Thymosin beta 4 is a pleiotropic actin-sequestering polypeptide that is involved in wound healing and developmental processes. Thymosin beta 4 gene silencing promotes differentiation of neural stem cells whereas thymosin beta 4 overexpression initiates cortical folding of developing brain hemispheres. A role of thymosin beta 4 in malignant gliomas has not yet been investigated. We analysed thymosin beta 4 staining on tissue microarrays and performed interrogations of the REMBRANDT and the Cancer Genome Atlas databases. We investigated thymosin beta 4 expression in seven established glioma cell lines and seven glioma-initiating cell lines and induced or silenced thymosin beta 4 expression by lentiviral transduction in LNT-229, U87MG and GS-2 cells to study the effects of altered thymosin beta 4 expression on gene expression, growth, clonogenicity, migration, invasion, self-renewal and differentiation capacity in vitro, and tumorigenicity in vivo. Thymosin beta 4 expression increased with grade of malignancy in gliomas. Thymosin beta 4 gene silencing in LNT-229 and U87MG glioma cells inhibited migration and invasion, promoted starvation-induced cell death in vitro and enhanced survival of glioma-bearing mice. Thymosin beta 4 gene silencing in GS-2 cells inhibited self-renewal and promoted differentiation in vitro and decreased tumorigenicity in vivo. Gene expression analysis suggested a thymosin beta 4-dependent regulation of mesenchymal signature genes and modulation of TGFβ and p53 signalling networks. We conclude that thymosin beta 4 should be explored as a novel molecular target for anti-glioma therapy.
Glioblastomas are highly aggressive primary brain tumours thought to originate from neural stem or progenitor cells (Galli et al., 2004; Zheng et al., 2008; Liu et al., 2010; Chen et al., 2012). Despite recent advances in multimodal treatment approaches including surgery, radiotherapy and chemotherapy, the median overall survival of patients with glioblastoma is still poor, i.e. in the range of 16 months in selected clinical trial populations (Stupp et al., 2005; Gilbert et al., 2011), and ∼11 months in population-based studies (Johnson and O'Neill, 2012). Recently, subgroups of malignant gliomas have been defined according to gene expression signatures. Increased invasiveness and poor prognosis were found to be associated with a mesenchymal gene expression signature, whereas expression of genes associated with neural differentiation was associated with improved patient survival (Verhaak et al., 2010). Furthermore, the expression of a mesenchymal gene signature was reported to be characteristic of a subpopulation of glioma cells termed glioma stem-like or glioma-initiating cells (Carro et al., 2010). These cells are believed to promote malignancy and evasion from conventional therapies (Singh et al., 2004; Bao et al., 2006; Chen et al., 2012). Glioma-initiating cells exhibit features of neural stem cells (Galli et al., 2004), and induction of differentiation in the glioma-initiating cells subpopulation might be a potential therapeutic approach to glioblastoma.
Thymosin beta 4 (TB4) is the most abundantly expressed member of the β-thymosin family of small pleiotropic polypeptides. Beta-thymosins were first isolated from calf thymus and thought to be thymic hormones (Low et al., 1979). However, the first reported cellular function of β-thymosins was buffering of actin monomers (Safer et al., 1990). To date, numerous functions beyond actin sequestering have been identified. For example, TB4 gene silencing in the heart led to a reduced number of mesenchymally differentiated cells and decreased migration of differentiated cells from the epicardial stem cell niche into the heart muscle, both during development and healing of ischaemic wounds (Bock-Marquette et al., 2004). Furthermore, TB4 activates epicardial stem cells and promotes reprogramming of cardiac fibroblasts to induced pluripotent stem cells (Smart et al., 2007; Qian et al., 2012). In the developing brain, TB4 expression is tightly associated with neurogenesis and regulates expansion of the stem cell pool of the early neuroepithelium (Roth et al., 1999; Wirsching et al., 2012), whereas TB4 gene silencing promotes the differentiation of neural stem cells in vitro (Mollinari et al., 2009). A role for TB4 has also been suggested in certain brain diseases. TB4 expression increased after focal brain ischaemia (Vartiainen et al., 1996), and after transient global hypoxia (Kim et al., 2006) in the rat, whereas proteomic analysis showed increased TB4 protein levels in the CSF of patients with Creutzfeldt-Jakob disease (Mohring et al., 2005). The pivotal role of TB4 in promoting cellular invasion and its role in brain development prompted us to investigate a putative involvement of TB4 in glioma pathogenesis.
Materials and methods
Tissue specimen and tissue microarray construction
Immunohistochemical stainings for TB4 expression were performed on a tissue microarray comprising 89 surgical glioma samples collected from patients who were treated at the Department of Neurosurgery, University Hospital of Zurich, Zurich, Switzerland between June 2003 and May 2009. All tumours were classified and graded according to the WHO classification of tumours of the CNS. For semi-quantitative analysis of TB4 staining intensity, we graded TB4 staining arbitrarily from 0–3 (0, negative; 1, weak; 2, moderate; 3, strong) for each tumour compartment, i.e. endothelial cells, inflammatory host cells and tumour cells. Compartments were identified based on morphology. Confirmatory immunohistological stainings were performed using antibodies against CD31 for endothelial cells, CD45 and CD11b for inflammatory host cells, and glial fibrillary acidic protein (GFAP) for tumour cells. In addition, we graded TB4 staining arbitrarily for each tissue sample from 0–3, based on the fraction of moderately or strongly TB4-positive cells (0, <20%; 1, 20–50%; 2, 50–80%; 3, >80% TB4-positive cells, respectively). Examples for weak/negative staining versus moderate/strong staining are given in Supplementary Fig. 1A.
Publically available microarray and clinical data of patients with glioma were acquired from the REpository for Molecular BRAin Neoplasia DaTa (REMBRANDT) using the data set available on 19 May 2011 (NCI, 2005), and from The Cancer Genome Atlas (TCGA) using the data set available on 15 December 2012 (McLendon et al., 2008). Gene expression and Kaplan-Meier survival data were queried following the REMBRANDT site’s instructions for ‘advanced search’ and through the caINTEGRATOR homepage (http://caintegrator2.nci.nih.gov) following the site’s instructions. The sample group for gene expression was restricted to WHO grade II/III gliomas (n = 184) and glioblastomas (n = 220). Survival data for the Kaplan-Meier analysis using the REMBRANDT database were retrieved for glioblastoma (n = 182). Samples with a 0.5-fold downregulation or a 2-fold upregulation of the target gene compared with median expression levels were defined as up- or downregulated, the other samples were defined as intermediate.
