Metformin reveals a mitochondrial copper addiction of mesenchymal cancer cells

The clinically approved drug metformin has been shown to selectively kill persister cancer cells through mechanisms that are not fully understood. To provide further mechanistic insights, we developed a drug surrogate that phenocopies metformin and can be labeled in situ by means of click chemistry. Firstly, we found this molecule to be more potent than metformin in several cancer cell models. Secondly, this technology enabled us to provide visual evidence of mitochondrial targeting with this class of drugs. A combination of fluorescence microscopy and cyclic voltammetry indicated that metformin targets mitochondrial copper, inducing the production of reactive oxygen species in this organelle, mitochondrial dysfunction and apoptosis. Importantly, this study revealed that mitochondrial copper is required for the maintenance of a mesenchymal state of human cancer cells, and that metformin can block the epithelial-to-mesenchymal transition, a biological process that normally accounts for the genesis of persister cancer cells, through direct copper targeting.


Introduction
Metformin is a clinically approved biguanide drug used in the management of type 2 diabetes [1]. The observation that treatments with metformin reduced risks of cancers in diabetic patients has prompted the search for mechanisms through which this molecule operates in cancer cells [2,3]. Metformin has been proposed to decrease glucose levels by activating AMP-activated protein kinase (AMPK) in hepatocytes resulting in a reduced activity of acetyl-CoA carboxylase and an induction of fatty acid oxidation [4]. Furthermore, it was shown that biguanides decrease the Krebs cycle [5] and inhibit complex I of the electron transport chain (ETC) as well as ATP synthase in cancer cells and in isolated mitochondria [6,7]. Furthermore, inhibition of complex I of the ETC by metformin was shown to reduce tumorigenesis [8]. Although it has been proposed that metformin acts on mitochondria and alters metabolism in cancer cells [9], there is of yet no direct visual evidence that biguanides physically accumulate in mitochondria. Importantly, metformin and the structural variant phenformin selectively target persister cancer cells [5] and synergize with conventional chemotherapy [10,11], through mechanisms that remain poorly understood. Cancer cells can undergo phenotypic alterations reminiscent of that observed during normal embryogenic development by going through epithelial-to-mesenchymal transition (EMT), to acquire physical properties allowing these cells to detach from primary sites and give rise to distal secondary tumors. Cells able to adopt such a metastable state have been shown to be refractory to conventional chemotherapies and these persister cancer cells have thus been associated to cancer relapse. In light of their activity in cancer, it is conceivable that biguanides directly alter EMT. Furthermore, biguanides interact with metals in vitro including copper [12,13], hinting towards a hypothesis whereby formation of metal complexes in cells could directly affect biological processes reliant on these metals, putative mechanistic targets of biguanides.
To investigate the subcellular localization of metformin in an unbiased manner, we designed a surrogate drug that could be labeled with a fluorophore in cells using click chemistry. A similar strategy was instrumental in identifying lysosomal iron as a mechanistic target of salinomycin in models of mesenchymal breast cancer cells, thereby illuminating a role of iron in the maintenance of these cells [14]. Using this approach, we identified mitochondria as a targeted organelle of metformin, providing unprecedented visual evidence of the accumulation of biguanides in this cellular compartment. Furthermore, we found that metformin directly impacts on mitochondrial copper levels in cancer cells, leading to mitochondrial stress and apoptosis. Importantly, we observed that mesenchymal cancer cells are addicted to copper and that metformin inhibits EMT in these cells. Taken together, our findings reveal mitochondrial copper as a druggable target in cancer cells that can be exploited to prevent the genesis of mesenchymal cancer cells.

General information of chemical synthesis
All starting materials were purchased from commercial sources and used without further purification, or purified according to Purification of Laboratory Chemicals (Armarego, W.L.F., Chai, C.L.L. 5 th edition). Reactions were monitored by thin-layer chromatography (TLC) using TLC silica gel coated aluminum plates 60F-254 (Merck). Metforminyn was purified using a Combiflash RF+ Teledyne Isco system. 1 H NMR spectra were recorded at 300 MHz and 13 C NMR at 75 MHz at 298 K. Chemical shifts were reported as δ values are expressed in ppm using the residual non-deuterated solvent as internal standard and coupling constants (J) in Hz. The following abbreviations are used: s, singlet; t, triplet; td, triplet doublet. UPLC trace and electrospray mass spectrum were obtained from a Waters Acquity SQD2 UPLC-MS system. HRMS was measured with a Bruker maXis LC-QTOF.

