Phase separation of FSP1 promotes ferroptosis

1Ferroptosis, a metabolic form of non-apoptotic cell death characterized by iron-dependent lipid peroxidation, has been defined only recently,. Ferroptosis has attracted tremendous interest because of its high relevance to human diseases such as neurodegenerative disorders, tissue damage during cold exposure, ischaemia–reperfusion injury and cancer–. In particular, triggering ferroptosis in the context of malignancies has emerged as a highly promising approach that shows synergistic effects with cancer immunotherapy and even kills therapy-resistant and metastatic cancers,,–. We recently showed that FSP1 represents a powerful backup system for the guardian of ferroptosis, known as GPX4, rendering tumours resistant to inhibition of this node,. However, because the first described FSP1-specific inhibitor iFSP1 (ref. ) does not qualify to be further developed as an anti-cancer drug, owing to its limited potential for medicinal chemistry development in terms of an unfavourable structure and substitution pattern, next-generation, efficacious in vivo FSP1 inhibitors for tumour treatment are urgently required.

2To identify possible in vivo-applicable FSP1-specific inhibitors as potential future drugs to combat difficult-to-treat cancers, we carefully revalidated hit compounds from ~10,000 drug-like small molecule compounds (as described previously) in terms of their potential for medicinal chemistry development, using cheminformatics tools for the prediction of physicochemical properties and drug-likeness. Hit validation studies identified the class of 3-phenylquinazolinones (represented by the lead compound icFSP1) as a class of potent pharmacological inhibitors of FSP1 (Fig. 1a). Preliminary structure–activity relationship studies have yet to identify compounds with substantial improvement over icFSP1 (Extended Data Fig. 1a). Treatment with icFSP1 caused marked lipid peroxidation and associated ferroptotic cell death in 4-hydroxytamoxifen (TAM)-inducible Gpx4-knockout mouse Pfa1 cells stably overexpressing human FSP1 (hFSP1) and in the human fibrosarcoma HT-1080 cell line (Fig. 1b–g and Extended Data Fig. 1b). icFSP1-induced cell death was rescued by ferroptosis inhibitors, but not by inhibitors targeting other forms of cell death, thus confirming its specificity for ferroptosis. Killing of (cancer) cells by targeting FSP1 usually requires co-treatment with other types of canonical ferroptosis inducers,, such as the system Xc− inhibitor erastin, the glutamate cysteine ligase (GCL) inhibitor l-buthionine sulfoximine (BSO), and the GPX4 inhibitors (1S,3R)-RSL3 (RSL3), ML210 and FIN56, as well as the iron oxidation compound FINO2 (refs.,,–), but not with compounds inducing other forms of cell death (Extended Data Fig. 1c–e). Thus, it was surprising that treatment of HT-1080 cells, the primary cell model in ferroptosis research, with icFSP1 alone or doxycycline-inducible FSP1 knockout for 72 h was sufficient to trigger ferroptosis (Fig. 1d and Extended Data Fig. 1f–h), in contrast to a panel of different human cancer cell lines (Extended Data Fig. 1c). To address whether icFSP1 may have off-target effects, HT-1080 and HEK293T cells and primary peripheral blood mononuclear cells (PBMCs) were exposed to the FSP1 inhibitors (icFSP1 and iFSP1) for 72 h and 24 h, respectively. In these experiments, icFSP1 did not show off-target effects even at higher concentrations as compared with iFSP1 (Extended Data Fig. 1h–j). Apart from this, FSP1-knockout cells treated with increasing concentrations of icFSP1 did not show any additional synergistic effects when cells were co-incubated with a GPX4 inhibitor (Extended Data Fig. 1k,l), indicating that icFSP1 should be considered to be a selective FSP1 inhibitor.

3To investigate whether icFSP1 also induces death in non-human cells, we included two different mouse cell lines and one rat fibroblast cell line in our study (that is, 4T1, B16F10 and Rat1, respectively), showing that co-treatment with RSL3 or Gpx4 knockout failed to synergistically kill these cells (Extended Data Fig. 2a–g). In line with this, lipid peroxidation and cell viability were not affected by icFSP1 in Gpx4-knockout Pfa1 cells stably overexpressing mouse FSP1 (mFSP1; Extended Data Fig. 2h,i), suggesting that icFSP1 specifically inhibits the human isoform. Additionally, we tested other orthologues of FSP1 in Pfa1 cells, overexpressing FSP1 from Gallus gallus (chicken) and Xenopus laevis (frog). Although FSP1 expression (except for that from X. laevis) fully prevented RSL3-induced ferroptosis, icFSP1 reduced cell viability only in cells overexpressing hFSP1 (Extended Data Fig. 2j–l). These data therefore reinforce the notion that icFSP1 is an hFSP1-specific inhibitor.

4To investigate the mechanism of action of these next-generation FSP1 inhibitors, the direct inhibitory activity of FSP1 was measured in vitro by using an established assay with recombinant hFSP1 enzyme (Fig. 2a). Whereas the enzymatic activity of hFSP1 was inhibited by iFSP1 in a cell-free system, as reported,, icFSP1 did not inhibit hFSP1 activity in the low-micromolar range, although it clearly affected cell viability and lipid peroxidation in Pfa1 cells overexpressing hFSP1 (Fig. 2b and Extended Data Fig. 3a–k). In fact, the estimated half-maximal inhibitory concentration (IC50) of icFSP1 in the in vitro assay (enzyme) was more than 100-fold higher than the half-maximal effective concentration (EC50) observed in Pfa1 cells (Fig. 2b and Extended Data Fig. 3i). These results strongly argue that icFSP1, as compared with iFSP1, uses a different mechanism of action to inhibit hFSP1 activity in cells.

5To shed light on the mechanism of action, we first analysed whether icFSP1 may decrease the expression levels of hFSP1. Immunoblotting of hFSP1-expressing H460 or HT-1080 cells after icFSP1 treatment for 48 h and 72 h showed that expression levels of hFSP1 were not affected by icFSP1 treatment (Extended Data Fig. 3l–n). Next, we considered that icFSP1 might change the subcellular localization of hFSP1 by detaching it from lipid membranes, thereby preventing its anti-ferroptotic function of scavenging phospholipid radicals through ubiquinone and/or vitamin E or K–. To this end, hFSP1 fused to enhanced green fluorescent protein and streptavidin (hFSP1–EGFP–Strep) was stably overexpressed in Pfa1 cells and its localization was monitored in response to icFSP1 treatment. Unlike iFSP1, icFSP1 markedly changed the subcellular localization of hFSP1–EGFP–Strep, as illustrated by the appearance of distinct foci and cellular condensates (Fig. 2c, Extended Data Fig. 3o and Supplementary Video 1). These condensates accumulated in cells in a time-dependent manner (Fig. 2d) and only occurred in cells expressing hFSP1 and not in those expressing mFSP1 (Extended Data Fig. 3p), corroborating that icFSP1 is specific for the human orthologue. To test whether the change in subcellular localization of hFSP1 causes induction of ferroptosis, Gpx4-knockout Pfa1 cells stably overexpressing hFSP1 fused to blue fluorescent protein (BFP) were established. hFSP1–BFP signal was monitored by live-cell imaging of cells co-stained with Liperfluo (a lipid hydroperoxide sensor) and propidium iodide, which can only stain nuclei when the plasma membrane becomes leaky. Straight after treatment of hFSP1–BFP-expressing cells with icFSP1, condensates were induced followed by Liperfluo oxidation, whereas lipid peroxide signals gradually increased in cells until cells became positive for propidium iodide as a measure for cell membrane rupture (Fig. 2e and Supplementary Video 2). These results indicate that changes in the subcellular localization of FSP1 precede lipid peroxidation and ferroptosis.

