RETRACTED ARTICLE: TGF-β2 initiates autophagy via Smad and non-Smad pathway to promote glioma cells’ invasion

The high-grade, human glioma cell lines U251, U87 MG and T98G were obtained from American Type Culture Collection (Manassas, VA, USA) and used for in vitro experiments. Tumor cells were maintained as monolayer cultures in Dulbecco’s Modified Eagle’s Medium (DMEM; GIBCO) supplemented with 5% fetal calf serum (FCS). Primary cell line (P3) was obtained from the Department of Biomedicine, University of Bergen. When indicated, the cells were treated with TGF-β2 (10 ng/ml; PeproTech) and/or the small molecule TGF-β receptor inhibitor LY2157299 (10 μM; Selleck), chloroquine (10 μM; Selleck), bafilomycin A1 (10 μM; Tocris) and rapamycin (20 μM; Selleck). The inhibitor was added 2 h before the addition of TGF-β2.

The siATG5, siATG7 shRNA-expressing vector was obtained from GenePharma (Shanghai, China). GFP-LC3 (pBABEpuro, 22,405)-expressing vectors were obtained from Addgene (Cambridge, MA, USA). For the stable downregulation of ATG5, ATG7 scrambled and shRNA pGIPZ vectors were purchased from Open Biosystems (Thermo Scientific) or obtained through the Open Biosystems library. Lentiviral and retroviral supernatants were prepared following the manufacturer’s instructions and provided by GenePharma. Lentiviral infections were carried out accordingly. A validated Stealth RNAi (GenePharma) specific to Smad2 was also transfected into the U251, T98, and U87 cells using Lipofectamine 2000 according to the manufacturer’s protocol.

The migratory ability of cells was determined by wound healing assays. The rate of wound closure was monitored at different time points under a microscope and quantified using ImageJ software. The invasion potential was determined on collagen coated Transwell assay (Corning, 8um). DMEM containing 0.1% FBS was added to the upper wells, while DMEM containing 10% FBS (a chemoattractant) was added to the lower wells. Transwell insert membranes were fixed and then stained with 0.25% eosin. Representative pictures of the membranes with cells were acquired at 40× magnification, and the total number of cells in 10 individual fields per membrane were counted.

In brief, the cells were harvested, and lysed with protein extraction agent (Beyotime, Beijing, China). 25–50 μg of proteins per sample per lane were loaded for sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE). Primary antibodies were incubated overnight at 4 °C. Rabbit polyclonal primary antibodies were used against LC3B, SQSTM1, MMP2, MMP9, vimentin, N-cadherin, β-catenin, Slug2, Akt, JNK, SMAD2 (1:1000; Cell Signaling), TGF-β2 (1:500; Abcam), phospho-SMAD2, phospho-Akt, and phospho-JNK (1:500; Cell Signaling). Purified mouse anti-β-actin was used (1:1000; Santa Cruz Biotechnology, Inc.). Secondary antibodies were incubated with anti-mouse immunoglobulin G (IgG), anti-rabbit IgG (1:5000; Santa Cruz Biotechnology) for 1 h at RT. The proteins were visualized using Millipore’s enhanced chemiluminescence (ECL) and detection system analyzed (ChemiDoc Touch, BioRad).

U87 cells were prepared for intracranial injection into NOD SCID male mice obtained from Vital River Laboratories. A total of five animals per condition were used, and animals were grouped as follows: PBS group, LY2157299 (75 mg/kg/d) group, CQ (25 mg/kg/d) group and LY2157299 (75 mg/kg/d) combined with CQ (25 mg/kg/d) group, drugs were used by oral application and 10ul U87 cells (1 × 10⁶/μl PBS) were injected in the cerebral cortex using a stereotactic frame. The mice were monitored and killed when they presented with neurological signs or after two months, during which time the tumor volumes and invading distance were monitored by MRI (General Electric, 3.0 T). Brains of three groups (PBS group, LY2157299 group and LY2157299 combined CQ group) were harvested and fixed in 4% paraformaldehyde for 48 h, embedded in paraffin and prepared for IHC. These experiments were approved by the Animal Care and Use Committee of Shandong University and conformed to the Animal Management Rule of the Chinese Ministry of Health (documentation 55, 2001).

