Targeting Mitochondrial Iron Metabolism Suppresses Tumor Growth and Metastasis by Inducing Mitochondrial Dysfunction and Mitophagy

Iron represents a vital element required for almost all forms of life (1). Because of its ability to transfer electrons by shuttling between ferrous and ferric forms, it is a crucial component participating in the electron transport chain within mitochondria, in DNA repair and replication (2), chromatin remodeling and epigenetic changes (3) and cellular metabolism (4).

In biological systems, iron can be incorporated into iron-sulfur [Fe-S] clusters and the heme molecule (2, 5). Both of these iron-containing structures serve as cofactors of many enzymes and are partially synthesized inside mitochondria, making it the central organelle in the cellular iron metabolism. At the same time, mitochondria are important for ATP production, metabolic reactions, calcium homeostasis, and programmed cell death (2, 4). Dysfunctional mitochondria are degraded by a specific process termed mitophagy, which is induced by iron deficiency (6, 7).

Iron is crucial for the rapidly proliferating cancer cells (8) and also for the cancer stem-like cells (9). Therefore, attempts to target iron metabolism to suppress tumor growth have been made (10). In certain cancer types, such as bladder cancer, gallium nitrate has been successfully applied, competitively replacing iron (11). Other approaches including antibodies against transferrin receptor, application of iron chelators, or a combination of both approaches have shown promising results in vitro (10, 12–14). Yet, the main remaining drawbacks of such strategies are the nonselective nature of the treatment and the associated perturbation of the systemic iron metabolism (14).

In recent years, a strategy for targeting small molecules into mitochondria via a triphenylphosphonium group (TPP⁺) proposed by Murphy and colleagues (15) has been applied for several compounds. Moreover, preferential accumulation of the drugs in cancer cells has been described, due to their higher inner mitochondrial membrane potential (16–19).

In this study, we have modified the classical iron chelator deferoxamine (DFO) by tagging it with the TPP⁺ group, thus repurposing an iron chelator used for treatment of iron overload into an anticancer agent. We demonstrate that such modification results in a preferential targeting of cancer cells, halting their proliferation and inducing cell death via interfering with their iron metabolism and destabilizing their mitochondria. This represents a novel approach towards targeting of cancer cells, which is of clinical relevance.

Cell lines were obtained from ATCC and used within 3 months from their thawing. Cells were tested for Mycoplasma (MycoAlert PLUS detection Kit; Lonza) and authenticated by short tandem repeat analysis (Generi Biotech).

mitoDFO—triphenyl (3,14,25-trihydroxy-2,10,13,21,24-pentaoxo-31-(10-(triphenyl phosphonio)decyl)-3,9,14,20,25,31-hexaazahentetracontan-41-yl) phosphonium chloride was prepared according to Scheme 1 described together with nuclear magnetic resonance results in Supplementary Data. mitoDFO-fluorescein isothiocyanate was prepared according to the protocol provided in Supplementary data. Both mitoDFO and mitoDFO-fluorescein isothiocyanate (mitoDFO-FITC) were dissolved in DMSO (10 mmol/L stock) and added to cells (max 0.4% DMSO). Under control conditions, the same volume of DMSO was added.

SDS-PAGE and Western blot analysis were performed as reported previously (9). Cells were lysed in RIPA buffer and protein content assessed by bicinchoninic acid method (Thermo Fisher Scientific). The list of primary antibodies is available as Supplementary Table S3.

The activity of aconitase was assessed by the Aconitase Activity Assay Kit (Sigma-Aldrich) according to the manufacturer's instructions. Aconitase activity is expressed as percentage of control cells treated with the solvent (DMSO). Control (lysate) and mitoDFO (lysate) columns correspond to the activity observed in the cell lysate treated with either DMSO or mitoDFO at 5 μmol/L. In addition, in-gel aconitase activity was assessed according to a published protocol (20).

Cellular heme content was evaluated as described previously (21), by boiling whole-cell lysate with 2 mol/L oxalic acid. Results are expressed as the difference between total and free protoporphyrin, relative to control.

Detailed description of autoradiography is provided in Supplementary Materials and Methods.

The oxygen consumption rate (OCR) and extracellular acidification rate (ECAR) were measured by the Seahorse XFe96 Analyzer (Agilent Technologies) according to the manufacturer's instructions (details in Supplementary Data).

Mitochondrial respiration in permeabilized cells was assessed using the high-resolution Oxygraph-2k respirometer (Oroboros Instruments), according to the standard procedure (22).

Mitochondria were isolated by differential centrifugation, and solubilized with digitonin, as described previously (23).

Blue native electrophoresis (BNE) and high resolution clear native electrophoresis (hrCNE) were used to assess the level and activity of mitochondrial respiratory complexes, as described in ref. 23. All antibodies used are listed in the Supplementary Table S3.

Total cellular ATP was assessed by the Cell Titer-Glo Luminescent Cell Viability Assay Kit (Promega) according to the manufacturer's instructions. Results are presented as percentage of control.

Mitochondrial superoxide levels were determined using the fluorescent probe MitoSOX Red (Thermo Fisher Scientific) and total cellular reactive oxygen species (ROS) by 2′,7′-dichlorodihydrofluorescein-diacetate (DCF-DA; Sigma-Aldrich). Mitochondrial membrane potential was assessed using tetramethylrhodamine methyl ester (TMRM; Sigma-Aldrich). Results are expressed as percentage of relative fluorescence units relative to the control.

Confocal microscopy images were acquired using a 63× water immersion objective in a Leica SP8 confocal microscope (Leica Microsystems; details in Supplementary Data).

