Mitochondria, the killer organelles and their weapons



Apoptosis is a cell-autonomous mode of death that is activated to eradicate superfluous, damaged, mutated, or aged cells. In addition to their role as the cell's powerhouse, mitochondria play a central role in the control of apoptosis. Thus, numerous pro-apoptotic molecules act on mitochondria and provoke the permeabilization of mitochondrial membranes. Soluble proteins contained in the mitochondrial intermembrane space are released through the outer membrane and participate in the organized destruction of the cell. Several among these lethal proteins can activate caspases, a class of cysteine proteases specifically activated in apoptosis, whereas others act in a caspase-independent fashion, by acting as nucleases (e.g., endonuclease G), nuclease activators (e.g., apoptosis-inducing factor), or serine proteases (e.g., Omi/HtrA2). In addition, mitochondria can generate reactive oxygen species, following uncoupling and/or inhibition of the respiratory chain. The diversity of mitochondrial factors participating in apoptosis emphasizes the central role of these organelles in apoptosis control and unravels novel mechanisms of cell death execution. © 2002 Wiley-Liss, Inc.


AIF, apoptosis inducing factor; ΔΨm, mitochondral transmembrane potential; MMP, mitochondrial membrane permeabilization.

Apoptosis is one of the fastest-growing research fields in Life Science. Tremendous advances over the last 10 years have made clear that this cell suicide programme is essential for development and for adult tissue homeostasis of all metazoan animals. A dysregulation of apoptosis is implicated in the appearance of several pathologies. Thus, abnormal inhibition of this process is a hallmark of neoplasia, whereas massive apoptosis has been linked to acute diseases such as stroke and septic shock as well as to chronic pathologies including AIDS and neurodegenerative disorders.

Since the most striking morphological alterations linked to apoptosis affect the nucleus (Kerr et al., 1972), it has initially been assumed that cellular suicide would be controlled at the nuclear level. One recurrent idea was that the specific transcriptional induction of “killer genes” would seal the cell's fate. However, in most models of apoptosis, de novo translation and transcription of macromolecules are dispensable for cell death to occur, implying that the suicide machinery is already established within the cell. Another idea that turned out to be wrong postulated the activation of DNases to seal the cell's fate. However, apoptotic modifications can be observed in cells from which the nucleus has been experimentally removed (cytoplasts) (Jacobson et al., 1994; Schulze-Osthoff et al., 1994), implying that apoptosis must be regulated at the cytoplasmic level.

The notion that a specific class of proteases, the “caspases” (cysteine aspartate-specific proteases), are involved in apoptosis, emerged from the use of the nematode Caenorrhabditis elegans. Genetic studies revealed that 131 cells out of 1,093 die during the development of the worm and that a specific set of genes are required for this process to occur. Cell death is induced by ced-3, ced-4, egl-1 and inhibited by ced-9 (Horvitz, 1999). The products of these four genes have counterparts in mammals. Thus, ced-3 is homologous to mammalian caspases. Caspases are synthesized as inactive precursors and undergo proteolytic maturation upon apoptosis induction. Ced-4 is an adaptor protein that is necessary for Ced-3 activation. The mammalian homolog of Ced-4 in mammals is Apaf-1 (apoptotic protease activating factor-1), which together with cytochrome c and dATP recruits pro-caspase-9 to form the so-called “apoptosome,” the pro-caspase-9 activation complex (see below). Ced-9 is an anti-apoptotic protein homologous to the human oncoprotein Bcl-2. In mammals, several proteins similar to Bcl-2 have been discovered, all carrying at least one “Bcl-2 Homology” (BH) domain. Bcl-2 family members are both anti-apoptotic (Bcl-2, Bcl-Xl, Bcl-w) and pro-apoptotic. Some of the pro-apoptotic members have three BH domains in their sequence (Bax, Bak, etc.), whereas others like Bim and Bid carry only one domain and for this reason are called “BH3-only.” The homologs of C. elegans pro-apoptotic protein Egl-1 stand in the “BH3-only” category. In C. elegans, Egl-1 interacts with Ced-9, thereby dissociating the interaction between Ced-9 and Ced-4. Ced-9 is a mitochondrial protein which normally tethers Ced-4 to mitochondrial membranes. Upon expression of Egl-1, Ced-4 dissociates from mitochondria and associates with the nuclear envelope. Moreover, once released, Ced-4 interacts with Ced-3 to cause its activation.

