Biological Role and Therapeutic Targeting of TGF-β₃ in Glioblastoma

Transforming growth factor (TGF)-β is a pleiotropic cytokine with multiple effects on cellular behavior including proliferation, migration, invasion, angiogenesis, and immune responsiveness. The three TGF-β isoforms, TGF-β₁, TGF-β₂, and TGF-β₃, exhibit high levels of similarity in their amino acid sequences; however, the tertiary structure of the active domain of TGF-β₃ differs from TGF-β₁ and TGF-β₂, enabling steric rearrangements which allow a more flexible binding to TGF-β receptor II (TβRII; refs. 1–5). TGF-β₃ has an isoform-specific function in embryonic palate fusion and wound healing (6–8). The deficits of TGF-β₃ knockout mice are almost exclusively restricted to impaired palate fusion and pulmonary abnormalities (9). In contrast, mice with a gene knockout for TGF-β₁ or TGF-β₂ exhibit multi-organ deficits, TGF-β₁ especially in the hematopoietic and vasculogenic system and TGF-β₂ in multiple embryonal developmental processes, including the development of heart, lung, neurons, and bones (10, 11). The different phenotypes of TGF-β isoform knockout mice point toward non-overlapping isotype-specific functions. Reports on TGF-β₃ in cancer are almost exclusively restricted to expression analyses and correlative studies without revealing an isotype-specific functional role (5).

Aberrant TGF-β signaling is considered to be a hallmark for the malignant phenotype of glioblastoma. All the three TGF-β isoforms, TGF-β₁, TGF-β₂, and TGF-β₃ are expressed in malignant gliomas in vivo (12, 13). However, pathogenic effects in glioblastoma have been only attributed to the isoforms TGF-β₁ and TGF-β₂ while there are no functional data for TGF-β₃ (14–16).

Molecular subtyping has identified TGF-β₃ as a gene highly expressed in the classical glioblastoma subtype (17). Still, the functional role of TGF-β₃ in the malignant phenotype of glioblastoma remains uncertain, and its mRNA expression levels in vivo are lower than those of TGF-β₁ and TGF-β₂ (13). Here, we characterize the expression and biological activity of TGF-β₃ using isoform-specific oligonucleotides in human and murine glioma models.

Nine human long-term malignant glioma cell lines (LTC), obtained in 1994, were previously described (18) and sent for authentication tests to the German Biological Resource Centre DSMZ in Braunschweig, Germany, in November 2013. The spontaneous murine astrocytoma (SMA) cell lines (SMA-497, SMA-540, and SMA-560) kindly provided by Dr D. Bigner (Durham, NC) were previously characterized (19–21). Murine GL-261 cells were obtained from the National Cancer Institute (Frederick, MD). Five human glioma-initiating cell (GIC) lines, established after informed consent and approval of the local ethics committees have been previously described (22–24).

Human and murine LTC were cultured in Dulbecco's modified Eagle's medium supplemented with 10% fetal calf serum. GIC were maintained in Neurobasal Medium supplemented with B-27 (20 μL/mL) and glutamine (10 μL/mL) from Invitrogen, fibroblast growth factor-2 and epidermal growth factor (20 ng/mL each; Peprotech). Recombinant TGF-β₁, TGF-β₂, and TGF-β₃ (R&D) were used as indicated. TGF-β₃-targeted oligonucleotides ISTH2020, ISTH2023, and control oligonucleotide C3_ISTH0047 were designed and provided by Isarna Therapeutics. ISTH2020 (sequence TTTGTTTACACTTCC) and ISTH2023 (sequence GAGTTTTTCCTTAGG) represent fully phosphorothioate locked nucleic acid (LNA)-modified antisense oligonucleotides (Supplementary Fig. S1). For overexpression of TGF-β₃, a plasmid containing an untagged human cDNA clone of TGF-β₃ (NM_003239) transfected in a pCMV6-XL5 vector was used (SC118071, Origene). Lipofection-based transfections were done with Lipofectamine RNAimax and Opti-MEM (Invitrogen).

