Animal ethics statement

Animal husbandry and experiments were approved by the committee of ethics of animal experiments of the state of Bavaria (Regierung Unterfranken, Würzburg; Regierung von Unterfranken, Würzburg; TS-30-2021, 55.2DMS2532-2-1832) and performed according to the European Animal Welfare laws and guidelines. Mice were kept on a 12-h light–dark cycle in individually ventilated cages at a constant temperature between 20–24 °C and 45–65% humidity and provided with food and water ad libitum in the animal facilities of the Friedrich-Alexander University of Erlangen-Nürnberg.

Cell lines and cell culture

MDA-MB-231, MCF7, H358, A549, BxPC3, 143B and U2OS cells were purchased from the American Type Culture Collection. Generation of various mouse KPC cell lines from KPC tumours is previously described15. They were cultured under standard conditions at 37 °C and 5% CO2 in DMEM (Gibco) supplemented with 10% foetal bovine serum (Gibco) and regularly tested for Mycoplasma contamination. Generation of 143BshZeb1 clones is previously described55. MDA-MB-231 shCtrl and shZeb1 cells56 were cultivated in the presence of puromycin (1 µg ml−1) for 7 days every 1 month to maintain stable transfection. CRISPR–Cas9-mediated knockout of EMT-TFs in KPC cells was carried out as previously described57. In brief, the sgRNAs targeting Zeb1 exon 2 (5′-GACCAGACAGTATTACCAGG-3′), Snai1 exon 1 (5′-GAGCTGCAGGACGCGTGTGT-3′) and Twist1 exon 1 (5′-CGGGAGCCCGCAGTCGTACG-3′) were cloned into pX459 (Addgene, 62988) and transiently transfected into KPC661 with Lipofectamine 3000, followed by selection with 4 µg ml−1 puromycin for 3 days and clonal expansion using FACS. Sequence-verified clones with biallelic indel mutations and protein loss were used. To induce a transient knockdown, cells were transfected with Silencer select siRNAs (Ambion; s229970 for siZeb1, 4390844 for siCtrl) at a final concentration of 50 nM, using Lipofectamine RNAiMAX transfection reagent (Thermo Fisher, 13778) according to the manufacturer’s instructions and cells were treated with indicated ferroptosis inducers 48 h after transfection.

H358 and BxPC3 drug-resistant cell lines were previously described17,58 and routinely maintained in 10 µM erlotinib or 40 nM gemcitabine, respectively. Doxorubicin-resistant U2OS cells (U2OS-Dox) were established by continuous treatment of parental cells (at 80% confluence) by stepwise increasing the concentration of doxorubicin (3.75–500 nM; Szabo Scandic) every 2 weeks over 4 months and maintained in 500 nM doxorubicin.

For TGFβ treatment, the medium was supplemented daily with 5 ng ml−1 TGFβ1 (PeproTech) for the indicated amount of time as specified in figure legends. As TGFβ1 was dissolved in a citric acid solution, the medium of control cells was supplemented with citric acid to a concentration of 500 nM.

Chemicals

3-(4,5-Dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide, erlotinib, gemcitabine and arachidonic acid were obtained from Sigma-Aldrich. CAY10566, erastin, etoposide, ferrostatin-1, ML210, (1S,3R)-RSL3, sc-26196 and oleic acid were obtained from Cayman Chemical. MF438 and ferrostatin-1 were from Med Chem Express. 1,2-Dimyristoyl-sn-glycero-3-phosphatidylcholine (DMPC), 1-pentadecanoyl-2-oleoyl(d7)-sn-glycero-3-phosphocholine (PC(15:0/18:1-d7)), 1,2-dimyristoyl-sn-glycero-3-phosphatidylethanolamine (DMPE), 1-pentadecanoyl-2-oleoyl(d7)-sn-glycero-3-phosphoethanolamine (PE(15:0/18:1-d7)), 1-pentadecanoyl-2-oleoyl(d7)-sn-glycero-3-phosphoinositol (PI(15:0/18:1-d7)), PE(16:0/20:4) and oxPC(16:0/20:4) (oxPAPC) were obtained from Avanti Polar Lipids, dissolved in chloroform, aliquoted and stored under argon and protected from light at −80 °C.

