The Antioxidant Transcription Factor Nrf2 Negatively Regulates Autophagy and Growth Arrest Induced by the Anticancer Redox Agent Mitoquinone*

Mitoquinone (MitoQ) is a synthetically modified, redox-active ubiquinone compound that accumulates predominantly in mitochondria. We found that MitoQ is 30-fold more cytotoxic to breast cancer cells than to healthy mammary cells. MitoQ treatment led to irreversible inhibition of clonogenic growth of breast cancer cells through a combination of autophagy and apoptotic cell death mechanisms. Relatively limited cytotoxicity was seen with the parent ubiquinone coenzyme Q10. Inhibition of cancer cell growth by MitoQ was associated with G1/S cell cycle arrest and phosphorylation of the checkpoint kinases Chk1 and Chk2. The possible role of oxidative stress in MitoQ activity was investigated by measuring the products of hydroethidine oxidation. Increases in ethidium and dihydroethidium levels, markers of one-electron oxidation of hydroethidine, were observed at cytotoxic concentrations of MitoQ. Keap1, an oxidative stress sensor protein that regulates the antioxidant transcription factor Nrf2, underwent oxidation, degradation, and dissociation from Nrf2 in MitoQ-treated cells. Nrf2 protein levels, nuclear localization, and transcriptional activity also increased following MitoQ treatment. Knockdown of Nrf2 caused a 2-fold increase in autophagy and an increase in G1 cell cycle arrest in response to MitoQ but had no apparent effect on apoptosis. The Nrf2-regulated enzyme NQO1 is partly responsible for controlling the level of autophagy. Keap1 and Nrf2 act as redox sensors for oxidative perturbations that lead to autophagy. MitoQ and similar compounds should be further evaluated for novel anticancer activity.

When superoxide is not rapidly detoxified, there is higher likelihood for generation of additional oxidants in mitochondria, leading to oxidative damage. Hence, targeting the mitochondria with agents capable of scavenging oxidants is an attractive cytoprotective strategy (1). Conjugation with a cationic triphenylphosphonium moiety is a method for targeting therapeutic molecules to the mitochondria (2). Mitoquinone (MitoQ) 3 is one such conjugate. It is derived from the mitochondrial respiratory chain component ubiquinone (3)(4)(5). The ubiquinone moiety is covalently linked by an aliphatic 10-carbon chain to a triphenylphosphonium cation. As a lipophilic cation, it can pass easily through lipid bilayers. Driven by the large membrane potential in the mitochondria, it selectively accumulates in the mitochondria. MitoQ is orally active and is being pursued as a potential therapy in degenerative conditions, such as Parkinson disease (2). Although initially designed as an antioxidant, mitochondrial redox cycling of mitoquinone has been shown to lead to superoxide production and apoptotic cell death (6,7).
Nrf2 (NF-E2-related factor 2) is a major transcription factor for antioxidant response element (ARE)-dependent gene transcription in response to agents that cause oxidative stress (including by the quinone oxidant tBHQ) (8). Nrf2 is critical in the regulation of antioxidant and cytoprotective genes. In the absence of oxidative stress, Nrf2 is sequestered within the cytosol by its inhibitory partner and principle regulator, Keap1 (Kelch-like ECH-associating protein 1). Nrf2 activity is normally limited by Keap1, which constitutively facilitates ubiquitination of Nrf2 and causes its ongoing proteasomal degradation (9). Keap1 can act as a sensor for oxidative stress due to its highly reactive cysteine residues. In vitro, in the presence of free radicals, Nrf2 is released from Keap1 and translocates to the nucleus, where it binds to the ARE and activates gene expression. One such Nrf2-regulated gene is NQO1 (NAD(P)H:quinine oxidoreductase 1). NQO1 is a cytoprotective flavoprotein that catalyzes the two-electron reduction of various quinones and redox dyes (10).
The role of mitochondrial oxidative phosphorylation and signaling in cancer cells is complex and controversial (11,12). Cancer cells have a greater mitochondrial membrane potential than healthy cells (13). Rhodamine 123, a lipophilic cationic probe, selectively accumulates in the mitochondria and consequently induces greater toxicity in cancer cells (14,15). Likewise, we found that mitoquinone has significant and selective anticancer activity, unlike untargeted ubiquinone (CoQ 10 ). Here, we demonstrate that the Nrf2-Keap1 proteins act as redox sensors that relay the signal for a failed cytoprotective response to MitoQ. Data are provided to elucidate the cellular response to MitoQ and to determine its novel underlying mechanism of action.

EXPERIMENTAL PROCEDURES
Synthesis of MitoQ-A new method was developed to synthesize MitoQ starting from the commercially available idebenone (see scheme in supplemental Fig. 1).
