Development of antitumor biguanides targeting energy metabolism and stress responses in the tumor microenvironment

To develop antitumor drugs capable of targeting energy metabolism in the tumor microenvironment, we produced a series of potent new biguanide derivatives via structural modification of the arylbiguanide scaffold. We then conducted biological screening using hypoxia inducible factor (HIF)-1- and unfolded protein response (UPR)-dependent reporter assays and selective cytotoxicity assay under low glucose conditions. Homologation studies of aryl-(CH2)n-biguanides (n = 0–6) yielded highly potent derivatives with an appropriate alkylene linker length (n = 5, 6). The o-chlorophenyl derivative 7l (n = 5) indicated the most potent inhibitory effects on HIF-1- and UPR-mediated transcriptional activation (IC50; 1.0 ± 0.1 μM, 7.5 ± 0.1 μM, respectively) and exhibited selective cytotoxicity toward HT29 cells under low glucose condition (IC50; 1.9 ± 0.1 μM). Additionally, the protein expression of HIF-1α induced by hypoxia and of GRP78 and GRP94 induced by glucose starvation was markedly suppressed by the biguanides, thereby inhibiting angiogenesis. Metabolic flux and fluorescence-activated cell sorting analyses of tumor cells revealed that the biguanides strongly inhibited oxidative phosphorylation and activated compensative glycolysis in the presence of glucose, whereas both were strongly suppressed in the absence of glucose, resulting in cellular energy depletion and apoptosis. These findings suggest that the pleiotropic effects of these biguanides may contribute to more selective and effective killing of cancer cells due to the suppression of various stress adaptation systems in the tumor microenvironment.

Biological evaluation. As a first-stage screening, cell-based luciferase reporter assays were undertaken in response to hypoxic and glucose deficiency stress. The luciferase assay was performed as reported previously 24 using our stable cell lines of HEK293 transfected with the HIF-1-responsive luciferase reporter plasmid p2.1 30,31 or the pGRP78pro160-luc plasmid containing the human glucose-regulated protein (GRP) 78 promoter with an ER stress response element 24,32 . Subsequently, a 3-(4,5-dimethylthiazol-2-yl)2,5-diphenyltetrazolium bromide (MTT) assay was performed on HT-29 colorectal adenocarcinoma cells in glucose-containing and glucose-free media in order to assess selective cytotoxicity under glucose-deprived conditions. As shown in Table 1, when bromine (7a) or iodine (7b) was used as the halogen substituent at the o-position, the inhibitory effects on HIF-1-and UPR-mediated transcription activations were stronger, but the selective cytotoxicity with low glucose was weaker than in the case of the chlorine substituent (entries [3][4][5]. Our SAR studies of biguanide derivatives found that upsilon steric parameters 33 correlated with cytotoxicity under low glucose conditions 24 24 in the para position, caused significant reductions in the cytotoxicity (Glc −). In the homologous series of aryl-(CH 2 ) n -biguanide (n = 0-6), hydrophobicity increased and biological activity became stronger with increasing alkylene linker length. (entry 9-14, Fig. 3). In particular, 7j (n = 5), 7k (n = 6) and the o-chloro phenyl analogue 7l (n = 5) showed potent inhibitory effects on both HIF-1-and UPRmediated transcriptional activation and were highly selectively cytotoxic under glucose-deprived conditions (entry [13][14][15]. We evaluated hydrophobicity for biguanides by Rm values calculated from TLC-relation factor (Rf) by using octylsililated silicagel plate 34,35 and calculated log D (clog D), distribution coefficients at physiological pH. According to the regression analyses of the homologation series data ( Figure S2A), there was a very good positive correlation between the methylene number of the alkylene linker and the hydrophobicity parameters (Rm; R 2 = 0.9758, clog D; R 2 = 0.9971) and between Rm and clog D values (R 2 = 0.9699). The Rm value of 7l was almost equal to that of 7j with the same linker length (n = 5) on the fitted regression line, but its clog D was far off the regression line to the hydrophobic side. Therefore, we analyzed the correlation of the IC 50 for each biological activity to Rm values for mono-substituted biguanides (entries [1][2][3][4][5][6][7][8][9][10][11][12][13][14][15]. Interestingly, the IC 50 of inhibitory activity of UPR-mediated transcriptional activation correlated well with the Rm value (R 2 = 0.8653, Figure S2B). However, the IC 50 for HIF-1-mediated transcriptional activation and cytotoxicity (G + and G −) did not correlate with the Rm in either case (R 2 = 0.3-0.7, Figure S2B-S2D). Since anti-diabetic biguanides are positively charged at neutral pH and have low hydrophobicity, they require organic cation transporters (OCTs) to facilitate their uptake into cells 36 . On the other hand, phenformin, which is more hydrophobic than metformin, was also reported to directly permeate the cell membrane 37 . In contrast, most of the biguanides shown here are much more hydrophobic than phenformin as predicted by their Rm values, and thus are more likely to cross the cell membrane by direct diffusion. Therefore, hydrophobic factors are considered to be of significance in these biological activities targeting the cancer microenvironment. In this connection, Bridges et al. reported a remarkable finding 38 . Namely, bis-substituted biguanides or those with a direct conjugated phenyl group inhibited isolated complex I, but did not inhibit mitochondrial respiration because they were not taken up by mitochondria regardless of hydrophobic parameters. This suggests a selective transporter across the mitochondrial inner membrane. As for our disubstituted biguanides with substituents at N1 position, 7m, 7n, and 7o, shown in Table 1 (entries 16-18), were not as potent, but exhibited HIF-1 and UPR inhibition and low glucose-selective cytotoxicity comparable to those of phenformin (1). These results indicate that the selective cytotoxicity of new biguanides is not solely attributed to mitochondrial damage.   24 . N.D., not determined. Concentration response curves for biguanide 7l can be found in Figure S3. www.nature.com/scientificreports/ Furthermore, compounds 2, 3 and 7l, which showed strong activity, were evaluated for selective cytotoxicity under glucose deprivation in A549 lung cancer and U87-MG glioma, both of which have been suggested to be more dependent of glycolytic processes for their energy metabolism than HT29 39,40 (Table 2). Among these compounds, 7l showed the most potent cytotoxicity under glucose deprivation in all tested cell lines. It is also noteworthy that 7l is effective against a wide variety of cancer cells and exhibits strong cytotoxicity, even against the highly malignant U87MG glioma, due to mitochondrial dysfunction and metabolic reprograming 41 .

Entry
The above screening led to the most promising compound 7l being selected for examination of its effect on the expression of HIF-1α and UPR-mediated protein expression. As shown in Fig. 4A and B, compound 7l clearly suppressed HIF-1α protein expression induced by hypoxia at concentrations of 1 μM, one-tenth of the effective concentration of compounds 2 and 3 24 . Additionally, compound 7l completely inhibited the induction of protein expression in GRP78 and GRP94, the major regulators of UPR, by glucose deprivation at concentrations of 3 μM.  