Plasmids

The generation of the LLGL2 and ANKRD1 luciferase reporter plasmids was previously described28,56. The promoter luciferase reporter plasmids were generated by amplifying the FADS2 promoter (−619 to +233 rel. to transcription start site (TSS)), the ACSL4 (−532 to +1008 rel. to TSS), the FASN promoter (−1305 to +196 rel. to TSS) and the SCD promoter (−976 to +69 rel. to TSS) from genomic DNA by PCR. The restriction sites XhoI and BglII were incorporated into the primers (Supplementary Table 1) and the amplicons were inserted into pGL4.10 (E6651, Promega). For the ELOVL5 luciferase reporter vector, the intronic region chr6: 53,326,815–53,325,856 (hg38) was amplified from genomic DNA. Restriction enzyme sites XhoI and BglII were incorporated into the primers and the amplicon was inserted into pGL4.23 (E8411, Promega). pCIneo-hZEB1 was a gift from M. M. Sanders (University of Minnesota).

Luciferase reporter assay

MCF7 cells were seeded in 24-well plates in triplicate at 20% density. The next day, they were transfected with the FuGENE HD transfection reagent (Promega, E2311) according to the manufacturer’s instructions, using 100 ng firefly luciferase reporter vector and 30 ng pRL-TK Renilla luciferase control reporter vector (Promega, E2241) together with 100 ng ZEB1 expression vector or the corresponding empty control vector, respectively. Cells were collected after 72 h and lysed in passive lysis buffer (Promega, E1941). Luciferase activity was measured using the Dual-Luciferase Reporter Assay system and a CentroXS3 LB 960 Luminometer (Berthold). Values of the firefly luciferase were normalized to their corresponding Renilla values, serving as a transfection control.

Chromatin profiling

ATAC-seq and ChIP–seq for ZEB1 were performed as described previously28. ChIP–seq for H3K27ac (rabbit anti-trimethyl histone H3K27ac, Millipore 07-449) was performed accordingly using MDA-MB-231 shCtrl or shZEB1 cells, except that the EGS crosslinking step was omitted. Crosslinking with 1% formaldehyde was performed for 5 min directly on the plate in cell growth medium.

Western blot analysis

For the analysis of whole cell protein, cells at 50–70% density were lysed in ice-cold lysis buffer (150 mM NaCl, 50 mM Tris-HCl, pH 8.0, 0.5% sodium deoxycholate (w/v), 0.1% SDS (v/v), 1% NP40 (v/v), 1 mM PMSF, 1× complete protease inhibitor cocktail (Roche, 04693132001) and 1× PhosphoStop (Roche, 4906837001)). Protein concentration was determined using the BCA Protein Assay (Thermo Fisher Scientific, 23225) according to the manufacturer’s instructions. Protein samples were separated by SDS–PAGE, followed by wet blot transfer onto nitrocellulose membranes (Roth, 4685.1). Primary antibodies were applied overnight at 4 °C and secondary antibodies were applied for 1 h at room temperature (RT). For protein detection Western Lightning Plus ECL solution (Perkin-Elmer, NEL105001EA) and the ChemiDoc MP Imaging System (Bio-Rad) with their respective software, ImageLab 6.1, were used. Western blot band quantification was performed using ImageJ v.153a. The following antibodies were used: rabbit anti-ZEB1 (1:2,000 dilution, HPA027524, Sigma-Aldrich), mouse anti-E-cadherin (1:5,000 dilution, 610182, BD Transduction Laboratories), mouse anti-β-actin (1:10,000 dilution, A5441, Sigma-Aldrich), rabbit anti-SCD (1:1,000 dilution, 23393-1-AP, Proteintech), rabbit anti-FASN (1:1,000 dilution, MA5-14887, Thermo), rabbit anti-FADS2 (1:1,000 dilution, 28034-1-AP, Proteintech) and rabbit anti-ELOVL5 (1:1,000 dilution, PA583879, Thermo), mouse anti-ACSL4 (1:1,000 dilution, Santa Cruz, sc-271800), rabbit anti-GAPDH (1:10,000 dilution, Cell Signalling, 2118), mouse anti-SNAIL (1:500, Cell Signalling, 3895), rabbit anti-TWIST (1:1,000 dilution, Abcam, ab50581), goat anti-Mouse IgG Peroxidase (1:10,000 dilution, 115-035-1463, Jackson ImmunoResearch); and goat anti-rabbit IgG peroxidase (1:10,000 dilution, 111-035-144, Jackson ImmunoResearch).

