Ferroptosis in Cancer Cell Biology

A major hallmark of cancer is successful evasion of regulated forms of cell death. Ferroptosis is a recently discovered type of regulated necrosis which, unlike apoptosis or necroptosis, is independent of caspase activity and receptor-interacting protein 1 (RIPK1) kinase activity. Instead, ferroptotic cells die following iron-dependent lipid peroxidation, a process which is antagonised by glutathione peroxidase 4 (GPX4) and ferroptosis suppressor protein 1 (FSP1). Importantly, tumour cells escaping other forms of cell death have been suggested to maintain or acquire sensitivity to ferroptosis. Therefore, therapeutic exploitation of ferroptosis in cancer has received increasing attention. Here, we systematically review current literature on ferroptosis signalling, cross-signalling to cellular metabolism in cancer and a potential role for ferroptosis in tumour suppression and tumour immunology. By summarising current findings on cell biology relevant to ferroptosis in cancer, we aim to point out new conceptual avenues for utilising ferroptosis in systemic treatment approaches for cancer.


Introduction
The emergence of electron transport chains as a means to generate a chemical gradient for the generation of adenosine triphosphate (ATP) is as ancient as the first single-celled organism undergoing photosynthesis to produce oxygen. Billions of years of evolution later, multicellular aerobic organisms generate ATP mainly by oxidative phosphorylation (OXPHOS) at the expense of atmospheric oxygen [1]. Herein, protein complexes belonging to the electron transport chain are located in the inner mitochondrial membrane where they transport electrons derived from nicotinamide adenine dinucleotide hydrogen (NADH) via redox reduction to the terminal electron acceptor oxygen (O 2 ), which is thereby reduced to water (H 2 O). NADH is one of the major products of the tricarboxylic acid (TCA) cycle within the mitochondrial lumen driven by metabolites such as Acetyl-CoA derived from glycolysis and catabolic fatty acid oxidation (β-oxidation) [1]. During electron transport, a small proportion of electrons leak out and react with oxygen molecules to generate highly reactive superoxide (O· −2 ) [2]. Superoxides can be transported into the cytosol using the mitochondrial permeability transition pore (mPTP) in the outer mitochondrial membrane. Thereby, OXPHOS is a major source of reactive oxygen species (ROS) in aerobic cells which can cause aberrant oxidation of proteins, lipids and DNA. lipid species within cellular membranes leading to membrane destabilisation and rupture. Lipid peroxidation can be triggered by cytosolic redox active iron (Fe 2+ ) shuttled into cells bound to transferrin via transferrin receptor (TFRC) endocytosis and endosomal release mediated by divalent metal transporter 1 (DMT1). In the presence of H2O2, Fe 2+ catalyses hydroxyl radical (HO•) generation in a Fenton reaction, which sets of a radical lipid peroxidation chain reaction. Lipoxygenase (LOX) can equally catalyse lipid peroxidation using Fe 2+ . As a required prerequisite for ferroptosis, Acyl-CoA synthetase long-chain family member 4 (ACSL4) and lysophosphatidylcholine acyltransferase 3 (LPCAT3) generate the pool of AA-containing target lipids. Glutathione peroxidase 4 (GPX4), in turn, hydrolyses lipid peroxides converting them into their respective non-toxic lipid alcohols (-OH). GPX4 requires glutathione (GSH) as a cofactor which upon its oxidation (GSSG) by GPX4 is reduced to GSH by glutathione reductase (GR). GSH synthesis depends on glutamate cysteine ligase (GCL) and glutathione synthetase (GSS) as well as on intracellular cystine shuttled into the cell in exchange for glutamate mediated by system xc -(SLC3A2 and SCL7A11/xCT). Independently of GSH, ferroptosis suppressor protein 1 (FSP1) generates ubiquinol from ubiquinone which acts as a lipophilic radical trapping agent within membranes thereby protecting from ferroptosis. Oxidative phosphorylation (OXPHOS) and the tricarboxylic acid (TCA) cycle have both been described to be required for ferroptosis triggered by cystine-depletion or system xcbut not GPX4 inhibition. Main peroxidation target PUFAs are arachidonic acid (AA) phosphatidylethanolamine (PE) lipid species within cellular membranes leading to membrane destabilisation and rupture. Lipid peroxidation can be triggered by cytosolic redox active iron (Fe 2+ ) shuttled into cells bound to transferrin via transferrin receptor (TFRC) endocytosis and endosomal release mediated by divalent metal transporter 1 (DMT1). In the presence of H 2 O 2 , Fe 2+ catalyses hydroxyl radical (HO·) generation in a Fenton reaction, which sets of a radical lipid peroxidation chain reaction. Lipoxygenase (LOX) can equally catalyse lipid peroxidation using Fe 2+ . As a required prerequisite for ferroptosis, Acyl-CoA synthetase long-chain family member 4 (ACSL4) and lysophosphatidylcholine acyltransferase 3 (LPCAT3) generate the pool of AA-containing target lipids. Glutathione peroxidase 4 (GPX4), in turn, hydrolyses lipid peroxides converting them into their respective non-toxic lipid alcohols (-OH). GPX4 requires glutathione (GSH) as a cofactor which upon its oxidation (GSSG) by GPX4 is reduced to GSH by glutathione reductase (GR). GSH synthesis depends on glutamate cysteine ligase (GCL) and glutathione synthetase (GSS) as well as on intracellular cystine shuttled into the cell in exchange for glutamate mediated by system xc -(SLC3A2 and SCL7A11/xCT). Independently of GSH, ferroptosis suppressor protein 1 (FSP1) generates ubiquinol from ubiquinone which acts as a lipophilic radical trapping agent within membranes thereby protecting from ferroptosis. Oxidative phosphorylation (OXPHOS) and the tricarboxylic acid (TCA) cycle have both been described to be required for ferroptosis triggered by cystine-depletion or system xcbut not GPX4 inhibition.

