Mitochondrial Targeting of Metformin Enhances Its Activity against Pancreatic Cancer

Cancer is one of the most serious pathologies in industrialized countries with rather grim prognosis (1). Emerging evidence indicates that cancer is primarily a metabolic disease arising in response to disturbance in cell energy homeostasis (2, 3). Many proto-oncogenes and tumor suppressors have been shown to regulate cell metabolism (4). A link between metabolic disorders and cancer is supported also by epidemiological evidence indicating that pathologies like type 2 diabetes mellitus (T2DM) are associated with increased risk of different types of malignancies, pancreatic cancer being a prime example (5). Retrospective as well as epidemiological and clinical studies indicate that therapy with metformin, the first-line drug of choice for treating T2DM, is associated with decreased incidence of cancer and increased survival rate of patients with different types of tumors (6–9).

Metformin is considered to be a promising drug in regard to treatment and prevention of pancreatic cancer, one of the most fatal human pathologies (10–12). The agent, which has been used for therapy of diabetes since 1950s (13), is recognized as a safe drug. Recent in vitro studies have demonstrated that metformin acts directly on tumor cells to suppress their proliferation (14–16), targeting also pancreatic cancer stem cells (17). The clinical impact of these observations is currently unclear, as the antiproliferative effects of metformin become apparent only at supra-pharmacological concentrations of the drug (18).

At the molecular level, metformin primarily targets mitochondria, inhibiting complex I of the respiratory chain (19, 20). Alternations in mitochondrial function are believed to be responsible for anticancer effects of metformin, as it restricts the ability of tumor cells to cope with energetic stress (21). This concept is supported by emerging literature showing that mitochondrial function is tightly linked to cancer (22).

Recent reports document that mitochondrial respiration is important for tumor initiation, progression, and metastasis (23–25). This led us to coin the hypothesis that targeting mitochondrial complexes may be an efficient way to treat cancer (26, 27). With this in mind, we designed, synthesized, and tested for their anticancer efficacy several mitochondrially targeted drugs acting via mitochondria that were tagged with triphenylphosphonium (TPP⁺; refs. 28, 29). This delocalized cationic group anchors small molecules with pro-oxidant function at the interface of the mitochondrial inner membrane (MIM) and matrix (30), allowing the drugs to accumulate at their primary site of action. We now applied TPP⁺ tagging to metformin in order to maximize its activity, and report that this approach enhances toxicity of the parental drug towards pancreatic cancer cells by three to four orders of magnitude, making mitochondrially targeted metformin (MitoMet), an exceptionally promising anti-cancer agent.

For information regarding the number of cells seeded in each experiment, see Supplementary Table S1.

All chemicals were purchased from Sigma-Aldrich if not stated otherwise. Stock solutions of metformin (Enzo Life Sciences) and the TPP⁺-modified compounds were prepared by their dissolving in water. For animal experiments, the compounds were dissolved in PBS.

The synthesis and physico-chemical properties of MitoMet (compound 9), norMitoMet (compound 7), C6 MitoMet (compound 10), and C6 norMitoMet (compound 8; Fig. 1) is described in detail in the Supplementary Methods. 11C-TPP was prepared as described earlier (31).

A panel of human pancreatic cancer cell lines and nonmalignant control cell lines was used. PANC-1, MiaPaCa-2, BxPC-3, AsPC-1 cells, BJ skin fibroblasts, MRC-5 lung fibroblasts, MCF-10A breast epithelial cells, and EA.hy926 endothelial hybridoma cells were purchased from ATCC within last 6 years. PaTu 8902 cells were obtained from DSMZ in 2011. HFP1 skin fibroblasts were a kind gift from K. Smetana (Institute of Anatomy, Charles University, Prague, Czech Republic) (32). The last authentication of the cell lines was performed in 2016 using STR profiling. Cells were routinely cultured in DMEM (PANC-1, PaTu 8902, BJ, MRC-5, HFP1, and EA.hy926; Lonza) or RPMI (BxPC-3 and AsPC-1 cells; Lonza) supplemented with 10% FBS (Life Technologies), nonessential amino acids (Life Technologies), l-glutamine, and antibiotics, at 37°C and 5% CO₂. MiaPaCa-2 cells were cultured in Gibco DMEM (Life Technologies) with the same supplements as for the other cell lines. DMEM containing 5% horse serum, 20 ng/mL epidermal growth factor (Life Technologies), 0.5 mg/mL hydrocortisone, 100 ng/mL cholera toxin, 10 μg/mL insulin, and antibiotics was used for MCF10A culture.

