The cell biology of ferroptosis

The execution and regulation of ferroptosis at the plasma membrane

Plasma membrane rupture is the terminal event in many forms of cell death. For some forms of non-apoptotic cell death, such as necroptosis and pyroptosis, this terminal process involves permeabilization of the plasma membrane by pore-forming proteins such as mixed lineage kinase domain-like pseudokinase (MLKL), gasdermin D and ninjurin 1 (refs. 11,12). Ninjurin 1 may facilitate the execution of ferroptosis in some but not all cells^9,13. Otherwise, there is little evidence that the execution of ferroptosis requires a specific pore-forming protein. Rather, ferroptosis execution and regulation involve a distinct set of lipid-centric mechanisms that govern plasma membrane integrity.

Ferroptosis execution at the plasma membrane

The execution of ferroptosis involves the peroxidation of PUFA-containing phospholipids and the formation of lipid hydroperoxides^14,15. Once formed, lipid hydroperoxides can react with iron to generate highly reactive lipid radicals and become further modified or truncated in ways that likely contribute to the execution of ferroptosis¹⁶. Lipid hydroperoxides may be synthesized by lipoxygenase enzymes¹⁷ or formed through chemical reactions between PUFAs, soluble reactive oxygen species (ROS) and iron¹⁰. It is controversial which mechanism is most important for ferroptosis and it seems likely that both mechanisms are relevant in one cell type or another^17–20. Within the membrane, dozens of different PUFA-containing phospholipid species can be peroxidized to contribute to the execution of ferroptosis, with PUFA-phosphatidylethanolamines being especially important in many cells^14,21–26. When cells are exposed to ferroptosis-inducing conditions, lipid peroxides seem to initially accumulate within the endoplasmic reticulum (ER) and other organelles and subsequently at the plasma membrane prior to membrane rupture^27–30 (Box 2).

Events downstream of plasma membrane lipid peroxidation that promote ferroptosis execution have recently been clarified (Fig. 1b). Increasing lipid peroxidation raises membrane tension, which in turn activates Piezo1 and transient receptor potential (TRP) mechanosensitive ion channels⁹. Channel opening leads to Ca²⁺ and Na⁺ influx and K⁺ efflux. These ion fluxes are enhanced by the simultaneous inactivation of the plasma membrane Na⁺/K⁺ ATPase, possibly due to altered interactions between this protein and peroxidized plasma membrane phospholipids⁹. Collectively, these changes lead to a loss of ionic homeostasis and osmotic cell swelling that eventually result in plasma membrane rupture. Plasma membrane rupture can be transiently delayed by polyethylene glycol molecules that block membrane nanopores to prevent the flow of ions and water across the plasma membrane^31,32. However, polyethylene glycol molecules only delay the onset of ferroptosis and do not completely prevent plasma membrane rupture over longer periods of time³¹. It seems likely that, as more lipids become peroxidized, this disrupts the biophysical properties of the plasma membrane³³ so severely that plasma membrane rupture becomes inevitable.

Suppression of membrane lipid peroxidation

Ferroptosis is normally inhibited by the continuous activity of coupled enzyme-metabolite systems that prevent the accumulation of membrane lipid peroxides to toxic levels. These systems are, in fact, the targets of small molecules that helped elucidate the ferroptosis mechanism in the first place (see Box 2 for a description of common ferroptosis inducers and inhibitors). The enzyme glutathione peroxidase 4 (GPX4) reduces potentially toxic lipid hydroperoxides to less dangerous lipid alcohols^34,35. GPX4 appears to be the most critical enzyme for preventing lipid hydroperoxide accumulation in most cells^35,36 (Fig. 1c). Several isoforms of GPX4 exist, including those designated as cytosolic, mitochondrial and nuclear. Cytosolic GPX4 seems the most essential for preventing ferroptosis, although mitochondrial GPX4 may also help inhibit ferroptosis in some cases^37–39. Cytosolic GPX4 does not contain an obvious plasma membrane-targeting element. However, modelling studies indicate that a patch of positively charged lysine and arginine residues may facilitate the electrostatic interactions with phospholipids that help keep GPX4 in the vicinity of the plasma membrane⁴⁰.

Ferroptosis suppressor protein 1 (FSP1; also known as AIFM2) inhibits plasma membrane lipid peroxidation in parallel to GPX4 (Fig. 1c). FSP1 is an NAD(P)H- and FAD-dependent oxidoreductase that reduces coenzyme Q10 (CoQ)^41,42 and vitamin K⁴³. The reduced forms of CoQ and vitamin K act as lipophilic radical-trapping antioxidants (RTAs) that can terminate lipid peroxidation chain reactions. FSP1 localizes specifically to the plasma membrane and to lipid droplets^41,42. During translation, the N-terminal initiator methionine of FSP1 is removed and the 14-carbon fatty acid myristate is covalently attached to the subsequent glycine residue^41,42. The irreversibly conjugated myristate anchors FSP1 to the plasma membrane and is necessary for its anti-ferroptotic activity^41,42. Targeting FSP1 selectively to the plasma membrane is sufficient to suppress ferroptosis, implying that this is the key site where endogenous RTAs function to suppress ferroptosis⁴¹. Co-immunoprecipitation and biochemistry data support the possibility that human FSP1 exists as a dimer⁴⁴, but whether this is the functional state of FSP1 at the membrane is unknown. In addition to CoQ and vitamin K, other endogenous metabolites can serve as RTA inhibitors of ferroptosis, including tetrahydrobiopterin^23,45, vitamin E and vitamin A⁴⁶. The relative contribution of these different metabolic ferroptosis suppressors appears to vary between cell types.

