Mitochondria Participate in Chemoresistance to Cisplatin in Human Ovarian Cancer Cells

Ovarian cancer is the most lethal genital malignancy and the fifth leading cause of cancer-related death among women (1). At stage 1, when the primary tumor is limited to ovaries, cure can be achieved by surgical resection of internal genitals (2). At stages 2 to 4, when the tumor progressively extends to the pelvis, peritoneal tissues, retroperitoneal lymph nodes, and extra-abdominal organs, the treatment of choice is cytoreduction followed by platinum/taxane combination chemotherapy, which can be administered in adjuvant and/or neoadjuvant settings. However, relapse is frequent (70% of patients), and recurrent tumors most often develop resistance to chemotherapy, which eventually leads to patient death (3).

Cisplatin is the most commonly used chemotherapeutic drug for the treatment of several types of cancers, including bladder, head and neck, lung, ovarian, and testicular cancers. It is administered by intravenous injection as an inactive prodrug that must undergo hydrolysis inside cancer cells to be activated (4). Activated cisplatin is a potent electrophile that can react with nucleophilic biomolecules, cross the nuclear membrane, and bind covalently to guanine and adenine in DNA, forming cytotoxic DNA adducts. However, DNA damage–induced cell death is not its only anticancer mechanism. It is indeed estimated that less than 1% of intracellular platinum is bound to nuclear DNA (nDNA), while most of cisplatin interacts with more accessible nucleophilic sites on other molecules, including mitochondrial DNA (mtDNA), RNA, phospholipids, tubulin, and cytosolic organelles (5). Cisplatin also acts as a pro-oxidant that can trigger the production of superoxide and other reactive oxygen species (ROS) inside cells, thus activating intrinsic and extrinsic proapoptotic pathways (6, 7). Cisplatin binding to mtDNA further creates adducts that are not repaired as efficiently as in nDNA (8), and damage to mtDNA genes encoding components of the electron transport chain (ETC) can compromise respiration, which subsequently leads to ROS generation (7).

In ovarian cancer, resistance to cisplatin can proceed through multiple mechanisms categorized as pre-, on-, post-, and off-target depending on whether they are implemented by cancer cells before nDNA damage, during the phase of damage signaling, during the implementation of the damage response, or independently of the effects of cisplatin on nDNA, respectively (9). Pretarget mechanisms include reduced drug uptake (10), increased drug efflux (11, 12), and drug inactivation (e.g., through a direct reaction of cisplatin with glutathione; ref. 13). On-target resistance mechanisms are those that improve cell ability to repair DNA damage or that allow cell replication by ignoring the damage. Posttarget resistance mechanisms arise after nDNA damage and affect the ability of the cells to induce cell death. Prevailing events are inactivation of p53 (14) and overexpression of Bcl2 (15). Finally, off-target mechanisms concern the alteration of processes that are not directly activated by cisplatin but are nevertheless able to counteract the toxic effects of the drug. Autophagy, for example, can be increased in drug-resistant ovarian cancers (16).

Recent discoveries in a variety of cancer types suggested an active participation of mitochondria in chemoresistance (17). Therefore, in this study, we hypothesized that cisplatin could alter energy metabolism and redox signaling in ovarian cancer cells. To test this hypothesis, we compared the bioenergetics of two pairs of matched cisplatin-sensitive and cisplatin-resistant SKOV-3/SKOV-3-R (18) and COV-362/COV-362-R cells. Our findings reveal that cisplatin-resistant human ovarian cancer cells undergo an oxidative switch that could be related to an increased mitochondrial turnover rate. Consequently, targeting autophagy in general and mitophagy in particular could be a strategy to control the proliferation of cisplatin-resistant human ovarian cancer cells.

SKOV-3 human ovarian adenocarcinoma cancer cells, originally from the ATCC (catalog no. HTB-77, RRID: CVCL_0532) were a kind gift of Dr. Shoshan (Karolinska Institute, Stockholm, Sweden) in 2005, and were routinely cultured in RPMI1640 (Gibco Life Technologies, catalog no. 61870-036) containing 11 mmol/L glucose, 2 mmol/L glutamax, and supplemented with 10% FBS. COV-362 human ovarian epithelial endometroid cancer cells (purchased from Sigma-Aldrich in 2018, catalog no. 07071910, RRID: CVCL_2420) were routinely cultured in DMEM containing 4.5 g/L glucose and 2 mmol/L glutamax, and supplemented with 10% FBS. Cisplatin-resistant SKOV-3 (SKOV-3-R) and COV-362 (COV-362-R) cells were obtained by treating wild-type SKOV-3 (SKOV-3) and wild-type COV-362 (COV-362) cells, respectively, with increasing concentrations of cisplatin (Teva) from 5 μmol/L to 10 μmol/L with increments of 1.5 μmol/L every 72 hours (18). After each selection, cells were cultured for minimum six passages and up to 25 passages in the absence of cisplatin to ensure phenotypic stability. Cell authentication was performed at passage 6 with a short tandem repeat (STR) test (GeneMapper, Applied Biosystems). DNA was isolated with a QIAmp DNA kit (Qiagen) and amplified by PCR using the PowerPlex 16 System Promega Amplification Kit. Fifteen markers (D3S1358, THO1, D21S11, D18S51, Penta E, D5S818, D13S317, D7S820, D16S539, CSF1PO, Penta D, Vwa, D8S1179, TPOX, FGA, and amelogenin) were used to obtain the genetic profile (Laboratoire de Biologie Moléculaire, Cliniques Universitaires St-Luc, Brussels, Belgium). All cell lines were checked for Mycoplasma every 2 months using the MycoAlert Plus Mycoplasma Detection Kit from Lonza (a biochemical test that selectively reports on the activity of Mycoplasma enzymes, catalog no. LT07-710), according to the manufacturer's instructions.