For lipofection-aided transfections, LTC were transfected at subconfluent conditions in serum-containing medium at a density of 75,000 cells/cm² by adding a transfection mix of transfection reagent and oligonucleotides prepared in Opti-MEM medium. After 12–24 hours, cells were washed and exposed to serum-free medium for 24–120 hours as indicated.

For gymnotic transfections, no transfection reagent was used. LTC were seeded at low densities (13,000 cells/cm²) and treated after 6–12 hours with the indicated concentrations of oligonucleotide in full serum-containing medium. On the third day after seeding, full medium supplemented with oligonucleotide was renewed. On the seventh day, the medium was removed and the cells were exposed to serum-free medium.

Cells were seeded and transfected as described. To obtain counts for dead and viable cells, cell culture supernatant was removed, cells were detached and counted including the cells of the supernatant in an automatic trypan blue-based cell counter (Vi-Cell, Beckman Coulter Inc.) in 3 or 4 replicates.

Supernatants were generated in serum-free medium after the indicated time points and centrifuged to remove cellular debris. Supernatants were concentrated using a centrifugal filter device (3 kD cutoff; Millipore). Whole-cell lysate preparation and quantification of protein levels were performed as previously described (23). After SDS-PAGE under reducing conditions, proteins were transferred to nitrocellulose membranes (Bio-Rad) and blocked in Tris-buffered saline containing 5% skim milk and 0.1% Tween 20 following antibody incubation (antibody information see supplementary methods).

Visualization of protein bands was performed with horseradish peroxidase (HRP)-coupled secondary antibodies (Santa Cruz Biotechnology) and enhanced chemoluminescence (Pierce/Thermo Fisher). Quantification of bands for correlation analyses was done using ImageJ software (Open Source).

mRNA extraction was done using the NucleoSpin RNA II system (Macherey-Nagel) including DNase treatment. cDNA was prepared using a cDNA reverse transcription kit (Applied Biosystems).

Gene expression was determined via real-time PCR (23) using ADP-ribosylation factor 1 (ARF-1) as a housekeeping gene with the ΔC_TT method for relative quantification. Primer sequences are indicated in the supplementary material section.

For invasion assays, glioma spheroids were generated by incubating 1,500 cells for 72 hours in full medium in 96-well plates precoated with 1% Noble Agar (Becton Dickinson). Thereafter, spheroids were embedded into a collagen matrix containing collagen type I in full medium at neutral pH in a 96-well plate. Sprouting of spheroids was monitored by photographs. For quantification, the median invaded distance of 30 cells was assessed using ImageJ software. The spheroid margin at the corresponding time point was used as a reference for measurement of the invaded distance of sprouting cells.

In vivo studies were performed as previously described (25). In brief, VM/Dk mice (Charles River) were stereotactically implanted into the right striatum with 5 × 10³ SMA glioma cells, C57Bl/6 mice with 2 × 10⁴ GL-261 cells and immunodeficient Crl:CD1-Foxn1^nunude mice (Charles River) with 10⁵ cells human LN-308 glioma cells in a volume of 2 μL of phosphate-buffered saline (Gibco, Life Technologies). Animal studies were approved by Cantonal Veterinary Office Zurich and Federal Food Safety and Veterinary Office (Permission number 202/2012). For ex vivo expression analyses without systemic treatment, the tumor or the non–tumor-bearing left hemisphere, considered as “normal brain,” were subjected to mRNA extraction on day 12 after implantation.

Systemic treatment with oligonucleotides was performed by subcutaneous injections as indicated. Brains were collected upon euthanization, embedded in cryomoulds in Shandon Cytochrome yellow (Thermo Scientific) and frozen. For histology, brains were cut in 8-μm sections using a Microm HM560 (Microchom HM560, Thermo Scientific) and every 20th section was stained with hematoxylin and eosin (H&E; ref. 25). Tumor volumes were determined using an approximation based on ellipsoid geometric primitive (26). For analysis of invasiveness in vivo, we counted satellite lesions as defined by tumor cell aggregates >10 cells at least 3 cell layers distant from the main tumor bulk on all H&E-stained sections.