Total RNA was isolated using the RNeasy mini kit (Qiagen) according to the manufacturer’s instructions. RNA was analyzed quantitatively using a Nanodrop (Nanodrop Technologies, Rockland, DE, USA). Total RNA (1 μg) was reverse transcribed into cDNA using a cDNA synthesis kit (Toyobo) according to the manufacturer’s instructions. RT-PCR was performed in a Roche LightCycler 2.0 detector with the Toyobo SYBR Green Supermix. The reactions were analyzed using SDS software (Version 2.4). The threshold cycles (CT) were calculated and the relative gene expression was analyzed after normalizing to glyceraldehyde 3-phosphate dehydrogenase (GAPDH). The human primers are listed in Additional file 1: Figure S1.

GBM cells were plated on glass slides in 24-well culture plates at a concentration of 2 × 10⁵ cells/well for 24 h and were subsequently treated with drugs for an additional 48 h in DMEM containing FBS. The cells were then fixed with a 4% formaldehyde solution in PBS, permeabilized with 0.5% Triton X-100 in PBS, stained with primary antibody overnight, and labeled with anti-mouse or anti-rabbit IgG conjugated with FITC and DyLight 594 (Santa Cruz Biotechnology). The cells were counterstained with DAPI and observed under an Olympus BX61 fluorescence microscope. Pictures were scanned using a DP71 CCD (charge-coupled device) digital camera.

Live cells were fluorescently labeled with 25 nM MitoTracker Red (Invitrogen, Molecular Probes) and were then fixed and permeabilized for antibody conjugating. All treatments were performed according to the manufacturer’s instructions. Mitochondrial morphology was analyzed using an Olympus BX61 fluorescence microscope.

For the △Ψm measurements, the cells were loaded with 50 nM tetramethylrhodamine methylester (JC-1) for 30 min at room temperature. The dye was present during the experiment according to the manufacturer. The mitochondrial membrane potential was analyzed using an Olympus BX61 fluorescence microscope equipped with a 40× objective and quantified by flow cytometry (Novocyte, ACEA).

For the quantitative determination of activated human TGF-β2, MMP2, MMP9 concentrations in cell culture supernatants, the quantitative sandwich enzyme immunoassay technique was used with commercially available, specific immunoassay kits for human TGF-β2 (R&D Systems), MMP2, and MMP9 (eBioscience). The minimum detectable dose of TGF-β2 was less than 7.0 pg/ml. The assay was performed in triplicate according to the manufacturer’s instructions.

The total ATP and lactate levels were measured by plating 2 × 10⁵ cells in 6-well plates overnight. The ells were counted and media samples were collected for ATP and lactate assays. The total ATP levels were determined using the Cell Titer-Glo Luminescent assay (Promega) according to the manufacturer’s instructions. Lactate was measured using the L-Lactate assay kit (Abcam). The data were expressed as moles of ATP and mg/dl lactate, and all values were normalized to the number of cells.

Results are presented as the mean ± SD from at least three independent experiments. The comparisons were performed using with two-tailed Student’s t test, Spearman correlation test, Pearson correlation test, Wilcoxon matched-pair test, one-way ANOVA and long-rank analyse. Significant differences: *p < 0.05; **p < 0.01;***p < 0.001.

