antibody cocktails for 2 hours at room temperature, rewashed, and incubated with streptavidin–HRP solution for 30 minutes at room temperature. The membranes were then washed again, incubated with Chemi Reagent Mix, imaged with Bio-Rad ChemiDoc MP Imaging System, and quantified with ImageJ.

Nude mice (NCr-Foxn1^nu) were purchased from Taconic. All mice were housed under an aseptic barrier facility and handled in compliance with the Institutional Animal Care and Utilization Committee, Northwestern University (Chicago, IL). Gliomas (2 × 10⁵ cells) were implanted into the right frontal lobes of mice aged 5–6 weeks. Engrafted mice were imaged 1–2 times weekly by bioluminescence imaging (BLI). BLI was performed by subcutaneous injection with 150 mg/kg d-luciferin potassium (Gold Biotechnology) in mice anesthetized via 2.5% isoflurane; each mouse was imaged 10 minutes after luciferin injection on an IVIS SPECTRUM imaging station. Overall survival endpoints were determined by neurologic symptoms or weight loss that exceeded 20% of the original body weight.

Differences between mean values of two groups were compared using two-sample t test, or between multiple groups by one-way ANOVA and post hoc Tukey test; P values less than 0.05 were considered significant. Kaplan–Meier and t tests were performed to compare survival between groups. Graph generation and statistical analyses were performed with GraphPad PRISM 5 software.

Additional methods can be found in the “Supplementary Methods” section.

The three major molecular subdivisions of diffusely infiltrative gliomas are, in order of increasing malignancy: (i) IDH1^mut with 1p/19q codeletion (oligodendrogliomas); (ii) IDH1^mut with intact 1p/19q (astrocytomas); (iii) IDH1^wt (GBM; ref. 22). In an analysis of 656 WHO grade II–IV gliomas in The Cancer Genome Atlas (TCGA) via GlioVis (23), F3 mRNA showed a strong inverse relationship with IDH1^mut in both low-grade (II-III) and high-grade (IV) gliomas (P < 0.001 in both grade II–III and grade IV tumors; Fig. 1A). On average, IDH1^mut gliomas with 1p/19q codeletion expressed slightly less F3 mRNA than IDH1^mut gliomas with intact 1p/19q (Fig. 1A), and also produced less mRNA from the gene encoding the major TF receptor, PAR2 (F2RL1), than either IDH1^wt gliomas or IDH1^mut gliomas with intact 1p/19q (Supplementary Fig. S1A). In patient-derived glioma cells with endogenous IDH1^mut, F3 mRNA and TF protein levels were markedly lower than in patient-derived IDH1^wt glioma cells (Fig. 1B and C), whereas PAR2 was present in all cells (Fig. 1C). All three IDH1^mut cell types expressed R132H IDH1 (Fig. 1C).

F3 contains a methylation-sensitive CpG island in its promoter region (24). F3 was markedly hypermethylated in IDH1^mut glioma cells (TB09, BT142, GBM164) relative to IDH1^wt glioma cells (GBM6, GBM12, GBM43; Fig. 1D), consistent with our prior analysis of TCGA gliomas (10). F3 methylation was highest in the 5′ end of the CpG island, as well as in CpG sites ≥2 kb 5′ to the CpG island, termed the “north CpG island shore.” This area is critical for the methylation-dependent suppression of many genes in cancers (25). To establish a direct causative link between IDH1^mut and F3 methylation, we analyzed publicly available data from human neural stem cells engineered to express IDH1^mut (GEO: GSE94962; ref. 26), and found the same pattern of induced F3 hypermethylation and F3 mRNA reduction as in our IDH1^mut cells and TCGA gliomas (Supplementary Fig. S1D and S1E). Treatment with a demethylating agent, decitabine (DAC), caused a 5.4-fold increase in F3 mRNA in IDH1^mut glioma cells (P = 0.018), but had no effect on F3 mRNA in IDH1^wt glioma cells (P = 0.35; Fig. 1E and F). DAC also increased the TF procoagulant activity (PCA) of IDH1^mut glioma cells by 2.3- to 19.9-fold, but not of IDH1^wt glioma cells (Supplementary Fig. S1F).

