Expansion and evolution of cell death programmes

Abstract Cell death has historically been subdivided into regulated and unregulated mechanisms. Apoptosis, a form of regulated cell death, reflects a cells decision to die in response to cues and is executed by intrinsic cellular machinery. Unregulated cell death (often called necrosis) is caused by overwhelming stress that is incompatible with cell survival. Emerging evidence, however, suggests that these two processes do not adequately explain the various cell death mechanisms. Recent data point to the existence of multiple nonapoptotic, regulated cell death mechanisms, some of which overlap or are mutually exclusive with apoptosis. Here we examine how and why these different cell death programmes have evolved, with an eye towards new cytoprotective therapeutic opportunities Cell death is a fundamental cellular response that has a crucial role in shaping our bodies during development and in regulating tissue homeostasis by eliminating unwanted cells. The first form of regulated or programmed cell death (PCD) to be characterized was apoptosis, which was described in Caenorhabitis elegans in the early 1990s. Subsequent genetic analysis of mammalian apoptosis presented a more complex picture, in which individual apoptosis genes from C. elegans have expanded into large multi-protein families (FIG. 1). These findings suggest that redundancy, functional specialization and compensatory regulation of mammalian apoptotic signalling and execution might be important features of mammalian apoptosis. Because of its conserved and uniform nature, apoptosis is frequently defined mechanistically as a pathway of regulated cell death that involves the sequential activation of caspases, a family of Cys proteases, and that is controlled both positively and negatively by B-cell lymphoma protein-2 (BCL2) family members. Assays have now been developed for multiple steps of the pathway, allowing the characterization of apoptotic death in vitro and in vivo. Apoptotic cell death is characterized by distinctive morphological features, including nuclear fragmentation, membrane blebbing and the formation of apoptotic bodies (see Ref. 2 for further details), that can be used to identify apoptotic cell death events. Genetic studies have shown that apoptosis has a significant role during normal mammalian development, especially in the central nervous system where genetic deficiency of apoptotic genes (such as caspase9, apoptotic protease-activating factor-1 (APAF1), or BCL2-associated X protein (BAX) and BCL2-antagonist/killer-1 (BAK) double mutant mice) results in significant abnormalities46. Apoptosis also functions to maintain homeostasis, especially in the immune system, because it eliminates unwanted cells. Dysregulation of apoptosis leads to various human diseases, such as cancer and autoimmunity. Inappropriate activation of cell death is also the leading cause of tissue injury and functional decline in a large number of acute diseases (such as stroke, myocardial infarction and brain trauma) and chronic diseases (such as diabetes and neurodegeneration). However, effective cytoprotective therapies for these diseases remain a major unmet medical need. To a significant extent, the limited success of cytoprotective drug development can be traced to the simplified view that cell death is either intrinsically regulated by apoptosis or that it is unregulated, caused by overwhelming stress (so-called necrosis). Necrosis possesses characteristic features, such as organelle swelling, mitochondrial dysfunction, massive oxidative stress and rapid plasma-membrane permeabilization, that are thought to be indicative of the catastrophic nature of cell death, rather than a result of cellular regulation. The general view of the relationship between apoptosis and necrosis is that milder insults to the cell cause apoptosis, whereas more intense insults induce uncontrollable necrosis. It is thought that such an apparently unregulated and hence untargetable process accounts for the bulk of cell death events in acute pathologies. However, in the past few years, evidence has emerged for a number of regulated non-apoptotic cell death pathways, including some with morphological features that were previously attributed to necrosis8. It has become clear that the simple apoptosisnecrosis classification does not adequately represent the complexity of endogenous cell death regulation. We are now just beginning to appreciate the important functions of regulated non-apoptotic cell death in homeostatic regulation

and development of human disease. In this review, we first provide a brief overview of apoptosis and then examine how mammals might have acquired more apoptotic genes and more complex apoptotic regulation through gene duplication. We then describe examples of regulated non-apoptotic cell death pathways and consider the possible relationships and evolutionary origins of the diverse cell death processes. Apoptosis the fundamental mechanism of PCD The mechanism of PCD elucidated by genetic studies in the nematode C. elegans defines a primordial paradigm of apoptosis9. In C. elegans, the activation of PCD is controlled by an elegant and simple pathway (FIG. 1). The initiation of apoptosis is regulated by transcriptional upregulation of egl1, a pro-apoptotic BCL2 homology-3 (BH3)-only member of the BCL2 protein family. Binding of EGL1 to antiapoptotic CED9 relieves the inhibition that CED-9 exerts on the adaptor protein CED4, allowing CED4 to bind and activate the Cys protease CED3, which in turn cleaves multiple specific cellular substrates to execute cell death. Analysis of apoptosis in mammalian cells has led to the identification of multiple mammalian homologues for each class of the C. elegans CED proteins. Extensive studies have revealed that the mechanism of mammalian apoptosis is similar to that of C. elegans, although it has become much more complex (FIG. 1). For example, although transcriptional upregulation of pro-apoptotic members of the BCL2 family can still play a part in the initiation of apoptosis in mammalian cells, many BH3- only, pro-apoptotic BCL2 family members can be activated in a number of different ways (including cleavage, phosphorylation, myristoylation and ubiquitylation 13 16) to regulate the initiation of apoptosis. In mammalian cells, the activated EGL1-like BH3- only members of the BCL2 family also inhibit CED9-like anti-apoptotic members of the BCL2 family (such as BCL2, BCL-XL and MCL1) by direct interaction (BOX 1; FIG. 1). However, a major mechanism by which anti-apoptotic members of the BCL2 family protect cells is not through the direct inhibition of caspase activation at the level of adaptor molecules (for example, the mammalian CED4- like APAF1 protein), but by protecting the integrity of the mitochondria (BOX 1; FIG. 1). BH3-only factors bind and antagonize anti-apoptotic BCL2 family members that reside in the outer mitochondrial membrane. Antiapoptotic BCL2 proteins exert their activity by preventing the pro-apoptotic multidomain BCL2 family members BAX and BAK from causing mitochondrial damage, and binding of BH3-only proteins relieves this inhibition. Additionally, a subclass of BH3-only factors might directly induce the formation of a BAXBAK channel, although how much this direct mechanism contributes to apoptosis remains a subject of debate. BAX and BAK have been proposed to form an oligomeric (at least tetrameric or larger) channel that leads to mitochondrial damage and cytochrome c release. In some cases, BAX and BAK can act through interactions with the components of the mitochondrial permeability pore, namely with the voltage-dependent anionic channel (VDAC). Mitochondrial damage might also be caused by BAX- and BAKindependent mechanisms, such as those induced by intramitochondrial K+ influx, or caused by the direct action of caspase2 on mitochondria (which occurs, curiously, independently of the protease activity of caspase-2). Mitochondrial damage and the release of mitochondrial proteins amplifies apoptotic signalling in mammalian cells, a step that is considered to be less important for apoptosis in lower organisms, including flies and worms. Cytochrome c, which is released from damaged mitochondria, promotes the formation of a heptameric apoptosome megacomplex of APAF1 and caspase9 (a member of the CED3-like Cys protease family). This leads to the conformational change and activation of caspase9 (FIG. 1). Activated caspase9 in turn cleaves and activates downstream caspases, including caspase3, caspase6 and caspase7, that carry out the execution phase of apoptosis. In addition to the intrinsic apoptosis pathway, which resembles PCD in C. elegans, mammalian cells also possess an extrinsic apoptosis pathway, which is induced by pro-apoptotic and pro-inflammatory cytokines (such as FAS ligand (FASL) and tumour-necrosis factor (TNF), which are ligands for the death-receptor family). By binding to death-domain receptors, FASL and TNF induce the formation of specific intracellular death-induced signalling complexes (DISCs23), which activate upstream caspases such as