Although large studies has been conducted [27, 28], the role of autophagy’s markers Light Chain 3B (LC3B), ATG5 and ATG7 in glioma prognosis was still controversial. By applying immunochemistry on samples with different pathologic grades obtained from patients in Qilu Hospital, China from 2012 to 2015, we found that the percentage of LC3B, ATG5, ATG7 positive cells was associated with the pathologic grade (Fig. 1a). During this IHC process, we also found abundant TGF-β2 positive cells in GBM samples accompanied with autophagy markers positive. To quantitatively analyze the immunochemistry samples, we used the IRS scoring model (IRS = SI (staining intensity) × PP (percentage of positive cells)) [29], Wilcoxon matched-pair test, and Spearman correlation test to assess the results. Then the autophagy related markers and TGF-β2 were both identified to be associated with the pathologic grades (Fig. 1b) and correlated with each other when searched in REMBRANDT database (Fig. 1c).

To further explore the relationship between autophagy and TGF-β, we concentrated on LC3B and TGF-β.The immunofluorescence images indicated that these two molecules were co-localized in glioma tissues (Fig. 1d). We verified these conclusions in primary tumor samples and qRT-PCR analysis showed similar results in normal brain tissues and glioma patients: Student’s t test, LC3B: GBM vs Normal, p = 0.0401, TGF-β2: GBM vs Normal, P = 0.0490; Pearson correlation test, r = 0.4101, p = 0.0144 (Fig. 1e).

Then based on the data we obtained from qRT-PCR, LC3B and TGF-β2 expression levels were also found to be associated with glioma grades in Chi-square test, whereas patient age and gender had no influence (Fig. 1f). We then added the survival time index and used the long-rank test to analyze. Data showed that high TGF-β2 expression indicated poor prognosis and short survival time in glioma patients (Fig. 1g). And among the samples with high TGF-β2 expression, robustly positive LC3B expression usually indicated a worse survival outcome (Fig. 1h). As the relationship between TGF-β2 and autophagy in glioma has not been explored, these results indicate that TGF-β2 and LC3B are highly expressed in glioma and correlated with each other. High expression levels of TGF-β2 with LC3B indicates poor prognoses in glioma patients.

To investigate the exact relationship between TGF-β2 and LC3B in glioma, we treated the tumor cells with a TGF-β2 dose gradient ranging from 0.1 ng/ml to 10 ng/ml, and after 24 h, Western blotting was used to detect autophagy levels. As an increase in the LC3B-II/I ratio is a credible marker of autophagy enhancement [30], TGF-β2 promoted endogenous LC3B-I to LC3B-II conversion in three different cell lines and one primary cell line in a dose-dependent manner (Fig. 2a Additional file 6: Figure S6a). To determine the proper time required for TGF-β stimulation, we set a gradient time course and also tested the conversion ratio by western blot. Our results indicated that 24 h was appropriate to induce glioma autophagy (Fig. 2b and Additional file 6: Figure 6b).

To confirm this phenomenon, we applied a lentiviral system and generated stably expressing GFP-LC3 cells to monitor autophagy (TGF-β2, 10 ng/ml, 24 h) (Fig. 2c). In autophagy studies, the analysis of GFP-LC3 localization is widely performed by counting LC3-GFP positive cells (LC3-GFP dots greater than four) [30]. We found that TGF-β2 treatment increased the percentage of LC3-GFP positive cells (Fig. 2c). As transmission electron microscopy is the gold standard for identifying autophagosome double-membrane structure, we used transmission electron microscopy and found that glioma autophagosomes increased after TGF-β2 treatment (Fig. 2d, Additional file 2: Figure S2a).

Considering the disadvantage of cell lines to represent glioma’s characters, we introduced primary cells in our in vitro experiment, which were obtained from newly surgical GBM samples in Qilu hospital and grades were confirmed by pathology IHC. As shown in western blot (Fig. 2e), TGF-β2 could also initiate autophagy in primary cell under the same condition (10 ng/ml; 24 h) which indicate that the results of cell lines may represent the primary cells to some extent in our experiment.