The Cancer Genome Atlas (TCGA) database and the R2 microarray analysis and visualization platform (http://hgserver1.amc.nl/cgi-bin/r2/main.cgi#, available on February 26, 2016) were used to perform survival analyses within the glioblastoma data set containing gene expression data.

Representative experiments, commonly performed three times with similar results, are shown. Statistical calculations were done using the software of GraphPad Prism Version 5 including two-sided unpaired t test (comparison of two groups), one-way ANOVA and Bonferroni post-hoc testing (multiple comparisons) and Spearman correlation coefficient for correlation analyses. A P value of 0.05 was considered statistically significant.

To assess whether TGF-β₃ expression correlates with survival, we used the database of the TCGA containing data on mRNA expression levels and clinical course of more than 500 patients with glioblastoma. High TGF-β₃ expression was associated with poor survival (13). We next asked whether this association varies within the molecular subtypes classified by Verhaak and colleagues (17). Stratifying patients into these subgroups, data of 82 patients with tumors of the neural, 88 of the proneural, 146 of the mesenchymal and 142 of the classic subtype were available (Supplementary Fig. S2). Notably, in the group of the neural subtype the survival benefit for patients with lower TGF-β₃ mRNA expression was highly significant. This was true both when segregating groups into high and low expression using the cutoff resulting in the highest association (P = 0.01, Supplementary Fig. S2A) and when using the median expression level as cutoff (P = 0.003; Supplementary Fig. S2B).

We next characterized our panel of 14 human glioma cell lines (9 LTC and 5 GIC) and 4 murine glioma cell lines for TGF-β₃ expression. TGF-β₃ mRNA was expressed in all human cell lines without apparent differences between LTC and GIC (Fig. 1A). TGF-β₃ mRNA expression levels were much lower than those of TGF-β₁ or TGF-β₂ in 8 of 9 LTC whereas GIC did not show this pattern (Supplementary Fig. S3A, data on human TGF-β_1,2 mRNA have been published previously, ref. 23). Murine cell lines expressed TGF-β₃ mRNA, too (Fig. 1B). For the murine cell lines, we further assessed the expression levels ex vivo, that is, we implanted the cells orthotopically in syngeneic mice and extracted mRNA from the non–tumor-bearing left hemisphere considered as normal brain and from the tumor after removal from the right hemisphere. Mean levels of TGF-β₃ mRNA in tumoral ex vivo samples were more than 3-fold higher in SMA-540 and SMA-560, comparable in GL-261 and more than 5-fold lower in SMA-497 compared with normal brain (Fig. 1C). SMA-497 showed the highest mRNA expression in vitro, but the lowest ex vivo. Comparing all TGF-β isoforms in vitro, TGF-β₁ was the predominant isoform in SMA-540 and SMA-560, while no significant difference between the isoforms was detected in SMA-497 and GL-261. Ex vivo, both TGF-β₁ and TGF-β₃ were preferentially expressed in SMA-560, TGF-β₃ was higher than TGF-β₂ in SMA-540 while in line with the expression profile in vitro, SMA-497 and GL-261 did not show significant differences between the isoforms (Supplementary Fig. S3B and S3C, data on murine TGF-β_1,2 mRNA expression have been published previously (27)).

We next analyzed TGF-β₃ protein levels in whole cell lysates and cell culture supernatants, which turned out to be challenging. The best antibody we identified (ab15537) still had some cross-reactivity with TGF-β₂ as assessed by immunoblot loaded with recombinant TGF-β_1/2 (Supplementary Fig. S4). Still, we considered the detected protein bands as a valid signal because treatment with TGF-β₃-specific oligonucleotides reduced the bands at 50 kDa and 12.5 kDa, the presumable proform and active form of TGF-β₃. Furthermore, these bands increased upon overexpression of TGF-β₃ (Fig. 1D). In whole cell lysates, the pro-form of TGF-β₃, but not the active form, was detected in all cell lines, with the highest levels in T98G, U87MG, and LN-308 among LTC, and in T-325 among GIC (Fig. 1E). In cell culture supernatants, the pro-form and the active forms of TGF-β₃ were detected, with highest levels in LN-428 and T98G, and lower levels in GIC (Fig. 1F). TGF-β₃ mRNA levels neither correlated with TGF-β₃ protein in whole cell lysates nor in the supernatant. However, in the cell culture supernatants, TGF-β₃ proform levels correlated with its active form (r = 0.64, P = 0.01).