Where indicated, cells were treated with the indicated doses of cisplatin, bafilomycin A1, H₂O₂, rotenone, N-acetyl-l-cysteine (NAC), antimycin A, oligomycin, or a combination thereof (all from Sigma-Aldrich). NAC was used as a 72-hour pretreatment at a 5 mmol/L concentration, followed by cisplatin treatment as indicated. To bypass glycolysis yet sustaining cell respiration, 4.5 g/L of galactose (Sigma-Aldrich) was added to glucose-deprived RPMI1640 (Sigma-Aldrich, catalog no. R1383) containing 2 mmol/L glutamax and supplemented with 10% FBS for SKOV-3/SKOV-3-R cells, or glucose-deprived DMEM (Sigma-Aldrich, catalog no. D5030) containing 2 mmol/L l-glutamine and supplemented with 10% FBS for COV-362/COV-362-R cells, as reported previously (19). For direct counting, cells were washed with PBS after treatment and stained with 0.5% crystal violet in a 10% ethanol solution for 30 to 60 minutes, and washed with water. After 24 hours of air-drying, 100 μL of methanol were added to each well and incubated for 20 minutes at room temperature on a bench rocker with a frequency of 20 oscillations per minute. Optical density was measured at 570 nm with a SpectraMax miniMax 300 imaging cytometer (Molecular Devices). Alternatively, cell number was determined using the SpectraMax minimax 300 automatic count on transmitted light images captured at indicated time points. For clonogenic assays, a range of 50 to 1,000 cells was seeded in 6-well plates and allowed to settle overnight. For every experiment, a control plate was seeded to obtain the plating efficiency (PE). After 48 hours of treatment with test compound(s), media were changed with fresh media. After colony formation, cells were fixed and stained with 0.5% crystal violet in a 10% ethanol solution for 30–60 minutes, washed with water, air dried, and counted. Results are expressed as surviving fraction (SF), where SF = #colonies/PE.

All in vivo experiments were conducted under approval of the Université catholique de Louvain (UCLouvain, Brussels, Belgium) authorities (Comité d'Ethique Facultaire pour l'Expérimentation Animale) according to national animal care regulations. Specific authorization was 2016/UCL/MD/018. Mice had access to water and food ad libitum. They were randomly assigned to a treatment group. In a first series of experiments (protocol 1), anesthetized (ketamine/xylazine) 8-week-old female Rj:NMRI-Foxn1^nu/nu nude mice (Elevage Janvier, RRID: IMSR_TAC:nmrinu) were subcutaneously injected with 10⁶ cancer cells (SKOV-3, SKOV-3-R, COV-362, or COV-362-R; one tumor per mouse) in PBS containing 10% growth factor–reduced (GFR) Matrigel (Corning, catalog no. L003975). From day +6 after tumor implantation, mice were intraperitoneally injected with cisplatin 3 mg/kg or vehicle every 3 days for seven times. Tumor size was measured every 2–3 days with an electronic caliper. After reaching a mean tumor volume of 300 mm³, mice were sacrificed by cervical dislocation under anesthesia and tumors were collected for further analyses. In a second series of experiments (protocol 2), mice were simultaneously injected with 1.5 × 10⁶ SKOV-3 and SKOV-3-R cancer cells in PBS containing 10% GFR-Matrigel in the left and right flanks, respectively. From day +18 after tumor implantation, mice were intraperitoneally injected with cisplatin 3 mg/kg or vehicle every 3 days. Tumor size was measured every 2–3 days with an electronic caliper. On day +29, mice were sacrificed by cervical dislocation under anesthesia and tumors were collected for further analyses.

Extracellular and intracellular glucose and lactate concentrations were measured using specific enzymatic assays on a CMA600 analyzer, as described previously (20). Glucose consumption and lactate production rates were calculated and normalized by total protein content obtained using the Bio-Rad protein assay. Oxygen consumption rates (OCR) were determined on a Seahorse XF96 bioenergetic analyzer using the XF Cell Mito Stress Kit (Agilent Technologies) according to the manufacturer's recommendations. Briefly, for experiments without concomitant cisplatin treatment, 10,000 cells per well were plated on XF96 culture plates 24 hours before experiments in RPMI1640 (SKOV-3 and SKOV-3-R) or DMEM (COV-362 and COV-362-R) medium containing 10% FBS. For experiments with cisplatin treatment, the protocol was similar except that 3,000 to 4,000 cells per well were plated 48 hours before experiments, which allowed to obtain the same final confluence than for cisplatin-untreated cells. On the day of analysis, culture media were replaced by DMEM containing 10 mmol/L glucose, 2 mmol/L glutamine, 1.85 g/L NaCl, 3 mg/L phenol red, pH 7.4. Cells were incubated for 1 hour in a CO₂-free incubator before analysis. Data were normalized to cell numbers measured right before oximetry using a SpectraMax miniMax 300 imaging cytometer. All other metabolic assays were performed on confluent cells. Intracellular ATP levels were measured using the CellTiter-Glo luminescent viability assay (Promega) on a Glomax 96 microplate luminometer (Promega) following manufacturer's instructions. Intracellular succinate was quantified using a succinate colorimetric assay kit (Sigma-Aldrich) following manufacturer's instructions.