Immunofluorescence and image acquisition

For immunofluorescence labelling, cells seeded onto sterile glass coverslips were fixed in 4% PFA, quenched and permeabilized in 0.2% Triton X-100/100 mM glycine/PBS, pH 7.4 and blocked in 3% BSA/PBS at room temperature. Primary and secondary antibodies were diluted in blocking solution and incubated for 1 h or 45 min, respectively, at room temperature in a humidified chamber protected from light. Nuclei were stained with DAPI (Sigma, D9542) before coverslips were mounted onto glass slides with CitiFluor AF1 solution (EMS, 17970-100). Images were acquired using a Leica DM5500 B microscope and processed using the Leica Application Suite X software. The following antibodies were used: rabbit anti-ZEB1 (1:250 dilution, HPA027524, Sigma-Aldrich), rabbit anti-vimentin (1:250 dilution, 5741, Cell Signalling), mouse anti-E-cadherin (1:250 dilution, 610182, BD Transduction Laboratories), Alexa Fluor 555 goat anti-rabbit IgG (H+L) (1:300 dilution, A21428, Life Technologies); Alexa Fluor 488 goat anti-mouse IgG (H+L) (1:300 dilution, A11029, Life Technologies).

Cell viability and death assay

To determine cell viability, cells were plated at 5–10% confluence in 96-well plates. After 16–24 h, cells were treated with vehicle or the indicated compounds at the given concentrations for 48–72 h. DMSO served as solvent control for most compounds, with the exception of ethanol (used for arachidonic acid and oleic acid) and PBS (used for gemcitabine). To determine the sensitization to ferroptosis by SCD and FADS2 inhibition, cells were pretreated with 5 µM MF438, 3 or 5 µM CAY10566 or 10 µM sc-26196 for the indicated time before the addition of ferroptosis inducers. To investigate the effect of exogenous fatty acid supplementation, cells were incubated with 10 µM 20:4/arachidonic acid or 500 µM 18:1/oleic acid for 12–16 h before ML210 was added. Ferroptosis was inhibited by co-treatment with ferrostatin-1 at a final concentration of 1 µM. Cell viability was either assessed by measuring cellular dehydrogenase activity via an MTT assay according to the manufacturer’s instructions or the confluence matrix using the live-cell imaging device Incucyte S3 (Sartorius). Confluence or absorbance data for the individual treatments were subtracted at the indicated time points either from their initial values or from positive controls, normalized to the mean value of control cells or the individual controls (each set to 100%) and optionally plotted to their respective, log-transformed drug concentrations. Cytotoxicity was measured using SYTOX Green nucleic acid stain (Thermo Fisher Scientific) at a final concentration of 5 nM. The cell death rate was calculated by normalizing the number of SYTOX Green-positive objects, indicating dead cells, to cell confluence (%) for each time point and condition.

Lung colonization

KPC(mix) cells were pretreated for 12 h with 10 µM 20:4/arachidonic acid, 500 µM 18:1/oleic acid or ethanol as vehicle control. Subsequently, 100 μl PBS containing 5 × 104 cells were injected into the tail vein of 8-week-old C57BL/6NRj mice (Janvier Labs). Littermates of both sexes were randomized for all treatment cohorts, monitored twice per week and killed 3 weeks after injection. Lungs were isolated, fixed in 4% paraformaldehyde and embedded in paraffin. Lung tissues were sectioned at 4-µM thickness and stained with haematoxylin and eosin solution. Per mice, metastatic lesions were screened on three sections separated by at least 200 μm. Quantification was performed by analysing the number of metastases as well as metastatic areas normalized to the respective lung area using ImageJ v.1.53a. For each treatment condition, nine mice were used in three independent experiments. The number and size of metastases never exceeded the maximal burden permitted by the local authorities.