Effects of 3-MA or Baf on autophagy flux mediated by TGF-β were also determined. Glioma cells were incubated with 3-MA (10 mmol/L) or Baf (10 nmol/L) during the treatment with TGF-β2 (Fig. 2f) and results indicated that Baf could effectively inhibit TGF-β induced autophagy except for 3-MA. According to the results of a microarray that compared the control and TGF-β2 treatment groups, the mRNA levels of the autophagy related genes BECN1, ATG5, ATG7 were also up-regulated after TGF-β2 treatment (Fig. 2g). To summarize, autophagy can be induced by TGF-β2 (10 ng/ml, 24 h) and blocked by Baf (10 nmol/L, 24 h). We next investigated the biological significance of this phenomenon.

To explore the biological meaning of TGF-β2-induced autophagy in oncogenesis, we first introduced a stable siRNA expression system targeting the essential autophagy gene ATG5 and ATG7. Efficiency of siRNA was examined by Western blot (Fig. 3a), and the ratio of LC3B-I to LC3B-II conversion and SQSTM1 (also well known as P62) degradation indicated that siATG5, siATG7 was an effective target for inhibiting autophagy flux. As we know from previous studies, TGF-β2 promotes glioma oncogenesis behavior [9, 31, 32]. We used methods involving CCK-8 and BrdU proliferation assays to assess cell proliferation and observed no obvious effect after TGF-β2 treatment in normal tumor cells (NTC) or autophagy-deficient tumor cells (Fig. 3b, Additional file 3: Figure S3).

Then we turned to examine the behavior of autophagy in glioma invasion and analyzed results with wound healing assays in U251, T98, and U87 cell lines. In stable siATG5 and siATG7 tumor cells, autophagy impaired groups exhibited delayed wound closure compared to scramble group tumor cells when exposed to TGF-β2 stimulation (Fig. 3c). And also we carried out a Transwell assay and found that fewer siATG5 and siATG7 cells migrated through the Transwell chamber micropores after TGF-β2 stimulation (Fig. 3d, Additional file 2: Figure S2b). Inefficient ECM degradation usually indicates potential changes in the function or expression level of matrix metalloproteinase (MMPs), which are involved in ECM degradation and the epithelial-mesenchymal transition (EMT). Summarizing the results above, TGF-β2 induced glioma autophagy participated in the invasive process but not tumor proliferation.

To identify the mechanism of autophagy-influenced glioma invasion, we conducted qRT-PCR to determine the expression levels of MMPs in normal glioma cells and Baf-treated tumor cells with or without TGF-β2 stimulation (Fig. 4a). The Baf treatment down-regulated the mRNA expression of the mostly MMP family in cells exposed to TGF-β2, and in contrast control group with TGF-β2 showed increasing expression levels of most MMPs (Fig. 4a, Additional file 4: Figure S4). MMP2 and MMP9, which are the most well-known MMPs for their important roles in tumor cell invasion [33, 34], were also down-regulated by Baf. And this finding was confirmed by Western blot experiments used to detect the cleaved, active forms of MMP2 and MMP9 (Fig. 4b). As the proteolytic cleavage of the ECM is usually governed by cell surface and soluble MMPs, we used an ELISA assay to detect soluble MMP2 and MMP9 in the supernatant of cultured NTC- and Baf-treated cells, similar results were expressed (Fig. 4c). In addition to degrading ECM, MMP2, 9 are also markers of EMT. We used Western blotting to examine other EMT markers to assess the potential association between EMT, TGF-β2 and autophagy. N-cadherin, Vimentin, Slug2, and β-catenin increased in normal tumor cells (NTC) treated with TGF-β2 but decreased in autophagy-inhibited cells however E-cadherin showed reverse effect (Fig. 4d). Immunofluorescence again demonstrated a direct change of Vimentin and N-cadherin between NTC and Baf-treated cells after treatment with TGF-β2 (Fig. 4e, f). Taken together, these results provide a potential explanation for how autophagy impairs glioma invasion: the inhibition of autophagy may lead to an impaired EMT and a failure in the invasive behavior of glioma cells.

