For a long time, cell death has been dismissed by biologists as an inevitable and, hence, spurious consequence of cellular life. A large body of experimental evidence accumulating over the past decades, however, has unveiled and characterized in ever greater detail a set of genetically encoded mechanisms for targeted elimination of superfluous, irreversibly damaged, and/or potentially harmful cells [1,2,3,4]. Intriguingly, regulated cell death (RCD) is not unique to multicellular life forms, a setting in which RCD has an obvious advantage for organismal homeostasis in both physiological and pathological settings [5,6,7,8,9], but is also found (in simplified variants) among unicellular eukaryotes living (at least for part of their life cycle) in colonies (such as several yeast species and Dictyostelium discoideum) [10,11,12,13,14,15], and at least in some prokaryotes (e.g., Escherichia coli) [16]. In striking contrast with accidental cell death (ACD)—the instantaneous and catastrophic demise of cells exposed to severe insults of physical (e.g., high pressures, temperatures, or osmotic forces), chemical (e.g., extreme pH variations), or mechanical (e.g., shear forces) nature—RCD relies on a dedicated molecular machinery, implying that it can be modulated (i.e., delayed or accelerated) by pharmacological or genetic interventions [5, 17].

Although the underlying molecular mechanisms exhibit considerable overlap (see below), RCD is involved in two diametrically opposed scenarios. On the one hand, RCD can occur in the absence of any exogenous environmental perturbation, hence operating as a built-in effector of physiological programs for development or tissue turnover [6, 18]. These completely physiological forms of RCD are generally referred to as programmed cell death (PCD). On the other hand, RCD can originate from perturbations of the intracellular or extracellular microenvironment, when such perturbations are too intense or prolonged for adaptative responses to cope with stress and restore cellular homeostasis [5]. Importantly, stress-driven RCD also constitutes a strategy for the preservation of a biological equilibrium, hence resembling adaptative stress responses. However, while adaptative stress responses operate at the cellular level (which—by extension—promotes the maintenance of homeostasis at the level of organism or colony), RCD directly operates at the level of the organism or colony in spite of cellular homeostasis [5]. Such a homeostatic function not only reflects the elimination of useless or potentially dangerous cells, but also the ability of dying cells to expose or release molecules that alert the organism or colony about a potential threat. Such danger signals are commonly referred to as damage-associated molecular patterns (DAMPs) or alarmins [19,20,21,22].

Cell death manifests with macroscopic morphological alterations. Together with the mechanisms whereby dead cells and their fragments are disposed of, such morphotypes have historically been employed to classify cell death into three different forms: (1) type I cell death or apoptosis, exhibiting cytoplasmic shrinkage, chromatin condensation (pyknosis), nuclear fragmentation (karyorrhexis), and plasma membrane blebbing, culminating with the formation of apparently intact small vesicles (commonly known as apoptotic bodies) that are efficiently taken up by neighboring cells with phagocytic activity and degraded within lysosomes; (2) type II cell death or autophagy, manifesting with extensive cytoplasmic vacuolization and similarly culminating with phagocytic uptake and consequent lysosomal degradation; and (3) type III cell death or necrosis, displaying no distinctive features of type I or II cell death and terminating with the disposal of cell corpses in the absence of obvious phagocytic and lysosomal involvement [23, 24]. Of note, this morphological classification is still extensively employed, irrespective of multiple limitations, and caveats. Starting from 2005, the Nomenclature Committee on Cell Death (NCDD) gathered on a regular basis (1) to address the issues related to the use of a nomenclature of cell death based on morphological grounds; (2) to precisely define major cell death modalities on a genetic, biochemical, pharmacological, and functional (rather than morphological) basis; (3) to distinguish essential (causal) from accessory (correlative) aspects of the death process; and (4) to identify consensus criteria for the identification of dead cells with irreversible plasma membrane permeabilization or complete cellular fragmentation [17, 25,26,27,28].

As the field continues to progress and novel signaling pathways that orchestrate RCD are still being characterized, we propose here an updated classification of cell death modalities centered on molecular and essential aspects of the process (Fig. 1 and Box 1). A major focus will be placed on the signal transduction modules involved in the initiation, execution, and propagation of cell death, as well as on the pathophysiological relevance of each of the main types of RCD.

Fig. 1
figure 1

Major cell death subroutines. Mammalian cells exposed to unrecoverable perturbations of the intracellular or extracellular microenvironment can activate one of many signal transduction cascades ultimately leading to their demise. Each of such regulated cell death (RCD) modes is initiated and propagated by molecular mechanisms that exhibit a considerable degree of interconnectivity. Moreover, each type of RCD can manifest with an entire spectrum of morphological features ranging from fully necrotic to fully apoptotic, and an immunomodulatory profile ranging from anti-inflammatory and tolerogenic to pro-inflammatory and immunogenic. ADCD: autophagy-dependent cell death, ICD: immunogenic cell death, LDCD: lysosome-dependent cell death, MPT: mitochondrial permeability transition.

Box 1 Operational definitions

Accidental cell death (ACD). Virtually instantaneous and uncontrollable form of cell death corresponding to the physical disassembly of the plasma membrane caused by extreme physical, chemical, or mechanical cues.

Anoikis. Specific variant of intrinsic apoptosis initiated by the loss of integrin-dependent anchorage.

Autophagy-dependent cell death. A form of RCD that mechanistically depends on the autophagic machinery (or components thereof).

Autosis. A specific instance of autophagy-dependent cell death that critically relies on the plasma membrane Na+/K+-ATPase.

Cell death. Irreversible degeneration of vital cellular functions (notably ATP production and preservation of redox homeostasis) culminating in the loss of cellular integrity (permanent plasma membrane permeabilization or cellular fragmentation).

Cellular senescence. Irreversible loss of proliferative potential associated with specific morphological and biochemical features, including the senescence-associated secretory phenotype (SASP). Cellular senescence does not constitute a form of RCD.

Efferocytosis. Mechanism whereby dead cells and fragments thereof are taken up by phagocytes and disposed.

Entotic cell death. A type of RCD that originates from actomyosin-dependent cell-in-cell internalization (entosis) and is executed by lysosomes.

Extrinsic apoptosis. Specific variant of RCD initiated by perturbations of the extracellular microenvironment detected by plasma membrane receptors, propagated by CASP8 and precipitated by executioner caspases, mainly CASP3.

Ferroptosis. A form of RCD initiated by oxidative perturbations of the intracellular microenvironment that is under constitutive control by GPX4 and can be inhibited by iron chelators and lipophilic antioxidants.

Immunogenic cell death. A form of RCD that is sufficient to activate an adaptive immune response in immunocompetent hosts.

Intrinsic apoptosis. Type of RCD initiated by perturbations of the extracellular or intracellular microenvironment, demarcated by MOMP, and precipitated by executioner caspases, mainly CASP3.

Lysosome-dependent cell death. A type of RCD demarcated by primary LMP and precipitated by cathepsins, with optional involvement of MOMP and caspases.

Mitochondrial permeability transition (MPT)-driven necrosis. Specific form of RCD triggered by perturbations of the intracellular microenvironment and relying on CYPD.

Mitotic catastrophe. Oncosuppressive mechanism for the control of mitosis-incompetent cells by RCD or cellular senescence. Per se, mitotic catastrophe does not constitute a form or RCD.

Mitotic death. Specific variant of RCD (most often, intrinsic apoptosis) driven by mitotic catastrophe.

Necroptosis. A modality of RCD triggered by perturbations of extracellular or intracellular homeostasis that critically depends on MLKL, RIPK3, and (at least in some settings) on the kinase activity of RIPK1.

NETotic cell death. A ROS-dependent modality of RCD restricted to cells of hematopoietic derivation and associated with NET extrusion.

Parthanatos. A modality of RCD initiated by PARP1 hyperactivation and precipitated by the consequent bioenergetic catastrophe coupled to AIF-dependent and MIF-dependent DNA degradation.

Programmed cell death (PCD). Particular form of RCD that occurs in strictly physiological scenarios, i.e., it does not relate to perturbations of homeostasis and hence does not occur in the context of failing adaptation to stress.

Pyroptosis. A type of RCD that critically depends on the formation of plasma membrane pores by members of the gasdermin protein family, often (but not always) as a consequence of inflammatory caspase activation.

Regulated cell death (RCD). Form of cell death that results from the activation of one or more signal transduction modules, and hence can be pharmacologically or genetically modulated (at least kinetically and to some extent).

Intrinsic apoptosis

Intrinsic apoptosis is a form of RCD initiated by a variety of microenvironmental perturbations including (but not limited to) growth factor withdrawal, DNA damage, endoplasmic reticulum (ER) stress, reactive oxygen species (ROS) overload, replication stress, microtubular alterations or mitotic defects [29,30,31,32,33,34]. Apoptotic cells retain plasma membrane integrity and metabolic activity (to some degree) as the process proceeds to completion, which—in vivo—allows for the rapid clearance by macrophages or other cells with phagocytic activity (a process commonly known as efferocytosis) [35]. Importantly, intrinsic (and extrinsic, see below) apoptosis and consequent efferocytosis are not always immunologically silent, as previously thought (see below) [36, 37]. In vitro, end-stage apoptosis is generally followed by complete breakdown of the plasma membrane and the acquisition of a necrotic morphotype (secondary necrosis), unless cultured cells display phagocytic activity [38], a process that has recently been linked to the pore-forming activity of gasdermin E (GSDME; best known as DFNA5) [39].