Precision-cut lung slices

Precision-cut lung slices (PCLSs) were obtained from 8-week-old female C57BL/6NRj mice (Jackson Laboratory) using a vibratome VT1200S (Leica). On the same day, fluorescently labelled MDA-MB-231 wild-type (mCherry) or shZeb1 (tdTomato) were pretreated with 5 µM MF438 or DMSO as vehicle control. The following day, single lung slices were incubated with 1 × 105 cells in low-adherent 48-well plates for 4 h. After transfer into fresh plates, PCLS tumour cell co-cultures were treated with 0.5 µM ML210 ± 1 µM ferrostatin-1 or DMSO as control. Imaging was performed using the EVOS system (M7000 Thermo Fisher) before and 4 days after treatment. Fluorescent signals were quantified and normalized to their respective PCLS areas using ImageJ v.1.53a. The resulting values from day 4 were divided by those from day 0, followed by normalization to their respective DMSO controls. PCLS viability was confirmed using CyQUANT LDH Cytotoxicity Assay (Invitrogen).

Analyses of mouse allograft tumours

Cryo-conserved tumour specimens were obtained from subcutaneous allografts described by Krebs et al.15. We analysed two tumours derived from mesenchymal KPC cell lines (KPC550 and KPC701) and epithelial and mixed KPC cell lines (KPC438, KPC661 and 792) (with wild-type Zeb1 or heterozygous alleles), as well as one tumour from the KPCZ-derived cell lines (346, 387, 426 and 519) with Zeb1-knockout alleles for protein expression by immunohistochemistry (IHC) and for abundance of phospholipids by UPLC–MS/MS. For IHC, serial sections (4 µm) were treated as described15. The following primary antibodies were used: polyclonal rabbit anti-Zeb1 (Novus Biologicals, NBP1-05987, diluted 1:150), polyclonal rabbit anti-FADS2 (Proteintech, 28034-1-AP, diluted 1:100), polyclonal rabbit anti-ELOVL5 (Novus Biologicals, NBP3-14304, diluted 1:50), monoclonal rabbit anti-FASN (Thermo Fisher Scientific, MA5-14887, diluted 1:50), polyclonal rabbit anti-SCD (Proteintech, 23393-1-AP, diluted 1:250), polyclonal rabbit anti-ACSL4 (Santa Cruz, sc365230, diluted 1:250) and polyclonal rabbit anti-4-HNE (Abcam, ab46545, diluted 1:250). Tumours were grouped and analysed in three different ways according to (1) Zeb1 expression (IHC score ≤1 versus >1); (2) Zeb1 genotype (wild type/KPC versus Zeb1 knockout/KPCZ); and (3) histological tumour grade (grade ≤2 versus >2). IHC expression in the tumour cells was scored from 0 (low) to 3 (high). For UPLC–MS a ~30–50-mg fresh-frozen cryo-conserved tumour sample was used for the procedure described in ‘Extraction and analysis of phospholipids’.

Zebrafish engraftment

Zebrafish larvae were injected as previously described59,60. In brief, 300 tdTomato-positive MDA-MB-231 shCtrl or shZeb1 cells, resuspended in 2% polyvinylpyrrolidone 40 (PVP40, Sigma)/DPBS were injected intravenously via the duct of Cuvier of Tg(fli:eGFP) × casper zebrafish larvae at 48 h after fertilization using capillary glass needles. For ferroptosis rescue experiments, cells were pretreated with Fer-1 24 h and 4 h before transplantation. Engraftment procedures have been previously described10. Engrafted individuals were imaged at 1 and 3 days after implantation using an epifluorescence stereo microscope. All images were analysed using a custom ImageJ MACRO. Data were normalized to the wild-type control group of each experimental population or, in the case of drug treatments, to the vehicle control group; two biological replicates were combined with at least 20 individuals per biological replicate.

RNA extraction and qRT–PCR

Total RNA was isolated and reversely transcribed using the RNeasy Plus Mini kit (QIAGEN, 74136) and the RevertAid First Strand cDNA Synthesis Kit (Thermo Fisher Scientific, K1622) according to the manufacturer’s instructions. cDNA was amplified in 384-well plates using gene-specific primers (Supplementary Table 2), with Power SYBR Green PCR Master Mix (Applied Biosystems, 4367659) according to the manufacturer’s protocol. Samples were run in triplicates in a LightCycler 480 (Roche) and normalized to GAPDH, ACTB or HPRT1.