6To interrogate whether hFSP1 condensates may localize to specific subcellular compartments, hFSP1–EGFP–Strep-expressing cells were co-stained with a number of organelle-specific markers. Reportedly, FSP1 localizes to different subcellular structures, including the endoplasmic reticulum (ER), Golgi apparatus, lipid droplets and perinuclear structures, (Extended Data Fig. 4a). Yet, treatment of cells with icFSP1 did not induce hFSP1 condensates that clearly colocalized with any of these subcellular structures. Moreover, hFSP1 condensates did not colocalize with other cell organelles, such as endosomes, lysosomes, mitochondria, ubiquitin-dependent aggresomes or stress granules (G3BP1) (Extended Data Fig. 4b,c). Following treatment with icFSP1, these condensates were also detectable in H460 cells (expressing only endogenous hFSP1) and in cells with even lower expression levels than those of endogenous hFSP1 using doxycycline-dependent, scalable expression of hFSP1 (Extended Data Fig. 4d–g). We further noted that hFSP1 condensates dynamically and freely moved around and fused in cells in response to icFSP1 treatment (Fig. 3a, Extended Data Fig. 4h and Supplementary Video 3), which appeared to be reversible after washing out the inhibitor (Fig. 3b and Supplementary Video 4). To investigate the state of condensates in more detail, we established fluorescence recovery after photobleaching (FRAP) analysis. These studies showed that only early-state condensates exhibited FRAP (Fig. 3c,d and Supplementary Video 5), in contrast to late-state condensates (Extended Data Fig. 5a,b and Supplementary Video 6).

7On the basis of these liquid droplet-like properties of hFSP1, we considered that hFSP1 condensates might involve phase separation. Phase separation is a physicochemical process characterized by the reversible formation of biomolecular condensates. These condensates are involved in the regulation of cellular signalling following stress and have been linked to human diseases, including cancer and neurodegeneration,–. In particular, phase separation is involved in the partitioning of target proteins in cancer tissue and promotes the formation of aggregates of neurodegenerative disease-related proteins. Thus, we asked whether hFSP1 has the propensity to generate condensates in a cell-free system. To this end, hFSP1–EGFP–Strep was immunoprecipitated from transfected HEK293T cells using the Strep-tag to isolate natively myristoylated hFSP1 from cells. For condensate formation assays, polyethylene glycol (PEG) was used as a molecular crowding agent. Purified hFSP1 was reconstituted with 10% PEG, whereupon hFSP1 immediately formed viscoelastic material, in contrast to immunoprecipitated EGFP–Strep controls (Fig. 3e and Extended Data Fig. 5c,d). To investigate whether icFSP1 alone can initiate hFSP1 condensates, purified hFSP1 was reconstituted with PEG and/or icFSP1. However, hFSP1 could form condensates in the presence of PEG regardless of icFSP1. These differences in condensate formation are presumably due to the fact that cell-free and cellular conditions differ greatly and condensates can form viscoelastic material as shown in cells. To reinforce our finding of hFSP1 condensation in a cell-free system through phase separation, we used recombinant hFSP1 without myristoylation (non-myr-FSP1) purified from Escherichia coli. Again, hFSP1 could form condensates in an FSP1 and PEG concentration-dependent manner (Extended Data Fig. 5e–h). Finally, we performed sedimentation assays to investigate whether hFSP1 condensates induced by PEG lead to stable viscoelastic material. Almost all hFSP1 could be recovered from supernatant fractions in pellets in the presence of PEG (Extended Data Fig. 5i,j), suggesting that hFSP1 has the propensity to form condensates induced by phase separation.

8Phase separation can lead to formation of membrane-less compartments in cells, where multivalent interactions, typically involving intrinsically disordered regions (IDRs) and low-complexity regions (LCRs), are known to be critical. Phase separation predictors revealed that hFSP1 contains two putative IDRs and one LCR in its sequence (Extended Data Fig. 6a), which may be required for condensate formation through phase separation. To analyse in detail the role of these predicted domains, the following series of hFSP1 deletion mutants was generated: ∆IDR1, ∆LCR, ∆N (∆IDR1 and ∆LCR) and ∆IDR2 (Fig. 4a). Two additional mutants were generated in which membrane localization is known to be affected,: (1) the G2A mutant, which is a myristoylation-defective mutant with strongly affected localization, abrogating the ferroptosis-suppressive function of FSP1 (ref. ), and (2) the Lyn11–G2A mutant, which contains the membrane-targeting Lyn11 sequence fused N-terminally to FSP1G2A and can suppress ferroptosis. After icFSP1 treatment, only wild-type hFSP1 and Lyn11–hFSP1G2A changed their subcellular localization to form hFSP1 condensates, and all other mutants did not undergo hFSP1 condensate formation following icFSP1 treatment (Fig. 4b). In line with this, pretreatment of cells with the myristoylation inhibitor IMP-1088 also abrogated hFSP1 condensation in cells (Extended Data Fig. 6b). To investigate whether the hFSP1 deletion mutants might have the propensity to form condensates in a cell-free system, purified hFSP1 mutants from transfected HEK293T cells were reconstituted with 10% PEG. Only wild-type hFSP1 and the G2A and Lyn11–G2A mutants formed condensates in the presence of PEG, whereas the IDR and LCR deletion mutants did not form condensates (Extended Data Fig. 6c). Next, we produced recombinant hFSP1 with myristoylation (myr-FSP1) purified from E. coli (Extended Data Fig. 6d) to test whether myristoylation may afford in vitro phase separation induced by PEG, lower salt concentrations and icFSP1 (Extended Data Fig. 6e,f). PEG and lower salt concentrations seemed to facilitate phase separation of myristoylated hFSP1 in vitro, whereas icFSP1 alone did not induce phase separation, suggesting that the cellular context (that is, other binding partners, the membrane environment, post-translational modifications, etc.) is important for phase separation in vitro. Furthermore, phase separation of myristoylated hFSP1 could be induced by icFSP1, whereas non-myristoylated hFSP1G2A did not form condensates, even when hFSP1 condensates of the wild-type enzyme were present in Pfa1 cells (Extended Data Fig. 6g). These data imply that the presence of a myristoylation tag facilitates condensate formation, as seen for other myristoylated proteins such as the enhancer of zeste 2 (EZH2) polycomb repressive complex 2 subunit. Moreover, myristoylation may function as a sticker enhancing polyphasic linkage, for phase separation that is modulated by icFSP1, as a ligand.

9To investigate the potential ferroptosis-suppressive functions of these mutants, cell viability experiments were performed using RSL3 and TAM-inducible Gpx4 knockout (Extended Data Fig. 6h,i). These analyses confirmed that only wild-type hFSP1 and Lyn11–hFSP1G2A could efficiently suppress ferroptosis. Hence, these data indicate that the IDRs and LCR may be required for the ferroptosis-suppressive role of FSP1, with icFSP1-induced ferroptosis triggered by hFSP1 condensation through phase separation.