The effect of tested compounds on cell proliferation was assessed by the crystal violet assay. Cell were exposed to the agents for 48 hours, unless stated otherwise, and fixed with 4% paraformaldehyde in PBS for 20 minutes at 37°C. Cells were than washed with PBS, stained with crystal violet (0.05% in water) for 1 hour. After 3 washing cycles, the crystal violet dye was extracted with 1% SDS and absorbance was determined at 595 nm.

Cells were seeded in the e-plate 96 (ACEA Biosciences) in 100 μL media per well and were transferred into xCelligence real-time cell analysis (RTCA) SP station (ACEA Biosciences) located in humidified 37°C chamber with 5% CO₂. Tested compounds were applied 24 hours post-plating. Impedance was measured in defined intervals for 100 hours. The data were evaluated using the RTCA software.

Treated cells and nontreated controls were harvested and lysed in RIPA buffer supplemented with protease and phosphatase inhibitors. Protein lysates were separated by SDS-PAGE and transferred to a nitrocellulose membrane (Bio-Rad Laboratories). After probing with specific antibodies, proteins were detected using SuperSignal West Femto Maximum Sensitivity Substrate or Pierce ECL WB Substrate (Thermo Scientific). Primary antibodies for cleaved caspase-3 (#9664), AMP-activated protein kinase α (AMPK; #2532), phospho-AMPKα (p-AMPK; Thr172; #2531), acetyl-CoA carboxylase (ACC; #3662), p-ACC (Ser79; #11818), raptor (#2280), p-raptor (Ser792; #2083), mammalian target of rapamycin (mTOR; #2972), p-mTOR (#2971), and β-actin (#5125) were purchased from Cell Signaling. PANC-1 cells used for AMPK pathway analysis were cultivated in media with decreased glucose concentration (1 g/L). Western blot (WB) signals were quantified using Quantity One analysis software (Bio-Rad).

PANC-1 or BJ cells were seeded in 12-well plates and allowed to attach overnight. After 48-hour incubation with norMitoMet, metformin or vehicle, both adherent and floating cells were collected, washed with PBS, resuspended in 100 μL of annexin binding buffer and incubated with 0.3 μL fluorescein isothiocyanate (FITC)-labeled annexin V (Apronex) for 30 minutes. Propidium iodide (PI) was added to identify cells with disrupted plasma membrane. Annexin V-positive fraction was determined by flow cytometry (FACS Calibur or LSRFortessa; BD Biosciences).

NDI1 and control pWPI vectors containing GFP were transfected into HEK293T cells using lipofectamine 3000 (Invitrogen) together with psPAX2 and pMD2.G packaging vectors. The resulting lentiviruses were used for transduction of parental PaTu 8902 cells. Fluorescence-activated cell sorting for GFP-positive cells using FACS Aria Fusion (BD Biosciences) was performed to select for NDI1-expressing and control cells.

Respiration of intact cells and CI, complex II (CII) or glycerol-3-phosphate dehydrogenase (G3PDH)-specific respiration in permeabilized cells was assessed using Oxygraph-2k (Oroboros). Procedure details are described in the Supplementary Methods.