Plasma membrane repair as an anti-ferroptotic mechanism

Plasma membrane repair processes oppose the terminal execution of ferroptosis. As noted above, lipid peroxidation can trigger an increase in intracellular Ca²⁺ levels³¹. Increased intracellular Ca²⁺ acts as a signal to recruit charged multivesicular body protein 5 (CHMP5) and CHMP6 — the components of the endosomal sorting complexes required for transport (ESCRT)-III — to the plasma membrane⁴⁷, where they participate in local membrane repair. Indeed, genetic silencing of CHMP5, CHMP6 or CHMP4B can enhance ferroptosis sensitivity to some degree^31,47 (Fig. 1b). However, membrane repair mediated by CHMP4B, CHMP5 and CHMP6 can be overwhelmed by high levels of lipid peroxidation^31,47. Of note, the ESCRT-III complex can also limit protein pore-dependent membrane damage during necroptosis and pyroptosis^48,49 and is therefore a negative regulator of several forms of non-apoptotic cell death.

Ferroptosis regulation by plasma membrane transporters and enzymes

Integral plasma membrane proteins can contribute to the execution and regulation of ferroptosis in many ways, including through the generation of ROS^1,50,51, and to the mediation of iron or metabolite trafficking^52–57. Here, we focus on proteins that are involved in the import of cysteine and selenium, two metabolites that contribute essentially to the negative regulation of ferroptosis (see also next section) (Fig. 1c).

Cysteine is required to synthesize sulfur-containing anti-ferroptotic molecules, including the GPX4 cofactor (reduced) glutathione (GSH)⁵⁸. The cysteine disulfide, cystine, can be imported by the cell surface system x_c⁻ cystine–glutamate antiporter. Cysteine can also be acquired indirectly via endocytosis of extracellular cysteine-rich proteins, such as albumin, and their subsequent lysosomal catabolism⁵⁹. Extracellular GSH can also directly fuel de novo intracellular GSH synthesis via the γ-glutamyl cycle. Plasma membrane-localized γ-glutamyltransferase 1 (GGT1) cleaves the usual γ-glutamyl bond that links glutamate to cysteine in the GSH tripeptide. The resulting cysteinylglycine dipeptide can then be imported and further catabolized by dipeptidases, including carnosine dipeptidase 2 (CNDP2), to release cysteine for use in GSH synthesis⁶⁰. Some brain cancer cells overcome the need for system x_c⁻-mediated cystine import via GGT1-mediated catabolism of extracellular GSH⁶¹. Moreover, in a rare kidney cancer (chromophobe renal cell carcinoma), the absence of GGT1 function renders these cells exquisitely dependent upon extracellular cystine for survival relative to normal tissues⁶². This helps illustrate how differences in protein expression between cells can result in unique sensitivities to pro-ferroptotic stimuli such as cystine deprivation.

Selenium is found within the active-site selenocysteine of GPX4 and a small number of other enzymes. In cancer cells, high levels of GPX4 translation promote ferroptosis resistance but, under conditions of low selenium, this GPX4 synthesis is inhibited by ribosome stalling and collisions, which causes early GPX4 translation termination (see also next section)⁶³. Selenium can enter the cell by endocytosis of the liver-secreted, selenium-containing carrier protein selenoprotein P, which is bound and endocytosed by cell surface receptors such as the LDL receptor-related protein 8 (LRP8)^63–65. In triple-negative breast cancer cells, LRP8 knockout sensitizes cells to ferroptosis inducers⁶³. In MYCN-amplified neuroblastoma cells, loss of LRP8 alone is sufficient to trigger ferroptosis, likely due to coincident low expression of system x_c⁻ subunits⁶⁵. Selenium can also enter the cell via the conversion of inorganic selenium (that is, selenite) to volatile selenide through a process that involves system x_c⁻-dependent cystine uptake, which provides thiols that can reduce extracellular inorganic selenium to selenide, which is then presumed to diffuse across the plasma membrane⁶⁶. How cystine and selenium uptake and metabolism are coordinated to ensure proper GPX4 function is an interesting open question.

Mitochondria

Mitochondria are multifunctional organelles that can regulate ferroptosis sensitivity in several different ways (Fig. 2). In cells treated with the system x_c⁻ inhibitor erastin, mitochondria are smaller, with disorganized cristae^1,80. This represents a potential morphological marker of ferroptosis. However, it is not clear whether changes in mitochondrial morphology contribute to the execution of ferroptosis or are merely a correlate of this process. For example, mitochondria can be eliminated from some cells with little effect on ferroptosis sensitivity⁸¹. In other cells, mitochondria seem to be essential for ferroptosis⁸². Below, we describe examples of the pro-ferroptotic and anti-ferroptotic effects of mitochondria, touching on specific examples of context-dependent effects and areas of disagreement in the literature.

Several mitochondrial activities reduce ferroptosis sensitivity. Mitofusin 1 (MFN1)-mediated mitochondrial fusion may reduce sensitivity to RSL3-induced ferroptosis, at least in cultured cancer cells⁸³. Likewise, in cells with dysfunctional mitochondria, activation of the mitochondrial stress response pathway mediated by metalloen-dopeptidase OMA1 and DAP3 binding cell death enhancer 1 (DELE1) can inhibit ferroptosis and delay cardiomyopathy⁸⁴. Here, DELE1 activates cytosolic activating transcription factor 4 (ATF4), which in turn promotes GSH synthesis and GPX4 protein stability. Mitochondria are also the site of CoQ metabolite synthesis⁸⁵. CoQ is an electron carrier in the mitochondrial electron transport chain (ETC) but also functions as an important anti-ferroptotic RTA^41,42. Key enzymatic machinery necessary for CoQ synthesis (for example, the COQ proteins) localize to the mitochondria. A specific proteolytically processed version of StAR-related lipid transfer protein 7 (STARD7) transports CoQ from the mitochondria to the plasma membrane, where it can be reduced by FSP1 to inhibit lipid peroxidation and ferroptosis⁸⁶. Whether STARD7 might act at membrane contact sites between the mitochondria and plasma membrane to promote efficient and directional CoQ transfer is unknown. Another important protective role for mitochondria is in the breakdown of PUFAs that could otherwise be incorporated into membrane phospholipids, peroxidized and thereby promote ferroptosis. This catabolic process requires the mitochondrial enzyme 2,4-dienoyl-CoA reductase 1 (DECR1) and, for reasons that remain unclear, appears especially relevant in prostate cancer^87,88. Collectively, these mitochondrial functions would tend to reduce ferroptosis sensitivity.

