Cutting Edge: ERK1 Mediates the Autocrine Positive Feedback Loop of TGF-β and Furin in Glioma-Initiating Cells

Glioblastoma is the most common malignant primary brain tumor and is largely refractory to current therapies (1). TGF-β has long been attributed a major role in the malignant phenotype of glioblastoma (2, 3). Cytostatic responses to TGF-β are selectively inhibited in glioblastoma cells (4) with TGF-β sustaining the migratory and invasive potential as well as suppressing antitumor immune surveillance. Glioma-initiating cells (GIC) have been proposed as the cell population responsible for tumor initiation and recurrence of gliomas (5), and have been attributed immunosuppressive properties (6, 7). TGF-β may maintain the stem cell–like properties and tumorigenic activity of GIC (8–11). TGF-β binding to the TGF-β receptor complex leads to the activation of both SMAD transcription factors (canonical/SMAD-dependent pathway) (12) and various SMAD-independent pathways including the PI3K/AKT pathway (13), the MEK (MAPK)/ERK, and the p38 and JNK pathways (14). Persistent activation of the TGF-β pathway in glioblastoma has been described as an autocrine loop involving TGF-β2, CREB1, SMAD3, and the PI3K/AKT pathway (2). Processing of pro–TGF-β1 and pro–TGF-β2 in the cytoplasm or in the extracellular matrix in glioma cells involves proprotein convertases subtilisin/kexin (PCSK) (15). PCSK are a class of nine enzymes with a fundamental role in the processing of diverse protein precursors (16). PCSK3, generally known as furin or paired basic amino acid-cleaving enzyme, is widely expressed and responsible for most of the processing events in the constitutive secretory pathway. Furin is located in a variety of cellular compartments including the endoplasmatic reticulum, trans-Golgi network, and cell surface (17). It is also considered the major endoprotease responsible for processing TGF-β at its (R/K)-2nX-R↓ site (where n = 0–3 aa) (18). TGF-β has been reported to induce expression of its converting enzyme furin in fibroblasts (19), epithelial cells (20), and hepatocellular carcinoma cells (21) involving SMAD, MAPK ERK1/p44, and ERK2/p42. This led us to explore the regulation of furin by these pathways in human GIC and specifically the existence and regulation of a potential autocrine TGF-β–furin loop as a potential therapeutic target for anti–TGF-β treatment in glioblastoma.

LN-308 and LN-229 kindly provided by N. de Tribolet (Lausanne, Switzerland) were cultured in DMEM (Life Technologies/Thermo Fisher, Madison, WI) supplemented with 1% l-glutamine (Life Technologies) and 10% FBS (Life Technologies). The human GIC T-325, T-269, ZH-161, S-24, and ZH-305 (22) were cultured in phenol red-free Neurobasal medium (NBM) (Life Technologies), supplemented with 2% B-27 without vitamin A (Life Technologies) (20 μl/ml), 1% l-glutamine (Life Technologies), and fibroblast growth factor and epidermal growth factor (FGF/EGF) (20 ng/ml each) (PeproTech, London, U.K.). Treatment of cells was performed in NBM containing 1% l-glutamine and FGF/EGF and the following reagents: PCSK inhibitor decanoyl-RVKR-CMK (3501; Tocris Bioscience, Bristol, U.K.), recombinant human TGF-β2 (R&D Systems, Minneapolis, MN), SD-208 (Scios, Fremont, CA), and MEK1/2 inhibitor U0126 (9903; Cell Signaling Technology, Danvers, MA).

Lentivirally-mediated gene silencing of furin was performed as described (23), using a furin specific short hairpin RNA (V3LHS_310002) cloned in a pGIPZ lentiviral vector and the respective pGIPZ vector carrying a non-targeting control (RHS4349) (Dharmacon, Lafayette, CO).

Transient gene silencing was performed by electroporation (Neon Transfection System, Invitrogen/Thermo Fisher) using 100 nM of the following ON-TARGET plus, small interfering RNA (siRNA) SMART pools from Dharmacon: TGF-β1 (L-012562-00), SMAD2 (L-003561-00), SMAD3 (L-020067-00), SMAD4 (L-003902-00), ERK1 (L-003592-00), ERK2 (L-003555-00), and non-targeting control (D-001810-10).

