Although there is little doubt that mitochondria occupy the center stage of the apoptotic theatre, an ongoing polemics concerns the molecular details linking mitochondrial membrane permeabilization to the apoptotic signal transducing machinery. Thus, AIF can be released from mitochondria in a caspase-dependent as well as a caspase-independent manner. Caspase and caspase-independent death effectors such as AIF may mediate apoptosis to a variable extent, depending on cellular context as well as the death-inducing stimulus.

AIF, a caspase-independent death effector?

Apoptosis-inducing factor (AIF) is a mitochondrion-localized flavoprotein with NADH oxidase activity that is encoded by a nuclear gene. 1,2,3 AIF has been shown to translocate from mitochondria to the cytosol as well as the nucleus when apoptosis is induced.1,4 Mitochondrion-localized (eutopic) AIF is thought to be inert, as far as apoptosis modulation is concerned. In contrast, extra-mitochondrial (ectopic) AIF causes cell death.5 AIF is believed to mediate caspase-independent death because inhibition of caspase activation (e.g. by knock-out of Apaf-1) or caspase activity (by addition of synthetic pseudosubstrates) does not abolish the proapoptotic action of AIF that is introduced into cells either by microinjection of recombinant AIF or by transfection of extramitochondrially targeted AIF.1,4,5,6 Moreover, in Caenorhabditis elegans, transgenic expression of AIF can induce cell death even in a context in which Ced3 (the principal C. elegans caspase) is inactive.7

AIF can bind to DNA via electrostatic interactions, and replacement mutations of a few (positively charged) arginine residues abolish the interaction of AIF with DNA as well as its apoptogenic potential.8 Recombinant mouse AIF can induce purified nuclei to undergo chromatin condensation and ‘large-scale’ DNA fragmentation (to ∼50 kBp), yet has little or no effect on naked DNA or heat-inactived (56°C, 30 min) nuclei, indicating that it has to interact with (an)other protein(s) to cause DNA degradation.1 In contrast, the oligonucleosomal ‘ladder type’ of DNA fragmentation is mostly mediated by caspase-activated DNAse, lysosomal DNAse II,9 as well as endonuclease G, another mitochondrial protein that translocates to the nucleus upon apoptosis induction.10 In C. elegans, AIF has been shown to interact with endonuclease G, both in genetic and in biochemical terms. In vitro, recombinant C. elegans AIF and endonuclease G synergize to mediate DNA degradation.7 Moreover, recombinant mammalian AIF and cyclophilin A proteins together form an active DNAse in vitro, and AIF is a more potent apoptosis inducer in cyclophilin A-expressing cells than in cyclophilin A knockout cells.11 The two known enzymatic activities of AIF (namely NADH oxidase and allosteric activator of DNAses) are probably mediated by separate domains within the molecule, indicating that AIF is a truly bifunctional protein.12

Many authors view apoptosis as a process that is near-to-synonymous to massive caspase activation.13,14 The existence of a caspase-independent death effector thus does not fit the dominant scheme. During the last 2 years, several papers advocated the hypothesis that AIF is indeed a caspase-dependent death effector, meaning that the translocation of AIF from the mitochondrion (where it would be inert) to the nucleus (where it would induce apoptosis) would be caspase-dependent. This mini-review is intended to critically evaluate the supposed caspase dependence of AIF.

Caspase-dependent AIF release from mitochondria

There is no doubt that caspases can trigger the permeabilization of the outer mitochondrial membrane, thereby triggering the release of cytochrome c and/or AIF. Thus, addition of recombinant caspases to purified mitochondria can trigger the release of cytochrome c,15,16,17 and caspase-2- or caspase-8-activated Bid (truncated Bid) can act on mitochondria to trigger the release of cytochrome c and AIF.17,18 When cytochrome c is microinjected into cells, it can cause AIF release from mitochondria (as well as the translocation of endogenous cytochrome c), and this release depends on the activation of caspases, meaning that it is suppressed by the addition of Z-VAD.fmk or by the deletion of the Apaf-1 gene.5,6,19,20 Thus, the primary activation of caspases may occur upstream of the release of AIF.