Kaplan-Meier survival data from TCGA (n = 465) were queried via the R2 microarray analysis and visualization platform (http://hgserver1.amc.nl/cgi-bin/r2/main.cgi). The averaged messenger RNA expression staining was scaled to 4775.5 for TMSB4X and 12.5 for TMSB4Y. The cut-off for the highest impact on survival was 7617.2 for TMSB4X and within the male population (n = 311) 9.5 for TMSB4Y.
For analysis of functional gene interactions, combined confidence scores were generated for each putative interaction by integration of experimental and predicted data using the Search Tool for the Retrieval of Interacting Genes/Proteins (STRING) Version 9.0 at http://string-db.org (Szklarczyk et al., 2011). Highest confidence settings were applied, thus merely integrating combined scores higher than 0.900. Cluster analysis was performed by application of the MCL algorithm.
LN-18, LNT-229 and LN-308 glioma cell lines were kindly provided by Dr. N. de Tribolet (Lausanne, Switzerland). T98G, U87MG and A172 glioma cell lines were purchased from the American Type Culture Collection. The TU159 cell line was generated in our laboratory (Bahr et al., 2003). GS-2, GS-3, GS-4, GS-5, GS-7, GS-8 and GS-9 glioma-initiating cells lines were provided by Dr. Katrin Lamszus (Gunther et al., 2008). All conventional cell lines were cultured in Dulbecco’s modified Eagle’s medium (Radnor) containing 10% foetal calf serum, 2 mM glutamine and penicillin (100 IU/ml)/streptomycin (100 µg/ml). All glioma-initiating cells lines were cultured in Neurobasal®-A medium supplemented with B27 (Invitrogen), 10 ng/ml basic fibroblast growth factor (bFGF), and 10 ng/ml epidermal growth factor (EGF) (BD Biosciences). Growth factors were replenished twice weekly. Conditioned supernatants were generated by seeding 2 million cells in T75 flasks overnight, then washing three times and adding serum-free Dulbecco’s modified Eagle’s medium for conditioning for 72 h.
A short-hairpin RNA expression cassette for targeting TB4 or a scrambled sequence were cloned into a custom-made SEW-based lentivirus (Demaison et al., 2002) under control of the U6 promoter. For targeting TB4, the following sequences were used: 5’-CTGAGATCGAGAAATTCGATAAG-3’, which is highly conserved and orthologous in mouse and human TMSB4X (Smart et al., 2007); and the sequence 5’-TGGCTGAGATCGAGAAATT-3’ (si_2). Both constructs were designed to co-express monomeric dsRed under the control of the SFFV promoter to assess the efficacy of stable transduction. The original GFP cassette from the SEW vector was excised. For TB4 overexpression, the coding sequence of human TMSB4X was cloned into a custom-made SEW-based lentivirus under the control of the SFFV promoter (Demaison et al., 2002), followed by the coding sequences of a 2A-autocleavage site, monomeric dsRed and WPRE. Lentivirus was produced and titred (Tabatabai et al., 2010) and target cells were transduced at a multiplicity of infection of 100 transducing lentiviral particles per cell.
Quantitative reverse transcriptase polymerase chain reaction
Total RNA was prepared using the NucleoSpin System (Macherey-Nagel) and complementary DNA transcribed using SuperScript® II reverse transcriptase (Invitrogen). For real-time PCR, complementary DNA amplification was monitored using SYBR® Green chemistry on the 7300 Real time PCR System (Applied Biosystems). The conditions for these PCR reactions were: 40 cycles of 95°C for 15 s, 60°C for 1 min, using the following specific primers: Arf1 fwd: 5'-GACCACGATCCTCTACAAGC-3', Arf1 rv: 5'-TCCCACACAGTGAAGCTGATG-3'; TB4 fwd: 5'-AAACCCGATATGGCTGAGAT-3’, TB4 rv: 5'-TGCTTCTCCTGTTCAATCGT-3', TB15A fwd: 5'-GCCTCCCAACAGCAGATTTCGA-3', TB15A rv: 5'-ACAGCATCTGCCATCTGGAACA-3', TB15B fwd: 5'-TCCTCCAAAGAGCAGATTTCAG-3', TB15B rv: 5'-GCATCTGCCATTTGGAATTTACA-3', TGFB1 fwd: 5'-GCCCTGGACACCAACTATTG-3', rev: 5'-CGTGTCCAGGCTCCAAATG-3', TGFB2 fwd: 5'- AAGCTTACACTGTCCCTGCTGC-3', rev: 5'-TGTGGAGGTGCCATCAATACCT-3'. Arf1 transcript levels were used as a house-keeping reference for relative quantification of messenger RNA expression levels using the ΔCT method. The samples for normal brain complementary DNA were purchased from Ambion (Applied Biosystems).
TGF-β-induced signalling was assessed by reporter assays using the SMAD-binding elements (SBE) containing reporter plasmid pGL3-SBE4-Luc (Zawel et al., 1998) kindly provided by Dr. B. Vogelstein, and the TGF-β-responsive plasminogen activator inhibitor 1 promoter fragment containing the reporter plasmid pGL2-3TP-Luc (Wrana et al., 1992), which was kindly provided by Dr. J. Massague. Dual luciferase/renilla assays were performed with co-transfection of 150 ng of the respective reporter construct and 20 ng of pRL-CMV. Luciferase activity was measured using a Mithras LB 940 microplate reader (Berthold) and normalized to constitutive renilla activity (pRL-CMV).
Denatured whole protein lysates (20 µg/lane) were separated on 10–13% acrylamide gels. After transfer to nitrocellulose (Bio-Rad), blots were blocked in PBS containing 5% skimmed milk and 0.05% Tween 20 and incubated overnight at 4°C with primary antibodies, washed in PBS and incubated for 1 h at room temperature with secondary antibodies. Primary antibodies were anti-TB4 (Immundiagnostik), anti-ILK (Lab Force) and anti-(p)Akt (Bioconcept). Visualization of protein bands was accomplished using horseradish peroxidase-coupled secondary antibodies (Santa Cruz Biotechnology) and the enhanced chemiluminescence technique (Thermo Fisher Scientific).
