Erastin, (2-[1-[4-[2-(4-Chlorophenoxy) acetyl]-1-piperazinyl]ethyl]-3-(2-ethoxyphenyl)-4(3H)-quinazolinone) was obtained from Cayman Chemicals (Ann Arbor, MI) and was dissolved in DMSO. Stock solutions were stored at −80C. Fresh drug solutions, prepared from the stock solutions, were used in all experiments.

Authenticated human ovarian OVCAR-8 and ADR-selected OVCAR-8 cells (Batist et al., 1986; Cowan et al., 1986) (NCI/ADR-RES cells) were obtained from the Division of Cancer Treatment and Diagnosis Tumor Repository, National Cancer Institute the NCI-Frederick Cancer Center (Frederick, MD, United States). The NCI/ADR-RES cells express MDR1 and have higher activities of SOD, Catalase, and GSH-dependent peroxidase1, and transferase (Batist et al., 1986; Cowan et al., 1986). However, ER was similarly cytotoxic to both cells (IC₅₀ 1.2 ± 0.10 × 10^-6 M vs. 0.8 ± 0.15 × 10^-6 M, for OVCAR-8 and NCI/ADR-RES cells, respectively (Sinha et al., 2024)). Cells were grown in Phenol Red-free RPMI 1640 media (pH = 7.0) supplemented with 10% fetal bovine serum and antibiotics. Cell cultures were incubated in an incubator at 37°C and 5% CO2 with saturating humidity. Cells were routinely used for 20–25 passages, after which the cells were discarded, and a new cell culture was started from the frozen stock.

Cells were treated with ER (2.5 µM) for 24 h, washed once with ice-cold PBS (pH 7.4) and collected in methanol by scraping. Dried cell lysates were stored at −80°C until analysis and then resuspended in 80 µL of 98:2 water/acetonitrile. System blanks comprised of the same resuspension solvent and a pooled quality control (QC) sample comprised of 15 µL aliquots of each sample were generated.

Samples were analyzed on a Vanquish Horizon ultra-high performance liquid chromatography system (UPLC, Thermo Scientific) coupled to an Orbitrap Lumos Tribrid high-resolution mass spectrometer (HRMS, Thermo Scientific). An EASY-Max NG ionization source was operated in the heated-electrospray ionization configuration. The source parameters were as follows: spray voltage of +4,000 V in positive ionization mode or −3,000 V in negative ionization mode, sheath gas of 50 arbitrary units (arb), auxiliary gas of 10 arb, sweep gas of 1 arb, ion transfer tube at 325°C, vaporizer at 350°C. Prior to measurement, the mass spectrometer was calibrated using FlexMix (Thermo Scientific). EASY-IC (Thermo Scientific) fluoranthene was used during data collection as a lock mass.

Chromatographic separation was carried out on a Kinetex F5 analytical column (2.1 inner diameter, 100 mm length, 100 Å, 2.6 µm particle size, Phenomenex) with corresponding guard cartridge at 30°C. Gradient elution was performed using HPLC-grade water with 0.1% acetic acid (A) and HPLC-grade acetonitrile with 0.1% acetic acid (B). Separation was performed as follows: 0% B from 0–2.0 min, 0%–100% B from 2.0 to 10.5 min, 100% B from 10.5 to 12.0 min, 100%–0% B from 12.0 to 13.0 min, 0% B from 13.0 to 20.0 min at a flow rate of 500 μL/min.