Lipid Peroxidation in Mitochondria
GPX4 is one of the major intracellular enzymes involved in hydrolysing lipid peroxides, thereby ensuring repair of lipid peroxide-perturbed cellular membranes. In a counter regulation step, GPX4 inhibition and GSH depletion induces the activation of 12/15 lipoxygenases (12/15-LOX), which are involved in lipid peroxidation [51,54]. Once 12/15-LOX are activated they can directly attack the mitochondrial membrane inducing local lipid peroxidation in neurons [54][55][56]. Therefore, although the current view of ferroptosis favours the idea of lipid peroxide accumulation in the plasma membrane leading to cellular rupture [25], lipid peroxides have also been shown to accumulate in the mitochondrial membrane during ferroptosis. In addition to 12/15-LOX regulating this process, CDGSH iron sulfur domain 1 (CISD1) is an iron-containing protein whose N-terminus is inserted into the outer mitochondrial membrane where it regulates mitochondrial iron uptake. Upon CISD1 deletion, iron accumulation inside the mitochondria facilitates the generation of mitochondrial lipid peroxides contributing to ferroptosis [31]. On the other hand, cholesterol hydroperoxide (ChOOH) species (5alpha-OOH, 6alpha/6beta-OOH, 7alpha/7beta-OOH) and phospholipid hydroperoxide (PLOOH) families (PCOOH, PEOOH, PSOOH) which are peroxidised in the cytosol can be transported to mitochondria via the sterol carrier protein 2 (SCP-2) upon cystine deprivation [55,57]. This suggests a potential role for these lipid peroxides in mitochondria during ferroptosis triggered via this route.
Taken together, mitochondrial metabolism is a main source of cellular ROS and contributes to ferroptosis in many cellular systems. However, mechanisms of how ferroptosis and mitochondria cross signal and whether xCT and GPX4 always signal in a single hierarchical pathway is poorly understood. Therefore, additional studies on the mechanistic contribution of mitochondria in ferroptosis are warranted.

Ferroptosis Discovery as an Oncogene-Selective Death
Although not named at the time, ferroptosis was first discovered as part of a synthetic lethality screen for small molecules selectively targeting Harvey rat sarcoma viral oncogene homolog (HRAS) G12V -mutant human foreskin fibroblasts (BJeLR) [58]. Next, the small molecules erastin [48] and RSL3 [59] were described to induce an oncogenic RAS-specific oxidative type of cell death independently of caspases. The observed HRAS G12V selectivity was proposed to stem from an increased expression of transferrin receptor TFRC and thereby increased intracellular iron levels in HRAS-mutant cells [59]. Moreover, Kirsten sarcoma viral oncogene homolog (KRAS)-mutant Calu-1 lung cancer cells exhibited higher sensitivity to erastin and silencing of KRAS by small hairpin (sh) RNA reduced erastin's efficacy in these cells. Additionally, A-673 cells, which harbour an activating BRAF V600E mutation, became more resistant to erastin treatment upon shRNA-mediated suppression of BRAF [48]. Supporting an increased ferroptosis sensitivity of oncogene-expressing cells, human mammary epithelial (HME) cells expressing mutant epithelial growth factor receptor (EGFR) were more sensitive to cystine deprivation-induced ferroptosis through increased mitogen-activated protein kinase (MAPK) pathway activation [60]. Additionally, in non-small-cell lung cancer (NSCLC) cell lines, it was shown that the level of MAPK pathway activity correlates with sensitivity to ferroptosis induced by cystine deprivation [60].