caspase8. The activation of caspase8 can in turn cleave downstream caspases, such as caspase3 and caspase7, to execute cell death; alternatively, caspase8 can cleave the BH3-only pro-apoptotic protein BID, which in turn amplifies the cell death signal by causing mitochondrial damage and cell death. The development of cytokine-mediated apoptosis programmes in higher multicellular organisms provides a crucial way to coordinate the regulation of cell numbers at the organismal level in response to the environmental stimuli. Evolutionary expansion of apoptosis Although the core regulators of PCD in C. elegans consist of only 4 genes egl1, ced3, ced4 and ced9 each has multiple mammalian homologues. On the basis of close homology in the key regions, the mammalian cedlike genes probably arose through gene duplication and might have evolved and been selected during evolution to meet the challenges that highly complex multicellular organisms face.

Multiple genes enable functional specialization. One direct consequence of this gene duplication is

specification different apoptosis regulators respond to different pro-apoptotic signals. The mammalian caspase family provides an excellent example of specification. Different caspase family members possess distinct functions in mammalian cells that are predominantly based on their subcellular localization and proteinprotein interactions rather than on their substrate specificities. Mammalian caspases were initially assigned to three major classes: apical or activator caspases, such as caspase -2, -4, 8, -9, 10 and 12, which initiate the caspase cascade in apoptosis; executioner caspases, such as caspase 3, 6 and 7, which act in the downstream execution steps of the process; and inflammatory caspases, such as caspase 1, -5 and 11, which mediate cell death and inflammatory responses. Further analyses suggest that there might be additional important distinctions between family members in the same class. Activator caspases have distinct roles that depend on the activating complexes that they are recruited to. For example, caspase8 specifically contributes to death-receptor signalling and normal proliferation of lymphocytes; caspase2 mediates genotoxic stress-mediated death; mouse caspase12 and human caspase4 mediate endoplasmic reticulum (ER )stress-mediated death; and caspase9 is activated by the apoptosome downstream of cytochrome c release. Such division of labour might provide a sensitive mechanism to allow complex multicellular organisms to detect and differentially respond to distinct environmental stimuli. A similar division of labour has been observed for the members of the BCL2 family. BCL2 family members can be subdivided into three major classes on the basis of structural and functional differences17 (BOX 1). However, recent studies suggest that subtle but important differences exist in each sub-class. In particular, although all BH3-only factors possess a similar EGL1-like mode of action, they can be activated non-redundantly by particular types of apoptotic signals. For example, BID mediates signalling by the TNF-family and by genotoxic stress; BIM plays a key part in inducing death in lymphoid and myeloid cells following cytokine withdrawal; NOXA and PUMA signal downstream of p53 (Ref. 34); and BMF has a key role in anoikis. The specificities of BH3-only factors not only result from their interactions with upstream regulators, they also result from their distinct binding preferences for different multidomain BCL2 family members. For example, the BH3 domain of BAD binds BCL2 and BCL-XL, whereas the BH3 domain of NOXA displays selectivity towards MCL1 and the BCL2-related protein A1 (Ref. 36). Such selectivity in the interactions of different members of the BCL2 family provides an important mechanistic basis for activating distinct apoptosis signalling pathways in response to different pro-apoptotic stimuli.

The benefits of redundancy. An additional and nonconflicting significance of the multiplication of

apoptosis regulators is the benefit that is provided by redundancy. The redundancy of apoptosis programmes in mammalian cells is shown by the compensatory upregulation of caspases in different caspase mutant mice the loss of one caspase can be compensated by the upregulation of another

caspase. Although the specification of different apoptotic regulators in a signal and/or subcellular compartmentalization manner provides a mechanism to fine-tune cellular responses, danger might arise if a specified response is lost because of a genetic mutation. In this case, it appears that upregulation of an alternative caspase, or of its regulators, can compensate for the loss of a specific caspase. Although different caspases exhibit specificity towards certain cleavage sites, such specificity is relative. When present in a sufficient concentration and given sufficient incubation time, most caspases can cleave most, if not all, of the caspase substrates that have been identified to date. The preferential cleavage of a selective subgroup of caspase substrates in the early phase of apoptosis may only reflect the proximity of the substrates to the activated caspases at that specific time or they may serve to modulate the kinetics of the process. The ability of multiple caspases to cleave common substrates might serve to insure the execution of apoptosis even when one caspase is lost. Furthermore, although upstream caspases can be activated preferentially during the early stages of apoptosis (such as caspase -8 and 10, which are preferentially activated in response to a FASL or TNF signal), all caspases are cleaved and activated during the later stages of apoptosis. Therefore, all of the caspases may contribute to the execution of apoptosis. Apoptosis in mammalian cells is genetically programmed to maximize the ability to kill the cells once an appropriate order is received.