Enzyme-linked immunosorbent assay
Conditioned supernatants were generated as described above. Next, 1 N HCl was added for 20 min to activate latent TGF-β. TGF-β1 and TGF-β2 protein levels were assessed by a commercially available ELISA kit (R&D Systems). Streptavidin–horseradish peroxidase was added for 20 min, then after washing substrate solution, was added for another 20 min and optical density measured at 450 nm using a Mithras LB 940 microplate reader (Berthold). The standard curve was calculated using recombinant TGF-β1 and TGF-β2 at pre-defined concentrations, and a computer generated 4-PL curve-fit (Excel, Microsoft).
All flow cytometry analyses were performed using a CyAN ADP flow cytometer (Beckman Coulter). For cell cycle analysis, cells were washed, resuspended in PBS and fixed by slowly adding ice-cold ethanol to a final concentration of 70% and incubation on ice for 60 min. Cells were washed and resuspended in buffer (PBS containing 0.5% bovine serum albumin, 0.02% NaN3, 1 mM EDTA) at 100 000 cells per 30 µl. Cells were incubated with 5 µl of a custom-made stock solution containing 2.5 mg/ml propidium iodide (Sigma-Aldrich), 0.1 mg/ml RNase A (Roth) and 0.05% Triton™ X-100 (Sigma-Aldrich) for 30 min at 4°C and subsequently washed and resuspended in buffer. For separation of cells by DNA content, a PE laser was used to measure the peak plane. For annexinV-propidium iodide analysis, cells were stained with propidium iodide (Sigma-Aldrich) and pacific blue-labelled annexinV (Lucerna).
Matrix metalloproteinase activity assay
For assessment of matrix metalloproteinase (MMP) activity, the fluorescence resonance energy transfer-based SensoLyte® 520 MMP Substrate Sampler Kit (Anaspec) was used. In brief, conditioned supernatants of LNT-229 glioma cells were incubated with the substrate QXL520™-γ-Abu-Pro-Cha-Abu-Smc-His-Ala-Dab(5-FAM)-Ala-Lys-NH238 (Smc = S-methyl-l-cysteine). Fluorescence emission was measured in a Mithras LB 940 microplate reader (Berthold) at excitation/emission = 490/535 nm. A standard curve for quantification was obtained using the pre-cleaved fragment 5-FAM-Pro-Leu-OH at predefined concentrations.
Affymetrix gene chip analysis
Total RNA was extracted from LNT-229 and U87MG glioma cell lines transfected either with short interfering TB4 or scrambled control small interfering RNA using the RNeasy® Kit (Qiagen). High RNA quality, as indicated by RNA integrity numbers of >9, was assured by using the Agilent 2100 Bioanalyzer (Agilent). RNA samples were processed using the 3’ IVT express kit (Affymetrix) starting from a total amount of 100 ng RNA. Labelled complementary RNA was hybridized to Affymetrix GeneChip Human genome U133 2.0 plus arrays, followed by washing, labelling and scanning according to standard protocols. The GeneSpring software (Agilent) was used for determination of normalized gene expression values.
Growth, Alamar blue, clonogenicity and sphere formation assays
For growth assays, 1000 LNT-229 or U87MG cells, or 5000 GS-2 cells per well were seeded in 24-well plates and triplicates were counted daily. Viable cells were identified using trypan blue dye exclusion.
For Alamar blue assays (Invitrogen), 1000 cells were seeded in 96-well plates and supplemented with 10% Alamar blue for 1 h. Fluorescence emission of metabolized Alamar blue was measured in a Mithras LB 940 microplate reader (Berthold at excitation/emission = 560/600 nm). For clonogenic survival assays of LNT-229 or U87MG, 500 cells were seeded in triplicate in 6-well plates. After 20 days, cells were stained using 0.5% crystal violet solution. Colonies of 50 or more cells were counted manually at ×10 magnification.
For sphere formation assays, GS-2 cells were initially separated from each other mechanically and with accutase (Chemie Brunschwig). Cells were then seeded in Neurobasal® medium supplemented with growth factors in 6-well plates at a density of 1 cell per 4 μl and sphere formation was assessed by manually counting spheres at ×10 magnification after 20 days. Sphere volume was calculated based on diameters of spheres assessed in 12 high power fields using ImageJ 1.40g software (NIH).
Migration and invasion assays
Migration of glioma cells was measured in transwell migration assays (24 wells, 8 μm pore size; BD Biosciences). In brief, 5 × 104 cells in 200 μl serum-free Dulbecco’s modified Eagle’s medium were added to each transwell insert. NIH 3T3-conditioned medium (700 µl) was used as a chemo-attractant in the lower wells. NIH 3T3 fibroblasts secrete a variety of growth factors, and therefore the conditioned medium from these cells is commonly used as a chemo-attractant (Albini et al., 1987). After an incubation period of 16 h, the cells on the lower side of the membrane were fixed in ice-cold methanol at 4°C, stained with hemalaun and sealed on slides. Quantification of cell migration was expressed as the mean count of stained cells in 11 random fields of each membrane.
For invasion assays, glioma spheroids were generated by incubating 1000 cells for 72 h in 96-well plates precoated with 1% Noble Agar (Difco Laboratories). Spheroids with a diameter of ∼200 µm were embedded into a collagen matrix containing collagen type I (Invitrogen), 10% foetal calf serum and 10% NaHCO3 in a 96-well plate. Sprouting of spheroids was monitored by daily photographs or by time-lapse video microscopy using a JuLI image viewer (NanoEnTek). For quantitation, the area covered by sprouting cells and the median invaded distance of 50 cells were assessed using the ImageJ 1.40g software (NIH). The spheroid margin at Day 0 was used as a reference point for measurement of the invaded distance of sprouting cells. For measurement of the invasion area, the area covered by the spheroid at Day 0 was subtracted from the overall area that was covered on subsequent days.