Human PBMC were prepared from healthy donor blood and cultured in RPMI 1640 (Life Technologies) supplemented with 10% FBS (Life Technologies), 20 mM l-glutamine, 50 μM β-mercaptoethanol (Sigma, St. Louis, MO), 10 U/ml penicillin (Sigma), 10 μg/ml streptomycin (Sigma), 10 mM sodium-pyruvate (Sigma), and 1 × MEM nonessential amino acids (Life Technologies). Media from GIC seeded at 10⁷ in 10 ml NBM for 72 h were concentrated 2.5-fold prior to addition to the PBMC. On day 0 PBMC were treated with 3 μg/ml concanavalin A (Sigma), 20 U/ml IL-2 (PeproTech), and GIC conditioned media. At day 3, treatment with IL-2 and exposure to GIC media were repeated, and at day 6 addition of IL-2 was repeated. On day 7 cells were analyzed by flow cytometry.

Immunoblot analysis was performed as previously described (23). Primary Abs were anti-furin (sc-20801; Santa Cruz Biotechnology, Dallas, TX), anti–β-actin (sc-1616; Santa Cruz), anti-human LAP/TGF-β1 (AF-246-NA; R&D Systems), anti–TGF-β1 (G1221; Promega, Madison, WI), anti–TGF-β2 (ab36495; Abcam, Cambridge, U.K.) and Abs from Cell Signaling Technology specific for SMAD1 (9743), pSMAD1/5 (Ser463/465) (9516), SMAD2 (3122), pSMAD2 (Ser456/467) (3108), SMAD3 (9513), pSMAD3 (Ser423/425) (9520), SMAD4 (9515), p44/42 MAPK (ERK1/2) (9102), pp44/42 MAPK (ERK1/2) (Thr202/Tyr204) (9106), AKT (9272), pAKT (Ser473) (9271), p38 MAPK (9212), and pp38 MAPK (Thr180/Tyr182) (9211). Secondary Abs were HRP-coupled goat anti-rabbit (sc-2004), donkey anti-goat (sc-2033) (Santa Cruz Biotechnology) or sheep anti-mouse Abs (NA931V; GE Healthcare, Amersham, U.K.).

RT-PCR was performed by using the Δ threshold cycle method with ARF1 as a housekeeping gene (see Supplemental Table I for primers).

Cellular and extracellular furin-specific activity (FSA) was measured as described (24), using the anti-furin Ab MAB15032 (R&D Systems) as capturing Ab in lysates and conditioned media of cells seeded at a density of 5 × 10⁶ cells in 7 ml NBM supplemented with l-glutamine and FGF/EGF for 48 h. FSA is expressed as the fluorescence intensity (λexc = 380 nm, λem = 460 nm) normalized to protein concentration as determined by Bradford assay (Bio-Rad, Hercules, CA). FSA was measured in concentrated conditioned media mixed with the appropriate amount of reaction buffer and analyzed in the same manner as previously described for the lysates (24).

APC-CD25 (17-0259) and PE-FOXP3 (12-4776) Abs were from eBioscience (San Diego, CA), FITC-CD4 (55346) from BD Biosciences (San Jose, CA). A FOXP3 fixation/permeabilization kit (eBioscience) was used for FOXP3 staining. Cells were analyzed with FACSVerse (BD Biosciences) and data analyzed with FlowJo software. Fluorescence minus one controls using fluorochrome-conjugated isotype control Abs from BD Biosciences (FITC-mouse IgG1, κ [555748], APC-mouse IgG1, κ [555751] and PE-rat IgG2A, κ [553930]) were used.

Seven of the nine PCSK, i.e., PCSK1, PCSK2, PCSK4, PCSK5, PCSK6, PCSK7, and furin, process their substrate at the consensus recognition sequence (R/K)X_n(R/K)↓, which is also the recognition site for processing TGF-β. We first investigated mRNA expression levels of these enzymes in a panel of five human GIC: T-325, T-269, ZH-161, S-24, and ZH-305. Furin was expressed in all cell lines and was one of the most abundant PCSK (Fig. 1A). Furin was detectable intracellularly in cell lysates as a double band of around 100 kDa, corresponding to the two glycosylated/sialylated forms of mature furin (25) except for T-269, where only the higher molecular mass isoform was detected. In the conditioned media, furin was detectable as a major band with an apparent molecular mass of around 90 kDa (Fig. 1B).