The critical step for intrinsic apoptosis is irreversible and widespread mitochondrial outer membrane permeabilization (MOMP) [40, 41], which is controlled by pro-apoptotic and anti-apoptotic members of the BCL2, apoptosis regulator (BCL2) protein family, a group of proteins sharing one to four BCL2 homology (BH) domains (i.e., BH1, BH2, BH3, and BH4) [29, 42, 43]. In response to apoptotic stimuli, MOMP is mediated by BCL2 associated X, apoptosis regulator (BAX), and/or BCL2 antagonist/killer 1 (BAK1; best known as BAK), both of which contain four BH domains and a conserved transmembrane domain [44,45,46]. Together with BOK, BCL2 family apoptosis regulator (BOK) [47], BAX and BAK are the only BCL2 family members characterized so far in mammalian cells for their ability to form pores across the outer mitochondrial membrane (OMM) and possibly other intracellular membranes [29, 42, 43]. In physiological conditions, BAX continuously cycles between the OMM and the cytosol, where it exhibits a quiescent monomeric or inactive dimeric conformation [48,49,50]. In contrast, BAK constitutively resides at the OMM, where it inserts within the lipid bilayer via its hydrophobic C-terminal α9 helix upon interaction with voltage dependent anion channel 2 (VDAC2) [51,52,53,54]. Of note, some degree of BAK retrotranslocation from the OMM to the cytosol has been documented [55]. Upon induction of apoptosis, BAX retrotranslocation ceases as the mitochondrial pools of BAX and BAK undergo direct or indirect activation (see below) by pro-apoptotic BH3-only proteins [48, 56,57,58,59].

These pro-apoptotic members of the BCL2 protein family (which contain a single BH3 domain) are activated transcriptionally or post-translationally as specific organelles or cellular compartments experience perturbations of homeostasis, de facto operating as cellular transducers of stress signaling [60,61,62,63]. Some BH3-only proteins—such as BCL2 binding component 3 (BBC3; best known as p53-upregulated modulator of apoptosis, PUMA), BCL2 like 11 (BCL2L11; best known as BCL2-interacting mediator of cell death, BIM), and phorbol-12-myristate-13-acetate-induced protein 1 (PMAIP1; best known as NOXA)—are mainly activated by transcriptional upregulation, while others—such as BH3 interacting domain death agonist (BID)—mostly undergo post-translational activation [64,65,66,67,68,69,70]. BID, BIM, PUMA, and NOXA share the ability to physically (but transiently) interact with the mitochondrial pool of BAX and/or BAK (hence being known as “activators”) to promote a series of conformational changes [59, 64, 67, 71,72,73,74] culminating with the dissociation of the core and latch domains of BCL2 effectors [75,76,77]. The current view is that activated BAX and BAK form homodimers (also heterodimers in specific settings), resulting in the release of BH3-only proteins and further dimer-by-dimer oligomerization [76,77,78,79,80,81,82,83]. Oligomerization ultimately leads to the assembly of a toroidal lipidic pore that alters mitochondrial permeability and causes profound rearrangements of the mitochondrial ultrastructure [78, 84,85,86]. In line with this model, it has recently been shown that (1) BAX can form rings or linear/arc-shaped oligomers that perforate the OMM [84, 85], and (2) MOMP proceeds upon the formation of pores (impinging on OMM curvature stress), which can vary in size depending on the number of BAX dimers recruited [87].

MOMP is antagonized by anti-apoptotic members of the BCL2 family, including BCL2 itself, BCL2 like 1 (BCL2L1; best known as BCL-XL), MCL1, BCL2 family apoptosis regulator (MCL1), BCL2 like 2 (BCL2L2; best known as BCL-W), and BCL2 related protein A1 (BCL2A1; best known—in human—as BFL-1) [29, 42, 43]. These pro-survival proteins contain all four BH domains, are generally inserted into the OMM or the ER membrane through their α9 helix, and mainly exert anti-apoptotic functions by directly binding pro-apoptotic members of the BCL2 family, an activity that generally—but not always—depends on a hydrophobic binding groove formed by BH1, BH2, and BH3 domains [88,89,90,91,92,93,94]. In addition, some anti-apoptotic BCL2 family members have been proposed to promote cellular survival by: (1) regulating Ca2+ homeostasis at the ER [95,96,97,98,99]; (2) promoting bioenergetic metabolism upon interaction with the F1FO ATP synthase [100,101,102,103,104]; and (3) contributing to the regulation of redox homeostasis [105,106,107,108,109]. However, the importance of these functions has been challenged by the generation of cell lines that lack all major anti-apoptotic and pro-apoptotic BCL2 family members [93]. Thus, most pro-survival BCL2 family members inhibit BAX and BAK by preventing their oligomerization and pore-forming activity either directly, upon physical sequestration at the OMM, or indirectly, following the sequestration of BH3-only activators [29, 64, 79, 110]. Of note, in physiological conditions, some anti-apoptotic BCL2 proteins, such as BCL-XL, exert a protective role by promoting the retrotranslocation of BAX and (to a lesser degree) BAK from the mitochondria to the cytoplasm, thus limiting their mitochondrial pool [48, 55, 111]. Evidence from T cells and platelets suggests that such retrotranslocation occurs in vivo, resulting in the physiological inhibition of BAK by BCL-XL [112]. Importantly, some BH3-only proteins including BCL2 associated agonist of cell death (BAD), Bcl2 modifying factor (BMF), or harakiri, BCL2 interacting protein (HRK) promote MOMP in the absence of a physical interaction with BAX or BAK. These BH3-only proteins, which are sometimes referred to as “sensitizers” or “inactivators” bind to anti-apoptotic BCL2 family members and hence limit their availability to sequester BAX, BAK, or BH3-only activators [58, 93].

Different BH3-only proteins have been suggested to preferentially bind specific anti-apoptotic BCL2 family members (e.g., BID, BIM, and PUMA potently bind all anti-apoptotic BCL2 family members; BAD preferentially interacts with BCL2, BCL-XL, and BCL-W; NOXA preferentially inhibits MCL1; and HRK preferentially inhibits BCL-XL) [57, 113, 114]. In vitro results suggest that the distinction between sensitizers and activators may be much less rigid than previously thought [79, 114,115,116,117]. However, overexpression of BH3-only sensitizers induces minimal apoptosis in cells lacking BID, BIM, PUMA, and NOXA [64], suggesting that BH3-only activators function downstream of BH3-only sensitizers. Of note, the interaction between anti-apoptotic and pro-apoptotic BCL2 family members has major therapeutic implications, with BCL2 representing the pharmacological target of the FDA-approved BH3 mimetic venetoclax (also known as ABT-199) and other molecules with a similar mechanism of action that are currently under development (e.g., the MCL1 inhibitor S63845) [118, 119]. Indeed, venetoclax kills chronic lymphocytic leukemia (CLL) cells by mimicking the activity of BH3-only proteins [120]. Recently, a mechanism of resistance to BH3 mimetics has been ascribed to the tight association between BCL-XL and BH3-only activators at subcellular membranes [121, 122]. The relevance of this mechanism for CLL patients under venetoclax treatment, however, remains to be elucidated.

Confirming the essential role of BCL2 family members for MOMP and the high degree of overlap between the machineries responsible for stress-driven RCD and PCD, the co-deletion of Bax and Bak1 not only renders a large panel of cell types profoundly resistant to diverse lethal stimuli [74], a phenotype that in some settings can be exacerbated by the co-deletion of Bok [123], but also causes perinatal lethality in mice as a consequence of severe developmental defects [124]. Along similar lines, Bcl2l11−/−Bmf−/− as well as Bid−/−Bcl2l11−/−Bbc3−/− mice die prematurely or display severe developmental defects, respectively [68, 125]. However, transformed cells lacking all major BH3 activators (i.e., BID, BIM, PUMA, and NOXA) can still undergo apoptosis in response to DNA-damaging agents or downregulation of BCL2, BCL-XL, and MCL1 [64]. This observation is in line with the notion that BAX and BAK can self-activate in the absence of anti-apoptotic BCL2 family members and pro-apoptotic BH3 proteins (according to a relatively slow kinetics) [64, 93]. Perhaps, BAX and BAK can even be activated independently of BH3-only proteins by the concerted action of the prolyl isomerase peptidylprolyl cis/trans isomerase, NIMA-interacting 1 (PIN1) and either tumor protein p53 (TP53; best known as p53) [126,127,128] or ATR serine/threonine kinase (ATR) [129, 130], several proteins containing BH-like motifs [131], as well as by detergents, heat, pH changes, or specific monoclonal antibodies [132]. That said, the actual pathophysiological relevance of non-canonical BAX and BAK activation remains to be formally established. Both anti-apoptotic and pro-apoptotic BCL2 proteins are also subjected to tight transcriptional and post-translational regulation, involving (but not limited to) proteasomal degradation, phosphorylation, and subcellular (re)localization [48, 108, 133,134,135,136,137,138]. Finally, it is becoming increasingly evident that mitochondrial size and shape [139,140,141] as well as lipid composition [142, 143] can influence the likelihood of mitochondria to undergo irreversible MOMP. These observations exemplify the number of factors involved in MOMP at the level of single mitochondria. Of note, active BAX and BAK have also been proposed to (1) permeabilize ER membranes, especially in response to reticular stress, leading to release of luminal ER chaperones into the cytosol [30, 144]; and (2) favor the activation of type 1 inositol trisphosphate receptors at the ER, resulting in the cytosolic leak of Ca2+ ions and consequent mitochondrial Ca2+ uptake [96, 145]. However, the actual relevance of ER permeabilization for intrinsic apoptosis remains to be elucidated. That said, the contact sites between mitochondria and the ER (which are commonly known as mitochondria-associated ER membranes) appear to regulate a plethora of cellular processes that influence RCD or its immunological consequences, including (but not limited to) ER stress signaling, the transfer of Ca2+ ions from the ER to mitochondria, and inflammatory reactions [146,147,148].