In whole cell lysates of mouse glioma cells, the levels of the pro-form of TGF-β₃ were similar while levels of the active form were again not detected (Fig. 1G). In cell culture supernatants, SMA-497 showed the highest levels of TGF-β₃ (pro- and active form) with low levels of active TGF-β₃ in the other cell lines (Fig. 1H).

We next confirmed the responsiveness to exogenous TGF-β with regard to isotype-specific functions at the level of canonical signaling. LN-18, U87MG, T98G, LN-308, or LN-229 cells were exposed to increasing concentrations of TGF-β_1/2/3. All cell lines showed a concentration-dependent induction of pSMAD2 by all TGF-β isoforms (Fig. 1I). No isoform-specific effect was identified.

We next analyzed the activity and specificity of TGF-β₃-targeted oligonucleotides (ISTH2020 and ISTH2023) in the human LTC LN-308 characterized by high endogenous TGF-β₁ and TGF-β₂ levels and high constitutive pSMAD2 and pSMAD1/5 phosphorylation (23, 28). We applied lipofection-based transfections in a nanomolar range and gymnotic transfection without transfection reagent in a micromolar range. Both oligonucleotides downregulated TGF-β₃ mRNA in a time- and concentration-dependent manner (Fig. 2). Both at 24 and 72 hours, TGF-β₃ mRNA was reduced by more than 75% at 25 nmol/L of ISTH2020 or ISTH2023 using lipofection-based transfection. Of note, in this model, mRNA levels of TGF-β₁ and TGF-β₂ are more than 100-fold higher than those of TGF-β₃. TGF-β₁ and TGF-β₂ were not reduced by more than 50% upon treatment with TGF-β₃-targeted oligonucleotides (Fig. 2A). With the gymnotic transfection method, mimicking the systemic administration of the oligonucleotides in vivo, ISTH2020 or ISTH2023 at 5 μmol/L led to more than 95% reduction of TGF-β₃ mRNA at day 6 and more than 80% reduction at day 8 of treatment, without affecting TGF-β₁ or TGF-β₂ mRNA by more than 2-fold (Fig. 2B). The effect of the oligonucleotides on TGF-β₃ was concentration-dependent with a reduction of more than 50% of TGF-β₃ mRNA up to 12.5 nmol/L (ISTH2020) and 25 nmol/L (ISTH2023; Fig. 2C) with lipofection and up to 0.6 μmol/L (ISTH2023) with gymnotic treatment (Fig. 2D). ISTH2020 and ISTH2023 showed comparable activity in the mouse glioma cell line SMA-560 with up to 60% reduction of TGF-β₃ mRNA at 20 nmol/L at 24 and 96 hours after lipofection-aided transfection (Fig. 2E) and up to 90% reduction of TGF-β₃ mRNA with gymnotic transfection at 5 μmol/L on day 6 (Fig. 2F). With lipofection-based transfection at 50 nmol/L, there was no effect on cell viability or proliferation between 24 and 96 hours after transfection (Supplementary Fig. S5A). Similarly, with the gymnotic transfection method at 2.5 μmol/L, there were neither significant effects on cell viability nor proliferation between day 4 and 8 of treatment (Supplementary Fig. S5B).