It would be incorrect, however, to assume that caspase activation in mammals follows the same rules as those established in C. elegans. Thus, the mammalian Ced-4 and Ced-9 equivalents, Apaf-1 and Bcl-2-like proteins, do not interact physically, and Apaf-1 is actually a cytosolic (not a mitochondrion-associated) protein (Haraguchi et al., 2001). Moreover, Apaf-1 possesses WD domains which are missing in Ced-4 and which have to interact with cytochrome c so that Apaf-1 can trigger caspase activation. In addition, the notion that caspases are critically involved in cell death does not hold true for many (most?) cases of mammalian cell death. Inhibition of caspases significantly inhibits the acquisition of some hallmarks of apoptosis such as the formation of nuclear apoptotic bodies. However, in most models of apoptosis induction, pharmacological caspase inhibition does not prevent cell death (Hirsch et al., 1997; Green and Kroemer, 1998; Kroemer et al., 1998). Although caspases may act as signal-transducing molecules which connect plasma membrane receptors to cytoplasmic events (in which case their inhibition prevents pro-apoptotic signaling at an early, receptor-proximal level), in most paradigms of apoptosis, caspase inhibition thus affects the way how a cell dies rather than the decision whether a cell dies.

Although the paper by Kerr, Wyllie, and Currie (Kerr et al., 1972) which coined the word “apoptosis” stated that several organelles like mitochondria, ER, and lysosomes did not undergo major modifications during apoptosis, recent evidence suggests that no cellular compartment is spared (Ferri and Kroemer, 2001). Mitochondria are particularly affected early during the apoptotic process, and they are now thought to act as central coordinators of cell death (Kroemer et al., 1995; Kroemer, 1997; Green and Kroemer, 1998; Green and Reed, 1998; Kroemer and Reed, 2000; Boya et al., 2001; Ferri and Kroemer, 2001). Indeed, several pro-apoptotic signal transduction and damage pathways converge on mitochondria to induce mitochondrial membrane permeabilization (MMP) and this phenomenon is under the control of Bcl-2-related proteins (Kroemer et al., 1995; Kroemer, 1997; Green and Kroemer, 1998; Green and Reed, 1998; Kroemer and Reed, 2000; Boya et al., 2001; Ferri and Kroemer, 2001). MMP differentially affects the inner and the outer mitochondrial membranes through mechanisms which are still a matter of debate (Martinou and Green, 2001; Zamzami and Kroemer, 2001). The inner membrane is characterized by a transmembrane potential (ΔΨm) generated through the activity of proton pumps of the respiratory chain. ΔΨm dissipates after the cells are induced to die (Zamzami et al., 1995). However, the inner membrane still retains matrix proteins, indicating that permeabilization of this membrane is rather partial. In contrast, the outer mitochondrial membrane becomes completely permeabilized to proteins, resulting in the leakage of potentially toxic mitochondrial intermembrane proteins that orchestrate the degradation phase of apoptosis. This review will summarize current knowledge on such lethal mitochondrial proteins.


Cytochrome c is encoded by a nuclear gene and translated into a precursor (apo-cytochrome c) that is unable to participate in apoptosis induction. The precursor is subsequently imported into mitochondria, where it is refolded into a globular protein at the same time as the cytochrome c lyase catalyzes the attachment of a hem moiety. Holo-cytochrome c (that is cytochrome c with its hem prosthetic group attached) remains sequestered in the mitochondrial intermembrane space where it serves its physiological function as an electron shuttle between complex III and IV of the mitochondrial respiratory chain. In 1996, Xiadong Wang and co-workers reported the surprising observation that holo-cytochrome c (but not apo-cytochrome c) is required for the activation of caspase-3 in a cell-free system (Liu et al., 1996). This result was welcomed with incredulity but was rapidly confirmed by several reports showing that cytochrome c release and caspase activation was blocked by the anti-apoptotic protein Bcl-2 (Kluck et al., 1997; Yang et al., 1997). Today it is an established fact that cytochrome c, once present in the cytosol, drives the assembly of a high molecular weight caspase activating complex termed the “apoptosome.” Cytochrome c binds to Apaf-1, within its C-terminal region rich in WD motifs (Zhou et al., 1997). This event facilitates the binding of ATP to Apaf-1 and exposes an oligomerization surface encompassing the N-terminal Ced-4 homologous region of the protein (Adrain et al., 1999). Oligomerization of Apaf-1 is accompanied by the recruitment of pro-caspase-9 to the CARD motif at the Apaf-1 N-terminus as both interactions are repressed by the C-terminus of the latter (Benedict et al., 2000). Proteolytic maturation of caspase-9 within the complex is then achieved through an autocatalytic cis-processing event within individual caspase-9 molecules. However, this maturation event is not required for the activation of pro-caspase-9 and uncleavable pro-caspase-9 mutants become enzymatically active upon simple interaction with Apaf-1, which functions as its obligate allosteric activator. Caspase-9 then functions as an initiator caspase which proteolyticlally activates the effector pro-caspase-9.


