Molecular mechanism of ferroptosis in brief

Ferroptosis is a recently discovered form of regulated (necrotic) cell death. Since its discovery in 2012, it has rapidly become recognized as a potential anticancer strategy.1 Ferroptosis is iron-dependent and can be induced by blocking the system X_C⁻ cystine/glutamate antiporter or glutathione peroxidase 4 (GPX4),2 resulting in a defective glutathione (GSH) redox system, where oxidized phosphatidylethanolamines (oxPEs) are crucial for ferroptosis execution (figure 1, treatment examples in table 1).1 3 4 More specifically, arachidonoyl and adrenoyl groups are oxidized in the endoplasmatic reticulum (ER)-associated compartments.5 Lipoxygenases (LOXs) oxidize those phosphatidylethanolamine (PE) species to cause cell death, and this can be blocked by tocopherols such as vitamin E (α-tocopherol). oxPEs cause an increase in membrane curvature that eventually leads to membrane destabilization, pore and micelle formation, and membrane rupture.6 Normally, glutamate is exchanged by cystine, which is then converted into cysteine, which is needed to produce GSH, an essential cofactor for GPX4. By blocking the X_C⁻ antiporter, not enough cystine is transported into the cell, leading to a decrease in the available GSH.7 When the pool of GSH is depleted, GPX4 can no longer function and lipid peroxidation can no longer be inhibited, leading to permeabilization of the cell membrane (figure 1). Alternatively, ferroptosis can be induced by directly blocking GPX4, which might result in a faster induction of ferroptotic cell death, as the GSH pool does not need to be depleted before the membrane is permeabilized.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Ferroptosis pathway in brief. Ferroptosis is an iron-dependent cell death modality. (1) Cystine is imported into the cell in exchange for glutamate by the system X_C⁻ antiporter. Cystine is converted into cysteine and used for GSH production. (2) GSH is an essential cofactor for GPX4. By blocking the system X_C‾ antiporter (eg, with erastin), GSH is depleted and GPX4 can no longer function. (3) GPX4 inhibits lipid peroxidation; directly inhibiting it (eg, by RSL3) induces ferroptosis. (4) Fe³⁺ bound to transferrin is transported into endosomes by binding to the transferrin receptor. (5) Fe³⁺ is converted to Fe²⁺ by STEAP3. Fe²⁺ is then transported to the labile iron pool by DMT1. (6) Fe²⁺ can be stored by binding to ferritin. (7) Fe²⁺ can be transported outside the cell by ferroportin. (8) Fe²⁺ is oxidized back to Fe³⁺ via the Fenton reaction,8 9 leading to production of hydroxyl radicals. (9) ACSl4 selectively enriches PUFAs in the plasma membrane. (10) Radicals formed from the Fenton reaction react with PUFAs, leading to oxPE formation and cell death. This can be inhibited by Fer-1.10 ACSl4, acyl–CoA synthetase long-chain family member 4; CoA, coenzyme A; DMT1, divalent metal transporter 1; Fer-1, ferrostatin-1; GSH, glutathione; GPX4, glutathione peroxidase 4; oxPE, oxidized phosphatidylethanolamine; PUFA, polyunsaturated fatty acid; RSL3, Ras-selective lethal 3.

Notably, iron has a considerable effect on lipid peroxidation in ferroptosis through the production of reactive oxygen species (ROS) by Fenton chemistry.8 9 Under physiological conditions, iron exists as ferric ions (Fe³⁺), which can bind to transferrin and enter cells via the transferrin receptor (figure 1). Once in endosomes, ferric ions are reduced to ferrous ions (Fe²⁺) in a reaction catalyzed by the ferric reductase STEAP3. Fe²⁺ is then transported through the endosomal membrane, with help from the divalent metal transporter 1, into the cytosol, creating a labile iron pool (figure 1). Excess iron is either transported outside the cell by ferroportin, an iron exporter, or stored in ferritin protein, which oxidizes ferrous ions back to ferric ions. Fe²⁺ present in the labile iron pool can interact with H₂O₂, resulting in oxidation of Fe²⁺ back to Fe³⁺ and formation of highly reactive hydroxyl radicals (OH·). OH· then reacts with polyunsaturated fatty acids (PUFAs) of phospholipids (PLs) in the cell membrane, converting PL into carbon-centered PL radicals (PL·). With the involvement of molecular oxygen (O₂) and adjacent PUFA, the primary lipid radical (L·) is converted into an alkoxyl radical (LO·), which can interact with Fe²⁺ and introduce a next round of lipid radical formation. Extensive production of lipid ROS promotes damage and/or destruction of the cell membrane, leading to cell death (figure 1). Several cell death inhibitors specific for ferroptosis have been developed. Ferrostatin-1 (Fer-1) can effectively block lipid peroxidation via a dual mechanism: on one hand, Fer-1 functions as a lipophilic radical scavenger reducing ROS formation, and on the other hand, it also blocks the 15-lipoxygenase (15LOX)/phosphatidylethanolamine-binding protein-1 (PEBP1) complex.10 This 15LOX/PEBP1 complex is responsible, together with the Fe-catalyzed radical reaction, for oxPEs formation, meaning that Fer-1 can block both pathways for oxPE formation.11 Acyl–coenzyme A (CoA) synthetase long-chain family member 4 (ACSL4) confers sensitivity to ferroptosis by selectively enriching membranes with PUFA, which are needed for ferroptosis execution.12 Also, it was recently discovered that peroxisomes can contribute to ferroptosis by synthesizing ether-linked PLs (so called plasmalogen).13 14 The plasmalogen from the peroxisome can be peroxidized as well, leading to ferroptosis. However, cells can downregulate the formation of plasmalogen, especially during cell state transitions, leading to ferroptosis resistance; this has been observed in carcinoma cells.15 These different pathways make ferroptosis a potential target for killing cancer cells and could be used in anticancer treatment.