Acidification of mitochondria was detected using the fluorescent protein mitoKEIMA as described previously (24). Results are expressed as the ratio between fluorescence intensity at 440 and 586 nm, relative to control.

The number of viable cells showing combined cytostatic and cytotoxic effect was assessed by crystal violet staining. IC₅₀ values for each condition were calculated by plotting percentage of viable cells versus compound concentration. After transforming x values to logarithmic scale, values were fitted to dose–response inhibition curves with normalized y values in the GraphPad Prism 5.0 software, and IC₅₀ values were interpolated and expressed as mean ± SEM.

The xCELLigence real-time cell analysis (RTCA) SP station (ACEA Biosciences) was used to determine the effect of mitoDFO and deferoxamine on cell growth. Impedance was monitored for 48 hours in 15-minute intervals. Results were analyzed using the RTCA software and are shown as cell index over time.

Similarly, the effect of the compounds on cellular viability was analyzed by the JuLI FL Fluorescence Live Cell Analyzer (NanoEnTek) for 48 hours. Results were analyzed with the JuLI FL PC software and are shown as percentage of confluence over time.

Cell death was assessed using the annexin V/propidium iodide (AV/PI) assay, as described previously (25). Results are expressed as percentage of dead cells (sum of AV⁺/PI⁻, AV⁻/PI⁺ and AV⁺/PI⁺ cells).

Cell cycle measurement was performed using Vybrant DyeCycle Violet Stain (Thermo Fisher Scientific), according to the manufacturer's instructions.

MDA-MB-231 and 4T1 cells were grown as spheroids for 2 days using three-dimensional (3D) Petri Dish (Microtissues; #12-81 large spheroids) according to the manufacturer's instructions. The invasion assay was performed as described previously (26).

Balb/c mice were injected with 1 × 10⁶ 4T1 cells, and NSG mice with either 1 × 10⁶ MDA-MB-231 cells or 3 × 10⁶ Bx-PC3 cells. When tumors reached the volume of 30–50 mm³ (quantified by ultrasound imaging, USI), each group was divided into two subgroups and treated intraperitoneally with either mitoDFO (1, 8, or 12 mg/kg), deferoxamine (8 or 80 mg/kg), or vehicle (2.5% DMSO in corn oil, 100 μL per dose) twice per week. Tumor volume was monitored by the USI instrument Vevo770 (VisualSonics). Results are presented as relative tumor size. Animal ethics was approved by the Czech Academy of Sciences and animal experiments were performed according to the Czech Republic Council guidelines for the Care and Use of Animals in Research and Teaching.

Lung metastases were evaluated as described previously (27), by selection of 4T1 cells with 60 μmol/L 6-thioguanine (details in Supplementary Materials and Methods).

Tissue iron levels were determined by the method of Torrance and Bothwell (28) in mg per μg of wet tissue. Iron content was calculated on the basis of a standard curve of ferric ammonium citrate.

Serum iron levels were assessed by the Liquid Fe 200 Kit (Erba Lachema) according to the manufacturer's instructions. Samples for the analysis were acquired by cardiac puncture, blood was clotted for 30 minutes and spun, and the resulting serum was analyzed. Hemolytic samples were excluded from the analysis.

Blood was collected into heparinized tubes and hematologic parameters measured at the Czech Centre for Phenogenomics, using the Mindray Bc-5300 Vet.

Tumors were fixed in 4% paraformaldehyde and analyzed by the Czech Centre for Phenogenomics principally as shown before (29). Images were acquired using the AxioScan.Z1 automated slide scanner (Carl Zeiss) and analyzed by the ImageJ software.

All results are expressed as mean ± SEM of at least three independent experiments. The comparison between experimental groups and control was performed by one- or two-way ANOVA followed by Sidak post test, or by Student t test using the GraphPad Prism 5.0 software. Differences were considered significant at P < 0.05.

Deferoxamine, a cell-impermeable iron chelator, has been shown to inhibit proliferation of cancer cells both in vitro and in vivo (13, 14, 30). Aiming to direct it into mitochondria, we tagged it with two TPP⁺ moieties via 10-carbon linker chains, referring to it as mitochondrially targeted deferoxamine (mitoDFO; for structure see Fig. 1A).

We first tested whether mitoDFO affects cellular iron metabolism. Our results show that mitoDFO induces a dysfunction in mitochondrial [Fe-S] cluster assembly, as the levels of [Fe-S] cluster-containing mitochondrial proteins NDUFA9, SDHB, UQCRFS1, and mtCO1 are markedly reduced upon mitoDFO treatment of MCF7 and MDA-MB-231 breast cancer cells (Fig. 1B). Furthermore, the levels of lipoylated proteins, which are dependent on the [Fe-S] cluster-containing lipoyl synthase (31), were diminished as a response to mitoDFO treatment (Fig. 1B). Interestingly, deferoxamine treatment (100 μmol/L) showed a much lower effect in both cell lines (Fig. 1B). We then studied the levels of enzymes responsible for [Fe-S] cluster synthesis. Our results show that while the level of the mitochondrially localized enzyme ISCU was greatly diminished by mitoDFO, the cytosolic enzymes were either not affected (CIAO1, MMS19) or even slightly increased (NUBP1 in MCF7 cells; Fig. 1C).