Other mitochondrial mechanisms increase ferroptosis sensitivity. For example, the mitochondrial tricarboxylic acid (TCA) cycle promotes ferroptosis sensitivity, at least in response to cystine deprivation⁸⁹. During glutaminolysis, glutamine is converted to glutamate and the TCA cycle intermediate α-ketoglutarate within mitochondria; both glutamate and α-ketoglutarate can promote ferroptosis^53,73,90. In these cases, the precise effector mechanism linking TCA cycle activity with the induction of ferroptosis remains obscure, although enhanced oxidative stress seems to be involved. Mitochondria are also central hubs for the use and metabolism of iron². Increased mitochondrial iron uptake mediated by mitoferrin 1 (MFRN1; encoded by SLC25A37) can enhance ferroptosis sensitivity, although the exact mechanism is unclear⁹¹. Disruption of mitochondrial biosynthesis of Fe–S clusters can also enhance ferroptosis sensitivity. Here, depletion of NFS1 cysteine desulfurase lowers the synthesis of Fe–S clusters, triggering a feedback iron starvation response that increases iron import and heightens ferroptosis sensitivity⁹². It might be the case that cystine deprivation caused by system x_c⁻ inhibition has unique effects on mitochondrial Fe–S biosynthesis given the need for cysteine-derived sulfur in this process. One model to explain observed differences in the ferroptosis phenotypes caused by cystine deprivation versus direct GPX4 inhibition¹⁰ is that cystine deprivation uniquely perturbs mitochondrial Fe–S cluster biosynthesis⁴⁵. Finally, mitochondrial outer membrane permeabilization, when it occurs in a partial and non-lethal manner throughout the cell, results in both activation of the ATF4 pathway and enhanced sensitivity to ferroptosis induced by GPX4 inhibitors⁹³. This may suggest that the loss of one or more mitochondrial anti-ferroptotic functions can, on balance, overcome the protective effect of ATF4 pathway activation.

In some cancer cells, mitochondrial ETC activity promotes ferroptosis, at least in response to cystine deprivation, as shown using specific ETC poisons⁸⁹. How the ETC promotes ferroptosis is not clear. One obvious candidate is the production of ROS that might be expected to initiate lipid peroxidation. Indeed, ROS scavengers that target mitochondria can attenuate ferroptosis in some models, implying that mitochondria-derived oxidants may contribute to cell death (Box 2). However, in other cancer cells, depleting the mitochondrial DNA, resulting in complete loss of ETC function, has no effect on ferroptosis sensitivity^1,94. Moreover, ETC function is necessary for the activity of two mitochondrially localized enzymes that may be able to generate reduced CoQ: dihydroorotate dehydrogenase (DHODH)³⁷ and glycerol-3-phosphate dehydrogenase 2 (GPD2)⁹⁵. Accordingly, loss of ETC function might be expected to enhance ferroptosis sensitivity by disrupting the synthesis of reduced mitochondrial CoQ. Overall, it seems likely that the role of the mitochondrial ETC varies by cell type and by the nature of the lethal stimulus. An emerging (and controversial)³⁸ concept is that the mitochondria might assume a greater importance in overall ferroptosis regulation when cytosolic anti-ferroptotic systems are disrupted^37,39,96. This model of a conditional role for mitochondria depending on circumstances could provide a reconciliation for some conflicting observations.

Endoplasmic reticulum

The ER plays an important role in ferroptosis regulation. POR and CYB5R1 are enzymes that reside in the ER membrane and use NADPH to produce ROS that can promote membrane lipid peroxidation^76,77 (Fig. 3a). Indeed, there is good evidence that the ER is one site where membrane lipid peroxidation is initiated during ferroptosis (Box 2). As noted above, whether lipid peroxidation spreads from the ER to the plasma membrane to execute ferroptosis, for example, through vesicular trafficking or phospholipid transfer at membrane contact sites, is an open question.

Ferroptosis is a lipid-dependent process. The ER is a major hub for lipid metabolism in the cell. Lipid desaturation, phospholipid synthesis, phospholipid remodelling and biogenesis of lipid droplets are all ER-localized activities (Fig. 3b). Thus, it is no surprise that ER-resident lipid metabolic enzymes contribute importantly to the regulation of ferroptosis sensitivity. Overall, the ferroptotic threshold of the cell can be determined by lipid composition and, more specifically, by the relative levels of different lipid species⁹⁷. Decreasing the ratio of more oxidizable PUFAs to less oxidizable monounsaturated fatty acids (MUFAs) is sufficient to convert cells from ferroptosis sensitive to ferroptosis resistant^28,98–101 whereas increasing the PUFA to MUFA ratio can have the opposite effect^22,77. ER-resident enzymes shape the PUFA and MUFA content of cellular lipid pools. For example, fatty acid desaturase 1 (FADS1) and FADS2, which introduce carbon–carbon double bonds into fatty acids, can enhance or reduce ferroptosis sensitivity, respectively, by synthesizing specific PUFAs or MUFAs^102,103. The overall activity of these desaturation processes can be governed within the ER by calcium levels, which are regulated by the tetraspanin membrane-spanning 4-domains subfamily A member 15 (MS4A15)¹⁰⁴.