For the induction of differentiation, glioma-initiating cells were cultured in full Neurobasal® medium supplemented with the indicated foetal calf serum concentrations for 5 days. Differentiation was assessed by counting cells that exhibited an adherent phenotype and by immunostaining for GFAP or nestin.
All experiments were performed according to the Swiss animal protection law. CD1nu/nu mice were purchased from Charles River Laboratories. Mice aged 8–12 weeks were anaesthetized and placed in a stereotaxic fixation device. A burr hole was drilled in the skull 2 mm lateral and 1 mm posterior to the bregma. The needle of a Hamilton syringe was introduced into a depth of 3 mm (Tabatabai et al., 2007). LNT-229 (75 × 104), U87MG (105) or GS-2 (2 × 105) cells were resuspended in PBS and were injected into the right striatum. GS-2 cells were dissociated before injection. Animal experiments with LNT-229 gliomas were performed twice. Animal experiments with U87MG and GS-2 gliomas were performed once. Animals were assessed clinically three times per week and sacrificed at a clinical score of 2 (Supplementary Table 1).
Immunocytology, histology and immunohistochemistry
Primary antibodies were monoclonal mouse anti-Nestin (Zytomed, 1:100), polyclonal rabbit anti-GFAP (Dako, 1:1000), monoclonal mouse anti-NeuN (Chemicon, 1:100), polyclonal rabbit anti-β-III-Tubulin (R&D Systems), monoclonal mouse anti-CD31 (Dako, 1:50), monoclonal mouse anti-CD45 (Dako, 1:50), polyclonal rabbit anti-CD11b (Abcam, 1:50) and polyclonal rabbit anti-thymosin β4 (Immundiagnostik, 1:100). For immunocytochemistry, cells were grown on poly-d-lysine-coated cell culture slides (BD Biosciences) and fixed in paraformaldehyde. Neurospheres were centrifuged onto glass slides using a Shandon Cytospin 4 (Thermo Fisher Scientific), dried briefly and fixed in paraformaldehyde. Cells were permeabilized with 0.2% Triton™ X-100 and 10% normal horse or swine serum (Vector) was used for blocking. For conventional histology, 5 µm paraffin sections were stained with haematoxylin and eosin. For immunohistochemistry, deparaffinized and rehydrated sections were boiled in EDTA buffer, pretreated with 1% H2O2, blocked in 10% swine serum or blocking solution (Candor Biosciences). Biotinylated secondary antibodies, streptavidin and diaminobenzidine were obtained from Dako and used according to standard procedures.
Quantitative data are expressed as mean or median and SEM, as indicated. For column statistics, unpaired t-test or one-way ANOVA followed by Tukey’s post hoc test was performed. The in vitro experiments reported here were performed at least three times in triplicate with similar results. For survival statistics Gehan-Breslow-Wilcoxon test was performed. All statistical analyses were performed using Prism 5 (GraphPad Software).
Thymosin beta 4 is overexpressed in malignant gliomas
We investigated TB4 levels in malignant gliomas in vivo by immunohistochemistry on a tissue microarray containing tissue cores from 89 gliomas of different WHO grades (17 WHO grade II, 24 WHO grade III, and 48 WHO grade IV) and four normal brain tissue control samples. Representative stainings are shown in Fig. 1A. The highest TB4 levels were observed in glioblastomas (Fig. 1B). We analysed TB4 staining in three different compartments: tumour cells, inflammatory host cells and endothelial cells (Supplementary Fig. 1A–D). This compartment-specific analysis revealed a correlation of TB4 staining intensities with increasing WHO grade in tumour cells (Fig. 1C), but not in inflammatory host cells or endothelial cells (Supplementary Fig. 1C and D). In glioblastoma samples, the TB4 staining intensity was higher in tumour cells than in host or endothelial cells (Fig. 1D). We also correlated TB4 staining data with clinical parameters (Supplementary Table 2). In glioblastomas, TB4 levels did not differ between younger versus older patients (Supplementary Fig. 1E), but TB4 levels were higher in recurrent versus primary glioblastomas (Supplementary Fig. 1F). For correlation with clinical outcome, we subdivided samples by low (0–2) versus high (>2) arbitrary TB4 levels. Median overall survival was lower in the group of patients with strongly TB4 positive glioblastomas (TB4 low: 15.1 months versus TB4 high: 9.9 months, P < 0.05) (Fig. 1E). A more detailed investigation of TB4 staining and its role for clinical outcome in histological glioma subtypes of WHO grades II or III was precluded because of sample size.
To expand and validate our immunohistochemical TB4 data in malignant gliomas, we performed a REMBRANDT database interrogation (NCI, 2005). TB4 messenger RNA expression in gliomas correlated with increasing WHO grade (Supplementary Fig. 1G) and glioblastoma patients with low (<0.5-fold of median) TB4 expressing tumours had longer overall survival than patients with high (>2-fold of median) TB4 expressing glioblastomas (Fig. 1F). Correlation of TB4 expression with survival in glioblastoma was further confirmed by a TCGA database interrogation (Supplementary Fig. 1H).
Considering that the gene coding for TB4 is located on the X-chromosome prompted us to further analyse gender differences and expression of the homologous gene TMSB4Y. REMBRANDT and TCGA analyses yielded no differences of TMSB4X expression in female or male patients. Moreover, TMSB4Y was expressed, if at all, only at very low levels in both data sets, although with a significant correlation to survival in the male population (data not shown).
Next, we analysed TB4 messenger RNA expression in seven established glioma cell lines (T98G, U87MG, LN-18, LNT-229, LN-308, A172, Tu159) and in seven glioma-initiating cells cultures (GS-2, GS-3, GS-4, GS-5, GS-7, GS-8, GS-9) using real-time reverse transcription-PCR analysis. TB4 was overexpressed up to 10-fold in all glioma-initiating cells cultures except for GS-4, and up to 22-fold in all long-term glioma cell lines relative to normal brain tissue samples used for reference (Supplementary Fig. 2). Immunocytochemistry indicated nuclear and cytoplasmic localization of TB4 in glioma cells, with a specific increase in cytoplasmic protrusions (Fig. 1G).