To identify the mechanism behind the decrease in the protein levels, we first analyzed the mRNA expression of iron-related genes. Our results show that mitoDFO and deferoxamine cause a small, yet significant reduction in the expression of genes related to [Fe-S] cluster biogenesis and utilization, as well as heme biosynthesis and degradation. On the other hand, the level of mRNA of genes related to iron uptake and storage was significantly upregulated (Supplementary Fig. S1A). It is of note that transcripts such as ACO1, ACO2, COX1, and TFRC clearly showed differential response to deferoxamine and mitoDFO (Supplementary Fig. S1A), documenting that the effect of the two chelators is not identical. Nevertheless, changes in gene expression were rather mild (2-fold), while changes in the corresponding protein levels were more profound especially for the mitochondrial proteins NDUFA9, SDHB, UQCRFS1, mtCO1, ISCU, and FECH, and these changes were not replicated with 100 μmol/L deferoxamine (Fig. 1B, C, E, and F). We thus propose that the main effect of mitoDFO occurs at the level of protein degradation, and we show that cotreatment with mitoDFO and the proteasome inhibitor MG-132 (5 μmol/L; 16 hours) prevented the decrease in protein levels of the proteasome-degraded ISCU (32), SDHB (33), FECH (34) and mtCO1 (35). On the other hand, the effect of mitoDFO on the level of NDUFA9 and UQCRFS1 was not affected by proteasome inhibition (Supplementary Fig. S1B), as both proteins are degraded by mitochondrial proteases (36, 37).

Another enzyme that depends on the correct assembly of [Fe-S] clusters is aconitase. In agreement with our previous results, mitoDFO significantly reduced the activity of both forms of aconitase (cytosolic and mitochondrial; Fig. 1D; Supplementary Fig. S1C). Interestingly, while deferoxamine had a similar effect on the activity of cytosolic ACO1, it had a much lower effect on the activity of mitochondrial ACO2 (Fig. 1D). Our results also show that mitoDFO not only affected aconitase activity when added to live cells, but decreased it even when added to whole-cell lysates, suggesting a direct effect of the compound on the aconitase proteins (Supplementary Fig. S1D). Aconitase 1 that lacks its [Fe-S] cluster serves as iron regulatory protein 1 (IRP1) and has affinity for the iron-responsive elements (IRE) on the mRNA of several iron regulators (38). Our results suggest that mitoDFO promotes increased binding of IRP1 to IREs and activates the IRP/IRE system, as the level of transferrin receptor 1 is increased, while the level of ferritin is decreased although the extent of the activation is lower compared with deferoxamine (Fig. 1E), being in line with only a slight change in the protein level of IRE-containing ferroportin and ACO2 (Fig. 1D and E). Similarly, the level of the cytosolic iron sensor FBXL5 was not changed or only slightly increased in response to mitoDFO, suggesting that the free labile iron pool in the cytosol is not diminished (Fig. 1E).

To further define the effect of mitoDFO on iron-containing proteins, we incubated cells loaded with ⁵⁵Fe with mitoDFO and deferoxamine, and visualized mitochondrial and cytosolic iron-containing proteins by autoradiography. Our data show that 24-hour incubation with 5 μmol/L mitoDFO led to a dramatic decrease in bands corresponding to respiratory supercomplexes, a general decrease of the ⁵⁵Fe signal in mitochondrial fraction, and a decrease in cytosolic ferritin (Fig. 1F). Deferoxamine treatment (50 μmol/L) did not affect mitochondrial proteins but led to a marked decrease in cytosolic ferritin (Fig. 1F). Long-term incubation with 1 μmol/L mitoDFO (60 hours) led to a decrease (MCF7) or absence (MDA-MB-231) of bands corresponding to iron-containing native proteins (Fig. 1F; Supplementary Fig. S1E) with the exception of ferritin, which was considerably increased (Supplementary Fig. S1E).

The second iron-containing cofactor is heme. Our results show that mitoDFO significantly reduced the level of mitochondrial FECH responsible for the last step of heme synthesis, while not affecting the levels of HMOX2 responsible for heme degradation. In line with this, phosphorylation of EIF2α induced by the heme-regulated inhibitor kinase (HRI) upon heme deficiency (39) was upregulated by mitoDFO (Fig. 1G). Although similar changes were observed with deferoxamine, they required much higher concentration (100 μmol/L; Fig. 1G). Furthermore, the total heme level (Fig. 1H) was lower upon treatment with mitoDFO, which is likely linked to lower levels of [Fe-S] cluster-containing ferrochelatase, resulting in lower heme biosynthesis, and diminished iron bioavailability. Once again, deferoxamine required a 10-fold higher concentration to reach a similar effect.

One of the main processes within mitochondria involves the transfer of electrons along the electron transport chain (ETC). Because the function of the ETC highly relies on [Fe-S] clusters and heme, we assessed the effect of mitoDFO on ETC-dependent oxygen consumption. Our results show that mitoDFO (10 μmol/L, 1 hour) completely suppressed mitochondrial respiration (Fig. 2A; Supplementary Fig. S2A) and enhanced basal ECAR (a surrogate marker of the glycolytic state; Fig. 2B; Supplementary Fig. S2B) in all cell lines studied.

To confirm this finding, we assessed the levels of ATP after the incubation of cells with mitoDFO in the absence or presence of the glycolysis inhibitor 2-deoxyglucose (2-DG). Surprisingly, mitoDFO (10 μmol/L, 1 hour) alone did not have any effect on the ATP content (Fig. 2C). However, the combination of mitoDFO and 2-DG dropped ATP levels significantly more than 2-DG alone. Such effect was not observed when cells were coincubated with mitoDFO and the complex V inhibitor oligomycin (2 μmol/L; Fig. 2C). Of note, mitoDFO treatment also induced an increase in protein level of some key glycolytic enzymes (HKII, LDHA) while simultaneously decreasing the citric acid cycle enzyme PDH-E1 (Supplementary Fig. S2C).