To investigate the processing of TGF-β1/2 by furin, lentiviral gene silencing of furin was performed in T-325, ZH-161, and ZH-305 cells, further referred to as shfurin cells (Fig. 1C–E). Furin gene silencing did not affect any other PCSK with the exception of PCSK6 in ZH-305 shfurin cells (data not shown). Decreased protein levels of both cellular and secreted furin (Fig. 1C) and a reduction in cellular and extracellular FSA (Fig. 1D) were confirmed in all furin knockdown cell lines. We then analyzed TGF-β1/2 processing in shfurin cells. For TGF-β1 we used an Ab to the N terminus of TGF-β1 detecting both pro–TGF-β1 (55 kDa) and the latency associated peptide of TGF-β1 (37 kDa), and an Ab to the C terminus detecting mature TGF-β1 (12.5 kDa) (see ZH-161/siTGF-β1 as a control, Fig. 1C). For TGF-β2 we used an Ab reacting with the C terminus of TGF-β2 detecting both pro–TGF-β2 (55 kDa) and mature TGF-β2 (12.5 kDa). Furin gene silencing resulted in increased levels of pro–TGF-β1 and correspondingly decreased levels of the two respective processing products latency associated peptide/TGF-β1 and mature TGF-β1 in ZH-161 and ZH-305, with TGF-β1 not being detectable in T-325 (Fig. 1C). In T-325 and ZH-305, showing detectable TGF-β2 protein levels, furin gene silencing increased levels of pro–TGF-β2 and decreased levels of mature TGF-β2. The effects of furin gene silencing on pro–TGF-β1/2 processing were similar to those obtained with the pan-proprotein convertase inhibitor decanoyl-RVKR-CMK (PCSK inhibitor), suggesting that furin is the main PCSK involved in pro–TGF-β1/2 processing here. Notably, in T-325 and ZH-161, pro–TGF-β processing takes place to a remarkable extent extracellularly as pro–TGF-β levels increased in shfurin cells and upon addition of the PCSK inhibitor in the conditioned media, but remained unchanged in the lysates. In ZH-305, the levels of pro–TGF-β2 increased both in cell lysates and in conditioned media following furin gene silencing, suggesting that pro–TGF-β2 processing takes place in part in the intracellular compartment in ZH-305 cells, too (Fig. 1C).

To confirm that reduced TGF-β levels affect downstream signaling activities, we analyzed the levels of pSMAD2, pSMAD3, and pSMAD1/5 as readouts for the activation of canonical TGF-β signal transduction, and of pAKT (Ser473), pERK1/2, and pp38 reflecting non-canonical TGF-β signaling (Fig. 1E). All shfurin cells showed reduced levels of pSMAD2 and pSMAD3. In ZH-161 and ZH-305, pSMAD1/5 was also reduced, pointing toward a reduction in basal bone morphogenetic protein signaling upon furin gene silencing, too. Indeed, inhibiting TGF-β by the TGF-β RI/activin receptor–like kinase (ALK)-5–specific inhibitor SD-208 did not decrease endogenous pSMAD1/5 levels in GIC, although the increase in pSMAD1/5 by exogenous TGF-β was blocked (26). The phosphorylation of AKT (Ser467) and p38 was reduced in T-325 shfurin cells, and ZH-305 shfurin cells showed a reduction in the phosphorylation of ERK and p38 as well.

TGF-β exerts several immunosuppressive functions, including the promotion of FOXP3⁺ regulatory T (Treg) cells. Previous studies have demonstrated that GIC induce Treg generation in vitro (6). Accordingly, we treated human PBMC with GIC-conditioned media and confirmed the induction of CD4⁺CD25⁺FOXP3⁺ cells: conditioned media of ZH-161 cells increased the fraction of CD4⁺CD25⁺FOXP3⁺ cells from 8.9 to 14.8 ± 0.1%. TGF-β2 treatment was used as a positive control. To test whether the inhibition of TGF-β processing affected the ability of GIC to induce Treg, we exposed PBMC to media derived from GIC treated with the PCSK inhibitor. This led to a decrease in the induction of the CD4⁺CD25⁺FOXP3⁺ population from 8.9 to 11.1 ± 0.5% (Fig. 1F), indicating that indeed the inhibition of TGF-β processing in GIC may reduce their immunosuppressive properties. Similar results were obtained for ZH-305 (data not shown).

