Metformin is a novel suppressor for transforming growth factor (TGF)-β1

Metformin is a widely used first-line antidiabetic drug that has been shown to protect against a variety of specific diseases in addition to diabetes, including cardiovascular disorders, polycystic ovary syndrome, and cancer. However, the precise mechanisms underlying the diverse therapeutic effects of metformin remain elusive. Here, we report that transforming growth factor-β1 (TGF-β1), which is involved in the pathogenesis of numerous diseases, is a novel target of metformin. Using a surface plasmon resonance-based assay, we identified the direct binding of metformin to TGF-β1 and found that metformin inhibits [125I]-TGF-β1 binding to its receptor. Furthermore, based on molecular docking and molecular dynamics simulations, metformin was predicted to interact with TGF-β1 at its receptor-binding domain. Single-molecule force spectroscopy revealed that metformin reduces the binding probability but not the binding force of TGF-β1 to its type II receptor. Consequently, metformin suppresses type II TGF-β1 receptor dimerization upon exposure to TGF-β1, which is essential for downstream signal transduction. Thus, our results indicate that metformin is a novel TGF-β suppressor with therapeutic potential for numerous diseases in which TGF-β1 hyperfunction is indicated.

target genes 6 . In our previous study, individual GFP-tagged Tβ RII molecules were imaged on the cell membrane using total internal reflection fluorescence microscopy (TIRFM) to study receptor activation. We demonstrated that Tβ RII occurs as a monomer in the resting state, and dimerizes upon TGF-β 1 stimulation 7 , which supports the idea that the receptor dimerization is essential for receptor activation.
Here, we identified a direct interaction between metformin and TGF-β 1 using a surface plasmon resonance (SPR)-based assay, and found that metformin inhibits [ 125 I]-TGF-β 1 binding to its receptor. We modeled the interaction mode between metformin and TGF-β 1 using molecular docking and molecular dynamics simulations. We further investigated the binding force between TGF-β 1 and Tβ RII using atomic force microscopy (AFM) and the dimerization of Tβ RII upon TGF-β 1 stimulation using single molecule fluorescence imaging. This study revealed a novel mechanism underlying the inhibition of the TGF-β 1 signaling pathway by metformin that explains the beneficial effects of metformin against organ fibrosis and cancer progression.