Thymosin beta 4 gene silencing in glioma cells enhances apoptotic cell death and decelerates clonogenic growth
To investigate the biological role of TB4 functions in glioma cells, we performed lentiviral short hairpin RNA-mediated gene silencing (short interfering TB4) in two cell lines with low and high TB4 expression levels, i.e. LNT-229 and U87MG (Supplementary Fig. 2, Fig. 2A and B). Assessment of dsRed co-expression revealed lentiviral transduction efficacies close to 100% (Supplementary Fig. 3). TB4 gene silencing did not affect the messenger RNA expression levels of thymosin beta 15 (TB15), another member of the β-thymosin family (data not shown), thus validating the specificity of the short hairpin RNA sequence used for targeting TB4 and excluding compensatory upregulation of TB15. In addition, we overexpressed TB4 in LNT-229 and U87MG cells by lentiviral vectors (Supplementary Fig. 4A).
The growth of LNT-229 cells was increased upon TB4 overexpression (Supplementary Fig. 4B) and decreased after TB4 gene silencing (Supplementary Fig. 4C), resulting in doubling times during exponential growth of 20.2 h for TB4-overexpressing cells compared with 22.1 h for empty vector control cells; and 25.6 h for short interfering TB4 transfected cells compared with 21.0 h for scrambled cells. Reduced growth of LNT-229 cells depleted of TB4 was paralleled by reduced metabolic activity (Supplementary Fig. 4D).
One of the key features of cancer cells is survival under nutrient-restricted conditions (Hanahan and Weinberg, 2011). Thus, we analysed cell viability by trypan blue exclusion in LNT-229 and U87MG glioma cells after 7 days of serum deprivation. Trypan blue-positive cells, i.e. the dead cell fraction was increased upon TB4 gene silencing (Fig. 2C). The effect of TB4 gene silencing on cell death under nutrient-restricted conditions was confirmed using a second short hairpin RNA sequence (Supplementary Fig. 5A–D). Flow cytometry-based analysis of the cell cycle indicated a 5-fold increase of the sub-G0/1 fraction in short interfering TB4 transfected cells whereas the cell cycle distribution upon starvation in the viable cell fractions was similar in both scrambled and short interfering TB4 LNT-229 cells (Fig. 2D and E). In serum-containing cell culture medium, however, dead cell fractions did not differ in scrambled and in short interfering TB4 transfected cells. We then performed annexinV and propidium iodide flow cytometry in scrambled and short interfering TB4 glioma cells to analyse whether enhanced starvation-induced cell death of TB4-depleted cells was paralleled by an increase in apoptosis. The percentage of apoptotic, i.e. annexinV-positive cells was 3.2-fold (48% versus 15%) higher in LNT-229 short interfering TB4 and 2.2-fold (61% versus 28%) higher in U87MG short interfering TB4 transfected cells. The percentage of dead, i.e. propidium iodide-positive cells was 3.9-fold (43% versus 11%) higher in LNT-229 short interfering TB4 and 2.6-fold (55% versus 21%) higher in U87MG short interfering TB4 transfected cells (Fig. 2F).
Next, we analysed clonogenicity, single colony morphology and actin filament staining in LNT-229 scrambled and short interfering TB4 transfected cells (Fig. 2G–L). After TB4 gene silencing, the number of colonies was not reduced, but single colonies were smaller (Fig. 2G and J) and more compact (Fig. 2H and K) and cellular protrusions were reduced in size and number (Fig. 2I and L). Similar data were observed in U87MG (data not shown).
Thymosin beta 4 gene silencing inhibits migration and invasion of glioma cells in vitro
The accumulation of TB4 in cytoplasmic protrusions (Fig. 1G) and the marked reduction of size and number of cytoplasmic protrusions in LNT-229 short interfering TB4 transfected cells (Fig. 2L) prompted us to investigate the effects of TB4 on migration and invasion in LNT-229 and U87MG cells. We analysed the invasiveness of LNT-229 scrambled or short interfering TB4 spheroids (Fig. 3A and B, Supplementary Fig. 5E and F), and U87MG scrambled or short interfering TB4 spheroids (Fig. 3C and D, Supplementary Video 1) in a 3D collagen spheroid invasion assay and assessed invasion after 24 h (Fig. 3 E–H, Supplementary Fig. 5G and H) and after 48 h (Fig. 3I–L, Supplementary Fig. 5I and J) in a serum-containing collagen matrix. The area covered by invading cells was reduced by 2.7-fold in LNT-229 short interfering TB4 and by 4.2-fold in U87MG short interfering TB4 transfected cells (Fig. 3M). The invaded distance was reduced by 2.3-fold in LNT-229 short interfering TB4 and by 4.1-fold in U87MG short interfering TB4 transfected cells (Fig. 3N). Similar results were observed after targeting TB4 with a second short hairpin RNA sequence in LNT-229 (Supplementary Fig. 5K and L). Transwell migration of LNT-229 and U87MG cells was decreased upon TB4 gene silencing and increased upon TB4 overexpression (Supplementary Fig. 6), but no effect of TB4 overexpression on spheroid invasion was observed (data not shown).
Thymosin beta 4 gene silencing prolongs symptom-free survival of glioma-bearing mice in vivo
We orthotopically implanted LNT-229 or U87MG scrambled or short interfering TB4 transfected cells into nude mice. The latency until the onset of neurological symptoms was prolonged in animals bearing TB4-depleted experimental gliomas (Fig. 4A and B). The median time interval until the onset of symptoms was increased from 44 to 56 days (P < 0.05) in LNT-229 glioma-bearing animals, and 45 to 54 days (P < 0.05) in U87MG glioma-bearing animals. Histological analysis revealed smaller and less invasive tumours in short interfering TB4 gliomas, and tumour volume was markedly reduced to ∼20–30% at Day 28 (Fig. 4C). At the onset of neurological symptoms, however, tumour size was similar in LNT-229 scrambled and short interfering TB4 gliomas (Supplementary Fig. 7).