Morphological changes during ferroptosis

Morphologically, ferroptosis is different from other cell death modalities, and especially from apoptosis. Ferroptosis is characterized mainly by mitochondrial aberrations. Shrunken mitochondria, reduced cristae, and rupture of the outer mitochondrial membrane are the most common manifestations.16 It should also be noted that the density of the mitochondrial membrane increases,1 and neither chromatin condensation nor blebbing occur. On the other hand, the main morphological changes in apoptosis are chromatin condensation and blebbing, together with cell shrinkage and nuclear fragmentation.17–19 Ferroptosis is also morphologically different from necroptosis, which is characterized by swelling of the cells and intracellular organelles, and membrane rupture.18 20 21

It was recently shown in human trophoblast cells (primary PHT cells and the BeWo human trophoblast cell line) that macroblebbing of the membrane occurred during ferroptosis.22 This macro-blebbing was caused by the increased membrane flexibility due to the presence of hydroxyperoxidated PE, as shown by analyzing the elasticity of a simulated double lipid layer membrane. This blebbing was also confirmed by Van der Meeren et al by using atomic force microscopy.18 In that study, membrane protrusions were observed after induction of ferroptosis in murine fibrosarcoma L929 cells and MCA205 cells, but this was accompanied by a decrease in overall elasticity of the cells. The more flexible oxPE might be responsible for bleb formation, but the elasticity of the cells still decreases during ferroptosis, presumably because the blebs containing oxPEs are released, causing overall decrease of elasticity. Of course, ferroptosis is also associated with cell rounding and plasma membrane permeabilization at the end of the cell death process.23 However, in necrosis induced by bacterial pore-forming toxin, membrane permeabilization is an active process mediated by ninjurin-1. In ferroptosis, pores are formed by an unknown mechanism (thought to be lipid peroxidation-dependent) leading to osmotic rupture of the membrane.24–26 Interestingly, ferroptosis could propagate to neighboring cells independently of cell rupture by Ca²⁺ fluxes through the ferroptotic pores.26 27 All these features make ferroptosis morphologically distinct from both apoptosis and necroptosis. This distinction might be used to morphologically distinguish ferroptosis from other regulated cell death modalities by using different microscopic methods, including those based on artificial intelligence.28

Damage-associated molecular patterns

Danger-associated molecular patterns (DAMPs) are endogenous intracellular molecules that in normal conditions contribute to different physiological processes. They are released (or exposed on the outer surface of the plasma membrane) when a cell is damaged or dying.29 Once released, DAMPs acquire immunostimulatory properties and bind to pattern-recognition receptors, such as Toll-like receptors (TLRs, eg, TLR4) and purinergic receptors (P₂X₇) to promote inflammation and activation of the innate immune system. DAMPs originate from the cytosol and mitochondria, and are extracellular matrix (ECM)-derived. In contrast to pathogen-associated molecular patterns, DAMPs are not associated with pathogens.30 The most well-known DAMPs, which play an important role as adjuvants in ICD, are ATP, high-mobility group 1 (HMGB1) and calreticulin (CRT) (figure 2). These molecules bind to P₂X₇ receptors, TLR4 or low-density lipoprotein receptor-related protein 1, respectively. By activating the immune system, DAMPs can enhance antitumor immunity during and after cancer treatment.31 32

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Hallmarks of ICD. The immunogenicity of cancer cell death is dependent on two main elements: adjuvanticity (green) and antigenicity (blue). HMGB1 binds to TLR4 on DCs, causing their activation and maturation.39 CRT is expressed on the surface of dying (apoptotic) cells and will bind to LRP1 on DCs.40 ATP is either actively secreted or passively released by dying cancer cells in a specific spatiotemporal pattern and will bind to P₂X₇ receptors on DCs, causing their activation and maturation and leading to inflammasome activation.41 Proinflammatory cytokines act as adjuvants and can be seen as iDAMPs activating the immune system cells.50 Anti-inflammatory cytokines cause inhibition of immune cells and are considered protumorigenic. Antigens are presented by DCs on MHC-I molecules to T cells in a process called cross-presentation.58 After antigen recognition, cytotoxic CD8⁺ T cells are formed and migrate to the tumor, where they eliminate antigen-expressing cancer cells.57 Memory T cells may be formed, establishing long-lasting antitumor immune responses and reducing the chance of tumor reoccurrence. CRT, calreticulin; DAMP, danger-associated molecular pattern; DC, dendritic cell; HMGB1, high-mobility group 1; ICD, immunogenic cell death; iDAMP, inducible danger-associated molecular pattern; LRP1, lipoprotein receptor-related protein 1; TCR, T-cell receptor.