As for BOK, it has been proposed that this BCL2 protein contributes to the regulation of ER homeostasis, as demonstrated by its prominent localization at the ER membrane [149] and the defective apoptotic response of Bok−/− cells to some ER stressors [150]. Moreover, BOK has recently been shown to induce MOMP in the absence of BAX and BAK and independently of other BCL2 family members [47, 151, 152]. In particular, BOK appears to be constitutively active and to be antagonized by an ER-associated degradation pathway rather than by anti-apoptotic BCL2 proteins [47]. BOK is also regulated by a mechanism involving the binding to inositol 1,4,5-trisphosphate (IP3) receptors, which reportedly limits its proteasomal degradation [153]. Of note, Bok−/−, Bax−/−Bok−/− as well as Bak1−/−Bok−/− mice display no obvious abnormalities (except for persistence of primordial follicle oocytes in aged Bax−/−Bok−/− females) [154, 155], implying that physiological functions of BOK can be compensated for by BAK and/or BAX.

MOMP directly promotes the cytosolic release of apoptogenic factors that normally reside in the mitochondrial intermembrane space [40, 44, 156]. These mitochondrial proteins include (but are not limited to) cytochrome c, somatic (CYCS), which usually operates as an electron shuttle in the mitochondrial respiratory chain [157,158,159,160], and diablo IAP-binding mitochondrial protein (DIABLO; also known as second mitochondrial activator of caspases, SMAC) [161,162,163]. The release of CYCS and SMAC to the cytosol is favored by mitochondrial cristae remodeling [164], which relies on the oligomerization and activation of OPA1, mitochondrial dynamin like GTPase (OPA1) [165], possibly preceded by the BAX-dependent and BAK-dependent activation of OMA1 zinc metallopeptidase (OMA1) [166, 167], and/or dynamin 1 like (DNM1L; best known as DRP1) [168]. Accordingly, nitric oxide (NO) has been shown to precipitate the release of apoptogenic factors from mitochondria upon direct nitrosylation of DRP1 (at least in some settings) [169,170,171]. The cytosolic pool of CYCS binds to apoptotic peptidase activating factor 1 (APAF1) and pro-caspase 9 (CASP9) in a deoxyATP-dependent manner to form the supramolecular complex known as apoptosome, which is responsible for CASP9 activation [160]. Recently, the structure of the apoptosome from multiple organisms including humans has been characterized at atomic resolution [172,173,174]. These studies revealed that the autocatalytic maturation of CASP9 within the apoptosome occurs through generation of CASP9 homodimers and CASP9-APAF1 heterodimers/multimers upon association of their respective caspase recruitment domains (CARDs) [175,176,177,178].

Activated CASP9 can catalyze the proteolytic activation of CASP3 and CASP7, which are widely perceived as the enzymes responsible for cell demolition during intrinsic (and extrinsic, see below) apoptosis in mammalian cells (and hence are commonly known as executioner caspases) [179, 180]. Cytosolic SMAC precipitates apoptosis by associating with members of the inhibitor of apoptosis (IAP) protein family, including X-linked inhibitor of apoptosis (XIAP) [162, 163, 181]. To acquire apoptogenic activity, SMAC must undergo a proteolytic maturation step that unleashes its latent IAP-binding domain, which is catalyzed by the inner membrane peptidase (IMP) complex [182] and perhaps by the inner mitochondrial membrane (IMM) protease presenilin associated rhomboid like (PARL) [183]. XIAP is the only IAP protein family member that counteracts the apoptotic cascade by stably binding to and hence physically blocking caspases [184, 185]. Conversely, baculoviral IAP repeat containing 2 (BIRC2; best known as c-IAP1) and BIRC3 (best known as c-IAP2) mostly do so as they (1) drive the upregulation of potent anti-apoptotic factors such as CASP8 and FADD like apoptosis regulator (CFLAR; best known as c-FLIP) [186]; (2) promote caspase inactivation by virtue of their E3 ubiquitin ligase activity [187,188,189,190,191,192,193,194,195]; (3) ubiquitinate receptor interacting serine/threonine kinase 1 (RIPK1) and hence trigger pro-survival NF-κB signaling [196,197,198]; and (4) perhaps promote SMAC degradation at mitochondria through a mechanism that depends on BCL2 proteins [199]. Of note, MOMP eventually leads to the dissipation of the mitochondrial transmembrane potential (Δψm)—mostly as a consequence of the respiratory impairment imposed by the loss of CYCS—and hence to the cessation of Δψm-dependent mitochondrial functions (including ATP synthesis and some forms of protein import) [200,201,202,203]. Intriguingly, BAK and BAX may not always be required for pro-apoptotic stimuli to promote CYTC release and consequent caspase activation, even in conditions in which mitochondrial permeability transition (MPT; see below) is disabled [204, 205]. This may suggest the existence of another—presently unidentified—mechanism for MOMP, possibly involving specific lipids like ceramide [206, 207]. The actual pathophysiological relevance of this potential mechanism remains obscure.

The catalytic activity of executioner caspases precipitates cellular demise and is responsible for many of the morphological and biochemical correlates of apoptosis, including DNA fragmentation [208], phosphatidylserine (PS) exposure [209, 210], and the formation of apoptotic bodies [211, 212]. CASP3 favors DNA fragmentation by catalyzing the proteolytic inactivation of DNA fragmentation factor subunit alpha (DFFA; best known as ICAD), hence unleashing the catalytic activity of DFFB (best known as CAD) [213,214,215]. Recent experimental evidence demonstrates that CASP3 promotes PS exposure by activating proteins involved in PS externalization, such as the phospholipid scramblases [216,217,218], or inactivating factors that mediate PS internalization, such as phospholipid flippases [219,220,221]. Thus, in response to apoptotic stimuli, active CASP3 reportedly cleaves (1) XK related protein 8 (XKR8), which interacts with basigin (BSG) or neuroplastin (NPTN) to form a phospholipid-scrambling complex responsible for PS exposure [216, 217], and (2) ATPase phospholipid transporting 11A (ATP11A) and ATP11C, resulting in inhibition of their flippase activity and PS exposure, as demonstrated by absent or reduced PS translocation on the cell surface of cells expressing a non-cleavable ATP11C or developing erythrocytes from Atp11a−/− mice [219,220,221]. That said, PS exposure may not universally accompany intrinsic (and extrinsic) apoptosis [222,223,224].

Of note, a large body of evidence suggests that executioner caspases precipitate intrinsic apoptosis, once a hitherto poorly defined point-of-no-return has been trespassed, but are not essential for it [17]. Accordingly, blocking post-mitochondrial caspase activation by genetic means or with specific pharmacological inhibitors, such as to N-benzyloxycarbonyl-Val-Ala-Asp(O-Me) fluoromethylketone (Z-VAD-fmk) and (3S)-5-(2,6-difluorophenoxy)-3-[[(2S)-3-methyl-1-oxo-2-[(2-quinolinylcarbonyl)amino]butyl]amino]-4-oxo-pentanoic acid hydrate (Q-VD-OPh), generally delays (but does not prevent) intrinsic apoptosis in vitro and in vivo (at least in the mammalian system), as it promotes a switch to other types of RCD [17, 225]. In addition, when MOMP affects a limited number of mitochondria, the consequent sublethal activation of caspases does not precipitate RCD but promotes genomic instability [226]. Finally, at least some cells exposed to transient apoptotic stimuli appear to survive MOMP affecting a limited number of mitochondria and the partial activation of executioner caspases by a hitherto poorly characterized process called anastasis (most likely constituting a robust adaptative response upstream of the boundary between cellular life and death) [226,227,228]. Altogether, these observations suggest that CASP3 and CASP7 mediate a facilitating, rather than indispensable, role in RCD (for an extensive discussion on this topic, please refer to ref [17]). This said, executioner caspases can positively or negatively regulate the emission of multiple DAMPs from dying cells, including immunostimulatory [229] as well as immunosuppressive [230] factors. Thus, pharmacological agents targeting executioner caspases may be unable to mediate bona fide cytoprotection, but may efficiently switch RCD modality. Interestingly, although CASP6 has long been considered as an executioner caspase based on its homology with CASP3 and CASP7, recent data on substrate specificity suggest that CASP6 may actually be involved in RCD initiation [179, 231, 232]. Additional investigation is required to elucidate the function of CASP6 in mammalian cells.