A Thymosin beta 4-dependent transcriptional network involves modulation of TGF-β signalling and regulates mesenchymal signature genes in glioma cells
Next, we aimed at analysing the mechanism by which TB4 exerts its effects in glioma cells. First, we assessed integrin-linked kinase (ILK), Akt and MMP2 expression after TB4 knockdown, because stabilization of ILK, phosphorylation of Akt and consecutive regulation of MMP2 expression by TB4 have been suggested in other cell types (Bock-Marquette et al., 2004; Fan et al., 2009). ILK protein levels were not decreased in LNT-229 short interfering TB4 transfected glioma cells and there was no effect of TB4 gene silencing on Akt phosphorylation on Ser473 (Supplementary Fig. 8A). We reasoned that phosphoinositide 3-kinase (PI3K) signalling might compensate for effects of short interfering TB4 on ILK levels, because PI3K is a major ILK activator. Indeed, upon treatment with the PI3K inhibitor wortmannin ILK protein levels were reduced in TB4-depleted cells, but with no effect on Akt phosphorylation. MMP2 protein levels also remained unaffected upon TB4 gene silencing (Supplementary Fig. 8B), but MMP activity was reduced in an assay that detects the activity of multiple MMPs (MMP1/2/3/7/8/9/12/13/14) (Supplementary Fig. 8C). These results indicate that TB4 stabilizes ILK in glioma cells and alters MMP activity, but does not affect Akt phosphorylation.
To understand the molecular network controlled by TB4 in glioma cells, we performed an Affymetrix chip-based transcriptome analysis of LNT-229 scrambled and short interfering TB4 transfected cells and retrieved a list of regulated candidate genes. At a 2.0-fold change cut-off, one thousand seven hundred and twenty-four probe sets were differentially detected, including 162 that were directed against a total of 116 complementary DNAs coding for transcription factors (Supplementary Table 3). Submission of all differentially expressed transcription factors to the Search Tool for the Retrieval of Interacting Genes/Proteins (STRING) identified two main clusters (Fig. 5A). One cluster involved mainly genes interacting with TP53 (green nodes), the other cluster was centred around the downstream TGF-β signalling modulators EP300 and FOS (red nodes). The latter cluster furthermore comprised the TGF-β signalling modulators CITED1 and CITED2, and interacted with a third cluster comprising the TGF-β signalling modulators SMAD6 and SMAD7 (brown nodes). All six identified TGF-β signaling modulators were downregulated in the Affymetrix gene chip in short interfering TB4 glioma cells.
As we had observed enhanced apoptosis and decreased migration and invasion upon TB4 gene silencing (Figs 2 and 3, Supplementary Fig. 5), we performed a gene ontology (GO) analysis with TB4-dependently differentially regulated genes. Gene ontology annotation identified 34 genes involved in migration (Supplementary Table 4) and 75 genes involved in apoptosis (Supplementary Table 5), with an overlap of 14 genes that are annotated to both processes, including the TGF-β signalling modulators TGFB2, THBS1, CITED2 and SMAD7. Furthermore, inhibitors of TGF-β signalling including ONECUT2 and FBN1 were upregulated in LNT-229 short interfering TB4 transfected cells. Conversely, TGF-β signalling transcriptional target genes including SNAI2, IRX1, FBXO32, DAPK1, ATG5, ATG7, TGFA, PDGFC, COL1A2 and FN1 were downregulated on TB4 gene silencing.
The central role of TGF-β signalling in gliomas prompted us to further validate it by quantitative real time PCR, ELISA and reporter assays. TGFB1 messenger RNA levels were reduced in LNT-229 short interfering TB4 transfected cells by 2.2-fold and in U87MG short interfering TB4 transfected cells by 3.6-fold (Supplementary Fig. 9A). TGFB2 messenger RNA levels were reduced in LNT-229 short interfering TB4 transfected cells by 4.3-fold and in U87MG short interfering TB4 transfected cells by 18.3-fold, respectively (Supplementary Fig. 9B). The levels of TGF-β1 and TGF-β2 protein in the supernatants of LNT-229 short interfering TB4 were decreased by 1.4-fold and 4.5-fold, respectively (Supplementary Fig. 9C and D). Furthermore, decreased TGF-β signalling in LNT-229 short interfering TB4 transfected cells was confirmed using a dual luciferase/renilla reporter assay (Supplementary Fig. 9E).
Inhibition of TGF-β signalling inhibits stem cell properties and promotes a more differentiated gene expression pattern of glioma cells due to decreased expression of the Sry-related HMG-box factor SOX2 (Ikushima et al., 2009) and SOX2 was downregulated in TB4-depleted LNT-229 cells (Supplementary Table 3). Thus, we compared the expression signature of TB4-depleted cells with recently established signatures of gliomas that take into account levels of differentiation (Freije et al., 2004; Phillips et al., 2006; Carro et al., 2010; Verhaak et al., 2010). A comparison of the signature of TB4-depleted cells with the signatures suggested by Freije et al. (2004) revealed upregulation of the neurogenesis signature genes BMP2 and HEY2, whereas the invasive signature genes COL6A3, THBS1 and FN1 and the proliferative signature genes TOP2A, CDKN3, PTTG1 and CENPF were downregulated (Freije et al., 2004). A comparison with the signatures suggested by Phillips et al. (2006) revealed an upregulation of the proneural signature gene GABBR1, silencing of the marker genes of the proliferative signature IQGAP3 and HMMR and silencing of the mesenchymal signature genes COLA4A1 and COL4A2 (Phillips et al., 2006). Of note, gene silencing of IQGAP3, HMMR, COLA4A1 and COL4A2 is also a feature of the proneural signature, as defined by the centroids of k-means clustering (Phillips et al., 2006). A shift of the messenger RNA expression pattern of LNT-229 short interfering TB4 transfected cells towards a proneural signature was furthermore apparent when comparing the TB4-depletion signature with the signatures established by Verhaak et al. (2010) (Supplementary Table 6).