Efficient ETC-linked respiration mainly relies on the assembly of respiratory complexes (RC) and supercomplexes (SC). The most known SC, referred to as respirasome, is composed of complex I (CI), CIII, and CIV (40). Given the profound effect of mitoDFO on the level of individual [Fe-S] cluster-containing subunits of the ETC and on mitochondrial respiration, we analyzed the level and organization of RCs and SCs, using blue native electrophoresis. A strong decrease in the level of the respirasome was observed in all cell lines after their exposure to 5 μmol/L mitoDFO for 16 hours (Fig. 2D; Supplementary Fig. S2D) while deferoxamine showed no effect (Supplementary Fig. S2E). This change was not accompanied by a decrease in lower forms of SCs, such as the SC composed of only CIII and CIV. On the other hand, it appears that an intermediate SC of CIII, migrating above CIII₂/CIV, is enhanced (Fig. 2D). Regarding individual RCs, we observed a minor decrease in the levels of CIII and CIV and no change in CV. CII, however, was significantly decreased in response to mitoDFO treatment. We also performed clear native electrophoresis to assess in-gel activity of complex I in mitoDFO-treated cells. Our results show that the activity of CI was lower in the respirasome as well as in other SCs comprising CI and in CI itself (Supplementary Fig. S3A).

Therefore, we studied the effect of short-term mitoDFO incubation on respiration via individual complexes using the high-resolution respirometer Oxygraph-2k. Supplementary Fig. S3B shows that 1-hour incubation of MCF7 cells with 10 μmol/L mitoDFO lowered CI-dependent respiration, while it did not significantly change CII-dependent respiration.

Given the differential effect of mitoDFO on mitochondria compared with deferoxamine, we next sought to demonstrate that our compound accumulates within mitochondria. To do so, we attached a FITC moiety to mitoDFO, and our results show that in MCF7 and MDA-MB-231 cells mitoDFO-FITC signal highly colocalizes with that of MitoTracker Deep Red (Fig. 3A).

Targeting TPP⁺ compounds to the inner mitochondrial membrane (IMM) is driven by their positive charge and the negative potential at the matrix face of the IMM (15). However, the presence of such compounds might lead to the dissipation of mitochondrial membrane potential (MMP), mitochondrial depolarization, and dysfunction (16, 17, 19). Our data show that addition of mitoDFO to cells led to a rapid and significant decrease of tetramethylrhodamine (TMRM) fluorescence, indicating a loss of MMP (Fig. 3B).

Mitochondria are known to be the main intracellular source of ROS (41), and mitochondrial depolarization and dysfunction highly increase their production. Therefore, we assessed the ability of mitoDFO to generate oxidative stress and found that it induced a rapid and substantial increase in mitochondrial superoxide and cellular ROS levels (Fig. 3C and D). Furthermore, coincubation of cells with mitoDFO and the antioxidant N-acetyl cysteine (NAC, 10 mmol/L) significantly suppressed cell death induced by mitoDFO (Fig. 3E) and the levels of intrinsic antioxidant defense enzymes CAT and SOD2 were slightly induced by a 16 hours treatment with mitoDFO (Supplementary Fig. S3C).

Mitochondria are highly dynamic organelles able to form a vast network. Such network is controlled by fusion, rendering highly efficient, nuclei-associated elongated mitochondria (42), and fission, resulting in small, fragmented, radially located mitochondria. The increased fission is a hallmark of mitochondrial damage and increased oxidative stress (41). Therefore, we assessed the status of the mitochondrial network by confocal microscopy and analyzed the intracellular distribution of mitochondria. Figure 4A shows that mitoDFO treatment (5 μmol/L, 24 hours) induces a slight, yet significant increase of radially localized mitochondria at the expense of their perinuclear localization in MCF7 cells. Furthermore, significant decrease in average mitochondrial volume and increase in the number of individual mitochondria in each cell were observed upon treatment with mitoDFO (Fig. 4B), consistent with more prominent mitochondrial fission. Of note, such increase in the number of individual mitochondria was not caused by increase in cellular size, as cell area was not affected by mitoDFO (Fig. 4B).

When fragmented mitochondria are not able to recover from an insult, the process of mitophagy is activated and damaged mitochondria are engulfed by lysosomes, where they are degraded (6, 43, 44). By using a pH-sensitive mitochondrial probe, we were able to study the acidification that occurs in mitochondria upon their engulfment. Our results show that incubation of cells with mitoDFO (5 μmol/L, 24 hours) led to increased mitochondrial acidification (Fig. 4C and D). Another indicator of mitochondrial acidification is the reporter protein mitoKEIMA, with excitation fluorescence dependent on the pH value (24). Consistent with the mitophagy-inducing effect of mitoDFO, mitoKEIMA-expressing cancer cells showed a lower ratio of neutral-to-acid-related fluorescence following exposure to the agent (Fig. 4E). Furthermore, we observed increase in PINK1 and BNIP3 levels, decrease in p62 and processing of LC3B upon addition of mitoDFO to cancer cells (Fig. 4F), pointing to mitophagy activation (44).