A specific variant of intrinsic apoptosis elicited by the loss of integrin-dependent attachment to the extracellular matrix is commonly known as anoikis [233, 234]. As such, anoikis is demarcated by MOMP and precipitated by the activation of executioner caspases, notably CASP3 [233]. At least in some settings, detachment from the extracellular matrix triggers MOMP upon activation of the BH3-only proteins BIM and BMF [137, 235]. Since anoikis prevents anchorage-independent proliferation and attachment to an improper matrix, it is generally considered as an oncosuppressive process [234, 236]. Accordingly, cancer cells need to acquire at least some degree of resistance to anoikis to initiate and progress though the so-called “metastatic cascade” [237,238,239]. Neoplastic cells can evade anoikis upon activation of mitogen-activated protein kinase 1 (MAPK1; best known as ERK2) caused by cellular aggregation and consequent epidermal growth factor receptor (EGFR) stabilization mediated by erb-b2 receptor tyrosine kinase 2 (ERBB2) [237, 240], or degradation of the negative ERK2 regulator BRCA1-associated protein (BRAP), which is favored by coiled-coil domain containing 178 (CCDC178) [241]. Once activated, ERK2 reportedly supports anoikis resistance by promoting the cytosolic sequestration of BIM in complex with dynein light chain LC8-type 1 (DYNLL1; best known as LC8) and beclin 1 (BECN1) [138, 238], or the transactivation of integrin subunit alpha 6 (ITGA6) via a mechanism dependent on KRAS [242].

Additional strategies that limit the sensitivity of malignant cells to anoikis encompass (but are not limited to): (1) activation of anti-apoptotic BCL2 proteins, including MCL1 stabilization as induced by fibroblast-derived insulin like growth factor-binding proteins (IGFBPs) [243] and increase in BCL2 expression levels as imposed by hepatitis B virus X protein [244]; (2) epigenetic silencing of adhesion-related genes, such as SHC adaptor protein 1 (SHC1) upon overexpression of the hematopoietic transcription factor IKAROS family zinc finger 3 (IKZF3; also known as AIOLOS) [245]; (3) perturbation of ITG-protein tyrosine kinase 2 (PTK2; best known as FAK) signaling, which usually suppresses anoikis [246,247,248,249]; (4) activation of the so-called “epithelial-to-mesenchymal transition” (EMT), which is associated with multiple signal transduction and metabolic modules for RCD resistance [242, 250, 251]; (5) targeting of Yes associated protein 1 (YAP1) by miR-200a or via a platelet-dependent mechanism [252, 253]; (6) increased antioxidant responses driven by the activating transcription factor 4 (ATF4)-mediated upregulation of heme oxygenase 1 (HMOX1) [254]; (7) autophagy activation [254, 255]; (8) upregulation of the molecular chaperone crystallin alpha B (CRYAB; also known as HSPB5) [256]; (9) signaling via AMPK and proliferation and apoptosis adaptor protein 15 (PEA15), which favors anchorage-independent cell growth [257]; (10) upregulation of matrix metallopeptidases (MMPs) by a mechanism involving the epidermal growth factor (EGF)-driven autocrine production of angiopoietin like 4 (ANGPTL4) [258]; (11) expression and phosphorylation of signal transducer and activator of transcription 3 (STAT3) [259]; and (12) rewiring of central carbon metabolism toward NAPDH synthesis, resulting in improved redox homeostasis [260, 261]. That said, it has become evident that the adaptation of cancer cells to the loss of attachment involves multiple processes beyond (but presumably highly interconnected to) anoikis resistance [234 262,263,264], suggesting that multiple barriers need to be overcome for the metastatic cascade to be initiated.

The NCCD proposes to define intrinsic apoptosis as a form of RCD initiated by perturbations of the intracellular or extracellular microenvironment, demarcated by MOMP and precipitated by executioner caspases, mainly CASP3 (Box 1).

Extrinsic apoptosis

Extrinsic apoptosis is an RCD modality initiated by perturbations of the extracellular microenvironment [265,266,267,268]. Extrinsic apoptosis is mostly driven by either of two types of plasma membrane receptors: (1) death receptors, whose activation depends on the binding of the cognate ligand(s), and (2) dependence receptors, whose activation occurs when the levels of their specific ligand drop below a specific threshold [267 269,270,271].

Death receptors include (but are not limited to): Fas cell surface death receptor (FAS; also known as CD95 or APO-1), and TNF receptor superfamily member 1A (TNFRSF1A; best known as TNFR1), 10a (TNFRSF10A; best known as TRAILR1 or DR4), and 10b (TNFRSF10B; best known as TRAILR2 or DR5) [269, 270, 272, 273]. As a general rule, death receptor ligation allows for the assembly of a dynamic multiprotein complex at the intracellular tail of the receptor, such as so-called “death-inducing signaling complex” (DISC), “complex I”, and “complex II”, which operate as molecular platforms to regulate the activation and functions of CASP8 (or CASP10, in a limited number of settings) [274,275,276]. In the case of FAS and TRAILRs, the cognate ligands—namely, FAS ligand (FASLG; also known as CD95L or APO-1L) and TNF superfamily member 10 (TNFSF10; best known as TRAIL), respectively—stabilize preformed receptor homotrimers to induce a conformational change at their intracellular tails that enables the death domain (DD)-dependent association of the adapter Fas associated via death domain (FADD) [277,278,279,280,281,282]. In turn, FADD drives DISC assembly by promoting the death effector domain (DED)-dependent recruitment of CASP8 (or CASP10) and multiple isoforms of c-FLIP. In contrast, TNFR1 signaling involves the association of TNFRSF1A associated via DD (TRADD), which acts as an adaptor for the assembly of complex I, generally consisting of TNF receptor associated factor 2 (TRAF2), TRAF5, c-IAP1, c-IAP2, RIPK1, and the linear ubiquitin chain assembly complex (LUBAC), a supramolecular entity consisting of SHANK associated RH domain interactor (SHARPIN), RANBP2-type, and C3HC4-type zinc finger containing 1 (RBCK1; best known as HOIL-1), and ring finger protein 31 (RNF31; best known as HOIP) [283,284,285,286,287]. Of note, the glycosylation state of some death receptors (e.g., FAS) has been shown to impact on the sensitivity of T lymphocytes to extrinsic apoptosis, hence influencing the termination of inflammatory responses [288,289,290]. The relevance of death receptor glycosylation for extrinsic apoptosis in other cell types has not been investigated in detail.

The molecular mechanisms regulating CASP8 activity upon death receptor stimulation have been extensively investigated. In particular, CASP8 maturation involves a cascade of events initiated by the binding of CASP8 to FADD at the DISC. This interaction enables the assembly of a linear filament of CASP8 molecules (depending on their DEDs) that facilitates homodimerization and consequent activation by autoproteolytic cleavage [291,292,293,294,295]. A key role in this setting is mediated by c-FLIP, which is a catalytically inactive close relative of CASP8 [296, 297]. Compelling evidence indicates that the short variant of c-FLIP (c-FLIPS) and its long counterpart (c-FLIPL) inhibit and activate CASP8, respectively, by modulating CASP8 oligomerization [298,299,300,301]. Active CASP8 reportedly cleaves c-FLIPL [302] and heterodimeric complexes of CASP8 with c-FLIPL (but not c-FLIPS) are endowed with limited enzymatic activity that favors CASP8 oligomerization and consequent activation [301]. c-FLIP isoforms and CASP8 seem to be recruited at the DISC to comparable levels [303], supporting the notion that elevated expression levels of c-FLIPL inhibit, rather than activate, extrinsic apoptosis possibly by disrupting CASP8 maturation [301, 304]. Of note, CFLAR (the gene encoding c-FLIP) is under direct transcriptional control by NF-κB, which largely contributes to pro-survival TNFR1 signaling in specific circumstances (see below) [287, 296, 305]. The enzymatic activity of CASP8 appears to be controlled by additional post-translational mechanisms, including (but not limited to): (1) phosphorylation at Y380, which inhibits the autoproteolytic activity of CASP8 upon FAS activation [306], (2) phosphorylation at T273, which is catalyzed by polo like kinase 3 (PLK3) at the DISC and promotes CASP8 apoptotic functions [307], and (3) deubiquitination, which decreases CASP8 activity and interrupts extrinsic apoptosis [302].

The execution of extrinsic apoptosis driven by death receptors follows two distinct pathways. In so-called “type I cells” (e.g., thymocytes and mature lymphocytes) the CASP8-dependent proteolytic maturation of executioner CASP3 and CASP7 suffices to drive RCD, which cannot be inhibited by the transgene-driven overexpression of anti-apoptotic BCL2 proteins, the co-deletion of Bax and Bak1, or the loss of BID [308, 309]. Conversely, in “type II cells” (e.g., hepatocytes, pancreatic β cells, and a majority of cancer cells), in which CASP3 and CASP7 activation is restrained by XIAP [310], extrinsic apoptosis requires the proteolytic cleavage of BID by CASP8 [70, 311, 312]. This leads to the generation of a truncated form of BID (tBID), which translocates to the OMM [313, 314] via a mechanism that, at least upon FAS stimulation, reportedly depends on the binding of modulator of apoptosis 1 (MOAP1) to the alleged BID receptor mitochondrial carrier 2 (MTCH2) [315, 316]. At the OMM, tBID operates as a BH3-only activator to engage BAX/BAK-dependent MOMP-driven and consequent CASP9-driven RCD. Although human CASP10 shares some degree of substrate specificity with CASP8 [317] and possibly contributes to extrinsic apoptosis in primary T cells [318], rodents including mice and rats lack a functional Casp10 gene, and the precise role of this caspase in death receptor-driven apoptosis in humans and other CASP10-proficient species remains a matter of controversy [319,320,321]. A recent study shows that—following FAS activation—CASP10 causes the dissociation of CASP8 from the DISC, thereby promoting cell survival [319]. FAT atypical cadherin 1 (FAT1) appears to mediate similar anti-apoptotic function by limiting the association of CASP8 with the DISC [322].