It was recently shown that alteration of the 1:1 IDH1^wt:IDH1^mut allele ratio, with subsequent reduction in global CpG methylation, develops in a subset of IDH1^mut gliomas during recurrence and malignant progression (27). Using the publicly accessible data from that study, we analyzed F3 methylation in matched original and recurrent IDH1^mut gliomas (European Genome-Phenome Archive database accession #EGAD00010001408). Of the 3 patients whose IDH1^mut gliomas acquired an allelic imbalance on recurrence, 2 had reduced tumor F3 methylation in the north shore and CpG island, whereas recurrent IDH1^mut tumors that maintained a 1:1 IDH1^wt:IDH1^mut ratio showed little change in F3 methylation (Supplementary Fig. S2A).

IDH1^mut is associated with significantly better survival in gliomas, but not in other cancers that often contain these mutations, including acute myeloid leukemia (AML; ref. 28), chondrosarcoma (29), melanoma (30), and cholangiocarcinoma (31). In comparison with these other types of IDH1^mut cancer, IDH1^mut gliomas showed a 5-fold median increase in F3 methylation (average F3 CpG β-value increase = 0.31 in IDH1^mut gliomas versus 0.06 in all other IDH1^mut cancers; Supplementary Fig. S2B). Likewise, IDH1^mut gliomas showed an 18.7% decrease in F3 mRNA log₂ compared with IDH1^wt gliomas, whereas the other IDH1^mut tumors only showed an average 6.1% decrease relative to their respective IDH1^wt counterparts (Supplementary Fig. S2C). Together, these experimental and in silico results suggest that TF is suppressed in IDH1^mut gliomas through CpG methylation, and that high F3 methylation and TF suppression are relatively unique to IDH1^mut gliomas.

TF-initiated signal transduction is known to enhance the aggressiveness of many cancers. Ectopic expression of TF via F3 cDNA transduction in IDH1^mut cells caused an increase in ERK1/2 and Akt activation, the latter being especially prominent in GBM164 cells (Fig. 2A and B). TF overexpression in IDH1^mut cells increased cell proliferation by 30% and 140% in TB09 and GBM164 cells, respectively (Fig. 2C). TF also increased IDH1^mut cell invasiveness (TB09 = 2.7-fold, P < 0.01; GBM164 = 8.8-fold, P < 0.001; Fig. 2D and E) and anchorage-independent growth (TB09 = 2.4-fold, P < 0.001; GBM164 = 2.6-fold, P < 0.001; Fig. 2F and G).

As is typical for most IDH1^mut gliomas, neither unmodified TB09 nor unmodified GBM164 cells grow as intracranial PDX in athymic mice. F3 transduction and associated ectopic expression of TF greatly increased intracranial engraftment and growth of GBM164 cells, as indicated by BLI (0/10 control vs. 10/10 TF+, P = 0.0007, Fisher exact test; Fig. 2H and I; Supplementary Fig. S3A). However, F3 cDNA had no effect on TB09 cell intracranial engraftment and growth (not shown). Intracranial growth of F3-modified GBM164 was initially slow, with a sharp increase in growth occurring after 3 months, resulting in a median survival of 6 months for animal subjects postengraftment, versus no deaths (and no tumor detected microscopically, not shown) in control GBM164 mice at day 180 (P = 0.02; Fig. 2I and J). Furthermore, while unmodified GBM164 can grow in the mouse flank, F3 cDNA greatly enhanced its engraftment frequency, growth rate, and subcutaneous spread (Supplementary Fig. S3B–S3D).