Table 2. Cytotoxicity of 2, 3 and 7l on HT29, A549 and U87MG cells. IC 50 (mean ± SD, n = 3) was determined by MTT assay on each cell following 48 h of drug treatment in the presence (Glc +) or absence (Glc −) of glucose. a Data from ref. 24 . Concentration response curves for biguanide 2, 3 and 7l on A549 and U87MG and 7l on HT29, A549 and U87MG can be found in Figures S3, S4  www.nature.com/scientificreports/ Phenformin and metformin are known to inhibit angiogenesis by reducing the production of angiogenesisrelated proteins 17 . The in vivo inhibitory effect of such compounds on angiogenesis can be evaluated by chick chorioallantoic membrane (CAM) assay 42,43 . We previously demonstrated that phenformin (1) and 2 inhibited neovascularization in chick embryos at 5 μg/CAM and 2 μg/CAM, respectively 24 , but here found that 7l significantly inhibited angiogenesis at a much lower dose, 0.5 μg/CAM, as shown in Fig. 4C. Birsoy et al. demonstrated that the sensitivity of cancer cells to biguanides, which are OXPHOS inhibitors, is enhanced under low glucose conditions and that OXPHOS is a major pathway required for optimal cancer cell proliferation under such conditions 44 . Therefore, in order to investigate the effects of new biguanides on the energy metabolism of tumor cells, we evaluated mitochondrial function using the oxygen consumption rate (OCR) as an indicator and glycolytic function using extracellular acidification rate (ECAR) in living HT29 cells under normal and low glucose conditions using the XFp extracellular flux analyzer 45,46 . ECAR is primarily a measure of lactate production and can be equated to glycolytic rate. We measured changes in OCR and ECAR of HT-29 cells in the absence or presence of biguanides 1, 2 and 3 in response to the sequential addition of the following mitochondrial inhibitors: the ATP synthase inhibitor oligomycin, the mitochondrial uncoupler FCCP, and the complex I/III inhibitors rotenone and antimycin A. As shown in Fig. 5A mitochondrial basal and maximal respirations were more strongly inhibited by 2 than by 1 and The energy phenotype profile diagram shows that the presence of biguanides led to a significant reduction in OCR both in the presence and absence of glucose in all cell lines; the order of the strength of inhibition was 7l > 3 > 2 > 1. In total, of the compounds examined, 7l inhibited OXPHOS most potently, such that the inhibitory effect was observed even at a concentration of 25 μM. In the presence of glucose, stronger OXPHOS inhibitor caused greater increases in ECAR. In addition, when energy production by mitochondrial The differences in OCR values between each compound treatment were statistically significant (*p < 0.05; **p < 0.005; ***p < 0.0005). Each data represents mean ± SD of triplicate experiments. www.nature.com/scientificreports/ respiration is inhibited by OXPHOS inhibitor, the glycolytic system is normally activated to compensate for the lack of energy. The limited availability of oxygen also enhances the glycolytic system via activation of HIF-1 signaling, resulting in lactic acidosis as a side effect. On the other hand, the present biguanide strongly inhibits protein expression and transcriptional activation of HIF-1α at lower concentrations then OXPHOS inhibition, which may prevent acidosis. Furthermore, recently, Minamishima et al., proposed a hopeful strategy to avoid acidosis by activating HIF-1 with a prolyl hydroxylase domain-containing protein 2 (PHD2) inhibitor 47 . Activation of HIF-1 signaling by PHD2 inhibitor would enhance the Cori cycle, which activate gluconeogenesis to reduce lactate. In cancer treatment, metformin has been reported to not cause acidosis or hypoglycemia in non-diabetic patients, but the cancer treatment effects and side effects of phenformin have not been well studied. New compound 7l showed potent low glucose-selective cytotoxicity, HIF-1 inhibition, and UPR inhibition, all of which were about 20-fold higher than phenformin, but no OXPHOS inhibition was observed at doses effective for these biological effects. These suggest that the adverse effect, lactic acidosis, may be alleviated with appropriate dosing or combination therapy 19 .

Fluorescence-activated cell sorting (FACS) analysis of cell death.