Cell viability
Cell viability was carried out by plating 1000 cells/well in 96-well plates using CellTiter-Blue viability assay according to the manufacturer's protocol. Cells were treated as indicated for 72 h. CellTiter-Blue reagent (G8081, Promega) was added after 72 h treatment and cells were incubated for 1-2 h before recording fluorescence intensities (ex. 560/20 nm; em. 590/10 nm) using a Perkin Elmer Wallac 1420 Victor2 Microplate Reader. Trypan blue exclusion and cell viability measurements were performed using an EVE Automated Cell Counter (NanoEnTek, #E1000) and EVE Cell Counting Slides. 10 μL of cell suspension were mixed with 10 μL of 0.4% Trypan solution and a final volume of 10 μL of this mixture used to count cells and assess viability according to the manufacturer's protocol.

Cell imaging
For immunofluorescence, cells were blocked with 2% BSA, 0.2% Tween-20/PBS (blocking buffer) for 20 min at RT. Cover-slips were incubated with 50 to 100 μL of diluted primary antibodies in blocking buffer 1 h at RT. Cover-slips were then washed three times with blocking buffer and incubated as described above with the appropriate secondary antibodies for 1 h. Cover-slips were washed three times with PBS and mounted using Vectashield Mounting Medium with DAPI (H-1200, VECTOR Laboratories). Fluorescence images were acquired using a Deltavision real-time microscope (Applied Precision). 60×/1.4NA and 100×/1.4NA objectives were used for 2D and 3D acquisitions that were deconvoluted with SoftWorx (Ratio conservative-15 iterations, Applied Precision) and processed with ImageJ.

Chemical labeling of Met in cells
Cells were cultured at~80% confluence and were treated with 0.3 mM Met for 3h. Cell were fixed with formaldehyde (2% in PBS, 12 min) prior to permeabilization (Triton X-100, 0.1% in PBS, 5 to 10 min) and washed three times with 1% BSA/PBS. The click reaction cocktail was prepared from Click-iT EdU Imaging kits (C10337, Life Technologies) according to the manufacturer's protocol. Briefly, mixing 430 μL of 1× Click-iT reaction buffer with 20 μL of CuSO 4 solution, 1.2 μL Alexa Fluor azide, 50 μL reaction buffer additive (sodium ascorbate) to reach a final volume of~500 μL. Cover-slips were incubated with the click reaction cocktail in the dark at RT for 30 min, then washed three times with PBS. Immunofluorescence was then performed as indicated.

Flow cytometry
Metal-specific probes were synthesized as previously described. Cells were treated as indicated in the figures. For mitochondrial copper and iron probes, cells were incubated with the relevant probe prior to being analyzed by flow cytometry. Cells were trypsinized (TrypLE Express Enzyme, Life Technologies, 12605010) and washed twice with ice cold PBS. For antibodies, cells were suspended in ice cold PBS containing 2% FBS and 1 mM EDTA (incubation buffer) and incubated for 20 min at 4˚C with the relevant antibody. Cells were then washed twice with ice cold PBS and suspended in incubation buffer prior to being analyzed by flow cytometry. For each condition, at least 100.000 cells were counted. Data were recorded on a BD Accuri C6 (BD Biosciences) and processed using Cell Quest (BD Biosciences) and FlowJo (FLOWJO, LLC) software.

Cellular levels of copper and iron
Cells were incubated with the relevant probes as described in Reagents section. Cells were then analyzed either by fluorescence microscopy or flow cytometry as described in Cell Imaging and Flow Cytometry sections.

Cyclic voltammetry
Cyclic voltammetry experiments were performed with a three-electrode cell under argon. Saturated Calomel Electrode (SCE) was used as reference, a steady Glassy Carbon (GC) electrode of diameter 3 mm was selected as working electrode and a Platinum wire as counter-electrode. All cyclic voltammograms were recorded at RT with a μ-autolab III from Metrohm using Nova software with a scan rate of 2 V/s. MeCN was used as a degassed HPLC grade from Carlo Erba.

ROS levels
Mitochondrial ROS was measured using MitoSOX Red Mitochondrial Superoxide Indicator (M36008, Thermo Fisher Scientific) according to the manufacturer's protocol. In brief, cells were treated for the indicated time with the red fluorescent dye, trypsinized, washed and resuspended in ice cold PBS containing 2% FBS and 1 mM EDTA prior to being analyzed by flow cytometry.