Prior to acquiring liquid chromatography-mass spectrometry (LC-MS) sample data, liquid chromatography-tandem mass spectrometry (LC-MS/MS) data were acquired using the AcquireX deep scan methodology. For AcquireX acquisition, LC-MS data were collected for a system blank (exclusion list) and pooled QC (inclusion list), then LC-MS/MS data were collected over seven QC injections. MS data were acquired at 120,000 resolution from m/z 100-1,000 with an RF lens of 60% and maximum injection time of 50 m. MS/MS data were acquired at 30,000 resolution using an isolation width of m/z 1.5, stepped assisted higher-energy collision induced dissociation was used with energy steps of 20, 35, and 60 normalized collision energy, and a maximum injection time of 54 ms. An intensity filter was applied with an intensity threshold of 2E4. Dynamic exclusion was used with the following parameters: exclude after n = 3 times; if occurs within 15 s; exclusion duration of 6 s; a low mass tolerance of 5 ppm mass error; a high mass tolerance of 5 ppm mass error; and excluding isotopes.

LC-MS sample data were collected in positive and negative ionization mode from individual samples, system blanks, and pooled QCs. Samples and system blanks were injected at 4 µL. The pooled QC was injected at different volumes (triplicate injections at 2 μL, 4 μL, and 6 µL) for signal response evaluation (Overdahl et al., 2023). MS data were acquired at 120,000 resolution from m/z 100-1,000 with an RF lens of 60% and maximum injection time of 50 ms.

Data files (.raw) were processed with Compound Discoverer 3.3.3.220 (ThermoFisher Scientific) to identify molecular features and, where possible, annotate them with chemical names. Features were annotated in CompoundDiscoverer based on MS/MS spectral matching in alignment with the Metabolomics Standards Initiative (Fiehn et al., 2007). Public, commercial and in-house MS/MS spectral libraries including NIST 2020, GNPS (access 04-01-2022), mzCloud (offline, endogenous metabolites) were utilized. An output table containing feature descriptors (e.g., m/z and retention time), annotation information, and peak areas were further processed using in-house R code via Jupyter Notebook for data formatting and analysis. Initial processing steps included formatting the output table from Compound Discoverer and assignment of MSI levels of confidence. MSI Level 2 features have an MS/MS match to either a database entry or an in-house library entry. MSI Level 4 and Level 5 features have either an MS/MS match but no database match, or no MS/MS match. Some MSI Level 2 features were manually promoted to MSI Level 1 after matching in-house m/z and retention time data acquired on the same analytical platform using identical conditions.

Subsequent processing steps included outlier removal, feature quality filtering via assessment of pooled QC signal response and dispersion ratio (i.e., comparison of signal variance in pooled QC replicate injections versus samples), data normalization, and multi- and univariate statistics. Sample outliers were identified and removed based on their total signal (i.e., sum of peak area) being greater than 1.5x the interquartile range. The signal response evaluation approach utilized here is detailed in Overdahl et al. (Overdahl et al., 2023) In short, data for each feature collected from the pooled QCs at three volumes was evaluated for a corresponding increase in signal. In this case, a Pearson correlation with p = 0.05 was used as the cutoff for significance, therefore only features displaying positive correlations within the p = 0.05 parameter were retained for further analysis. Feature abundances were row sum normalized to the total signal of all features for a given sample to account for differences in input material.

Statistical analyses were also performed using R via Jupyter Notebook. Data were first examined using unsupervised multivariate statistics via principal component analysis (PCA). Data were first log₁₀-transformed to reduce the weighting of abundant features in the PCA computation. PCA score plots (PC1– PC3) and their corresponding loading plots were generated for both positive and negative mode data. Mid-level data fusion was utilized to combine the principal components obtained from PCA of positive and negative mode ionization mode data to give a singular visual output based on comprehensive metabolomics information. PCA plots were annotated according to genotype and treatment group in order to examine variance between sample groups as well as sample variance among those within a single group. Two-way ANOVA was used to detect significant differences followed by a post-hoc Tukey’s Honest Significance Difference (HSD) test for identifying pairwise differences while adjusting for multiple hypothesis testing. Log₂ fold changes (FC) were also calculated for each pairwise comparisons. Features with an adjusted p-value <0.05 and a log₂FC > 1 or < -1 were considered significantly statistically different. Volcano plots were generated for each comparison as well as box plots of peak areas for all statistically significant features.