Based on the sample data classified as proneural, proliferative or mesenchymal in previous studies, Carro et al. (2010) defined a subset of transcription factors that function as master regulators of a mesenchymal expression signature, while suppressing a proneural signature in malignant gliomas (Freije et al., 2004; Nigro et al., 2005; Phillips et al., 2006; Carro et al., 2010). Comparison of this comprehensive data set with the set of genes that were differentially regulated in LNT-229 short interfering TB4 transfected cells revealed an upregulation of 20 genes of the proneural signature and silencing of 23 proliferative signature genes (Supplementary Table 7). Furthermore, 10 transcription factors that had been predicted by an ARACNe-based master regulator algorithm to be frequently connected to the mesenchymal signature, or had consensus enrichment in the promoters of mesenchymal signature genes (Carro et al., 2010), respectively, were downregulated in LNT-229 short interfering TB4 transfected cells (Fig. 5B). Conversely, the inhibitor of the mesenchymal signature SATB1 was upregulated upon TB4 gene silencing (Fig. 5B). These data indicate that TB4 gene silencing inhibits the mesenchymal signature and shifts the balance towards a more differentiated expression pattern. In line with this, a set of 20 genes that are downregulated during in vitro astrocyte development were also downregulated in LNT-229 short interfering TB4 transfected cells (Supplementary Table 8) (Cahoy et al., 2008).
Thymosin beta 4 gene silencing in GS-2 cells reduces stemness in vitro and in vivo
The analyses summarized above suggested that TB4 promotes the mesenchymal signature in glioma cells. To further analyse the functional relevance of this finding, we investigated the role of TB4 in the glioma stem-like cell line GS-2 (Gunther et al., 2008). We efficiently depleted TB4 expression in GS-2 cells as assessed by quantitative reverse transcription-PCR and immunoblot (Fig. 6A and B). TB4 depletion decreased cellular growth (Fig. 6C) and sphere formation (Fig. 6D). The sphere volume was reduced in GS-2 short interfering TB4 to 16.12 ± 1.97% (P < 0.0001; Fig. 6E).
Next, we assessed the differentiation capacity of GS-2 short interfering TB4 versus GS-2 scrambled cells using increasing foetal calf serum concentrations as a stimulus to induce differentiation. Foetal calf serum treatment yielded a maximum of 5.8 ± 0.9-fold increase (P < 0.0001) of cells exhibiting an adherent, differentiated phenotype in GS-2 cells transfected with short interfering TB4 (Fig. 6F). Cells that acquired this phenotype expressed GFAP, but not nestin. Moreover, there was a reduction of nestin-positive cells in GS-2 cells transfected with short interfering TB4 spheres as compared with GS-2 cells transfected with scrambled RNA, whereas the number of cells expressing GFAP was increased (Fig. 6G). Representative images of GS-2 scrambled and short interfering TB4 transfected cells grown as spheres or supplemented with 5% foetal calf serum for 7 days are shown in Fig. 6H and representative stainings of GS-2 scrambled and short interfering TB4 transfected cells are outlined in Fig. 6I.
Finally, we assessed the tumorigenicity of GS-2 short interfering TB4 transfected cells in vivo. Even after implantation of 200 000 cells, we did not observe any neurological symptoms in GS-2 short interfering TB4-glioma-bearing mice until Day 140 whereas GS-2 scrambled-glioma-bearing mice developed neurological symptoms within 69–87 (median 78) days (Fig. 6J). After 140 days, we sacrificed the asymptomatic GS-2 short interfering TB4-glioma-bearing animals and analysed the brains by histology. Tumour formation had occurred only in three of seven animals (Fig. 7A and B). The expression of nestin was reduced (Fig. 7C and D). Expression of GFAP was similar in both groups (Fig. 7E and F). We did not detect the neural lineage marker NeuN in GS-2 scrambled gliomas (Fig. 7G) whereas NeuN-positive cells were present in GS-2 short interfering TB4 gliomas (Fig. 7H).
TB4 is a key regulator of cancer hallmarks including migration, invasion, cell survival and stem cell activation (Bock-Marquette et al., 2004; Smart et al., 2007; Fan et al., 2009). In neural stem cells, TB4 gene silencing promotes neural differentiation, whereas overexpression of TB4 in the developing brain leads to an expansion of the neuroglial stem and progenitor cell pool and induction of a mesenchymal phenotype (Mollinari et al., 2009; Wirsching et al., 2012). The subpopulation of stem-like cells in gliomas termed glioma-initiating cells exhibits features of neuroglial progenitor cells, including self-renewal and multilineage differentiation, and expresses a mesenchymal gene signature (Galli et al., 2004; Bao et al., 2006; Carro et al., 2010; Chen et al., 2012). The role of TB4 during brain development and for mesenchymal transformation prompted us to hypothesize a role of TB4 in malignant glioma.
As a first step to investigate a role for TB4 in malignant glioma, we performed tissue microarray, REMBRANDT and TCGA analyses (Fig. 1 and Supplementary Fig. 1). TB4 expression correlated with ascending grades of malignancy and with survival. Tissue microarray data suggested that the correlation of TB4 staining intensity with increasing WHO grade was because of an increase of TB4 staining intensity within the tumour cell compartment (Fig. 1 and Supplementary Fig. 1). Thus, we focused our further analyses on TB4 in glioma cells.
Both, long-term glioma and glioma-initiating cells lines expressed high levels of TB4 in vitro (Supplementary Fig. 2). Silencing of TB4 in long-term glioma cells promoted apoptotic cell death upon nutrient depletion (Fig. 2 and Supplementary Fig. 5), inhibited migration and invasion (Fig. 3 and Supplementary Fig. 5) and increased survival of glioma-bearing nude mice (Fig. 4). At the onset of clinical symptoms, LNT-229 short interfering TB4 transfected gliomas resembled LNT-229 scrambled gliomas in terms of tumour size and invasiveness, suggesting a compensation for TB4 gene silencing in an advanced stages of the disease (Supplementary Fig. 7).