Mitochondrial membrane potential
Measurements were performed using the MitoProbe JC-1 assay kit (M34152, Thermo Fischer Scientific) according to the manufacturer's protocol. In brief, cells were treated with 2 μM JC-1 or CCCP (m-chlorophenylhydrazone) control for the indicated time. Cells were then trypsinized, washed and resuspended in ice cold PBS containing 2% FBS and 1 mM EDTA prior to being analyzed by flow cytometry.

Cell death
Cells were treated as indicated in the figures. After treatment, cell death was quantified using Annexin V-FITC (A)/Propidium Iodide (PI) assay according to the manufacturer's protocol (Annexin V-FITC apoptosis detection kit II, 556570, BD Pharmingen). Data were analyzed by a LSRFortessa flow cytometer (BD Biosciences, San Jose, CA) and processed using Cell Quest (BD Biosciences) and FlowJo (FLOWJO, LLC) software.

ICP-MS analyses
Glass vials equipped with Teflon septa were cleaned with nitric acid 65% (VWR, Suprapur, 1.00441.0250), washed with ultrapure water (Sigma Aldrich, 000000001012620500) and dried. Cells were treated as indicated and harvested using trypsinization (TrypLE Express Enzyme, Life Technologies, 12605010) followed by two washes with DPBS (Life Technologies, 14190169). Cells were then transferred to the cleaned vials and this material was lyophilized. Samples were subsequently digested using nitric acid 65% (VWR, Suprapur, 1.00441.0250) overnight followed by heating to 80˚C for 2 h. Samples were diluted with ultrapure water (Sigma Aldrich, 000000001012620500) to a total concentration of 0.475N nitric acid and transferred to metal-free centrifuge vials (VWR, 89049-172) for subsequent ICP-MS analyses. Cu concentrations were measured using an Agilent 7900 ICP-QMS in low-resolution mode. Sample introduction was achieved with a micro-nebulizer (Micro-Mist, 0.2ml/min) through a Scott spray chamber. Cu was measured using a collision-reaction interface with helium gas (5mL/min) to remove polyatomic interferences. Scandium and indium internal standards were injected after inline mixing with the samples to verify the absence of signal drift and matrix effects. A mix of certified standards was measured at concentrations spanning those of the samples to convert count measurements to concentrations in the solution. Uncertainties on sample concentrations are calculated using algebraic propagation of ICP-MS blank and sample counts uncertainties. Results were normalized by dry weight and cell number.

Quantification and statistical analyses
All results are presented as means with their standard deviation (SD), unless stated otherwise. Data are representative of at least three independent biological replicates, unless stated otherwise. Unpaired t-tests were used to calculate significance unless stated otherwise. P values are defined as: � p � 0.05, �� p � 0.01, ��� p � 0.001, ���� p � 0.0001.

Development of a potent clickable metformin surrogate
We set out to develop a surrogate of metformin that would allow for direct visualization of the subcellular sites of action of biguanide drugs. To this end, we synthesized an alkyne-containing analogue we named metforminyn (Met) (Figs 1 and S1, S2, S3 and S4). This analogue was designed after metformin and phenformin, where distal methyl/benzyl substituents have been replaced by a terminal alkyne suitable for labeling in cells by means of click chemistry [15][16][17][18][19]. We anticipated that introducing a functional group at this position would lead to a small molecule probe that retains its activity against cancer cells. A comparative evaluation of metformin and Met against two triple negative human breast cancer cell lines, and circulating tumor cells (CTC) from patients with colorectal cancer [20], showed that the synthetic analogue exhibited a higher potency compared to metformin (Fig 1).
We then set out to investigate the effect of metformin and Met on cancer cells undergoing EMT. This process of EMT can be artificially induced from normal epithelial mammary cells or triple negative breast cancer cells, leading to cells with a pronounced mesenchymal phenotype [21,22]. Importantly, we found that Met was more potent compared to metformin in these models, exhibiting a higher toxicity against MDA-MB-468 triple negative breast cancer cells transiently induced into a mesenchymal state using epidermal growth factor (EGF) [23] (Fig 1). A similar trend was observed for HMLER CD44 high /CD24 low mesenchymal cells compared to their HMLER CD44 low /CD24 high [21] epithelial counterpart (Fig 1).
To further investigate whether Met induced similar phenotypic alteration compared to metformin, we comparatively evaluated the effect of these molecules on levels of protein previously shown to be altered by metformin. Notably, we observed that metformin and Met increased levels of phosphorylated AMPKα-Thr172 and Glyceraldehyde 3-phosphate dehydrogenase (GADPH) whereas it decreased levels of GLUT1 (Glucose transporter 1) and the phosphorylated form of insulin-like growth factor 1 (IGF1 receptor-P) receptor (S5 Fig). Taken together, these data indicate that Met functionally phenocopied metformin in several models including mesenchymal cancer cells making it a suitable surrogate for subsequent cell imaging studies.