Newborn 10-day-old rats of Wistar strain were used to obtain interscapular brown adipose tissue. The tissue was homogenized in media containing 320 mmol/L sucrose, 10 mmol/L Tris-HCl, 1 mmol/L EDTA, and 0.5 mg/mL BSA with pH adjusted to 7.4. Mitochondria were isolated by differential centrifugation and resuspended in homogenization media with no BSA added. Isolated mitochondria were stored at −80°C. Frozen-thawed mitochondria suspended in the K medium (80 mmol/L KCl, 10 mmol/L Tris-HCl, 5 mmol/L K-phosphate, 3 mmol/L MgCl₂, 1 mmol/L EDTA, pH 7.4) were utilized for high-resolution respirometry using Oxygraph-2k. The inhibitory effect of metformin and norMitoMet on GPDH-mediated respiration stimulated by 10 mmol/L glycerol 3-phosphate (G3P) was determined by stepwise addition of tested compounds directly into the chamber of the Oxygraph-2k instrument.

Extracellular acidification rates (ECAR) and oxygen consumption rates (OCR), respective measures of glycolytic flux and mitochondrial respiration, were assessed for a panel of pancreatic cancer cell lines using the Seahorse XF-24 analyzer (Seahorse Biosciences). Cells were plated in the Seahorse XF24 cell culture microplates in standard culture media. After 24 hours, the medium was replaced with Seahorse XF base medium supplemented with 0.2% BSA and 10 mmol/L glucose, and the microplates were placed in non-CO₂ incubator for 30 to 60 minutes. The assay protocol consisted of four consecutive injection steps in which 1 μmol/L oligomycin, 0.5 μmol/L FCCP, 1 μmol/L FCCP, and the combination of 100 mmol/L 2-deoxyglucose, 1 μmol/L rotenone, and 1 μg/mL antimycin A were added. Maximal respiration was determined as the maximal OCR stimulated by FCCP. The elevated rate of glycolysis after oligomycin addition is referred to as glycolytic capacity. After terminating the measurement, cells were lysed in the RIPA buffer and the protein content was determined using the Pierce BCA Protein Assay Kit (Thermo Scientific). Data were normalized to the amount of protein present in each well of the microplate.

Cells were seeded in 12-well plates, left to attach overnight, and treated as indicated. 15 minutes before collecting the cells, 5 μmol/L 2′,7′-dichlorofluorescin diacetate (DCF-DA) and 50 nmol/L tetramethylrhodamine methyl ester (TMRM), probes for monitoring ROS production and ΔΨ_m,i, respectively, were added. Harvested cells were resuspended in PBS containing 50 nmol/L TMRM and analyzed by flow cytometry (FACS Calibur). The level of TMRM fluorescence in cells with ΔΨ_m,i dissipated by CCCP pretreatment was used as a baseline for ΔΨ_m,i measurements. MitoSOX Red dye (Life Technologies) applied at 1.25 μmol/L concentration for 15 minutes was used to detect mitochondrial superoxide production by flow cytometry.

The recently deposited crystal structure of yeast CI from Yarrowia lipolytica (PDB ID 4wz7; ref. 33) was used for modeling. The geometry of a set of possible tautomeric forms of MitoMet was optimized using the DFT-D method (34) with TPSS functional and TZVP basis set (35). The effect of water solvation was treated implicitly using COSMO (36) with ϵ = 78.4. All optimizations were performed in the TurboMole suite of programs. Optimized geometry of two most stable forms of MitoMet was used for the docking study. The Python Molecular Viewer (PMV 1.5.6 rc3) was used to set the docking parameters. MitoMet was then allowed to sample docking poses in a box (90 × 90 × 90 grid points, 1.0 Å spacing) covering the lower part of the peripheral arm (Q module) and the transmembrane P_P module of the membrane arm. Results of five separate docking runs for each MitoMet form were collected employing AutoDock Vina version 1.1.2. The program 3V (37) was used to identify internal cavities connecting the iron-sulfur clusters with the ubiquinone (UbQ) binding cavity in the crystal structure.