Gene Ontology analysis

The target genes list of Zeb1-bound promoters was obtained from Feldker et al.28, and the differentially expressed genes between the KPC wild-type and KPC Zeb1-knockout conditions1 were used to perform Gene Ontology term analysis using ClusterProfiler R package (v.4.0)11,61. Ensemble Gene symbols were fed into the tool with the default settings and the top 20 biological processes were selected for plotting.

Motif analysis

Analysis of transcription factor binding motifs was performed using LASAGNA-Search: an integrated web tool for transcription factor binding site search and visualization62.

Extraction and analysis of phospholipids

Lipids were extracted from cell pellets or tumour allografts by the sequential addition of PBS, methanol, chloroform and saline (at a final ratio of 14:34:35:17)63,64. The chloroform layer was recovered, brought to dryness using an Concentrator Plus System (Eppendorf) and the lipid film was dissolved in methanol. Internal standards (0.2 nmol each): Phosphatidylcholine (14:0/14:0), PE(14:0/14:0), PC(15:0/18:1-d7), PE(15:0/18:1-d7) and/or PI(15:0/18:1-d7).

Phospholipids were separated on an Acquity UPLC BEH C8 column (130 Å, 1.7 μm, 2.1 × 100 mm, Waters) using an Acquity UPLC (Waters) coupled to a QTRAP 5500 mass spectrometer (Sciex)65,66 or an ExionLC AD UPLC (Sciex) coupled to a QTRAP 6500+ mass spectrometer (Sciex), both equipped with an IonDrive Turbo V Ion Source and a TurboIonSpray probe12. For the latter, chromatographic separation was performed at 45 °C and 0.75 ml min−1 using mobile phase A (water:acetonitrile, 90:10, 2 mM ammonium acetate) and mobile phase B (water:acetonitrile, 5:95, 2 mM ammonium acetate). The gradient was ramped from 75 to 85% B over 5 min and increased to 100% B within 2 min, followed by isocratic elution for 2 min. The MS source and compound parameters for the QTRAP 6500+ mass spectrometer are shown in Supplementary Table 3. Phospholipids were analysed in the negative ion mode by multiple reaction monitoring (MRM), and the mean signal of the two fatty acid anion fragments was calculated.

Mass spectra were obtained and processed with Analyst (v.1.6.3 or v.1.7.1)67. To calculate absolute PE and phosphatidylinositol quantities, signals were normalized to protein content and a subgroup-specific deuterated internal standard. To calculate relative intensities, all analysed signals within the subgroup were summarized (=100%) and the signals of individual lipids were expressed as percentage of this sum. The fractions of PUFAs, MUFAs and SFAs in phospholipids were calculated from mean signal intensities divided by two and equally distributed to the sn-1 or sn-2 fatty acids. The proportions of PUFA- and non-PUFA-containing phospholipid species in Figs. 2c and 3c and Extended Data Fig. 2e (pie charts) summarize the relative intensities of phospholipids species either containing at least one PUFA or exclusively carrying SFA/MUFA.

Extraction and analysis of oxidized phospholipids

Oxidized phospholipids were extracted as described above and analysed using an ExionLC AD UHPLC system (Sciex) coupled to a QTRAP 6500+ mass spectrometer (Sciex) by MRM in a negative ion mode. The MS source and compound parameters are shown in Supplementary Table 3.