Idebenone (3.38 g, 0.01 mol) was stirred in 50 ml of dichloromethane containing triethylamine (2 g, 0.02 mol) in an ice bath. To this mixture, methanesulfonyl chloride (2.24 g, 0.02 mol) in 20 ml of dichloromethane was added dropwise. The mixture was stirred on ice for 1 h, and the solution was shaken with 1 M HCl until washings became acidic. The solution was dried, and solvent was removed using a rotary evaporator. The orange semisolid was chromatographed on a silica gel column, and the orange band was eluted with ether-hexane (1:1) solvent. Homogeneous fractions were combined, and solvent was subsequently removed to obtain an orange solid. The yield was almost quantitative. Thin layer chromatography (TLC) on silica gel plates showed a single spot. The methane sulfonyl ester was dissolved in 50 ml of dry acetone, and lithium bromide (1.75 g, 0.02 mol) was added and stirred at room temperature for 24 h. The solvent acetone was removed by evaporation, and the product was extracted with 50 ml of ether. The ether extract was washed with water (2 ϫ 25 ml), the orange solution was dried, and solvent was removed. The compound was purified by silica gel column chromatography, and the bromoderivative of idebenone was eluted with ether/hexane (1:1). Homogeneous fractions were combined, and solvent was removed to get an orange solid. The yield was 3.56 g (89%). TLC in ether/hexane solvent showed a single spot that moved ahead of the starting methanesulfonyl ester.
The bromoderivative was reduced with sodium borohydride (NaBH 4 ) to the corresponding quinol. The bromoderivative was dissolved in 50 ml methanol, and sodium borohydride (1.5 g, 0.04 mol) was slowly added in small aliquots with stirring. After the addition, stirring was continued for an additional 30 min, and the colorless solution was acidified with dilute hydrochloric acid and extracted with 50 ml of chloroform. The organic layer was dried, solvent was removed by blowing nitrogen gas, and the volatiles were removed by vacuum. The resulting colorless oil (quinol) was immediately used in the next step to prevent further oxidation in air to the corresponding quinone.
In 25 ml of dioxane, triphenylphosphene (3.9 g, 0.015 mol) was added to the quinol and kept refluxed under a nitrogen atmosphere for 48 h. The solvent was partially removed by rotary evaporation, and 50 ml of ether was added with stirring.
The precipitated product was collected by decantation, and the residual ether was removed by blowing nitrogen gas. The semisolid product was dissolved in 5 ml of dichloromethane and added to the solution in drops to 50 ml of ether with stirring. The precipitated product was collected as before by decantation. The dried product was dissolved in dichloromethane and precipitated from ether, and this process was repeated four times. The purified product consisted of mitoquinol with trace amounts of the oxidized compound mitoquinone. The product was dissolved in 50 ml of ethanol and was gently bubbled with air for 24 h to allow complete oxidation of mitoquinol to mitoquinone. The final quantity of mitoquinone was 3.1 g with an overall yield of 47%. The purity was confirmed by HPLC and LC/MS (mass ϭ 583).
Colony Formation Assay-Cell survival was determined using a colony formation assay after the indicated treatments for 6 h and replating for 7 days followed by washing, fixation, and staining with Wright's Giemsa stain (Sigma) (18). Colonies were counted using a light microscope and triplicate samples from each experiment. Colonies with greater than 30 cells each were counted.
Assessment of Antiproliferative Activity-Growth inhibition was assessed by the sulforhodamine B (SRB; Sigma-Aldrich) assay as described previously (18). GI 50 values were calculated using Prism 4 software (GraphPad Software, Inc.).
Flow Cytometric Analysis-For cell cycle analyses, cells were fixed in 70% ethanol and stained with PI after treatment with 5 g/ml RNase A. Cells were analyzed by a FACSCaliber flow cytometer (BD Biosciences), and cell cycle phases were quantitated using the CellQuest program. For apoptosis analyses, cells were stained with annexin V and PI (16). Both floating and attached cells were collected for FACS analyses.
Ultrastructural Analysis of Autophagy by Transmission Electron Microscopy-Transmission electron microscopy analysis was carried out using either high pressure freezing/freeze substitution or chemical fixation/resin flat embedding as described previously (19).
Quantitation of Percentage of Autophagic Cells-Autophagosomes were identified and counted with a fluorescence micro-scope using a 40ϫ objective. Autophagosomes were identified from the phase-contrast image using the intracellular structure images obtained by electron microscopy for comparison. A representative example is provided below the graph in Fig. 3D. The percentage of cells with autophagosomes was quantitated by counting cells that stained orange after treatment with acridine orange to stain the acidic vesicular organelles (see supplemental Fig. 2) (20).
Immunoprecipitation and Western Blot Analysis-The procedures for lysate preparation, immunoprecipitation, and Western blotting were performed as reported previously (18,21).
Hydroethidine Assay with HPLC Analysis-For the detection of superoxide and/or other reactive oxygen species (ROS), cells were incubated in DMEM with 2% FBS and treated with 10 M HE during the last 30 min of incubation with MitoQ. After the incubation, the cells were processed and analyzed by HPLC with electrochemical detection, as described previously (23).
