Low-GPX4 drives a sustained drug-tolerant persister state in TNBC by a targetable adaptive FSP1 upregulation

Abstract

Metastatic relapses in Triple-Negative Breast Cancer (TNBC) patients with residual disease pose a significant clinical challenge. In this study, we longitudinally modelled cellular state transition from dormant drug-tolerant persister (DDTP) to proliferative (PDTP) cell state across TNBC subtypes. We identified specific molecular and phenotypic alterations that characterize the DTP states in TNBC cells that are maintained upon re-gaining proliferation. We found that Basal-Like proliferative DTPs stably acquired mesenchymal traits, while luminal androgen receptor-positive TNBC DTPs undergo partial Epithelial-to-Mesenchymal Transition (EMT). TNBC DTP cells exhibit reduced expression of glutathione peroxidase-4 (GPX4), conferring susceptibility to ferroptosis inducers. Mechanistically, GPX4 downregulation promotes EMT in TNBC, supported by an inverse correlation between GPX4 and EMT marker vimentin (VIM) expression that also serves as a predictor of survival in TNBC patients undergoing chemotherapy. The genetic, pharmacological, or chemotherapy-induced suppression of GPX4 in TNBC cells leads to robust upregulation of ferroptosis suppressor protein-1 (FSP1). The clinical significance of these findings is established by a strong predictive value of FSP1^high/VIM^high signature for worst survival and incomplete pathological response in chemotherapy-treated TNBC patients. Further, targeting FSP1 re-sensitizes cells to chemotherapy, while combined inhibition of FSP1 and GPX4 is selectively lethal in proliferative DTP TNBC cells by inducing ferroptosis.

Keywords: Combination therapy; Drug-tolerant persister cells; EMT; FSP1; GPX4; TNBC.

Graphical abstract

Fig. 1

Longitudinal modelling of drug tolerance to generate proliferative persisters, PDTP cells from different TNBC subtypes. (A) Schematic representation of the experimental flow for developing TNBC PDTP from dormant DTP (DDTP) cells, upon chemotherapy treatment of parental cells, PC. TNBC cell line representatives from different molecular subtypes MDA-MB-468, HCC70, HS578T, and MDA-MB-453, were treated with chemotherapy (IC_80-95) cytotoxic doses in vitro. The proliferation of residual DDTP tumor cells was observed after culturing them without drugs. (B) Phase contrast micrographs (20× magnification) of the TNBC cell lines in different time-defined longitudinal cellular phases during the development of proliferative drug tolerant persister from each subtype (colour bars in left represent subtypes: light blue-BL-1, dark blue- BL-2, green-MSL and orange-LAR) using paclitaxel (0.5–1 μM) treatment as a representative chemotherapy agent. Scale bar, 10 μm (C) Proliferation rates of TNBC cell lines under different phases of development of PDTPs are calculated by counting the cells in each time point shown and represented as folds of parental cells, where the value is defined as 1 (colour key, green to red depicts cell number from 100 to 0 %). (D) Drug tolerance represented as folds cell viability of untreated PC on a scale of 0–1, was calculated by the percent of cells surviving the persister deriving paclitaxel doses at each phase assuming that untreated PC is with minimal tolerance (colour key, gradient of blue to yellow depicts 0–100 % tolerance). (E) Tolerance to paclitaxel in different TNBC cell lines represented as Folds of paclitaxel IC₅₀ in TNBC cell lines. (F) Cell aspect ratio was calculated as described under the material and method section in the MDA-468, HCC70, and MDA-453 cell lines at different phases of DTP development. (G) Dose-response curves of varying concentrations of paclitaxel in the PC, Paclitaxel derived DDTP and PDTP upon 72 h treatment of Paclitaxel in the indicated cell lines after 72 h of treatment presented as means of three replicates. (H) Colony seeding potential detected by colony formation assay in PC and PDTP-P of respective TNBC cell lines. (I) Tumor volume growth curves of MDA-468 PC and PDTP-P orthotropic xenograft mice, ∗∗P < 0.005 (Wilcoxon test). (J) Tumor weight at the end of the experiment from MDA-468 PC and PDTP-P orthotropic xenograft mice, ∗P < 0.01 (t-test). (K) Representative images of tumors developed from MDA-468 PC and PDTP-P cells in mice excised after day 65 are shown (scale, 1 cm). (L) Representative confocal micrographs (63x) and 3x zoomed inserts of immunofluorescence staining of Ki67 (red) expression and DAPI nuclear staining (blue) in the tumor tissues sections developed from MDA-468 PC and PDTP-P. Scale bar, 50 μm. (M) Schematic representation summarizing the development and characterization of DDTP and PDTP from PC is shown in Fig. 1.