Oxygen flow per cell was analyzed using an O2k-respirometer (Oroboros Instrument), according to manufacturer's instructions. Briefly, one million confluent cells per mL were analyzed over time, and the following drugs were sequentially used: oligomycin, CCCP (increasing doses from 100 nmol/L up to plateau), antimycin A, and rotenone. The O₂ flow was assessed after each treatment.

Cells cultured on glass coverslips were fixed in 4% formaldehyde, permeabilized with 0.1% Triton X-100 in PBS containing 0.1% Tween 20, and blocked with 5% BSA. Immunostaining was then performed as described previously (21). Primary antibodies were a rabbit polyclonal against mitochondrial import receptor subunit TOMM20 (Invitrogen, catalog no. PA5-52843, RRID: AB_2648808) and a mouse mAb against p62 (Santa Cruz Biotechnology, catalog no. sc-28359, RRID: AB_628279). Secondary antibodies were an Alexa Fluor 488–conjugated goat anti-rabbit (Invitrogen, catalog no. A-11034, RRID: AB_2576217) and an Alexa Fluor 594–conjugated goat anti-mouse (Thermo Fisher Scientific, catalog no. A-11005, RRID: AB_141372). Nuclei were stained with 4′,6-diamidino-2-phenylindole dihydrochloride (DAPI, 1 μg/mL, Sigma-Aldrich). Cells were counted in 28 different random fields. Images of mitochondria were captured by structured illumination fluorescence microscopy using an ApoTome-equipped AxioImager.z1 microscope (Zeiss). Mitochondrial surface and network analyses were performed using the ImageJ software (NIH, Bethesda, MD) following a previously described methodology with the MINA plugin (22). Colocalization of mitochondria and p62 was quantified using the AxioVision software (Zeiss).

mtDNA copy number was determined following the procedure detailed in ref. 23. Briefly, total DNA was isolated with a QIAmp DNA kit (Qiagen) following manufacturer's instructions. RT-qPCR was performed using TaqMan Universal Master Mix II with UNG (Applied Biosystems) on a ViiA 7417 Real-Time Instrument (Life Technologies). nDNA was quantified using the RNAseP VIC 2′-chloro-7′phenyl-1,4-dichloro-6-carboxy-fluorescein-labeled probe (Thermo Fisher Scientific). mtDNA was quantified using as primers: forward: 5′-GTA CCC ACG TAA AGA CGT TAG G-3′; reverse: 3′-TAC TGC TAA ATC CAC CTT CG-5′; and as labeled probe 5′-CCC ATG AGG TGG CAA GAA AT-3′ FAM 5(6)-carboxyfluorescein. mtDNA content was normalized to nDNA content, as described previously (24).

Cells at about 30% of confluence were synchronized overnight in RPMI1640 supplemented with 0.1% FBS, then cultivated in RPMI1640 supplemented with 10% FBS for a time equivalent to their doubling time. After synchronization, cells were treated with 10 μmol/L of cisplatin for 24 to 96 hours, trypsinized, centrifuged at 500 × g, and washed twice with PBS. Cells were then fixed by adding 700 μL of ice-cold ethanol 100% in 300 μL of cell suspension in PBS. Cells were then washed twice with 1 mL of Tris buffer with Triton X-100 0.2% v/v (TST), and finally resuspended in 300 μL of PBS with RNase (0.2 mg/mL) and propidium iodide (5 μg/mL). At least 20,000 to 30,000 events were recorded using a BD FACSCalibur flow cytometer. The FlowJo cell-cycle analysis tool (BD Biosciences) was used to interpret data according to DNA content. Polyploidy was analyzed using flow cytometry after propidium iodide staining according to nDNA content per cell.

Cisplatin–DNA adducts were quantified using the rat mAb CP9/19 targeting cisplatin-modified DNA (Abcam, catalog no. ab103261, RRID: AB_10715243) according to manufacturer's recommendations. Briefly, cells at 50% of confluence were cultured for 48 hours ± cisplatin, detached with trypsin 0.5% in PBS, fixed with ice-cold ethanol 100% in 300 μL of cell suspension in PBS, and permeabilized with Triton x-100 0.1% in PBS for 10 minutes on ice. They were then incubated for 18 hours with the anti-cisplatin–modified DNA antibody, washed with PBS, and incubated with Alexa Fluor 488 donkey anti-rat IgG secondary antibody (Thermo Fisher Scientific, catalog no. #A-21208, RRID: AB_2535794). The fluorescence intensity of FITC was measured using a BD FACSCanto II flow cytometer, and at least 20,000 events were recorded in each sample. Results were analyzed with FlowJo.