Transcriptional Activity of Nrf2-Nuclear extracts were prepared from MDA-MB-231 cells using the nuclear extract kit by Active Motif (Carlsbad, CA). The presence of nucleophosmin and absence of GAPDH protein were used as a measure of the purity of nuclear extracts. An ELISA-based assay consisting of an immobilized oligonucleotide containing the ARE consensus binding site (5Ј-GTCACAGTGACTCAGCAGAATCTG-3Ј) was used to measure Nrf2 DNA binding activity (TransAM kits, Active Motif, catalog no. 50296). Nrf2 from 2.5 g of nuclear extract/sample was allowed to bind to the ARE on the 96-well plates. A primary antibody against Nrf2 was then used to detect bound Nrf2. A secondary antibody conjugated to horseradish peroxidase (HRP) provided a colorimetric readout at 450 nm. Nuclear extracts from COS-7 cells transfected with Nrf2 were included as the positive control.
Statistical Analysis-Data shown are representative of at least three independent experiments, unless noted otherwise. Error bars on graphs in all figures denote S.D. values. Student's t test was used to ascertain statistically significant differences (p Ͻ 0.05).

Cytotoxicity of MitoQ in Healthy Cells and Cancer Cell Lines-
The growth-inhibitory activity of MitoQ was tested in the breast cancer cell lines MDA-MB-231 and MCF-7 and in MCF12A primary mammary epithelial cells using the SRB dye assay (Fig. 1A, left). The effects of MitoQ were compared with untargeted CoQ 10 in the same cell lines (Fig. 1A, right). Breast cancer cell lines were more sensitive to MitoQ than healthy mammary cells. The GI 50 values for MitoQ in MDA-MB-231, MCF-7, and MCF12A cell lines were 296 nM, 113 nM, and Ͼ10 M, respectively. In contrast, the GI 50 values for CoQ 10 in these cells were 8.2 M (MDA-MB-231) and Ͼ10 M (for MCF-7 and MCF-12A). Thus, MitoQ exhibits selective antiproliferative effects in breast cancer cells as compared with healthy mammary cells.
One of the hallmark traits of cancer cells is their ability to undergo isolated clonal growth. Results obtained from the SRB growth inhibition assay were confirmed using a colony formation assay for cancer cell killing (Fig. 1B). MDA-MB-231 cells were exposed to MitoQ for 6 h, followed by removal of media and replating. MDA-MB-231 cells were evaluated for their colony formation ability for up to 7 days in drug-free media. MitoQ elicited a dose-dependent decrease in clonogenic growth.
Cell Cycle Arrest by MitoQ-To better understand the mechanism of cytotoxicity by MitoQ, we analyzed the cell cycle and apoptosis profile of MDA-MB-231 cells following drug exposure at 48 h (Fig. 1C). MitoQ arrested MDA-MB-231 cells in the G 1 /S phase of the cell cycle. Compared with control cells, the percentage of G 1 cells increased from 46.8% to 50.6 and 68.2% after exposure to 0.1 and 1 M MitoQ, respectively. A concentration-dependent increase in sub-G 1 DNA-containing cells, indicative of apoptosis, was also detected. With 1 M MitoQ, 10.8% of cells were apoptotic compared with 1.6% in control cells. The cell cycle arrest by MitoQ was compared with that by the untargeted ubiquinone, CoQ 10 . In contrast to MitoQ, 1 M CoQ 10 produced a smaller G 1 /S phase arrest and produced significantly fewer sub-G 1 DNA-containing cells (2.9%) (Fig. 1C). To determine the mechanism by which MDA-MB-231 cells undergo G 1 /S cell cycle arrest, we assayed the phosphorylation of cell cycle checkpoint kinases Chk1 and Chk2. These proteins help maintain the cell cycle checkpoint and may also promote apoptosis (24) via a complex network of phosphorylation-dephosphorylation events (25). MDA-MB-231 cells exposed to 0.1 M MitoQ showed significant phosphorylation of phospho-Chk1-Ser 317 , phospho-Chk1-Ser 345 , and phospho-Chk2-Thr 68 (Fig. 1D). Doxorubicin served as a positive control. The data show that MitoQ causes cell cycle arrest and checkpoint kinase activation in cancer cells.
Although 40 -60% of the cells underwent growth inhibition ( Fig. 1A) with 1 M MitoQ, this did not appear to translate into apoptotic events (ϳ11%, as indicated by the number of sub-G 1 DNA-containing cells). We performed FACS analysis of annexin V-PI staining of live cells treated with 1 M MitoQ ( Fig.  2A) to further quantify apoptosis. A time course study with MitoQ showed no appreciable apoptosis prior to 24 h. Follow-ing treatment with MitoQ for 48 and 72 h, 21.8 and 28.8% of cells were annexin V-positive compared with 3% in control cells. Comparable results were obtained using Hoechst-PI staining to quantify apoptosis (not shown). We conclude that apoptosis only accounts for part of the mechanism by which MitoQ elicits its anticancer growth inhibition.
To test whether apoptosis by MitoQ was dependent on caspases, we pretreated cells with 10 M benzyloxycarbonyl-VAD-fluoromethyl ketone for 30 min followed by MitoQ for 48 h (Fig. 2B). Pretreatment with benzyloxycarbonyl-VAD-fluoromethyl ketone did not appreciably lower the total percentage of annexin V-positive cells (18.6% compared with 22.4%). Cells treated with recombinant human TRAIL (10 ng/ml for 16 h) were used as positive control for caspase-dependent apoptosis and showed abrogation of apoptosis in the presence of benzy- loxycarbonyl-VAD-fluoromethyl ketone. We also found no significant difference in the percentage of MitoQ-induced annexin V-positive cells in MCF7 cells, which lack caspase-3 ( Fig. 2C). Hence, we conclude that MitoQ-induced apoptosis is not caspase-dependent.