Mitochondrial dysfunction, increased oxidative stress and enhanced mitophagy can halt the cell cycle and eventually cause cell death. Therefore, we evaluated the effect of mitoDFO on cellular proliferation and viability. Our results show that mitoDFO is at least 40-fold more potent than the parental deferoxamine in decreasing the number of viable cancer cells of breast, ovarian, and pancreatic origin (Fig. 5A; Supplementary Table S1). To further understand the inhibitory effect of mitoDFO, we performed real-time monitoring assays with low concentrations of mitoDFO (1–5 μmol/L). The results show that at concentrations ≤ 2 μmol/L, the effect of mitoDFO is primarily cytostatic, with cell death appearing at concentrations of 5 μmol/L (Fig. 5B; Supplementary Fig. S3D and S3E). Once again, real-time monitoring confirmed that mitoDFO is more potent than deferoxamine in inhibiting cellular proliferation and inducing cell death, as deferoxamine increased proliferation of cells at low micromolar concentrations and induced cell death only at a concentration of 100 μmol/L (Fig. 5B, top right).

Furthermore, there is a difference in the cytostatic effect of mitoDFO between most malignant and normal cells, as cancer cells in general respond by considerably lower proliferation while nonmalignant BJ fibroblasts continue to grow under the same conditions (Fig. 5B; Supplementary Fig. S3D). In addition, nonmalignant cell lines BJ, MRC5 and HFP1 show higher IC₅₀ values for mitoDFO compared with malignant cell lines tested, with several exceptions (Supplementary Table S1), and we observed accumulation of cells in the G₁-phase in MCF7 and MDA-MB-231 cells, but not in normal BJ fibroblasts, upon exposure to mitoDFO for 24 hours (Fig. 5C). However, such effect was not observed in OVCAR-3 or Bx-PC3 cells incubated with mitoDFO (Supplementary Fig. S3F), probably because at 24 hours mitoDFO showed very little cytostatic effect in OVCAR-3 cells, and already a strong cytotoxic effect in Bx-PC3 cells (Supplementary Fig. S3E).

Importantly, all tested cancer cell lines (even those with lower cytostatic effect of mitoDFO) showed enhanced cytotoxic effect of the compound observed by annexin V/PI staining at concentrations ranging from 5 to 40 μmol/L compared with their nonmalignant counterparts that exhibited pronounced cell death only at 40 μmol/L (Fig. 5D and E). Interestingly, deferoxamine did not show any effect on cell-cycle distribution when added at 5 μmol/L, and was only effective in stopping proliferation or inducing cytotoxicity at concentrations of 100–200 μmol/L, where it also affected nonmalignant BJ fibroblasts (Fig. 5B, C, and E; Supplementary Fig. S3D).

Finally, to determine whether the iron-chelating ability of mitoDFO was important for the cytotoxic effect, we exposed cells to mitoDFO preincubated with equimolar concentrations of iron citrate. Our data show that the complex of mitoDFO with iron is much less cytotoxic to malignant cells than the agent alone, confirming the importance of the iron-chelating properties of mitoDFO for cell death induction (Fig. 5F).

Our in vitro data provided a solid evidence that mitoDFO is effective in suppressing cellular proliferation and inducing death of cancer cells. Therefore, we next tested its efficacy in vivo using both syngeneic and xenograft mice models. In the syngeneic model, we found that mitoDFO (8 mg/kg, i.p.) considerably suppressed the growth of tumors, in some cases causing even a decrease in tumor size (Fig. 6A). Furthermore, Ki67 staining was markedly reduced in mitoDFO-treated tumors (Supplementary Figs. S4A and S4B). In contrast, deferoxamine failed to affect tumor progression, even when administered at a 10-fold higher dose (8 and 80 mg/kg, i.p; Fig. 6A).

A similar growth-slowing effect of mitoDFO was observed in the xenograft models of breast and pancreatic cancer. In the first case, while a dose of 1 mg/kg had no effect, a dose of 8 mg/kg mitoDFO (i.p.) significantly decreased the growth rate of MDA-MB-231 tumors, although with less potency than in the syngeneic model (Fig. 6B). Similarly, the Bx-PC3 xenografts showed that mitoDFO (12 mg/kg i.p.) significantly inhibited the growth of tumors (Fig. 6C).

To assess whether the induction of mitophagy observed in cultured cells also occurs in vivo, we analyzed the expression of several mitophagy markers in 4T1-derived tumors. Our results show that mitoDFO treatment increased expression of PINK and BNIP3 while reducing the levels of p62 (Fig. 6D). Similarly, we observed an increase in transferrin receptor 1 and a decrease in FECH and SDHB, consistent with in vitro results. These changes were not detected in the liver tissue of the same animals (Supplementary Fig. S4C). Furthermore, mice weight was not affected by mitoDFO treatment (Supplementary Figs. S4D and S4E).

To define whether mitoDFO preferentially targets cancer cells in vivo and does not significantly affect the systemic iron metabolism, we assessed the level of iron in the tumor, serum, spleen and liver in treated and control tumor-bearing Balb/c and NSG mice. None of these parameters was significantly different between the control and treated animals (Fig. 6E–H). Similarly, the number of red blood cells, their hemoglobin content, hematocrit and mean corpuscular volume in Balb/c mice did not differ between the two groups (Fig. 6I and J; Supplementary Fig. S4F and S4G). In addition, white blood cells (WBC) were assessed and, although mitoDFO-treated animals showed a general increase in WBC number (Supplementary Fig. S4H), the percentage of each individual type of cell was similar in both groups (Supplementary Fig. S4I).

We have shown that mitoDFO effectively inhibits growth of cancer cells, highlighting its potential for cancer treatment. However, the most life-threatening aspect of cancer is development of metastases, depending on the ability of cancer cells to invade surrounding tissues (45). Therefore, we tested the ability of mitoDFO to affect the invasive behavior of cancer cells using a 3D collagen invasion assay (26). Our results show that mitoDFO is highly efficient in inhibiting cell invasion at noncytotoxic concentrations (Fig. 7A and B). Again, the effect of mitoDFO was much more potent than that of deferoxamine.