We next investigated the exact mechanism of the process. Because autophagy is usually activated through mTOR-dependent and mTOR-independent pathways [35, 36], we used Western blotting to determine whether either of these pathways functioned in this process. Inhibiting the AKT-PI3K-mTOR pathway by 3-MA did not attenuate TGF-β2-induced autophagy (Fig. 5a). Thus, we assessed the mTOR-independent pathway using the c-Jun NH2-terminal kinase (JNK) pathway inhibitor SP600125. This treatment efficiently down-regulated the LC3B-II level in TGF-β2 treated glioma cells, indicating JNK pathway was one potential pathway responsible for TGF-β2 induced autophagy. TGF-β signals also went through downstream Smad and non-Smad pathways. The JNK pathway is a non-Smad pathway, and therefore we examined whether the Smad pathway also participated in autophagy induction. After transient siRNA with siSmad2, the addition of TGF-β2 did not stimulate autophagy in glioma, suggesting that the Smad pathway was also involved in autophagy induction (Fig. 5b).

Based on previous study, TGF-β2 function requires sustained autocrine signaling by tumor cells [31]. Thus, we explored whether autophagy would in turn affect TGF-β2 autocrine signaling (Fig. 5c), finding that rapamycin elevated the endogenous TGF-β2 levels in glioma cell and enhanced Smad2 phosphorylation. However, in siATG5-treated tumor cells, the Western blotting results showed reverse effect (Fig. 5c). The ELISA assay for TGF-β2 also demonstrated these phenomenon (Fig. 5d).

We next explored how autophagy influenced TGF-β2. The qRT-PCR results demonstrated that enhancing autophagy by rapamycin increased the R-Smad expression levels (Smad2, Smad3) and decreased the I-Smad expression level (Smad7). In contrast, CQ down-regulated Smad2 and Smad3 and up-regulated Smad7 (Fig. 5e). Microarray also indicated that promoting autophagy could also in turn regulate TGF-β genes’ expression (Fig. 5f). LY21457299 is a clinical applied TGF-β inhibitors, however Western blotting demonstrated that LY2157299 had little effect on the endogenous TGF-β2 levels. After blocking autophagy flux with CQ or siATG5, the TGF-β2 level decreased (Fig. 5g).

Lactate was also reported to promote TGF-β synthesis and MMP2 production in glioma cells [37]. And lactate was associated with autophagy process tightly. We hypothesis that autophagy may mediate TGF-β2 autocrine via lactate secretion. In our experiments, lactate synthesis was increased in TGF-β2-stimulated tumor cells however decreased after siATG5 or CQ-treatment (Fig. 5h). And when we added exogenous lactate in cultured tumor cells, results confirmed that lactate could stimulate TGF-β2 and MMP2 expression and EMT-related protein changes, however CQ, except for LY2157299, could inhibit this process (Fig. 5i). Therefore, we summarized that TGF-β2 could induced autophagy in glioma via both Smad and non-Smad pathway (JNK pathway), and inhibiting autophagy could also in turn mediated endogenous TGF-β2 level in an autocrine fashion which might resulted from the changing of lactate level secreted by glioma cells.

In addition to the signal pathway activation and invasion related genes variation, there might be other mechanisms leading to the failure of TGF-β2 caused by autophagy inhibition. As tumor cell invasion is a highly energy- and material-consuming process [23, 24] and autophagy is tightly associated with tumor metabolism [26, 38, 39], We thus hypothesized that TGF-β2-induced autophagy is enhanced to generate more ATP and materials for invasion. First we introduced the MitoTracker Red probe to observe the mitochondrial morphology and localization. Based on one newly issued research [40], we identified the mitochondria morphology as fragmented, intermediate and tubular according to the length of the mitochondria (Fig. 6a, Additional file 4: Figure S4a). In our experiment, tubular form was found to be a more favorable form in the cell invasion apex and underwent more trafficking. TGF-β2-treated NTCs favored the tubular mitochondria, whereas autophagy inhibited tumor cells formed more fragmented mitochondria.