Mitochondrial Membrane Depolarization by MitoQ-To determine whether MitoQ causes disruption of the mitochondrial membrane, we studied the mitochondrial membrane potential in MDA-MB-231 breast cancer cells after drug expo-sure. Conversion of red fluorescent aggregates to green fluorescent monomers of the JC-1 dye upon membrane depolarization marks changes in membrane potential. Exposure of cells to MitoQ (0.1 and 1 M) led to a rapid membrane depolarization within 60 min of exposure (Fig. 2D). Membrane depolarization appeared to be sustained for up to 24 h after drug exposure. In contrast to MitoQ, CoQ 10 did not exhibit any appreciable effect at comparable concentrations. VP-16 was included as a positive control (26). Next, the release of cytochrome c from the mito- chondria to the cytosol was measured as an indicator of mitochondrial damage after MitoQ treatment (1 M). A distinct increase in cytosolic cytochrome c protein levels was observed at 6 h (Fig. 2E). Thus, MitoQ exposure is accompanied by a rapid membrane depolarization and cytochrome c release that may be partly responsible for the apoptosis induced in breast cancer cells.
Induction of Autophagy by MitoQ-Our findings indicate that apoptosis may not be the major mechanism responsible for the antiproliferative activity of MitoQ. We therefore hypothesized that MitoQ may induce autophagy that leads to inhibition of proliferation. Autophagy is one of the consequences of stressing cells, and it is emerging as an important mechanism to explain drug responses in cancer cells (27)(28)(29). Autophagy is characterized by formation of autophagic vacuoles (autophagosomes) (30,31), which can be seen by transmission electron microscopy (31). Autophagosomes were observed after 24 h of incubation with 1 M MitoQ (Fig. 3A). The presence of mito-chondria within the autophagic vacuole ("mitophagy") was also noted (labeled by arrowheads) (32). Mitophagy is a potential mechanism for turnover of dysfunctional mitochondria, in agreement with the rapid membrane depolarization and cytochrome c release (Fig. 2D) (32). Also shown are representative images of an early autophagic vacuole with a double membrane (labeled DM) and the fusion of two late stage autophagic vacuoles with degraded cellular contents (labeled L) after MitoQ treatment. Additionally, the formation of the proteolytic and lipidated maturation product of the LC3 protein LC3-I (microtubule-associated light chain 3 protein, termed LC3-II) correlates with the extent of autophagic activity (31)(32)(33)(34). We found an increase of LC3-II protein at 24 and 72 h after MitoQ treatment of MDA-MB-231 cells (Fig. 3B). Rapamycin (at 10 M, 24 h) served as the positive control. To confirm that the increase in LC3-II was not transient but indicative of autophagic increase in LC3-II that is subsequently degraded by lysosomal hydrolases after formation of autolysosomes (autophagic Rapamycin was used as a positive control. Increase in LC3-II was confirmed to be from autophagic flux using lysosomal protease inhibitors pepstatin A (10 g/ml) and E64d (10 g/ml) in the presence or absence of MitoQ (right). The graphs below show the densitometric quantitation of LC3-II protein levels normalized to ␣-tubulin levels from a representative analysis. C, measurement of apoptosis following inhibition of autophagy by bafilomycin A using annexin V-PI staining and flow cytometry. MDA-MB-231 cells were treated with MitoQ in the presence or absence of 5 nM bafilomycin A (2-h pretreatment). D, the percentage of cells with autophagosomes (Autophagic) was counted in MitoQ-or rapamycin-treated cells. Error bars, S.D. flux), we tested the LC3-II in the presence of E64d and pepstatin A, which inhibit lysosomal proteases (31,35). The presence of lysosomal inhibitors substantially stabilized the LC3-II induction by MitoQ (Fig. 3B, right). It should be noted that due to the massive induction of LC3-II in the presence of lysosomal inhibitors, the Western blot membranes had to be exposed for different periods of time to show clear and quantifiable bands for all samples. Finally, we found that pretreatment with the autophagy inhibitor bafilomycin A1 triggers increased apoptosis in response to 1 M MitoQ (Fig. 3C). Bafilomycin acts as a vacuolar H ϩ -ATPase inhibitor and prevents acidification of lysosomes and rematuration/fusion of autophagic vacuoles (36). Bafilomycin A1 (a 2-h pretreatment with 5 nM) synergistically increased the percentage of annexin V-positive cells by MitoQ. This result is in agreement with other recent reports showing increased apoptosis by autophagy-inducing agents in the presence of bafilomycin A1 (37,38). A time course of autophagosome induction in response to MitoQ is shown in Fig. 3D. The percentages of cells with detectable autophagosomes were counted manually. After 72 h, about 31% of cells were autophagic in response to MitoQ (representative cells are shown below the graph). Collectively, this demonstrates that MitoQ induces autophagy at growth-inhibitory concentrations in MDA-MB-231 cells.