10Because icFSP1 was specific for the human enzyme and did not inhibit its mouse counterpart (Extended Data Fig. 2), we addressed the underlying molecular mechanisms that may account for this species-specific inhibitory activity (Extended Data Fig. 7a). The phase separation predictor revealed that mFSP1 also harbours two predicted IDRs and an LCR with some sequence differences at its N terminus. Thus, we decided to generate a chimeric enzyme (that is, hmFSP1) consisting of the first 27 residues of hFSP1 (comprising IDR1 and the LCR) fused to residues 28–373 of mFSP1. However, hmFsp1 did not form condensates (Extended Data Fig. 7b,c), implying that other amino acids contribute to sensitizing cells to icFSP1-mediated condensate formation. On the basis of amino acid differences between the human and mouse orthologues, we generated a series of point mutations in human FSP1 and stably expressed the corresponding mutants in Pfa1 cells (Extended Data Fig. 7d). These studies allowed us to identify the S187, L217 and Q319 residues of hFSP1 as being critical for icFSP1-dependent condensate formation and ferroptosis induction, as substitution at these sites rendered the mutant proteins resistant (Fig. 4c–e and Extended Data Fig. 7d,e). Given that (1) binding of icFSP1 to FSP1 was not affected by these substitutions (Extended Data Fig. 7f) and (2) reverse substitutions at these three positions to the human residues allowed mFSP1 to form icFSP1-dependent condensates and induce ferroptosis (Extended Data Fig. 7g–i), S187, L217 and Q319 may have critical effects on FSP1–FSP1 interactions to trigger phase separation in cells in the presence of icFSP1 (although the precise binding site of icFSP1 to wild-type hFSP1 remains to be structurally resolved).

11To evaluate whether icFSP1 might be applicable for in vivo use, metabolic stability and pharmacokinetics analyses were performed, showing that icFSP1 has clearly improved microsomal stability in mice and maximum concentration in mouse plasma as compared with iFSP1 (Extended Data Fig. 8a,b). Furthermore, to evaluate the efficacy of icFSP1 in a tumour-bearing mouse model, Gpx4 and Fsp1 double knockout B16F10 cells overexpressing hFSP1 were subcutaneously injected into female C57BL/6J mice. After tumours reached approximately 25–50 mm3 in size, mice were randomized and treated intraperitoneally with vehicle or icFSP1 twice daily. icFSP1 treatment significantly inhibited tumour growth and decreased tumour weight, without affecting body weight (Fig. 4f and Extended Data Fig. 8c). Notably, treatment of tumour-bearing mice with icFSP1 markedly increased the abundance of hFSP1 condensates and immunoreactivity to 4-hydroxynonenal (4-HNE), a lipid peroxidation breakdown product (Fig. 4g and Extended Data Fig. 8d). These data indicate that icFSP1 may trigger phase separation of hFSP1 and thereby impair tumour growth in vivo. To substantiate these findings, we established a Gpx4 and Fsp1 double-knockout B16F10 mutant cell line overexpressing hFSP1Q319K to study condensate formation in vivo. Like the results obtained with Pfa1 and H460 cells, melanoma cell lines dependent on hFSP1Q319K were resistant to icFSP1 in cultured cells and in vivo (Extended Data Fig. 8e–j); in line with this, FSP1 condensates were not observed after icFSP1 treatment in vivo (Fig. 4h and Extended Data Fig. 8h). To investigate whether icFSP1 may also work in the human context, we used a human melanoma cell line (A375) and a human lung cancer cell line (H460), which are known to express substantial levels of FSP1 (ref. ) and which can survive after withdrawal of radical trapping agents even when GPX4 is genetically deleted (that is, with GPX4 knockout),. In fact, GPX4-knockout cells were highly sensitive to icFSP1 treatment, and the tumour growth of GPX4-knockout cells was substantially inhibited in the corresponding xenograft tumour model (Extended Data Fig. 9a–i). In conclusion, our results indicate that the hFSP1-specific inhibitor icFSP1 may trigger phase separation of FSP1 and synergize with canonical ferroptosis inhibitors to induce ferroptosis, as a viable way for efficient eradication of certain cancer entities.

12Here we report on a yet-unrecognized class of in vivo-efficacious hFSP1 inhibitors, that is, 3-phenylquinazolinones, that exhibit a unique mechanism of action in which they trigger ferroptosis through dissociation of FSP1 from the membrane and formation of FSP1 condensates involving phase separation (Fig. 4i). Our experiments show that icFSP1-mediated phase separation requires several molecular features in hFSP1, such as N-terminal myristoylation, specific amino acid residues (S187, L217 and Q319), and the IDRs and LCR, resulting in lipid peroxidation and ferroptosis under GPX4-inhibited conditions. In the absence of an experimentally determined three-dimensional structure of FSP1, we mapped the amino acid residues that contribute to icFSP1-induced phase separation of FSP1, on the basis of our mutational analysis, onto the structure predicted by AlphaFold2 (refs.,). Although the globular fold prediction had a high-confidence IDR1 region, it was annotated as an uncertain region, in contrast to the LCR, which was predicted to exhibit an α-helical conformation with an intermediate score. The role of the LCR needs to be experimentally studied, to examine how this region may contribute to phase separation and potentially interact with the globular domain of FSP1. These interactions may well be modulated by icFSP1 and should help in understanding the underlying structural mechanisms. Moreover, considering that icFSP1 initiates formation of FSP1 condensates, including of plasma membrane-localized mutants, immediately after treatment and that icFSP1 by itself cannot trigger condensation of myristoylated FSP1 in vitro, icFSP1 may induce condensates by modulating FSP1–membrane interactions and thereby reducing the membrane-binding affinity of FSP1, similar to the KRAS inhibitor Cmpd2 (ref. ) and other modulators (for example, Ca2+ for recoverin) of myristoyl–ligand switches. In support of our hypothesis, the three-dimensional structural model of FSP1 (refs.,) showed that residues S187, L217 and Q319 are all located on the surface of FSP1, with Q319 in particular close to the expected membrane-binding surface. Considering the known relevance of charged and polar residues for protein–protein interactions during phase separation, changing S187 to cysteine, L217 to arginine and Q319 to lysine should increase the positive surface charge or reduce polarity and may thereby impair phase separation in cells (Extended Data Fig. 7i).

13The concept in which ligands modulate the driving forces for phase separation is known as polyphasic linkage,. Conceptually, icFSP1 would modulate the phase transition of FSP1, probably through interactions that directly or indirectly involve residues such as S187, L217 and Q319, and disruption of interactions by mutagenesis changes the phase boundary. In particular, myristoylation appears to be indispensable for this process in which icFSP1, as a ligand, preferentially binds to the myristoylated form of FSP1 (Extended Data Fig. 6g).

14Although inhibition of FSP1 alone is generally not sufficient to drive cancer cell death through ferroptosis,, a subset of cancer cell lines can potentially be sensitive to FSP1 inhibition alone under certain conditions, as was observed for HT-1080 cells. In this respect, database analysis might be helpful to predict the sensitivity of cancer cell lines to FSP1 inhibition (Extended Data Fig. 10). In light of the fact that Fsp1-knockout mice are fully viable and that icFSP1 does not show any observed off-target activity and does not affect body weight even at high concentrations (Extended Data Figs. 1 and 8), FSP1 should be regarded as an attractive target for tumour treatment. Thus, future studies should be geared to developing pharmacological approaches that simultaneously target both the cyst(e)ine–GSH–GPX4 node and the FSP1 system, allowing for efficient tumour cell eradication by triggering ferroptosis as a new anti-cancer paradigm.