Metformin targets mitochondrial copper
Cancer cells treated with Met were subjected to labeling in cells by means of click chemistry [14] to identify putative mechanistic sites of action of biguanides (Fig 2). In breast cancer cells, labeled Met co-localized with cytochrome c, a cellular component that uniquely characterizes mitochondria, along with weaker nuclear staining and larger foci that resembled that observed for nucleolar proteins (Figs 2 and S6). In contrast, labeled Met could not be detected in the nucleus of CTC that responded to treatment (Fig 2), consistent with the idea that mitochondria are functional target organelles of biguanides. Remarkably, labeling of Met in cells, a chemical reaction that is normally catalyzed by Cu + , could be performed in situ without the need to incubate cells with additional copper catalyst (S6 Fig). Thus, the natural mitochondrial abundance of endogenous copper was sufficient to promote labeling of Met in native conditions. This provided further evidence that Met chemically interacted with copper in this organelle, in agreement with the previously reported capacity of metformin to form complexes with copper in vitro [12,13]. Moreover, in agreement with metformin directly targeting mitochondria, fluorescence of labeled Met was significantly reduced upon co-treatment with excess metformin competitor (S7 Fig). This finding encouraged us to investigate the effect of metformin and Met on the mitochondrial levels of Cu 2+ using a selective turn-off fluorescent probe [24]. Metformin and Met induced a reduction in fluorescence of this turn-off probe in MDA-MB-468 cells, as defined by microscopy and fluorescence activated cell sorting (FACS), indicating increased levels of Cu 2+ (Fig 2). Furthermore, we found that metformin and Met decreased mitochondrial Cu + levels using a selective turn-on fluorescent probe [25] (Fig 2). Importantly, metformin altered the redox properties of copper as defined by the occurrence of cyclic voltammetry peaks characteristic of quasi-reversible couples (Fig 2), and treatment with metformin or Met led to reduced protein level of the copper uptake protein 1 (Ctr1) in these cells (S8 Fig). These data indicated that increased levels of mitochondrial Cu 2+ occurred as a result of biguanides directly interacting with, and promoting oxidation of Cu + in this organelle, rather than by increasing cellular uptake of copper. For instance, Met was more effective at promoting oxidation of Cu + and preventing reduction of Cu + compared to metformin. This indicated that Met is more effective in stabilizing copper ions compared to metformin, which was in line with the higher potency of this analogue in cell based assays compared to the parent molecule. It is noteworthy that metformin and Met also increased levels of mitochondrial Fe 2+ as defined by flow cytometry using a selective turn-off probe (Fig 2) [26], and cyclic voltammetry showed that metformin and Met altered the redox properties of this metal, with metformin being more effective than Met (S9 Fig). These data supported a model whereby biguanides can inhibit complex I by hijacking iron from Fe-S clusters. Alternatively, Fe 3+ may also be consumed, acting as an electron scavenger of Cu + oxidation upon treatment with metformin. Consistent with metformin acting as a pro-oxidant of copper in mitochondria and altering the ionic balance of copper and iron, treatment of MDA-MB-468 cells with metformin or Met led to increased levels of reactive oxygen species (ROS) in this organelle accompanied by the reduction of mitochondrial membrane potentials and an alteration of mitochondrial morphology (S10 Fig). Furthermore, cell death induced by metformin or Met was rescued by the caspase 3 inhibitor Z-VAD-FMK but not the ferroptosis and necrosis inhibitors ferrostatin-1 and necrostatin-1 [27], respectively, which was in line with apoptotic cell death taking place as a result of mitochondrial dysfunction (S11 Fig) [28].