A large body of evidence demonstrates that death receptor ligation does not necessarily culminate in RCD. In particular, TNFR1 activation can have diverse outcomes depending on multiple variables, such as the post-translational modification status of RIPK1, which has a direct impact on the assembly of pro-survival vs. pro-death signaling complexes [323,324,325]. Thus, following tumor necrosis factor (TNF) stimulation, RIPK1 is recruited at complex I in a TRADD-independent manner, followed by RIPK1 polyubiquitination by c-IAP1, c-IAP2, and LUBAC [196, 324 326,327,328,329]. Polyubiquitinated RIPK1 promotes cell survival and inflammation by acting as a scaffold for the sequential recruitment of TGF-beta activated kinase 1/MAP3K7-binding protein 2 (TAB2), TAB3, and mitogen-activated protein kinase kinase kinase 7 (MAP3K7; best known as TAK1), which can drive mitogen-activated protein kinase (MAPK) signaling or IκB kinase (IKK)-dependent NF-κB activation [283, 287 330,331,332,333]. Moreover, the phosphorylation of RIPK1 by TAK1, the IKK complex or mitogen-activated protein kinase-activated protein kinase 2 (MAPKAPK2; best known as MK2) appears to alter its ability to interact with FADD and CASP8, hence preventing the variants of TNF-induced RCD that depends on RIPK1 kinase activity and favoring RIPK1-independent TRADD-, FADD-, and CASP8-driven apoptosis [285 334,335,336]. Conversely, in the presence of so-called “SMAC mimetics” (which de facto operate as IAP inhibitors) [337], RIPK1 is deubiquitinated by CYLD lysine 63 deubiquitinase (CYLD), favoring its release from complex I and its association with FADD and CASP8 in the cytosol to form complex II, which drives extrinsic apoptosis [338]. Complex II formation also requires TRAF2 ubiquitination by HECT domain and ankyrin repeat containing E3 ubiquitin protein ligase 1 (HACE1) [339]. To add a further layer of complexity, the proteasomal degradation of TRAF2 appears to be prevented (at least in hepatocytes) by RIPK1, independently of its kinase activity [340, 341]. Of note, TNFR1 can also activate alternative RCD modalities, such as necroptosis (see below).

Death receptor signaling can also lead to NF-κB activation, generally resulting in cell survival associated with a robust inflammatory response [272, 342]. The ability of some death receptors including TNFR1 to promote NF-κB activation over CASP8 activation appears to depend on the degree of receptor oligomerization (i.e., trimerization vs. higher-order multimerization) [343], the scaffolding (i.e., non-enzymatic) functions of CASP8, and the consequent assembly of TNFR1-like complexes containing RIPK1 and LUBAC [272, 286, 344]. Upon TRAILR activation, LUBAC reportedly ubiquitinates both CASP8 and RIPK1 while promoting the recruitment of IKK to complex I [286], which also explains the requirement of LUBAC for the inhibition of TNF-induced cell death [345]. In line with this notion, TNF alpha-induced protein 3 (TNFAIP3; best known as A20) inhibits CASP8 activation downstream of TRAILRs in glioblastoma cells, owing to its ability to polyubiquitinate RIPK1 [346, 347]. A recent study suggests that the ability of TRAILR2 to dispatch pro-survival rather than pro-apoptotic signals may depend on its preferential localization outside of lipid rafts [348]. It remains to be demonstrated whether the same also applies to other death receptors.

The family of dependence receptors consists of approximately 20 members, including: (1) the netrin 1 (NTN1) receptors DCC netrin 1 receptor (DCC), unc-5 netrin receptor A (UNC5A), UNC5B, UNC5C, and UNC5D; (2) the neurotrophin receptor neurotrophic receptor tyrosine kinase 3 (NTRK3); and (3) the sonic hedgehog (SHH) receptor patched 1 (PTCH1) [267, 349, 350]. Intriguingly, dependence receptors promote cell survival, proliferation and differentiation in physiological conditions (when their cognate ligands are normally available), but activate distinct (and not completely elucidated) lethal signaling cascades (generally impinging on caspase activation) once ligand availability falls below a specific threshold level [350]. Thus, in the absence of their respective ligands: (1) DCC is cleaved by CASP3 and this promotes its association with adaptor protein, phosphotyrosine interacting with PH domain and leucine zipper 1 (APPL1) and CASP9, resulting in the activation of the CASP9-CASP3 cascade [350, 351]; (2) PTCH1 interacts with the cytosolic adaptor four and a half LIM domains 2 (FHL2; best known as DRAL), hence favoring the assembly of a CASP9-activating complex consisting of caspase recruitment domain family member 8 (CARD8; also known as TUCAN) and neural precursor cell expressed, developmentally down-regulated 4, E3 ubiquitin protein ligase (NEDD4) [352,353,354]; (3) UNC5B enables the protein phosphatase 2 (PP2A)-mediated activating dephosphorylation of death associated protein kinase 1 (DAPK1), which is known to promote p53-dependent RCD [355,356,357]; and (4) UNC5D and NTRK3 are subjected to CASP3 cleavage generating intracellular fragments that translocate either into the nucleus to trigger the E2F transcription factor 1 (E2F1)-driven expression of pro-apoptotic genes (as in the case of UNC5D) or at mitochondria to activate CASP9 upon MOMP (as in the case of NTRK3) [358, 359].

Dependence receptor-driven RCD has been involved in multiple pathophysiological settings, and exerts robust oncosuppressive functions [350]. Accordingly, neoplastic cells often escape from dependence receptor-mediated RCD by (1) upregulating the expression of their cognate ligands such as NTN1 [360,361,362]; (2) inactivating, downregulating, or losing gene(s) encoding specific dependence receptors, including DCC, UNC5C, and NTRK3 [350 363,364,365,366,367,368,369]; or (3) silencing signal transducers operating downstream of dependence receptors—such as DAPK1—via epigenetic mechanisms [370]. That said, whether the actual pathophysiological relevance of dependence receptor signaling stems from the initiation of extrinsic apoptosis remains to be formally established. Of note, in specific cell types, some members of the toll-like receptor (TLR) protein family including toll like receptor 3 (TLR3) have also been suggested to trigger RCD by a mechanism that involves toll like receptor adaptor molecule 1 (TICAM1; best known as TRIF), and ultimately impinges on CASP8 activation [371, 372]. However, it remains unclear whether TLR3 and other TLRs actually initiate a private RCD program that directly engages CASP8, or whether they promote RCD upon the activation of an NF-κB-dependent autocrine/paracrine signaling pathway involving TNF.

We propose to define extrinsic apoptosis as a type of RCD initiated by perturbations of the extracellular microenvironment that are detected by plasma membrane receptors, propagated by CASP8 (with the optional involvement of MOMP), and precipitated by executioner caspases, mainly CASP3 (Box 1).

MPT-driven necrosis

MPT-driven necrosis is a form of RCD initiated by specific perturbations of the intracellular microenvironment such as severe oxidative stress and cytosolic Ca2+ overload, which generally manifests with a necrotic morphotype [373, 374]. The term MPT refers to an abrupt loss of the impermeability of the IMM to small solutes, resulting in rapid Δψm dissipation, osmotic breakdown of both mitochondrial membranes, and RCD [373, 374].

At the biochemical level, MPT-driven necrosis has been proposed to follow the opening of the so-called “permeability transition pore complex” (PTPC), a supramolecular complex assembled at the junctions between the IMM and OMM [103, 374]. The composition, regulation, and precise mechanism of action of the PTPC are still under intense investigation and matter of a vivid debate [373, 375]. To date, peptidylprolyl isomerase F (PPIF; best known as cyclophilin D, CYPD) is the only protein whose in vivo requirement for MPT induction has been formally validated with robust genetic tools (although there is consensus around the notion that CYPD does not constitute the pore-forming unit of the PTPC) [376,377,378,379]. Accordingly, pharmacological inhibitors of CYPD including cyclosporin A (CsA) [376 379,380,381], sanglifehrin A (SfA) [382, 383], and JW47 [384] limit MPT-driven necrosis and confer protection in multiple rodent models of disease in which oxidative stress and cytosolic Ca2+ overload constitute major etiological determinants (e.g., neuronal, cardiac, and renal ischemia/reperfusion). Along similar lines, CYPD degradation through a mechanism initiated by the overexpression of HCLS1 associated protein X-1 (HAX1) abolishes MPT-driven necrosis and limits the demise of cardiomyocytes experiencing ischemia/reperfusion in vivo [385]. Nonetheless, a large randomized clinical study completed in 2015 (the CIRCUS trial) failed to confirm previous findings from 2008 [386] on the cardioprotective effects of cyclosporine administered before percutaneous coronary intervention to patients with acute myocardial infarction [387]. Although multiple caveats linked to the methods employed to measure infarct size and the use of a specific pharmacological CsA formulation can be invoked to explain the negative results of the CIRCUS trial [388], the elevated interconnectivity of RCD subroutines (notably, intrinsic apoptosis and MPT-driven necrosis) may have played a key role in this setting.