Expansion or reduction. Although the findings discussed above lead us to suggest that apoptotic pathways expanded during evolution, another view is also possible. C. elegans, along with its
evolutionary relatives, might have experienced a reduction in the complexity of the apoptotic machinery from primordial ancestors with apoptotic machinery that might have been closer to that of humans. However, we find this scenario less likely, because no such developed apoptotic machinery, resembling that of mammals, has been identified in more primitive organisms, such as plants, fungi and bacteria.

Non-apoptotic mechanisms of cell death Emerging evidence suggests that apoptosis is not the only mechanism of cellular suicide, but rather that cells might choose one of many mechanisms to die when they are ready, with apoptosis often representing the top choice. In C. elegans, most developmental cell death events occur through the apoptotic mechanism and only rare events, such as the developmental death of a linker cell, are nonapoptotic. Type I cell death, which displays the characteristic morphology of apoptosis as discussed above, is also the most frequently observed form of death during normal mouse development. Although induction of morphologically non-apoptotic developmental cell death becomes prominent in mice in vivo when apoptotic machinery is genetically disrupted (such as in caspase-deficient motor neurons), one might question whether non-apoptotic cell death can also occur under normal conditions when apoptosis is possible. One could argue that non-apoptotic mechanisms represent rudimentary back-up forms of cell death that are only relevant in rare circumstances when the apoptotic machinery is genetically unavailable. However, an understanding of the molecular mechanisms underlying non-apoptotic cell death in vitro has recently begun to emerge, and we are now beginning to appreciate the importance of these processes. Here, we summarize the data that describes three emerging regulated nonapoptotic cell death pathways: type II cell death, necroptosis and poly(ADP ribose) polymerase 1 (PARP1)-mediated necrotic death.

Type II cell death. Type II cell death is characterized by the accumulation of double-membraneenclosed vesicles. These vesicles are characteristic of autophagy and so type II cell death is often called autophagic cell death, although the role of autophagy in this type of cell death is still under debate. Autophagy is an evolutionarily- conserved intracellular catabolic mechanism that operates at low levels under normal conditions to mediate the degradation of cytoplasmic components, protein aggregates and expired intracellular organelles (for example, the ER and mitochondria) by forming

double-membrane-enclosed vesicles called autophagosomes. The contents of autophagosomes are degraded by lysosomal enzymes after autophagosomes fuse with lysosomes. Under normal conditions, autophagy has an important role in maintaining intracellular homeostasis. In this role, autophagy removes damaged or dysfunctional organelles and misfolded proteins that can be detrimental to survival (for example, of neurons). Under conditions of nutrient deprivation, autophagy promotes cell survival by degrading disposable intracellular contents, thereby generating energy and building blocks for protein synthesis. Autophagy is regulated by a large group of ATG (autophagy-related) genes that are conserved from yeast to humans (BOX 2). Activation of autophagy during Drosophila melanogaster metamorphosis has been well established. Using genetic elimination of Atg genes, it was recently shown that a lack of autophagy attenuated the degradation of D. melanogaster salivary glands by blocking PCD. Thus, this study provides evidence for the role of autophagy in developmental cell death. However, the lack of cell death efects in autophagy-deficient mutant mice (such as ATG5 or ATG7 mouse mutants), as well as in ATG7-deficient Drosophila, raises the question of whether autophagy per se or only part of the autophagy pathway is involved in type II cell death during development. Autophagy might contribute to cell death that is induced by viruses. Human immunodeficiency virus1 (HIV1) induces the accumulation of Beclin-1 protein and cell death in uninfected bystander CD4+ T cells when the HIV1 envelope protein (Env) interacts with the chemokine receptor CXCR4. Although the mechanism by which autophagy is activated by Env is not yet clear, this result suggests that autophagy might contribute to cell death in a non-cell-autonomous manner, providing a mechanism by which a virus might induce cell death independently of viral replication. Autophagy can assume the killer role when apoptosis is unavailable. For example, autophagy mediates cell death in apoptosis-deficient BAX/ BAK/ cells in response to genotoxic or ER stress stimuli. It is possible that autophagy is induced, albeit at low levels, under normal conditions, but that it becomes exacerbated in response to cellular stress during apoptosis-deficient conditions to promote cell death. In this regard, it is interesting to note that Beclin-1 the mammalian homologue of yeast Atg6 and a regulator of the type III phosphatidylinositol 3-kinase VPS34 has a BH3 domain and interacts with BCL2, which reflects the convergent regulation of apoptotic and autophagic cell death55. Furthermore, a subset of BH3-only pro-apoptotic BCL2 family members, including BNIP3 and BIK, can induce the activation of autophagic cell death. Although the mechanism by which BNIP3 and BIK induce autophagy is not clear, and although it remains to be seen whether autophagy is induced as a secondary consequence of mitochondrial damage, such studies raise the possibility that autophagic cell death might be induced in a manner similar to that of apoptosis.