Acyl-CoA synthetase long-chain (ACSL) enzymes ‘activate’ both PUFAs and MUFAs to PUFA-CoA and MUFA-CoA species, which can then be incorporated into membrane phospholipids, triacylglycerols (TAGs) or other lipids. ACSL4 and ACSL3 appear generally to function in opposition, with ACSL4-driven PUFA metabolism increasing ferroptosis sensitivity and ACSL3-dependent MUFA metabolism promoting ferroptosis resistance^{24,28,30,99,105,106}. However, exceptions to this general scheme exist, for example, the pro-ferroptotic role of ACSL3 in KRAS-mutant lung cancer cells¹⁰⁷. Further work is required to establish where in the cell these enzymes function to regulate ferroptosis. Genetic rescue data suggests that ER-localized ACSL4 may be critical for ferroptosis regulation²⁷. However, ACSL4 can also sometimes be found in lipid droplets, the Golgi apparatus or at the plasma membrane^24,108,109. ACSL3 can also localize to both the ER and to lipid droplets¹¹⁰. Which of these localizations is most important for ferroptosis regulation remains somewhat mysterious and could differ by cell type. How the activities of these enzymes at the ER impact plasma membrane lipid composition and peroxidation²⁷ is also not at all clear.

The ER is the site of de novo phospholipid synthesis¹¹¹. There is evidence from large-scale genetic screens that ER-resident enzymes involved in de novo phospholipid synthesis, including glycerol-3-phosphate acyltransferase 4 (GPAT4), can promote ferroptosis sensitivity^27,111. However, the role of de novo phospholipid synthesis is less explored than the role of phospholipid remodelling (that is, the Lands cycle). The Lands cycle involves cleavage of one acyl chain of a phospholipid by a phospholipase, followed by reacylation of the lysophospholipid with an acyl-CoA to reform the phospholipid. Calcium-independent phospholipase A2 (also known as PLA2G6, PNPLA9 or iPLA2β) can cleave oxidized PUFA acyl chains from phospholipids, thereby directly limiting the spread of lipid peroxidation and the onset of ferroptosis^112–114. A related enzyme, PLA2G4C, may also negatively regulate ferroptosis in KRAS-mutant lung cancer in a similar manner¹⁰⁷. A distinct lysophospholipase enzyme, abhydrolase domain-containing protein 12 (ABHD12), cleaves PUFA-containing lysophosphatidylserine and oxidized phosphatidylserines specifically, and inhibition of this enzyme is sufficient to increase PUFA-phosphatidylserine (that is, 18:0/20:4 phosphatidylserine) levels and boost ferroptosis sensitivity. Lysophospholipids are reacylated (or, in de novo synthesis, acylated) by phospholipid acyltransferases²². For example, reacylation of lysophospholipids with PUFAs is catalysed by ER-resident enzymes, including lysophosphatidylcholine acyltransferase 3 (LPCAT3; also known as MBOAT5), 1-acylglycerol-3-phosphate O-acyltransferase 3 (AGPAT3)^25,105,115 and MBOAT7 (ref. 24), while ER-resident MBOAT1 and MBOAT2 acylate lysophospholipids with MUFAs^116,117. Presumably, there is competition between phospholipid acyltransferases to reacylate (or acylate) the same substrates, and the relative levels of different enzymes and substrates will help determine the overall composition of the membrane and, in turn, the ferroptosis sensitivity of the cell. Although LPCAT3 is an ER-resident enzyme¹¹⁸, the phospholipase PLA2G6 may be found through the cytosol with the possibility of being recruited to the vicinity of the plasma membrane¹¹⁹. Thus, it is also unclear whether all ferroptosis-relevant lipid synthesis and remodelling events occur in the ER, or whether localized remodelling in other endomembrane compartments and/or at the plasma membrane also helps shape ferroptosis sensitivity.

In addition to enzymes involved in phospholipid metabolism, the ER houses the enzymes of the mevalonate pathway (Fig. 3c). Although this pathway is best known for the synthesis of cholesterol, several intermediates within this pathway are important in ferroptosis suppression. For example, an intermediate in cholesterol synthesis, 7-dehydrocholesterol, is highly prone to oxidation and can act as a sacrificial endogenous RTA to limit membrane lipid peroxidation^120,121. A different metabolite, squalene, has also been implicated in ferroptosis suppression¹²², but the mechanism remains unclear. Another product of the mevalonate pathway is farnesyl pyrophosphate, a key precursor for the synthesis of vitamin K in the Golgi¹²³ and CoQ in the mitochondria⁸⁵. Finally, another mevalonate pathway product, isopentenyl pyrophosphate is necessary for isopentylation of the Sec-tRNA and it thereby modulates selenoprotein translation. Whether isopentylation impacts GPX4 levels to an extent that would affect ferroptosis sensitivity remains to be directly tested¹²⁴.

Beyond its roles in lipid synthesis, the ER is a site for the processing of important transcription factors that modulate the cellular lipid and redox environments (Fig. 3d). Sterol regulatory element-binding proteins (SREBPs) are ER-resident transcription factors that regulate the expression of many enzymes related to lipid metabolism. Tethered to the ER by two transmembrane domains, SREBPs undergo regulated trafficking to the Golgi, where they are proteolytically processed to release the active transcription factor, which traffics to the nucleus to initiate transcription. A signalling pathway involving phosphatidylinositol 3-kinase (PI3K), AKT and mechanistic target of rapamycin complex 1 (mTORC1), can activate SREBP1 to protect cancer cells from ferroptosis¹²⁵. SREBP1-dependent expression of stearoyl-CoA desaturase 1 (SCD1) is required for this effect¹²⁵. SCD1 catalyses the synthesis of long-chain MUFAs (for example, oleate); thus, increasing MUFA production should reduce membrane susceptibility to peroxidation and ferroptosis sensitivity. A second transcription factor that is processed via the ER is nuclear factor erythroid 2-related factor 1 (NFE2L1). NFE2L1 undergoes a complex series of post-translational modifications, including glycosylation in the ER and subsequent deglycosylation in the cytosol by N-glycanase 1 (NGLY1), as well as proteolytic cleavage by DNA-damage inducible 1 homologue 2 (DDI2), to yield an active transcription factor^126–129. The NGLY1–NFE2L1 axis helps inhibit ferroptosis in some cells by promoting GPX4 protein stability indirectly, possibly through regulation of proteasome gene expression or function^130,131.