DAMPs are an important component in ICD by providing adjuvanticity to dying cells.33 It has been established that DAMPs are emitted at a precise moment and that the release of all the DAMPs in a specific sequence is required for effective activation of the immune system. Apoptosis was previously thought to be tolerogenic,34 but in certain cases, and more specifically in apoptotic cancer cells, it could activate the immune system.35–38 This immunogenic apoptosis was associated with the release of HMGB1 and exposure of CRT, thus causing immune cell activation.39–41 HMGB1 and ATP release have also been observed in necroptosis.42 43 This indicates that both apoptosis and necroptosis show hallmarks of ICD and could be exploited in cancer therapy. However, cancer cells are known to be resistant to those cell death types, which diminishes their therapeutic potential.44 Therefore, there is a need for triggering other cell death modalities.

Cytokines released during ICD

DAMPs are not the only factors released during cell death that have adjuvant properties. Cytokines are a diverse group of molecules that can be either proinflammatory or anti-inflammatory.45 Proinflammatory cytokines can act as adjuvants too during ICD, whereas anti-inflammatory cytokines dampen the immune response directed at the dying cells (figure 2). Interestingly, cytokines are more effective than DAMPs as they function at lower concentrations in the range of picogram per millilitre.46 For example, in one study, the DAMP heat shock protein 70 (HSP70) was found to be expressed at the nanogram scale, whereas the cytokines (mainly interleukins) were in the picogram scale.46 However, a clear distinction between proinflammatory and anti-inflammatory cytokines cannot be made, as many of them share common signaling pathways and the eventual outcome depends on the cell type, the presence of other cytokines, and the tumor microenvironment. In this regard, Abiko et al showed that in tumors resected from patients with ovarian cancer, the tumor-promoting factor prodeath ligand 1 was expressed more strongly on tumor cells due to increased secretion of interferon gamma (IFN-γ) by CD8⁺ T cells.47 However, IFN-γ can also have a tumor-suppressive effect due to its upregulation of antigen processing and presentation.48 This might be explained by the more complex microenvironment in the tumors, where different cell types interact with each other and the eventual effect on the immune system depends on the overall balance between proinflammatory and anti-inflammatory factors.49

It should be noted that cytokine secretion is induced during cell death (eg, necroptosis), and so these cytokines can be seen as inducible danger-associated molecular patterns (iDAMPs, as proposed by Yatim et al50). Together with the DAMPs already present inside cells (such as ATP and HMGB1), iDAMPs are released during cell death, after which they interact with immune cells to induce activation of the immune system. Thus, both DAMPs and iDAMPs are required for efficient immune activation. While DAMPs are generally viewed as one of the main determinants of immunogenicity, the presence of iDAMPs suggests that immunogenicity not only depends on the release of DAMPs but also requires stress-induced transcription of proinflammatory cytokines.51 Moreover, cytokines (eg, type I interferons) might act in an autocrine fashion on cancer cells to increase their immunogenicity.52 Therefore, analysis of the cytokine secretion pattern is important in assessing the immunogenicity of cell death.

During and even after cell death, cytokines can have a significant effect on its immunogenicity. It has been observed that even when the cell membrane has been permeabilized in necroptotic cells, the corpses still release cytokines, such as interleukin (IL)-6, CXCL1, CXCL2 and CCL2due to RIPK3 activation.53 This continued cytokine secretion upregulates phagocytosis by bone marrow-derived macrophages and enhances the immune response by promoting cross-presentation (presentation of antigen in the context of major histocompatibility complex (MHC)-I molecules). Thus, the cytokines act as adjuvants during ICD. However, previous studies have shown that when macrophages take up necroptotic L929 fibrosarcoma cells induced by tumor necrosis factor-alpha (TNF-α), no proinflammatory cytokines are produced and an effective immune response is not generated.19 54 This might have been due to the use of different inducers. Orozco et al53 used a cell line with a constitutively active RIPK3 to trigger necroptosis, whereas Brouckaert et al induced necroptosis with TNF-α.54 The two methods might induce different expression patterns of cytokines and thereby lead to different immune responses. The interaction of different stress-induced signaling pathways will play an important role in the eventual immunogenicity of dying cells.

Different cell death modalities eventually lead to different immunogenicity outcomes. For efficient ICD, all the previously mentioned parameters, such as DAMPs and iDAMPs (responsible for adjuvanticity) and antigens (for antigenicity, see next section), need to be present in a specific spatiotemporal pattern. For example, Zhang et al showed in CT26 cells that ATP and HMBG1 release from apoptotic cells peaked at 6 hours and from necroptotic cells at 8 hours.55 Also, CRT expression at the membrane was observed after an hour of necroptosis induction. In strong contrast, in apoptotic MCF7 cells, ATP and HMGB1 release peaked only after 12 and 48 hours, respectively,56 and membrane CRT expression was observed only after 10 hours. Despite these differences, both cell death modalities have been shown to be immunogenic in nature. This means that different cell death modalities may differ in their modulation of DAMP exposure, leading to activation of the immune system. This highlights the importance of the cell death type in ICD. The hallmarks of ICD should always be studied in terms of time kinetics and pattern of DAMPs, iDAMPs and antigen expression in order to properly correlate these variables with the responses to ICD.

Modulation of antigenicity during ICD

Another major determinant of immunogenic cancer cell death is antigenicity.33 It is known that efficient induction of antitumor immunity requires exposure of the relevant antigens. Antigens can be derived from intracellular proteins that become degraded to peptides by proteasomes.57 The peptides are transported to the endoplasmic reticulum, where they bind to newly produced MHC-I molecules. The MHC-I molecules, with the peptides bound in their groove, are transported to the plasma membrane, where they can be recognized by T cells. Normally, antigens bound to MHC-I are self-antigens that do not activate T cells because they have been selected for self-recognition during their development. Therefore, normal cells do not elicit an immune response, except in autoimmune diseases. However, since tumors may contain a large number of mutations, the mutated antigens presented by tumor cells are recognized as foreign.