Necroptosis as a form of regulated necrosis. Necroptosis (BOX 2), which represents a type of

programmed necrosis, is another intriguing example of a regulated nonapoptotic cell death mechanism. Its discovery was prompted by observations that classic apoptotic stimuli, such as deathdomain-receptor engagement by corresponding ligands, can lead to non-apoptotic cell death (as assessed using morphological criteria) when apoptosis is inhibited by caspase inhibitors or through mutations in caspase8 or FAS-associated death-domain protein (FADD). Although necroptosis is activated by the same stimuli that initiate apoptosis, the morphological features of necroptosis organelle swelling, rapid mitochondrial dysfunction, plasma membrane permeabilization and lack of nuclear fragmentation are characteristic of pathological necrosis, which is presumed to be unregulated death that is caused by overwhelming stress. Emerging evidence suggests that the initiation of the necroptotic programme by TNF occurs at the receptor level through the recruitment and activation of an intracellular signalling complex that involves the adaptor molecule RIP1 (but not the TRADDRIP1 complex, which mediates the activation of nuclear factor-B (NFB) and apoptosis). Activation of necroptosis requires the kinase activity of RIP1, which is not required for NFB and apoptosis signalling. The distinct nature of apoptosis and necroptosis was further emphasized by the discovery, in a random cell-based screen, of a series of structurally distinct small

molecule necrostatins, all of which specifically and efficiently inhibit necroptosis, but not TNFinduced apoptosis. The mechanism that leads to the execution of necroptosis downstream of RIP1 kinase activation remains unclear. RIP1 is translocated into the mitochondria, which leads to the disruption of the association of ADPATP translocase (ANT) with cyclophilin D and this might explain the rapid mitochondrial dysfunction that is associated with necroptosis. Interestingly, the protective effect of the necroptosis inhibitor necrostatin1 in an in vivo heart ischaemiareperfusion injury was dependent on the expression of cyclophilin D. However, the mechanistic details of this step remain to be determined.

PARP1-mediated necrotic death. PARP1 is a nuclear enzyme that has a key role in maintaining

genome stability. PARP1 is rapidly activated by DNA-strand breaks and recruits DNA-repair factors by attaching ADPribose units to chromatin-associated proteins. Loss of PARP1 leads to an increased sensitivity to DNA damage, which prompted the development of PARP1 inhibitors as chemopotentiators of DNA-damaging anticancer agents. However, over-activation of PARP1 can lead to caspase-independent cell death. PARP1 can mediate cell death in a number of different scenarios (BOX 2). In one scenario, alkylating DNA damage promotes rapid PARP1-mediated depletion of cytosolic NAD+, which leads to necrotic death by energy collapse in glycolytic cells (these cells depend on cytosolic NAD+ for glycolysis and energy generation). This mechanism can be viewed as an extension of the genome-surveillance function of PARP1, as it provides an elegant way to differentially regulate DNA-damage responses in rapidly proliferating glycolytic cells and in cells in a vegetative state, relying on mitochondrial respiration for maintaining ATP levels. Rapidly proliferating glycolytic cells might pose a significant danger for the organism if they accumulate DNA damage and, hence, have to be efficiently eliminated; by contrast, vegetative cells can be allowed more time to complete DNA repair. Along these lines, PARP1 activation also leads to the specific release of the inflammatory cytokine high mobility group proteinB1 (HMGB1), which can alert immune cells to the presence of dangerous cells with damaged DNA77. PARP1 also mediates cell death that is induced by secondary DNA damage associated with acute neuronal injury. In this case, excitatory neuronal cell death leads to the translocation of the poly (ADPribose)-polymer into the cytosol, triggering translocation of protein apoptosisinducing factor (AIF) from the mitochondria to the nuclei, where it mediates cell death. Interestingly, DNA-damage-induced PARP1-mediated cell death involves TNF receptor-associated factor-2 (TRAF2) RIP1-dependent activation of c-Jun Nterminal kinase1 (JNK1, also known as MAPK8), which contributes to mitochondrial dysfunction and necrotic death. However, the relationship between this process and necroptosis remains unclear. Currently, it is difficult to make definitive conclusions regarding the relationship of the two pathways of PARP1induced cell death. However, it appears possible that NAD+ depletion and AIF activity act in the same pathway and that their relative contributions to cell demise might depend on the specific cell milieu. Although the fine details of PARP1-dependent cell death remain to be sorted, it is clear that PARP1-mediated cell death is an important subject for investigation because of its role in various human pathologies (BOX 3). The roots of regulated cell death Curiously, although the phenomenon of apoptosis was first described in the context of developmental regulation, the function of apoptosis in development is not clearcut; C. elegans ced3 (loss-offunction) or ced4 (loss-offunction) mutants are developmentally normal81. If PCD does not provide a developmental advantage for nematodes that have it compared with ancient variants that lacked it, what factors might have led to the selection and evolution of PCD mechanisms? Interestingly, ced3 and ced4 mutants were found to be significantly more sensitive to death caused by Salmonella typhimurium infection compared with wild-type worms, raising the possibility that the host defence response, rather than developmental cell death, might be the primordial function of apoptosis. The host defence function of apoptosis has been expanded in mammalian cells. A large family of