It has been reported that the antiproliferative activity of biguanides is mediated by the induction of apoptosis in breast cancer cells and is markedly enhanced under glucose-deficient conditions 48 . To evaluate the cell death of HT29 cells via biguanide treatment in media with variable glucose levels, flow cytometric analysis was performed by double staining with Annexin V-FITC and propidium iodide (PI). HT29 cells were treated with compounds 2, 3, and 7l at concentrations of 10 µM, 10 µM, and 5 µM, respectively, for 48 h and subjected to FACS analysis. The total ratios of early (Annexin V-FITC+ /PI−) and late (Annexin V-FITC+/PI+) apoptosis induced by compounds 2, 3, and 7l in the low glucose medium were 53.8%, 47.9%, and 70.8%, respectively (Fig. 6). On the other hand, minimal cell death was observed in cells exposed to the compounds in the presence of glucose, suggesting that biguanide treatment selectively promoted apoptosis at low levels of glucose. These results are consistent with previous reports by Ben Sahra et al., who found that energy depletion with metformin and 2-deoxy-d-glucose (2-DG) induced apoptosis in prostate cancer cells 49 .

Conclusions
To develop therapeutic agents targeting energy metabolism in the tumor microenvironment, we developed a series of potent new biguanide derivatives via structural modification of the aryl biguanide scaffold and screening using HIF-1-and UPR-dependent reporter assays and selective cytotoxicity assays under low glucose conditions. Homologation studies of aryl-(CH 2 ) n -biguanides (n = 0-6) yielded strong derivatives with appropriate alkylene linker lengths (n = 5, 6). The o-chlorophenyl derivative 7l was found to be the most promising compound, such that its inhibitory activities were tens of times stronger than those of phenformin (1). Furthermore, biguanide 7l, at the lowest dose of the series of compounds synthesized here, also markedly reduced the protein expression of HIF-1α induced by hypoxia and the protein expression of GRP78 and GRP94 induced by glucose starvation, in addition to inhibiting angiogenesis. Although these biguanides resulted in potent HIF-1 inhibitory activity, no selective cytotoxicity was observed under hypoxic conditions (1% O 2 ). It has been reported that the sensitivity www.nature.com/scientificreports/ of cancer cells to OXPHOS inhibitors, such as biguanides, is enhanced by low levels of glucose, especially in cells with abnormal mitochondrial function and impaired glucose utilization 44 . Therefore, among the diverse biological activities of biguanides, their modulating effects on energy metabolism may be the main contributor to their selective cytotoxicity under low glucose conditions. Metabolic flux analysis of tumor cells revealed that these newly produced biguanides strongly inhibit OXPHOS and activate compensative glycolysis in the presence of glucose, whereas both are strongly suppressed in the absence of glucose, resulting in cellular energy depletion and apoptosis. Since apoptosis induced by energy stress in combination with metformin and 2-DG is mediated by AMPK 50 , the apoptosis induced by energy depletion in the presence of the biguanides reported herein may also be related to AMPK signaling. We have previously shown that the biguanide derivatives suppress HIF-1 and UPR-dependent transcriptional activation and the expression of c-Myc and ATF4 proteins, while they activate AMPK 24,26 . Biguanides will be utilized effectively through synthetic lethal strategies by exploiting the specific metabolic vulnerability of transformed cells or by inducing changes in the microenvironment with drugs 49,51 . In this paper, we present new potent biguanides that can efficiently inhibit mitochondrial respiration under glucose deprivation conditions in tumor microenvironment, ultimately leading to cell death due to energy depletion. In addition, they can potently suppress HIF-1 and UPR signaling, which are key factors for adaptation to cancer microenvironmental stress. As a result, the new biguanides may have multifunctional properties that disable stress response mechanisms, such as hypoxia, endoplasmic reticulum stress, and metabolic stress, and cause synthetic lethality in combination with potent OXPHOS inhibition.