15Lip-1 (Selleckchem, cat. no. S7699), doxycycline (Dox; Sigma, cat. no. D9891), RSL3 (Cayman, cat. no. 19288), BSO (Sigma, cat. no. B2515), iFSP1 (ChemDiv, cat. no. 8009-2626), icFSP1 (ChemDiv, cat. no. L892-0224 or custom synthesis by Intonation Research Laboratories), erastin (Merck, cat. no. 329600), ML210 (Cayman, cat. no. Cay23282-1), FIN56 (Cayman, cat. no. Cay25180-5), FINO2 (Cayman, cat. no. Cay25096-1), deferoxamine mesylate salt (DFO; Sigma, cat. no. 138-14-7), ferrostatin-1 (Fer-1; Sigma, cat. no. SML0583), zVAD-FMK (zVAD; Enzo Life Sciences, cat. no. ALX-260-02), necrostatin-1s (Nec-1s; Enzo Life Sciences, cat. no. BV-2263-5), MCC950 (Sigma, cat. no. 5381200001), olaparib (Selleckchem, cat. no. S1060), staurosporine (STS; Cayman, cat. no. 81590), recombinant human tumour necrosis factor (TNF; R&D Systems, cat. no. NBP2-35076), Smac mimic (BV-6; Selleckchem, cat. no. S7597), nigericin (Thermo Fisher, cat. no. N1495), IMP-1088 (Cayman, cat. no. Cay25366-1) and lipopolysaccharide (LPS; Sigma, cat. no. L2880) were used in this study.

16Five- to six-week-old female C57BL/6J and athymic nude mice were obtained from Charles River, and 6- to 7-week-old mice were used for the experiments. All mice were kept under standard conditions with water and food provided ad libitum and in a controlled environment (22 ± 2 °C, 55% ± 5% humidity, 12-h light/12-h dark cycle) in the Helmholtz Munich animal facility under SPF-IVC standard conditions. All experiments were performed in compliance with the German Animal Welfare Law and were approved by the institutional committee on animal experimentation and the government of Upper Bavaria (ROB-55.2-2532.Vet_02-17-167).

17TAM-inducible Gpx4-knockout mouse immortalized fibroblasts (referred to as Pfa1 cells) were reported previously. Genomic Gpx4 deletion can be achieved by TAM-inducible activation of Cre recombinase using the CreERT2/loxP system. HT-1080, 786-O, A375, NCI-H460 (H460), MDA-MB-436, HT-29, B16F10, and 4T1 cells were purchased from ATCC. THP-1 cells were obtained from DSMZ). Human PBMC cells were purchased from Tebu-bio (cat. no. 088SER-PBMC-F). SUDHL5, SUDHL6, DOHH2 and OCI-Ly19 cells were a gift from S. Hailfinger. Rat1 cells were a gift from Medizinische Hochschule Hannover. Pfa1, HT-1080, 786-O, A375, HT-29, Rat1 and B16F10 cells were cultured in high-glucose DMEM (4.5 g l–1 glucose) with 10% FBS, 2 mM l-glutamine and 1% penicillin-streptomycin. H460, MDA-MB-436, THP-1, PBMC, SUDHL5, SUDHL6, DOHH2 and 4T1 cells were cultured in RPMI GlutaMax with 10% FBS and 1% penicillin-streptomycin. OCI-Ly19 cells were cultured in IMDM with 10% FBS and 1% penicillin-streptomycin. To generate cell lines with stable overexpression, appropriate antibiotics (1 µg ml–1 puromycin, 10 µg ml–1 blasticidin and 0.5–1.0 mg ml–1 G418) were used. GPX4-knockout human A375 and H460 cells were cultured in the presence of 1 µM Lip-1 for maintenance. All cells were cultured at 37 °C with 5% CO2 and verified to be negative for mycoplasma.

18Cells were seeded on 96-well plates and cultured overnight. All cell number conditions for each cell line are described in Supplementary Table 1. The next day, the medium was changed to medium containing the following compounds: RSL3, ML210, erastin, FIN56, FINO2, BSO, iFSP1, icFSP1, Lip-1, DFO, Fer-1, zVAD, Nec-1s, MCC950, olaparib, STS, TNF, Smac mimic or nigericin at the indicated concentrations. For TAM and Dox treatment, cells were seeded with compounds at the same time. Cell viability was determined 1 h (for nigericin), 24–48 h (for RSL3, ML210, erastin, FIN56, FINO2, iFSP1, icFSP1, STS, TNF, Smac mimic and zVAD) or 72 h (for BSO, icFSP1, TAM and Dox) after the start of treatment using AquaBluer (MultiTarget Pharmaceuticals, cat. no. 6015) as an indicator of viable cells according to the manufacturer’s protocol. For apoptosis induction, HT-1080 cells were incubated with different concentrations of STS for 24 h. For necroptosis induction, HT-29 cells were incubated with different concentrations of TNF with Smac mimic (400 nM) and zVAD (30 µM) for 24 h. For pyroptosis induction, THP-1 cells stimulated with LPS (1 µg ml–1, 2 h) were incubated with nigericin for 1 h. For ferroptosis induction, cells were incubated with ferroptosis inducers for 24–72 h.

19As readout, fluorescence was measured at an excitation/emission wavelength of 540/590 nm using a SpectraMax M5 microplate reader with SoftMax Pro v.7 (Molecular Devices) after 4 h of incubation with AquaBluer in normal cell culture conditions. The relative cell viability (%) was calculated as follows: (fluorescence of samples – background)/(fluorescence of appropriate control samples – background) × 100.

20Cells were seeded on 96-well plates and cultured overnight. The next day, the medium was changed to medium containing compounds and cells were incubated for another 24 h. Cell death rates were measured using the Cytotoxicity Detection Kit (LDH) (Roche, cat. no. 11644793001) in principle following the manufacturer’s protocol. In brief, cell culture supernatant was collected as a medium sample. Cells were then lysed with PBS containing 0.1% Triton X-100 as a lysate sample. The medium and lysate samples were individually mixed with reagents on microplates, and the absorbance was measured at 492 nm using a SpectraMax M5 microplate reader after incubation for 15–30 min at room temperature. The cell death ratio was calculated by LDH release (%) as follows: (absorbance of medium samples – background)/((absorbance of lysate samples – background) + (absorbance of medium samples – background)) × 100.

21Wild-type and Gpx4-knockout Pfa1 cells stably overexpressing hFSP1 were seeded on separate 384-well plates (500 cells per well) and screened with a library of small molecule inhibitor compounds as reported previously. Cell viability of the different cell lines was assessed 48 h after the start of treatment using AquaBluer. Compounds showing selective lethality in Gpx4-knockout Pfa1 cells stably overexpressing hFSP1 were then validated in cell viability assays and in vitro FSP1 enzymatic assays.