Immunocompromised, athymic female Balb c/nu-nu mice (Charles River Laboratories) were subcutaneously injected with 2 × 10⁶ PANC-1 or 5 × 10⁶ PaTu 8902 cells per animal. The grafted PANC-1 cells formed slow-growing tumors after 2 week lag phase. When the PANC-1 tumors reached about 5 mm³, mice were divided into norMitoMet, metformin, and control groups receiving treatment (125 μmol/kg of norMitoMet or 1500 μmol/kg of metformin) or the vehicle by oral gavage 3 times per week (Mo/We/Fri). PaTu 8902 cells formed fast-growing tumors several days after tumor cell implantation. Treatment was started when tumors reached about 80 mm³. To overcome the possible low bioavailability of the agents after oral delivery, mice were treated intraperitoneally using 4.4 μmol/kg of norMitoMet (maximal tolerated dose), 1,500 μmol/kg of metformin or vehicle administrated daily. Tumor growth was monitored using the Vevo770 ultrasound imaging device equipped with the RMV708 or RMV704 probe (VisualSonics). All mice were cared for and maintained in accordance with the Animal Welfare Act of the Czech Republic.

Statistical analysis was performed using GraphPad Prism 6 software. Statistical significance was determined by one-way ANOVA followed by Tukey's or Dunnett's multiple comparison tests. The results from xenograft experiments were statistically evaluated by ANOVA followed by Sidak's multiple comparison test. Statistical significance is reported as follows: ns, P > 0.05; *, P ≤ 0.05; **, P ≤ 0.01, ***, P ≤ 0.001; and ****, P ≤ 0.0001. Data are expressed as mean ± SEM.

We prepared a panel of mitochondrially targeted analogs of metformin (Fig. 1). This includes MitoMet comprising a metformin core tagged with the TPP⁺ group via a 10-C spacer (compound 9) as well as norMitoMet, lacking a methyl group on the nitrogen adjacent to the 10-C spacer (compound 7). To see the importance of the length of the spacer, we prepared corresponding compounds featuring a 6-C spacer (compound 8 and 10, respectively). These agents were tested for their toxicity toward the human pancreatic cancer cell line PANC-1. Figure 2A and B documents a surprising finding that MitoMet was some three to four orders of magnitude more efficient than the parental metformin. For example, the IC₅₀ value for norMitoMet, which was about 20-fold more efficient than MitoMet, was 0.9 μmol/L, whereas it was 14 mmol/L for metformin. The corresponding 6-C analogs of MitoMet and norMitoMet were at least 100-fold less efficient than their 10-C counterparts. Figure 2C and D shows that 11C-TPP, that is, a compound with straight undecyl chain tagged with TPP⁺, was more efficient in suppressing viability of PANC-1 cells than was norMitoMet. However, 11C-TPP was similarly toxic toward a panel of nonmalignant cell lines as toward PANC-1 cells, whereas norMitoMet was much less toxic, indicating selectivity of the mitochondrially targeted agent. Toxicity of norMitoMet was not dependent on glucose level, which was found for metformin (Fig. 2E and F), consistent with the literature (38).

In the next set of experiments, we explored the effect of the most efficient of the newly synthesized agents norMitoMet, using a panel of pancreatic cancer cells. The agent suppressed viability in all of them with a considerable difference for the individual lines. Figure 2G and H and Supplementary Table S2 document a similar strong effect of the agent after 48-hour treatment on PANC-1 and MiaPaCa-2 cells, with the IC₅₀ values of 0.9 and 0.8 μmol/L, respectively. PaTu 8902 cells were slightly more resistant (IC₅₀ = 2.3 μmol/L), whereas BxPC-3 and AsPC-1 were most resistant, with IC₅₀ comparable to noncancer BJ fibroblasts (20.0 and 17.1 μmol/L, respectively, compared to 13.2 μmol/L for BJ cells).