MitoQ-induced Modification of Oxidative Stress Sensor Keap1 and Dissociation from the Antioxidant Transcription
Factor Nrf2-To elucidate the nature of the cellular stress from MitoQ, experiments were performed to investigate if MitoQ generates ROS in breast cancer cells. The effect of MitoQ (1 M) on ROS production in MDA-MB-231 cells was tested using a hydroethidine (HE) assay with HPLC-based detection of HEderived oxidation products (39,40) (Fig. 4A). In the presence of 1 M MitoQ, there was a significant increase in the HE oxidation products ethidium (E ϩ ) and diethidium (E ϩ -E ϩ ). Formation of E ϩ and E ϩ -E ϩ is attributed to increased levels of oneelectron oxidizing species. The increase in 2-hydroxyethidium (2-OH-E ϩ ) was not statistically significant. Note that the intracellular level of HE did not change significantly with different treatments. Thus, the observed changes are not due to variations in the availability of the intracellular probe for oxidants (HE) (6,23). These data are in agreement with a previous report on detection of superoxide generated by MitoQ (6). The formation of E ϩ and E ϩ -E ϩ levels indicates that MitoQ stimulates intracellular generation of one-electron oxidizing species.
To define the molecular mechanism of MitoQ-induced growth arrest, we investigated the ability of oxidants generated by MitoQ to modify Keap1 and engage Nrf2 antioxidant transcriptional activity. The cysteine residues (Cys 273 , Cys 288 , and Cys 151 ) on Keap1 act as redox sensors that undergo reversible disulfide bond formation that can be detected as Keap1 dimers with altered electrophoretic mobility under non-reducing conditions (9). MDA-MB-231 cells were exposed to MitoQ, and whole cell lysates were used for Western blot analysis in the presence or absence of a reducing agent, ␤-mercaptoethanol. The oxidant tBHQ was used as a positive control (9). We detected a higher molecular weight band corresponding to oxidized Keap1 protein in the presence of MitoQ (10 M, 2 h). This modified Keap1 band was not detected under reducing condi-tions (left lanes), indicating that it represents disulfide-bonded (oxidized) Keap1 (Fig. 4B). Upon modification, Keap1 is known to undergo proteasome-independent degradation (41). Consistent with this, significant Keap1 degradation was observed after exposure to MitoQ (Fig. 4C). Thus, MitoQ induces oxidative modification and degradation of the oxidant-sensitive regulatory protein Keap1.
We next evaluated whether MitoQ-induced Keap1 modification leads to disruption of Keap1 binding to Nrf2. Keap1 was immunoprecipitated from lysates of MDA-MB-231 cells exposed to MitoQ (1 M, 6 or 24 h). The anti-Keap1 immunoprecipitates were subjected to immunoblot analysis with anti-Nrf2 antibody. A significant decrease in Keap1-Nrf2 interaction was detected after MitoQ treatment (Fig. 4D, top). tBHQ (50 M, 16 h) was used as a positive control. To confirm that the association between Keap1 and Nrf2 was disrupted by treatment with MitoQ, Nrf2 was also immunoprecipitated from the whole cell lysates, and immunoblots were probed with anti-Keap1 antibody (Fig. 4D, bottom). The results confirm a reduced amount of Keap1 protein bound to Nrf2 after MitoQ treatment.
Increased Nuclear Protein Levels and Transcriptional Activity of Nrf2-Disruption of Keap1 binding stabilizes Nrf2 protein levels (9). Release from Keap1 under conditions of oxidative stress also triggers the nuclear translocation of Nrf2, where it functions as an antioxidant transcription factor (8). We used whole cell lysates and nuclear fractions from MitoQ-treated MDA-MB-231 cells to investigate Nrf2 protein levels (Fig. 5A). After a 24-h exposure to MitoQ, a 3-fold increase in nuclear protein levels was observed (Fig. 5A). tBHQ was used as a positive control and also showed a greater than 3-fold increase in nuclear Nrf2. Next, to confirm nuclear translocation of Nrf2, we performed confocal microscopy analysis in cells exposed to 1 M MitoQ (Fig. 5B). A clear increase in nuclear Nrf2 protein was observed after 6 h with MitoQ. Finally, the transcriptional activity of Nrf2 was measured with an assay using an immobilized oligonucleotide containing the ARE consensus binding site (Fig. 5C). tBHQ was used as a positive control. Treatment of MDA-MD-231 cells with MitoQ for 24 h led to a 3.5-fold increase in transcriptional activity of Nrf2.