Apart from their potential influence on ferroptosis execution, oncogenic RAS isoforms and induction of cellular ROS have been extensively studied. In this regards, oncogenic NRAS G12D and HRAS G12V expression was shown to activate ROS production and to trigger a p38 mitogen-activated protein kinase (MAPK)-mediated oxidative stress response [61]. Moreover, RAS-stimulated ROS production was shown to be mediated via the activation of nicotinamide adenine dinucleotide phosphate hydrogen (NADPH)-oxidases (NOX) regulated by the PI3K/Rac1 and RAF/MEK/ERK RAS-effector pathways [62,63]. KRAS also induced translocation of p47 phox , a subunit of NADPH oxidase 1 (NOX1), to the plasma membrane thereby promoting its activation and aiding cellular transformation [64]. Additionally, in the context of inactivation of the tumour suppressor p16, KRAS G12V expression also up-regulated NADPH oxidase 4 (NOX4) [65]. Thereby, oncogenic RAS isoforms may also influence lipid peroxidation and ferroptosis through enhanced feeding into the general cellular ROS pool.
Whilst these and other studies have demonstrated an induction of ROS under acute oncogenic RAS overexpression in vitro, which may promote cellular transformation at the cost of elevated ROS, the question remained how tumours expressing oncogenic KRAS from the endogenous locus would handle ROS in vivo. Interestingly, expression of KRAS G12D , BRAF V619E and MYC ERT2 oncogenes from their respective endogenous loci activated nuclear NRF2, a major transcription factor inducing antioxidant defence genes [66,67]. Importantly, NRF2 is responsible for the positive regulation of different genes involved in GSH synthesis including xCT, glutamate-cysteine ligase catalytic subunit (GCLC) and glutamate-cysteine ligase modifier subunit (GCLM) [68][69][70]. In addition, NRF2 promotes expression of Ferritin (FTH) [71] which may act as a scavenger for redox active iron suggesting a protective function of NRF2 in ferroptosis. In line with this suggestion, hepatocellular carcinoma cells became more sensitive to ferroptosis inducers upon deletion or pharmacological inhibition of NRF2 [71,72]. NRF2 interacts with KEAP1, a tumour suppressor protein, which also regulates the expression of the ATP binding cassette (ABC)-family transporter multidrug resistance protein 1 (MRP1). Interestingly, MRP1 was shown to sensitise to ferroptosis by mediating glutathione efflux [73]. Additionally, it was shown that there is a high co-occurrence of KEAP1 and KRAS mutations in human lung cancers which elevates cellular rates of glutaminolysis [74]. Increased glutamate shuttling into the TCA cycle, namely glutaminolysis, may compete with glutamate requirement for cystine antiport which might affect GSH levels. Yet, NRF2 was also shown to be activated by withaferin A leading to induction of its bona-fide target gene heme oxygenase 1 (HMOX1) which causes an excess of cytosolic labile iron through catalysing its release from haeme promoting ferroptosis in neuroblastoma [75]. Hence, NRF2 activation and target gene expression can lead to opposing outcomes for ferroptosis and its actual effect on ferroptosis is possibly cell type specific.
Also arguing against mutant KRAS as a marker of ferroptosis sensitivity, artesunate (ART) was shown to induce ferroptosis in an iron-and ROS-dependent manner in pancreatic ductal adenocarcinoma (PDAC) cell lines irrespective of KRAS status [76]. Furthermore, erastin treatment induced growth inhibition in acute myeloid leukaemia (AML) cells with an NRAS Q61L mutation (HL-60), but not in other cell lines harbouring RAS mutations (e.g., NRAS G12D or KRAS A18D ) or RAS wild type (WT) suggesting other genetic or non-genetic factors influencing ferroptosis sensitivity [77]. Moreover, there is evidence suggesting that oncogenic KRAS mutations may in fact protect cells from ferroptosis. In this regards it was shown that rhabdomyosarcoma cells (RMS13) expressing either NRAS G12V , HRAS G12V or KRAS G12V were more resistant to ferroptosis than respective empty vector control cells [78]. Interestingly it was recently shown that fibroblasts, expressing oncogenic KRAS G12V , are protected from hydrogen peroxide-induced cell death by up-regulating xCT which allows for KRAS-induced tumourigenicity in vivo [79].
It was demonstrated that the mitochondrial TCA cycle is required for ferroptosis [42]. Interestingly, a shift in the metabolic pathway in KRAS-driven cancers can enhance different characteristics like glutaminolysis, glycolysis or nutrient uptake and thereby affect the TCA cycle [80]. Moreover, NADPH is needed to maintain the TCA cycle, fatty acid synthesis and glutamine metabolism and is used by glutathione reductase (GR) to reduce oxidised GSSG to GSH. Of note, KRAS was shown to elevate NADPH levels through metabolic reprogramming which may enable an improved rate of GSH regeneration and ferroptosis protection [81,82].
Intriguingly, cells expressing homozygous KRAS G12D/G12D demonstrated upregulation of genes related to glycolysis, enhanced glucose consumption and significantly increased levels of TCA cycle enzymes. Moreover, it was shown that homozygous KRAS mutant cells shuttle glucose towards glutathione synthesis, again suggesting that these cells may be more protected from ROS and thereby more resistant to ferroptosis [83].
Apart from oncogene status, in a panel of 117 cancer cell lines from different tissues, sensitivity to erastin-induced cell death was examined to define possible additional determinants of erastin sensitivity.