Oxidized phospholipids were identified from the fragments of [M-H]− (oxidized phosphatidylethanolamine, oxidized phosphatidylinositol) or [M+OAc]− (oxidized phosphatidylcholine) ions indicated in Supplementary Table 4. Signals were analysed only when retention times applied to the effective carbon number model and were within predefined ranges (Supplementary Tables 5–7). The definition of retention time windows was supported by reference phospholipids, oxPAPC (Avanti Polar Lipids) and oxidized PE(16:0/20:4). The latter was obtained by enzymatic oxygenation of PE(16:0/20:4) (Avanti Polar Lipids) with a lipoxidase (type V) from glycine max (soybean, L6632; batch no. SLCC4512; Sigma-Aldrich)68. The retention time windows for oxPAPC (1[O]: 2.8–4.6 min; 2[O]: 2.9–4.3 min; and 3[O]: 1.45–2.3 min) and oxidized PE(16:0/20:4) (1[O]: 2.79-2.96 min; 2[O]: 2.84-3.04 min) were extended to include potential regioisomers69 and adapted to additional phosphatidylcholine, phosphatidylethanolamine or phosphatidylinositol species based on the effective carbon number model, as listed in Supplementary Tables 5–7. Oxidized phospholipids were quantified based on the most intensive, specific transition to the oxidized fatty acid anions. To calculate the amount of phospholipids with one [1O], two [2O] or three oxygens [3O] incorporated, the individual signals within the defined retention time windows were summarized without discriminating between isomers and normalized to DMPC (oxidized phosphatidylcholine, oxidized phosphatidylinositol) or DMPE (oxidized phosphatidylethanolamine) and the cell number.

Public database queries

For the Cancer Dependency Map dataset (Figs. 1b and 3d), the CERES score for each gene of interest or tumour was downloaded from the DepMap portal (https://depmap.org/portal). The relationship between compound sensitivity and gene expression (Figs. 1b and 3d) was analysed using datasets from the Cancer Therapeutics Response Portal (http://portals.broadinstitute.org/ctrp.v2.1/)70,71,72. The correlation of expression analysis (Extended Data Fig. 4b.) was performed with cBioPortal using expression data from 732 cell lines of solid cancers from the Cancer Cell Line Encyclopedia (https://www.cbioportal.org/study/summary?id=ccle_broad_2019). For transcriptomic analysis of publicly available RNA-seq datasets from patients with breast cancer, processed MET500 RNA-seq samples (https://xenabrowser.net/datapages/?cohort=MET500)25 were downloaded from the UCSC Xena Browser, and Ensembl gene IDs were converted to gene symbols using annotations from Ensembl Release 108. Samples in each compendium were then scored for their position in the EM spectrum. In brief, the expression matrix was scaled (mean centred, with s.d. set to 1) and the previously described method73 was applied, using the KS gene signature for tumour cells. For GSVA analysis74, we used custom gene sets for MUFAs (SCD and FASN), PUFAs (FADS2, ELOVL5 and ACSL4) and the (single-gene) ZEB1. After ranking samples by EM score, GSVA was run with the ssGSEA method, and the GSVA scores were visualized with R’s pheatmap package. The regression line for scatter-plots was calculated using a linear model, as implemented in R. For further meta-analysis of published datasets for survival of patients with breast cancer, colon cancer and pancreatic cancer, we used KM-Plotter (http://kmplot.com/analysis)33. Settings were ‘all samples’ or as indicated in the figures for each tumour entity and for all analyses: ‘auto select best cut off’, ‘user selected probe sets’ with selection of the recommended probe set, ‘compute median survival’ and ‘censore at threshold’.

Statistics and reproducibility

Data analysis was performed using GraphPad Prism 9 software and OriginPro 2021 software. For multiple comparisons, ordinary or repeated measures one-way or two-way ANOVA with Dunnett’s, Tukey’s or Sidak’s post hoc tests were applied. For the comparison of two groups, multiple t-tests with false discovery rate of 5% using a two-stage linear step-up procedure by Benjamini, Krieger and Yekutieli or two-tailed unpaired t-tests were used. P values <0.05 were considered statistically significant. Mouse/fish numbers represent the biological replicates. Sample size and replicates are indicated in the figure legends. All experiments presented in the Article were repeated in at least three independent biological replicates. Data are presented as mean or mean ± s.e.m. of n observations, where n represents the individual number of experiments, unless otherwise specified in the figure legends. Precise P values are in the figure legends. No data were excluded from the analyses. The investigators were blinded to allocation during experiments (IHC analyses) and outcome assessment. No statistical methods were used to predetermine sample sizes but our sample sizes are similar to those reported in previous publications15,17. Data distribution was assumed to be normal but this was not formally tested.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.