22For lipid peroxidation assays, 100,000 cells per well were seeded on a 12-well plate 1 day before the experiments. The next day, cells were treated with 2.5 µM icFSP1 for 3 h and then incubated with 1.5 µM C11-BODIPY 581/591 (Invitrogen, cat. no. D3861) for 30 min in a 5% CO2 atmosphere at 37 °C. Subsequently, cells were washed once with PBS, trypsinized and then resuspended in 500 µl PBS. Cells were passed through a 40-μm cell strainer and analysed with a flow cytometer (CytoFLEX, Beckman Coulter) with a 488-nm laser for excitation. Data were collected from the FITC detector (for the oxidized form of BODIPY) with a 525/40-nm bandpass filter and from the PE detector (for the reduced form of BODIPY) with a 585/42-nm bandpass filter using CytExpert v.2.4 (Beckman Coulter). At least 10,000 events were analysed per sample. Data were analysed using FlowJo software (FlowJo). The ratio of fluorescence of C11-BODIPY 581/591 (lipid peroxidation) (FITC/PE ratio (oxidized/reduced ratio)) was calculated as follows: (median FITC-A fluorescence – median FITC-A fluorescence of unstained samples)/(median PE-A fluorescence – median PE-A fluorescence of unstained samples). An example gating strategy is shown in Supplementary Fig. 1.

23Two million cells were seeded on 15-cm dishes 1 day before the experiments. The next day, cells were treated with 5 µM icFSP1 to induce lipid peroxidation. Five hours later, cells were collected, sampled to liquid nitrogen and stored at –80 °C. Lipids from cells were extracted using the methyl-tert-butyl ether (MTBE) method. In brief, cell pellets collected in PBS containing dibutylhydroxytoluene (BHT; 100 µM) and diethylenetriamine pentaacetate (DTPA; 100 µM) were washed and centrifuged. SPLASH LIPIDOMIX (Avanti Polar Lipids) was added (2.5 µl), and samples were incubated on ice for 15 min. After addition of ice-cold methanol (375 µl) and MTBE (1,250 µl), samples were vortexed and incubated for 1 h at 4 °C (orbital shaker, 32 rpm). Phase separation was induced by adding water (375 µl), and samples were vortexed, incubated for 10 min at 4 °C (orbital shaker, 32 rpm) and centrifuged to separate the organic and aqueous phases (10 min, 4 °C, 1,500g). The organic phase was collected, dried in a vacuum evaporator and redissolved in 100 µl isopropanol. Lipid extracts were transferred to glass vials for LC–MS analysis.

24Reversed-phase LC was carried out on a Shimadzu ExionLC equipped with an Accucore C30 column (150 × 2.1 mm, 2.6 µm, 150 Å; Thermo Fisher Scientific). Lipids were separated by gradient elution with solvent A (1:1 (v/v) acetonitrile/water) and solvent B (85:15:5 (v/v/v) isopropanol/acetonitrile/water), both containing 5 mM NH4HCO2 and 0.1% (v/v) formic acid. Separation was performed at 50 °C with a flow rate of 0.3 ml min–1 using the following gradient: 0–20 min, increase from 10% to 86% B (curve 4); 20–22 min, increase from 86% to 95% B (curve 5); 22–26 min, 95% isocratic; 26–26.1 min, decrease from 95% to 10% B (curve 5), which was followed by re-equilibration for 5 min at 10% B. MS analysis was performed on a Sciex 7500 system equipped with an electrospray ionization (ESI) source and operated in negative-ion mode. Products were analysed in MRM mode monitoring transitions from the parent ion to the daughter ion, as described in Supplementary Table 1, with the following parameters: TEM, 500 °C; GS1, 40; GS2, 70; CUR, 40; CAD, 9; IS, −3,000 V.

25The area under the curve for the parent to daughter ion transition was integrated and normalized by appropriate lipid species, PC(15:0/18:1(d7)) or PE(15:0/18:1(d7)), from the SPLASH LIPIDOMIX Mass Spec Standard (Avanti Polar Lipids). Normalized peak areas were further log transformed and auto-scaled in MetaboAnalyst online platform v.5.0 (https://www.metaboanalyst.ca). Zero values were replaced by 0.2 times the minimum value detected for a given oxidized lipid in the samples. Oxidized lipids showing a significant difference (ANOVA, adjusted P value (false discovery rate (FDR)) cut-off of 0.05) between samples were used for the heat maps. The heat maps were created in GraphPad Prism 9. The colour scheme corresponds to auto-scaled log-transformed fold change relative to the mean log value for the samples.

26Cells were lysed in LCW lysis buffer (0.5% Triton X-100, 0.5% sodium deoxycholate salt, 150 mM NaCl, 20 mM Tris-HCl, 10 mM EDTA and 30 mM sodium pyrophosphate tetrabasic decahydrate) supplemented with protease and phosphatase inhibitor cocktail (cOmplete and phoSTOP; Roche, cat. nos. 04693116001 and 4906837001) and centrifuged at 20,000g for 1 h at 4 °C. The supernatant was sampled by adding 6× SDS sample buffer (375 mM Tris-HCl pH 6.8, 9% SDS, 50% glycerol, 9% β-mercaptoethanol and 0.03% bromophenol blue). After heating at 98 °C (55 °C for xCT) for 3 min, the samples were resolved on 12% SDS–PAGE gels (Bio-Rad, cat. no. 4568043 or 4568046) and subsequently electroblotted onto PVDF membrane (Bio-Rad, cat. no. 170-4156). The membranes were blocked with 5% skim milk (Roth, cat. no. T145.2) in TBS-T (20 mM Tris-HCl, 150 mM NaCl and 0.1% Tween-20) and then probed with primary antibodies, diluted in first antibody dilution buffer (TBS-T with 5% BSA and 0.1% NaN3 (Sigma, cat. no. S2002)), against GPX4 (1:1,000; Abcam, cat. no. ab125066), valosin-containing protein (VCP; 1:1,0000; Abcam, cat. no. ab11433 or ab109240), Flag tag (1:5,000; Cell Signaling Technology, cat. no. 2368), HA tag (1:1,000; clone 3F10, homemade), hFSP1 (1:1,000; Santa Cruz, cat. no. sc-377120, AMID), mFSP1 (1:500; clone AIFM2 1A1, rat IgG2a), hFSP1 (1:10; clone 6D8, rat IgG2a), mFSP1 (1:5; clone AIFM2 14D7, rat IgG2b), human SLC7A11 (1:10; rat IgG2a monoclonal antibody against an N-terminal peptide of human xCT, clone 3A12-1-1, developed in house), mouse SLC7A11 (1:1,000; Cell Signaling Technology, cat. no. 98051), ACSL4 (1:1,000; clone A-5, Santa Cruz, cat. no. sc-271800) or β-actin-HRP (1:50,000; Sigma, cat. no. A3854) diluted in 5% skim milk in TBS-T overnight. After membranes were washed and probed with appropriate secondary antibodies diluted in 5% skim milk in TBS-T, antibody–antigen complexes were detected with the ChemiDoc Imaging System with Image Lab v.6.0 (Bio-Rad). Representative images are shown after adjustment to the appropriate brightness and angle using ImageJ/Fiji software (v.1.52 and v.1.53).