Interestingly, tissue origin was a much stronger predictor of ferroptosis sensitivity in cancer cell lines than oncogene mutational status [34]. Within this study, diffuse large B-cell lymphoma (DLBCL) cell lines were identified as particularly sensitive to ferroptosis [34]. Although, it was described that suspension cells are more sensitive to small molecules which induce growth inhibition, this was not the underlying mechanism for increased ferroptosis sensitivity in DLBCL which remains to be addressed [34,84]. Apart from tissue origin, another important factor determining ferroptosis sensitivity may be cellular differentiation. Interestingly, in melanoma, ferroptosis sensitivity did not correlate with MAPK pathway activity but instead depended on the dedifferentiation status [34]. Moreover, cells with an expression signature indicative of an epithelial-to-mesenchymal transition (EMT) were more sensitive to GPX4 inhibitors [85]. Interestingly, high cell confluence inhibits ferroptosis via loss of YAP-mediated transcriptional upregulation of TFRC and ACSL4, an experimental variable which therefore needs to be tightly controlled in studies aiming for the identification of factors influencing ferroptosis sensitivity [86].

In Vivo Relevance of Ferroptosis: Lessons Learned from Knockout Mice
One of the key bottlenecks for protection from ferroptosis is the availability of GSH, which serves as a redox equivalent for GPXs including GPX4. As mentioned above, one important source of intracellular cysteine for GSH synthesis is cystine imported via xCT (=Slc7a11). xCT is highly expressed in neurons and brush border membranes of the kidney and duodenum [37]. Furthermore, it has been reported to be expressed in the thyroid gland [87]. Expression can be induced by oxidative stress stimuli such as hydrogen peroxide, oxygen and sodium arsenite leading to elevated GSH production [87,88]. In line with a role in cystine import, xCT-deficient mice display increased cystine concentrations in blood plasma and decreased intracellular GSH levels in comparison to WT mice. However, cellular cysteine levels are comparable between knockout (KO) and WT littermates, suggesting compensatory cysteine synthesis via the transsulfuration pathway [87]. Mouse embryonic fibroblasts (MEFs) isolated from Slc7a11 KO mice fail to survive in cell culture except when supplemented with 2-mercaptoethanol or N-acetyl cysteine (NAC) both described to serve as alternative cystine sources for cells. These data suggest that cells grown in 2D culture in vitro may lack compensation via the transsulfuration pathway. Interestingly, Slc7a11 KO mice are protected from neurotoxic insults induced by 6-hydroxydopamine (6-OHDA), which can trigger Parkinson's Disease (PD) via decreasing extracellular glutamate levels in the brain [89].
During GSH synthesis, glutamate and cysteine are ligated by the glutamate-cysteine ligase (GCL) forming the GSH precursor γ-glutamylcysteine (see Figure 1). GCL is a heterodimer consisting of the catalytic subunit GCLC and the regulatory subunit GCLM [90]. Deletion of Gclc is embryonically lethal before E8.5 as mice are unable to synthesise GSH [91,92]. Moreover, Gclc knockout embryos fail to gastrulate and instead present with increased cell death assayed by TUNEL staining [92]. In contrast, Gclm-deficient mice develop normally but synthesise approximately 75-90% less GSH in comparison to WT littermates, suggesting that either 10-25% of cellular GSH are sufficient for healthy embryonic development or Gclc may fulfil additional functions during embryonic development [93]. Glutathione synthetase (GSS) catalyses condensation of glycine and γ-glutamylcysteine to generate GSH. Mice deficient in Gss die at E7.5, whilst heterozygous littermates survive and do not display an overt phenotype [94]. These data underline the important role of GSH synthesis for normal embryogenesis and development and may in the case of Gclc-deficiency hint at a connection with cell death prevention during embryonic development.
The transsulfuration pathway represents an alternative strategy for cysteine generation from methionine, in which the cystathionine beta-synthase (CBS) processes homocysteine [95,96]. Cbs deletion in mice results in severe growth retardation and postnatal death 5 weeks after birth. This is accompanied by elevated levels of the cysteine precursor homocysteine in plasma, generating a mouse model for severe homocystinuria [97,98]. Interestingly, a recently developed CBS inhibitor has been identified to induce ferroptosis in different cancer cell lines [99]. Yet, whether any aspect of the Cbs KO phenotype is related to overt induction of ferroptosis is unknown.