Fig. 2

Proliferative chemotherapy-tolerant persisters attain mesenchymal phenotype with increased migration and invasion properties. (A) TEM images of intercellular spaces revealed loss of cell-cell contact in PDTP cells and intact tight junctions between cells in PC (indicated in zoomed inserts). Scale bars, 2, 5, or 10 μm, as shown. Quantification using ImageJ software of the cell area of individual PC and PDTP cells, with 30 cells analysed per group. (B) Representative confocal microscopy images of PC and PDTP cells after immunofluorescence staining for analyzing the expression and localization of E-cadherin and vimentin. Green fluorescence indicates E-cadherin and red fluorescence indicates Vimentin. Scale bar: 10 μm. (C) Bar graphs representing the quantification of mean fluorescent intensity of PC and PDTP cells for E-cadherin and Vimentin immunofluorescence staining, as shown in B, with statistical significance of p < 0.001. (D) Protein expression of E-cadherin and vimentin in PC and PDTP-P as assessed by Western blot analysis in 4 TNBC subtypes, α-Tubulin protein expression was used as the loading control. ImageJ software was used for quantification of Western blot band intensities of PC and PDTP-P and presented as a heat map of fold changes of PC shown in the right panel. (E) Representative confocal micrographs (63x) and 3x zoomed inserts of immunofluorescence staining of E-Cadherin (green), Vimentin (red) expression and DAPI nuclear staining (blue) from tumors developed from MDA-468 PC and PDTP-P and its quantitation was done using ImageJ Software in F. Scale bar, 50 μm. (G) Brightfield microscopy images (20× magnification) of cell migration through scratch wound healing assays in TNBC PC cells and PDTP-P were recorded at 0 and 12 h. The red dotted lines indicate the migration front of the cells. (H) Bar graphs representing the migration rate and the percentage of wound closure in PC and PDTP-P TNBC cells calculated as indicated in the material and method section. (I) Brightfield microscopy images showing transwell matrigel invasion assay in PDTP-P and PC TNBC cells. The fold difference in invasion capacity compared to PC is presented in heatmap. (J) Schematic representation summarizing main results of phenotypic changes in PDTP TNBC cells shown in Fig. 2.