Using the xCelligence apparatus, we assessed the effect of norMitoMet on proliferation of pancreatic cancer cells. Figure 3A shows plots for proliferation of PANC-1 cells in the presence of various concentrations of norMitoMet, and Fig. 3B shows the derived normalized slopes. These data indicate that the effect of norMitoMet on cell proliferation develops gradually in terms of days. After 48 to 72 hours, the cell growth ceased and the number of cells started to decline as documented by the negative values of the growth curve slope. The xCelligence assay was used for calculation of IC₅₀ values for inhibition of proliferation of individual pancreatic cancer cell lines and noncancer BJ fibroblasts. As presented in Fig. 3C and Supplementary Table S2, the level of suppression was time-dependent and the cell lines were susceptible similarly as shown for viability (cf. Fig. 2G and H and Supplementary Table S2) with PANC-1 cells slightly more susceptible in this case than MiaPaCa-2 cells. The decreased proliferation rate induced by norMitoMet was associated with cell-cycle arrest as documented by increased G₀ fraction and decreased S and G₂ fractions (Supplementary Fig. S1). To see whether toxicity of norMitoMet results also in activation of apoptosis, we tested PANC-1 cells for binding of annexin V and for cleavage of caspase-3. Figure 3D and E document that norMitoMet triggers apoptosis at levels as low as 1 μmol/L, whereas metformin was inactive even at 5 mmol/L and at 48 hours of treatment. In BJ fibroblasts, norMitoMet did not induce any apoptosis even when used in 10 times higher concentration than in PANC-1 cells (Supplementary Fig. S2).

Potential mechanisms of metformin's antitumorigenic effect include activation of AMPK signaling (11, 12). To compare the effect of norMitoMet and metformin on the AMPK pathway, we analyzed the phosphorylation status of AMPK and its downstream targets ACC, raptor, and mTOR. Figure 3F and Supplementary Fig. S3 document that norMitoMet activates the AMPK pathway when used in 3 to 4 order of magnitude lower concentration compared to metformin. As metformin-induced AMPK activation is believed to be related to reduced cellular energy charge resulting from the inhibition of respiration (11), we compared the effect of metformin and norMitoMet on the ATP content in PANC-1 cells. Intracellular ATP levels were decreased by the compounds used in the same concentration range that activates AMPK (Supplementary Fig. S4).

We next assessed the effect of norMitoMet on mitochondrial respiration. Mitochondrial complex I (CI)-dependent respiration was suppressed in PANC-1 cells by norMitoMet with IC₅₀ of 4.9 μmol/L, whereas it was some 3 orders of magnitude higher for metformin (Fig. 4A). Figure 4B documents that the IC₅₀ values for inhibition of respiration via CI are similar (some 2.4–8 μmol/L) for all pancreatic cancer cell lines tested. The sensitivity of the cells to norMitoMet-induced inhibition of CI respiration does not correlate with their susceptibility to the toxic effects of the agent (cf. Fig. 2H). The IC₅₀ values derived from dose responses to acutely applied norMitoMET are most probably underestimated, as this compound is characterized by time-dependent activity. In PANC-1, norMitoMET is able to fully inhibit CI-mediated respiration but also routine respiration of intact cells in a dose as low as 2 μmol/L and 24-hour incubation time (Supplementary Fig. S5). Figure 4C and D show that CII is a very weak target for norMitoMet, as the CII-dependent respiration was suppressed only at levels of the agent >100 μmol/L. Because it has been reported that G3PDH-dependent respiration is a major target for the effect of metformin in liver cells (39), we tested the effect of norMitoMet and metformin on G3PDH-dependent respiration in brown adipose tissue mitochondria, where a considerable portion of respiration is driven by G3P. Figure 4E documents that norMitoMet suppressed G3PDH respiration considerably at levels >100 μmol/L and metformin at levels >10 mmol/L. This finding for norMitoMet is comparable with its effect on CII-dependent respiration. Comparing the contribution of the three different types of respiration (CI-, CII-, and G3PDH-dependent) revealed that PANC-1 cells respire similarly via CI and CII, whereas G3PDH contributes only by ∼10% to total respiration (Fig. 4F).

To further document CI as a molecular target of MitoMet, we stably overexpressed the yeast NADH dehydrogenase NDI1 in PaTu 8902 cells and compared the effect of norMitoMet on respiratory rate in NDI1-expressing and control vector-transfected cells. NDI1 expression increased the routine respiratory rate and the respiration supported by NADH-linked substrates, whereas it resulted in slight decrease in succinate-stimulated oxygen consumption (Supplementary Fig. S6). In NDI1-expressing cells, norMitoMet was not able to fully inhibit the NADH-linked respiration contrary to control cells (Fig. 4G). Moreover, the sensitivity to antiproliferative effects of norMitoMet and rotenone were suppressed in NDI1-transfected cells, as documented by elevated IC₅₀ values for these compounds (Fig. 4H). Thus, NDI1 expression allows recovery of mitochondrial electron-transport activity in norMitoMet-treated cells and, at the same time, reduces the impact of norMitoMet treatment on cellular viability.