Finally, the overall function of the ER may be impaired in cells undergoing ferroptosis, at least when induced by cystine deprivation. Cystine deprivation can trigger a global integrated stress response and ER stress response, including changes in gene expression and the viscosity and pH of the ER^58,132. Whether these changes contribute directly to the execution of ferroptosis is not known. However, certain changes, such as the transcriptional upregulation of SLC7A11 (which encodes the transport subunit of system x_c⁻)⁵⁸, might be expected to reduce ferroptosis sensitivity, while changes in viscosity or pH could very well modulate the function of one or more ER-resident enzymes or mechanisms noted above. In sum, the ER has a multifaceted role in ferroptosis regulation.

Lipid droplets

Lipid droplets are dynamic ER-derived lipid storage organelles that have a context-specific role in ferroptosis regulation¹³³ (Fig. 4). Lipid droplets consist of a neutral lipid core, composed mostly of TAG and steryl esters, encircled by a phospholipid monolayer containing integral and peripheral regulatory proteins¹³⁴. Different fatty acids can be stored in or released from lipid droplet-resident lipids with varying effects on ferroptosis sensitivity. For example, cancer cells grown in acidic conditions that mimic the tumour microenvironment accumulate lipid droplets, and inhibition of lipid droplet synthesis channels PUFAs away from TAGs and into phospholipids, sensitizing cells to ferroptosis¹³⁵. In mice, oral gavage with PUFA-rich oils together with blockade of lipid droplet biosynthesis increases PUFA-containing phospholipids and is sufficient to trigger ferroptosis in tumour xenografts¹³⁵. Conceptually similar, in patient-derived glioblastoma cells, if PUFAs are not sequestered in lipid droplets away from membrane phospholipids they can promote ferroptosis sensitivity¹³⁶. Here, storage of PUFAs in lipid droplets versus incorporation into membrane phospholipids is genotype dependent, with CDKN2A-null glioblastoma cells specifically showing reduced PUFA sequestration in lipid droplets and enhanced ferroptosis induction¹³⁶. Finally, in cell-cycle-arrested cancer cells, PUFAs can be channelled into lipid droplets, promoting ferroptosis resistance¹³⁷. Thus, in some cells, the preferential sequestration of PUFAs in lipid droplets, away from membrane phospholipids, limits ferroptosis sensitivity.

However, lipid droplets do not always seem to protect against ferroptosis. In clear cell renal cell carcinoma cells, hypoxia-inducible factor 2α (HIF2α; also known as EPAS1) regulates the expression of hypoxia-inducible lipid droplet-associated protein (HILPDA). HILPDA inhibits patatin-like phospholipase domain-containing 2 (PNPLA2; also known as ATGL), the rate-limiting enzyme in TAG lipolysis. HILPDA expression increases lipid droplet numbers and PUFA-TAG levels, concomitant with increased ferroptosis sensitivity¹³⁸. Conversely, HILPDA depletion reduces ferroptosis sensitivity¹³⁸. Furthermore, system x_c⁻ inhibition together with dehydroascorbic acid (that is, oxidized vitamin C) treatment can trigger ferroptosis that involves the oxidation of lipid droplets¹³⁹. In this scenario, inhibition of diacylglycerol O-acyltransferases (DGATs; the rate-limiting enzymes in TAG synthesis) prevents ferroptosis¹³⁹. Thus, lipid droplets do not always serve as a protective enclosure that prevents PUFA oxidation. Oxidized TAG species have been measured by lipidomics¹⁴⁰, but whether the oxidation of TAGs is common and whether oxidized PUFAs are released from TAGs to propagate ferroptosis is unclear. The mechanisms that influence TAG oxidation are also unknown. Given the localization of FSP1 to lipid droplets one can speculate that FSP1 recycles RTAs at this site to protect TAGs from oxidation, but this possibility remains untested. Furthermore, the mechanisms that determine PUFA sequestration or release via lipolysis or lipophagy are also outstanding questions.

Finally, in some contexts, lipid droplet formation does not affect ferroptosis sensitivity positively or negatively. For example, in fibrosarcoma cells incubated with extracellular MUFAs, the number of lipid droplets is increased, but disrupting lipid droplet synthesis using small-molecule DGAT1 and DGAT2 inhibitors does not alter ferroptosis sensitivity or the ability of exogenous MUFAs to inhibit ferroptosis²⁸. Inhibition of the DGAT enzymes also has no effect on ferroptosis sensitivity in an osteosarcoma cell line⁴¹, further highlighting the context-specific role of lipid droplets in ferroptosis.

Peroxisomes

Peroxisomes have important roles in several metabolic processes, including very-long-chain fatty acid breakdown and the catabolism of hydrogen peroxide by catalase. However, peroxisomes are currently best understood to contribute to the regulation of ferroptosis sensitivity via the synthesis of ether phospholipids. PUFA-containing ether phospholipids are abundant in the plasma membrane, and these lipids contribute to ferroptosis execution in some contexts. Ether lipid synthesis is a multistep process that involves collaboration between several compartments of the cell. A key precursor molecule, 1-O-alkyl-glycerol-3-phosphate (AGP), is synthesized in the peroxisomes by FAR1, GNPAT, and AGPS and then shuttled to the ER, where further acylation, head group addition and other modifications can be introduced. Genetic disruption of peroxisome biogenesis or peroxisomal AGP synthesis suppresses ferroptosis in renal cell carcinoma models, hepatocellular carcinoma models and endometrial carcinoma models, concomitant with a reduction in the levels of PUFA-containing ether lipid species²⁵. However, there are also cell-specific differences. For example, deletion of AGPS in fibrosarcoma cells eliminates PUFA-containing ether lipids without altering ferroptosis sensitivity²⁷. Moreover, in Caenorhabditis elegans, disruption of the worm AGPS orthologue (ads-1) actually enhances ferroptosis sensitivity in germ cells¹⁰¹. Here, the protective effect of ADS-1-dependent ether lipid synthesis may be explained by the fact that, in C. elegans (versus mammalian cells), ether lipids are more likely to be acylated with anti-ferroptotic MUFA species than with pro-ferroptotic PUFA species¹⁴¹. These findings highlight how the same metabolic pathway can have different effects on ferroptosis depending on the nature of the final products that are synthesized.