Even though most pro-caspases have a cytosolic localization, certain pro-caspases are present in the intermembrane space of mitochondria, including pro-caspases-2, -3, and -9 (Mancini et al., 1998; Krajewski et al., 1999; Susin et al., 1999a). The relative amount of pro-caspase-9 and -3 found in mitochondria versus the cytosol depends on the cell type. In SHEP neuroblastoma cells (but not in HeLa cells), the mitochondrial intermembrane space also contains mature, pre-processed caspase-9 (Costantini et al., 2002). This pre-processed caspase-9 is likely to have little or no enzymatic activity in the absence of interaction with Apaf-1. It is still unclear via which mechanisms certain pro-caspases are segregated into mitochondria. Upon apoptosis induction, pro-caspases are released from mitochondria, and it is reasonable to assume that they participate in apoptosis induction. Sequestration of pro-caspase-9 to the mitochondrial intermembrane space of some cell types, for instance neurons, may constitute an additional mechanism to prevent caspase activation.


Inhibitor of apoptosis proteins (IAPs) such as XIAP, c-IAP1, and c-IAP2 can effectively bind to and inhibit processes caspase-3 and -9. The functional unit within each IAP is the so-called baculovirus IAP repeat (BIR). Most mammalians IAP have more than one BIR, each one exhibiting a different function. For example, in XIAP, the third BIR domain (BIR3) potently inhibits the activity of processed caspase-9, whereas the linker region between BIR1 and BIR2 selectively targets active caspase-3 (Takahashi et al., 1998; Sun et al., 2000). In cells undergoing apoptosis, caspases are liberated from IAP blockade, a process that is made possible by a protein called Smac (Second Mitochondrial Activator of Caspases) (Du et al., 2000) or Diablo (Direct IAP Binding protein with Low pI) (Verhagen et al., 2000). Smac/DIABLO is synthesized as a 239 amino acid precursor protein that is targeted to mitochondria via its N-terminal domain. This mitochondrial localization sequence (MLS) is proteolytically removed after import into the intermembrane space. In response to various apoptotic stimuli, mature Smac/DIABLO is released into the cytosol where it inhibits IAPs. Subsequent in vitro experiments have revealed an interaction between Smac/DIABLO and the BIR2 and BIR3 of XIAP that was abolished by mutations in the N-terminal region of Smac/DIABLO. Conversely, a peptide corresponding to this N-terminus suffices to facilitate caspase activation in cellular extracts. Three pro-apoptotic proteins from Drosophila, Hid, Reaper, and Grim are similar to Smac in their N-terminus, starting with an alanine residue (AVPS) which is responsible for their IAP inhibitory activity (Abrams, 1999; Wu et al., 2000; Srinivasula et al., 2001). By binding to IAP, Smac displaces active caspases or prevents IAPs binding active caspases and thus promote death of the cell. For example, when Smac binds to XIAP it prevents it from binding processed caspase-9 and thus promotes death following UV irradiation (Ekert et al., 2001). However, caspase activation facilitated by Smac may involve additional, yet to be discovered mechanisms that are independent from its interaction with XIAP (Creagh et al., 2001).