mammalian NLR (NOD-like receptor) proteins, homologous to C. elegans CED4 and mammalian APAF1, function to regulate the activation of caspases in response to intracellular pathogens. NLRs contain three distinct domains: an Nterminal caspase-recruitment domain (CARD) or pyrin effector domain (with the exception of neuronal apoptosis-inhibitory protein (NAIP) and, possibly, NOD5 (also known as NLRX1)); a nucleotidebinding and oligomerization domain (the so-called NACHT domain); and a variable number of Cterminal Leu-rich repeats (LRR s) (BOX 1). The 22 identified NLRs can be divided into two large sub-classes the NODs (NOD15), which activate the RIP2NF-B pathway, and the NLRs, consisting of NALP114 (NALP is named after NACHT, LRR - and pyrin domain (PYD)containing proteins), IPAF (also known as NLRC4) and NAIP, which promote caspase1 activation. In addition, the NLR protein CIITA (major histocompatibility complex (MHC) class II trans-activator) serves the function of master regulator of MHC class II transcription. At least some of the NALPs function by recruiting the adaptor protein ASC through a homotypic PYDPYD interaction, and ASC in turn recruits caspase1 through a CARDCARD interaction. Oligomerization of NALPs brings the inflammatory caspases, such as caspase1 and caspase5, into close proximity to promote their activation. This NALP protein complex is called the inflammasome. The activation of caspase1 in turn leads to the processing and release of interleukin1 (IL-1) and IL18, which serve important proinflammatory functions. The NALP1-mediated caspase1 activation is also regulated by anti-apoptotic proteins of the BCL2 family, providing a possible role for BCL2 family proteins in regulating the host defence response. Interestingly, the origins of the NLR family can be traced to non-apoptotic regulators in simple organisms, potentially providing an exciting insight into the evolutionary origins of mammalian apoptosis. Mammalian NLRs functionally and structurally resemble NLR-like proteins, such as plant R-gene-encoded proteins, that are present in many primitive organisms (FIG. 2). The Rgene-encoded proteins contain LRRs, which sense intracellular pathogens and pathogen-induced danger signals and activate inflammatory signalling. In a manner that is highly analogous to the plant innate immune system, mammalian inflammasomes recognize signals that are presented by intracellular pathogens and mediate the activation of death in infected cells (such as macrophages) through caspase1-dependent apoptosis (pyroptosis) and induction of inflammatory signalling through caspase1- dependent processing of IL-1, IL-18 and probably IL-33. Similar to hundreds of plant Rgene-encoded proteins, distinct mammalian inflammasome complexes sense particular conserved pathogen-associated molecular patterns (PAMPs; these include components of the bacterial cell wall, bacterial flagellin, and bacterial and viral RNA) and pathogen-induced danger-associated molecular patterns (DAMPs; these include changes in ATP levels and intracellular potassium concentrations). For example, the NALP1 inflammasome senses bacterial cell wall peptidoglycans, whereas the IPAF inflammasome is activated by cytosolic flagellin. In addition, recent studies showed that plant Rgene-encoded proteins can sense PAMPs in the nuclei to induce transcriptional changes that promote an immune response. It is important to determine whether this can also represent an additional mode of action of mammalian NLRs. Curiously, whereas the plant innate immune response can lead to cell death at the infection site, involving measurable induction of caspase-like activity, it is carried out by non-caspase molecules, such as legumains. Plant proteins displaying sequence homology and structure similarity to caspases, termed metacaspases, probably represent a different branch of the evolutionary tree. Based on these observations, the mammalian inflammasome might be a convergence point between a simple ced-gene-mediated apoptotic pathway and the NLR-mediated innate immune pathway, providing capabilities for PAMP or DAMP sensing by NLRs, as well as caspase activation. Therefore, the mammalian apoptotic machinery might have evolved by converging the simple apoptotic machinery, such as that found in C. elegans, with a separate and even more ancient and highly conserved innate immune system. Furthermore, because mammalian CED4-like factors (such as NLRs, APAF1 and p53-induced protein with a death domain (PIDD)) function similarly in caspase activation, development of the mammalian apoptosis machinery might have been primarily driven by the requirements for a more efficient host defence system, rather than the more complex homeostatic and developmental regulation in higher eukaryotes. Although retaining the basic layout of

the PCD mechanism in C. elegans, the convergence of this pathway with NLR regulation led to the acquisition of a number of unique features that are characteristic of mammalian apoptosis, including direct regulation of the mammalian CED4 orthologues by specific activators (such as cytochrome c, ATP and pathogens) and divergence of the developmental, genotoxic and inflammatory pathways, which are regulated by separate sets of activating complexes (the apoptosome, PIDDosome and inflammasome, respectively) (BOX 1). Evolution of non-apoptotic cell death mechanisms Whereas PARP1-mediated non-apoptotic cell death probably evolved as part of the cellular response to DNA damage, other non-apoptotic cell death mechanisms might have evolved as a part of host defence responses. Autophagy has a well-established role in defending against viral and bacterial invasion. Sindbis virus, a single-stranded RNA virus of the Togavirus family, causes encephalitis, which can be ameliorated by overexpression of Beclin-1 in transgenic mice. Bacteria that escape into the cytosol from the endosomes can be engulfed by macroautophagy. Furthermore, autophagic machinery promotes the clearance of extracellular bacteria that are recognized by the TLR4 receptor. One could speculate that autophagic cell death might have arisen from the need to ensure the survival of the whole organism through sacrificing infected cells. Both type I and II interferons (IFNs) modulate macroautophagy as a part of the antiviral response. In this regard, it is interesting that treatment of human HeLa cells with IFN leads to the induction of autophagy through the activation of death-associated protein (DAP) kinases, and overactivation of DAP kinases can lead to autophagic cell death. Thus, autophagic cell death might be part of the antiviral response that is activated by IFNs. The functional significance of necroptosis in the host defence response is less clear. However, RIP1, a kinase that is crucially involved in the activation of necroptosis, has an important role mediating the activation of NFB and IFN genes in the context of the innate immune system, although apparently in a kinase-independent manner. RIP1 signalling can be triggered by extracellular pathogen-sensing Toll-like receptors, components of invading pathogens and inflammatory cytokines, such as TNF. The closest D. melanogaster homologue of RIP1, IMD, is also a key mediator of the innate immune response to Gram-negative bacteria, acting upstream of a number of factors that are also linked to RIP1 signalling in mammalian cells, including the D. melanogaster homologues of FADD, caspase8, IKK (inhibitor of nuclear factor (NF)-B (IB) kinase) complex and NFB. Furthermore, IMD can also mediate caspasedependent apoptosis, similar to the recently discovered RIP1-dependent, caspase8-mediated apoptosis cascade in mammalian cells. Curiously, IMD lacks a kinase domain, so kinasedependent induction of necroptosis might be an evolutionarily novel addition to the repertoire of RIP1 functions. Therefore, RIP1 might promote both the induction of specific inflammatory signalling (by NF-B and IFN upregulation) and the elimination of infected cells through a pro-inflammatory process (necrosis). The promotion of necrosis can by itself serve to potentiate the antibacterial response by causing a leakage of cellular contents and the specific release of pro-inflammatory mediators, such as IL-6. Recently discovered cell-death-independent activation of autophagy by TLR4RIP1 as a potential mechanism for bacterial clearance also fits well with this notion. These data suggest that the evolution of the innate immune response might have led to the acquisition of the RIP1-kinase-mediated necroptotic response by the IMD pathway. Opportunities for cytoprotective therapy Catastrophic cell death is the main underlying cause of death and lifelong disabilities in a broad range of human diseases, from acute disorders (such as stroke, myocardial infarction, brain and spinal cord trauma and septic shock) to chronic neurodegenerative conditions. Cell death is also an important compounding factor as a side effect of chemotherapy, many inflammatory diseases, diabetes and other conditions. Therefore, the development of efficient strategies to inhibit pathological cell death remains a key challenge of cell death research and a crucial unmet medical need. The discovery of apoptosis and the development of specific genetic and small molecule methods to inhibit