ROS levels were determined using CM-H₂DCFDA (Invitrogen) freshly prepared by dissolution in DMSO. Briefly, culture medium was removed, and 10,000,000 cells were washed with PBS for 10 minutes. Cells were resuspended in prewarmed PBS containing 1 μmol/L of CM-H₂DCFDA and incubated for 45 minutes. Cells were then washed with PBS, and incubated in complete medium for 15 minutes to allow the dye to respond to oxidation. H₂O₂ (50 μmol/L) was used as a positive control. Fluorescence intensity was measured using a BD FACSCanto II flow cytometer, and at least 10,000 events were recorded in each sample.

Western blotting was performed as described previously (25). Primary antibodies were a rabbit polyclonal against LC3 (MBL International, catalog no. PD014, RRID: AB_843283) and a mouse monoclonal against β-actin (Sigma-Aldrich, catalog no. A5441, RRID: AB_476744). Secondary antibodies were a HRP-coupled goat anti-rabbit (Jackson ImmunoResearch, catalog no. AB2307391, RRID: AB_2307391) and anti-mouse (Jackson ImmunoResearch, catalog no. 115-035-003, RRID: AB_10015289). Staining was revealed with an Amersham Imager 600 (GE Healthcare). Data were analyzed using the ImageJ software.

All data are expressed as means ± SEM. Note that error bars are sometimes smaller than symbols. n refers to the total number of replicates and N to the number of independent experiments per condition. Data were analyzed using GraphPad Prism 7.0. Survival curve fitting was performed using Matlab. Student t test, one-way ANOVA, and two-way ANOVA were used where appropriate. P < 0.05 was considered to be statistically significant.

To characterize metabolic changes associated to cisplatin chemoresistance in ovarian cancer, we selected SKOV-3 and COV-362 human ovarian adenocarcinoma cancer cells as a models. Cisplatin-resistant SKOV-3-R cells were previously generated by treating SKOV-3 cells with increasing doses of cisplatin (18). We applied the same selection protocol to generate cisplatin-resistant COV-362-R from COV-362 human ovarian epithelial endometroid cancer cells. After selection, cells were cultured for 3 to 6 weeks in the absence of cisplatin to ensure phenotypic stability.

To test their resistance, SKOV-3 and SKOV-3-R cells were challenged with increasing doses of cisplatin. As expected, direct cell count after 24 or 48 hours of treatment showed that SKOV-3-R were significantly more resistant than SKOV-3 cells to doses of cisplatin ranging from 10 μmol/L to 50 μmol/L (Fig. 1A). Doses above 4 μmol/L of cisplatin for 48 hours completely inhibited the clonogenicity of SKOV-3, whereas about 10% of SKOV-3-R cells were still clonogenic at 10 μmol/L of the drug (Fig. 1B, left). LC₅₀s were approximately 270 nmol/L and approximately 1.86 μmol/L of cisplatin for SKOV-3 and SKOV-3-R cells, respectively (Fig. 1B, right). Similarly, compared with COV-362, COV-362-R cells were more resistant to increasing doses of cisplatin (Fig. 1C) and more clonogenic (Fig. 1D, left). LC₅₀s were approximately 7 nmol/L and approximately 110 nmol/L of cisplatin for COV-362 and COV-362-R cells, respectively (Fig. 1D, right).

We next examined cell recovery. Cells were treated for 48 hours ± 10 μmol/L of cisplatin, washed, and cultured for increasing periods of time in drug-free medium. Although all cell types did recover in this assay, direct cell counting revealed that SKOV-3 were significantly slower than SKOV-3-R cells to repopulate (Fig. 1E, left), and COV-362 were significantly slower than COV-362-R cells to repopulate (Fig. 1E, right). Of important note, untreated SKOV-3 and SKOV-3-R cells proliferated at the same rate, indicating that chemoresistance to cisplatin did not result from an altered cell proliferation rate in this model. Untreated COV-362-R cells proliferated faster than COV-362 cells. Together, these first sets of data validated SKOV-3/SKOV-3-R and COV-362/COV-362-R as isogenic models to study the metabolic determinants of cisplatin chemoresistance in human ovarian cancer cells in vitro.

In mice, using protocol 1 (see Materials and Methods), SKOV-3 cells (10⁶ in GFR-Matrigel) generated fast-growing tumors (Fig. 1F, left). SKOV-3-R cells also generated fast-growing tumors, but tumor engraftment was delayed. Despite a trend for SKOV-3 (P = 0.085), tumors generated with either SKOV-3 or SKOV-3-R cells were insensitive to a clinically relevant treatment of 7 cycles of 3 mg/kg of cisplatin every 3 days (26), starting on day +6 postimplantation, as shown by tumor doubling times analyzed from the time when tumors reached 100 mm³ (Fig. 1F, right). Using the same implantation protocol, COV-362 and COV-362-R cells did not generate tumors in the lifetime of immunodeficient mice. We concluded that, compared with the in vitro situation, additional parameters influenced the response to cisplatin in vivo.

To address this concern, we repeated the experiment using protocol 2, where mice were simultaneously implanted with SKOV-3 and SKOV-3-R cells (1.5 × 10⁶ cells in GFR-Matrigel) in the left and right flanks, respectively. To avoid interfering with tumor implantation, the treatment was started on day +18 after implantation. Using this experimental protocol, SKOV-3 tumors were found to be sensitive to the clinically relevant treatment of 3 mg/kg of cisplatin every 3 days, as shown by tumor doubling times analyzed from the time when tumors reached 400 mm³ (Fig. 1G). Comparatively, SKOV-3-R cells did not generate tumors in the lifetime of the animals, irrespectively of treatment with cisplatin.