Downstream of GSH synthesis, GPX4 is a GSH-dependent key regulator of the ferroptosis pathway. Human GPX4 contains a selenocysteine encoded by a "Stop" codon within its catalytic domain. GPX4 has been shown to be indispensable for embryogenesis as Gpx4 KO mice die in utero at E7.5 and display abnormal organ compartmentalisation [100,101]. Interestingly, mice with an inactivating serine exchange mutation in the enzymatically active selenocysteine of Gpx4 die at the same stage of embryonic development whereas mice with heterozygous loss of selenocysteine within Gpx4 are born but display defects in spermatogenesis [52,102]. Of note, mice containing a cysteine instead of a selenocysteine in the catalytically active site of Gpx4 are viable but display seizures and cell death in interneurons when on a mixed genetic background [103,104]. Whole-body inducible knockout of Gpx4 increased oxidative stress and mitochondrial dysfunction in Gpx4-depleted organs with decreased activity of electron transport chain members complex I and IV demonstrating that Gpx4 is an important protector of mitochondrial integrity [105]. Adult mice die within two weeks after systemic induction of Gpx4 KO caused by acute renal failure demonstrating its essential role also for adult tissue homeostasis [55,105]. Interestingly, Gpx4 deletion in hematopoietic cells causes receptor-interacting protein 3 (Rip3)-dependent cell death in erythroid precursor cells resulting in anaemia in mice [106]. Mechanistically, the authors show that GPX4 deletion leads to glutathionylation and inactivation of caspase 8, which triggers necroptosis independently of tumor necrosis factor (TNF). Moreover, T-cell-specific deletion of Gpx4 resulted in T cell ferroptosis which was dependent on receptor-interacting protein 1 (Rip1) and Rip3, suggesting that Gpx4 may be essential for T-cell mediated immune responses [107]. Thereby, these two studies strongly suggest an interaction between ferroptosis and necroptosis takes place in vivo. This has also recently been discussed in Florean et al. [19].
Mice harbouring a x-chromosomal deletion of the important PUFA-regulator Acsl4 are viable but adipocyte-specific deletion results in decreased levels of PUFA-derived fatty acyl-CoAs which fuels lipid peroxidation [108,109].
Thereby, protection from ferroptosis plays a crucial role for tissue homeostasis, whilst many proteins involved in ferroptosis also fulfil metabolic functions unrelated to regulated cell death which may be causative for some of the phenotypes seen in KO mouse models.

Cancer Mouse Models and Genetic Evidence for Ferroptosis in Cancer
In mouse models of mammary tumours, lymphomas and sarcomas, Gclm deficiency leading to a significant reduction of cellular GSH caused impaired tumour initiation and progression [110]. Moreover, murine MC38 colon cancer and Pan02 pancreatic cancer cells, harbouring a CRISPR-Cas9-mediated Slc7a11 knockout displayed impaired in vivo tumour growth accompanied by increased ROS levels and decreased GSH levels [111]. Interestingly, Arensman et al. showed that while cystine uptake by Slc7a11 is essential for T cell viability and proliferation in vitro, it is dispensable for T cells in vivo and does not impact T cell infiltration in tumours in immunocompetent mice. These results show that targeting xCT unlike GPX4 [107] may represent an effective cancer treatment strategy without compromising anti-tumour immunity. Recently, it has been demonstrated that Gpx4 KO significantly impairs tumour growth upon Fer-1 withdrawal, whilst additional Fsp1 KO enhances this effect [39].

Ferroptosis as a p53-Mediated Tumour-Suppressive Mechanism
Cell cycle arrest, cell death and senescence can, in many cases, serve tumour suppression. Therefore, p53, a master regulator of cell cycle arrest and transcription of intrinsic apoptosis genes, has long been assumed to exclusively suppress tumours via these mechanisms [112][113][114]. Surprisingly, mutating three lysine residues within the DNA-binding domain of p53 (p53 3KR ), which prevents their acetylation, was sufficient to suppress tumour growth in mice despite impaired functionality of this mutant in inducing cell cycle arrest, senescence and apoptosis [115]. Strikingly, it was discovered that the p53 3KR mutant was still capable of suppressing xCT transcription via direct promotor binding, thereby sensitising cells to ferroptosis upon ROS-induced stress [116]. Hence, ferroptosis induction has been identified to belong to the arsenal of tumour-suppressive activities of p53. Importantly, mutating a fourth identified acetylation site in p53 (K98) in addition to the three previously described ones led to a complete loss of its tumour suppressor activity. Importantly, this mutant had also lost the capacity to induce ferroptosis sensitisation through xCT suppression, which indicates that the acetylation of the fourth acetylated lysine residue K98 in p53 in addition to the other three sites is vital for tumour suppression and ferroptosis induction by p53 [117]. Interestingly, cytosolic accumulation of p53 mutants typically found in cancer mutants was shown to bind and thereby sequester NRF2, preventing nuclear translocation of NRF2 and induction of its target genes including xCT [118]. These data might offer an explanation as to how p53 mutants may indirectly promote suppression of xCT expression irrespective of DNA binding. In addition, cytoplasmic accumulation preceding nuclear translocation may contribute to wild type p53-mediated xCT suppression. Another study demonstrated that the INK4 locus alternative reading frame (ARF) protein functions as a p53-independent tumour suppressor by limiting NRF2-mediated xCT induction resulting in tumour growth suppression [68]. Thereby, ferroptosis sensitivity is also linked to ARF status and its interaction with NRF2. Interestingly, APR-246, a clinical reactivator of mutant p53 was shown to decrease intracellular glutathione levels through direct binding of cysteines within GSH, leading to an increase of cellular ROS [119]. Yet, it remains to be established whether APR-246-mediated decrease in cellular GSH also stems from p53-mediated xCT suppression and resulting decrease of cystine import. Interestingly, a p53 Ser47 single nucleotide polymorphism (SNP) was identified in a population of African descent which rendered cells more resistant to RSL3 treatment and generated knock-in mice more prone to spontaneous tumour development [120]. This Ser47 variant impaired phosphorylation on adjacent Ser46, which was important for p53 to induce spontaneous cell death in different cell lines [121]. Moreover, a lack of this phosphorylation in Ser46 decreased the ability of Ser47 to bind to p53 target genes which might explain decreased RSL3 sensitivity by de-repression of xCT [120].