We also observed a strong anti-metastatic effect of mitoDFO in vivo in the syngeneic model of 4T1 lung metastatic breast cancer cells. We assessed the total number of lung-metastasized 4T1 cells in control and mitoDFO-treated tumor-bearing animals (8 mg/kg i.p.), and observed an almost complete inhibition of lung metastases in mitoDFO-treated mice (Fig. 7C). On the other hand, deferoxamine treatment (8 and 80 mg/kg) had no effect on the number of lung metastases (Fig. 7C).

Iron, an essential micronutrient, has a role in many cellular processes, such as DNA replication, cellular metabolism, and mitochondrial respiration (1, 2). Cancer cells show higher demand for iron due to their proliferative nature and altered metabolic needs, (8–10, 46). Several reports have shown that higher iron uptake and decreased iron export are associated with increased tumorigenicity and poor clinical outcome (9, 46). Therefore, iron chelators have been tested as possible anticancer agents (8, 10, 12–14, 30, 47–49), and shown to inhibit proliferation of cancer cells in vitro (13) and delay the growth of tumors (10, 12, 49). However, low intracellular availability of the agents and concomitant side effects linked to considerable alterations in systemic iron metabolism have been observed (14).

Within cells, iron is primarily present either in heme or [Fe-S] clusters (5). Because both are synthesized by mitochondria, these organelles are crucial for cellular iron metabolism (2, 4). We have, therefore, modified the structure of the iron chelator deferoxamine by tagging it with two triphenylphosphonium groups (TPP⁺), to target it to mitochondria (15). Our data show that mitoDFO accumulates in mitochondria, as 1-hour incubation with fluorescently labeled mitoDFO-FITC was sufficient to observe almost exclusive mitochondrial localization of FITC staining.

When designing mitoDFO, we tested deferoxamine tagged with one or two TPP⁺ groups and found that mitoDFO with two TPP+ moieties is more potent (Supplementary Table S2). Therefore, this structure was used in all further experiments. The likely explanation of its enhanced potency is that a more lipophilic double-TPP⁺ mitoDFO is better at carrying the highly hydrophilic deferoxamine moiety across the membrane. On the other hand, the addition of the FITC moiety to mitoDFO decreased its effect, probably due to its bigger size and decreased lipophilicity (Supplementary Table S2). Of note, mitoDFO presented here is different to the mitochondrially targeted deferoxamine synthesized by Alta and colleagues (referred to as TPP-DFO; ref. 50), who reported it as a nontoxic chelator with antioxidant and protective activity against iron overload, with some evidence of its biological effects and virtually no results on its anticancer effect. Furthermore, Alta's structure presents a two-carbon linker between the TPP group and the mother compound, and contains a single TPP group, making it much less lipophilic when compared with mitoDFO. Similarly, TPP-DFO was designed so that the labile imine bond can be hydrolyzed under physiologic conditions, potentially releasing free deferoxamine inside cells, whereas mitoDFO comprises an amine bond that is much less prone to intracellular hydrolysis. These differences point to the fact that both compounds have distinct mechanisms of action, and opposite effects on cells, that is, TPP-DFO is a protective compound while mitoDFO is a toxic agent in cancer cells. In fact, the IC₅₀ reported for TPP-DFO in A2780 human cells is 135 μmol/L, while IC₅₀ values of mitoDFO are in the range of 2–7 μmol/L. This suggests that the length and identity of the linker, the chemical structure of the TPP⁺ attachment and the number of TPP⁺ groups present, considerably affect the activity of the synthesized compounds.

As deferoxamine is known to induce transferrin receptor 1 and reduce cytosolic ferritin as a consequence of iron starvation, we tested whether mitoDFO exerts a similar effect. Furthermore, we tested whether mitoDFO interferes with iron-dependent processes inside mitochondria, including [Fe-S] cluster and heme biosynthesis. Our results show that mitoDFO diminishes (i) mitochondrial [Fe-S] cluster biogenesis as evidenced by the decrease in ISCU, (ii) heme synthesis by downregulation of the last enzyme of heme biosynthesis, FECH, and (iii) the levels of [Fe-S] cluster/heme-containing subunits of individual mitochondrial respiratory complexes: NDUFA9 (CI), SDHB (CII), UQCRFS1 (CIII), and mtCO1 (CIV). The decrease in SDHB, ISCU, and FECH has already been reported in situations where [Fe-S] cluster biogenesis is impaired (51, 52) and thus is in agreement with our observations. Moreover, the decrease in [Fe-S] cluster availability was further confirmed by the aconitase in-gel activity assay, which showed a marked decrease in enzymatic activity in mitoDFO-treated cells. Of note, the enzymatic activity of mitochondrial aconitase (ACO2) was strongly reduced by mitoDFO, while such effect was much milder with deferoxamine. Aconitase 1 that lacks the [Fe-S] cluster behaves as IRP1 and binds the IRE sequences of genes related to iron metabolism, posttranscriptionally decreasing their translation (ferritin, FTH) or stabilizing their mRNA (transferrin receptor 1, TFR1). We observed that mitoDFO increased the level of TFR1 and decreased the level of FTH, although the effect was less prominent compared with deferoxamine. Moreover, mitoDFO also decreased the level of lipoylated proteins, possibly due to [Fe-S] cluster-dependent activity of lipoyl synthase (31, 52, 53), thus linking [Fe-S] cluster deficiency with other metabolic pathways, in agreement with low PDH E2 and αKGDH E2 following repression of [Fe-S] cluster biogenesis by lipopolysaccharide (51). Similarly, phosphorylation of EIF2α was increased by mitoDFO treatment, probably due to activation of the HRI kinase resulting from mitoDFO-induced decrease in heme levels.