Both models were characterized metabolically using a variety of standardized biochemical assays. Compared with SKOV-3, SKOV-3-R cells consumed more glucose and produced more lactate, indicating accelerated glycolysis (i.e., glycolysis coupled to lactic fermentation), but the lactate/glucose ratio revealed no change in the glycolytic yield (Fig. 2A). SKOV-3-R cells also had a higher respiratory spare capacity, as revealed under uncoupled conditions (Fig. 2B). In the second model, COV-362-R had a similar rate of glucose consumption compared with COV-362 cells but produced less lactate, yet the lactate/glucose ratio was unaltered (P = 0.0671, Fig. 2C). COV-362-R cells also had an increased OCR and a higher respiratory spare capacity (Fig. 2D). These findings indicated that cisplatin-resistant human ovarian cancer cells have a better capacity to perform oxidative phosphorylation (OXPHOS) than matched cisplatin-sensitive cells.

To study the molecular mechanisms underpinning the observed oxidative switch, we further investigated OXPHOS. In the SKOV-3 model, enzymatic measurements revealed that the maximal activity of ETC Complex I was highly significantly (P < 0.001) increased in SKOV-3-R compared with SKOV-3 cells, while the activities of Complexes II, III, and IV were unchanged (Supplementary Fig. S1A, left). Accordingly, upon complex I inhibition with 500 nmol/L of rotenone, the OCR of SKOV-3-R was better preserved than that of SKOV-3 cells (Supplementary Fig. S1B). Conversely, succinate, which donates electrons to Complex II, was similarly abundant in SKOV-3-R compared with SKOV-3 cells (Supplementary Fig. S1C). These observations explain, at least in part, the increased respiratory spare capacity of SKOV-3-R versus SKOV-3 cells. Comparatively, in the COV-362 model, there were no significant changes in the activities of ETC complexes (Supplementary Fig. S1A, right), OCR sensitivity to rotenone (Supplementary Fig. S1B) and succinate levels (Supplementary Fig. S1C) when comparing sensitive to resistant cells. However, respiration coupled to OXPHOS (Supplementary Fig. S1D) and citrate synthase activity (Supplementary Fig. S1E) were increased in COV-362-R cells. These observations explain, at least in part, the increased OCR and respiratory spare capacity of COV-362 versus COV-362-R cells. Of note, succinate levels were higher in COV-362 and COV-362-R compared with SKOV-3 cells (Supplementary Fig. S1C), further supporting a better respiration coupled to OXPHOS in the former cells.

To evaluate whether the observed increased OXPHOS capacity of resistant cells could be mobilized upon treatment, we measured cell respiration following acute exposure to cisplatin. Cisplatin activated both basal respiration and the spare capacity of SKOV-3-R (Fig. 2E) and COV-362-R (Fig. 2F) cells, but at different concentrations (10 μmol/L and 5 μmol/L, respectively), which may reflect the different sensitivities of the cells to the drug (Fig. 1A–D). At 10 μmol/L of cisplatin, the OCR of COV-362 and COV3-62-R cells went back to that of the corresponding untreated cells, still remaining significantly higher in COV-362-R versus COV-362 cells (Fig. 2F). These findings suggested that cisplatin-resistant cells could potentially exploit OXPHOS to resist to the treatment.

One determinant of mitochondrial functions and OXPHOS capacity is organelle dynamics (27), so we examined mitochondrial morphology in our cell models.

For the SKOV-3/SKOV-3-R pair, representative pictures are shown in Fig. 3A, where mitochondria are labeled in green (TOMM20 staining) and nuclei in blue (DAPI). In untreated conditions, we detected no difference in the number of individual mitochondria (Fig. 3B), in the number of mitochondrial tubules (Fig. 3C) nor in the total mitochondrial area (Fig. 3D) per cell. Conversely, mtDNA normalized to nDNA content was decreased in SKOV-3-R compared with SKOV-3 cells (Fig. 3E), while nDNA content per cell was unchanged (Supplementary Fig. S2A). When present for 48 hours, cisplatin (10 μmol/L) significantly increased the number of individual mitochondria (Fig. 3B) as well as the mitochondrial surface (Fig. 3D) per cell in SKOV-3-R cells. It had no effects on SKOV-3 cells. Polyploidy was increased only in SKOV-3 cells (Supplementary Fig. S2B).

For the COV-362/COV-362-R pair, representative pictures are shown in Fig. 3F. In untreated conditions, COV-362-R had more individual mitochondria (Fig. 3G), more mitochondrial networks (Fig. 3H) and an increased mitochondrial surface (Fig. 3I) per cell compared with COV-362 cells. mtDNA normalized to nDNA content was unchanged (Fig. 3J). Cisplatin (10 μmol/L) incubated for 48 hours significantly increased the mitochondrial surface per cell in COV-362 cells (Fig. 3I), while the number of individual mitochondria (Fig. 3G) and the number of mitochondrial networks (Fig. 3H) were unchanged. The number of individual mitochondria (Fig. 3G) and the number of mitochondrial networks (Fig. 3H) were further increased by cisplatin in COV-362-R cells, while the mitochondrial surface did not change (Fig. 3I). mtDNA normalized to nDNA content was unchanged (Fig. 3J).