We next examined the effect of TF knockdown in patient-derived IDH1^wt glioma cells that express high levels of TF (Fig. 1B and C). GBM6 has amplification of the constitutively active EGFRvIII (EGFRvIII^amp), GBM12 has EGFR^wt amplification, and GBM43 has an inactivating NF1 mutation (21, 32). Two different F3-targeted shRNA suppressed TF expression following tumor cell transduction, with construct #2 showing greatest suppression (Fig. 3A and B); F3 shRNA #2 was therefore used in all subsequent experiments. Knocking down TF reduced ERK1/2 and Akt phosphorylation in EGFR-driven GBM6 and GBM12 cells, but had negligible effect on pERK1/2 and pAkt in NF1-driven GBM43 cells (Fig. 3B). TF knockdown reduced cell proliferation by 48.9% in GBM6 and by 55.7% in GBM12, but had no significant effect on GBM43 proliferation (Fig. 3C). However, the invasiveness of all 3 IDH1^wt cell types was reduced by 32%–47% (P < 0.001; Fig. 3D). TF knockdown also reduced anchorage-independent growth by 2.3-fold and 14.6-fold in GBM6 and GBM12 cells, respectively (P < 0.001), whereas GBM43 cells were unable to form colonies in soft agar, even under control conditions (Fig. 3E).

TF knockdown markedly impaired the intracranial growth of IDH1^wt, EGFR-driven GBM6 and GBM12 PDX. The median survival of mice bearing GBM6 was 85% longer when TF was suppressed (77.0 vs. 41.5 days, P = 0.001; Fig. 3F). The effect of TF suppression was even more striking in GBM12 PDX, decreasing the bioluminescence of intracranially injected cells to undetectable levels (Fig. 3G). None of the F3 shRNA GBM12 mice died at 90 days, whereas the median survival for mice bearing GBM12 vector control (vc) shRNA was 41.0 days postengraftment, and all were dead before 60 days (P < 0.001; Fig. 3G). Microscopic analysis at 90 days found no detectable tumors in mice injected with GBM12 expressing F3 shRNA (Supplementary Fig. S3E). However, F3 shRNA had no effect on tumor growth or median survival of mice engrafted with the NF1^mut GBM43 PDX (Fig. 3H). To determine whether the malignant behavior of F3 shRNA-transduced cells could be rescued, knockdown cells were transduced with F3 cDNA-encoding lentivirus resulting in elevated TF expression (F3 rescue; Fig. 3I). Interestingly, F3 rescue increased tumor growth beyond that of corresponding unmodified parental cells, even in GBM43 (Supplementary Fig. S3F and S3G; Fig. 3F–H).

We previously demonstrated that low TF immunostaining correlated with significantly longer overall survival in patients with IDH1^wt GBM (10). Consistent with this, analysis of F3 and survival in IDH1^wt TCGA GBM patients using Cutoff Finder (33) showed that low F3 mRNA correlated with 55% longer median overall survival (20.1 vs. 13.0 months, HR = 0.58, 95% CI = 0.47–0.73, P < 0.0001; Fig. 3J). Low F3 mRNA was also a significant favorable prognostic marker for overall survival in TCGA GBMs (HR = 0.65; 95% CI = 0.47–0.91; P = 0.011), independent of IDH1^mut and patient age (Supplementary Table S4). An analysis of 10,844 PANCAN TCGA cancers showed a similar relationship between lower F3 mRNA and longer survival (7.9 years vs. 5.3 years, HR = 1.7; 95% CI = 1.6–1.9; P < 0.0001; Supplementary Fig. S3H).

We next investigated the effects of TF/F3 manipulation in IDH^mut and IDH^wt gliomas using a human phospho-kinase antibody array. In IDH1^mut TB09 cells, F3 transduction and expression increased cellular β-catenin, as well as phosphorylation of PRAS40, JNK1/2, and MSK1, and decreased phosphorylation of STAT3, p53, HSP60, and WNK1 (Fig. 4A). In IDH1^mut GBM164 cells, TF overexpression increased production of β-catenin and phosphorylation of ERK1/2, HSP60, and Akt1/2/3 (Fig. 4B). TF knockdown in IDH1^wt GBM6 and GBM12 cells lowered expression of β-catenin and reduced phosphorylation of EGFR, ERK1/2, and Akt1/2/3 (Fig. 4C and D), whereas IDH1^wt GBM43 cells only showed lower β-catenin expression and reduced WNK1 phosphorylation (Fig. 4E).

In a reverse-phase protein array (RPPA) analysis of TCGA gliomas, high F3 mRNA (defined by the 75th percentile) correlated with increased expression of proteins, and the phosphorylation of proteins, involved with EGFR (q < 0.0001), mTOR (q < 0.0001), and extracellular matrix receptor (q < 0.05) KEGG signaling pathways, as indicated by pathway enrichment analysis (Fig. 4F). Gliomas with high F3 also showed biological process enrichment for proteins involved in cell migration (q < 0.001), inhibition of anoikis (q < 0.05), and cell survival (q < 0.001; Fig. 4F).