Furthermore, we observed that mitoDFO treatment resulted in lower level of ⁵⁵Fe-labeled proteins within mitochondria and the cytosol, and extended incubation (60 hours) led to an even stronger effect. An exception was the storage protein ferritin, which accumulated ⁵⁵Fe, likely representing a “sink” for iron that cannot be incorporated into [Fe-S] clusters and heme.

Taken together, our data show that mitoDFO disturbs the [Fe-S] cluster biogenesis, leading to a decrease in the level and activity of [Fe-S] cluster/heme-containing proteins, confirming that mitoDFO deprives cancer cells of biologically active iron.

Mitochondrial ETC is instrumental for generation of the electrochemical gradient coupled with ATP synthesis and is also the main source of ROS (41, 54). We found that mitoDFO almost completely abolished oxygen consumption and led to a significant destabilization and disassembly of SCs including the respirasome, which is essential for CI-dependent respiration (40) an effect not observed in the case of deferoxamine. This is consistent with our previous reports on several TPP⁺-containing compounds that inhibit CI in malignant cells by blocking the movement of electrons from the catalytic site of CI to its coenzyme Q binding domain (16, 18). Of note, the effect of mitoDFO on the assembly of SCs is unlikely to explain the fast decrease in the OCR described above, because long time incubation (≥6 hours) is required to affect SC assembly while a decrease in OCR is evident within minutes. Importantly, we also show that mitoDFO leads to significant increase in ECAR and combination of the glycolysis inhibitor 2-DG with mitoDFO resulted in profound decrease in cellular ATP. This suggests that cancer cells, in response to mitoDFO, switch their bioenergetics toward glycolysis in order to generate ATP.

Because mitochondria are the main cellular source of ROS (41, 54), we assessed the effect of mitoDFO on ROS level and found that both total cellular ROS and mitochondrial superoxide levels were significantly increased and the antioxidant NAC was able to significantly protect against mitoDFO-induced cytotoxicity, confirming the role of ROS in the response to mitoDFO. The incomplete protection could reflect the fact that NAC is a water-soluble general ROS scavenger not directed to mitochondria.

Dysfunctional mitochondria differ in their intracellular localization and architecture (6, 55) and are divided into smaller organelles by the process of fission (56). The resulting mitochondrial fragments could then be degraded by mitophagy, thus maintaining a healthy mitochondrial pool (6, 42, 44). Our results show that mitoDFO induces mitochondrial fragmentation and decreases their perinuclear localization, increasing also mitochondrial acidification and PINK1-dependent mitophagy, supporting the finding that intracellular iron chelation induces engulfment of mitochondria by lysosomes (7). Because mitophagy maintains a pool of healthy mitochondria, it is likely that the observed induction of mitophagy is an adaptive response of cancer cells that are trying to maintain functional mitochondria and reduce generation of ROS (57). A report by Seebacher and colleagues linked novel cytotoxic iron chelators with lysosomal destabilization (58), which could also impact the process of mitophagy. However, we did not detect any effect of mitoDFO on lysosomal permeability assessed by the acridine orange method (59).

Delocalized lipophilic cations have been reported to accumulate preferentially in cancer cells (16–19, 60) due to their higher mitochondrial membrane potential (61, 62). Our results show that by tagging deferoxamine with TPP⁺, we increased its antiproliferative ability over 40-fold in several cancer cell lines. The IC₅₀ values were in the range of 1–8 μmol/L (with two exceptions being higher than 25 μmol/L), and normal fibroblasts were more resistant to the agent than most breast, ovarian, or pancreatic cancer cells. A detailed analysis by real-time cell monitoring revealed that mitoDFO exerted a potent cytostatic effect at concentrations lower than 5 μmol/L, consistent with accumulation of cells in G₁-phase of cell cycle, while nonmalignant fibroblasts were unaffected. Furthermore, the cytotoxic effect seen at higher concentrations was considerably more pronounced in cancer cells than in their nonmalignant counterparts and was present even in cell lines that were less responsive to the cytostatic effect of mitoDFO. Importantly, the cytotoxic effect elicited by mitoDFO was dependent on the iron-chelating activity as mitoDFO precomplexed with iron was significantly less toxic toward malignant cells.

Because the in vitro data suggested a profound cytostatic and cytotoxic effect of mitoDFO on cancer cells, the next step was to investigate the effect of the agent on tumor growth and metastasis in vivo. Our data show a significant antitumor effect of mitoDFO in both immuno-competent (Balb/c) and immuno-suppressed (NSG) mice, with a significant reduction of Ki67 staining in the treated syngeneic tumors, documenting an effect of mitoDFO on the proliferation of cancer cells in vivo. Furthermore, our analysis showed decreased level of ISCU, SDHB, and FECH proteins in mitoDFO-treated tumors, while transferrin receptor 1 was increased, suggesting the inhibition of [Fe-S] cluster and heme biogenesis and activation of the IRP/IRE system similarly to the in vitro data. Moreover, we also detected the induction of PINK1, suggesting active mitophagy. Such changes were not observed in liver tissue of mitoDFO-treated mice, and our results illustrate that mitoDFO does not affect systemic iron metabolism as serum, spleen, and liver iron and standard hematologic parameters were not changed. Furthermore, mice weight was not affected by mitoDFO treatment. This strengthens the premise that mitoDFO preferentially affects malignant cells while sparing normal cells, and allows us to postulate that mitoDFO may be a lead compound in the design of novel anticancer therapies with high efficacy and lower general toxicity.