Golgi apparatus

The Golgi apparatus is a complex stack of membranous compartments with a role in protein and vesicle trafficking and lipid metabolism. Disrupting Golgi structure pharmacologically can enhance ferroptosis sensitivity in HeLa and other cancer cells, possibly by increasing intracellular oxidative stress¹⁴². Mechanistically, this may involve the disruption of endogenous RTA metabolism. UBIA prenyltransferase domain-containing protein 1 (UBIAD1) undergoes regulated cycling between the ER and Golgi apparatus¹⁴³. UBIAD1 utilizes geranylgeranyl pyrophosphate in the synthesis of menaquinone 4 from menadione¹⁴⁴. Menaquinone 4 (a vitamin K-family metabolite) acts as an endogenous lipophilic RTA that can be recycled by FSP1 and vitamin K epoxide reductase complex subunit 1 (VKORC1; an ER-localized enzyme¹⁴⁵) to suppress ferroptosis^43,146,147. UBIAD1 is also proposed to directly mediate CoQ synthesis in the Golgi to prevent lipid peroxidation¹⁴⁸, which would be important as this could potentially provide a source of non-mitochondrial CoQ in the secretory pathway and plasma membrane that is accessible by FSP1. The relative contributions of UBIAD1-synthesized CoQ¹⁴⁸ and STARD7 transport of mitochondrially synthesized CoQ⁸⁶ to the pool of plasma membrane CoQ remain unknown. Of note, UBIAD1 binds and inhibits the sterol-accelerated degradation of 3-hydroxy-3-methylglutaryl (HMG)-CoA reductase in the ER¹⁴⁹. Binding of geranylgeranyl pyrophosphate releases UBIAD1 from HMG-CoA reductase and promotes UBIAD1 trafficking to the Golgi¹⁴⁹. Thus, UBIAD1-dependent regulation of the mevalonate pathway may indirectly influence de novo synthesis of CoQ in the mitochondria by controlling the synthesis of CoQ precursors (for example, farnesyl pyrophosphate).

Lysosomes

Lipid peroxidation may occur early on in this organelle under ferroptosis-inducing conditions¹⁵⁰. Moreover, lysosomal de-acidification generally inhibits ferroptosis¹⁵⁰. These effects are likely explained by the role of lysosomes in iron metabolism. Iron can be imported into the cell via the transferrin–transferrin receptor (TFRC) system. Here, Fe³⁺–transferrin complexes are internalized by TFRC and eventually trafficked to and liberated within the acidic environment of the lysosome. TFRC silencing inhibits ferroptosis, presumably by limiting the uptake of iron into the cell^52,53. Iron storage proteins are also degraded in the lysosome. The adaptor protein nuclear receptor coactivator 4 (NCOA4) links iron-loaded ferritin to the lysosomal catabolic machinery through a selective ferritin autophagy pathway known as ferritinophagy; NCOA4 disruption reduces ferroptosis sensitivity by preventing iron retrieval from ferritin and thereby limiting the amount of loosely coordinated (labile) intracellular iron available to participate in redox reactions^151,152. Some evidence indicates that NCOA4-mediated ferritin catabolism may be more important for ferroptosis in response to system x_c⁻ inhibition than direct GPX4 inhibition¹⁵³, an example of context-dependent ferroptosis regulation. More broadly, lysosomal catabolic processes are regulated in a multilayered manner by the transcription factor nuclear factor erythroid 2-related factor 2 (NRF2; also known as NFE2L2); one way that loss of NRF2 can enhance ferroptosis sensitivity is by altering lysosome function to increase the pool of labile intracellular iron that can promote lipid peroxidation¹⁵⁴.

In addition to regulating iron metabolism, the catabolic function of lysosomes can be engaged either to promote or to prevent ferroptosis. Chaperone-mediated autophagy (CMA) is a degradative process that terminates in lysosomal protein degradation. CMA requires heat shock 70 kDa protein 8 (HSPA8) and the lysosome-associated membrane glycoprotein 2 (LAMP2). System x_c⁻ inhibitors can increase LAMP2 expression, the association of HSPA8 and LAMP2 with GPX4, and the degradation of GPX4 via CMA, thereby enhancing ferroptosis¹⁵⁵. GPX4 degradation via CMA is reduced by creatine kinase B (CKB), which, acting as a protein kinase, phosphorylates GPX4 at Ser¹⁰⁴ to limit GPX4 interaction with HSPA8 and LAMP2 (ref. 156). While CMA of GPX4 tends to promote ferroptosis, the catabolism of other proteins can inhibit ferroptosis. Namely, lysosomal cathepsin D-mediated catabolism of cysteine-rich extracellular proteins such as albumin, followed by cystinosin-mediated export of cystine from the lysosome to the cytosol, can enhance GSH synthesis and inhibit ferroptosis when extracellular free cystine is limiting⁵⁹. Albumin is abundant outside the cell, potentially providing an important alternative cysteine supply. Within the cell, one response to cysteine deprivation is activation of the integrated stress response, including increased expression of the transcription factor ATF4 (ref. 157). Depletion of lysosomal cystine specifically increases ATF4 expression and reduces ferroptosis sensitivity¹⁵⁸.