Smac is not the only protein that promotes cell death via its interaction with IAP. Recently, using XIAP as a bait, four different groups have identified HtrA2/Omi as a novel IAP binding protein (Hegde et al., 2001; Martins et al., 2001; Suzuki et al., 2001; Verhagen et al., 2001). The precursor form of HtrA2 is a 50-kDa protein. The N-terminal part that bears the MLS is processed after import into the mitochondria, generating a mature 36-kDa form with an N-terminus resembling that of Smac. HtrA2 belongs to a family of serine proteases that is well conserved from bacteria to humans. Htr stands for high temperature requirement (Faccio et al., 2000; Gray et al., 2000). The bacterial HtrA gene product is one of the best-characterized proteins of the family. HtrA endoprotease is localized within the periplasmic space of bacteria and its presence is necessary for bacterial thermotolerance (Lipinska et al., 1990). Moreover, it has recently been shown that bacterial HtrA has a dual role acting as a chaperone at normal temperature and as an active protease at high temperatures (Spiess et al., 1999). In normal human cells, HtrA2 is confined to the mitochondrial intermembrane space. Upon apoptosis induction with different agents including staurosporine, TRAIL, or UV irradiation, HtrA2 is released into the cytosol (Hegde et al., 2001; Martins et al., 2001; Suzuki et al., 2001; Verhagen et al., 2001) where the mature form of the protein, but not its precursor, binds to IAPs in a similar manner as does Smac/DIABLO, thereby facilitating caspase activation. However, at difference with Smac/DIABLO, overexpression of extramitochondrially targeted HtrA2 induces an unusual form of cell death without membrane blebbing or apoptotic body formation, while the integrity of the plasma membrane is maintained (Suzuki et al., 2001). Moreover, HtrA2 overexpression can induce cell death in the presence of caspase inhibitors as well as in Apaf-1−/− and caspase-9−/− cells (Hegde et al., 2001). Mutation of the N-terminus (which removes the anti-IAP activity) and simultaneous mutation of aminoacid residues required for serine-protease activity completely abolished the death-inducing function of HtrA2 (Suzuki et al., 2001). Thus, HtrA2 promotes cell death via two different mechanisms: one relies on IAP inhibition and involves a significant increase of caspase activity, whereas the second depends on its serine protease activity and is caspase-independent (Hegde et al., 2001; Suzuki et al., 2001).


Shortly before cytochrome c was identified as the mitochondrial activator of cytosolic caspases, our laboratory described a bioactivity recovered from atractyloside-treated mitochondria that induced apoptosis independently of caspases (Zamzami et al., 1996). The protein responsible for this activity was purified and named apoptosis inducing factor (AIF) (Susin et al., 1996). AIF gene is largely conserved between mice and humans (92% identity) (Susin et al., 1999b). The AIF precursor is a 67-kDa protein that is organized in three domains: an N-terminal region that bears a mitochondrial localization sequence (MLS), a central spacer sequence, and a C-terminal part that shows an important similarity with bacterial oxidoreductases and harbors two putative nuclear localization sequences. After the precursor is imported into the mitochondria, the N-terminal part carrying the MLS is cleaved off, giving raise to the 57-kDa mature form of AIF. After an apoptotic insult, AIF translocates to the cytosol and the nucleus, where it induces peripheral chromatin condensation and high molecular weight (50 kbp) DNA fragmentation. The mitochondrio-nuclear translocation of AIF appears to be a universal feature of apoptosis occurring in human or mouse cells. Intriguingly, an AIF homolog was also found undergo a similar translocation process during developmental cell death of Dictyostelium discoideum (Arnoult et al., 2001), a fungus which lacks caspases. Overexpression of the anti-apoptotic protein Bcl-2 blocks the AIF redistribution in mammalian cell lines (Susin et al., 1996, 1999b; Daugas et al., 2000). When AIF is microinjected into the cytoplasm of intact cells, it induces several hallmarks of apoptosis, like ΔΨm dissipation, phosphatidylserine exposure, and nuclear apoptosis (Ferri et al., 2000; Susin et al., 2000; Loeffler et al., 2001). The nuclear and mitochondrial effects of AIF can be recapitulated in cell-free systems (Susin et al., 1999b, 2000). Similar effects are obtained when cells are transfected with an AIF deletion mutant in which the MLS has been removed, thereby causing AIF to accumulate in the extramitochondrial compartment and in particular in nuclei (Loeffler et al., 2001). None of these alterations is inhibited by caspase inhibitors or by Bcl-2 overexpression, and AIF can induce chromatin condensation in cells lacking Apaf-1 or caspase-3 (Ferri et al., 2000; Susin et al., 2000; Loeffler et al., 2001).

Like cytochrome c, AIF possesses a second enzymatic activity different from its apoptotic potential. Because it can stably bind FAD, AIF falls in the category of flavoproteins. In line with its primary sequence, AIF has oxydoreducatase properties. It displays NAD(P)H oxidase as well as monodehydroascorbate reductase activities. Moreover, it can catalyze the reduction of cytochrome c in the presence of NADH (Miramar et al., 2001). Regardless of the presence or the absence of FAD and/or NAD(P)H, AIF can induce nuclear apoptosis, which clearly indicates that the apoptotic and the oxidoreductase function are independent (Miramar et al., 2001). The cellular targets of AIF remain elusive. However, an endogenous inhibitor of AIF was recently discovered, namely heat shock protein (Hsp) 70. Hsp70 physically interacts with AIF and inhibits its apoptotic effects both in vitro and in intact cells. Surprisingly, the chaperone function of Hsp70 is not required to block AIF (Ravagnan et al., 2001). Further studies will have to define the critical amino acids on AIF for the interaction with Hsp70 and establish whether Hsp70 is the sole endogenous AIF inhibitor. More generally, the identification of AIF target(s) will tell whether this factor originates a novel pathway of nuclear apoptosis or whether it is an activator of known nucleases.