As cancer spreading and poor clinical outcome is connected with metastasis formation (45, 63), we analyzed the migrastatic properties of mitoDFO. Our results show significantly reduced cellular invasion in the 3D in vitro spheroid model and a strong migrastatic activity in the mouse model of lung metastasis formation by 4T1 cells (27) while deferoxamine failed to show such effect. This confirms the critical importance of effective mitochondrial metabolism for the energy demanding process of cancer cell invasiveness (64). It is also possible that highly migratory cells possess higher ROS level and mitoDFO-induced increase in ROS exceeds the sustainable ROS level, leading to cytostatic and cytotoxic effect, thus stopping cellular migration.

Interestingly, our results show that mitoDFO is more efficient in delaying the tumor growth of mouse 4T1 cells in immunocompetent Balb/c mice compared with immunocompromised animals with breast or pancreatic human cells. It is possible that the in vivo effect of mitoDFO could be partially dependent on immune system, in agreement with finding that induction of mitophagy increases antigen presentation of malignant cells in vivo, resulting in immune system-mediated recognition and elimination of cancer cells (65).

In summary, mitoDFO represents a novel approach that targets iron metabolism of cancer cells, representing an agent with considerable cytostatic, cytotoxic, and migrastatic properties derived from an FDA-approved drug for treating iron overload. Targeting mitochondrial iron metabolism thus represents one of the possible strategies that might be the basis for the development of next-generation antineoplastic treatments. Of importance, mitoDFO targets a critical process of the cell, the synthesis of biologically active iron-containing cofactors, required for a wide range of cancer cells to maintain high rates of proliferation and the ensuing tumor progression. This also means that our new compound does not target a single protein or pathway, by virtue of which, frequent complications that compromise cancer therapy, such as genetic instability and multiple mutations, are unlikely to affect the response to mitoDFO.

J. Stursa reports a patent for PCT/CZ2018/050036, WO/2019/015701, EP3655414, US20200078379 pending. L. Werner reports grants from Czech Academy of Sciences during the conduct of the study; in addition, L. Werner has a patent for PCT/CZ2018/050036, WO/2019/015701, EP3655414, US20200078379 (deferoxamine derivatives as medicaments) pending. J. Truksa reports grants from Czech Science Foundation and Ministry of Education, Youth, and Sport of the Czech Republic during the conduct of the study; in addition, J. Truksa has a patent for PCT/CZ2018/050036, WO/2019/015701, EP3655414, US20200078379 (deferoxamine derivatives as medicaments) pending. No disclosures were reported by the other authors.

C. Sandoval-Acuña: Conceptualization, data curation, formal analysis, supervision, investigation, methodology, writing–original draft, writing–review and editing. N. Torrealba: Data curation, formal analysis, investigation, methodology, writing–review and editing. V. Tomkova: Investigation, methodology, writing–review and editing. S.B. Jadhav: Investigation, methodology. K. Blazkova: Investigation, methodology. L. Merta: Investigation, methodology. S. Lettlova: Investigation, methodology, writing–review and editing. M.K. Adamcová: Resources. D. Rosel: Supervision, validation, methodology, writing–review and editing. J. Brábek: Supervision, validation, methodology, writing–review and editing. J. Neuzil: Conceptualization, funding acquisition, methodology, project administration, writing–review and editing. J. Stursa: Funding acquisition, investigation, methodology, writing–original draft, writing–review and editing. L. Werner: Resources, funding acquisition, investigation, methodology, project administration, writing–review and editing. J. Truksa: Conceptualization, resources, data curation, supervision, funding acquisition, validation, investigation, methodology, writing–original draft, project administration, writing–review and editing.

This work has been supported by GACR grants 16-12816S and 18-13103S, and the institutional support of the Czech Academy of Sciences RVO: 86652036 to J. Truksa, and GACR grants 17-01192J and 18-10832S, an AZV grant 17-30138A, and funding from the Australian Research Council to J. Neuzil It was further supported by the MEYS of CR within the LQ1604 National Sustainability Program II (Project BIOCEV-FAR). The results were obtained in cooperation with Czech Centre for Phenogenomics supported by the MEYS of CR no. LM2018126 and by MEYS and ERDF OP VaVpI CZ.1.05/2.1.00/19.039. This publication is a result of the project implementation: “The equipment for metabolomic and cell analyses,” registration number CZ.1.05/2.1.00/19.0400, supported by Research and Development for Innovations Operational Programme (RDIOP) cofinanced by European regional development fund and the state budget of the Czech Republic. The CMS-Biocev (Biophysical techniques/Crystallization/Diffraction/Structural mass spectrometry) was supported by MEYS CR (LM2015043). L. Werner was supported in part by the Research Programme 18—Strategy AV21 of the Czech Academy of Sciences.

The authors thank Roberto Bravo-Sagua from INTA, University of Chile, for his cordial assistance with microscopy images processing and analysis. The authors acknowledge Imaging Methods Core Facility at BIOCEV, institution supported by the MEYS CR (LM2018129 Czech-BioImaging) and ERDF (project No. CZ.02.1.01/0.0/0.0/16_013/0001775 and CZ.02.1.01/0.0/0.0/18_046/0016045) for their support with obtaining imaging and flow cytometry data presented in this article. Graphical abstract was created using BioRender.

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.