To localize the possible binding site for MitoMet in CI, we performed molecular modeling using the recently published crystal structure of Yarrowia lipolytica CI resolved at 3.6 Å (33). We have optimized the geometry of five cationic forms of MitoMet in order to identify the most probable protonation state. At the DFT-D level we have identified two nearly isoenergetic, most stable structures, whereas the remaining structures represent minima less stable by tens of kcal/mol. The two structures are axially chiral forms of one protonation state. The interconversion of these two forms is connected with a barrier of ∼7 kcal/mol. We have thus used both these forms for the subsequent docking study. As summarized in Fig. 4I, we have identified similar high affinity (sub-micromolar) binding mode for both MitoMet forms inside the UbQ-binding pocket. The poses share the same binding cavity as well as orientation of the metformin moiety with the predicted position of UbQ (33, 40). Supported by experimental observations, this suggests that MitoMET may affect UbQ interaction with CI, that is, its function within CI.

To learn more about the bioenergetics in the studied pancreatic cancer cell lines, we utilized the Seahorse instrument to assess their OCR and ECAR. Figure 5A presents the OCR and ECAR curves for PANC-1 and BxPC-3 cells, whereas Fig. 5B shows OCR and ECAR values for all five studied lines, documenting both the basal and maximal respiration, as well as glycolysis and glycolytic capacity. The results revealed an inverse correlation between respiration and glycolysis for the tested lines. We next calculated the ratios between maximal OCR and ECAR values, which is plotted in Fig. 5C. This shows that the highest ratio was found for PANC-1 cells that are highly susceptible to norMitoMet, whereas the least susceptible BxPC-3 and AsPC-1 cells showed the lowest OCR/ECAR ratio, suggesting that the level of toxicity of norMitoMet to pancreatic cancer cells is driven by the respiratory and glycolytic state of the cells. In other words, the higher respiratory and the lower glycolytic activity of the cells, the more susceptible they are to the mitochondrially targeted analog of metformin.

Because agents targeting mitochondrial complexes are expected to alter the mitochondrial function, we tested norMitoMet for its effect on ΔΨ_m,i and ROS generation. Figure 6A documents that 5 μmol/L norMitoMet caused strong dissipation of ΔΨ_m,i. Similarly it caused considerable generation of ROS as assessed using the probe DCF-DA (Fig. 6B). Because interference with mitochondrial complexes is assumed to cause generation of superoxide within mitochondria, we also used the MitoSOX probe. Figure 6C reveals that already at 1 μmol/L and at 6 hours, norMitoMet caused significant increase in mitochondrial superoxide. The increase in ROS production at least partially mediates the norMitoMet-induced apoptosis in PANC-1 cells, as documented by the decreased level of apoptosis in cells treated with the anti-oxidant NAC (Fig. 6D and E).

Finally, we tested the effect of norMitoMet on the growth of experimental pancreatic cancer. For this, xenografts were prepared by subcutaneous implantation of PANC-1 and PaTu 8902 cells in nude mice. Figure 6F and G reveals about 50% inhibition of tumor progression in norMitoMet-treated mice. In PANC-1-derived tumors, norMitoMet applied orally displayed similar tumor suppression effect as metformin used in 10-fold higher dose. In the very aggressive PaTu 8902-derived tumors, metformin dosed via intraperitoneal route was not able to suppress the tumor progression, whereas 2 orders of magnitude lower dose of norMitoMet significantly reduced the growth of tumors. NorMitoMet was found to inhibit with a similar extent also MiaPaCa-2-derived tumors (Supplementary Fig. S7A). We did not observe any effect of treatment on body weight of the animals or their behavioral pattern at any dosing (Supplementary Fig. S7B).