At odds with CYPD, several other proteins that had previously been hypothesized to mediate a non-redundant role within the PTPC turned out to be dispensable for MPT in vivo, based on relatively robust genetic models [373]. Thus, an inducible cardiomyocyte-specific deletion of solute carrier family 25 member 3 (Slc25a3, which codes for the inorganic phosphate carrier) in mice does not affect the ability of mitochondria to undergo MPT in vitro, as it establishes partial PTPC desensitization in cellula and slightly mitigates cardiac injury upon ischemia/reperfusion in vivo (~10% reduction in ischemic area over area at risk) [389]. Similar findings have been obtained for distinct isoforms of the IMM integral protein adenine nucleotide translocator (ANT) and the OMM protein VDAC. In particular, the concurrent knockout or knockdown of Slc25a4 and Slc25a5, which encode ANT1 and ANT2, respectively [390], or that of Vdac1, Vdac2, and Vdac3 [391, 392] fails to prevent the induction of MPT by oxidative stress or Ca2+ overload. However, mitochondria isolated from Slc25a4−/−Slc25a5−/− mouse livers are desensitized to Ca2+-driven MPT to a similar extent than mitochondria exposed to CsA [390]. Moreover, Slc25a31 encodes another ANT isoform (i.e., ANT4), that (at least in some mouse tissues) may compensate for the absence of ANT1 and ANT2 [393, 394]. These results reflect a consistent degree of genetic and functional redundancy among the components of the molecular machinery for MPT [373].

Several lines of evidence suggest that the mitochondrial F1FO ATPase mediates a non-redundant role within the PTPC. Initially, the c-ring of the F1FO ATPase [395,396,397,398] as well as F1FO ATPase dimers [399] have been proposed to constitute the long-sought PTPC pore-forming unit. A specific interaction between CYPD and the lateral stalk of the F1FO ATPase, as well as the ability of Ca2+ ions (which are potent MPT inducers) to bind to ATP synthase, H+ transporting, mitochondrial F1 complex, beta polypeptide (ATP5B) [100], lend further support to this interpretation [395, 400, 401]. However, very recent findings seem to exclude the possibility that the F1FO ATPase constitutes the pore-forming component of the PTPC [402,403,404,405]. First, it seems unlikely for c-rings (which exist as pores across the IMM) to lose their lipid plugs in relatively physiological conditions [402]. Second, mitochondria from human cells lacking all the genes coding for the c subunit of the F1FO ATP synthase, i.e., ATP synthase, H+ transporting, mitochondrial Fo complex subunit C1 (subunit 9; ATP5G1), subunit C2 (subunit 9; ATP5G2), and subunit C3 (subunit 9; ATP5G3), reportedly retain the ability to undergo MPT in response to Ca2+ overload [403]. Finally, cells lacking ATP synthase, H+ transporting, mitochondrial F1 complex, O subunit (ATP5O; best known as OSCP), or the membrane domain of the b subunit of the F1FO ATP synthase (encoded by ATP5F1) appear to preserve normal PTPC activity [405]. That said, the implication of the F1FO ATPase or components thereof in MPT-driven necrosis remains a matter of intensive investigation. An RNA interference (RNAi)-based screening identified SPG7, paraplegin matrix AAA peptidase subunit (SPG7) as an essential component of the PTPC acting as part of VDAC-containing and CYPD-containing hetero-oligomers [406]. Despite the availability of Spg7−/− mice, the actual involvement of SPG7 in MPT-derived necrosis in vivo remains to be validated.

Several physical or functional PTPC interactors have been shown to regulate MPT-driven necrosis. These include: (1) pro- and anti-apoptotic BCL2 family members such as BAX, BAK, and BID [407,408,409,410], as well as BCL2 and BCL-XL [411,412,413,414]; (2) DRP1, which appears to promote PTPC opening in response to chronic β adrenergic receptor stimulation, via a mechanism that relies on DRP1 phosphorylation by calcium/calmodulin dependent protein kinase II (CAMK2G; best known as CaMKII) [415]; and (3) p53, which participates in MPT-driven necrosis upon physical interaction with CYPD [416]. The latter interaction has been shown to participate in the pathogenesis of ischemic stroke in mice [416]. Its pathophysiological relevance in humans, however, remains to be elucidated. Recent findings lend additional support to the relevance of tight Ca2+ homeostasis at the mitochondrial level for cellular and organismal fitness. Thus, perturbing the activity of the IMM Ca2+ uniporter, consisting of mitochondrial calcium uniporter (MCU), single-pass membrane protein with aspartate-rich tail 1 (SMDT1; also known as EMRE), mitochondrial calcium uptake 1 (MICU1) and MICU2, reportedly affects mouse survival and liver regeneration after partial hepatectomy by promoting mitochondrial Ca2+ overload and MPT-driven necrosis [417]. Along similar lines, the loss of mitochondrial m-AAA proteases of the IMM, which regulate the assembly of the IMM Ca2+ uniporter, induces mitochondrial Ca2+ overload, PTPC opening, and neuronal cell death [418]. Adult mice subjected to the cardiomyocyte-specific deletion of Mcu are protected against cardiac ischemia/reperfusion as a consequence of MTP inhibition [419]. Moreover, the inducible cardiomyocyte-specific deletion of solute carrier family 8 member B1 (Slc8b1, which encodes a mitochondrial potassium-dependent sodium/calcium exchanger) in mice reportedly provokes sudden lethality owing to heart failure imposed by MTP-regulated necrosis upon mitochondrial Ca2+ overload [420]. Finally, rap guanine nucleotide exchange factor 3 (RAPGEF3; best known as EPAC1) appears to trigger PTPC opening by increasing mitochondrial Ca2+ levels through interaction with VDAC1, heat shock protein family A (Hsp70) member 9 (HSPA9; best known as GRP75), and inositol 1,4,5-trisphosphate receptor type 1 (ITPR1; best known as IP3R1), and the knockout of Rapgef3 protects mice against myocardial ischemia/reperfusion injury [421]. However, EPAC1 activation with bicarbonate reportedly decreases mitochondrial Ca2+ uptake, stimulates ATP production, and inhibits multiple forms of RCD including MPT-driven necrosis in rat cardiomyocytes [422]. The precise reasons underlying this apparent discrepancy remain to be elucidated.

We propose to define MPT-driven necrosis as a form of RCD triggered by perturbations of the intracellular microenvironment and relying on CYPD (Box 1).


Necroptosis is a form of RCD initiated by perturbations of the extracellular or intracellular microenvironment detected by specific death receptors, including (but not limited to) FAS and TNFR1 [423,424,425,426,427], or pathogen recognition receptors (PRRs), including TLR3, TLR4, and Z-DNA binding protein 1 (ZBP1; also known as DAI) [428,429,430]. It is now clear that necroptosis (which generally manifests with a necrotic morphotype) not only mediates adaptative functions upon failing responses to stress, but also participates in developmental safeguard programs (to ensure the elimination of potentially defective organisms before parturition), as well as in the maintenance of adult T-cell homeostasis (de facto serving as a PCD subroutine, at least in specific settings) [2 431,432,433].

At the molecular level, necroptosis critically depends on the sequential activation of RIPK3 and mixed lineage kinase domain like pseudokinase (MLKL) [434, 435]. Upon necroptosis initiation by TNFR1, RIPK3 is activated by RIPK1 (provided that CASP8 is inactive, see below) through a mechanism involving the physical interaction between their respective RIP homotypic interaction motif (RHIM) domains and RIPK1 catalytic activity [436,437,438]. Accordingly, chemical inhibitors of RIPK1 including necrostatin-1 (Nec-1) and derivatives (e.g., Nec-1s) robustly inhibit TNFR1-driven necroptosis, in vitro and in vivo [425, 427]. Alternatively, RIPK3 can be activated following the RHIM-dependent interaction with (1) TRIF upon either TLR3 activation by double-stranded RNA (dsRNA) within endosomes, or TLR4 activation by lipopolysaccharide (LPS) or various DAMPs at the plasma membrane [428]; or (2) ZBP1, which operates as a sensor for cytosolic DNA-promoting type I interferon (IFN) synthesis and NF-κB activation [439,440,441]. Active RIPK3 catalyzes the phosphorylation of MLKL, resulting in the formation of MLKL oligomers (most likely trimers or tetramers) that translocate to the plasma membrane, where they bind specific phosphatidylinositol phosphate species by a roll-over mechanism and hence trigger plasma membrane permeabilization [435 442,443,444,445,446,447,448,449,450,451,452,453].

Although the essential contribution of MLKL to necroptosis has been confirmed by genetic studies [435] as well as by pharmacological (i.e., inhibition of MLKL with necrosulfonamide, NSA) interventions [442], the precise mechanism through which MLKL executes necroptosis is not completely understood. Recent studies ascribe to the heat shock protein 90 kDa alpha family class A member 1 (HSP90AA1; best known as HSP90) a specific and non-redundant role in MLKL oligomerization and translocation [454, 455]. Moreover, it has also been reported that MLKL oligomerization promotes a cascade of intracellular events involving (1) Ca2+ influx, which is presumably mediated by the MLKL target transient receptor potential cation channel subfamily M member 7 (TRPM7) [449]; and (2) PS exposure, which seems to be directly operated by MLKL [456]. This is followed by the formation of PS-exposing plasma membrane bubbles whose breakdown and release is negatively regulated—in conditions of limited MLKL activation—by the antagonistic activity of the endosomal sorting complex required for transport (ESCRT)-III machinery [456, 457]. Once localized at the plasma membrane, MLKL reportedly activates cell-surface proteases of the ADAM family, which can promote the shedding of plasma membrane-associated proteins [458], or form Mg2+ permeant channels [459]. Of note, active MLKL also appears to translocate to the nucleus, but the relevance of this phenomenon for necroptosis remains to be investigated [460]. Previous data supporting the involvement of PGAM family member 5, serine/threonine protein phosphatase, mitochondrial (PGAM5)- and DRP1-driven mitochondrial fragmentation in necroptosis [461] have been conclusively invalidated [435, 446 462,463,464,465,466], confirming that necroptotic signaling can proceed normally independent of mitochondria. Of note, the core components of necroptosis are poorly conserved across the animal kingdom, as some species lack RIPK3 and/or MLKL [467]. Moreover, a few non-canonical instances of pseudonecroptotic RCD involving MLKL (but not RIPK3) [468] or RIPK3 (but not MLKL) [469] have been described. These observations reinforce the notion that the signaling pathways leading to RCD display a hitherto incompletely understood degree of interconnectivity.