In contrast to the above-mentioned studies showing that p53 can induce ferroptosis through suppression of xCT expression [116,118], in human colorectal cancer (CRC) cells, p53 was shown to promote SLC7A11 expression [122]. Furthermore, the same study showed that loss of p53 inhibits accumulation of dipeptidyl-peptidase-4 (DPP4) in the nucleus, which results in enhanced plasma-membrane associated DPP4-dependent lipid peroxidation via ROS-generating NOX enzymes resulting in ferroptosis [122]. Thereby, loss of p53 may equally sensitise cells to ferroptosis in certain contexts. In addition, treatment with the MDM2 inhibitor nutlin-3 stabilised WT p53, reduced cellular ferroptosis sensitivity and induced p21 [123]. p21 upregulation in turn promoted intracellular glutathione storage leading to reduced accumulation of lipid ROS also in the presence of p53 transcriptional activity [123].
Although many of these studies propose that there is a relationship between tumour suppression and ferroptosis sensitivity, so far, there is no genetic evidence that p53 expression directly induces ferroptosis and thereby mediates tumour suppression in vivo. Moreover, results obtained with loss of p53 do not account for the NRF2 sequestering function seen for mutated p53 in the cytosol. Therefore, future studies will have to unravel whether ferroptosis can suppress tumour growth and to what extent it is part of the constitutive p53-mediated tumour-suppressive machinery.

Ferroptosis-Inducing Therapy (FIT) for Cancer Treatment
A breakthrough discovery for a potential use of ferroptosis-inducing therapy (FIT) for cancer treatment came with the finding that cells with acquired resistance to the human epidermal growth factor receptor (HER1/HER2) tyrosinkinase inhibitor Lapatinib, so called persister cells with a high mesenchymal state, were selectively sensitised to the induction of ferroptosis [85,124]. These studies for the first time pointed out that cells which had escaped other means of killing may be selectively sensitised to ferroptosis. Moreover, immunotherapy was recently shown to sensitise to tumour cell ferroptosis (see Section 9). Considering that xCT is often aberrantly expressed in many cancers [125,126], ferroptosis induction may just prove to be a weak spot of cancer.
However, the two main ferroptosis inducers used in vitro, RSL3 and erastin, do not meet pharmacokinetic standards for in vivo application yet due to poor water solubility and metabolic instability [34,127,128]. To circumvent this problem, several efforts have been undertaken to render erastin more suitable for in vivo application. In one approach, triple-negative breast cancer (TNBC) cells, which highly express folate receptor, were treated with erastin packaged in exosomes covered with folate to specifically target folate receptor-overexpressing TNBC [127]. Another metabolically more stable form of erastin is piperazine-coupled erastin which has demonstrated anti-tumour activity in a xenograft model using human fibrosarcoma HT-1080 cells [15,34]. In addition, imidazole-ketone erastin, a metabolically stable variant of erastin, was shown to reduce tumour growth in a SU-DHL-6 DLBCL xenograft model [128].
Whilst small molecule backbones for RSL3 and erastin will have to be further optimised for clinical application, interestingly, a number of food and drug administration (FDA)-approved drugs have been identified to function via the induction of ferroptosis in different cancer entities: Sorafenib, is an FDA-approved multi-kinase inhibitor for treatment of advanced renal cell carcinoma (RCC) and advanced hepatocellular carcinoma (HCC). Molecularly, Sorafenib was shown to inhibit system xc- [38]. Moreover, cell death induced in HCC by sorafenib was suppressed by ferropstatin-1 and iron-chelators [129]. It has further been reported that the tumour suppressor retinoblastoma protein (RB1) suppresses ferroptosis induced by sorafenib treatment [130] suggesting a possible biomarker for sorafenib treatment-induced ferroptosis.
Altretamine (hexamethylmelamine), an FDA-approved alkylating antineoplastic drug, is used for the treatment of ovarian cancer [135]. It has also been shown to inhibit GPX4 and effectively kill U-2932 DLBCL cells in vitro [136].
Statins, such as cerivastatin and simvastatin, have been shown to reduce the synthesis of Coenzyme Q10 via blocking the mevalonate pathway and thereby induce ferroptosis in the human fibrosarcoma cell line HT-1080 [85].