Autocrine production of TGF-β may be necessary to maintain GIC stemness and TGF-β signaling in the tumor microenvironment (27); however, the mechanisms of how GIC maintain high TGF-β activity are not well understood (2). We therefore evaluated the effect of TGF-β2 on furin levels. Indeed, TGF-β2 induced the expression of furin and FSA in all cell lines except T-269 (Fig. 2A–C). To investigate whether the expression of PCSK other than furin was affected by TGF-β2, we analyzed the mRNA levels of the other six PCSK in ZH-161 (Fig. 2D), T-325, and ZH-305 cells (data not shown) treated with TGF-β2. Indeed, none of them were changed, pointing toward a specific effect of TGF-β2 on furin expression. Concentration and time dependency of furin induction was investigated in ZH-161 cells (Fig. 2E–G). Because we observed an induction on protein levels after 24 h (Fig. 2F) and the maximum effect was achieved with 2.5 ng/ml TGF-β2 (Fig. 2G), we selected these conditions for future experiments. We proceeded to study the signal transduction mechanisms involved in the regulation of furin by TGF-β2. Exposure of T-325, ZH-161, and ZH-305 to the ALK-5 inhibitor SD-208 had no effect on constitutive furin protein levels, but abolished TGF-β2–mediated furin induction (Fig. 2H). The transient gene silencing of SMAD2, SMAD3, or SMAD4 did not abrogate the induction of furin expression by TGF-β2, indicating involvement of SMAD-independent signal transduction (Fig. 2I).

Indeed, blocking the MEK1/2 branch of the non-canonical TGF-β pathway by U0126 decreased constitutive furin levels and attenuated the increase in furin levels in response to TGF-β2. U0126 also decreased constitutive furin levels in T-269 cells that are the TGF-β2/furin–non-responsive model (Fig. 3A). The inhibitory effect of U0126 on the induction of furin expression by TGF-β2 translated into a reduction in FSA (Supplemental Fig. 1A). Similarly, the selective gene silencing of ERK1, but not of ERK2, reduced furin basal levels in all cell lines and attenuated the induction of furin by TGF-β2 (Fig. 3B, 3C). Gene silencing of ERK1, but not of ERK2, reduced furin on mRNA levels in both basal conditions and upon TGF-β2 treatment (Supplemental Fig. 1B). To address whether the control of furin by ERK1 and TGF-β2 is associated with a glioma stem cell–like phenotype, we examined the same GIC cultured in differentiating conditions (Supplemental Fig. 1C) and the long-term glioma cell lines, LN-308 and LN-229 (Supplemental Fig. 1D). In both cases ERK1 gene silencing reduced basal furin levels and attenuated the induction of furin by TGF-β2, suggesting that this molecular pattern is shared between cells with stemness properties and more differentiated tumor cells. The reduction in furin levels following ERK1 gene silencing was associated with a reduction in pro–TGF-β1/2 processing, with increased pro–TGF-β levels and decreased mature TGF-β levels (Supplemental Fig. 1E). MEK1/2 inhibition did not significantly affect the number of spheres, but the spheres formed by GIC (T-325, ZH-161, and ZH-305) treated with U0126 were smaller, indicating an inhibitory effect on cell proliferation as confirmed by reduced MTT metabolism (Fig. 3D). This rather implies a role for the TGF-β–ERK1–furin loop in sustaining the expansion of more differentiated, i.e., progenitor-like glioma cells, rather than in directly promoting GIC sphere formation. Targeting ERK1/2 also attenuated the immunosuppressive properties of GIC because treatment of ZH-161 and ZH-305 cells with U0126 decreased the induction of CD4⁺CD25⁺FOXP3⁺ cells by conditioned media of ZH-161 (Supplemental Fig. 1F) and ZH-305 (data not shown). Thus, our findings reveal that the MEK1/2 branch of the non-canonical signaling pathway, often deregulated in glioma stem cells (28), plays a pivotal role in the control of furin levels in GIC via ERK1. Altogether, we report that TGF-β induces furin activity in GIC and we specifically identify the ERK1 pathway as an essential to maintain furin activity. Our findings indicate that the disruption of ERK1 signaling would be a therapeutic strategy combating TGF-β activity in GIC.

M.W. has received research grants from Acceleron, Actelion, Alpinia Institute, Bayer, Isarna, Merck Sharp & Dohme, Merck & Co., Novocure, Piqur Therapeutics, and Roche. He has received honoraria for lectures or advisory board participation from Celldex, Immunocellular Therapeutics, Isarna, Magforce, Merck Sharp & Dohme, Merck & Co., Northwest Biotherapeutics, Novocure, Pfizer, Roche, and Teva. The other authors have no financial conflicts of interest.