Death receptor (in particular TNFR1) engagement is the trigger for RIPK3 activation best characterized so far. As mentioned above, the biological outcome of TNFR1 signaling spans from cell survival and activation (i.e., cytokine secretion) to multiple subroutines of RCD, depending on a variety of cell-intrinsic (e.g., expression levels of the proteins involved in the process) and cell-extrinsic (e.g., intensity and duration of TNF stimulation) factors [283]. In particular, the activation of RIPK3 downstream of TNFR1 ligation relies on the formation of a RIPK1-containing and RIPK3-containing amyloid-like signaling complex commonly known as necrosome [436, 470], wherein first RIPK1 and then RIPK3 undergo a series of trans-phosphorylation or auto-phosphorylation events that are required for MLKL recruitment and necroptosis activation [437, 438, 442, 471]. Major negative regulators of the necrosome include: (1) STIP1 homology and U-box containing protein 1 (STUB1; also known as CHIP), which promotes RIPK1 and RIPK3 ubiquitination followed by lysosomal degradation [472, 473]; (2) A20, which inhibits necrosome assembly by deubiquitinating RIPK3 [473, 474]; (3) protein phosphatase, Mg2+/Mn2+ dependent 1B (PPM1B), which prevents MLKL recruitment to the necrosome by dephosphorylating RIPK3 [475]; and (4) aurora kinase A (AURKA), which mediates inhibitory function upon physical interaction with RIPK1 and RIPK3 [476]. RIPK3 activation also depends on its physical association with a HSP90-containing and cell division cycle 37 (CDC37)-containing co-chaperone complex [477]. In addition, the assembly of the necrosome upon TNFR1 stimulation impinges on two conditions: (1) pharmacological or genetic CASP8 inactivation [478, 479], and (2) RIPK1 deubiquitination-dependent phosphorylation (at least in some settings), which can be favored by exogenously provided SMAC mimetics, ensuring the release of RIPK1 from complex I (see above) [334, 335, 480, 481].

As for the first condition, compelling experimental findings demonstrate that the concerted activity of CASP8, FADD, and c-FLIPL tonically inhibits necroptosis [432, 466, 478, 479 482,483,484]. Thus, the embryonic lethality imposed on mice by the loss of Casp8 or Fadd can be rescued by concurrent ablation of Ripk3 or Mlkl, even though these double knockout animals generally display lymphoproliferative and/or systemic autoimmune disorders as adults [432, 466, 484, 485]. Of note, Cflar−/− mice require the concomitant knockout of Ripk3 and Fadd to develop into adulthood, which underscores the inhibitory role of c-FLIP in both necroptosis and extrinsic apoptosis reported above [483, 486]. Along similar lines, the concurrent deletion of Ripk3 averts perturbations of cutaneous and intestinal homeostasis imposed by the tissue-specific ablation of Fadd or Casp8 [483, 487, 488]. Moreover, the proliferative defects of Casp8−/− or Fadd−/− T cells can be rescued by the administration of the RIPK1 inhibitor Nec-1 or the concomitant ablation of Ripk3 [489]. Necroptosis is also tonically inhibited by c-IAPs, owing to their ability to ubiquitinate RIPK1 [490,491,492,493]. Accordingly, necroptosis relies on the deubiquitinating activity of CYLD [338], which is also a proteolytic target of CASP8 [494,495,496]. Finally, some components of the TNFR1 signaling cascade reportedly regulate necroptosis either in a negative manner, by catalyzing the inhibitory phosphorylation of RIPK1 (e.g., the IKK complex and MK2) [335, 336] and constitutively interacting with (and thus preventing the activation of) MLKL (e.g., TRAF2) [497], or in a positive manner, by favoring the activating phosphorylation of RIPK1 or RIPK3 upon prolonged activation (e.g., TAK1) [334, 498]. In this context, CYLD also contributes to necroptosis by deubiquitinating—and hence suppressing the anti-necroptotic activity of—TRAF2 [497].

That said, mounting evidence indicates that necroptosis driven by several stimuli—in some circumstances even TNFR1 activation—does not necessarily rely on RIPK1. Thus, in contrast to Ripk3−/− mice that are viable and fertile, the Ripk1−/− genotype causes perinatal lethality [482], which cannot be prevented by the ablation of Ripk3, Casp8, or Fadd alone, but can be rescued by the co-deletion of Ripk3 and Casp8, Fadd or Tnfrsf1a [482 499,500,501]. Moreover, Ripk1−/− cells display increased sensitivity to necroptosis and/or extrinsic apoptosis induced by a set of innate immune stimuli [499]. Conditional knockout mouse models demonstrate the key role of RIPK1 for the preservation of intestinal and cutaneous homeostasis and survival [502, 503]. In particular, mice lacking Ripk1 in intestinal epithelial cells display increased rates of spontaneous CASP8-driven apoptosis and develop severe inflammatory lesions leading to premature death, a detrimental phenotype that can be prevented by co-deleting Fadd or (to a lesser degree) Tnfrsf1a [502]. Likewise, the absence of Ripk1 from keratinocytes promotes spontaneous necroptosis and consequent cutaneous inflammation, which can be prevented by the co-deletion of Ripk3, Mlkl, or Zbp1 but not Fadd [440, 502]. Collectively, these results suggest that (at least in some settings) RIPK1 can inhibit (rather than activate) RIPK3-dependent necroptosis and/or CASP8-dependent extrinsic apoptosis [504]. At least in some settings, this reflects the major role of RIPK1 in NF-κB activation [505,506,507].

Intriguingly, the pro-survival role of RIPK1 in development seems to be independent of both its kinase activity and RIPK3 binding, as demonstrated by the fact that mice genetically engineered to express a kinase-dead variant of RIPK1 (e.g., RIPK1K45A) are viable and fertile [447, 499, 508]. Moreover, it has recently been reported that the autophagic receptor optineurin (OPTN) [509] actively regulates the proteasomal turnover of RIPK1, as the loss of OPTN induces axonal degeneration via RIPK1-dependent necroptosis [510]. Inhibitor of nuclear factor kappa B kinase subunit gamma (IKBKG; best known as NEMO) also prevents RIPK1-driven intestinal inflammation and epithelial cell death, although the underlying molecular mechanisms remain poorly understood [511] Finally, when catalytically inactive or inhibited by specific pharmacological agents such as Nec-1, RIPK1 (and, at least under certain circumstances, RIPK3) reportedly contributes to specific forms of CASP8-dependent apoptosis (see above) [335, 336, 446, 447, 481 512,513,514,515,516]. The current view ascribes the opposing roles of RIPK1 (and—at least in part—RIPK3) in promoting or inhibiting RCD to its kinase-dependent vs. kinase-independent (i.e., scaffolding) functions, respectively [4, 517].

As mentioned above, RIPK3 can be activated by proteins involved in innate immunity to invading pathogens including TRIF and ZBP1 [428, 439]. Thus, in the absence of CASP8 activity, stimulation of TLR3 or TLR4 by their respective ligands promotes necroptosis upon the interaction between TRIF and RIPK3 and the consequent activation of MLKL [428]. Accordingly, the synthetic TLR3 ligand polyinosinic-polycytidylic acid (polyI:C) or the co-administration of low-dose LPS and the caspase inhibitor Z-VAD-fmk trigger necroptosis in dendritic cells (DCs) [518] or microglial cells [519], respectively. In this context, IFN alpha and beta receptor subunit 1 (IFNAR1) and IFN gamma receptor 1 (IFNGR1) also appear to have pro-necroptotic functions [520,521,522,523]. Thus, Ifnar1−/− macrophages are resistant to RCD induced by LPS or polyI:C in the context of caspase inhibition, which would otherwise trigger a necroptotic process relying on TRIF and tonic IFN-stimulated gene factor 3 (ISGF3) signaling [523]. Genetic studies demonstrate that the lethality imposed to mice by the Ripk1−/−Tnfrsf1a−/− genotype is delayed (but not prevented) by the co-deletion of Ticam1 or Ifnar1 [482]. Moreover, Ripk1−/− cells are more sensitive to necroptosis induced by polyI:C or type I IFN [482]. However, Tnfrsf1a−/−Ripk1−/−Ripk3−/− mice develop into adulthood, suggesting the existence of additional RIPK3 activators [482].