The main cells capable of antigen presentation are dendritic cells (DCs),58 which can take up tumor antigens from dying cancer cells, and in the presence of DAMPs and iDAMPs become mature and activated. CD4⁺ helper T cells are required for efficient activation via the costimulatory molecules CD40/CD40L (figure 2). Activated DCs migrate to the lymph nodes and prime naive cytotoxic CD8⁺ T cells. Antigen-specific cytotoxic T cells might be formed, which can migrate back to the tumor, recognize the tumor cells presenting mutated antigens on MHC-I molecules, and kill them, thus leading to eradication of the tumor. Cross-presentation is the corner stone of cancer immunotherapy because it can lead to long-lasting protection against tumors. Of course, tumor cells have developed mechanisms of resistance to cross presentation, such as downregulation of MHC-I expression.57

Antigens can be tumor-specific antigens (TSAs), tumor-associated antigens (TAAs) or neoepitopes.59 60 TSAs are specific only for tumors. However, many cancers have developed resistance mechanisms to eliminate TSA expression in order to remain hidden from immune surveillance, for example, by downregulating the expression of TSAs and thereby eliminating antigen presentation.60 TAAs can also be used for antigen presentation, but this type of antigen is not specific for tumor cells. They are mainly proteins that are normally expressed only during embryogenesis or in immune privileged areas such as the brain or gonads.59 61 Therefore, TAAs are less suitable for ICD-based therapies as the immune cells do not have sufficient specificity for the cancer. On the other hand, neoepitopes arise from mutations in cancer cells and are therefore specific for the tumor.61 This makes neoepitopes suitable for use in therapy because a specific immune response can be mounted against the cancer. However, the process of identifying neoepitopes remains very technically challenging. Also, in the tumor, there is a natural selection process of cancer cells that express limited antigens, and this reduces the efficacy of activating the immune system.62

The cell death type may also modulate antigenicity during ICD, leading to different therapeutic outcomes. Many cell lines (eg, colon carcinoma CT26 cells) used in the field of ICD already express non-endogenous antigens (eg, virally derived peptides), and thus, they will likely be recognized by the immune system as TSAs.63 In this context, it has been shown that both apoptotic and necroptotic CT26 colorectal cancer cells, which express the virally derived peptide AH1 (an immunodominant Ag), can activate the immune system of mice and protect against the tumor.64 However, when a knock-out model for AH1 was used, only necroptotic CT26 cells were still immunogenic, whereas the apoptotic CT26 cells could not activate the immune system. The authors showed that only necroptotic CT26 cells could stimulate the formation of T cells specific for CT26 neoantigens, which means that necroptosis is a form of ICD with very strong immunogenic potential. While apoptosis seemed to be immunogenic, this was only caused because of the AH1 antigen, because when this antigen was removed, antigenicity was insufficient for efficient immune activation. Importantly, the type of cell death can also affect the antigenicity of dying cancer cells and thereby modulate their immunogenicity. The models should be selected critically because non-endogenous immunodominant antigens might mask the intrinsic antigenicity of different cell death modalities.

In conclusion, ICD is a complex phenomenon that requires at least three factors to be present in a specific spatiotemporal pattern (figure 2). To study ICD, both adjuvanticity and antigenicity need to be investigated as both are at least equally important features necessary for efficient activation of the immune system against the tumor (figure 2). Either of them alone will not lead to ICD but to an immunologically silent form of cell death. As ICD leads to activation of immune cells, and possibly to establishment of immunological memory, this might be an effective alternative for current immunotherapies or as part of a combined treatment modality. Current anticancer therapies mainly trigger apoptosis, but since resistance against apoptosis is a frequent occurrence in cancer,44 combined with the fact that antigenicity might be hindered during apoptosis,64 this might not be the ideal choice. Therefore, it is important to investigate whether other cell death modalities, such as necroptosis and ferroptosis, are immunogenic and to exploit them in immunotherapy in more clinically relevant models.

Ferroptosis and innate immunity

Ferroptosis is associated with inflammation and innate immunity. Li and colleagues68 demonstrated that ferroptotic cell death occurs in the early phase of reperfusion during intestinal ischemia–reperfusion injury (IRI).68 In that study, the authors showed that the expression of ACSL4, a key enzyme regulating lipid composition, was upregulated under ischemic conditions and that it contributed to reperfusion-induced ferroptotic injury. Though a direct link between inflammation and ferroptosis was not studied in that work,68 the study suggested that IRI-induced ferroptosis might lead to infiltration of immune cells in the injured intestine. In another study related to myocardial IRI,69 it was demonstrated that inhibition of cell death with Fer-1 in the heart model subjected to IRI leads to reduction of several oxidized lipids associated with ferroptosis. The authors discovered that ferroptotically dying cells trigger TLR4/Trif signaling in endothelial cells, which promotes recruitment of neutrophils to the injured heart. The relevance of ferroptosis to mediating innate immunity has also been demonstrated in acute renal tubular necrosis and IRI,70 where leukocyte infiltration in the model of cremaster muscle IRI was observed. Moreover, infiltration into the postischemic area after treatment with Fer-1 was less than in controls.70 The aforementioned studies demonstrate an emerging role of ferroptotic cell death in the innate inflammatory response during tissue injury and IRI. However, this needs further examination to identify the main players and executors of ferroptosis-induced inflammation during IRI.