pathological apoptosis in vivo have shown that pathological cell death can, indeed, be targeted for therapeutic benefit. However, the success of anti-apoptotic therapies has been limited, perhaps because of our lack of understanding of the complexity of cell death regulation in mammalian cells. The appreciation of such complexity leads us to suggest that when considering the possibility of inhibiting a specific pathological mechanism of cell death, we must consider several issues. Does one particular form of cell death have a major role in the injury (FIG. 3)? Alternatively, are several cell death mechanisms operational in the injured tissue? If several mechanisms are operational, then combination therapy might provide maximal benefit. Whether a combination therapy will be effective depends on the contribution of each form of death, not only to the tissue injury, but also to the functional decline of the tissue (FIG. 3c). We must also consider whether a possible backup cell death mechanism that might be activated in the event of the primary mechanism is being inhibited. Given these possibilities, cytoprotective treatment might not only be improved by combination treatment, it might require combination treatment (FIG. 3d). It is important to keep in mind that although apoptosis may be the preferred type of physiological cell death, the option to die by apoptosis might not always be available under in vivo conditions. Situations that involve an imbalance of RO Sgeneration and RO S-detoxification, limited energy metabolism or a lack of proper protein synthesis might restrict the ability of cells to activate apoptotic cell death. Under such circumstances, cells might choose to die through one of the alternative cell death pathways. The existence of other cell death options suggests that there might be some plasticity in the choice of cell death programmes, with apoptosis being only one of the spectrum of available regulated cell death options. Furthermore, rather than considering a hierarchal regulation of multiple cell death mechanisms, in which cells first activate apoptosis and only undergo non-apoptotic cell death if apoptosis is inhibited, we propose that commitment to apoptosis might not even be necessary for the activation of non-apoptotic cell demise. In other words, non-apoptotic signalling can be initiated independently through alternative mechanisms to carry out the order of cellular execution under conditions that are ill-suited for apoptosis, but ideal for non-apoptotic cell death (FIG. 4). The discovery of the apoptotic programme opened the door for the development of specific cell-death-targeting smart therapies. Pathological cell death processes might represent a conscious decision of the cell in response to specific pathological signals, which would allow the development of specific approaches to influence this choice by small molecule or protein-based agents that could target the apoptotic signalling machinery. This has already led to the development of multiple classes of agent, such as BCL2 and inhibitor of apoptosis (IAP) proteins, that specifically trigger apoptosis in cancer cells. Although these agents are still in the early stages of clinical development, preliminary evidence is promising. At the same time, therapies to eliminate catastrophic tissue damage and functional decline, which are associated with pathological cell death in various human pathologies (from stroke to myocardial infarction), are still limited. The recent discovery of regulated nonapoptotic cell death might offer a new hope for treating these diseases. Although the study of these forms of cell death is still in its infancy, a number of promising results have already been generated (BOX 3). Conclusion Less than 2 decades ago, cell death was categorically considered to be passive and uninteresting. The discovery of mammalian homologues of C. elegans cell death genes led to the understanding that cell death, in the form of apoptosis, can be a highly regulated cellular mechanism. In-depth characterization of mammalian apoptosis uncovered both conservation of the basic layout of apoptotic signalling in C. elegans and evolutionary expansion of the protein families of apoptotic regulators. Moreover, these features allowed specialized activation of cell death responses by various upstream stimuli, improved integration of apoptosis with other cellular signalling and metabolic pathways, and increased fidelity of apoptosis execution, due to the redundancy in the functions of individual members of apoptosis regulatory families. Recent characterization of the inflammasome pathway of caspase1 activation revealed that the evolution of mammalian apoptosis probably involved

convergence of the primitive apoptosis machinery with the innate immune system. In the past few years we have also begun to appreciate that apoptosis is not the only form of regulated cell death. Three of the best-understood examples of non-apoptotic cell death are type II cell death, necroptosis and PARP1- mediated necrotic death (see above). Although these processes can serve functions that are complimentary or reinforce apoptosis, it is likely that they have evolved to serve specific non-redundant functions in responses to pathogen infection, nutrient and energy deprivation, and DNA damage. The changing perception of regulated cell death as an array of diverse responses, rather than a single apoptotic pathway, implicates complexity and provides novel opportunities for cytoprotective therapies. In particular, the discovery of the specific regulated, morphologically necrotic, non-apoptotic cell death mechanisms suggests that at least a subset of necrotic pathological cell death might also be regulated by cellular mechanisms and, therefore, could be amendable to therapeutic drug development. Understanding how cell death operates under the specific conditions of particular human diseases might bring in a new era of cytoprotective drug development. Figure 1 | Evolutionary expansion of C. elegans apoptotic machinery in mammalian cells. Side-by-side comparison of the Caenorhabditis elegans CED protein pathway and the core apoptotic machinery in mammalian cells shows the conservation of the general outline of the pathway. Extension of the apoptotic machinery can also be observed at every step of the pathway, including multiple B-cell lymphoma protein-2 (BCL2) homology-3 (BH3)-onlyprotein activating signals, complex regulation of the BCL2 family and the addition of mitochondrial cytochrome c release, which drives the formation of an apoptosome and activation of the upstream caspases (first caspase9 and then the executioner caspases, such as caspase3 and caspase7). Added complexity is provided by the existence of multiple family members in each class of the apoptotic nonregulators, with both redundant and

redundant functions. These regulators provide fail-safe apoptosis machinery that can generate specialized responses to various upstream stimuli. Possible direct activation of BAX and BAK by BH3-only proteins is indicated by a dotted line. APAF1, apoptotic protease-activating factor-1; BAK, BCL2-antagonist/killer-1; BAX, BCL2-associated X protein.