Despite a considerable progress in molecular medicine with focus on neoplastic diseases, pancreatic cancer is still on the rise. Apart from patients where resection is an option, there is no current cure for this pathology, with the prescribed therapeutic regimen only modestly increasing the 5-year survival of patients and with some 80% relapse of the disease (41, 42). About 90% of pancreatic cancer patients are positive for oncogenic K-Ras (41). Interestingly, a recent report showed that ablation of K-Ras in pancreatic cancer causes demise of majority of the malignant cells, with a small population surviving, featuring high level of mitochondrial respiration and characteristics of cancer stem-like cells, capable of tumor initiation (23). This indicates that tumor initiation/progression may be driven by or be dependent on mitochondrial function. This is consistent with growing body of reports that tumors are aberrant tissues with deregulated metabolism and that metabolic reprogramming is critical for tumor initiation, progression and metastasis, critically involving mitochondria (2–4, 23–25), and that metabolism can be a target for anticancer therapy (43).

An intriguing anticancer target, yet to be fully exploited, are mitochondria, in particular mitochondrial respiratory complexes (26, 27). Mitochondrial CII has been recently reported as a novel target for the anticancer agent α-tocopheryl succinate (α-TOS; refs. 44–46). CI has also been suggested as a target, for example in the context of breast cancer (47), and, interestingly, for metformin (19, 20), a drug of choice for T2DM, which has also been implied as a potential agent against pancreatic cancer (11, 12, 48). In pancreatic cancer, metformin is believed to act via regulation of tumor cell metabolism (49).

Because metformin is very inefficient against cancer cells, including pancreatic cancer cells, suppressing their proliferation/inducing apoptosis only at high concentrations (20, 50), we decided to adapt a novel approach that is based on sending the agent where it matters, that is, to the interface of the MIM and the matrix, the site of mitochondrial complexes, by means of tagging it with the delocalized cationic TPP⁺ group. Very unexpectedly, this increased the efficacy of the mitochondrially targeted agent compared to that of metformin by 3 to 4 orders of magnitude. Of the analogs we synthetized, norMitoMet, featuring a 10-C linker between the metformin and TPP⁺ moieties and lacking one methyl on the metformin structure, was the most efficient one.

Using high-resolution respirometry, we found that norMitoMet preferentially suppressed respiration via CI, with the IC₅₀ values some 100-fold lower than those for CII. It inhibited with similar efficacy as found for CII-dependent respiration also respiration dependent on G3PDH, a major target of metformin in liver cells (39). Because G3P-driven respiration contributed to overall mitochondrial respiration only marginally, we ruled it out as a major target for MitoMet. Expression of the yeast NADH dehydrogenase NDI1, which is known to be able to bypass CI and reverse the effects of metformin in cancer cells (20), was able to partially rescue the effect of MitoMet on respiration and cellular viability, further corroborating CI as a molecular target of the agent.

To explore the possible binding site for MitoMet in CI, we performed molecular modeling of its interaction with the respiratory complex using its recently published crystal structure (33). Our previous modeling of interaction of mitochondrially targeted vitamin E succinate (MitoVES; refs. 28, 29) documents association of the TPP moiety of the agent with the interface between the MIM and matrix, whereas the succinyl moiety interacts with the proximal UbQ-site of CII (51). On the contrary, the whole molecule of MitoMet resides inside CI, within its UbQ site. While shortening of the spacer of MitoVES from 11-C to 5-C results in losing the anticancer activity of the agent due to the notion that the free carboxyl group of the succinyl moiety cannot reach the proximal UbQ site, this is unlikely the reason for MitoMet. Rather, the fact that C6 MitoMet is less efficient than MitoMet with 10-C spacer can be explained likely due to higher hydrophobicity of the latter.

Importantly, we found that MitoMet suppressed pancreatic cancer in two mouse models by some 50% using 10- to 20-fold lower dose than found for metformin by us and as reported in the literature for the latter (20, 50). The anticancer efficacy of the agent did not show the 3 to 4 log gain in efficacy compared to metformin found for toxicity toward cultured pancreatic cancer cells. This may be due to the mode of administration of the drug and its resulting slower uptake following gavage. Indeed, norMitoMet administrated intraperitoneally suppressed the growth of very aggressive tumors derived from PaTu 8902 cell line, which were not affected by 300-fold higher doses of metformin. We are currently developing pro-drugs based on MitoMet that would result in better uptake of the drug and its delivery to the pancreatic cancer tissue.