Results
Metformin directly interacts with TGF-β1, inhibits binding to its receptor, and attenuates downstream signaling. The SPR-based assay suggested that the binding of metformin to TGF-β 1 occurred with a K D value of 15.9 μ M (Fig. 1a). Next, we measured the effect of metformin on the specific binding of [ 125 I]-TGF-β 1 to 3T3 fibroblasts. As shown in Fig. 1b, metformin inhibited [ 125 I]-TGF-β 1 binding to its receptor in 3T3 fibroblasts in a dose-dependent manner (log[IC50] = − 4.16 ± 0.53). In contrast, HCT 116 cells that harbor a loss of function mutation in Tβ RII displayed negligible binding of [ 125 I]-TGF-β 1. And metformin has little effect on the [ 125 I]-TGF-β 1 binding to HCT 116 cells, suggesting that metformin inhibits the binding of [ 125 I]-TGF-β 1 to Tβ RII (see Supplementary Fig. S1). Consistent with these results, western blotting analyses indicated that metformin decreased TGF-β 1-induced phosphorylation of Smad2 and Smad3 Metformin interacts with TGF-β 1 to block its binding with receptors and attenuates downstream signaling (a) Sensorgrams for the binding of metformin and TGF-β 1. TGF-β 1 was covalently coupled to a CM5 chip, and metformin was injected in a two-fold dilution concentration series ranging from 62.5 μ M to 1.9 μ M. The steady-state values were calculated from the sensorgrams and plotted against the concentration. The data were fit to a single site binding model to calculate the K D . (b) Metformin inhibited [ 125 I]-TGF-β 1 receptor binding to 3T3 mouse fibroblasts. The results are expressed as the percentage of specific binding in the absence of metformin (n = 4). (c) Metformin and TGF-β 1 (0.5 ng/mL) were premixed for 2 h and then 3T3 cells were treated with the mixture for 30 minutes. Western blot analysis and quantification of phosphorylated-Smad2 (p-Smad2), Smad2, p-Smad3, Smad3 and GAPDH were performed. Data are mean ± SEM from 4 independent experiments. Kruskal-Wallis ANOVA combined with post-hoc Dunn's multiple comparison test (two tailed) was performed. * P < 0.05 vs. TGF-β 1 group.
Metformin antagonizes TGF-β1 signaling via a direct interaction with TGF-β1. To further assess the potential binding of metformin to TGF-β 1 and its receptor, we performed molecular docking and molecular dynamics (MD) simulation. The binding of metformin to TGF-β 1 was stable, determined by the root mean square deviation (RMSD) of metformin relative to TGF-β 1 (Fig. 3a,b). Metformin tended to bind in a cave-like structure consisting of the β -strand1 and β -strand2 of TGF-β 1 (Fig. 3c). Importantly, this site was partially overlaid with the binding interface of Tβ RII. The residues in direct contact with TGF-β 1 are depicted in Fig. 3d. The binding of metformin was largely attributed to the shape complementarity and hydrogen bond interaction between the guanidine group and Arg25. In addition, the nonpolar components (methyl groups) of metformin were nestled in the hydrophobic bottom of the cave. Thus, the binding stability of metformin to TGF-β 1 was further evaluated according to the binding free energy using Molecular Mechanics/Poisson Boltzmann Surface Area methods. The estimated value of the binding free energy (ΔG bind ) was − 68.50 kJ/mol, which was considered to be sufficiently strong for such a small compound. However, metformin could not stably bind to the putative binding site of Tβ RII(extracellular domain) (Fig. 3e). The molecule quickly diffused away from the initial binding site during the molecular simulation (Fig. 3f). Metformin inhibits the binding probability of TGF-β1 and TβRII but has no effect on the binding force of TGF-β1 and TβRII. To determine how metformin inhibits the binding of TGF-β 1 to its receptor, we next performed TGF-β 1-Tβ RII binding force measurements on live cells using AFM-based single-molecule force spectroscopy. Blocking experiments were performed by the addition of free TGF-β 1 monoclonal antibodies. As shown in Fig. 4a,b, the force distribution histogram displayed a single maximum by a Gaussian fit and the binding probability was less than 30%, indicating that the single molecule forces were measured. In the cells treated with metformin (50 μ M), similar binding forces (measured as the averaged histogram peak value) were observed for TGF-β 1 with Tβ RII on the cell surface as the control (Mann-Whitney U test with exact method, control vs. metformin: 49.5 ± 1.3 vs.49.3 ± 1.4 pN, P > 0.05, Fig. 4c). However, metformin decreased the binding probabilities from 21.7 ± 3.5% to 9.9 ± 1.2%, which was similar to the result of the TGF-β 1 antibody treatment (6.4 ± 1.9%, Fig. 4d).

Metformin inhibits TGF-β1-induced TβRII dimerization.
Because TGF-β 1-induced Tβ RII dimerization is a consequence of TGF-β ligand-receptor interaction and essential for receptor activation 7 , we next determined the effect of metformin on the formation of ligand-induced Tβ RII dimers. By analyzing the photobleaching traces (Fig. 5a), we found that 88.8% (778 out of 876 spots from 14 fixed cells) of individual Tβ RII-GFP molecules were monomers because they bleached in one step (Fig. 5b), 10.7% (94 of 876 spots) were dimers because they bleached in two steps (Fig. 5c), and that 0.5% (4 of 876) bleached in three steps. Following the TGF-β 1 stimulation, 67.7% (529 of 781 spots) bleached in one step as monomers, 31.6% (247 of 781spots from sixteen fixed cells) bleached in two steps as dimers and 0.6% (5 of 781) bleached in three steps. As shown in Fig. 5d, metformin inhibited the percentage of dimers induced by TGF-β 1 in a dose-dependent manner.