DAMPs in ferroptotic cancer cell death

High-mobility group 1

Release of DAMPs during ferroptotic immunogenic cancer cell death is being actively studied and has been described for several DAMPs, such as HMGB165 66 71 72 and ATP65 (figure 2). HMGB1 release from ferroptotic cells was detected for the first time by Wen et al in human fibrosarcoma HT1080, human pancreatic carcinoma cells PANC-1 and immortalized mouse embryonic fibroblasts (MEF) following treatment with two different classes of ferroptotic inducers: class I (erastin and sorafenib) and class II (RSL3 and FIN56).66 Of note, the release of HMGB1 from ferroptotic cells was significantly decreased when ATG5 and ATG7, genes regulating the autophagy pathways, had been knocked out, suggesting that the autophagy pathway is required for the release of HMGB1 in ferroptosis. Importantly, extracellular HMGB1 released by dying cancer cells is a potent inducer of innate and adaptive immunity and is one of the main players in mediating the adjuvanticity of ICD (figure 2). HMGB1 binds to AGER/RAGE73 and TLR474 on the surface of antigen-presenting cells, driving the immune response against cancer cells. Initially, it was shown that HMGB1 evoked immunogenicity of apoptotic cancer cells via the MYD88/TLR4 signaling pathway.39 75 However, Wen and colleagues showed that in ferroptotic MEF cells, HMGB1-mediated inflammation requires signaling through AGER/RAGE but not TLR4.66 How dying cancer cells prioritize HMGB1-mediated inflammation and decide between AGER/RAGE or TLR4 signaling remains unknown and needs further investigation.

Recently, we described the release of HMGB1 from mouse ferroptotic fibrosarcoma MCA205 cells (figure 3) induced with RSL3 (figure 1).65 We distinguished between ‘early’ and ‘late’ ferroptotic cells based on Annexin V/Sytox Blue staining. Early ferroptotic cells (1 hour after ferroptosis induction) had 2 x higher single Annexin V⁺ population and 11 x lower Annexin V⁺Sytox Blue⁺ population than late ferroptotic cells (24 hours after stimulation with RSL3). We found that HMGB1 release gradually increased during the progression of cancer cells towards late ferroptosis and was maximal 24 hours after RSL3 stimulation, when all cancer cells were stained double positive for Annexin V and Sytox Blue. Moreover, to better mimic the situation in situ, the levels of HMGB1 were measured by co-culturing early and late ferroptotic cells with bone-marrow derived dendritic cells (BMDCs). We found that the level of HMGB1 was maximal when the BMDCs were co-cultured with early ferroptotic cancer cells, indicating the immunogenicity of early but not late ferroptotic cancer cells (table 1). Thus, these two studies65 66 describe the role of HMGB1 mainly as DAMPs that can contribute to the immunogenic potential of ferroptotic cancer cells (figure 3), similar to their role in the immunogenicity of other regulated cancer cell death modalities such as apoptosis39 and necroptosis.43 All these data confirm that HMGB1 is a key DAMP in different ICD modalities involved in the activation of innate76 and adaptive immunity.39

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

Immunogenic potential of ferroptosis. Several factors might influence the immunogenicity of ferroptosis. Early, but not late, ferroptotic cancer cells are immunogenic in vitro and in vivo, and HMGB1 and ATP are released by early ferroptotic cancer cells.65 it is unknown if CRT is exposed on early ferroptotic cells, and this requires investigation. Notably, HMGB1 and ATP are not released by late ferroptotic cells. HMGB1 might act through the AGER/RAGE or TLR2/TLR4/TLR9 pathway and activate innate and adaptive immunity. Moreover, knockdown of the HMGB1 gene decreases erastin-induced ferroptosis,72 therefore pointing to the regulatory role of HMGB1 in ferroptosis (*). Another component potentially modifying the immunogenicity of ferroptotic cell death might hide behind oxPLs and other oxidized products generated during ferroptosis. OxPLs (eg, oxPE) can play a dual role in ferroptosis, inducing either tumor promotion or tumor suppression. However, the role of oxidized oxPLS in the immunogenicity of ferroptosis is not known. CRT, calreticulin; HMGB1, high-mobility group 1; IL, interleukin; oxPE, oxidized phosphatidylethanolamine; oxPL, oxidized phospholipid; oxPS, oxidized phosphatidylserine.

Unexpectedly, another study discovered a role of HMGB1 not related to immunogenicity but to regulation of ferroptosis.72 Ye and colleagues generated knockdown of HMGB1 in the HL-60 human acute myeloid leukemia line cell expressing NRAS^Q61L and analyzed the capacity of the cells to undergo ferroptosis. Interestingly, when HMGB1-deficient HL-60 cells were treated with erastin (inhibitor of the cystine-glutamate antiporter system X_C‾, figure 1), both ROS generation and cell death via an iron-mediated lysosomal pathway decreased. These data suggest that in addition to its function as a DAMP,65 66 HMGB1 is also a positive regulator of ferroptotic cell death itself. One can speculate that human tumors deficient in HMGB1 (eg, human breast cancers)77 are not only deficient in induction of anti-tumor immunity because ferroptosis would not be immunogenic,65 but also that such tumors would be resistant to ferroptosis. Future studies will shed light on the dual role of HMGB1 in ferroptosis as a DAMP and as a regulator of ferroptosis (figure 3), and additional work is needed to validate the possible dual role of HMGB1 in ferroptosis in human cancers.