Thereby, the induction of ferroptosis may underlie treatment efficacies of several already approved cancer drugs which may shorten clinical development for the concept of therapeutic induction of ferroptosis in human cancers [17,19]. Which ones these are apart from the already identified ones above will have to be discovered in the future. Current compounds known to induce ferroptosis are summarised in Table 1. Sulfasalazine System xc-inhibitor Yes/Rheumatoid arthritis and inflammatory bowel diseases treatment [24,[131][132][133][134] Withaferin A Gpx4 inactivation/Keap1 inactivation No/Clinical trial for schizophrenia [75,142]

Ferroptosis and Inflammation in Pathology
A hallmark of regulated necrosis pathways is the loss of plasma membrane integrity [143]. This distinct feature of regulated necrosis allows for the release of cytosolic content and exposure of this content to the surrounding tissue. Many of these released cytosolic factors function as immunogenic signals, so called damage-associated molecular patterns (DAMPs) [144]. Not only DAMPs but also alarmins that are released upon cell death or injury have the ability to induce inflammatory immune responses [145]. Mechanistically, DAMPs and alarmins operate as ligands that stimulate pattern recognition receptor (PRR)-expressing immune cells resulting in an inflammatory process, termed necroinflammation if regulated necrosis was the initiator of immune activation [144]. Although precise mechanisms remain to be established, accumulating evidence suggests that ferroptosis may be inflammatory through lipid peroxidation-mediated plasma membrane rupture [146] leading to sterile inflammation [36]. As such, the small molecule inhibitor Fer-1 reduced immune cell infiltrations into diseased tissues in models of acute kidney injury (AKI) [147,148]. Fer-1 treatment also decreased cytokine and chemokine expression levels (C-X-C-motif chemokine 2 (CXCL-2), interleukin 6 (IL-6), p65 subunit of NF-κB [147], interleukin 33 (IL-33), TNF-α, monocyte chemotactic protein 1 (MCP-1) [148]), suggesting that ferroptotic DAMPs may have the capacity to elicit secondary immune cell activation and cytokine production or that production of these alarmins is directly blocked by Fer-1. Interestingly, the application of the caspase inhibitor zVAD-fmk and the RIPK1 inhibitor necrostatin-1 appeared to be futile for folic acid-induced AKI amelioration [148]. Mononuclear interstitial infiltrations were also observed to occur upon GPX4 deletion in kidneys [55]. These data suggest that inflammatory processes and activation of the immune system in acute kidney injury are induced through ferroptotic cell death.
Furthermore, Trolox, another radical scavenger known to block ferroptosis, was equally shown to decrease expression levels of proinflammatory cytokines (TNF-α, IL-6, interleukin 1 beta (IL-1β)) in steatohepatitis [153]. This metabolic liver disease is characterised by the progression from steatosis into fibrosis through inflammatory processes [154]. Overall, it is strongly suggested that ferroptosis is closely associated with overt inflammatory signatures exhibiting elevated levels of proinflammatory cytokines in damaged tissues [147,148,151,153]. Collectively, these results imply that ferroptosis exerts a critical role in early inflammatory processes and that the prevention of necroinflammation provides a novel therapeutic strategy in combating inflammatory tissue damage and disease.
Although other regulated necrosis pathways such as necroptosis and pyroptosis were shown to mitigate inflammation in several diseases [155], a growing body of literature suggests that ferroptosis may be present from the outset of inflammatory [20,148,149,153] potentially triggering or sensitising to other inflammatory events. Importantly, immunogenic cell death results in the recruitment and activation of immune cells through alarmins which in turn has been hypothesised to induce further necrotic cell death pathways, culminating in a necroinflammatory auto-amplification loop [156]. However, the exact molecular pathways/machineries driving primary and secondary necroinflammation and the underlying hierarchical sequence of events remains yet to be disentangled.