Determination of changes in plasma membrane potential was carried out using DiBAC4. Briefly, 30 min prior to time of examination, DiBAC₄(3) (Invitrogen; Molecular Probes) was added to 1-mL of cells to a final concentration of 150 ng/mL and incubation was continued at 37°C, 7% CO₂ atmosphere. Flow cytometric analysis was carried out using a BD Fortessa flow cytometer and FACSDiVA software (Becton Dickinson Immunocytometry Systems, San Jose, CA). Single cells were gated on a FCS-area versus width dot plot. DiBAC₄(3) was excited at 488 nm and detected at 530 nm. For each sample, 10,000 cells were examined on a DiBAC₄(3) histogram to determine the changes in mean fluorescent intensity (MFI). An increase or decrease in DiBAC₄(3) fluorescence shows cellular depolarization or hyperpolarization, respectively.

The metabolomic profiles of OVCAR-8 (WT) and NCI/ADR-RES (R) cells were explored using principal component analysis (PCA), Figure 1 and Supplementary Figure S1. This approach allowed for the overall metabolomics profiles to be compared between samples using unsupervised multivariate statistics. Untreated (blue) WT and R groups were visually and statistically different in the positive ionization mode, Figure 1A, (PERMANOVA, F = 4.9, p = 0.005) as well as the negative ionization mode, Figure 1B, (F = 5.5, p = 0.006). Erastin treatment (red) resulted in the visual and statistical differentiation from untreated cells in the positive (F = 20, p = 0.001) and negative ionization mode (F = 24.2, p = 0.001) data as well as increased the visual separation between WT and R cells, implying differential response. Given the clear changes in metabolomic profiles between WT, R, and ER-treated cells, we investigated the individual metabolites which contributed to the observed profiles differences.

Redox balance (i.e., balance of oxidants and antioxidants) is crucial in the cellular maintenance of reactive oxygen species (ROS). Both reduced form glutathione (GSH) and oxidized form glutathione disulfide (GSSG) are crucial cellular metabolites involved in redox balance. Metabolomics analysis indicated different levels of GSH and GSSG between WT and R cells (Figures 2A, B). Further, the ratio of GSH/GSSG, a metric of oxidative stress (Kemp et al., 2008; Chen et al., 2024) was found to be differential in the untreated condition (p-value = 0.03, fold change = 2.6) between WT and R cells (Figure 2C). Levels of GSH and GSSG as well as the GSH:GSSG ratio significantly decreased in WT and R cells in response to ER treatment.

Ophthalmic acid (OPH), formed via reaction of ɣ-glutamylcysteine and glutathione synthase, has been suggested to be a biomarker of oxidative stress in cells and tissues following depletion of cellular glutathione (Soga et al., 2006). Since ER treatment significantly decreased GSH in both OVCAR-8 and NCI/ADR-RES cells, we examined the formation of OPH in these cells. Untargeted metabolomics indicated a significant increase in OPH (Figure 2D) in ER-treated WT (p-value = 4.3 × 10⁻⁸, fold change = 18) and R (p-value = 1.0 × 10⁻¹⁰, fold change = 34) cells. Intriguingly, OPH levels were greater in ER-treated R cells.

Our metabolomic profile also indicated that ER treatment concomitantly decreased taurine formation in both cells. Taurine has been reported to be a natural antioxidant and biomarker of oxidative stress in cells (Surai et al., 2021; Baliou et al., 2021). Similar to GSH and GSSG, taurine was found to be significantly lower (p-value = 9.5 × 10⁻⁶) in untreated R cells compared to WT cells. Taurine levels significantly decreased in both WT and R cells in response to ER treatment (Figure 2E).