Fig. 3

TNBC PDTPs are more sensitive to ferroptosis induction due to elevated mitochondrial damage, lipid ROS, labile iron, coupled with glutathione depletion as a result of GPX4 pathway suppression. (A) Representative TEM micrographs and their zoomed inserts indicating change specific to mitochondrial structure in PDTP-P, PDTP-D, and PDTP-C as compared to PC in the indicated TNBC cell lines. Scale bar- 2 μm, as indicated. (B) Violin plots showing the number of damaged mitochondria in TNBC PC and PDTP cells from TEM micrograph analysis. (C) Bar graph showing H₂O_2-induced cellular ROS by using H₂DCFDA assay in live cells in TNBC PCs and their PDTP-P, represented as folds of untreated cells as a control. (D) Representative images show Oil red O staining of MDA-MB-468 PC and PDTP-P. Scale bar, 10 μm. The maximum intensity of Oil red staining (from ∼35 different microscopic fields) is quantified using ImageJ and plotted. (E) Representative fluorescence microscopy images (20× magnification) acquired by Incucyte live-cell analysis system to detect lipid peroxidation after exposure to Erastin (1 μM) for 4 h, followed by incubation with lipid probe C11 BODIPY 581/591 in PC, PDTP-P, and PDTP-C of the indicated TNBC cell lines. Fluorescence channels showing non-oxidized (red) oxidized (green) C11 probe intensities and their overlay. Scale bar, 50 μm. (F) FACS analysis of C11 BODIPY stained MDA-468 PC and MDA-468 PDTP cells after treatment with GPX4-in-3 (1 nM, 3h) and GPX4-in-5 (0.75 nM, 3h) with and without Lip-1 to assay lipid peroxidation. (G) Graphs showing the percentage of cells gated for PE⁺/FITC⁺ in each treatment from at least 3 independent sets of experiments. (H) Graphs showing glutathione levels (relative luminance units, RLU) in indicated PC and PDTPs cells in untreated and Erastin (1 μM, 4h), treated conditions using GSH-Glo Glutathione assay kit as described under the method section. (I) Bar graphs showing cellular levels (nmol) of ferrous, ferric, and total iron in PC and PDTPs after Erastin treatment (1 μM, 4 h) using the iron assay kit. Results represent the mean of 3 replicates. (J) Labile iron (Fe²⁺) in MDA-468 PC and PDTP was analysed using FerroOrange Live cell dye in untreated and GPX4-in-3 treated (1 μM, 3h) cells. Scale bar, 10 μm Fluorescence intensities were quantified using ImageJ and plotted in the adjacent graph. (K) The protein expression levels of GPX4 and NRF2 in parental TNBC cell lines and PDTP-P were analysed by WB analysis. ImageJ software was used to quantify the band intensities and fold change compared to PC from replicates, which is presented as a heatmap. (L) Representative confocal micrographs (63x) and 3x zoomed inserts of immunofluorescence staining of GPX4 (green) expression and DAPI nuclear staining (blue) from tumors developed from MDA-468 PC and PDTP-P, fluorescence intensity of GPX4 staining quantified by ImageJ software and presented in a violin plot. Scale bar, 50 μm. (M) MTT assay was performed to determine the viability of indicated TNBC cell lines and PDTP-P and PDTP-D as compared to PC treated with RSL3, FIN56, and Erastin. Percent cell death was calculated in each case and presented as folds of PC in the heatmap as the mean from triplicate experiments. (N) Dose-response curves of GPX4-in-3 and GPX4-in-5 in PC and PDTP-P cells are shown with fold change (FC of PC) of IC₅₀. (O) Schematic representation of the proposed model for GPX4, GSH inhibition, and an increase in lipid ROS leading to ferroptosis vulnerability in PDTP cells is depicted.

Fig. 4

Modulation of GPX4 activity and expression in TNBC cells revealed its role in EMT phenotype as well as in induction of compensatory FSP1 expression. (A) TNBC cells from different subtypes were treated with RSL3 (1–3 μM) for 72h to select RSL3-tolerant cells and were grown till recovery of proliferative colonies (15–20 days) of PDTP-RSL3 cells. Phase contrast images of morphological changes in PDTP-P and PDTP-RSL3 as compared to PC of indicated TNBC cell lines. Magnification: 10X, scale bar, 10 μm (B) Representative immunofluorescence staining of PDTP-P, PDTP-RSL3 and PC of MDA-MB-468 cell lines to assess the expression and localization of E-Cadherin, vimentin, GPX4 and Actin (in green). Nuclear staining (Blue) with DAPI was performed. Scale bar, 10 μm. (C) Fluorescence intensity of the different proteins shown in B was quantified using ImageJ Software and presented as mean ± SD in the bar graphs with statistical significance between PC, PDTP-P and PDTP-RSL3. (D) The protein expression levels of E-Cadherin, vimentin, GPX4 and housekeeping protein in PDTP-P, PDTP-RSL3 and PC in TNBC cell lines MDA -MB-468 and HS578T were analysed using western blots and ImageJ software was used to quantify the band intensities of proteins and the results were presented as fold changes compared to PC in a heatmap from 3 independent experiments. (E) Representative confocal images of immunofluorescence staining of GPX4, E-Cadherin (green) and vimentin (red) in MDA-MB-468 PDTP-P and PC in GPX4 knockdown or GPX4 overexpression conditions. Scale bar, 10 μm (F) Fluorescence intensity of different proteins stained in E were quantified using ImageJ Software. Expression of the respective proteins, as indicated in GPX4 knockdown and overexpression conditions in PDTP-P and PC MDA-MB-468, was compared to control vectors. (G) Western blot analysis of FSP1 expression in RSL3-PDTP (H), in 48h-treated cells with GPX4 inhibitors RSL3 and FIN56, 0.25 μM concentration (I), and in shRNA mediated GPX4 knockdown TNBC cells. Densitometric analysis of WB in control cells, RSL3-PDTP or shGPX4 cells shown in the right panel as folds of PC, UT (untreated) or shRNA control cells. (J) The proposed model for GPX4 expression and its effect on EMT, and upregulation of FSP1 expression is summarized using the schematic.