We next asked whether treatment with TGF-β₃-targeted oligonucleotides affects downstream signaling. With lipofection, pSMAD2 and pSMAD1/5 levels were time- and concentration-dependently downregulated by both oligonucleotides (Fig. 3A and B). Gymnotic delivery of 2.5 μmol/L ISTH2020 or ISTH2023 reduced pSMAD2 earliest on day 7 with the most pronounced effect on day 8. SMAD1/5 phosphorylation was reduced on day 8 (Fig. 3C). The effect on SMAD2 and SMAD1/5 phosphorylation was also concentration-dependent as assessed with ISTH2023, however, lower concentrations than 2.5 μmol/L had no effect on SMAD phosphorylation (Fig. 3D). Of note, TGF-β₁ and TGF-β₂ protein levels in the cell culture supernatant were unaffected after 120-hour exposure of oligonucleotides (25 nmol/L) with lipofection-aided transfection (Supplementary Fig. S6A) and on day 8 of gymnotic treatment with 2.5 μmol/L (Supplementary Fig. S6B). We next extended our analysis to the human LTC T98G, and the murine cell line SMA-560. In T98G and SMA-560 cells, both TGF-β₃-targeted oligonucleotides reduced SMAD2 phosphorylation, too (Fig. 3E).

TGF-β isoforms are involved in regulating tumor cell invasiveness (29) and TGF-β₃ has been attributed isoform-specific effects in wound healing (8), a process involving cell migration and invasion. We therefore assessed whether inhibition of TGF-β₃ affected invasiveness of LN-308 glioma cells. Invasiveness was reduced by silencing of TGF-β₃ with either ISTH2020 or ISTH2023 (Fig. 4).

We next asked whether the inhibitory effect on SMAD phosphorylation by TGF-β₃-targeted oligonucleotides is preserved upon additional stimulation with other TGF-β isoforms. Surprisingly, SMAD2 and SMAD1/5 phosphorylation was induced less by TGF-β_1/2 at 0.5 ng/mL in the presence of TGF-β₃-targeted oligonucleotides. This effect still held true for SMAD2 phosphorylation at higher concentrations (5 ng/mL) of TGF-β₁ and TGF-β₂ (Fig. 5).

The TGF-β₃-targeted oligonucleotides ISTH2020 and ISTH2023 were shown to exhibit no liver toxicity upon systemic administration as assessed by plasma alanin-transaminase (ALT) in murine models (M. Janicot, unpublished observation). To verify target inhibition in vivo, we systemically treated mice bearing syngeneic (SMA-560) or xenograft (LN-308) gliomas with control or ISTH2023 oligonucleotides. Tumors derived from SMA-560 cells showed significantly higher mRNA expression of TGF-β₃ than normal brain. Upon treatment with ISTH2023, tumoral mRNA levels of TGF-β₃ were reduced by more than 30% (Fig. 6A). Importantly, PAI-1 as a transcriptional down-stream target was significantly reduced in ISTH2023-treated tumors (Fig. 6B). Similar to the in vitro data, levels of TGF-β_1/2 were not altered by ISTH2023 (Fig. 6C).

In the xenograft LN-308 model, ISTH2023 significantly reduced human TGF-β₃ (Fig. 6D) and human PAI-1 (Fig. 6E) mRNA levels while tumoral human TGF-β_1/2 was not affected significantly (Fig. 6F).

We next asked whether treatment with TGF-β₃-targeted oligonucleotides affects the phenotype of experimental gliomas in vivo. Tumor size as assessed by volumetric measurement of H&E-stained slides was not significantly affected by ISTH2023 either in the SMA-560 (Fig. 6G) or in the LN-308 model (Fig. 6J). However, a trend toward a reduced tumor size was observed in both models in mice treated with ISTH2023. In the SMA-560 model, we assessed the number of satellite lesions as a morphologic surrogate marker of tumor invasiveness. The mean number of satellite lesions per brain section was 5 in the control group versus 3 in the group treated with ISTH2023. This reduction was not significant (P = 0.41) when comparing the mean number of satellites per section of the 4 brains per group, but was significant (P = 0.0036) when comparing the results of all H&E-stained brain sections per group (n = 149 “control” versus n = 148 “ISTH2023”; Fig. 6H, representative images shown in Fig. 6I). In the LN-308 model, assessment of satellite lesions was not meaningful because the mean number of satellites per brain section was less than 1 in either group (representative images shown in Fig. 6K).

