Combination of oxidative phosphorylation and platelet-derived growth factor inhibitors for the treatment of ovarian cancers

Abstract

The combination therapy of FDA approved oxidative phosphorylation inhibitors (metformin and atovaquone) and platelet derived growth factor inhibitors (sunitinib and sorafenib) could target ovarian cancer stem-like cells (CSCs) and carcinoma associated mesenchymal stem cell enrichment of CSCs while reducing off-target effects. Results of a 48-hour drug exposure ovarian cancer-MSC tumoroid model showed additive effects with atovaquone combined with sunitinib or sorafenib being most effective.

Introduction

Despite aggressive surgical debulking and first-line platinum-based therapy, high grade serous ovarian cancer (HGSC) patients relapse at a ~ 70% rate, attributable to chemoresistance and peritoneal metastasis¹. Chemoresistance and new tumor growth following primary therapy has been ascribed to ovarian cancer stem-like cells (CSCs), through a complex interplay with other cells within the tumor microenvironment (TME)^1–6. Stromal cells within the TME including carcinoma-associated MSCs (CA-MSCs) play critical supportive roles in CSC proliferation, metastasis and chemoresistance⁷. This interaction is mediated by a complex cytokine network involving IL-6, SDF-1, TGF-β, BMP2, BMP4 and CCL5^8–12. Indeed, co-culturing ovarian cancer cells with MSCs increases the proliferation, invasiveness, and platinum resistance of the malignant cells via growth factor signaling such as PDGF/RTK pathways^8,13–15. Given CA-MSCs’ role in enriching and activating CSCs, simultaneous targeting of both CA-MSC and CSC cell populations may provide a more effective strategy to reduce ovarian cancer recurrence.

Although ovarian cancer cells upregulate glycolysis to support proliferation, oxidative phosphorylation (OXPHOS) remains crucial for CSCs, tumor initiation, and acquisition of chemotherapy resistance^16–18. OXPHOS is an essential energy generating process in nutrient-starved regions of tumors which do not have abundant glucose for glycolysis¹⁹. CSCs have been observed to have increased mitochondrial activity and an increased reliance on OXPHOS compared to other tumor cells¹⁹. Inhibition of OXPHOS can be employed to kill CSCs by inhibiting components of the electron transport chain: mitochondrial complex I inhibitor (with metformin) and mitochondrial complex III inhibitor (with atovaquone).

Concurrently, PDGF signaling is active in stromal MSC²⁰ and has been linked to the formation and maintenance of glioma, gastric cancer, lung cancer and breast cancer among others^21–26. Downstream of the PDGF-BB/PDGFR interaction, several mechanistic pathways have been shown to increase tumorigenicity, metastasis and CSC phenotypes, including epithelial to mesenchymal transition and the Hedgehog signaling pathway^27–30. Importantly, the multitargeted tyrosine kinase inhibitor, sunitinib, has been explored in randomized phase II settings, demonstrating moderate single agent efficacy in targeting stromal components in ovarian and other cancers. Therefore, the PDGF pathway is a potential target in ovarian cancers with poor prognosis.

Monotherapies targeting either OXPHOS or receptor tyrosine kinase (RTK) signaling have shown limited clinical benefit in ovarian cancer, likely due to compensatory mechanisms within CSCs and their stromal niche⁸. Thus, simultaneous inhibition of OXPHOS (to target CSC metabolism) and RTK signaling (to disrupt supportive stroma) attacks both critical arms of ovarian tumor persistence, offering a promising strategy to overcome resistance and achieve durable therapeutic responses Fig. 1. This combination may also decrease drug-based toxicity in vivo due to reduced dosages required for the therapeutic window. Therefore, in this work, the combined targeting of OXPHOS and PDGF signaling was investigated in OVCAR3/MSC tumoroids (Fig. 1). HGSC cell line (OVCAR3) and adipose-derived primary human MSC were combined in hanging drop arrays to self-assemble as OVCAR3/MSC tumoroids. To account for faster proliferation of OVCAR3 than MSC, tumoroids were generated with a starting ratio of 1 cancer cell to 3 MSC (Fig. 2A). Tumoroids self-assembled by day 3 and increased in size through day 7 (Fig. 2C).

Fig. 1. Combination therapy of oxidative phosphorylation inhibitors and platelet derived growth factor inhibitors.