27All plasmids for this study were constructed using standard molecular biology techniques and verified by sequencing as follows. A human FSP1 cDNA (NM_001198696.2, 1008:C>T) was cloned from previously reported vectors. Codon-optimized sequences for Mus musculus (mouse) FSP1 (NP_001034283.1), Rattus norvegicus (rat) FSP1 (NP_001132955.1), Gallus gallus (chicken) FSP1 (XP_421597.1) and Xenopus laevis (frog) FSP1 (NP_001091397.1) were cloned into the p442 vector. Codon-optimized sequences for human FSP1 (NP_001185625.1) and mouse Fsp1 (NP_001034283.1) were synthesized by TWIST Bioscience and subcloned into 141-IRES-puro vector. To generate deletion mutants or perform subcloning, desired DNA sequences were first amplified using KOD One PCR master mix (Sigma, cat. no. KMM-201NV) or PrimeSTAR Max DNA polymerase master mix (Takara Bio, cat. no. R045A) and resulting PCR products were purified by Wizard SV Gel&PCR Clean-up System (Promega, cat. no. A9285) according to the manufacturer’s protocol. Ligation reactions of PCR products or single guide RNA (sgRNA) duplexes with digested vectors were performed using T4 ligase (NEB, cat. no. M0202L) or In-Fusion cloning enzymes (Takara Bio, cat. no. 639649 or 638948) according to the manufacturer’s protocol. Subsequently, reaction mixtures were transformed into stable competent cells (NEB, cat. no. C3040H). Plasmids were isolated using the QIAprep Spin Miniprep kit (Qiagen, cat. no. 27106) according to the manufacturer’s protocol; the correct inserts of plasmids were confirmed by sequencing.

28HEK293T cells were used to produce lentiviral particles. The ecotropic envelope protein of murine leukaemia virus (MLV) was used for mouse-derived cells, while the amphitropic envelope protein VSV-G was used for human-derived cells. A third-generation lentiviral packaging system consisting of transfer plasmids, envelope plasmids (pEcoEnv-IRES-puro or pHCMV-EcoEnv (ecotropic particles) or pMD2.G (pantropic particles)) and packaging plasmids (pMDLg_pRRE and pRSV_Rev or psPAX2) was co-lipofected into HEK293T cells using transfection reagent (PEI MAX (Polysciences, cat. no. 24765) or X-tremeGENE HP reagent (Roche, cat. no. 06366236001)). Viral particle-containing cell culture supernatant was collected 48–72 h after transfection, filtered through a 0.45-µm PVDF filter (Millipore, cat. no. SLHV033RS) and then used for lentiviral transduction.

29Cells were seeded on 12- or 6-well plates in medium supplemented with 10 µg ml–1 protamine sulfate and lentivirus was incubated with cells overnight. The next day, the cell culture medium was replaced with fresh medium containing appropriate antibiotics, such as puromycin (Gibco, cat. no. A11138-03; 1 µg ml–1), blasticidin (Invitrogen, cat. no. A1113903; 10 µg ml–1) or G418 (Invitrogen, cat. no. 10131-035; 1 mg ml–1)) and cells were cultured until non-transduced cells were dead.

30sgRNAs were designed to target critical exons of the genes of interest, and gene knockout was confirmed by western blotting. sgRNAs were cloned into BsmBI-digested lentiCRISPRv2-blast, lentiCRISPRv2-puro and lentiGuide-neo vectors (Addgene, cat. nos. 98293, 98290 and 139449). All sequences for sgRNAs are listed in Supplementary Table 1.

31To generate knockout cells, MDA-MB-436, 786-O, A375, H460, B16F10 and 4T1 cells were transiently co-transfected with the desired sgRNAs expressed from lentiCRISPRv2-blast and lentiCRISPRv2-puro using X-tremeGENE HP reagent as described previously,. One day after transfection, selection was started with puromycin (1 µg ml–1) and blasticidin (10 µg ml–1). After selection for 2–3 days, single-cell clones were isolated, and knockout clones were validated by immunoblotting and sequencing of genomic DNA.

32To generate Dox-inducible knockout cells, lentiviruses from pCW-Cas9-blast (Addgene, cat. no. 83481) were transduced into HT-1080 cells followed by selection and establishment of single-cell clones as described previously. Lentiviruses generated from lentiGuide-neo vectors harbouring sgRNAs targeting FSP1 were transduced into HT-1080 pCW-Cas9-blast cells, followed by selection with G418 as above. After Dox induction, loss of FSP1 expression was confirmed by immunoblotting.

33To generate Dox-inducible FSP1–EGFP-expressing cells, H460 FSP1-knockout cells were transduced with lentivirus (pCW-FSP1WT-EGFP-blast or pCW-FSP1Q319K-EGFP-blast). After Dox treatment of cells, scalable FSP1 expression was confirmed by immunoblotting.

34Gpx4-knockout 4T1 cells and Gpx4 and Fsp1 double-knockout B16F10 cells were transfected with 141-IRES-puro, 141-hFSP1WT or hFSP1Q319K-IRES-puro and 141-mFsp1-IRES-puro vectors using X-tremeGENE HP reagent. One day after transfection, selection was started with puromycin (1 µg ml–1) and Lip-1 was removed from the medium to select cells with stable FSP1 expression. To obtain cells with stable expression, cells were maintained under the selection condition.

35Recombinant hFSP1 protein (non-myr-FSP1) was produced in BL21 E. coli and purified by affinity chromatography with a Ni-NTA system as described previously.

36For protein isolation by pulldown, HEK293T cells were seeded on 10- or 15-cm dishes and transfected with constructs encoding EGFP–Strep-tagged protein. After washing with PBS, cells were lysed in LCW lysis buffer supplemented with protease and phosphatase inhibitor cocktail and 1 mM dithiothreitol (DTT). Cell extracts were collected with a cell scraper and centrifuged at 20,000g for 1 h at 4 °C. Supernatants were incubated with MagStrep ‘type3’ XT beads (Biozol, cat. no. 2-4090-002) at 4 °C for 1–2 h. Beads were washed twice with washing buffer (100 mM Tris-HCl, 150 mM NaCl and 1 mM EDTA). EGFP–Strep-tagged proteins were eluted with elution buffer (100 mM Tris-HCl, 150 mM NaCl, 1 mM EDTA and 50 mM biotin), followed by dilution to 3 µM with TBS (50 mM Tris-HCl and 150 mM NaCl). Protein concentration was estimated from the absorbance at 280 nm measured with a UV5Nano spectrophotometer (Mettler Toledo). The coefficient was calculated by using ExPASy ProtParam (https://web.expasy.org/protparam/).

37Myristoylated protein was obtained by coexpressing N-myristoyltransferase 1 (hsNMT1) with FSP1. E. coli BL21 cells were transformed with constructs encoding hsNMT1 (petCDF vector, spectinomycin resistance) and FSP1 (FSP1–EGFP with a C-terminal His6 tag, petM13, kanamycin resistance). Purification was performed as for wild-type FSP1 with a final step of gel filtration chromatography. Purified protein was obtained and confirmed by denaturing SDS gel. Confirmation of the presence of myristoylated protein was obtained with MS.

38Myristoylation was confirmed by LC–ESI–MS (Waters Synapt XS). Proteins were separated on an Acquity UPLC Protein BEH C4 column (0.4 ml min–1; buffer A, 0.1% formic acid in water; buffer B, 0.1% formic acid in acetonitrile), and data were analysed using Masslynx v.4.2 (Waters).

39For resazurin assays, enzyme reactions were prepared in TBS (50 mM Tris-HCl and 150 mM NaCl) containing 50 nM non-myr-FSP1, 200 μM NADH and inhibitor (iFSP1 or icFSP1). After the addition of 100 μM resazurin sodium salt (Sigma, cat. no. R7017), fluorescence intensity (F; excitation/emission wavelength of 540/590 nm) was measured every 1 min using a SpectraMax iD5 microplate reader with SoftMax Pro v.7 (Molecular Devices) at 37 °C. Reactions with an equivalent volume of DMSO and without resazurin were used to calculate IC50 values. Curve fitting and calculation of IC50 values were conducted using GraphPad Prism 9.