The TGF-β pathway has been attributed a key role in the pathogenesis of glioblastoma with regard to immunosuppression, invasion, angiogenesis, and maintenance of the stem cell phenotype (30, 31). Pharmacological strategies to interfere with the TGF-β pathway via inhibition of the kinase activity of TGF-β-receptor type I were promising in murine models, but so far disappointing in the clinic (29, 32–34). A ligand-based approach using AP-12009, an antisense oligonucleotide supposed to target TGF-β₂, administered intratumorally, suggested non-inferiority to alkylating chemotherapy in a phase II clinical trial, however, interpretation of the trial results remained controversial (35, 36).

Targeting the ligands rather than their bona fide receptors might have different therapeutic safety, tolerability and efficacy profiles. Beyond AP-12009, a TGF-β₂-antisense-modified tumor cell vaccine as another ligand-based approach of targeting the TGF-β pathway in glioblastoma has been tested in the clinic in a phase I trial (37). Targeting TGF-β expression through integrin inhibition has yielded promising results in vitro (38) but was not active in glioblastoma patients (39). TGF-β₁ and TGF-β₂ were considered as the most important isoforms in glioblastoma while little is known on the role of TGF-β₃ in the context of glioblastoma (13).

Here, we present a comprehensive analysis of TGF-β₃ mRNA and protein expression in human and murine in vitro models and show that TGF-β₃ expression can be specifically inhibited by oligonucleotides in vitro and in vivo. This inhibition led to downregulation of SMAD signaling in vitro and of the SMAD-dependent target gene PAI-1 in vivo despite the presence of TGF-β₁ and TGF-β₂.

The fact that high TGF-β₃ mRNA expression correlates with poor survival in glioblastoma patients in the TCGA database (13) suggests targeting of TGF-β₃ as a promising strategy. A previous study performed using the Rembrandt database showed that expression levels of more than 2-fold of TGF-β₁ (P = 0.02) or more than 5-fold of TGF-β₂ (P = 0.05) correlated with poor survival while more than 2-fold expression of TGF-β₃ was without prognostic significance (P = 0.08; ref. 16). Divergent results regarding a prognostic role of TGF-β₃ expression might be due to different samples in the respective databases with tumors exhibiting genetic heterogeneity or the use of different cutoffs. This would be in line with our observation that the association of TGF-β₃ mRNA expression with poor survival is most prominent in the neural subtype of the Verhaak classification (Supplementary Fig. S2). These data suggest that therapeutic targeting of TGF-β₃ might be most effective in these tumors.

We show that TGF-β₃ expression in human glioma cell lines is less abundant than that of TGF-β₁ and TGF-β₂, in line with a previous study based on glioblastoma tissue samples (13), however, GIC did not show this pattern (Supplementary Fig. S3A). Intracellular protein levels of TGF-β₃ were detected in all human and murine cell lines while the 12.5 kDa secreted form of TGF-β₃ was present in the majority of human LTC, but not GIC (Fig. 1D–F). This might reflect a different importance of this isoform in these cell types. All three TGF-β isoforms induced SMAD2 phosphorylation in a concentration-dependent manner (Fig. 1I), not revealing isoform-specific differences. However, specific time- and concentration-dependent inhibition of endogenous TGF-β₃ by oligonucleotides (Fig. 2) reduced SMAD2 and SMAD1/5 phosphorylation despite the concurrent presence of TGF-β_1/2 which would be expected to be major and sufficient drivers of baseline phosphorylation levels (Fig. 3). The gymnotic transfection method with omission of transfection reagent and micromolar concentrations intends to mimic the systemic administration of the drug in vivo and led to a similarly effective inhibition of TGF-β₃ expression and reduction of SMAD2 and SMAD1/5 phosphorylation as the conventional lipofection-based transfection method (Figs. 2 and 3).