Endo G

Recently, a further mitochondrial factor translocating to the nucleus has been isolated: endonuclease G (Endo G) (Li et al., 2001). Endo G is a mitochondrial nuclease encoded by a nuclear gene that was already described but whose role in apoptosis was never suspected. Endo G can be recovered from mitochondria treated with the active form of the pro-apoptotic protein Bid. Once liberated into the cytosol, Endo G translocates towards the nucleus where it generated oligonucleosomal DNA fragmentation even in the presence of caspase inhibitors (Li et al., 2001). Subsequent studies have demonstrated that Endo G catalyses both high molecular weight DNA cleavage and oligonucleosomal DNA breakdown in a sequential fashion. Moreover, Endo G cooperates with exonuclease and DNase I to facilitate DNA processing (Widlak et al., 2001). Endo G thus may act in a similar fashion as CAD (caspase-activated DNAse), a DNAse whose activation critically relies on caspases. However, additional studies are necessary to fully evaluate the importance of Endo G in nuclear apoptosis with respect to others effectors. Thus, Samejima et al. have recently shown that oligonucleosomal fragmentation is blocked in CAD−/− chicken DT40 lymphoma cells, whereas high molecular weight fragmentation still occurs (Samejima et al., 2001).


Mitochondria presumably release all soluble intermembrane proteins through the permeabilized outer membrane when apoptosis is induced. Accordingly, mass spectroscopic studies performed on atractyloside or t-Bid-treated mitochondria revealed the presence of dozens of different proteins (Patterson et al., 2000; Van Loo et al., 2002). At present, it is unclear which among these proteins actually participate in the apoptotic process as catabolic effectors. Pro-caspase-3 was found to form a ternary complex together with molecular chaperones heat shock protein (Hsp) Hsp60 (most abundant in the mitochondrial matrix) and its co-chaperone Hsp10 (most abundant in the intermembrane space) in HeLa and Jurkat cells. Remarkably, the association with the chaperones accelerated the maturation of pro-caspase-3 (Samali et al., 1999; Xanthoudakis et al., 1999). However, it is unclear if and how Hsp60 can translocate from the mitochondrial matrix to the cytosol to facilitate caspase activation.


As can be deduced from the beforementioned list of proteins, mitochondria constitute authentic “poison cupboards” (Ernshaw, 1999) (Fig. 1). Upon induction of apoptosis, they release several potentially lethal proteins that either participate in caspase activation (cytochrome c, Smac/DIABLO, HtrA2) and/or can induce cell death in a caspase-independent fashion (AIF, Endo G, HtrA2). At present, it is difficult to weight the relative contribution of each of these factors to apoptosis. Knock-out of cytochrome c obviously induces major defects in oxidative phosphorylation and is embryonic lethal on day 10 post coitum. Embryonic stem (ES) cells lacking cytochrome c are relatively resistant to apoptosis induction by staurosporin and UV irradiation (Li et al., 2000). Knock-out of the aif gene revealed that this factor is rate limiting for apoptosis in vitro, as well as for mouse development (Joza et al., 2001). aif/Y embryonic stems (ES) cells are resistant to serum deprivation and vitamin K3 (menadione). More importantly, AIF is indispensable for the process of cavitation, the very first wave of apoptotic cell death during embryogenesis (Joza et al., 2001). This may be the explanation why the knock-out of AIF is embryonic lethal. It has also been reported that Endo G is important for cell death occurring during the development of C. elegans, meaning that Endo G-deficient nematodes possess extra cells (Parrish et al., 2001). Altogether these knock-out studies underline the importance of mitochondrion-triggered caspase-dependent and caspase-independent pathways for cell death regulation. In a way, they confirm the general notion that MMP constitute a major checkpoint in apoptotic cell fate decisions.

Figure 1.

Contribution to apoptosis of different proteins released from the mitochondrial intermembrane space. For details and explanations see main text.


This work has been supported by a special grant from the Ligue Nationale contre le Cancer, as well as grants from ANRS and the European Commission (QLG1-CT-1999-00739 to G.K.).