Are oxPEs species required for the immunogenicity of ferroptosis?

In addition to well-known DAMPs, such as CRT, HMGB1, ATP, and IFNs (figure 2), oxidized phospholipids (oxPLs) (figure 1) might also play an important modulatory role in the immunogenicity of ferroptotic cancer cells and in the subsequent immune responses.4 82 The execution of ferroptosis includes oxidation of different types of PLs in the plasma membrane.5 12 83 84 During ferroptosis, enzymatic oxidation occurs, primarily of PE, due to shortage of antioxidant protection (GPX4) and/or LOX-mediated PL peroxidation5 12 83 (figure 1). There is no knowledge available yet about a direct influence of oxPL derived from ferroptotic cells and their role in modulating the immune system. However, in several recent studies, it has been shown that oxPL derived from or associated with other regulated cancer cell death modalities might affect the function of APCs and modulate the immune responses. In this regard, it has been reported that the oxidized version of externalized phosphatidylserine (PS) positively affects the engulfment of apoptotic cells by macrophages.85 86 Tyurin and colleagues85 compared the efficiency of the engulfment of apoptotic wild-type HL-60 cells (low in oxidized phosphatidylserine (oxPS)) and linoleic acid (LA)-enriched HL-60 cells (high in oxPS) by RAW264.7-derived and THP-1-derived macrophages. In apoptotic LA-HL-60 cells, increased oxPS led to enhanced phagocytosis by macrophages compared with control HL-60 cells. Not only oxPS but also oxPE, which is generated during ferroptosis (figure 1), can interfere with the clearance of apoptotic cells by professional phagocytes.87 Uderhardt and colleagues87 demonstrated that milk fat globule-EGF factor 8 (MFG-E8)-dependent uptake of apoptotic cells by inflammatory macrophages was limited in the presence of PE oxidation products. The authors showed that PE oxidation products, mediated by resident macrophages expressing 12/15-lipoxygenase (12/15-LO), blocked phagocytosis of apoptotic cells. Interestingly, in the absence of 12/15-LO, clearance of apoptotic cells by inflammatory monocytes was restored, leading to T-cell responses against antigens derived from the apoptotic cells. Considering these data and knowing that PE is one of the major PLs undergoing oxidation during ferroptosis (figure 1), it is conceivable that the engulfment of ferroptotic cancer cells by APC, such as DCs and macrophages, could be modulated by oxPE and its oxidation products (discussed in the next section). Although several studies have already demonstrated that ferroptotic cancer cells are efficiently engulfed by APCs,65 88 the role of oxPEs was not addressed. Thus, the effect of oxPEs generated during ferroptosis on phagocytosis and the immunogenicity of ferroptotic cancer cells are unknown, and additional studies can bring further insights into the role of oxPL in the immunogenicity of ferroptotic cancer cells (figure 3).

Importantly, it is not only the oxidation products derived from dying cancer cells that can affect the activity of antigen-presenting cells, because tumor-resident immune cells themselves can produce oxidized lipids, which affect other APCs and thus further modulate the anticancer immune responses. In this regard, it has been shown that cancer-associated DCs accumulated lipids in so-called lipid bodies, which weakened the function of DCs, and so they were not able to effectively stimulate antigen-specific T cells.89 90 This might have been mediated by the oxidative products in the lipid bodies, such as oxidatively truncated electrophilic lipids, which interact with the major stress-induced peptide chaperone HSP70, thereby preventing the presentation of peptides–MHC-I on the DC surface.89 Of note, it was recently demonstrated that polymorphonuclear myeloid-derived suppressor cells (PMN-MDSCs) interfere with cross-presentation of DCs, the process of interference being associated with lipid transfer.91 PMN-MDSC, a part of tumor microenvironment, produces a large amount of ROS, which generate various species of oxidized lipids, such as oxidized triacylglycerols or prostaglandin E2, which might affect DCs and impair the induction of specific antitumor CD8⁺ T-cell responses. In another intriguing study, it has been reported that immunotherapy-activated CD8⁺ T cells promote tumor cell lipid peroxidation and sensitize tumors to ferroptosis through production of IFN-γ.92 In their study, Wang and colleagues demonstrated that IFN-γ released from CD8⁺T-cells downregulates the expression of SLC3A2 and SLC7A11 (subunits of the glutamate–cystine antiporter system X_C^‾) and diminishes the uptake of cystine by tumor cells, and thus promotes cancer cell lipid peroxidation and ferroptosis. Altogether, these results show an important role of lipid oxidation products in shaping the response of the adaptive immune system to cancer. There is a lot of speculation about probable effects of ferroptotic cancer cells, which contain large numbers of various oxidized and peroxidized lipids and PLs,5 12 83 84 on DCs, CD8⁺ T cells and, overall, on the anticancer immune response. Therefore, future studies are needed to clarify whether ferroptosis-derived oxPL are responsible for modulating the immunogenicity of ferroptosis and to explain the non-immunogenicity of late ferroptotic cells65 (figure 3).