Box 1 | mediators

Major

classes

of

apoptosis

Caspases Caspases are a family of Cys proteases (humans have 11 caspases) that cleave their substrates after Asp residues. Caspases contain three main domains: a prodomain and large (p20) and small (p10) catalytic subunits. The large domain contains the active site Cys residue. Activation of caspases involves the proteolytic cleavage of zymogens, the removal of the prodomain and separation of the p20 and p10 subunits, or allosteric conformational changes. The prodomains of activator and inflammatory caspases contain proteinprotein-interaction domains (such as the caspaserecruitment domain (CARD) and the death-effector domain (DED)) that link them to apoptosis signalling molecules21. BCL 2 family The B-cell lymphoma protein-2 (BCL2) family is subdivided into three subclasses: anti-apoptotic (such as BCL2, BCL-XL and MCL1), multidomain pro-apoptotic (BAX and BAK) and BH3-only (such as BID, BIM, BAD, NOXA and PUMA). The BH1, BH2 and BH3 domains of multidomain family members form a hydrophobic cleft that serves as a heterodimerization interface for the BH3 domains of BH3-only proteins. The BH4 domain of anti-apoptotic BCL2 family members directly interacts with a voltagedependent anion channel and inhibits apoptotic mitochondrial changes113. Anti-apoptotic BCL2 family members can be cleaved by caspases, which results in the loss of the BH4 domain and in the proteins exhibiting proapoptotic, rather than anti-apoptotic, activity114. BH1/BH2/BH3, BCL2 homology1/2/3 domain; TM, transmembrane domain. APAF1, NLR and PIDD adaptors NLR (nucleotide-binding and oligomerization domain (NOD)-like receptor), APAF1 (apoptotic protease-activating factor-1) and PIDD (p53-induced protein with a death domain) activate caspases following pathogenic infection (caspase1), cytochrome c release (caspase9) and genotoxic stress (caspase2), respectively. The NLR family (22 proteins in humans) can be subdivided into NOD factors (which sense bacterial peptidoglycans and signal through RIP2 to activate NFB), inflammasome-forming NALPs, IPAF, NAIP and CIITA (which regulates major histocompatibility (MHC) class II transcription). Activation of APAF1, NLRs and PIDD proteins leads to the formation of large multimeric caspase-activating complexes 115,116. Caspase recruitment can be direct (through CARD domains) or can involve additional adaptors (such as RAIDD (also known as CRADD) interacting with the death-domain (DD) of PIDD116). The nucleotide-binding and oligomerization domain (the NACHT domain) and NBARC domains are required for oligomerization. The NBARC domain of APAF1 also binds to dATP, which is required for apoptosome formation. The WD40 repeats of APAF1 bind cytochrome c. The Leu-rich repeats (LRRs) of NLRs probably have a key role in sensing pathogen-associated molecular patterns83. CIITA, major histocompatibility complex (MHC) class II trans-activator; IPAF, ICE-protease activating factor; NAIP, neuronal apoptosis-inhibitory protein; NALP, NACHT, LRR- and pyrin domain (PYD)-containing proteins; NFB, nuclear factor-B.

Box 2 | Three pathways of non-apoptotic cell death

Autophagic cell death. Eighteen yeast autophagy-related (ATG) genes, which are required for autophagosome (AP) formation, have been identified (APAtg proteins). Several mammalian homologues have been identified (shown in figure panel a). The activity of the class III phosphatidylinositol 3-kinase complex is subject to inhibition by B-cell lymphoma protein-2 (BCL2) and to activation by the tumour suppressor UVRAG (UV radiation resistance-associated gene protein). Nutrient-deprivation signalling induces autophagy through the suppression of mammalian target of rapamycin (mTOR) activity. Functional analysis has implicated the same core ATG proteins in the formation of different types of autophagosomes, including those that have pro-death roles118. The molecular machinery of mammalian autophagosomes is still only partially characterized. Intrinsic differences in the composition of autophagosome protein machinery might determine its function as a pro-survival mechanism, rather than a pro-death mechanism. PAS, phagophore assembly site. See REF. 117 for further details. This panel is modified, with permission, from ref. 117 (2007) Cold Spring Harbor Laboratory Press. Necroptosis. The activation of RIP1 kinase increases reactive oxygen species (ROS) production (from the mitochondrial respiratory chain and the RIP1Rac1 NADPH oxidase complex) and activates c-Jun N-terminal kinase (JNK) kinase (which might be crucial for the execution of necroptotic cell death in some cell types). Autophagy can be prominently activated during necroptosis, but it only appears to contribute to cell death in some cell types. Other execution steps, including the activation of phospholipase A2, lipoxygenases and acid sphingomyelinase, have been described. The exact roles of these steps remain to be elucidated. Given the similarity of many of the downstream execution steps in necroptosis to those attributed to classic unregulated necrosis, the main difference between these two mechanisms might be in the method of activation (regulated by internal signalling mechanisms rather than caused by overwhelming stress) of similar relatively nonspecific execution events. SMase, sphingomyelinase; TNF, tumour-necrosis factor-; TNFR, TNF receptor. PARP1-mediated cell death. Two pathways are shown: energy collapse and apoptosis-inducing factor (AIF) translocation. In addition to the cell death role, poly(ADPribose) polymerase1 (PARP1) is involved in initiating DNA repair. Genetic deletion of PARP1 leads to sensitivity to DNA-damaging agents. Oxidoreductase activity of AIF also has a homeostatic role, as it is crucial for mitochondrial complex I function124 and a mutation in mice that is associated with severe progressive neurodegeneration has been mapped to AIF124. TRAF2, TNF receptor-associated factor-2.