Fig. 5

Low GPX4, high VIM, and FSP1, are strong indicators of the worst outcome in chemotherapy-treated TNBC patients with high-grade and lymph node metastasis. (A) Kaplan-Meier Plot representing relapse-free survival and distant metastasis-free survival in LN + ve, high-grade TNBC chemotherapy-treated patients with high or low gene expression of GPX4. (B) Relapse-free survival in LN + ve TNBC chemotherapy-treated patients (upper) and in LN + ve, high-grade TNBC chemotherapy-treated patients (lower) with high or low gene expression of VIM. (C) Overall survival of in LN + ve, chemotherapy-treated TNBC patients (upper) and all breast cancers (lower) with high and low AIFM2 gene expression. ROC curve analysis of the AIFM2-VIM-gene signature (E) for pathological response to chemotherapy in LN + high-grade TNBC patients (D) as performed using ROCPlotter, https://www.rocplot.org/. (F) Volcano plot showing Differentially Expressed Genes analysed from gene expression Affymetrix dataset (GSE6434) using GEOR on GEO having breast cancer patient cohorts with docetaxel sensitive and resistant tumors. Significantly different and relevant genes were marked on the plot showing fold change and p-value (Log2 FC and Log10 p value). Distribution of GPX4 gene expression between resistant and sensitive tumors is plotted in the dot plot below. Epithelial score was calculated as described under the method section for the docetaxel-resistant and sensitive tumors and plotted in the bar graph. (G) Scatter plot showing correlation analysis of GPX4 expression with the GSS, EMT signature genes, and IREB2 expression in Basal tumors using TCGA RNASeq Dataset. (H) Scatter plot showing correlation analysis between GPX4 and EMT signature genes using singe-cell RNA sequencing dataset form TNBC patients. The black line represents linear regression; the grey area indicates the limits of the confidence intervals. (I) Correlation analysis of GPX4 with GSS and EMT score in patient-derived xenograft models in breast cancer. (J) Confocal micrographs showing representative three-channel overlap of immunofluorescence co-staining of GPX4 (green) and vimentin (red) proteins in paraffinized human breast cancer tissue sections, DAPI (blue) was used for nuclear staining. Areas marked under white squares were shown as zoomed inserts in the right panel where white arrows indicate cells showing co-expression of GPX4 and vimentin. Magnification, 63x and scale bar, 50 μm. (K) Waterfall bar graph showing the Mander's overlap coefficients (MOC) for relative contribution of green (GPX4) and red (vimentin) channel signals to the co-localized immunofluorescence signals in 54 human breast tissues. Spearman's R correlation between the groups was used to compute the correlation between MOC K1 and K2 and p value was determined by two-tailed t-test. (L) Dot plots (left) showing analysed colocalization coefficient for relative co-expression of GPX4 (Green channel) and vimentin (red channel) in human breast tissue used for immunofluorescence co-staining. Each dot in the plot represents an individual breast tissue categorised under normal breast tissue (NBT, n = 8), primary tumor (PT, n = 36) and metastasis tumor (MT, n = 10). Dot plot (right) showing the colocalization coefficient for GPX4 and vimentin co-expression in tissue sections of human breast cancer (n = 7) and its matched normal tissue (n = 7) after immunofluorescence co-staining and analysis. Unpaired two-tailed t-test was used to analyze the significant difference between two groups and p values are indicated on the graphs.