Experimental section
General synthetic procedure. All 13  General procedure for the synthesis of biguanides 2, 3 and 7a-7q 24 . The reaction procedure followed the literature procedure. The amine derivatives (compounds 4, 5 and 6a-6q, 0.3-6.23 mmol) was added to a solution of dicyandiamide (1.0 eq.) in 0.8-5.0 mL of CH 3 CN, and then TMSCl (1.1 eq.) was slowly added dropwise to the mixture. With Initiator 2.0, the mixture was stirred and irradiated for 10-30 min at 130 or 150 °C. After the mixture was cooled down to approximately 50 °C, iPrOH (3.0 eq.) was added slowly and the mixture was further stirred and irradiated at 125 °C for 1 min. The precipitate was filtrated and washed with CH 3 CN twice to give the biguanide hydrochloride salt. 24 Following the general procedure for the synthesis of biguanides, the reaction of 4 (540 mg, 4.0 mmol) was performed (130 °C, 10 min, 12.0 mL CH 3 CN) to give 2 hydrochloride (764 mg, 75%). The analytical sample was obtained by recrystallization from EtOH to obtain 2 dihydrochloride as a colorless powder. mp 171.5-172.5 °C (lit. 173-174 °C) 24 .
Then treated with various concentrations of compounds in the normal or glucose-free medium for 48 h. Then the medium was replaced with fresh growth medium, and cells were cultured for further 16 h. Subsequently, 10 μL of thiazolyl blue tetrazolium bromide (Sigma-Aldrich, St Louis, MO, USA) solution (0.5 mg/mL) was added to each well. After 4 h incubation at 37 °C, the medium was removed, 100 μL of DMSO was added then absorbance of each well was measured at 570 nm by MULTISKAN JX plate reader. Relative cell survival (mean ± SD of triplicate determinations) was calculated by setting each of the control absorbance from non-drug treated cells as 100%. www.nature.com/scientificreports/ ture. Then the membranes were incubated, at 4 °C for 10 h with the anti-HIF-1α antibody (diluted 1:1,000), at room temperature for 1 h with anti-KDEL antibody (diluted 1:1,000), or at room temperature for 1 h with the anti-β-actin antibody (diluted 1:3,000). The membranes were washed with TBS-T containing 5% nonfat dry milk or 1% BSA at room temperature, and then incubated at room temperature for 1 h with appropriate horseradish peroxidase-labeled secondary: Anti-Mouse IgG (whole molecule) peroxidase conjugate (A4416; Sigma-Aldrich) and Anti-Goat IgG (whole molecule) peroxidase conjugate (A5420, Sigma-Aldrich). After washing with TBS-T, the specific signals were detected with an enhanced chemiluminescence detection system (Pierce Western Blotting Substrate, Thermo Scientific) or chemiluminescence detection system (Immobilon Western Chemiluminescent HRP substrate, Merck Millipore, Billerica, MA, USA) and visualized with a Fujifilm Luminescent Image Analyzer LAS-3000 (Fujifilm, Tokyo, Japan).

Rm value.
OCR and ECAR measurements for the analysis of cell energy phenotypes. Cellular OCR and ECAR of live cells were measured using a Seahorse XFp Extracellular Flux Analyzer (Agilent, Santa Clara, CA, USA). The cellular mitochondrial function and glycolytic rate were analyzed via the XF Cell Mito Stress Test and XF Glycolysis Stress Test kit (Agilent), respectively according to the manufacturer's protocol. In brief, HT29, A549 or U87MG cells (2.5 × 10 4 cells/well) were seeded into an XFp cell culture microplate and incubated for 24 h. The sensor cartridges of the XFp analyzer were hydrated in a 37 °C non-CO 2 incubator the day before the experiment. To examine the mitochondrial function, injection port A on the sensor cartridge was loaded with 1.5 μM oligomycin (complex V inhibitor), port B was loaded with 2 μM carbonyl cyanide-4-(trifluoromethoxy) phenylhydrazone (FCCP), and port C with 0.5 μM rotenone/antimycin A (inhibitors of complex I and complex III). During sensor calibration, cells were treated with 5-50 μM of test compounds and incubated in a 37 °C non-CO 2 incubator together with 180 μM of assay medium (XF base medium with 10 mM glucose, 1 mM pyruvate, and 2 mM l-glutamine, pH 7.4) for 1 h. The