Box 3 | The contribution of non-apoptotic cell death to pathological injury Recent studies have shown that non-apoptotic cell death is not just a nuisance of in vitro experimentation it makes important contributions to pathological regulation in vivo. Autophagy has been prominently observed in various disease models and its inhibition can provide therapeutic benefit. For example, recent studies suggest that Beclin-1-dependent autophagy promotes injury in mouse models of heart ischaemia reperfusion injury and heart failure. In addition, autophagy suppresses apoptosis in MYC-dependent lymphomas, promoting tumour growth. Conversely, genetic analysis has clearly established that Beclin-1 functions as a tumour suppressor. Autophagy promotes the survival of cancer cells that express apoptosis-inhibiting BCL2 family members under the conditions of hypoxia. However, inhibiting autophagy under such conditions unleashes necrosis, which might promote inflammation and, ultimately, tumorigenesis. These data suggest a complex role of autophagy in tumour formation. As the morphology of necroptotic cell death is similar to that of necrosis, it has been investigated whether pathological necrosis might actually represent, if only in part, regulated necroptosis. Indeed, administration of necroptosis inhibitor necrostastin1 (Nec1) provides significant tissue protection and functional improvements in a range of acute tissue injuries in vivo in mouse models (brain and heart ischaemiareperfusion) by mechanisms that are clearly distinct from the inhibition of pathological apoptosis. In particular, in a mouse model of stroke, the effect of Nec1 was both temporally and mechanistically (measured through caspase3 activation) distinct from that of apoptosis inhibitors. Furthermore, the protective effect of Nec1 did not require co-administration of caspase inhibitors, although the pan-caspase inhibitor zVAD. fmk and Nec1 did have an additive effect (FIG. 3). Poly(ADPribose) polymerase1 (PARP1) inhibitors provide significant protective effects in mouse brain and heart ischaemiareperfusion injury, mouse models of colitis and other inflammatory diseases, neurodegeneration and diabetes mellitus (reviewed in Refs 131133). PARP1 inhibitors have also emerged as promising anti-cancer agents, increasing the sensitivity of resistant cancer cells to various DNA-damaging agents and also selectively killing some tumour cells. Figure 2 | Mammalian NLR proteins and NBLRR proteins from simpler organisms have similar functions. Although NLR (panel a) and NBLRR (panel b) proteins (such as those encoded by plant Rgenes) have variable domain architecture83,89,90, all NLR and NBLRR proteins are characterized by the presence of: Leu-rich repeats (LRR), which have significant variability (because of alternative splicing136) and are involved in pathogen-associated molecular pattern (PAMP) sensing; nucleotidebinding ATPase oligomerization domains (NACHT or NB; these make-up the oligomerization domain); and effector domains (CARD and pyrin in NLRs; TIR and coiled coil (CC) in NBLRRs), which are involved in the recruitment of downstream factors, such as caspase1 or RIP2 in the case of IPAF and NOD proteins, or additional adaptors, such as ASC, in the case of NALP3 protein83. The mammalian NLRs (panel a) also display multiple functional similarities with NBLRRs (panel b) in that they function as part of the innate immune response. Other similarities include: NLRs and NBLRRs

both require SGT1HSP90 binding to maintain the proteins in an inactive, but signal-competent state; they are both activated by PAMPs and danger-associated molecular patterns (DAMPs) in vivo and in vitro; and their cell death and inflammatory responses are activated through effector domains. These similarities suggest that mammalian NLR proteins (and other APAF1-like molecules, such as APAF1 and PIDD) probably evolved from primitive NBLRR proteins. ASC, apoptotic speck protein; IPAF, ICE-protease activating factor; NB-LRR, nucleotide-binding Leu-rich-repeat; NLR, nucleotidebinding and oligomerization domain (NOD)-like receptor; NALP3, NACHT, LRR- and pyrin domain (PYD)- containing protein-3; TIR, translocated intimin receptor. Figure 3 | Plasticity of cell death activation in vivo. The existence of multiple cell death mechanisms suggests that careful consideration should be given to determine which mechanism (or mechanisms) is primarily activated in any particular injury paradigm if therapeutic approaches are to prove useful. As an example, multiple outcomes of antiapoptotic or anti-necroptotic therapies can be anticipated depending on the specific injury paradigm. a | Predominant necrotic death might occur when the endogenous conditions prohibit apoptosis. Use of necrostatin1 (Nec1), which inhibits RIP1-dependent necroptosis, might provide maximal cytoprotective benefit as a single agent under these circumstances. b | Alternatively, apoptotic cell death might be predominant in cell populations that are not subjected to excessive external stress, or that are intrinsically deficient in necroptosis activation58,63. zVAD.fmk prevents apoptosis by inhibiting caspases. c | A mixture of apoptotic and necroptotic cell death might occur, leading to a significant, but partial, cytoprotective effect of each treatment and an additive effect of combination therapy. d | Apoptosis might be the predominant primary form of cell death. However, inhibition of apoptosis might result in the activation of necroptosis. In this scenario, neither apoptosis nor necroptosis inhibitors might work as single agents, and combined treatment could provide maximal cytoprotective benefit. Figure 4 | Activation of alternative cell fates following TNF stimulation.

In many cell types, apoptosis is not the default response to tumour-necrosis factor (TNF) stimulation multiple cell fates can be independently adopted. a | In many cases, activation of nuclear factor-B (NFB) signalling, resulting from NFB-activating complex I formation and RIP1 polyubiquitylation136, is a primary response to TNF. In this scenario, RIP1 ubiquitylation limits the formation of pro-apoptotic signalling complexes137. In addition, NF-B transcriptionally upregulates the expression of pro-survival genes137. The low level of caspase8 cleaves RIP1, which inhibits necroptosis. b | Activation of apoptosis as a primary response to TNF requires specific apoptosispromoting conditions, such as the presence of protein synthesis inhibitors (cycloheximide), the overexpression of zinc finger-like protein (ZFRA) polypeptide138 or a deficiency in focal adhesion kinase (FAK) kinase signalling139. This leads to efficient pro-apoptotic death-inducing-signalling complex (DISC) formation136 and caspase8 activation. c | Activation of necroptosis as a primary response to TNF requires suppression of apoptotic signalling or, at least, caspase activity (by caspase inhibitors). Rapid loss of ATP140143, conditions of excessive reactive oxygen or nitrogen species production144 or ischaemic conditions63,130 can provide environments that are nonpermissive to apoptosis. Notably, all three of these pathways represent independent cell fates that are selected on the basis of the specifics of the cellular regulation. IKK, inhibitor of NF-B (IB) kinase; NIK, NF-B-inducing kinase; RIP1, receptor-interacting protein-1; TRADD, TNF receptor type 1-associated death domain protein; TRAF2, TNF receptor-associated factor-2; Ub, ubiquitin.