Together, these observations indicated that resistance to cisplatin was associated to fitter (i.e., less abundant yet more functional) mitochondria in our models, but for different reasons. Compared with matched sensitive cells, SKOV-3-R cells had more oxidative mitochondria due to increased Complex I activity (Supplementary Fig. S1A and S1B) while mitochondrial organization was largely preserved (Fig. 3A–E), and COV-362-R cells had more oxidative mitochondria due to an increased coupling between respiration and OXPHOS (Supplementary Fig. S1C–S1E), with more abundant mitochondria (Fig. 3F–J).

We then tested whether increased OXPHOS was a cause or a consequence of resistance to cisplatin. Inhibiting cell respiration with ETC Complex III inhibitor antimycin A (500 nmol/L) was more cytotoxic for SKOV-3-R than for SKOV-3 cells (Supplementary Fig. S3A). However, the opposite was observed in the COV-362/COV-362-R model (Supplementary Fig. S3B). Inhibiting OXPHOS with ATP synthase inhibitor oligomycin (1 μmol/L) in SKOV-3 and SKOV-3-R cells (Supplementary Fig. S3C, left) and in COV-362 and COV-362-R cells (Fig. 3D, left) was similarly effective in sensitive versus resistant cells, and resulted in an increased glycolytic rate reflected by an increased 2-deoxyglucose-sensitive (2-DG) extracellular acidification rate (ECAR; Supplementary Fig. S3C and S3D, middle left). Oligomycin alone was not significantly cytotoxic for any of the cell types (Supplementary Fig. S3C and S3D, middle right). However, it increased the cytotoxicity of cisplatin (10 μmol/L) in both sensitive and resistant cells (Supplementary Fig. S3C and S3D, right). Inhibiting OXPHOS can thus sensitize human ovarian cancer cells to cisplatin, but without discriminating sensitive and resistant cells.

We also tested a potential contribution of glycolysis to cisplatin chemoresistance. SKOV-3 and SKOV-3-R cells cultured in the presence of galactose instead of glucose (19) experienced a decreased 2-DG–sensitive ECAR (Supplementary Fig. S3E, middle), indicating a reduced glycolytic metabolism. OCR was slightly reduced in SKOV-3, but not in SKOV-3-R cells (Supplementary Fig. SE3, left). The treatment was cytostatic (Supplementary Fig. S3E, right). SKOV-3 cells kept their sensitivity to cisplatin (10 μmol/L) and SKOV-3-R cells were still resistant to the chemotherapy. Similar effects were seen in COV-362 and COV-362-R cells, except that galactose was less efficient than glucose to sustain the high basal OCR of COV-362-R cells (Supplementary Fig. S3F).

Because modulating ETC activity, OXPHOS, and glycolysis had no specific effects on cisplatin-resistant cells, the link between metabolism and resistance could be more subtle; the reason why we inspected the cell cycle, focusing on the SKOV-3/SKOV-3-R model. Exposure of the cells to cisplatin (10 μmol/L) induced an S-phase arrest in SKOV-3, but not in SKOV-3-R cells (Fig. 4A). This could be due to direct or indirect effects of cisplatin, or to altered DNA repair mechanisms.

Using specific antibodies to detect cisplatin adducts on DNA, we observed that cisplatin (10 μmol/L, 48 hours) induced similar DNA damage in SKOV-3 versus SKOV-3-R and in COV-362 versus COV-362-R cells (Fig. 4B), indicating that cisplatin equally reached its DNA target in sensitive and resistant cancer cells.

Cancer cell killing by cisplatin also depends on cisplatin-induced ROS production (6, 7). Interestingly, basal ROS production was higher in SKOV-3-R compared with SKOV-3 cells (Fig. 4C). We therefore envisioned that increased basal ROS levels could prime antioxidant defenses in resistant cells. However, challenging the cells with 5 mmol/L H₂O₂ for 48 hours did not spare SKOV-3-R compared with SKOV-3 cells (Fig. 4D). Conversely, pretreating the cells for 24 or 48 hours with 5 mmol/L of glutathione analogue NAC did not modulate the cytotoxicity of cisplatin (Fig. 4E). It increased rather than decreased the clonogenicity of cisplatin-treated SKOV-3, with no significant effects on SKOV-3-R cells (Fig. 4F). Overall, we concluded that resistance to cisplatin was not linked to an alteration of cisplatin-induced DNA damage nor to altered antioxidant defenses in human ovarian cancer cells.

Finally, we aimed to understand why cisplatin-resistant ovarian cancer cells had fitter mitochondria compared with their cisplatin-sensitive counterparts. We hypothesized that resistant cells could recycle mitochondria faster. Indeed, compared with SKOV-3, SKOV-3-R cells had a higher basal content in LC3-I (Fig. 5A), that is, the precursor form of LC3-II associated with both autophagy and mitophagy (28). Upon cisplatin treatment, the expression of both LC3-I and LC3-II was increased in SKOV-3-R, while it was unchanged in SKOV-3 cells, and the ratio LC3-II/LC3-I was increased as well (Fig. 5A).