Ovarian cancer stem cells (CSCs) have an increased reliance on oxidative phosphorylation (OXPHOS) because it can be utilized to generate ATP in a nutrient-scarce environment with little glucose. Metformin and atovaquone inhibit mitochondrial complexes I and III, respectively, which are vital components of the electron transport chain necessary for ATP generation via OXPHOS. Additionally, stromal mesenchymal stem cells (MSC) enrich for CSC via PDGF signaling which activates CSC supporting pathways such as epithelial to mesenchymal transition and Hedgehog in the cancer cells. Sunitinib and sorafenib are receptor tyrosine kinase inhibitors that block the PDGF receptor to interfere with the stromal-mediated signaling to promote cancer cell stemness. By combining both OXPHOS inhibitors with PDGF inhibitors, CSC are targeted via two distinct pathways, which we hypothesize will result in synergistic CSC killing activity.

Fig. 2. Ovarian cancer cells and mesenchymal stem cells self-assemble into three-dimensional tumoroids.

a Generation of multi-cellular tumoroids composed of OVCAR3 and human adipose derived mesenchymal stem cells (MSC) with 1:3 ratio (OVCAR3/MSC tumoroids). b Representative confocal image of high grade serous ovarian cancer tumoroids with OVCAR3 (green, GFP) and MSC (red, mCherry). c Representative bright field images of tumoroids i) 3, ii) 5, and iii) 7 days after initiating the suspension growth.

On day 5, tumoroids were treated with an OXPHOS inhibitor (OXPHOSi)—metformin (MET) or atovaquone (ATQ)—or a PDGF inhibitor (PDGFi)—sunitinib (SUN) or sorafenib (SOR)—or a combination of the two drug classes at various concentrations. The final drug concentrations of MET added to the tumoroids was 5.00 mM, 1.08 mM, 0.23 mM and 0.05 mM. The final drug concentrations of ATQ, SOR, and SUN added to the tumoroids was 60.0 µM, 3.30 µM, 0.18 µM, and 0.01 µM.

By day 7, OVCAR3 were the predominant cell type in the tumoroid as visualized by confocal imaging of GFP-labeled OVCAR3 and smURFP-labeled MSC (Fig. 2B). The viability in response to single agents or combination of two agents (one OXPHOSi with one PDGFi) was measured using Calcein AM intensity, normalized to vehicle control tumoroids (Fig. 3). Maximum doses were chosen based on solubility limits in the aqueous cell culture medium environment that tumoroids were treated in. Notably, all four drugs have low solubility in aqueous solution, which prevented complete IC₅₀ curves from being generated for single agents. The lack in understanding of the entire dynamic range of the dose-response curve may limit clinical applications and necessitate more safety studies. However, the solubility limitations of these drugs will persist in aqueous clinical applications, which may restrict their dose range. At the highest achievable concentrations, the most effective single agents were SUN and ATQ, with average normalized viability of 52–66% and 60–67%, respectively (Fig. 3). SOR was not far behind with 57–74% viability (Fig. 3). However, MET showed very little effect on ovarian cancer viability as a single agent, with 94–114% viability (Fig. 3).

Fig. 3. Viability response of tumoroids to single agent and combination treatment of OXPHOS inhibitors and PDGF inhibitors.

Heat map graphs of the viability response of tumoroids containing OVCAR3 and human adipose derived mesenchymal stem cells (MSCs) when exposed to drug combinations of (a) atovaquone (ATQ) and sorafenib (SOR), (b) ATQ and sunitinib (SUN), (c) metformin (MET) and SOR, and (d) MET and SUN. The tumoroids were exposed to the drugs for 48 hours and then viability was measured by adding Calcein AM and measuring fluorescence intensity on a plate reader. Percentage of viability was calculated by normalizing the drug-free control tumoroids. Analysis was performed on 20 tumoroids/biological replicate from 3−5 biological replicates. Two-way ANOVA was performed to determine statistical significance between conditions (Supplementary Fig. 1).

Given the moderate efficacy of these single agents observed previously^31–36 and the limited dose range due to solubility, the synergy between the OXPHOSi and PDGFi classes of drugs was explored. All four combinations of OXPHOSi with PDGFi resulted in a lower viability compared to either single agent (Fig. 3). The most effective combinations were achieved using the highest concentration of each single agent. ATQ in combination with SUN (36% viability) or with SOR (45% viability) showed the greatest reduction in viability. Interestingly, the most effective combination of MET + SOR was not at the highest concentrations of each drug (53% viability 'Fig. 3'), but rather at the highest concentration of SOR with a moderate concentration of MET (0.23 mM) (38%). MET in combination with SUN was the least effective (89% viability).