M1 versus M2-Type Immunity in Cancer: Potential Implications for Ferroptosis
Necroptotic cancer cells can release DAMPs, triggering inflammation in normal tissues [157]. Recently, it was shown that ferroptosis can equally allow for the release of the known DAMP high-mobility group protein B1 (HMGB1) in a manner dependent on the cellular autophagy machinery [158]. Apart from this, it is conceivable that the induction of ferroptosis in cancer cells might elicit innate immune responses which can influence tumours via various ways. The M1/M2 polarisation axis proposed by Mills et al. in 2000 provides a simplistic approach to distinguish between two types of macrophages with different metabolic programmes [159]. Whereas the M1 macrophage phenotype was reported to result in tumour regression [160] and inhibition of tumour growth [161], M2 macrophages have been shown to exhibit tumour-promoting activity as they produce angiogenesis factors [162] and inhibit M1 macrophages, thereby suppressing anti-tumour immunity. The importance of macrophage polarisation in the tumour microenvironment was further demonstrated by showing a chemokine-mediated decrease in M2-like macrophages in the vicinity of the tumour leading to reduced tumour growth [163]. In view of the intricate dynamics between innate and adaptive immune responses, it is important to highlight that M1/M2 macrophages promote adaptive Th1/Th2 lymphocyte responses, respectively [159]. M1/Th1 responses are associated with the release of proinflammatory cytokines, such as TNF [164] as well as IFN-γ, a cytokine that is in turn able to stimulate M1 polarisation in a feed-forward loop [165] as well as downregulate xCT [137]. In contrast, macrophage-induced Th2 responses are closely linked to anti-inflammatory mediators, such as IL-4 and IL-10 [164,166]. Interestingly, cytokines released from Th2 T-cell clones have been shown to avert cytokine production by Th1 cells, essentially strengthening the suppression of M1/kill type activity [159,167]. Furthermore, tumour cells have developed mechanisms to stimulate tumour cell migration and to maintain a high M2 to M1 ratio in the tumour microenvironment, thereby working in their best interest [168][169][170]. Of note, the majority of tumour-associated macrophages (TAM) in the tumour microenvironment is predominantly composed of M2 macrophages [171]. As a result, anti-cancer macrophage innate conversion therapies have emerged in recent years, directly targeting intracellular signalling pathways to increase the M1/M2 ratio [172,173]. Beatty et al. showed that the combination therapy of gemcitabine and agonist CD40 antibody obtained macrophage-dependent tumour regression in vivo which was suggested to be attributable to M1 activity [174].
As the M1/M2 dichotomy influences inflammatory events in opposite directions, it is of great interest to elucidate the quality of a ferroptotic secretome released from dying cancer cells as it may either trigger M2 tumour-promoting activity or initiate an M1 anti-tumour immune response. Given that ferroptosis in tissues has been causatively linked to the presence of TNF, this may suggest the potential for the promotion of an M1-type immune microenvironment. Supporting the idea of ferroptosis to be inflammatory in the context of cancer, ferroptosis triggered in neuroblastoma by withaferin A led to an influx of immune cells into the tumour [75]. Moreover, recent research has proposed that sensitising tumour cells to ferroptosis enhances the efficacy of combinatorial cancer immune-activatory therapies [175]. Interestingly, xCT was reported to be down-regulated through both immunotherapy-activated IFNγ [137,175] as well as radiotherapy-induced ataxia-telangiectasia mutated gene (ATM), resulting in increased tumour cell lipid peroxidation and tumoural ferroptosis [175]. Furthermore, immunotherapy-induced cytokines, namely IFNγ and TNF-α, were shown to provoke dedifferentiation in melanoma cells which was accompanied by resistance development. Intriguingly, this transition by tumour cells resulted in an increased susceptibility to ferroptosis induction and affected their overall immunosuppressive abilities [176]. Therefore, not only the induction but the sensitisation of tumour cells to ferroptosis may supply a novel tool for anti-tumour immunity and may serve the optimisation of established immunotherapies.

Conclusions
Many cancers highly express xCT, suggesting a selective dependency on either cystine or GSH, which may be exploited therapeutically by targeting these signalling nodes. Moreover, cancer cells appear to acquire ferroptosis sensitivity as part of an escape strategy against other targeted therapies, posing an opportunity for FIT in therapeutic management of relapse. Whilst the pathway will require additional in-depth characterisation and the validity of genetic induction of ferroptosis remains to be tested in genetically engineered mouse models of cancer, intriguingly, immunotherapy seems to positively interact with ferroptosis in vivo. These and other data strongly suggest ferroptosis to either be directly immunogenic or to prime an inflammatory response to other forms of regulated necrosis in the tumour microenvironment. Collectively, we propose that tumour cell ferroptosis may promote four possible outcomes in cancer: (A) if immuno-silent and completely killing, tumour cell ferroptosis should suppress tumours, (B) if immuno-silent but fractional killing causes selection, ferroptosis may result in tumour promotion through selection of the fittest cancer cell clone, (C) if an M1-type immunity is triggered, ferroptosis should be tumour-suppressive via activation of anti-tumour immunity and (D) if an M2-type immunity is triggered, ferroptosis would be overall tumour-protective, as it would aid shielding tumours against anti-tumour immune attack (proposed concepts summarised in Figure 2). Importantly, mixed responses within these four response types are likely to occur due to tumour heterogeneity. Thus, it will be important in future work to determine not only cellular determinants of ferroptosis sensitivity and resistance but also systemic responses and mechanisms of how these are interlinked with other types of regulated cell death in order to fully harness the potential of FIT for cancer treatment.
protective, as it would aid shielding tumours against anti-tumour immune attack (proposed concepts summarised in Figure 2). Importantly, mixed responses within these four response types are likely to occur due to tumour heterogeneity. Thus, it will be important in future work to determine not only cellular determinants of ferroptosis sensitivity and resistance but also systemic responses and mechanisms of how these are interlinked with other types of regulated cell death in order to fully harness the potential of FIT for cancer treatment.

Conflicts of Interest:
The authors declare no conflict of interest.