Knockdown of Nrf2 Increases Autophagy and G 1 Cell Cycle Arrest but Not Apoptosis from MitoQ-The critical role of Nrf2 as an antioxidant defense mechanism against the MitoQ-induced cellular effects of cell cycle arrest, apoptosis, and autophagy was investigated using siRNA against Nrf2. We identified the optimal conditions for knockdown of Nrf2 protein (100 nM siRNA oligonucleotides for 24 h) (Fig. 6A). We observed ϳ84% knockdown of Nrf2 under these conditions. Non-coding siRNA oligonucleotides were used as controls. The relative levels of LC3-II in Nrf2 siRNA-transfected cells were higher than in control siRNA-transfected cells at 24 h. The 1.4-fold increase in LC3-II in Nrf2-siRNA cells remained relatively steady for 72 h after transfection. After exposure of Nrf2 and control siRNA cells to MitoQ for 24 or 72h, we tested for autophagy. A significantly greater level of LC3-II protein was detected in Nrf2 siRNA cells compared with control siRNA, confirming greater levels of autophagy in Nrf2 knockdown cells exposed to MitoQ (Fig. 6B). The difference was most significant at 24 h. A detailed time course is shown in supplemental Fig. 3. We confirmed the role of Nrf2 in down-regulating autophagy by observing greater percentages of cells with autophagosomes in Nrf2 siRNA-transfected cells (Fig. 6C). We also observed a greater percentage of cells in G 1 with MitoQ in the Nrf2 siRNA cells compared with cells with control siRNA (Fig. 6D). We conclude that this increase corresponds to a greater percentage of cells in autophagic growth arrest. Finally, we measured apoptosis under knockdown conditions at various time points after adding MitoQ using annexin V-PI staining and FACS analysis. We did not observe a significant difference in the percentage of annexin V-positive cells at any tested time point (Fig. 6D, right). We also did not find significant differences in the sub-G 1 frac-tion of cells (indicative of subdiploid apoptotic cells) following treatment of Nrf2 siRNA cells with MitoQ. Thus, MitoQ-induced Nrf2 signaling is a critical antioxidant defense mechanism that negatively controls autophagy in response to ROS generated by MitoQ.

NQO1-deficient MDA-MB-231 Cells Undergo Higher Levels of Autophagy and G 1 Cell Cycle Arrest but Not Apoptosis from
MitoQ-NQO1 is a downstream target of Nrf2. We hypothesized that NQO1 might be one of the antioxidant enzymes regulated by Nrf2 that could negatively control the level of autophagy after MitoQ treatment. MDA-MB-231 cells have no detectable NQO1 protein under regular growth conditions FIGURE 4. Oxidation of Keap1 and its dissociation from Nrf2 in response to MitoQ. A, the generation of ROS by MitoQ was assayed by measuring HE and its oxidation products by HPLC with electrochemical detection in cells treated with HE (10 M) for the final 30 min of the MitoQ exposure. HPLC with electrochemical detection peak areas of HE, 2-OH-E ϩ , E ϩ , and E ϩ -E ϩ extracted from MDA-MB-231 cells are shown after normalization to the protein concentration in the cell lysates. Each panel includes the structure of the analyzed compound. B, Keap1 protein oxidation was confirmed by the appearance of a higher molecular weight Keap1 band that only appears under non-reducing conditions (without ␤-mercaptoethanol). tBHQ (50 M) was used as a positive control. Keap1 (68 kDa) and modified Keap1 (Ͼ120 kDa) are indicated with arrows. Quantitation of the bands representative of the oxidized and total Keap1 proteins is shown after normalization to ␣-tubulin levels. C, degradation of Keap1 in response to MitoQ was analyzed by Western blotting after exposure to 3, 6, or 24 h with 1 M MitoQ. tBHQ (at 50 M) was used as a positive control. D, interaction between Keap1 and Nrf2 is disrupted after MitoQ treatment. Top, Keap1 was immunoprecipitated (IP) from MitoQ-or tBHQ-treated lysates, followed by Western blotting (WB) with anti-Nrf2 antibody. Bottom, Nrf2 was immunoprecipitated from Mito-Q or tBHQ-treated lysates, followed by Western blotting with anti-Keap1 antibody. Input control samples were precleared with agarose beads and were not immunoprecipitated prior to Western blotting. Error bars, S.D. (42). Previously generated isogenic MDA-MB-231 cells stably transfected with NQO1 were used for studying the role of NQO1 (17). We first confirmed the NQO1 status in these cells using Western blot analysis (Fig. 7A, inset). Next we probed the growth inhibition profile by using two approaches. We tested cellular proliferation using the SRB dye assay and cell cycle arrest using propidium iodide staining/FACS analyses. MDA-MB-231 NQO1-negative cells showed greater levels of growth inhibition in response to increasing concentrations of MitoQ at 72 h (Fig. 7A). MDA-MB-231-NQO1-negative cells also showed higher percentage of cells in the G 1 cell cycle phase by FACS analysis at 48 h with 0.1 M MitoQ (Fig. 7B). In contrast, NQO1-positive and NQO1-negative cells showed no significant difference in the total percentage of apoptotic cells (Fig.  7C). It was noted that the NQO1-negative cells had a greater percentage of cells that were in early apoptosis following MitoQ, as indicated by the annexin V-positive and propidium iodide-negative fraction of cells. However, the total percentage of annexin V-positive cells was not significantly different between the two cell lines. These results were confirmed using Hoechst-PI staining of live cells (not shown). We then assayed for autophagy using LC3 Western blotting and light fluorescence microscopy. MDA-MB-231-NQO1-negative cells showed higher levels of LC3-II than MDA-MB-231-NQO1positive cells in response to MitoQ (Fig. 7D). The higher levels of LC3-II in the absence of NQO1 were also confirmed using the NQO1 inhibitor dicumarol in MDA-MB-231-NQO1-positive cells (43). Cells pretreated with dicumarol showed greater levels of LC3-II in response to MitoQ than with MitoQ alone (Fig. 7E). The NQO1-negative cells also showed a greater percentage of cells with autophagosomes as seen by microscopy (Fig. 7F). These results suggest that NQO1 is, at least in part, responsible for the downstream signaling from Nrf2 and controlling the level of autophagy in response to MitoQ.