Caspase activation upstream of the release of AIF undoubtedly is important in cell death induction via death receptors, that is the extrinsic pathway, when caspase-8 (and perhaps caspase-10) are activated in a receptor-proximal manner21. Moreover, there are arguments suggesting that caspase-2 activation can occur within the nucleus, for instance after DNA damage, and that this caspase activation process thus occurs upstream of the mitochondrial membrane permeabilization.22,23 Similarly, it has been published that caspase-12 (which exists in mice but not in humans) would be activated in a direct manner in response to endoplasmic reticulum-targeted stress, before mitochondria are activated.24 Thus, in several examples of the intrinsic pathway, primary activation of signal-transducing caspases can occur upstream of the mitochondrial phase of apoptosis.25

In C. elegans, inducible expression of Egl1 (a proapoptotic BH3-only protein from the Bcl-2 family) can trigger the release of GFP-targeted AIF (and endonuclease-G) in vivo, in the developing embryo. This effect is greatly attenuated (from 100 to 17%) – but not completely abolished – in animals lacking a functional Ced3 protein, suggesting that caspase activation is somehow required for the release of AIF from mitochondria.7 Accordingly, in mammalian cell-free systems, the addition of recombinant t-Bid or Bax protein to isolated mitochondria can lead to the release of cytochrome c without that AIF (and endonuclease-G) would be released.26,27 Although this is at odds with other reports showing that recombinant Bax and t-Bid can release AIF (and endonuclease-G) from isolated mitochondria,18,28,29 it suggests that in some particular circumstances, proapoptotic protein of the Bcl-2 family can mediate the selective release of cytochrome c from mitochondria, which simultaneously retain the caspase-independent death effectors AIF and endonuclease-G. Accordingly, it has been found in several cellular models of apoptosis that cytochrome c can be released before AIF translocates to the nucleus, at least within the limits of immunofluorescence detection,26,27,30 and that the release of cytochrome c (but not that of AIF) occurs in a caspase-independent manner (Table 1a). This led to the proposal that apoptosis would follow a general scheme in which different primary stimuli including selective cytochrome c release from mitochondria would stimulate the activation of caspases,27 which then would trigger the release of AIF and endonuclease from mitochondria (Figure 1a).

Table 1 Evidence in favor of a (a) caspase-dependent and a (b) caspase-independent AIF release mechanism in mammalian cells
Figure 1
figure 1

Two alternative models for the hierarchical relationship between AIF and caspases. (a) A model of caspase-dependent AIF release. Stimulation of the intrinsic pathway (which leads to cytochrome c release and apoptosome-dependent caspase activation), stimulation of the extrinsic pathway (via death receptors hardwiring to caspase-8), primary DNA damage (which leads to nuclear caspase-2 activation) or endoplasmic reticulum (ER)-specific stress (which can stimulate caspase-12, in the mouse) stimulate explosive caspase actication, either directly or through an action on Bid (which then stimulates the mitochondrial amplification loop). This leads to apoptosis. The mitochondrial release of AIF would be a postcaspase event and irrelevant to the cell death process. (b) A model of mitochondrial apoptosis control that involves a cross talk between two alternative cell death execution pathways. Proapoptotic Bcl-2 family members, upstream signaling caspases, and a number of small signal-transducing molecules can cause a variable degree of mitochondrial membrane permeabilization, leading to the release of caspase activators and caspase-independent death effectors. Depending on the abundance of intracellular inhibitors and activators, one or the other of the two death pathway come into action and cause full-blown apoptosis (with caspase activation) or subapoptosis (without caspase activation). Subapoptosis does lead to cell death, yet lacks some of the hallmarks of advanced apoptosis (such as oligonucleosomal chromatinolysis and karyorrhexis with the formation of nuclear apoptotic bodies)