Among IDH1^mut and IDH1^wt cells studied, the most consistent phenotypic and molecular changes resulting from TF modulation were altered invasiveness and β-catenin expression. In an analysis of 670 WHO grade II–IV TCGA gliomas, F3 mRNA strongly correlated with β-catenin mRNA, CTNNB1 (P < 0.001; R² = 0.098; Fig. 5A). As β-catenin is known to promote invasion in many cancers, including glioma (34), we hypothesized that the proinvasive effects of TF are mediated predominately through β-catenin. When unmodified IDH1^wt cells were cultured in serum-free conditions (to eliminate any serum-based TF and FVIIa), β-catenin protein expression increased and peaked at 2–8 hours after treatment with exogenous recombinant TF (recTF) and FVIIa (Fig. 5B). Treatment of IDH1^wt cells for 24 hours with β-catenin inhibitor XAV-939 (35) was only able to partially inhibit cell proliferation in GBM6 and GBM12 at 100 μmol/L, and had no effect on GBM43 proliferation (Fig. 5C). Treatment of IDH1^wt cells with XAV-939 decreased cell viability at the highest concentration for all cells by 29%–53% (P < 0.001), with GBM6 cells showing slightly more sensitivity (Fig. 5D). XAV-939 treatment increased cell apoptosis of GBM6 at 10 μmol/L and 100 μmol/L by 7.2% and 12.6% respectively, but did not affect apoptosis in GBM12 or GBM43 cells (Fig. 5E). In contrast, XAV-939 was much more potent against chemoinvasion in all 3 IDH1^wt cell types (Fig. 5F and G). These findings suggest that the proinvasive effect of TF in gliomas is dependent on β-catenin.

RTKs, such as EGFR, are well known to stimulate downstream TF production (36, 37). However, our phospho-kinase data indicated that TF may also, in turn, contribute to RTK activation (Fig. 4). In serum-free media, exogenous recTF/FVIIa increased pEGFR, as well as downstream pAkt and pERK1/2, in EGFRvIII^amp GBM6 and EGFR^amp GBM12 cells (Fig. 6A). In NF1^mut GBM43 cells, recTF/FVIIa had no effect on pEGFR, a weaker effect on pAkt, and a less durable effect on pERK1/2, especially in comparison with GBM12 (Fig. 6A). The TF/FVIIa complex causes activation of PAR2 and, to a lesser degree, PAR1, PAR3, and PAR4 (38). Of the four major PAR genes, IDH1^wt glioma cells express mainly F2RL1, encoding PAR2 (Supplementary Fig. S4A). A PAR2-specific activating peptide (PAR2-AP) modestly increased cell proliferation and chemoinvasion in GBM6 and GBM43 cells, but only increased chemoinvasion in GBM12 cells (Supplementary Fig. S4B and S4C). PAR2-AP triggered EGFR, Akt, and ERK1/2 phosphorylation to a much greater extent in GBM6 and GBM12 than in GBM43 (Supplementary Fig. S4D). A pharmacologic PAR2 inhibitor, GB83, reduced cell proliferation and chemoinvasion in all 3 IDH1^wt cells (Supplementary Fig. S4F and S4G).

G protein–coupled receptors (GPCR), like PAR2, can transactivate RTKs through autocrine/paracrine mechanisms that involve extracellular release of RTK ligands, or by intracellular, ligand-independent mechanisms involving kinases such as Src (39). To determine whether TF transactivates EGFR through PAR2 in gliomas, the effect of recTF/FVIIa on EGFR kinase activity and phosphorylation was tested in conjunction with the following specific pharmacologic inhibitors: (i) cetuximab, an extracellular inhibitor of EGFR activation; (ii) AG1478, which inhibits intracellular EGFR phosphorylation; (iii) PAR2 antagonist GB83; (iv) Src antagonist SKI-606. Both cetuximab and AG1478 blocked the ability of EGF to activate EGFR, whereas GB83 did not (Fig. 6B). In contrast, both AG1478 and GB83 blocked the effect of recTF/FVIIa on EGFR, but cetuximab did not (Fig. 6B). SKI-606 reduced recTF/FVIIa-induced EGFR activation (Fig. 6C), and only AG1478, not cetuximab, was able to block the enhancing activity of TF/FVIIa on proliferation (Fig. 6D). These results suggest that TF-PAR2 transactivates EGFR through cytoplasmic domain interactions that involve Src, even when an extracellular EGFR inhibitor is present.