Discussion
In the present study, we demonstrated that metformin antagonized TGF-β 1 signaling by interacting with the TGF-β 1 ligand, thereby blocking the binding of TGF-β 1 to Tβ RII and resulting in decreased downstream signaling. This finding provides new indications for metformin, including numerous TGF-β 1 hyperfunction-associated diseases. Based on the interaction between metformin and TGF-β 1, new compounds with similar properties could be further developed.
It is generally accepted that metformin acts via the activation of AMPK and the inhibition of mitochondrial respiratory-chain complex 1 8 . Our previous study has shown that metformin inhibits TGF-β 1 induced collagen synthesis independent of AMPK activation 5 . Here, we further discovered that metformin antagonizes TGF-β 1 signaling by directly binding to TGF-β 1 which is independent of AMPK activation. Consistent with the well-established role of TGF-β 1 in the exacerbation of fibrosis 9 , our previous study and other studies have shown that metformin treatment attenuates cardiac fibrosis 5 , liver fibrosis 10 and renal fibrosis 11 . In addition, metformin has been shown to inhibit TGF-β 1-induced EMT which plays a key role in carcinoma progression and organ fibrosis 12,13 . Moreover, clinical trials have suggested that metformin is associated with decreased cancer risk and improved prognosis in cancer patients 14,15 . These findings support the idea that metformin exerts a protective effect against organ fibrosis and malignant tumor progression by blocking TGF-β 1. In addition to fibrosis and tumors, TGF-β is involved in numerous other diseases. Thus, targeting the TGF-β signaling pathway has become attractive for drug development. Currently, therapeutic strategies against the TGF-β family include three approaches: 1) inhibition at the translational level using antisense oligonucleotides, 2) inhibition of the ligand-receptor interaction using ligand traps and anti-receptor monoclonal antibodies, and 3) inhibition of the receptor-mediated signaling cascade using inhibitors and aptamers of TGF-β receptor kinases 9,16 . However, these approaches have specific challenges that limit their application, such as the limited ability of an antisense oligonucleotides and monoclonal antibodies to reach the targeted tissue 17,18 . In contrast, metformin is a small molecule compound that can easily reach the targeted tissue. Inhibitors of TGF-β receptor kinases have side effects that occur through the potential cross-inhibition of other kinases 9 . Conversely, metformin has been shown to be safe and have fewer side effects over decades of use. In addition, metformin has beneficial effects beyond targeting TGF-β 1 and based on the interaction mode between metformin and TGF-β 1, additional compounds can be developed to target TGF-β with higher specificity and potency.
In summary, our study identified metformin as a novel TGF-β 1 suppressor, and this action underlies the pleiotropic effects of the drug. This finding strongly supports the clinical use of metformin as a treatment for numerous diseases beyond diabetes where TGF-β 1 signaling malfunctions are indicated. In addition, our study provides insights that can be used in the development of new compounds targeting TGF-β 1.

Equipment and settings.
For single molecule fluorescence imaging, a custom-built TIRFM system was used as previously described 7 . The experiment was performed on TIRFM with 100X/1.45NA Plan Apochromat