40For NADH consumption assays, enzyme reactions were prepared in PBS (Gibco, cat. no. 14190094) containing 25 nM non-myr-FSP1, 200 μM menadione (Sigma, cat. no. M5625) or 200 µM CoQ0 (Sigma, cat. no. D9150), and inhibitor (iFSP1 or icFSP1). After the addition of 200 μM NADH, absorbance at 340 nm at 37 °C was measured every 30 s using a SpectraMax M5 microplate reader (Molecular Devices). Reactions without NADH and without enzyme were used to normalize the results. Curve fitting was conducted using GraphPad Prism 9.

41Purified EGFP–Strep or hFSP1–EGFP–Strep tagged protein was diluted in TBS supplemented with 1 mM DTT. Purified Strep-tagged proteins were then mixed with PEG 3350 (Sigma, cat. no. P3640) and/or icFSP1; final concentrations of the proteins, PEG and icFSP1 are indicated in each figure legend. For confocal microscopy analysis, samples were immediately transferred onto objective slides and EGFP signal was quickly captured using an LSM880 microscope with Zen Black software (v.2.3, ZWISS) with a ×63 water-immersion objective. For confocal microscopy analysis, recombinant C-terminally GFP-tagged FSP1 and myr-FSP1 were measured at a 15 μM protein concentration in PBS (pH 7.4; 300 mM NaCl) or at a 10 μM protein concentration in PBS (pH 7.4; 150 mM NaCl). Confocal fluorescence microscopy was performed at 255 °C on a Leica TCS SP8 confocal microscope using a ×63 water-immersion objective. Samples were excited with a 488-nm laser (GFP) and imaged at 498–545 nm.

42To measure turbidity, different concentrations of non-myr-FSP1 and PEG were reconstituted in 10 µl in a 384-well plate and the absorbance at 600 nm was measured using a SpectraMax iD5 microplate reader (Molecular Devices). To show non-myr-FSP1 condensation in PCR tubes, images were acquired using a smartphone. Representative bright-field images of non-myr-FSP1 condensates on an objective slide were captured using an Eclipse Ts2 microscope (Nikon) with a ×40 objective.

43For sedimentation assays, recombinant non-myr-FSP1 was mixed with the same amount of TBS with 0% or 20% PEG and 1 mM DTT and samples were centrifuged at 2,500g for 5 min. The supernatant was collected in a new tube and the pellet was resuspended in TBS supplemented with 1 mM DTT. Supernatant and resuspended non-myr-FSP1 were collected by adding 6× SDS sample buffer and subsequently resolved by SDS–PAGE. One gel was subjected to western blotting and probed with anti-FSP1 (1:1,000; Santa Cruz, cat. no. sc-377120, AMID). The other gel was immediately stained with Coomassie staining solution (1 mg ml–1 Coomassie Brilliant Blue G-250 (Sigma, cat. no. 1154440025), 50% methanol and 10% acetic acid) for 15 min and then soaked in washing buffer (70% methanol and 7% acetic acid). The washing buffer was heated using a microwave, and the buffer was changed until protein bands gave clear signals.

44Saturation transfer difference experiments were performed on a Bruker Avance III HD spectrometer at 600-MHz 1H frequency using an H/N/C triple-resonance cryogenic probe. Spectra were recorded at 10 °C with 5 µM recombinant hFSP1 (mutant) and 100-fold molar excess of icFSP1 in PBS containing 150 mM NaCl, 1% (v/v) DMSO-d6 and 10% (v/v) D2O for deuterium-lock. The saturation time was 2.5 s, and the on and off frequencies were 0.68 and −17 ppm, respectively. NMR spectra were processed using Topspin 4.0.6 (Bruker).

45All confocal microscopy images were acquired using an LSM880 microscope (Zeiss) with a ×63 objective and the corresponding appropriate filter sets for fluorophores and analysed with Zen Blue software (v.3.2, ZWISS) or ImageJ/Fiji unless noted otherwise. Cells were seeded on µ-Slide 8-well slides (Ibidi, cat. no. 80826) 1 day before the experiments. The next day, the medium was changed to fresh cell culture medium supplemented with 2.5 µM icFSP1. After incubation for the indicated times, cells were fixed and stained according to the following procedure: fixation with 4% paraformaldehyde for 5–10 min; permeabilization and blocking for 15 min with 0.3% Triton X-100 and 10 mg ml–1 BSA in PBS; and incubation at 4 °C overnight with primary antibodies or undiluted supernatant for anti-AIFM2 (FSP1; clone 14D7, homemade). Antibody dilutions were as follows: 1:10 for anti-YPYDVPDYA-tag (HA; clone 3F10) and 1:100 for anti-calnexin (Abcam, cat. no. ab22595), anti-GM130 (clone EP892Y, Abcam, cat. no. ab52649), anti-EEA1 (clone C45B10, Cell Signaling Technology, cat. no. 3288) and anti-LAMP1 (clone H4A3, Santa Cruz, cat. no. sc-20011)) in primary antibody dilution buffer. Cells were further stained with appropriate fluorophore-conjugated secondary antibodies (1:500 dilution) and DAPI (1:10,000 dilution) in TBS-T or PBS for 1–2 h at room temperature, avoiding light. Finally, all samples were mounted in Aqua-/PolyMount (Polysciences, cat. no. 18606-20) and dried at 4 °C overnight. Staining of mitochondria, lipid droplets and aggresomes was performed using MitoTracker Red CMXRos (20 nM; Invitrogen, cat. no. M7512), LipidSpot 610 (1:1,000; Biotium, cat. no. 70069-T) and the Proteostat detection kit (1:10,000; Enzo, cat. no. ENZ-51035-0025), respectively, according to the manufacturer’s protocol.

46Pfa1 cells (20,000 cells) were seeded on µ-Slide VI 0.4 slides (Ibidi, cat. no. 80606) 1 day before the experiments. The next day, the medium was changed to high-glucose DMEM supplemented with 10% FBS, 2 mM l-glutamine, 1% penicillin-streptomycin, 2.5 µM icFSP1 and 10 mM HEPES. After incubation with icFSP1 for 2–4 h, 2–5 rectangular areas that each contained more than three FSP1 condensates were selected as bleaching areas. One image acquired before bleaching was considered to correspond to time ‘0’. Subsequently, the selected areas were photobleached using the maximum intensity of the laser and FRAP was monitored at minimum intervals (~5 s) using an LSM880 microscope (Zeiss).

47To quantify the FRAP rate, a region of interest (ROI) for each condensate (i) in the photobleached area and one condensate (c) in a non-photobleached area was determined using ImageJ/Fiji and the mean fluorescence intensity of condensate i at time t, Fi(t) was obtained. After determining each time of fluorescence value, Fi(t) was normalized by the value of Fi(0) to obtain the relative fluorescence (RFi(t)) of each bleached condensate area. To reflect quenching effects during observation and photobleaching, each RFi(t) value was normalized by relative fluorescence value at time t of condensate (c) in non-bleached condensate areas (RFc(t)) as follows: Fi(t) = RFi(t)/RFc(t) = (Fi(t)/Fi(0))/(Fc(t)/Fc(0)). Finally, the FRAP rate (%) at time t in the particles was calculated as the mean of Fi(t) × 100 as described previously.