To discriminate between halted versus increased autophagy and to test the biological significance of these observations, we blocked autophagy using bafilomycin A1 (10 nmol/L), and inspected the colocalization of mitochondria with p62, based on the fact that p62 interacts with mitochondria to address them to mitophagic degradation (29). Bafilomycin caused the accumulation of p62-labeled mitochondria in SKOV-3-R but not in SKOV-3 cells (Fig. 5B), and mtDNA accumulated (Supplementary Fig. S4). Similar results were found in COV-362-R compared with COV-362 cells (Fig. 5C), suggesting that cisplatin-resistant ovarian carcinoma cells could rely on mitophagy more heavily that their cisplatin-sensitive counterparts.

Autophagy can be increased in cisplatin-resistant ovarian cancers (16), and inhibiting autophagy is a promising anticancer strategy (30). Accordingly, we observed that low doses of bafilomycin A1 (5 to 10 nmol/L) highly significantly repressed SKOV-3-R (Fig. 6A) and COV-362-R (Fig. 6B) cell expansion in vitro. However, similar effects were seen in SKOV-3 and COV-362 cells (Fig. 6A and B). When used in combination, cisplatin (10 μmol/L) and bafilomycin A1 (10 nmol/L) had additive effects on the number of SKOV-3-R, but not on the number of SKOV-3 cells (Fig. 6C, left). Comparatively, additive effects of the two treatments were observed both in COV-362-R and in COV-362 cells (Fig. 6C, right). Interestingly, bafilomycin A1 inhibited the clonogenicity of SKOV-3 cells by approximately 90% and the clonogenicity of SKOV-3-R cells by approximately 99%, thus showing selectivity for cisplatin-resistant cells (Fig. 6D, left), which was confirmed for COV-362-R compared with COV-362 cells (Fig. 6D, right).

Together, these data indicated that autophagy/mitophagy supports the clonogenicity of human ovarian cancer cell, preferentially cisplatin-resistant cells. Our data thus support the future evaluation of mitophagy inhibition as a potential treatment for cisplatin-resistant ovarian cancers.

The main finding of this study is that acquired resistance to cisplatin by human ovarian cancer cells is associated to an oxidative switch relying on enhanced mitochondrial preservation after treatment. Consequently, targeting cell respiration (with antimycin A or oligomycin) or autophagy/mitophagy (with bafilomycin A1) were identified as efficient strategies to control the proliferation of cisplatin-resistant cells. Among these treatments, only autophagy/mitophagy inhibition was more efficient to reduce the clonogenicity of cisplatin-resistant compared with cisplatin-sensitive cells, thus showing some selectivity.

While many mechanisms (detailed in the introduction) can account for resistance to cisplatin in ovarian cancer, our main objective in this study was to identify metabolic changes associated to resistance and simultaneously present in two independent experimental cell models. In cisplatin-resistant cancer cells, we identified a common switch to a more oxidative metabolism, which was further stimulated by acute exposure to cisplatin. The switch proceeded through different molecular changes involving either an increased activity of ETC Complex I or improved coupling between the ETC and OXPHOS. Interestingly, SKOV-3-R cells also had an increased glycolytic metabolism (i.e., glycolysis coupled to lactic fermentation), but this was not shared by the second model. Bypassing glycolysis by replacing glucose by galactose (19) was cytostatic, but it did not chemosensitize cisplatin-resistant cancer cells, indicating that previously reported attempts to resensitize ovarian cancer cells to cisplatin via inhibition of glycolysis (31, 32) might be only applicable to a limited number of cases. This proposition is supported by Dar and colleagues (33) who analyzed 13 established and 12 patient-derived ovarian cancer cell lines, reporting high metabolic heterogeneity in general, but a preferential dependency on glycolysis for cisplatin-sensitive cells and on OXPHOS for cisplatin-resistant ones. Xu and colleagues (31) further reported a correlation between high Bcl2 expression and elevated OXPHOS in cisplatin-resistant SKOV3/DDP, another resistant variant of SKOV-3 cells. OCR was higher in resistant versus sensitive patient-derived xenografts. Consequently, metformin (inhibiting ETC Complex I; ref. 34), antimycin A (inhibiting ETC Complex III; our study) and oligomycin (inhibiting ATP synthase; our study) efficiently killed cisplatin-resistant human ovarian cancer cells. However, these treatments were not selective, as sensitive cancer cells died as well.

Although we here provide molecular explanations for the oxidative switch in resistant cells, the biological significance of the switch was obscure. High OXPHOS activities inevitably increase electron leak from the ETC, hence mitochondrial superoxide production, and mtROS can promote tumor aggressiveness by stimulating protumoral signaling pathways and/or by priming antioxidant defenses (mitohormesis; ref. 35). However, neither targeting ROS (with NAC) nor challenging antioxidant defenses (with H₂O₂) resensitized our resistant cells to cisplatin. Moreover, we found no change of cell proliferation that could have accounted for chemoresistance. However, in our models, the in vitro selection of resistance to cisplatin was associated to increased clonogenicity and delayed tumor take in mice. These characteristics are shared by ovarian cancer stem cells (CSC), which also have a higher rate of OXPHOS (36, 37) and are more resistant to cisplatin (38, 39) than non-CSCs. Whether treating cancer cells with increasing doses of cisplatin selects CSCs certainly warrants further investigation.

Interestingly, fitter mitochondria (i.e., more active yet less abundant) in cisplatin-resistant cells were also better preserved following acute treatment with cisplatin, with generally better networking and an increased surface compared with sensitive cells. Comparatively, cisplatin equally reached the nuclear compartment in both cases, producing similar nDNA damage in sensitive and resistant cells. Nevertheless, a S-phase cell-cycle blockade was bypassed by resistant cells, indicating a different DNA damage repair strategy in SKOV-3 versus SKOV-3-R cells.