Nicotinamide adenine dinucleotide (NAD⁺) and nicotinamide adenine dinucleotide phosphate (NADP⁺) are essential for maintaining cellular redox homeostasis and for modulating numerous biological events, including cellular metabolism (Biniecka et al., 2023; McKay-Corkum et al., 2023). NAD⁺ was significantly depleted in ER-treated WT and R cells (Figure 3A). Intriguingly, levels of NAD⁺ in untreated WT cells were significantly lower (p-value = 0.009) compared to R cells. NADP⁺ was significantly depleted in ER-treated WT cells (p-value = 0.03) but not ER-treated R cells (Figure 3B).

Furthermore, depletion of NAD+/NADH has been shown to leads to membrane depolarization, mitophagy and cell death via apoptosis or ferroptosis (McKay-Corkum et al., 2023; Zhu et al., 2022). Therefore, we examined effects of ER on cellular stability utilizing DiBAC4(3), a dye specific for studying plasma membrane polarization/depolarization in cells. Results presented in Figure 3C clearly shows that ER induced significant polarization of both cells within 4 h of treatment. Moreover, NCI/ADR-RES (R) cells were more sensitive to this depolarization than OVCAR-8 cells (WT).

Key energy metabolism pathways such as the tricarboxylic acid (TCA) cycle and glutaminolysis have been implicated in ferroptosis (Liu et al., 2023; Martinez-Reyes and Chandel, 2020). Untargeted metabolomics analysis indicated a significant increase in citric and isocitric acid in ER-treated WT and R cells (Figure 4A). Other TCA metabolites were differential following ER treatment in either WT or R cells (Figures 4B–D), such as α-ketoglutaric acid (p-value = 4.0 × 10⁻⁴) and malic acid (p-value = 1.7 × 10⁻⁵) which increased in ER-treated R cells. Conversely, succinic acid increased only in ER-treated WT cells (p-value = 7.3 × 10⁻⁵).

Glutamine serves as an important source of carbon to replenish the TCA cycle, as its degradation via glutaminolysis produces glutamate and subsequently the TCA cycle metabolite α-ketoglutarate (Figure 4E) as well as glutathione (Jin et al., 2023). Untargeted metabolomics analysis revealed that both glutamic acid and glutamine levels significantly decreased in WT and R cells in response to ER treatment (Figures 4F,G).

Primary functions of L-carnitines are to transport activated long chain fatty acids (long chain fatty acyl-CoAs) into the mitochondria for degradation by β-oxidation and generation of ATP (Virmani and Cirulli, 2022; Farahzadi et al., 2023). As our previous studies in colon cancer cell lines had shown significant modulation of L-carnitines during drug-induced ferroptosis (Sinha et al., 2023), we sought to examine effects of ER on carnitines in ovarian cells. Metabolomics analysis indicated decreased levels of all identified L-carnitines (Figures 5A–E) in response to ER treatment. L-carnitine and two of its acyl analogues 2-methylbutyryl-L-carnitine and iso/butyryl-L-carnitine decreased in both WT and R cells, however this change was smaller for 2-methylbutyryl-L-carnitine and iso/butyryl-L-carnitine in R cells as the levels in untreated R cells were lower than untreated WT cells (p-values = 4.9 × 10⁻⁶ and 1.9 × 10⁻⁴, respectively). Acetyl-DL-carnitine levels were significantly higher in untreated R cells than WT cells (p-value = 5.8 × 10⁻¹²), whereas hexanoyl-L-carnitine was significantly higher in untreated WT cells than R (p-value = 6.6 × 10⁻⁹).

Ferroptosis results from the peroxidation of polyunsaturated fatty acids (PUFAs) in the cell membrane, such as arachidonic acid. PUFAs are continuously oxidized and reduced under normal cellular conditions; however, the depletion of GSH can lead to an accumulation of oxidized lipids, ultimately leading to cell membrane damage and cell death. Interestingly, arachidonic acid levels were similar in WT, WT-ER, and R cells, whereas R-ER was found to have a slight increase (p-value = 0.03, fold-change = 1.3, Figure 5F).