Alterations of the cellular morphology from spindle-like to a more dense and adhesive pattern upon TB4 depletion suggested altered actin dynamics to be the underlying mechanism by which TB4 mediates its cellular functions (Fig. 2). However, enhanced starvation-induced apoptotic cell death pointed to a functional significance of TB4 beyond its well-established role as an actin-buffering polypeptide (Fan et al., 2009). Thus, we reasoned that alteration of intracellular signalling pathways may contribute to the observed in vitro and in vivo effects of TB4 gene silencing in glioma. To further elucidate the underlying mechanism of TB4 gene silencing-mediated effects, we first investigated known interactions by which TB4 modulates cellular functions in other cell types. During cardiac development and in migrating endothelial cells TB4 stabilizes ILK, thus promoting Akt phosphorylation and MMP2 expression (Bock-Marquette et al., 2004; Fan et al., 2009). Furthermore, TB4-induced epithelial-mesenchymal transition and malignant progression of colon adenoma to carcinoma is mediated by ILK stabilization and activation (Huang et al., 2007). In turn, TB4 depletion in colon cancer cells leads to reduced ILK activity, reduced tumour volumes and cell migration (Ricci-Vitiani et al., 2010). In LNT-229 glioma cells, ILK protein levels were not reduced upon TB4 gene silencing, unless PI3K signalling was inhibited by wortmannin. However, even strongly decreased ILK protein levels did not affect Akt phosphorylation or MMP2 protein level (Supplementary Fig. 8) suggesting compensatory mechanisms for downregulation of ILK.
To unravel the molecular network of TB4 in glioma cells, we performed a genome-wide screening for genes that were differentially transcribed upon TB4 gene silencing (Fig. 5). In the past decade, several molecular subgroups of malignant gliomas based on genetic signatures have been defined and correlated to clinical outcome (Phillips et al., 2006; Verhaak et al., 2010). Expression of proneural marker genes correlates with better clinical outcome whereas mesenchymal and proliferative signatures were associated with poor outcome (Freije et al., 2004; Phillips et al., 2006; Carro et al., 2010). Consequently, we compared the TB4-dependently regulated set of genes with the genetic signatures of these previously published molecular subgroups. This comparison revealed regulation of various mesenchymal signature genes in the sense of a shift towards a more differentiated molecular subgroup in short interfering TB4 transfected cells (Fig. 5 and Supplementary Tables 6 and 7).
As glioma-initiating cells more closely reflect the functional relevance of a shift towards a more differentiated molecular subgroup than long-term glioma cell lines (Galli et al., 2004; Singh et al., 2004) we analysed the role of TB4 expression in the glioma-initiating cells line GS-2 (Gunther et al., 2008). TB4 depletion in GS-2 cells inhibited self-renewal, decreased sphere size and increased differentiation capacity in vitro (Fig. 6), and strongly reduced tumorigenicity and increased differentiation capacity in vivo (Fig. 7). These data are supported by previous reports suggesting a role of TB4 for stemness during brain and cardiac development (Bock-Marquette et al., 2004; Wirsching et al., 2012).
GS-4 cells share molecular features with GS-2 (Gunther et al., 2008), but, as opposed to GS-2, GS-4 express little if any TB4 (Supplementary Fig. 2). In line with our observations on the role of TB4 for stemness and differentiation, GS-4 cells are not tumorigenic in nude mice, express lower levels of the progenitor marker nestin, higher levels of the differentiation markers MAP2 and BAI1, and exhibit an adherent growth pattern (Gunther et al., 2008).
Functional interactome analysis of the TB4-dependently regulated set of genes using STRING and subsequent cluster analysis revealed a network involving p53 and TGF-β signalling (Fig. 5). Of note, recent studies have demonstrated that TGF-β mediates promotion of stemness in gliomas (Anido et al., 2010), thus suggesting a mechanism by which a short interfering TB4-mediated decrease of TGF-β signalling may have decreased stemness in GS-2 cells (Figs 6 and 7).
We further confirmed altered TGF-β signalling by data derived from quantitative reverse transcription-PCR, ELISA and reporter assays indicating a decreased TGF-β1 and TGF-β2 transcription and signalling upon TB4 depletion (Supplementary Fig. 9). One well-established function of TGF-β in gliomas is immunomodulation in the tumour microenvironment (Massague, 2008). Immunomodulation, however, is unlikely to have played a significant role for the prolonged survival in our orthotopic xenograft models, as these experiments were performed in immunodeficient mice (Figs 4 and 6).
In summary, TB4 expression correlates with glioma grades and patients’ survival, and regulates key malignant features in glioma models, including cell survival, invasiveness and stemness. Thereby, TB4 modulates core molecular networks including p53 and TGF-β signalling. We conclude that TB4 might be a novel key molecule integrating multiple hallmarks and molecular networks in malignant gliomas and should thus be further explored as a putative therapeutic target. To date, the only molecular inhibitor known to interfere with TB4 is Photorabdus toxin complex 3, which inhibits the interaction of TB4 with G-actin (Lang et al., 2010), but inhibition of the TB4-actin interaction alone is likely to result in activation of other cancer relevant pathways (Fan et al., 2009). Antisense strategies for targeting gene expression are conceivable, but have had limited success in cancer treatment so far. Recently, a first antisense therapy has been approved by the United States Food and Drug Administration to target apolipoprotein B in familial hypercholesterolaemia (Gotto and Moon, 2013). Limitations for such an approach in malignant glioma include tumour heterogeneity and the uncertainty of how efficient downregulation would have to be to exert activity. Yet, the design of pharmacological inhibitors of TB4 function seems feasible, given innovative high-throughput screening of small molecule libraries (Zhang et al., 2009). Small molecule inhibitors could either target TB4 function directly or its down-stream mediators. Thus, a thorough investigation of molecular interactions in the TB4 network in gliomas is needed.
We thank Caroline Herrmann (Department of Preclinical Imaging and Radiopharmacy, University of Tübingen, Tübingen, Germany, chairman: Bernd Pichler), Silvia Dolski and Matthias Scholl (Laboratory of Molecular Neuro-Oncology, Department of Neurology, University Hospital Zurich, Zurich, Switzerland) for outstanding technical assistance. We also thank Thorsten Wachtmeister and René Deenen (Center for Biological and Medical Research, BMFZ, Heinrich Heine University, Düsseldorf) for help with microarry hybridization and the evaluation of microarray data, and Katrin Lamszus (Laboratory for Brain Tumor Biology, Department of Neurosurgery, University Medical Center Hamburg-Eppendorf, Hamburg, Germany) for GS cells.
This work was supported by the Deutsche Forschungsgemeinschaft (SFB 773, A6), by the NCCR Neural Plasticity and Repair (P4) and by the Krebsliga Zürich.
Supplementary material is available at Brain online.
thymosin beta 4
The Cancer Genome Atlas