Among IDH1^wt gliomas, the extent to which TF-PAR2 activated EGFR correlated with the effect of TF knockdown on in vitro proliferation and in vivo growth (Figs. 3 and 6; Supplementary Fig. S3). F3 cDNA promoted cell proliferation and PDX growth of IDH1^mut GBM164 cells, while having little effect on these properties in IDH1^mut TB09 cells (Fig. 2; Supplementary Fig. S3). We hypothesized that the differential effect of TF on GBM164 versus TB09 may be due to GBM164 expressing one or more RTKs that TB09 does not. On phospho-kinase immunoblots, both TB09 and GBM164 cells showed low basal pEGFR, which did not increase in response to F3 cDNA (Fig. 4A and B). Other major RTKs involved in glioma malignancy, including platelet-derived growth factor receptor alpha (PDGFRα), PDGFRβ, and Met, were extremely low or undetectable in TB09 cells (Fig. 6E). In contrast, PDGFRβ was readily evident in GBM164, and was phosphorylated after TF induction (Fig. 6E). Treatment with exogenous recTF/FVIIa also triggered PDGFRβ phosphorylation in GBM164, with corresponding increases in pAkt and pERK1/2 (Fig. 6F). Treating TF-expressing GBM164 cells with the PDGFRβ inhibitor (sunitinib malate) prior to implantation blocked in vivo growth (Supplementary Fig. S3A), although pretreatment with PDGF-BB ligand was unable to facilitate the growth of GBM164 control cells (not shown). We then sought to determine whether PDGFRβ expression in TB09 cells modified to express TF would increase cell proliferation and/or in vivo growth (Fig. 6G). TB09 cells expressing both TF and PDGFRβ had increased pPDGFRβ (Fig. 6G) and greater proliferation than either in isolation (Fig. 6H), but were still unable to grow in vivo (not shown). Together, these data suggest that TF-PAR2 is capable of intracellular RTK transactivation, and that the effect of TF on cell proliferation and tumorigenesis may depend, in part, on the extent to which RTKs are expressed and activated.

The existence of “Tissue Factor” has been known since the 19^th century, when it was discovered that an unknown factor, present in homogenized extracts from animal tissues, could trigger blood clotting (40). TF is a highly conserved protein, found even in ancient invertebrates like the horseshoe crab, and plays a critical role not just in coagulation, but also in immunity and wound healing (41). High TF activity is common in many cancers, and enhances cell proliferation, invasion, and metastasis. Our data indicate that its consistent suppression in IDH1^mut gliomas contributes to the less aggressive behavior seen in this unique subset of tumors, and that blocking TF signaling could prove effective in suppressing the malignant phenotypes of IDH1^wt gliomas.

IDH1^mut leads to global hypermethylation, a common finding among all IDH1^mut cancers. However, IDH1^mut is only associated with reduced malignancy in gliomas, for reasons that remain unclear. One clue is that the targets of IDH1^mut-related hypermethylation vary greatly depending on the cellular and microenvironmental contexts in which the mutation occurs (42). Strong, consistent hypermethylation and downregulation of F3 appears to be a distinct feature of IDH1^mut gliomas relative to other IDH1^mut tumors (Supplementary Fig. S2B and S2C). Because of events like F3 suppression, IDH1^mut may inadvertently impose limits on glioma malignancy—limits that gradually disappear as the tumor evolves beyond the need for functional IDH1^mut and D2HG (Supplementary Fig. S2A) (27). Demethylating agents like DAC may therefore have undesirable side effects in IDH1^mut gliomas. In this study, we found that treating IDH1^mut TB09 with DAC increased TF levels to that observed for high TF-producing IDH1^wt GBM12 (Supplementary Fig. S1F; of note, the patient from which GBM12 was derived died of pulmonary embolism; ref. 20). However, DAC still resulted in a net decrease in cell proliferation (Supplementary Fig. S1G). This may be due to its relatively broad mechanism of molecular action, wherein it not only demethylates DNA, but also results in stalled replication forks and the induction of DNA damage (43, 44). F3 methylation often diminishes in IDH1^mut gliomas when they lose the normal 1:1 IDH1^wt:IDH1^mut allele ratio and progress in malignancy (Supplementary Fig. S2A). In a patient with paired pretherapy and postmortem glioma samples, the loss of IDH1^mut correlated with markedly increased TF expression and hypercoagulation (Supplementary Fig. S5).