The profound effect of inhibition of TGF-β₃ on SMAD phosphorylation, albeit expressed at mRNA levels much lower than those of TGF-β_1/2 in these cells, suggests a major distinct regulatory role of this isoform. The reduction in phosphorylated SMAD levels by inhibition of TGF-β₃ leading to reduced invasiveness in the LN-308 model (Fig. 4) points toward biological and clinical relevance of targeting TGF-β₃ in glioblastoma.

The hypothesis that TGF-β₃ might exert major biological effects despite its low expression levels and concurrent presence of the other TGF-β isoforms is further supported by our observation that TGF-β₃-specific oligonucleotides reduce SMAD phosphorylation even in the presence of exogenous TGF-β_1/2 (Fig. 5). For the three TGF-β isoforms, auto-feedback loops inducing their own expression mediated by different transcription factors according to the respective isoform with differential promotor regions have been suggested (16, 40, 41). An impaired auto-feedback loop through resulting from TGF-β₃ depletion might explain the profound effects on SMAD phosphorylation despite the presence of the other isoforms. TGF-β₃-dependent epithelial–mesenchymal transition is downstream of a TGF-β_1- and TGF-β₂-induced upregulation of the E-cadherin repressors snail and slug (42), supporting the concept of hierarchical importance of the different isoforms. Potentially, TGF-β₃ functions as a gatekeeper controlling downstream signaling despite high expression of TGF-β₁ and TGF-β₂ isoforms.

Treatment with the TGF-β₃-specific oligonucleotides in vivo also resulted in specific target downregulation without significant effects on the other TGF-β isoforms. Biological relevance of reduced TGF-β₃ levels was demonstrated by profound downregulation of the SMAD-dependent target gene PAI-1 both in the syngeneic SMA-560 and xenograft LN-308 model (Fig. 6). Short-term inhibition of TGF-β₃in vivo showed only a minor, non-significant reduction of tumor volumes (Fig. 6), in line with the observation that there are no significant effects on cell viability nor proliferation in vitro (Supplementary Fig. S5). Reduced TGF-β₃ levels were associated with a minor reduction of tumor invasiveness in vivo (Fig. 6), but prolonged exposure to inhibitors is probably necessary for more prominent effects. Furthermore, inhibition of TGF-β₃ alone might not be sufficient to block tumor invasiveness, given the complexity in vivo involving other drivers of tumor invasion and potential escape mechanisms.

In summary, we demonstrate that isoform-specific targeting of TGF-β₃ is feasible and effective by subcutaneous injection of a suitable oligonucleotide.

Pharmacological inhibition of TGF-β₃ in glioblastoma may therefore represent a promising strategy warranting further investigation.

K. Seystahl is a consultant/advisory board member for Roch. P. Roth is a consultant/advisory board member for Roche, MSD, and Molecular Partners and has an expert testimony from Novartis. M. Weller reports receiving commercial research grant from Isarna and has received speakers bureau honoraria from Isarna. K. Hasenbach was employed by Isarna Therapeutics and M. Janicot is employed by Isarna Therapeutics. No potential conflicts of interest were disclosed by the other authors.

Conception and design: K. Seystahl, I. Burghardt, M. Janicot, P. Roth, M. Weller

Development of methodology: K. Seystahl, P. Roth, Michael Weller

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): K. Seystahl, A. Papachristodoulou, I. Burghardt, H. Schneider, Michael Weller

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): K. Seystahl, A. Papachristodoulou, I. Burghardt, H. Schneider, M. Janicot, P. Roth, Michael Weller

Writing, review, and/or revision of the manuscript: K. Seystahl, A. Papachristodoulou, K. Hasenbach, M. Janicot, P. Roth, Michael Weller

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): K. Hasenbach, Michael Weller

Study supervision: P. Roth, Michael Weller

The authors thank Monika Kruszyńska for her expert technical assistance.

This study was supported by a research grant from Isarna (Munich, Germany), a grant from the Swiss Cancer League/Oncosuisse to I. Burghardt and M. Weller (project number KFS-3305-08-2013), and by a grant of the Canton of Zurich (HSM-2) to M. Weller and P. Roth.

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.