The surface of the lysosome is an important site where the nutrient-sensing mTORC1 can be activated. In some cells, active mTORC1 promotes ferroptosis by favouring mRNA translation and the consumption of limited intracellular cysteine supplies in the synthesis of proteins rather than GSH⁹⁰. In other cells, active mTORC1 inhibits ferroptosis by stimulating the translation of GPX4 or the activation of SREBP1, which, as noted above, promotes anti-ferroptotic MUFA synthesis^125,159. Which of these pro-ferroptotic and anti-ferroptotic effects dominates at a given time seems likely to vary by cell line and context.

The nucleus

Morphological alterations of the nucleus (for example, karyorrhexis) are defining features of apoptosis¹⁶⁰. By contrast, during ferroptosis, the nuclei appear to remain largely intact and DNA damage is neither a defining feature nor a necessary mechanistic event in this process¹⁶¹. The nuclear envelope of mammalian cells is continuous with the ER membrane and includes PUFA-containing phospholipids that may be susceptible to lipid peroxidation¹⁶². A functional role for nuclear lipid peroxidation in ferroptosis has not been reported. However, increased levels of the DNA-damage marker 8-oxo-2′-deoxyguanosine (8-oxo-dG) can correlate with the induction of ferroptosis in vivo¹⁶³. While DNA damage may not contribute to the execution of ferroptosis, one possibility is that peroxidation of nuclear membrane phospholipids leads to the formation of lipid fragments that locally damage DNA during ferroptosis.

Within the nucleus, transcription is a major activity and contributes importantly to the regulation of ferroptosis sensitivity. As one might expect from the context-dependent nature of ferroptosis regulation, diverse transcriptional processes occurring in the nucleus help set the ferroptosis threshold of a given cell. A non-exhaustive list of transcription factors and epigenetic regulators that can regulate ferroptosis sensitivity include p53 (refs. 6,147,164), NRF2 (refs. 154,165–168), NFE2L1 (refs. 130,131), ZEB1 (ref. 169), STAT1 (ref. 170), ATF3 (ref. 171), YAP1 and TAZ^50,172, HIF2α¹³⁸, PPARα¹⁷³, NUPR1 (ref. 174), MYCN^65,175,176, and BAP1 (refs. 7,177). These proteins regulate the expression of genes (for example, SLC7A11, ACSL4, HILPDA, FSP1, NCOA4, LCN2, VKORC1L1 and LRP8) that, in turn, govern ferroptosis sensitivity in different ways, typically by regulating the levels of proteins or metabolites that impinge on membrane lipid peroxidation. Again, context-specific regulation appears to be the rule, with some transcription factors having an important role in ferroptosis regulation in one cell type but not in another¹⁷⁸.

Lipid metabolism

Fatty acid synthesis, trafficking and flux help dictate ferroptosis sensitivity by controlling the abundance and localization of more oxidizable versus less oxidizable lipids. These processes involve interactions between plasma membrane transport processes and enzymes and signalling pathways that are distributed throughout the cell. For example, cytosolic acetyl-CoA carboxylase 1 (ACC1) synthesizes malonyl-CoA, which can be used both in the synthesis of long-chain saturated fatty acids by fatty acid synthase (FASN) and in the elongation of imported PUFAs⁵⁵ into longer-chain-length PUFA species. The desaturation of long-chain saturated fatty acids by SCD1, FADS1 and FADS2 occurs in the ER, which is also the site where fatty acids are activated by ACSL enzymes and incorporated into phospholipids or TAGs by a host of acyltransferases. ACC1, FASN and SCD1 can all regulate ferroptosis sensitivity but seem to do so in different ways in different cells and situations^{125,179–181}. Furthermore, signalling networks impact the function of these enzymes. In one example, under conditions of glucose limitation, cytosolic AMPK is activated and can phosphorylate and inhibit ACC1 and ACC2, which ultimately lowers ferroptosis sensitivity by reducing the abundance of oxidizable PUFA-containing membrane phospholipids (for example, PE 18:0/22:4)^30,182, perhaps by limiting the malonyl-CoA-dependent elongation of shorter-chain PUFAs. In another example, oncogenic KRAS activation can increase the expression of FASN, which in turn leads to more incorporation of saturated fatty acids and MUFAs into membrane phosphatidylcholines in place of PUFAs, reducing ferroptosis sensitivity¹⁰⁷. Likewise, in FLT3-mutant acute myeloid leukaemia, the expression of FASN and SCD1 is increased, resulting in reduced ferroptosis sensitivity¹⁷⁶. Outside the context of cancer, how different signalling networks may dynamically regulate lipid metabolism to influence lipid peroxidation and ferroptosis sensitivity is mostly unclear.

How lipids are channelled into different classes of membrane phospholipids, to either suppress or enhance ferroptosis, is also not well understood. For example, phosphatidylethanolamine is a key phospholipid that is often oxidized during ferroptosis²¹. The synthesis of phosphatidylethanolamine involves the transport of phosphatidylserine from the ER to mitochondria at membrane contact sites known as mitochondria-associated membranes¹⁸³ (Fig. 3b). Within the mitochondria, phosphatidylserine is enzymatically converted into phosphatidylethanolamine and then transported back to the ER, at which point it can traffic to other compartments in the cell (for example, plasma membrane) via the secretory pathway or lipid transfer proteins¹⁸³ (Fig. 3b). How these trafficking events may contribute to ferroptosis is not clear. In particular, while oxidized phosphatidylethanolamine has been implicated in ferroptosis, its trafficking within the cell has not been directly studied in the context of ferroptosis. Likewise, the flux of PUFAs into TAGs, concomitant with the biogenesis of lipid droplets from the ER, can be protective against ferroptosis in some cases by restricting the incorporation of PUFAs into phospholipids. However, how decisions to channel fatty acids into different lipid species and to traffic phospholipids from one compartment to another are made is not well understood. Additionally, it is possible that cellular decisions about lipid fluxes are made for reasons completely independent of cell death regulation (for example, connected to membrane remodelling events needed for cell division)^117,184 and that increased sensitivity or resistance to ferroptosis is a by-product of these changes.