Caspase-independent AIF release from mitochondria

In several paradigms of cell death, AIF is released from mitochondria in a caspase-independent manner. Thus, in the slime mold Dictyostelium discoideum, which lacks caspase homologues, AIF undergoes a mitochondrio-nuclear translocation upon induction of programmed cell death.31 This provides phylogenetic evidence that AIF is a death-inducing factor that comes into action before evolution invents caspases.32 In mammals there are numerous experimental situations in which the release of AIF occurs before that of cytochrome c and/or in the absence of detectable caspase activation (Table 1b). This applies to a variety of different experimental situations, including infection of T cells with HIV-1,19 photoreceptor degeneration,33 neonatal brain damage34 or myocard infarction.35 AIF release from mitochondria has been documented by different methods, either by transfection of cells with an AIF-GFP fusion construct (which allows to monitor AIF release by videomicroscopy)5 or by immunofluorescence analysis (after fixation, permeabilization and staining of cells with a variety of different polyclonal or monoclonal antibodies).1 Again, it has been found in a cornucopia of different models that caspase inhibition by pan-caspase inhibitors such as N-benzyloxycarbonyl-Val-Ala-Asp-fluoromethylketone (Z-VAD.fmk) or N-benzyloxycarbonyl-Asp(Ome)fluoromethylketone (BAF) does not affect the AIF translocation. For instance, in vivo, in a rat model of neonatal brain damage induced by ischemia reperfusion, intrathecal administration of Z-VAD.fmk does prevent the caspase-mediated degradation of fodrin, yet has no significant effect on the mitochondrial release of AIF.34 Similarly, results suggesting that pharmacological caspase inhibition or knockout of genes essential for activation of the apoptosome caspase activation complex (e.g. Apaf-1 or caspase-9) do not influence the mitochondrio-nuclear translocation of AIF have been reported for numerous in vitro cell death models (Table 1b). Moreover, in some models, the release of AIF clearly occurs independently from that of cytochrome c, for example, in the early apoptosis of primary T cells induced by staurosporine36,37 or in human embryonic fibroblasts infected by herpes simplex virus-1.38 Altogether these data suggest a different scheme from that evoked above. AIF (and other caspase-independent death effectors such as endonuclease G) would be released from mitochondria in a caspase-independent manner and thus escape from any kind of hierarchical subordination to caspases. In the advent that caspases are not activated, for instance in the presence of endogenous or exogenous inhibitors, AIF and its cooperating factors would take the relay and participate in cellular demise (Figure 1b).

Resolving the dilemma: different modes of AIF release?

The data (Table 1) and the resulting schemes (Figure 1a and b) are clearly contradictory. Supposing that there is no major bias in the methodologies that were employed (meaning that all papers contained in either Tables 1a or b are wrong), there are two possibilities to resolve this dilemma. First, one might tend to simplification and establish a ‘rule’. If science was a democracy, then this rule (compare the lengths ofTables 1a and b) would say that AIF is released from mitochondria in a caspase-independent way. If the dogma that apoptosis equals caspase activation had to prevail, then the rule would say that caspase-independent AIF release constitutes a case of non-apoptotic cell death and that, by consequence, the mitochondrio-nuclear translocation of AIF would be caspase-dependent in ‘true apoptosis’.

As a second choice (which we prefer), one might accept the complexity of cell death regulation and execution. If AIF is sometimes released in a caspase-dependent manner and often released in a caspase-independent manner, then one must assume that there are several different mechanisms to release AIF from mitochondria. When caspases play a role in proapoptotic signal transduction, then they are placed upstream of mitochondria (and by extension of AIF). However, in the case that the signal linking the lethal condition to mitochondrial membrane permeabilization is caspase-independent, then AIF release can occur independently from (and sometimes upstream of) caspase activation (Figure 1b).

It is an ongoing conundrum how the apoptotic mitochondrial membrane permeabilization is achieved in molecular terms. Thus, present theories are unable to explain how the loss-of-function mutation of the mitochondrial intermembrane proteins Omi/HtrA2 can increase the susceptibility of mitochondria to undergo membrane pemeabilization.39 In vitro reconstitution experiments have suggested that proapoptotic members of the Bcl-2 family (alone or in conjunction with VDAC) would form channels with a precise molecular cutoff, meaning that they would allow for the passage of cytochrome c (16.6 kDa) but not that of AIF (57 kDa).40,41 Yet other reports, however, disclaim this molecular sieve effect and suggest that composite channels formed by t-Bid, Bax and mitochondrial membranes form giant pores that allow for the passage of solutes up to 2000 kDa.42 In addition, the formation of small pores in the inner mitochondrial membrane with consequent volume changes may lead to the rupture of the outer mitochondrial membrane.43,44,45 It has also been suggested that cytochrome c and AIF may be tethered to the inner mitochondrial membrane, perhaps by electrostatic interactions with negatively charged lipids (such as cardiolipin) that are particularly abundant in that localization.26,46 In that case, the release of different apoptogenic proteins from mitochondria would involve a multistep process requiring (nonspecific?) membrane permeabilization as well as (specific?) desorption of the proteins from the inner membrane in a manner that is yet to be understood. This may explain why in some experimental systems AIF is released well before cytochrome c or vice versa. Another degree of complexity in the regulation of protein release may reside in the existence of multiple submitochondrial compartments, for example, in the separation of the intermembrane space (which contains 15% of cytochrome c and presumably most of the AIF) and the intracristae space (which contains 85% of cytochrome c), meaning that the microanatomy of mitochondria (and its dynamic change subject to fusion/fission events, volume regulation and permeability transition) will impinge on the release of apoptogenic factors from the cell's Pandora's box.47,48 If this speculation is true, the central enigma of apoptosis research – the mechanisms of lethal mitochondrial change – is still to be resolved.