Copper is an essential component of the epithelial-to-mesenchymal transition
The pronounced effect of Met against mesenchymal cancer cells, accumulation of this biguanide in mitochondria and its effect on copper homeostasis in this organelle led us to investigate a putative dependency of mesenchymal cancer cells on mitochondrial copper. Western blot analysis revealed that EGF-induced EMT led to a significant increase of copper-containing proteins including superoxide dismutase 1 (SOD1) and cytochrome c oxidase subunit 4 (Cox4), along with increased levels of copper transporter 1 (Ctr1) (Fig 3). We also observed an increase of mitochondrial copper and total copper levels in EGF-treated samples using a combination of FACS and inductively coupled plasma mass spectrometry (ICP-MS), respectively (Fig 3). Consistent with previous findings showing that mitochondrial load is higher in persister cancer cells [29,30], these data suggested that this cell state requires higher needs for energy production, a biological process that heavily relies on metalloproteins and metal co-factors essential to the mitochondrial ETC [31]. Importantly, addition of copper synergized with EGF treatments to shift MDA-MB-468 cancer cells toward a mesenchymal phenotype as defined by increased levels of mesenchymal markers (e.g. Vimentin, Fibronectin) and the proportion of cells harboring a CD44 high /CD24 low cell surface marker pattern (Figs 3 and S12). Importantly, we also observed that copper synergized with TGF-β treatment to shift HMLER CD44 low /CD24 high cells and DU-145 prostate cancer cells toward a mesenchymal phenotype as defined by increased ratios of cells with a CD44 high /CD24 low or CD44 high /CD133 high cell surface marker pattern, respectively, demonstrating a general role of copper in the regulation of EMT in different systems (S12 Fig). Finally, induction of EMT in MDA-MB-468 cells using EGF was prevented by metformin and Met as defined by the subcellular localization of E-cadherin and phalloidin staining, levels of mesenchymal markers fibronectin, vimentin and EMT-transcription factors (TF) Zeb1, Snail and Slug as well as the proportion of cells harboring a CD44 high /CD24 low cell surface marker pattern (Figs 3 and S13). Similarly, this effect of metformin was also observed in MCF-7 breast cancer and in DU-145 prostate cancer cells (S13 Fig). Alternatively, this data may also reflect a preferential cell killing of Met when cells are in a mesenchymal state. Taken together, these data reveal a dependency of mesenchymal cancer cells on mitochondrial copper, providing a rationale for the heightened sensitivity of these cells to biguanides.

Conclusion
To provide insights into mechanisms underlying the effect of metformin in cancer cells undergoing EMT or cancer cells exhibiting a mesenchymal phenotype, we have developed a surrogate probe of this drug that can be chemically labeled in cells by means of click chemistry. This probe is both more potent than metformin in cancer models and enables the visual detection of sites of action of biguanides in fixed cells. We provide visual evidence that biguanides preferentially target mitochondria in several cancer cell lines including cancer cells isolated from patients. Furthermore, the fact that azide-alkyne labeling of Met can be performed in cells in absence of added copper catalyst and that metformin promotes oxidation of Cu + in vitro validate mitochondrial copper as a putative mechanistic target of metformin. In line with this, the biologically more potent analogue Met alters copper redox potentials to a greater extent than metformin, whereas the latter preferentially affects that of iron in vitro. Importantly, we show that metformin impacts levels of mitochondrial Cu 2+ and Fe 2+ , providing a mechanistic rationale underlying inhibition of the ETC and ATP-synthase and suggesting that effects on AMPK occurs downstream of mitochondrial targeting. This investigation shows that cancer cells in a mesenchymal state are more sensitive to biguanides compared to their epithelial state and that this effect relies on mitochondrial copper targeting. Increased levels of mitochondrial coppercontaining and copper uptake proteins during EMT, together with the synergistic effect of copper and growth factors of cytokines to induce EMT, support a prevalent role of mitochondrial copper in the maintenance of a mesenchymal state of cancer cells. We show a central role of copper in the regulation of the EMT program, a process associated to the genesis of persister cancer cells, metastasis and cancer relapse. In this context, inhibition of EMT by biguanides strongly argues in favor of a cross-talk between mitochondria and the nucleus, potentially involving mitochondrial metabolites that could regulate EMT plasticity. Moreover, it is conceivable that in models of diabetes [32], metformin targets mitochondrial copper to promote mitochondrial dysfunction, forcing cells to switch to a lower yielding anaerobic mode. These findings strengthen the notion that cancer cells exhibiting a mesenchymal phenotype are susceptible to mitochondrial copper targeting, which paves the way toward the establishment of novel therapeutic strategies [33][34][35][36].