Ferroptosis amplification mechanisms

Rising plasma membrane lipid peroxidation can trigger feedforward mechanisms that accelerate this process (Fig. 1b). First, lipid peroxidation leads to water entry and cell swelling, increasing plasma membrane tension, which then further activates mechanosensitive Piezo1 and TRP ion channels, accelerating this loop⁹. Second, lipid peroxidation can stimulate the activity of protein kinase Cβ (PKCβII), which in turn phosphorylates ACSL4 at Thr³²⁸, increasing the activity of this lipid metabolic enzyme and enhancing the synthesis of oxidizable PUFA-containing membrane phospholipids¹⁸⁵. Where exactly in the cell this pool of ACSL4 resides is not entirely clear; ACSL4 is thought to localize to the ER but a pool of this enzyme may also be found at the plasma membrane and is therefore possibly able to modulate plasma membrane lipid composition directly. Third, plasma membrane lipid peroxidation can cause the protein CAVIN1 to be released from plasma membrane caveolae, allowing it to migrate into the cytosol, where it can bind to and promote the degradation of NRF2 (ref. 186), thereby limiting the protective response mediated by this transcription factor. Fourth, increasing lipid peroxidation can lead to the hyperoxidation of mitochondrial peroxiredoxin 3 (PRDX3), which then translocates to the plasma membrane and inhibits the function of the system x_c⁻ cystine–glutamate antiporter, which will tend to increase ferroptosis sensitivity¹⁸⁷. These amplification mechanisms may be engaged to overcome the powerful lipid peroxidation defence and membrane repair mechanisms that operate at the plasma membrane, ensuring that the execution of ferroptosis proceeds to completion once this process has been initiated¹⁸⁸.

Feedback suppression of ferroptosis sensitivity

While some mechanisms may serve to accelerate ferroptosis, others operating in parallel will tend to restrain this process. These mechanisms can also be distributed spatially within the cell. For example, in breast cancer cells exposed to GPX4 inhibitors, the pentaspanin protein prominin 2 (PROM2) can be upregulated and stimulate the release of extracellular vesicles containing iron-loaded ferritin from the cell, thereby attenuating lipid peroxidation¹⁸⁹. In another example, in tumour-infiltrating neutrophils, signals emanating from cell surface receptors eventually lead to upregulation of aconitate decarboxylase 1 (ACOD1) expression in the mitochondria, which synthesizes the metabolite itaconate. Itaconate can then diffuse out of the mitochondria, bind the NRF2 negative regulator KEAP1 in the cytosol and stabilize NRF2 to suppress ferroptosis¹⁹⁰.

Intracellular trafficking

Lipid peroxidation occurring within the cell in response to GPX4 inhibition can have a profound effect on the function of the secretory pathway. As noted above, in some cells, GPX4 inhibition stimulates the secretion of extracellular vesicles¹⁸⁹. Additionally, in response to the same stimulus, TFRC is depleted from the Golgi apparatus and the endosomal recycling compartment and accumulates at the plasma membrane¹⁹¹. Why TFRC gets ‘stuck’ at the plasma membrane following GPX4 inhibition is unclear, especially since a different cell surface protein, EGFR, traffics normally in response to GPX4 inhibitor treatment¹⁹¹. This TFRC protein redistribution within the cell is a potential ferroptosis molecular marker. Functionally, reduced trafficking of a key iron metabolism-related protein such as TFRC may also tend to reduce ferroptosis sensitivity by limiting iron uptake. The trafficking of other proteins within the secretory pathway, such as the innate immune regulator stimulator of interferon genes (STING)¹⁹², is also inhibited following GPX4 inactivation and increased levels of lipid peroxidation. It is unclear whether the inhibition of STING trafficking contributes to ferroptosis execution. However, this example suggests that a pro-ferroptotic stimulus such as GPX4 inactivation can have diverse effects on protein trafficking that are likely to alter cellular responses to other stimuli. Understanding how intracellular protein and lipid transport networks may change over time in response to ferroptotic stress to either reduce or increase ferroptosis sensitivity is an interesting frontier area.

Transcription programmes

Different transcription factors can induce gene expression programmes that ultimately regulate ferroptosis sensitivity by modulating the function of multiple pathways and organelles in parallel. NRF2 will generally promote ferroptosis resistance by positively regulating the expression of several factors: GSH biosynthetic enzymes that localize to the cytosol¹³⁰, FSP1 (refs. 165,166), which is required at the plasma membrane, and proteins involved in iron homeostasis found at the lysosome¹⁵⁴. However, these mechanisms are not entirely universal. High NRF2 expression can, in some cancer cells, tend to promote ferroptosis sensitivity by upregulating the expression of a plasma membrane-localized transporter (MRP1; encoded by ABCC1) that exports GSH from the cell⁵⁷. The tumour suppressor protein p53 also seems to be capable of either promoting or inhibiting ferroptosis through various molecular mechanisms, depending on the cell type and even on the nature of the ferroptosis-inducing stimulus^{6,117,164,193}. Likewise, it seems possible that similar competing mechanisms could exist for other transcriptional regulators of ferroptosis. For example, SREBP1 can positively regulate the expression of ER-resident SCD1 (refs. 125,180) but also of cytosolic enzymes involved in malonyl-CoA and NADPH synthesis¹⁹⁴. The coordinate expression of these genes may help enable anti-ferroptotic MUFA synthesis but could conceivably also be involved in enhancing ferroptosis in other ways as described above (for example, NADPH-dependent ROS synthesis that initiates ferroptosis). While it is uncomfortable to not have universal models of ferroptosis regulation by individual transcription factors that apply to all cell types and contexts, it does help illustrate the importance and value of detailed investigations of ferroptosis regulation across systems, with careful attention to how ferroptosis is being induced.