Fig. 6

Chemotherapy-induced sustained FSP1 upregulation in TNBC is a therapeutic target that synergizes with chemotherapy and GPX4 inhibitor RSL3. (A) Protein expression of FSP1 in PC and PDTP-P as assessed by Western blot analysis in 4 TNBC subtypes after 48 h of chemotherapy treatment, α-tubulin protein expression was used as loading control. ImageJ software was used to quantify the band intensities, and the results were presented as fold changes compared to PC in a heatmap, based on three independent experiments. (B) Representative confocal micrographs and 3x zoomed inserts showing immunofluorescence staining of FSP1 (red) in PC and PDTP-P in different TNBC cell lines. Scale bar: 10 μm. Dot plot representing the quantification and statistical analysis of mean fluorescence intensity for FSP1 immunofluorescence staining in PC and PDTP cells (C) Western blot analysis FSP1 in parental TNBC cell lines with different chemotherapeutic agents (Pac, Paclitaxel; Doxo, Doxorubicin; Cis, Cisplatin; CPA, Cyclophosphamide; and 5FU, 5- Fluorouracil calculated doses corresponding to IC₇₅) after 48h of treatment and with paclitaxel (1 μM) with different timepoints. α tubulin served as a loading control. ImageJ software was used for quantification of Western blot band intensities of PC and PDTP-P in C presented as heatmap below the WB as folds of PC. (D) Representative confocal micrographs (63x) of immunofluorescence staining of FSP1 (red) expression and DAPI nuclear staining (blue) MDA-468 PC treated with chemotherapy agents for 48h. Scale bar, 10 μm Dot plot representing the quantification and statistical analysis of mean fluorescence intensity for FSP1 immunofluorescence staining in MDA 468 PC cells treated with chemotherapy agents. (E) Western blot analysis FSP1 in parental TNBC cell lines after 24 and 48h of treatment with paclitaxel. Alpha-Tubulin and GAPDH served as a loading control. ImageJ software was used to quantify the band intensities, and the results were presented as fold changes compared to UT(Untreated) in a heatmap (F) Representative confocal micrographs (63x) and 3x zoomed inserts of immunofluorescence staining of FSP1 (green) expression and DAPI nuclear staining (blue) from tumors developed from MDA-468 PC and PDTP-P. Scale bar, 50 μm. The mean fluorescence intensity of FSP1 IF staining was plotted in the dot plot (right). (G) MTT assay was performed to check the viability of mouse MMTV-R26^Met TNBC tumor-derived cell lines in response to the indicated increasing doses of chemotherapy treatment for 48h, data presented as mean percent cell viability in the heatmap. (H) Western blot analysis FSP1 in MGT mouse TNBC cell lines treated with 1 μM of docetaxel or doxorubicin, 10 μM cisplatin, or 5FU for 12h. ImageJ quantification of Western blot band intensities in H is presented below as a heatmap below the WB as folds of untreated (UT) control cells. Heatmap showing percent cell viability (I) and crystal violet staining showing colony formation assay (J) of MDA-468 PC, PDTP-P, and in PDTP-P cells with shRNA mediated FSP1 knockdown after 72h of chemotherapy treatment at different doses. (K) Effect of iFSP1 on chemotherapy sensitivity in MDA-468 PC upon 48h of different chemotherapy treatments shown as percent cell viability in the heatmaps. (L) Effect of iFSP1 pre-treatment in MDA-468 PDTP-P on chemosensitivity presented as heatmaps of percent cell viability using MTT assays. (M) Dose-response matrix of RSL3 and iFSP1 combination therapy in MDA-468 PC and PDTP-P shown as percent cell viability as determined by MTT assay. The table below shows the Combination Index (CI) values calculated using CompuSyn software for the effective dose combinations indicating drug synergy in MDA-468 PC. (N) PC and PDTP were treated with iFSP1 and RSL3 and the level of lipid peroxidation levels were determined using C11 BODIPY in IncuCyte. Scale bar, 10 μm Bar graph representing red/green ratio in PC and PDTP cells along with statistical analysis. (O) MTT assay was performed in PDTP by treating them with RSL3, iFSP1 or both with or without liproxstatin1. Bright field images of the same were acquired. Scale bar, 50 μm. (P) Kaplan–Meier curve showing overall survival of breast cancer patients with high AIFM2 mRNA from the METABRIC dataset (n = 995) bifurcated on the basis of low GPX4 and high GPX4 expression (median cut-off) is shown within this group. Log-rank test p-value (p) is shown. (Q) Schematic representation to summarize data shown in Fig. 6 that suggests FSP1 upregulation in TNBC post-paclitaxel treatment and in PDTP renders them vulnerable to iFSP1 and a combination of iFSP1 with GPX4 inhibitor in a synergistic manner.

Fig. 7

Schematic model summarizing the main findings of the study. (A) Derivation of PDTP from different subtypes of TNBC. (B) Identification of common targetable vulnerabilities of TNBC PDTPs (C) Regulation of PDTP phenotypes by GPX4, its prognostic and therapeutic potential in combination with FSP1 in TNBC.