DISCUSSION
MitoQ was first designed as an antioxidant intended to block mitochondrial oxidative damage by preventing lipid peroxidation (4). Within the mitochondria, MitoQ undergoes redox cycling by complex II and is reduced to mitoquinol. In contrast to the toxic effects of MitoQ on tumor cells, this compound is protective to normal cells (4,44,45). Consistent with these previous findings, our preliminary data suggest that MitoQ does not cause cytotoxicity in healthy mammary epithelial cells. Sensitivity of breast cancer cells (MDA-MB-231 and MCF-7) to MitoQ was 30-fold higher than for healthy mammary epithelial cells (MCF-12A). MitoQ also enhanced the anticancer cytotoxicity of doxorubicin when tested in breast cancer cells (not shown). We therefore hypothesized that MitoQ might possess unique anticancer activity as a single agent.
Our studies show that MitoQ is cytotoxic and induces both autophagy and apoptosis in human breast cancer cells. MitoQ caused irreversible inhibition of clonogenic cell growth, a G 1 /S cell cycle arrest, and checkpoint kinase activation. Rapid mitochondrial membrane depolarization was accompanied by a release of cytochrome c, a response that was not observed with the untargeted ubiquinone (CoQ 10 ). MitoQ treatment led to a greater percentage of cells undergoing autophagy than apoptosis under the same conditions. Irreversible growth arrest, observed as a loss of clonogenicity, suggests that autophagy might ultimately lead to cell death in response to MitoQ. Agents such as rapamycin and tunicamycin that preferentially cause a growth arrest in the G 1 /S phase of the cell cycle are also known to induce autophagy (34). Inhibitors of mitochondrial electron transport chain complexes also induce a caspase-independent, autophagic cell death (46,47). Autophagy under such circumstances occurs as a consequence of mitochondrially generated oxidants, including superoxide, hydrogen peroxide, and hydroxyl radicals (48). We therefore hypothesized that autophagy is a plausible mechanism for MitoQ-induced growth inhibition. Cell death following autophagy is characterized by morphological changes that are distinct from apoptosis (49). It involves the caspase-independent degradation of organelles, as observed with MitoQ in caspase-deficient MCF-7 cells or in MDA-MB-231 cells pretreated with a pan-caspase inhibitor. The role of autophagy in cancer chemotherapy is not fully understood (49,50). Although it could be a mechanism for drug resistance (i.e. through the degradation of unnecessary cellular components), it could also serve as a mechanism for drug-induced cell death (i.e. due to the degradation of critical cellular components). The observation that a short term exposure to MitoQ leads to an irreversible loss of clonogenic growth sug-   1-h pretreatment). F, percentage of cells with autophagosomes in the isogenic cell lines followed by MitoQ treatment were counted as in Fig. 3D. *, statistical significance compared with control siRNA cells. G, schematic model for the proposed mechanism for the anticancer cytotoxicity induced by MitoQ (see "Discussion"). Error bars, S.D.
gests that autophagy was unable to mitigate the cytotoxic and irreversible damage by MitoQ and is not a successful means of drug resistance. Cell death following autophagy occurs if the cells are unable to repair the damage. We found that cells exposed to an inhibitor of autophagy (bafilomycin A1) together with MitoQ underwent increased apoptosis compared with cells treated with MitoQ alone. This switch to apoptosis is a common observation for autophagic cells (37,38).
Our findings are also in concordance with previous reports relating the production of ROS to the induction of autophagic cell death in cancer cells (27,47,50). The pro-oxidant arsenic trioxide and celecoxib-derived OSU-03012 induce ROS and autophagy in cancer cells (27,51). We speculate that the production of ROS might be involved in the induction of autophagic cell death by MitoQ. In contrast, other to-be-determined mechanisms might be responsible for the apoptosis induced by MitoQ. A combination of autophagy and apoptosis has also been observed after camptothecin, tamoxifen, and oridonin treatment in breast cancer cells (52,53).