To calculate the synergy between the two drugs in each combination, viability data was input into three different synergy models: Bliss, Loewe, and MuSyC using the web applications Synergy Finder 3.0³⁷ and MuSyC³⁸. The Bliss model predicted all four combinations to be additive, with ATQ + SUN having the highest score of 2.29 ± 5.16 (Table 1). The Loewe model predicted 3 of the 4 combinations to be additive, with MET + SUN being antagonistic. The highest Loewe score of 6.29 ± 7.56 was received by ATQ + SOR. The MuSyC model results differed most from the Bliss and Loewe models, with ATQ + SUN and MET + SUN predicted to have increased efficacy and ATQ + SUN and MET + SOR predicted to have decreased efficacy (Table 1). Across the three models, the only drug combination that received consistent predictions was ATQ + SOR which was predicted to be additive based on Bliss and Loewe and have increased efficacy based on MuSyC. This combination was also one of the most effective, with 41% viability when treated with the highest concentrations of both atovaquone and sorafenib (Fig. 2).

Table 1.

Drug synergy quantification of the four combinations of the OXPHOS and PDGF drugs using Synergy Finder 3.0 for the Bliss and Loewe synergy scores and the multi-dimensional synergy of combinations (MuSyC) framework based on the Calcein AM viability assay data

Drug combination	Bliss synergy score	Loewe synergy score	MuSyC efficacy (beta [CI])
ATQ + SOR	1.26 ± 7.56	6.29 ± 7.56	0.31 [0.06, 62.8]
ATQ + SUN	2.29 ± 5.16	3.08 ± 5.16	−8.06 [−5.58, 7.31]
MET + SOR	2.17 ± 10.4	3.57 ± 10.4	−0.58 [−0.78, 0.35]
MET + SUN	−5.21 ± 3.22	−16.7 ± 3.22	99.2 [65.4, 3298]

For the Bliss and Loewe models, interactions between two drugs are considered antagonistic if the synergy score is < −10, additive from −10 to 10, and synergistic if < 10. In the MuSyC model, beta (b) is defined as the percent increase in the effect of the combination over the most efficacious single agent based on the fitted E_max. A beta (b) value > 0 indicates there is increased efficacy between the two drugs.

While the combination of OXPHOSi and PDGFi did not show synergistic effects in cancer cell killing as predicted, they did show additive effects for the most part. The incongruencies between synergy models may be explained in part by the incomplete IC₅₀ curves due to solubility limitations and relatively high viability as the maximum response. The MuSyC model relies more on this information, which may explain why it differed from the other models while Bliss and Loewe were more consistent with each other. Additionally, both OXPHOS and PDGF signaling are enhanced in CSC, thus more potent effects may be observed in a tumoroid composed entirely of CSC^16–18. However, most tumors are composed of a mix of cancer cell types, so this alternative would be less clinically applicable (Fig. 3).

Despite metformin providing improved progression-free survival at 5 years for both diabetic and nondiabetic patients, clinical trials are still ongoing, and metformin does not show efficacy in all ovarian cancer patients³⁹. Additionally, only moderate efficacy has been observed in phase I and II trials with RTK inhibitors sunitinib and sonidegib single agents^31–36,40. However, there are a few preclinical and clinical studies demonstrating benefit with combinations of OXPHOS and RTK inhibition in various cancer models. Evans et al. showed that combining OXPHOS inhibition with the multi-kinase inhibitor cabozantinib delayed resistance in triple-negative breast cancer⁴¹. Similarly, synergy between metformin and sorafenib was observed in preclinical models of hepatocellular carcinoma^42–44. This combination was investigated in a clinical trial (NCT02101736), but its results have not yet been published⁴⁵. A few other clinical trials have tested combining metformin with other targeted agents (e.g., mTOR inhibitors, EGFR inhibitors), but not RTK (PDGFR/VEGFR) inhibitors directly^46,47. Strikingly, triple negative breast cancer was effectively treated with albumin nanoparticles co-loaded with atovaquone and sorafenib⁴⁸. Therefore, combining OXPHOSi and PDGFi could be an attractive combination therapy approach in future clinical trials with agents that are highly bioavailable.