Effect of ferroptosis on macrophages

As mentioned previously, ferroptosis affects immune cells in the tumor microenvironment by activating DCs.65 However, the tumor microenvironment is very complex and contains other cell types.93 Ferroptosis might also affect the other cells in the tumor microenvironment. It has been shown that macrophages can behave differently during ferroptosis induction.94 M1 macrophages, which are considered proinflammatory, are very resistant to ferroptosis because of their high levels of inducible nitric oxide synthase. This is in contrast to the alternatively activated M2 macrophages, which are considered anti-inflammatory and are more sensitive to ferroptosis. Secretion of nitric oxide (NO·) inhibits the 15LOX-catalyzed peroxidation of PUFAs in M2 macrophages, thus blocking ferroptosis execution. Hence, ferroptosis can also kill the tumor-promoting M2 macrophages, removing their immunosuppressive effects while preserving the immunostimulatory role of M1 macrophages. However, M1 macrophages are not only more resistant to ferroptosis but can also confer this resistance to other cells94 and contribute to inflammation-driven carcinogenesis. It has been shown that mouse lung epithelial cells were more resistant to ferroptosis in the presence of M1 macrophages due to NO· secretion.94 It is unknown if they can also confer this resistance to cancer cells, which would reduce the therapeutic efficiency of ferroptosis. This is in stark contrast to the widely accepted notion that M1 macrophages possess antitumoral effects. Notably, ferroptotically dying cancer cells can also directly affect macrophages. An intriguing aspect of ferroptosis was recently presented in regard to polarization of tumor-associated macrophages.95 The authors showed that autophagy-dependent ferroptosis of cancer cells leads to procancer polarization of pancreatic tumor-associated macrophages. The process was shown to be regulated through the AGER/RAGE pathway and STAT3-dependent fatty acid oxidation. Ferroptotic pancreatic ductal adenocarcinoma cells release KRAS^G12D, which is taken up by macrophages and causes them to switch to an M2-like protumor phenotype.95 Whether this effect is characteristic of all types of cancer cells undergoing ferroptosis awaits thorough investigation. By modulating the switch between M1 and M2 macrophages, the immune system might be more efficiently activated and might generate strong and long-lasting antitumor immunity.

Lactate and acidification

Lactate is an important metabolite abundant in the tumor microenvironment because cancer cells often switch to glycolysis.96 However, it was recently shown that lactate is not merely a waste product but can be used by the cancer cells to promote ferroptosis resistance.97 Lactate uptake causes an increase in ATP production, leading downstream to upregulation of stearoyl–CoA desaturase-1, which produces monounsaturated fatty acids. Lactate can also directly inhibit ACSL4, which, as mentioned earlier, can enrich PUFAs in the plasma membrane. This combined effect thus decreases the available PUFAs needed for ferroptosis and thereby inhibits cell death. In addition, lactate acidifies the tumor microenvironment, leading to increased iron storage and decreased iron availability for the Fenton reaction. Acidification of the tumor microenvironment also increases apoptosis resistance and enhances the invasiveness of tumors, thereby contributing to metastasis.98 99 Moreover, it was recently shown that the acidic environment drives a stem cell-like phenotype in melanoma cells and thereby increases therapy resistance.100 Overall, a more acidic tumor microenvironment promotes cell death resistance. Therefore, it contributes to therapy resistance and increases the metastatic potential of tumors. However, it was recently reported that in tumor acidosis, long-chain PUFAs not present inside lipid droplets can be peroxidized, leading to increased ferroptosis.101 Use of diacylglycerol O-acyltransferase (DGAT) to inhibit lipid droplet formation resulted in the presence of more long-chain PUFAs available for peroxidation, and an antitumor effect in mice was observed due to ferroptotic cell death induction in cancer cells. It was discovered that acidosis causes autocrine transforming growth factor (TGF)-β secretion leading to enhanced lipid droplet formation.102 Interestingly, TGF-β also caused lipid droplet formation in DCs, resulting in DC dysfunction and less effective immune responses.103 By blocking both TGF-β and DGAT, DC activity recovered, demonstrating an efficient antitumor response.

Ferroptosis in spheroids: challenges and new opportunities

Ferroptosis and cell death modalities in general are still studied mostly in two-dimensional models (cells growing on a flat surface). However, this does not fully resemble the three-dimensional (3D) in vivo situation, where cells are often less responsive to cell death induction therapy. 3D models, such as spheroids, resemble the in vivo situation more closely because the cells produce ECM, which acts as an extra barrier for drugs to cross in order to induce tumor cell death,104–106 and more cell-to-cell contacts are present as well as cell-to-ECM contacts. Therefore, there has been a shift in recent years towards more relevant 3D models, including cancer spheroids.104 107 However, the more complex 3D models need more sophisticated equipment for analysis of cell death.104 Different cell types need to be distinguished, which still requires flow cytometry analysis or confocal microscopy. Recently, a new method has been developed that allows analysis of cell death in intact spheroids without the need to destroy the 3D structure.104

It was recently shown by using spheroids composed of human non-small cell lung tumor cells that the inner cells of the spheroids require nuclear factor erythroid 2-related factor 2 (NRF2) for their survival (figure 4).108 NRF2 is a basic leucine zipper (bZIP) protein that may regulate the expression of the antioxidant proteins protecting against the oxidative damage triggered by injury and inflammation.109 NRF2 and GPX4 are presumably needed to protect the cancer cells from ROS generation, which is toxic for the cells inside the spheroids and tumors. By inhibiting the different antioxidant pathways, this protection is circumvented and an overload of ROS in cancer might lead to efficient ferroptosis induction, eliminating all cancer cells (table 1).