Targeting of CI by MitoMet is linked to deregulation of the mitochondrial function, as proposed for class five mitocans acting by targeting the electron transport chain (26, 27). The mechanism by which suppression of CI-dependent respiration by MitoMet is relayed to cell-cycle arrest and induction of apoptosis in pancreatic cancer cells is currently unclear, although a parallel can be drawn with the proposed effect of metformin. This agent has been suggested to act by bioenergetics deregulation, such as modulating the AMPK/mTOR pathway, an important regulator of cell-cycle progression (11, 12, 52), which we show to be affected also by MitoMet. It cannot be excluded that there are additional mechanisms by which MitoMet acts. For example elevated ROS production induced by MitoMet may have beneficial effects, as cancer cells are more vulnerable to oxidative stress than normal cells (26). Indeed, we have demonstrated that MitoMet-induced apoptosis in pancreatic cancer cells is ROS-dependent, as it was suppressed by pretreatment with NAC. An innovative approach to understanding the effect of agents like MitoMet stems from the recent train of thought that respiration is a prerequisite for tumor initiation and progression, as well as for the metastatic disease (2–4, 23–25). That anticancer agents, including metformin, deregulate cancer cell metabolism has been proposed (49). A very attractive option is that suppression of respiration observed for MitoMet and pancreatic cancer cells is linked to generation of essential metabolites that are substrates for important biosynthetic pathways, such as the de novo pyrimidine synthesis as shown recently (53, 54). Related to the perception of the importance of respiration for tumor growth, we found that MitoMet was more toxic toward pancreatic cancer cell lines that were more dependent on respiration and less on glycolysis.

In conclusion, we have designed, synthesized and tested a novel anticancer agent based on the most frequently prescribed anti-T2DM drug, metformin. We report here on an unprecedented finding that mitochondrially targeted metformin (MitoMet) is more toxic toward pancreatic cancer cells by some 3 to 4 orders of magnitude compared to the parental compound. MitoMet is, therefore, a very promising anticancer drug against a pathology that is at present largely beyond treatment. The attractiveness of MitoMet stems from the fact that it is based on an approved and widely used drug, facilitating its potential translation into a drug of choice against pancreatic cancer.

No potential conflicts of interest were disclosed.

Conception and design: S. Boukalova, L. Werner, L. Dong, J. Neuzil

Development of methodology: S. Boukalova, L. Werner, L. Dong, Z. Drahota, J. Neuzil

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S. Boukalova, Z. Ezrova, J. Cerny, A. Bezawork-Geleta, A. Pecinova, L. Dong, Z. Drahota

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S. Boukalova, J. Stursa, L. Werner, J. Cerny, A. Bezawork-Geleta, J. Neuzil

Writing, review, and/or revision of the manuscript: S. Boukalova, J. Stursa, L. Werner, A. Bezawork-Geleta, L. Dong, J. Neuzil

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): L. Werner

Study supervision: S. Boukalova

Other (Design of the synthesis; synthesis of mitochondrially targeted metformin derivatives; purification, analysis and identification of prepared metformin derivatives.): J. Stursa

The NDI1-containing pWPI vectors and the empty counterparts were a generous gift from Professor Navdeep S. Chandel.

This work was supported in part by Australian Research Council Discovery grant, Czech Science Foundation grant (GA15-02203S), and Czech Ministry of Health grant (AZV 16-31604.A) to J. Neuzil. Further support was provided by BIOCEV CZ.1.05/1.1.00/02.0109 and Mitenal CZ.2.16/3.1.00/21531 from the ERDF, RVO: 86652036 and the Ministry of Education, Youth and Sports of the Czech Republic (LO1220) at the CZ-OPENSCREEN: National infrastructure for chemical biology.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.