Recently, the mechanism underlying ZBP1-mediated necroptosis and its regulation by RIPK1 has been elucidated [440, 441]. ZBP1 acts at the initial steps of necroptosis by mediating the sequential activation of RIPK3 and MLKL. Moreover, mice expressing a variant of RIPK1 mutated in the RHIM domain die perinatally, a phenotype that can be rescued by concurrent Ripk3, Mlkl, or Zbp1 (but not Ticam1) deletion, as well as by the knock-in of catalytically inactive RIPK3 or RIPK3 mutated in the RHIM domain [440, 441]. This suggests that the RHIM of RIPK1 acts as an inhibitor of ZBP1-driven necroptosis, most likely because it prevents the interaction between ZBP1 and RIPK3. Further investigation is required to clarify the mechanisms of ZBP1 activation in this context and its relevance for development and homeostatic tissue regulation. Importantly, multiple components of the molecular machinery for necroptosis—including ZBP1, RIPK3, MLKL, and TNFR1 (mainly via NF-κB)—impinge on the control of the so-called “inflammasome”, a supramolecular platform for the activation of CASP1 and consequent secretion of mature interleukin 1 beta (IL1B; best known as IL-1β) and IL18 [524,525,526,527,528,529]. Discussing in detail these links—which exemplify the complex interconnection between RCD signaling and inflammatory responses—goes beyond the scope of this review [4, 8 530,531,532].

In summary, we propose to define necroptosis as a type of RCD triggered by perturbations of extracellular or intracellular homeostasis that critically depends on MLKL, RIPK3, and (at least in some settings) on the kinase activity of RIPK1 (Box 1).


Ferroptosis is a form of RCD initiated by specific perturbations of the intracellular microenvironment, notably severe lipid peroxidation, which relies on ROS generation and iron availability [533,534,535,536]. The molecular mechanisms precipitating ferroptosis have begun to emerge [534], and (so far) ferroptotic RCD has been linked to toxic lipid peroxide accrual [537, 538]. Ferroptosis occurs independently of caspases, necrosome components and CYPD, and the molecular machinery for autophagy [539], manifests with a necrotic morphotype (with a predominance of mitochondrial alterations encompassing shrinkage, an electron-dense ultrastructure, reduced/disappeared cristae, and ruptured OMM) [374], and is potentially associated with a consistent release of immunostimulatory DAMPs [540, 541]. Interestingly, BCL2 has been suggested to limit the physiological demise of neuron progenitors failing to differentiate via a mechanism that (1) does not depend on BAX and caspases, and (2) can be suppressed by ferroptosis inhibitors [542]. The actual implication of BCL2 in the regulation of ferroptosis, however, remains to be firmly established.

Some of the molecular circuitries regulating the initial steps of ferroptosis have been recently unveiled by employing (1) specific ferroptosis-inducing agents, including erastin [543, 544], RSL3 [543, 544], and FIN56 [545]; and (2) specific ferroptosis-inhibiting agents, including ferrostatins [539, 546] and liproxstatins [547]. In particular, the reduced glutathione (GSH)-dependent enzyme glutathione peroxidase 4 (GPX4)—which is directly targeted by RSL3—has emerged as the main endogenous inhibitor of ferroptosis by virtue of its ability to limit lipid peroxidation by catalyzing the GSH-dependent reduction of lipid hydroperoxides to lipid alcohols [547,548,549,550]. In line with this notion, erastin triggers ferroptosis by (indirectly) affecting the catalytic cycle of GPX4 via a mechanism that involves the inhibition of the cystine/glutamate antiporter system xc and consequent decrease in intracellular cysteine (which derives from cystine reduction in the cytoplasm) and GSH (which is synthesized from cysteine) [539, 548, 549, 551]. Accordingly, depleting GSH with L-buthionine sulfoximine (BSO)—an inhibitor of the glutamate–cysteine ligase complex—can induce ferroptotic RCD (at least in some cases) [547]. Moreover, the toxicity of high extracellular glutamate may depend (at least in part) on the activation of ferroptosis through cysteine imbalance [534, 538, 552]. Of relevance for cancer therapy, the pronounced addiction of triple-negative breast carcinoma to glutamine relates (at least in part) to its ability to drive cystine uptake via xc, implying that xc may constitute a therapeutic target in this setting [553, 554]. Moreover, the FDA-approved tyrosine kinase inhibitor sorafenib can trigger ferroptosis in distinct cellular models by depleting GSH upon system xc inhibition [551 555,556,557], while altretamine (an FDA-approved alkylating agent) has been recently identified as a potential inhibitor of GPX4 by a regulatory network genome-wide system strategy [558]. Thus, the antineoplastic effects of sorafenib and altretamine may partially stem from the activation of ferroptosis. Notably, the demise of neurons caused by inhibition of xc was initially referred to as oxytosis, oxidative glutamate toxicity, or excitotoxicity, and was linked to alterations in intracellular Ca2+ homeostasis [559,560,561]. It remains unclear to which extent oxytosis can be mechanistically discriminated from ferroptosis and MPT-driven necrosis in diverse cellular contexts.

Recent evidence indicates that ferroptosis involves the preferential oxidation of specific phosphatidylethanolamine-containing polyunsaturated fatty acids (PUFAs) such as arachidonic and adrenic acid [562]. In line with a critical requirement for oxidizable PUFAs, genetic and/or pharmacological inhibition of acyl-CoA synthetase long chain family member 4 (ACSL4) and lysophosphatidylcholine acyltransferase 3 (LPCAT3), both of which are involved in the incorporation of long PUFAs into cellular membranes, protects cells against ferroptosis (at least in some settings) [562,563,564,565]. Lipid hydroperoxides can be formed by autoxidation or via enzymatic reactions catalyzed by lipoxygenases (LOXs) or cyclooxygenases (COXs). In the context of ferroptosis, PUFA peroxidation seems to be mainly regulated by the mutually antagonistic activity of LOXs (which directly catalyze lipid peroxidation) and GPX4 (which indirectly inhibits it) [550, 566]. Although arachidonate 15-lipoxygenase (ALOX15) was initially thought to play a major role in this setting, the deletion of Alox15 fails to rescue the renal phenotype imposed by the Gpx4−/− genotype (see below) [547], suggesting that multiple LOXs are involved in PUFA peroxidation and consequent ferroptosis in some mouse tissues. Accordingly, oxidized PUFAs accumulate upon GPX4 inactivation and this can result in PUFA fragmentation and ferroptosis [539, 547]. This lethal cascade can be prevented by antioxidant agents such as ferrostatin-1 (Fer-1), liproxstatin-1 (Lip-1) as well as by vitamin E, coenzyme Q10 and their analogs, all of which efficiently limit lipid peroxidation by operating as ROS scavengers [539, 547, 550, 562 567,568,569]. Of note, the catalytic sites of LOXs contain di-iron centers [570]. This may explain: (1) the ferroptosis-inhibiting effect of iron depletion by either chelators [539, 543, 548] or phosphorylase kinase catalytic subunit gamma 2 (PHKG2) knockdown [566], and (2) the ferroptosis-promoting effect of increased intracellular iron availability consequent to import by the circulating iron carrier transferrin (TF) [571, 572], degradation of ferritin (a cellular iron storage complex) by a specific autophagic mechanism known as ferritinophagy [573, 574], disruption of iron homeostasis induced by nanoparticles [541], or administration of a bioavailable iron form [575]. Alternatively, the critical dependency of ferroptosis on iron can also be ascribed to the ability of this heavy metal to promote non-enzymatic lipid oxidation via lysosomal Fenton reactions [538, 572, 576, 577].

Additional ferroptosis regulators described so far include: (1) the mevalonate pathway component farnesyl-diphosphate farnesyltransferase 1 (FDFT1; best known as SQS) [545]; (2) the transsulfuration pathway enzyme cysteinyl-tRNA synthetase (CARS) [578]; (3) heat shock protein family B (small) member 1 (HSPB1; best known as HSP27) [579] and heat shock protein family A (Hsp70) member 5 (HSPA5) [580]; (4) glutaminolysis [571]; (5) components of the MAPK signaling pathway [539, 581]; (6) the nuclear factor, erythroid 2 like 2 (NFE2L2; best known NRF2) signaling pathway [582]; (7) metallothionein 1G (MT1G) [583]; (8) dipeptidyl peptidase 4 (DPP4) [584]; (9) Fanconi anemia complementation group D2 (FANCD2) [585]; and (10) CDGSH iron sulfur domain 1 (CISD1; also known as mitoNEET) [586]. Elucidating the precise role of these proteins or signaling pathways in ferroptosis requires further investigation.

Accumulating evidence demonstrates that the pro-survival functions of GPX4 contribute to development and homeostatic tissue maintenance. Gpx4−/− mice display embryonic lethality with complete penetrance [547, 550, 587, 588]. Moreover, the inducible or tissue-specific ablation of Gpx4 in mice provokes a variety of pathological conditions, including acute renal or hepatic injury [547, 563, 589], neurodegeneration [550, 590, 591], and defective immunity to infection [567], all of which can be prevented or mitigated by ferroptosis-inhibiting strategies. A similar protective effect is observed in GPX4-independent models of renal ischemic or toxic injury [540, 592], Parkinson disease [593], and other human pathologies [594]. Moreover, ferroptosis appears to operate as a bona fide oncosuppressive mechanism [548 595,596,597,598]. It has been proposed—but remains to be formally established—that part of the multipronged oncosuppressor functions of p53 may derive from the transcriptional downregulation of components of system xc, which would impinge on specific post-translational modifications of p53 [596, 598]. Accordingly, the ability of ATF4 to upregulate system xc and stabilize GPX4 (upon HSPA5 transactivation) is causally involved in some models of oncogenesis and chemoresistance to ferroptosis induction [580,