48For co-staining or washout analyses, Pfa1 cells (15,000–30,000 cells) were seeded on µ-Dish 35-mm low dishes (Ibidi, cat. no. 80136) and incubated overnight. The next day, the cell culture medium was changed to FluoroBrite DMEM (Gibco, cat. no. A1896701) supplemented with 10% FBS, 2 mM l-glutamine and 1% penicillin-streptomycin. Live-cell microscopy was performed using the 3D Cell Explorer (Nanolive) with Eve v.1.8.2 software and the corresponding appropriate filter sets. During imaging, cells were maintained at 37 °C and 5% CO2 using a temperature-controlled incubation chamber. For co-staining analysis, cells were pretreated for 1 h with 5 µM Liperfluo (Dojindo, cat. no. L248-10) and then changed to FluoroBrite DMEM containing 0.2 µg ml–1 propidium iodide (Sigma, cat. no. P4170) and acquisition was started using Nanolive. After recording one image, 1 mM of icFSP1 in FluoroBrite DMEM was added to the dishes (final concentration of 10 µM) while continuing to record images. Images were acquired every 10 min for more than 4 h, and GFP, BFP and RFP filter sets were used to acquire signal. For washout experiments, high-glucose DMEM was changed to FluoroBrite DMEM before the experiments followed by data acquisition using Nanolive. After recording a few images, 0.25 mM of icFSP1 in FluoroBrite DMEM was added to the dishes (final concentration of 2.5 µM) and recording of images continued for 4 h. Thereafter, the dishes were carefully washed once with fresh FluoroBrite DMEM without icFSP1 and refilled with medium. Image acquisition was then immediately restarted. Images were recorded every 5 min for one more hour; that is, the total duration of data acquisition was around 5 h.

49To determine the number of condensates in cells, Pfa1 cells (15,000–20,000 cells) were seeded on µ-Slide 8-well slides (Ibidi, cat. no. 80826) and incubated overnight. The next day, the medium was changed to high-glucose DMEM supplemented with 10% FBS, 2 mM l-glutamine, 1% penicillin-streptomycin, 2.5 µM icFSP1 and Hoechst. Immediately thereafter, the focus was adjusted and Hoechst and EGFP images were recorded using an Axio Observer Z1 imaging system with VisView v.4.0 (Visitron Systems, ZWISS) with a ×20 air objective and a CCD camera (CoolSnap ES2, Photometrics) with the corresponding filter sets. During imaging, cells were maintained at 37 °C and 5% CO2 using a temperature-controlled incubation chamber. The imaging software ImageJ/Fiji was used for visualization, and CellProfiler (v.4.1.3, Broad Institute) was used to count the condensates in each cell.

50All mice were obtained from Charles River. For syngeneic subcutaneous tumour experiments, Gpx4 and Fsp1 double-knockout B16F10 cells stably overexpressing hFSP1–HA (1 × 106 cells in 100 µl PBS) were injected subcutaneously into the right flank of 7-week-old female C57BL/6J mice. After tumours reached approximately 25–50 mm3 in size, mice were randomized and treated with vehicle or icFSP1 (50 mg kg–1, Intonation) by intraperitoneal injection twice daily for 4–5 days. To generate tumour samples for staining, Gpx4 and Fsp1 double-knockout B16F10 cells stably expressing hFSP1WT–HA or hFSP1Q319K–HA (1 × 106 cells in 100 µl PBS) were injected subcutaneously into the right flank of 7-week-old female C57BL/6J mice. After tumours reached approximately 25 mm3 in size, mice were randomized and treated with vehicle or icFSP1 (50 mg kg–1, Intonation) by intraperitoneal injection twice daily.

51For xenograft subcutaneous tumour experiments, GPX4-knockout A375 cells (5 × 106 cells in 100 µl PBS) were injected subcutaneously into the right flank of 7-week-old female athymic nude mice. After tumours reached approximately 25–100 mm3 in size, mice were randomized and treated with vehicle or icFSP1 (50 mg kg–1, Intonation) by intraperitoneal injection twice daily for the first 4 days and once daily afterwards.

52For xenograft subcutaneous tumour experiments, GPX4-knockout H460 cells (5 × 106 cells in 100 µl PBS) were injected subcutaneously into the right flank of 6-week-old female athymic nude mice. After tumours reached approximately 100 mm3 in size, mice were randomized and treated with vehicle or icFSP1 (50 mg kg–1, Intonation) by intraperitoneal injection twice daily.

53icFSP1 was dissolved in 45% PEG E 300 (Sigma, cat. no. 91462-1KG) and 55% PBS (Gibco, cat. no. 14190094). Tumours were measured by calliper every day, and tumour volume was calculated using the following formula: tumour volume = length × width2 × 0.52. When the tumour was greater than 1,000 mm3 in size at measurement or the tumour became necrotic, tumours were considered to have reached the humane endpoint. When tumours reached the humane endpoint, the experiment was stopped and no further study was conducted.

54Dissected tissues were fixed in 4% paraformaldehyde in PBS overnight at 4 °C. For immunofluorescence staining, fixed tissues were incubated in 20% sucrose in PBS overnight at 4 °C, followed by embedding in OCT mounting compound (TissueTek, Sakura) on dry ice and storage at –80 °C. Frozen tissues were cut into 5-µm-thick sections using a Cryostat Microm HM 560 (Thermo Fisher Scientific) at –30 °C. Tissue sections were postfixed with 1% paraformaldehyde in PBS for 10 min and subsequently fixed with 67% ethanol and 33% acetic acid for 10 min. Sections were incubated with blocking solution (5% goat serum and 0.3% Triton X-100 in PBS) for 30 min and then incubated with primary antibodies (anti-HA (clone, 3F10; 1:10; developed in house), anti-4-HNE (JaICA, cat. no. HNEJ-2; 1:50) and anti-AIFM2 (FSP1, clone 14D7; undiluted; developed in house)) diluted in blocking solution overnight at 4 °C. The next day, sections were incubated with appropriate fluorophore-conjugated secondary antibodies (goat anti-rat Alexa Fluor 488 IgG (H+L) (1:500; A-11006, Invitrogen), goat anti-mouse IgG H&L Alexa Fluor 647 (1:500; ab150115, Abcam) and donkey anti-rat IgG Alexa Fluor 555 (1:500; ab150154, Abcam)) in secondary dilution buffer (1% BSA and 0.3% Triton X-100 in PBS) for 2 h at room temperature. DNA was visualized with DAPI staining for 5 min, and slides were mounted in Aqua-/PolyMount. Images were obtained using an LSM880 microscope (Zeiss) and analysed with Zen Blue or ImageJ/Fiji software.

55All studies were performed by Bienta/Enamine Ltd.

56All data shown are the mean ± s.e.m. or mean ± s.d., and the number (n) in each figure legend represents biological or technical replicates as specified. All experiments (except those described otherwise in the legend) were performed independently at least twice. For mouse experiments, at least three animals were included per group once or twice. Two-tailed Student’s t tests and one-way or two-way ANOVA followed by Bonferroni’s, Dunnett’s, Tukey’s or Sidak’s multiple-comparison tests were performed using GraphPad Prism 9 (GraphPad Software) (also see figure legends for more detail). The results of the statistical analyses are presented in each figure. P < 0.05 was considered to be statistically significant.

57Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

58Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41586-023-06255-6.