Taken together, inhibition of OXPHOS-dependent CSC metabolism and inhibition of stromal cell-mediated PDGF signaling have an additive effect when used in combination and are less effective as single therapies in the OVCAR3/MSC tumoroids. Atovaquone combined with either sunitinib or sorafenib provided the best results, but further in vivo studies are needed for validation. While not synergistic, this combination therapy strategy could be useful in providing more cancer cell death with less toxicity from each individual agent. Additionally, this work highlights the challenges of targeted compound solubility and underscores the need for methods to improve solubility, bioavailability and co-delivery of OXPHOSi and PDGFi.

Methods

Tumoroid generation

OVCAR3 (National Cancer Institute (NCI) Division of Cancer Treatment and Diagnosis (DCTD) Tumor Repository) and adipose-derived mesenchymal stem cells (MSCs) (Lonza Walkersville Inc., MD) were maintained in 2D culture plates in RPMI 1640 medium with 10% FBS and 1% antibiotic-antimycotic or Adipose Derived Stem Cell Basal Medium with 10% FBS, 1% L-glutamine, and 1% antibiotic-antimycotic, respectively. OVCAR3 and MSC were collected from flasks using 0.25% trypsin and 0.05% trypsin, respectively, and combined in a ratio of 1:3 (OVCAR3:MSC) in phenol red-free complete RPMI 1640 with a total cell concentration of 5,000 cells/mL. This 1:3 ratio has been used previously to achieve a final concentration between 5−20% MSC in tumoroids by day 5, which is consistent with the abundance in ascites of approximately 1−11%^12,49. Then 20 µL of the cell suspension was added to each well in a 384-well hanging drop plate.

Drug stock solutions and dilutions

Drug stock solutions of sunitinib (9 mM) and sorafenib (30 mM) were made in dimethyl sulfoxide (DMSO). Stock solutions of metformin (500 mM) were made in UltraPure™ DNase/RNase-Free Distilled Water (Invitrogen), and stock solutions of atovaquone (20 mM) were made in tetrahydrofuran (THF) (Sigma-Aldrich). All drugs were purchased from MedChemExpress. On day 5 after the tumoroids were plated, drug dilutions were made in cell culture medium to achieve 20X the final concentration experienced by the tumoroids. For each combination, 1 µL of each 20X drug solution was added to the wells. For combinations where that included 0 µM/mM concentration, vehicle controls (medium + DMSO/TFH/UltraPure™ Dnase/Rnase-Free Distilled water) were added to the wells.

Viability assay

The viability assay was performed on day 7 after the tumoroids were plated. Calcein AM (Invitrogen) was diluted to 32 µM in 1X PBS, and 2 µL was added to each well. The plates were incubated for 1 hour at 37 °C and then the fluorescence intensity (excitation/emission 495/517 nm) were measured using a BioTek Synergy HT microplate reader. Analysis was performed on 20 tumoroids/biological replicate from 3-5 biological replicates. Two-way ANOVA was performed to determine statistical significance between conditions

Imaging

Brightfield and epifluorescence microscopy images were taken on an Olympus IX83 microscope a calibrated phase contrast microscope (Olympus IX81, Japan equipped with ORCA R2 Cooled CCD camera and CellSens software). Overall, 3-5 tumoroids were imaged for each drug dose.

Synergy calculations

Viability data from the Calcein AM viability assay were compiled and analyzed using Synergy Finder 3.0. The curve fitting algorithm for both the Bliss and Loewe models was a four-parameter logistic regression (LL4). The viability data was also input into MuSyC.

Author contributions

All authors edited and approved the final version of the manuscript. The author contributions following CREDIT guidelines are as follows: Conceptualization, G.M.; Data curation, K.B., T.R.; Formal analysis, K.B., T.R.; Funding acquisition, G.M.; Investigation, K.B., T.R.; Methodology, K.B., T.R., G.M.; Project administration, G.M.; Resources, G.M.; Supervision, G.M.; Validation, K.B., T.R.; Writing—original draft, K.B., T.R., G.M.; Writing—review and editing, K.B., T.R., G.M.

Data availability

The data that support the findings of this study are available from Deep Blue Data (https://deepblue.lib.umich.edu/data).

Code availability

No new codes were generated in this work.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s44294-026-00134-x.