On the basis of our data, we propose that cisplatin-resistant ovarian cancer cells have a faster mitochondrial turnover rate. This hypothesis is supported by the highly significant increase in p62-labeled mitochondria upon autophagy blockade in our resistant cells, suggesting elevated mitophagy, while mtDNA content was preserved, suggesting elevated mitochondrial biogenesis. Comparatively, sensitive cells did not show this profile. Others reported increased mitochondrial biogenesis and fusion in cisplatin-treated resistant ovarian cancer cells (40–43). For treatment, a further discrimination between autophagy and mitophagy was not necessary, as several previous studies correlated cisplatin resistance in ovarian cancer cells with a high rate of autophagy (32, 44–48), and bafilomycin A1 almost equally efficiently killed both cisplatin-sensitive and cisplatin-resistant ovarian cancer cells. However, clonogenic assays revealed a more intense inhibitory effect of bafilomycin A1 on cisplatin-resistant cells, which we believe could be linked to an enrichment in stem cell features in our selection protocol. Accordingly, an increased autophagic rate is a characteristic of ovarian CSCs (48, 49), calling for a future examination of the contribution of mitochondrial turnover to the oxidative profile of CSCs in this type of cancer.

A limitation of our study is that the chemosensitive/chemoresistant phenotypes that we selected in vitro were partially lost in vivo. Indeed, if in a first series of experiments the increased time for tumor take in mice for SKOV-3-R compared with SKOV-3 cells was an evident difference between the two variants, tumors turned out to be equivalently insensitive to a clinically relevant regiment of cisplatin (7 cycles of 3 mg/kg of cisplatin every 3 days; ref. 26). While tumors formed from resistant cells remained resistant, those formed from sensitive cells were resistant as well. It suggested that, compared with the in vitro situation where culture parameters are well controlled, several pathways supporting cisplatin resistance are potentially activated under the influence of the in vivo tumor microenvironment (e.g., suboptimal tumor perfusion, hypoxia, acidosis). In this context, it is interesting to note that hypoxia stimulates autophagy in ovarian cancers in mice (50, 51), which calls for a systematic evaluation of interventions targeting autophagy/mitophagy in ovarian cancer models in mice. While these considerations stay valid, a second series of experiments showed that, if the treatment was started later after tumor implantation, SKOV-3 cells retained sensitivity to cisplatin in vivo, indicating that cisplatin could also interfere with the early stages of tumor development in this model.

In summary, our study used matched cisplatin-sensitive and cisplatin-resistant human ovarian cancer cells that were metabolically compared. Our data support that resistant cells undergo metabolic reprogramming to an ovarian CSC-like phenotype characterized by increased OXPHOS, mitochondrial turnover and autophagy. While cisplatin-resistant cells were sensitive to OXPHOS and autophagy/mitophagy inhibition, further studies are needed to firmly establish their CSC nature.

L. Zampieri reports receiving grants from the European Union during the conduct of the study. D. Grasso reports receiving grants from the European Union, from Belgian Télévie, and grants from UCLouvain during the conduct of the study. P. Sonveaux reports receiving grants from the European Union, from Belgian Fonds National de la Recherche Scientifique (F.R.S.-FNRS), from Belgian Télévie, and grants from UCLouvain during the conduct of the study. No potential conflicts of interest were disclosed by the other authors.

L.X. Zampieri: Conceptualization, data curation, formal analysis, validation, investigation, visualization, writing-original draft, writing-review and editing. D. Grasso: Data curation, formal analysis, investigation, writing-original draft, writing-review and editing. C. Bouzin: Resources, formal analysis, supervision, visualization, methodology, writing-review and editing. D. Brusa: Resources, formal analysis, supervision, visualization, methodology, writing-review and editing. R. Rossignol: Conceptualization, resources, supervision, validation, methodology, writing-review and editing. P. Sonveaux: Conceptualization, resources, software, formal analysis, supervision, funding acquisition, validation, methodology, writing-original draft, project administration, writing-review and editing.

This work was supported by European Union's Horizon 2020 research innovation program under the Marie Skłodowska-Curie grant agreements no. 722605 TRANSMIT and no. 642623 RADIATE, the Belgian Fonds National de la Recherche Scientifique (F.R.S.-FNRS, CDR J.0135.18), the Belgian Télévie (project no. 7.4617.17), and the UCLouvain Bourses du Patrimoine. L. Zampieri is a PhD fellow of Marie Skłodowska-Curie grant no. 722605 TRANSMIT. D. Grasso is a PhD fellow of Marie Skłodowska-Curie grant no. 642623 RADIATE, a PhD fellow of the Belgian Télévie, and was supported by an UCLouvain Bourses du Patrimoine. P. Sonveaux is a F.R.S.-FNRS senior research associate. The authors thank Maria Shoshan (Karolinska Institute, Stockholm, Sweden) for the kind gift of SKOV-3 and SKOV-3-R cells; Thibaut Vazeille, Loïc Hamelin, Marie Bedin, and Elodie Dumon for excellent technical assistance; and Vincent Haufroid at the Louvain Center for Toxicology and Applied Pharmacology (LTAP, UCLouvain).

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.