• Download figure
• Open in new tab
• Download powerpoint

Figure 4

Complex interplay of ferroptosis with the tumor microenvironment. Hypoxia has an important influence on ferroptosis. (1) HIF-1α is upregulated in hypoxic conditions. (2) In turn, HIF-1α causes CA9 to be upregulated.114 CA9 then converts CO₂ to H⁺ and HCO₃−, further acidifying the extracellular environment and leading to increased iron solubility. (3) Increased extracellular iron leads to a decrease in the labile iron pool via increased intracellular iron storage. (4) HIF-2α is also upregulated in hypoxia, leading to stronger expression of HILPDA.118 (5) HILPDA then selectively enriches PUFAs in the plasma membrane, creating more substrate for lipid peroxidation. (6) In contrast, NRF2 and GPX4 protect cancer cells from ferroptosis by blocking lipid peroxidation.108 Due to hypoxia, many cancer types become dependent on GPX4 for their survival. (7) However, PUFAs can be stored in LDS by DGAT, making peroxidation impossible and thereby reducing sensitivity to ferroptosis. (8) Acidosis causes autocrine TGF-β secretion, leading to increased LD formation. CA9, carbonic anhydrase 9; DGAT, diacylglycerol O-acyltransferase; GPX4, glutathione peroxidase 4; HIF1α, hypoxia-inducible factor 1α; HIF2α, hypoxia-inducible factor 2α; HILPDA, hypoxia-inducible, lipid droplet-associated protein; NRF2, nuclear factor erythroid 2-related factor 2; LD, lipid droplet; oxPE, oxidized phosphatidylethanolamine; PUFA, polyunsaturated fatty acid.

Moreover, the compactness and stiffness of spheroids resemble the in vivo hypoxic tumor microenvironment,110 which can cause drug resistance through cell cycle arrest and downregulation of procell death proteins (eg, caspase-3 in apoptosis).111 Hypoxia is an important factor that might affect cellular responses to anticancer therapy by modulating the cell death responses. Hypoxia and lactate (discussed previously) are strongly correlated, as hypoxia can cause cancer cells to switch to glycolysis in order to survive.⁹¹ In spheroids, a hypoxic core might lead to upregulation of the hypoxia-inducible factor pathway (figure 4). In this regard, hypoxia-inducible factor-1α can cause overexpression of membrane-bound carbonic anhydrase 9.112 Carbonic anhydrase nine causes CO₂ to be converted to H⁺ and HCO₃⁻ in order to regulate the extracellular pH for better cellular metabolism, but this creates an opportunity for cancer cells to resist ferroptosis. On one hand, this lowered extracellular pH might increase soluble iron in the blood, which results in decreased iron intake and increased iron storage. Since cancer cells are in a constant state of oxidative stress, they might become less susceptible to ferroptosis by increasing iron storage in the cellular compartments,113 thereby reducing the labile iron pool (figure 4), generating less free radicals, and decreasing cell death via the Fenton reaction occurring in ferroptosis.9 By mimicking this oxidative stress using cancer spheroids, better models can be obtained to study ferroptosis and screen for novel ferroptotic cell death inducers. However, on the other hand, it has already been observed that in some tumors, cancer cells become dependent on GPX4 for their survival.114 115 Due to intracellular iron, lipid peroxidation could occur, but this is prevented by GPX4. Lipid peroxidation normally occurs at a slow rate because PUFAs are required to react with the free radicals generated by iron.3 In GPX4-dependent cancers, however, lipid peroxidation occurs frequently. This abundance of lipid peroxides formation, needed for ferroptosis execution, is generated by another protein of the HIF pathway, namely, HIF-2α (figure 4). HIF-2α is needed for survival of cancer cells in hypoxia, but this protein also creates a vulnerability for ferroptosis.116 In renal clear-cell carcinomas, HIF-2α selectively enriches PUFAs, the rate-limiting substrates for lipid peroxidation and thus for ferroptosis, by activating the expression of hypoxia-inducible, lipid droplet-associated protein.116 Since GPX4 is still present inside the cancer cells, GPX4 keeps the process of lipid peroxidation in check. This makes the these cancers dependent on GPX4 for survival. By targeting this enzyme, ferroptosis might be induced more easily. Therefore, targeting GPX4 might lead to more effective cell death induction in cancer.

However, HIF-2α also promotes cell-cycle progression in hypoxic cells, which correlates with enhanced c-Myc promoter binding. Enhanced c-Myc activity likely contributes to HIF-2α-mediated neoplastic progression following loss of the von Hippel–Lindau tumor suppressor and influences the behavior of hypoxic tumor cells.117 This suggests that HIF-2α might actually promote cancer cell proliferation in hypoxia. Also, HIF-2α modulates lipid storage to sustain ER homeostasis, particularly under conditions of nutrient and oxygen limitation, thereby enhancing tumor cell survival.118 Hence, HIF-2α might have different effects on the outcome of ferroptosis treatment. In addition, hypoxia provides a tumor microenvironment that allows selection of cell death-resistant cell types.119 It will be interesting to evaluate how both the protective signals and the intrinsic sensitivity to cell death in cancer affect cancer treatment efficiency.

In conclusion, it is important to keep in mind that in 3D models, including spheroids, many factors can influence each other (eg, cell adhesion, stiffness, concentration gradient, hypoxia, etc), which will have an impact on the overall resistance to ferroptotic cell death and its immunogenicity, and thus to therapy. Nonetheless, 3D models are more reliable for studying cell death responses to cancer therapy, which might require a multidisciplinary approach combining different methods to obtain a complete view of cell death in a three-dimensional setting.