Mitochondrial trafficking can also be classified as three forms [41]: perinuclear, polarized and infiltrating (Fig. 6b, Additional file 5: Figure S5a) according to the distance from nuclei to mitochondria. Then TGF-β group showed an increased percentage of the polarized form and Baf group showed more perinuclear form (one-way ANOVA, polarized form, U251 Con 40 ± 2.1% vs TGF 72 ± 6.2% vs Baf 30 ± 5.5% p = 0.019, perinuclei Con 52 ± 3.7% vs TGF 20 ± 4.8% vs Baf 65 ± 6.2% p = 0.026). That indicated Baf stopped TGF-β2 stimulation reprogramming mitochondrial trafficking to fuel tumor cell invasion at a long distance (Fig. 6b).

Apart from above two factors, mitochondrial membrane potential (△Ψm) is also a vital factor for ATP generation. We used a JC-1 probe to detect the changes of △Ψm in mitochondria. JC-1 probe is an ideal probe that forms J-aggregates with red fluorescence in high △Ψm conditions and J-monomers with green fluorescence in low △Ψm conditions [42]. We observed red fluorescence increasing after TGF-β2 treatment in NTCs and green fluorescence enhancing in Baf treatment tumor cells though also treated with TGF-β2 (Fig. 6c, Additional file 5: Figure S5b). This phenomenon was confirmed by flow cytometry (FCM): one-way ANOVA, green percentage, U251 Con 4.55 ± 3.26% vs TGF 1.82 ± 0.67% vs Baf 15.55 ± 3.28% p = 0.038 (Fig. 6d).

As we could observe the changes of mito-fussion and fission in TGF-β2 and Baf treated groups, the exact molecular signal pathway were also explored by Western blot according to related studies [40, 43]. Results indicated that the phosphorylation of Drp1 and the expression of Opa3 and Mitofusion1 was responsible for the effect (Fig. 6e).

Besides the mitochondria changing, tumor cells were also characterized for aerobic glycolysis (Warburg-like metabolism). We thus recurred to XFe24 analysis (Seahorse) and found that TGF-β2 also altered metabolism program towards to aerobic glycolysis which resulted in a higher consumption rate of oxygen (OCAR) and an increased extracellular acidification (ECAR) to sustain ATP providing and then inhibiting autophagy could block this process (Fig. 6f). ATP assay kit and L-lactate testing kit also confirmed the same results (Fig. 6g, h). These results above provided a further explanation for how inhibiting autophagy might impair TGF-β2-stimulated glioma invasion: the mitochondria relocalization, morphological changes, down regulation in membrane potential and aerobic glycolysis reprogramming.

We used tumor xenograft models to test this finding and explore the clinical significance. After planting U87 cells in severe combined immune deficiency (SCID) mice, we assigned the animals to the following treatment groups: PBS group, LY2157299 group, CQ group and LY2157299 combined with CQ group. LY2157299 combined with CQ was the most favorable treatment for the mice with tumor xenografts according to the survival time and tumor volumes results monitored by 3 Tesla MRI (Fig. 7a, b), while groups with single CQ showed limited cure effect on tumors which was also in accord with other reference [44]. After 50 days, samples were obtained from the SCID mice and stained with HE, images showed that brains of groups with PBS was infiltrated with tumor cells among the normal brain and tumors borders, while combination treatment with LY2157299 and CQ showed sharp and clear borders between normal tissues and tumor (Fig. 7c). This may indicated a more moderate invasiveness in dual-drug used groups. With IHC, we could also observed a decreased expression level of TGF-β in LY2157299 treated groups (P = 0.0476) and combination drugs groups showed the least expression level (P = 0.0056) (Fig. 7d).The in vivo results showed the potential benefits of combined treatment using TGF-β and autophagy pathway inhibitors as CQ could improve the effect of TGF-β inhibitors. This finding provides additional evidence that TGF-β may partially act via autophagy to function in glioma invasion.