Given the nature of its redox cycling, it is not surprising that MitoQ was also found to generate ROS in isolated mitochondria (6). Oxidative perturbations within cells can trigger numerous defense signals that protect cells and tissues from irreversible damage. Based on our findings, we propose a model for the mechanism for MitoQ-induced cell death via autophagy and signaling by the Nrf2-Keap1 pathway (Fig. 7G). Nrf2 is a critical transcription factor that regulates antioxidant responses by increasing the expression of antioxidant enzymes. The cysteine-rich, redox-sensitive protein Keap1 acts as a negative regulator of Nrf2. We propose that the MitoQ-induced oxidation of Keap1 is a redox sensor for the cellular response and results in subsequent degradation of Keap1. This modification of Keap1 leads to the release, stabilization, and nuclear translocation of Nrf2 protein, ultimately resulting in increased Nrf2 transcriptional activity. Although we observed degradation of Keap1, this could not directly account for the disruption of Keap1-Nrf2 complex formation. We draw this conclusion based on the observation that Nrf2 protein stabilization and translocation (in Fig. 5) are dependent on the oxidation-induced degradation of Keap1 and disruption of Keap1-Nrf2 complex formation (9,21). Our reverse immunoprecipitation experiments also confirm the reduced protein-protein interaction. Earlier studies with oxidant-producing agents, such as tBHQ, sulforoaphane, and quercetin, showed a similar engagement of the Keap1-Nrf2 pathway (54,55). Our data demonstrate the critical role of Nrf2 in controlling the level of autophagy and growth inhibition in response to oxidative stress. We speculate that the induction of an Nrf2-mediated antioxidant response limits but does not abolish the induction of autophagy and growth arrest by MitoQ. Hence, Nrf2 is a cytoprotective signal that limits the extent of growth arrest and autophagy under basal conditions. In the absence of Nrf2 (e.g. following knockdown by siRNA), the antioxidant defense to MitoQ is decreased, resulting in increased unquenched ROS and increased autophagy and G 1 cell cycle arrest. We note that Nrf2 transfection alone caused a slight increase in the LC3-II protein level, which may indicate that Nrf2 is also responsible for regulating basal levels of autophagy in these cancer cells. This find-ing is unique and requires further investigation because, to our knowledge, there are no known Nrf2-responsive genes that inhibit or stimulate autophagy. With MitoQ, we observed an increase in the levels of LC3-I, in addition to LC3-II (Fig. 3B). We speculate that this previously unreported increase in LC3-I is unique to oxidative stress and is related to Nrf2 because it was not observed with rapamycin or in the absence of Nrf2 (Fig. 6B). We investigated NQO1, an Nrf2-regulated enzyme, for its role in regulating autophagy and apoptosis. Our data suggest that NQO1, in agreement with the role for Nrf2, limits the extent of autophagy induced by MitoQ. Isogenic cells deficient in NQO1 had elevated levels of autophagy and G 1 cell cycle arrest (Fig. 7). There was no observed effect on apoptosis by MitoQ. We speculate that NQO1 limits the oxidant-induced autophagy that is triggered by MitoQ. The role for NQO1 in controlling autophagy is also evident in the higher levels of LC3-II observed under basal conditions in the NQO1-deficient MDA-MB-231 cells, compared with other cancer cell lines (35). This is in agreement with the protective role for NQO1 observed in response to menadione in NQO1-null mice (56).
It is likely that the control of autophagy downstream of Nrf2 occurs with multiple antioxidant enzymes playing a role in clearing the reactive oxygen species. It is also plausible that other proteins indirectly regulated by Nrf2 might directly and negatively regulate autophagic processes.
The combination of MitoQ and the knockdown of Nrf2 may also point to a novel therapeutic opportunity to enhance the antitumor effects of MitoQ. Recent reports propose a new model for Nrf2-Keap1 signaling where the dissociation of Nrf2 from Keap1 is proposed to be an in vitro phenomenon resulting from the experimental use of excess drug concentrations (57). This model suggests that there are two separate pools of Nrf2 (free and bound to Keap1) that translocate to the nucleus to mediate antioxidant signaling. Our data cannot rule out this mechanism, but the combined effects of MitoQ on Keap1 and Nrf2 are consistent with the view that Keap1 down-regulates Nrf2 in the absence of oxidative stress.
Our observation that oxidative stress mediates autophagic signaling by an antiproliferative agent underscores the need to understand the multifaceted responses to oxidative stress induced by cancer drugs. Autophagic mechanisms can trigger a reversible growth arrest that allows cells to escape death after elimination of damaged cellular components. This defense mechanism could be responsible for the failure of many promising investigational drugs to be successfully translated into the clinic. In the case of MitoQ, the autophagic growth arrest and apoptosis appear to be irreversible, as manifested by reduced clonogenic growth even after removal of the drug. It is likely that MitoQ operates in a manner that is similar mechanistically to tBHQ and that it acts as a weak electrophile. It is conceivable that MitoQ is more cytotoxic to cancer cells than CoQ 10 because of the enhanced intracellular accumulation of MitoQ. We recently showed that MitoQ can reverse the cardiotoxic effects of doxorubicin, a widely used cancer chemotherapy drug, and that MitoQ enhances doxorubicin-mediated cytotoxicity in breast cancer cells (58). These attributes of MitoQ provide us with a unique opportunity to synergize with doxorubicin to better destroy breast tumor cells while providing protection from the cardiotoxic side effects of doxorubicin. These investigations are currently under way in our laboratory.