plate was immediately placed into the calibrated XFp Extracellular Flux Analyzer. For the glycolysis stress test, injection port A on the sensor cartridge was loaded with 10 mM glucose, port B was loaded with 1 μM Oligomycin, and port C with 50 mM 2-deoxy-D-glucose. During the sensor calibration, cells were treated with 5-50 μM of test compounds and incubated in a 37 °C non-CO 2 incubator together with 180 μL of assay medium (XF base medium with 2 mM l-glutamine, pH 7.4) for 1 h. The plate was immediately placed into the calibrated XFp Extracellular Flux Analyzer. The each average of the OCR values measured at three points during the first 15 min in the profile of mitochondrial function and the ECAR values measured simultaneously were used as the respiratory inhibitory activity of biguanides in the presence of glucose (Glc +). Similarly, the mean values of ECAR and OCR determined from the glycolytic functional profile were used as respiratory inhibitory activity under glucose-depleted conditions (Glc −). Then, the OXPHOS and ECAR were measured after treatment of HT29 with biguanides 1, 2, 3 and 7l under 10 mM glucose (Glc +) and glucose deprivation conditions (Glc −) in a non-CO 2 incubator. DMEM-based medium supplemented with 10 mM glucose, 1 mM sodium pyruvate, and 2 mM l-glutamine at final concentrations was prepared for the normal glucose assay medium, and 2 mM l-glutamine supplemented medium was used for the glucose-free assay condition.
CAM assay in fertilized chicken eggs. The antiangiogenic effect of compounds in vivo was evaluated by CAM assay as reported procedure 42 . Briefly, fertilized chicken eggs were incubated at 37.5 °C in a humidified incubator with forced air circulation. Ovalbumin (3 mL) was removed from 3-day-old embryonated eggs. Then a small hole was drilled on each shell of the egg and capped, and the eggs were incubated at 39.0-39.5 °C. After a 1-day incubation, each compound saline solution, with 2.0% DMSO and 1.0% methylcellulose, was applied on the center of silicon rings (outer diameter, 5 mm; inner diameter, 3 mm; height, 1 mm) that were placed on each of the CAMs, and the eggs were incubated at 39.0-39.5 °C for 2 days. A white 20% intralipos solution (Otsuka Pharma-ceutical, Japan) was injected beneath the CAM to enhance visibility of the overlying superficial CAM vessels. The images were captured with a digital camera and scored on the seventh embryonic day. Saline solution, with 2.0% DMSO and 1.0% methylcellulose, was used as a vehicle. Eight to ten eggs were used in total for each data point. The inhibition point was scored by an estimation of the area of the avascular zone. The inhibition ratios were calculated from the following formula: Inhibition ratio (%) = [1 − (point for control CAM/point for drug treated CAM)] × 100.

FACS analysis of cell death.
Apoptosis and cell death were analyzed using the Annexin V-FITC Detection kit according to its instruction manual. Briefly, HT29 cells were seeded at 5 × 10 5 cells/100-mm dish. After exposure to 2, 3 or 7l at concentrations of 10 or 5 μM in normal or glucose-free medium for 48 h, the cells were washed twice with cold PBS, and suspended in a 1× binding buffer at a concentration of 5 × 10 5 cells/mL, after which Annexin V-FITC and propidium iodide were added. Following incubation for 5 min at room temperature in the dark, 1× binding buffer (500 μL) was added to each tube. The cells were analyzed using a BD FACSVerse flow cytometer and BD FACSuite v1.0.5.3841 software (BD Biosciences) and the fraction of the cell population in different quadrants was determined using quadrant statistics.

Statistical analysis.
For the results of OCR and ECAR measurements, statistical differences between the three groups of samples treated with three compounds were evaluated by Student's t-test as indicated in each figures (*p < 0.05; **p < 0.005; ***p < 0.0005). Dose-response results of the reporter assays and cell viability assays are presented as the mean ± SD of three independent experiments.