Figure 5. Metformin inhibits TGF-β1-induced TβRII dimerization. (a) Typical single-molecule image of
Tβ RII-GFP on the HeLa cell membrane. After transfection with Tβ RII-GFP for 4 h, HeLa cells were imaged using total internal reflection fluorescence microscopy (TIRFM). The diffraction-limited spots (5 × 5 pixel regions) enclosed with green circles represent the signals from individual Tβ RII-GFP molecules, and they were chosen for the intensity analysis. Scale bar, 5 μ m. (b,c) Two representative time course graphs of GFP emissions after background correction demonstrating one-and two-step bleaching, respectively. The arrows indicate the bleaching steps. The individual Tβ RII-GFP molecules were monomers when they were bleached in one step (b) and dimers when they were bleached in two steps (c). (d) Metformin inhibited TGF-β 1-induced Tβ RII dimerization as shown by single-molecule imaging. Fraction of two-step bleaching events for Tβ RII-GFP molecules (counted spots were set as 100%) was represented as the dimer percentage. Prior to single-molecule fluorescence imaging, metformin and TGF-β 1 (10 ng/mL) were premixed for 2 h, and HeLa cells were then treated with the mixture for 15 min at 37 °C. The data were presented as the mean ± SEM (n = 5-16). ANOVA combined with Tukey's post-hoc test (two tailed) was used. * P < 0.05 vs. control group, # P < 0.05 vs. TGF-β 1 group.
TIR objective (Olympus, Japan) and a 14-bit back-illuminated electron-multiplying charge-coupled device camera (Andor iXon DU-897 BV). Imaging was performed at room temperature. GFP was excited at 488-nm by an argon laser (Melles Griot,Carlsbad, CA, USA) with the power of 1 mW measured after the laser passing through the objective. Movies of 200-300 frames were acquired for each sample at a frame rate of 10 Hz. Sequences of images were stored directly to a computer hard drive for subsequent analysis by IQ live cell imaging software (Andor Technology, BT, UK). SPR spectroscopy. Experiments were performed at 25 °C using a Biacore T200 and the data were analyzed using Biacore T200 evaluation software 2.0 (GE Healthcare, Stockholm, Sweden). TGF-β 1 was covalently coupled to a CM5 chip (GE Healthcare) and metformin was injected in a two-fold dilution concentration series ranging from 62.5 μ M to 1.9 μ M. The steady-state values were calculated from the sensorgrams and plotted against the concentrations. The data were fit to a single site binding model to calculate the K D .
TGF-β1 binding assay. The binding experiments were performed as previously described 19 . Briefly, 3T3 cells (American Type Culture Collection, ATCC ® CRL-2752 ™ ) or HCT 116 cells (American Type Culture Collection, ATCC ® CRL-247 ™ ) were seeded onto a 24-well-plate and cultured in DMEM supplemented with 10% Fetal Bovine Serum and antibiotics (100 U/mL penicillin-streptomycin). When cells were at a near-confluent stage, 50 pM [ 125 I]-TGF-β 1 with or without different concentrations of metformin were added. After 4 h at 4 °C, the medium was removed and cells were washed five times with ice-cold binding buffer (50 mM HEPES, 128 mM NaCl, 5 mM KCl, 5 mM MgSO4, and 1.2 mM CaCl2). The cells were then solubilized using binding buffer containing 1%Triton X-100 and the radioactivity was measured. Non-specific binding was determined in the presence of unlabeled TGF-β 1 (10 nM).
Western blotting analysis. Western blotting analyses were performed using specific antibodies as previously described 5  Theoretical study: molecular docking and molecular dynamics simulation. The geometry structure of metformin was optimized with Hartree-Forck methods at 6-31 + G* level of theory. The crystal structures of TGF-β 1 and the extracellular domain of Tβ RII, were retrieved from the PDB archives (3KFD) 20 . Autodock4.2 suite 21 was first applied to predict the preferential binding poses of ligand (metformin) in both TGF-β 1 and Tβ RII. Then the structure of both TGF-β 1 and Tβ RII bound with metformin were obtained for further evaluation by MD simulation. Amber99SB-ILDN forcefield 22 for protein and General Amber force field 23 for ligand was used. The charge parameters of ligand were taken from restrained electrostatic potential calculation 24 . The protein-ligand complex was solvated with TIP3P water. Sodium and Chloride ions were added to neutralize the system. All simulations were carried out with the GROMACS4.6.1 packages 25 and were run in NPT ensemble. The temperature (T = 300k) and pressure (p = 1atm) was kept constant using velocity scaling methods and Berendsen barostat methods, respectively. Based on the results of simulation, Molecular Mechanics/Poisson Boltzmann Surface Area methods 26 was used to estimate the binding free energy of metformin on protein.
Single molecule fluorescence imaging. Hela cells (American Type Culture Collection, ATCC ® CCL-2 ™ ) were transfected with Tβ RII-GFP plasmid for 4 h. Prior to the single-molecule fluorescence imaging, the cells were treated with TGF-β 1 (10 ng/ml for 15 min) and different concentrations of metformin and then washed twice and fixed. The single-molecule fluorescence intensity and photobleaching steps were also analyzed as previously reported 7 .
AFM tips preparation and AFM force measurements. TGF-β 1-modified AFM tips (type: NP-10, Bruker, Santa Barbara, CA, USA) were prepared as previously reported 27 . Hela cells were transfected with the Tβ RII-GFP plasmid for 24 h, and the force measurements were performed on a PicoSPM II system with a PicoScan 3000 controller and a large scanner (Agilent, Santa Clara, CA, USA). The AFM scanner was mounted on an inverted fluorescence microscope (Olympus IX71, Japan). The fluorescent protein-labeled cells were used to guide the AFM tips on the cell expressing Tβ RII. All of the force curves were measured with the contact mode at room temperature using a soft cantilever (0.06 N m −1 ). The loading rate of the force measurements was 1.0 × 10 4 pN/s. The force curves were recorded using PicoScan 5 software (Molecular Imaging, Tempe, AZ) and analyzed using a user-defined program in MATLAB (MathWorks Corp., Beijing, China).