When TF complexes with FVIIa, it can either activate cell-free factor X in the clotting cascade, or it can activate PAR2 on cell surfaces by cleaving PAR2's extracellular tail, leading to intracellular activation of associated G proteins. Under normal circumstances, PAR2 activation stimulates processes involved in wound healing, like cell proliferation, migration, and angiogenesis, but cancers use it to enhance growth and metastasis (13). Indeed, this is one reason why cancer is often referred to as “the wound that does not heal” (45). Prior work by others showed that suppression of TF forces U373 glioma cells into a dormant, undetectable state in vivo for up to a full year, followed by rapid tumor growth after TF restoration (16). Our data indicate that the methylation-dependent suppression of TF in IDH1^mut gliomas could help explain why IDH1^mut gliomas are less malignant, and are difficult to culture and engraft. Along those lines, it would also be of interest to “rescue” F3/TF expression in nonneoplastic progenitor cells engineered to overexpress IDH1^mut (Supplementary Fig. S1D and S1E), and observe the effects on cell behavior. Another approach would be to induce IDH1^mut in an established IDH1^wt cancer cell line. However, as such a cell model is inherently more artificial, it may not have the same biologic relevance as direct manipulation of actual patient-derived IDH1^mut and IDH1^wt glioma cells (46).

In this study, modification of TF expression in IDH1^mut and IDH1^wt cells led to changes in multiple biological processes, including cell migration, adhesion, and proliferation. TF knockdown in IDH1^wt gliomas reduced in vitro proliferation and in vivo growth, but only in cells driven by EGFR. TF knockdown completely prevented the in vivo growth of GBM12, in which amplified EGFR^wt depends on interaction with other proteins for its activation, and TF knockdown delayed but did not prevent growth of EGFRvIII^amp GBM6, which retains some EGFR activity even when TF is suppressed (Figs. 3 and 6A). In contrast, GBM43, in which NF1^mut exerts procancer effects downstream of EGFR, was completely resistant to the antigrowth effects of TF suppression. However, F3 cDNA still managed to enhance GBM43 growth (Fig. 3H) and pAKT and pERK1/2 (Fig. 6A); this may be explained by the differences in RTKs expressed by the three IDH1^wt glioma cells (Supplementary Fig. S4E). Whereas GBM43 does not express EGFR, it does express PDGFRβ, an RTK that is activated by TF. Furthermore, pharmacologic PAR2 inhibition by GB83 inhibited GBM43 cell proliferation (Supplementary Fig. S4F), suggesting that either GBM43 cells still benefit from PAR2 activation through mechanisms independent of EGFR, or that GB83 has off-target effects.

TF has previously been reported to be a downstream effector of EGFR and participates in EGFR signaling (37, 47). But to our knowledge, the current study is the first to provide detailed evidence that it acts as an upstream activator of EGFR and other RTKs through PAR2, even increasing EGFRvIII phosphorylation beyond its constitutively active baseline (Figs. 4 and 6). Of potential clinical relevance is the discovery that TF does this via purely intracellular mediators, and can activate EGFR even in the presence of cetuximab (Fig. 6B). Cetuximab is clinically effective against RTK-driven carcinomas arising in the large intestine and head and neck (48), but it is still not curative, and transactivation of EGFR by TF-PAR2 is a potential mechanism of therapeutic resistance—one that warrants further investigation in those cancers. Our hypothesis that TF is dependent on RTKs to promote proliferation and tumorigenesis is also consistent with the IDH1^mut data, in which PDGFRβ-expressing GBM164 cells benefited much more from TF induction compared with RTK-deficient TB09 cells, in terms of enhanced proliferation and in vivo growth (Figs. 2 and 6). These novel TF–PAR2–RTK activity relationships therefore suggest potential combination therapies for treating IDH1^wt gliomas.

In this study, the most consistent pathway and phenotypic changes in response to TF modulation among all cells was invasiveness, with indication of elevated β-catenin expression as a key downstream effector of TF-stimulated invasion (Figs. 2–5). The proinvasive activity of TF appears to be initiated by and dependent on TF interaction with PAR2, as PAR2-AP and GB83 increased and decreased invasion, respectively (Supplementary Fig. S4C and S4F). Activated PAR2 is known to stabilize β-catenin (49), and β-catenin, in turn, is known to facilitate cell–cell adhesion and cell–matrix adhesion during cancer cell invasion (50). Consistent with this, analysis of TCGA gene expression data revealed that ECM-receptor interactions positively correlate with glioma F3 mRNA (Fig. 4F). Thus, the progrowth effects of TF may depend primarily on RTK interactions, whereas the proinvasive effects of TF appear to be predominately mediated through β-catenin.

In sum, these data demonstrate that the activation of PAR2 by the TF/FVIIa complex has two major consequences in IDH1^wt gliomas: (i) promotion of cell proliferation and tumor formation through intracellular stimulation of RTK; (ii) β-catenin–dependent upregulation of invasiveness (Supplementary Fig. S6). Because F3 hypermethylation occurs to a much greater extent in IDH1^mut gliomas than in other IDH1^mut cancers, and IDH1^mut has only been consistently shown to be a favorable prognostic marker in gliomas, not other cancer types, TF suppression may be central to the lesser malignancy of IDH1^mut gliomas. Clinically, this work has multiple implications, including: (i) counteracting IDH1^mut activity in gliomas, either directly by enzyme inhibition or indirectly through demethylating drugs, might paradoxically increase tumor aggressiveness and VTE risk; (ii) TF-PAR2 is a potential mechanism of resistance to extracellular EGFR inhibitors currently in use against other cancers; (iii) PAR2 may be an effective anticancer target, without disrupting the normal hemostatic activity of TF.

No potential conflicts of interest were disclosed.

Conception and design: D. Unruh, C.D. James, C. Horbinski

Development of methodology: D. Unruh, C.D. James, C. Horbinski

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): D. Unruh, M. Drumm, Y.D. Li, Q.F. Haider, A.M. Saratsis, J.N. Sarkaria, C. Horbinski

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): D. Unruh, B. Wray, M. Drumm, J. Lamano, Y.D. Li, Q.F. Haider, A.M. Saratsis, C. Horbinski

Writing, review, and/or revision of the manuscript: D. Unruh, S. Mirkov, A.M. Saratsis, D.M. Scholtens, J.N. Sarkaria, C.D. James, C. Horbinski

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S. Mirkov, R. Javier, K. McCortney, C.D. James, C. Horbinski

Study supervision: C. Horbinski

Other (performed in vitro and in vivo experiments): S. Mirkov

We thank Dr. Hai Yan for the TB09 cells, Lisa Magnusson for her assistance with the glioma PDX models, and Alicia Steffens for IHC and histochemical staining. This work was supported by NIH grants K08CA155764 and R01NS102669 (to C. Horbinski). D. Unruh was supported by NIH grants T32CA070085 and F32CA216996. C. Horbinski, D.M. Scholtens, and C.D. James were also supported by the Northwestern SPORE in Brain Cancer P50CA221747. The Mayo Clinic Brain Tumor Patient-Derived Xenograft National Resource is supported by Mayo Clinic, the Mayo SPORE in Brain Cancer P50CA108961 and NIH grant R24NS92940. Histology and fluorescent microscopy services were provided by the Mouse Histology and Phenotyping Laboratory and the Center for Advanced Microscopy/Nikon Imaging Center, respectively; both are supported by NCI P30CA060553 awarded to the Robert H. Lurie Comprehensive Cancer Center.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.