The Central Executioner of Apoptosis: Multiple Connections between Protease Activation and Mitochondria in Fas/APO-1/CD95- and Ceramide-induced Apoptosis

Human CEM-C7.H2 lymphoma cells were transfected with a PHD1.2 crmA cDNA (1.46 kb) cloned in the sense orientation into a β-actin STneo B vector (crmA cells) or a vector-only control (Neo). Three different clones hyperexpressing crmA at the protein level yielded similar functional results. Results are shown for the C7.H2/D1.2/2E8 clone. Alternatively, CEM-C7.H2 cells were transfected with pEF-tTA 2A10, a doxycyclin-inhibitable transactivator (tTa) and super-transfected with a tTa-repressed bcl-2 construct in a tk-Hyg selection vector (pUGD10-3 Bcl-2 tkHyg; reference 29; results are shown for one out of two clones yielding similar data). Bcl-2 expression was repressed by culture on doxycycline (10 ng/ml, 48 h), as described (30). Apoptosis was induced by stimulation of 1–5 × 10⁶ cells/ml with the Fas–cross-linking IgM monoclonal antibody CH-11 (500 ng/ml; Immunotech, Marseille, France) in the presence or absence of the membrane-permeant–specific inhibitor of ICE, Ac-YVAD.cmk (100 μM; Bachem, Basel, Switzerland), or alternatively with C₂ ceramide (50 μM; Sigma Chemical Co., St. Louis, MO).

To evaluate the ΔΨ_m, cells (10⁶/ml) were incubated with the cationic lipophilic dye chloromethyl-X-rosamine (CMXRos; 150 nM; Molecular Probes, Inc., Eugene, OR; reference 15). As a control, cells were simultaneously treated with the uncoupling agent carbonyl cyanide m-chlorophenylhydrazone (mClCCP; 50 μM; Sigma Chemical Co.). CMXRos incorporates into mitochondria driven by the ΔΨ_m (15) and reacts with thiol residues to form covalent aldehyde-fixable thiol ester bonds (31). After fixation (4% paraformaldehyde in PBS for 15 min at room temperature), cells were washed and stained for the detection of chromatinolysis using the TUNEL method, as described (31). In one series of experiments, cells were stained with the potential-sensitive dye DiOC₆(3) (15 min, 37°C, 40 nM) together with a biotin–Annexin V conjugate (50 × dilution; revealed by streptavidine-phycoerythrine at 5 μg/ml, following the manufacturer's protocol; Boehringer Mannheim GmbH, Mannheim, Germany), followed by sorting of DiOC₆(3)^low Annexin V⁺, DiOC₆ (3) Annexin V⁻ and DiOC₆(3)^high Annexin V⁺ cells on an Elite cytofluorometer (Coulter Corp., Miami, FL), as described (6, 15).

100 μl cytosols (10⁷ cells/100 μl in cell-free system [CFS] buffer [220 mM mannitol, 68 mM sucrose, 2 mM NaCl, 2.5 mM PO₄H₂K, 0.5 mM EGTA, 2 mM MgCl₂, 5 mM pyruvate, 0.1 mM PMSF, 1 mM dithiotreitol, 10 mM Hepes-NaOH]) and pH 7.4 buffer (supplemented with additional protease inhibitors: 1 μg/ml leupeptin, 1 μg/ml pepstatin A, 50 μg/ml antipain, 10 μg/ml chymopapain) were prepared by five freeze/thaw cycles in liquid nitrogen, followed by centrifugation (1.5 × 10⁵ g, 4°C, 1 h) as described (32). The protein concentration in the supernatant was determined by the Bradford assay (Bio Rad Labs., Hercules, CA). ICE activity was measured using a fluorogenic substrate containing the cleavage site YVAD, 4-(4′-deimethylaminophenylazo) benzoic-YVADAPV-5-(-2-aminoethyl-amino) naphtalene-1-sulfonic acid (Bachem), as described (22), using a spectrofluorometer (Kontron SFM 25; Kontron AG, Zurich, Switzerland). The capacity of cytosols or purified recombinant human CPP32 activity to cleave the CPP32 recognition site DEVD was determined using Ac-DEVD-amino-4-methylcoumarin (Bachem) as fluorogenic substrate (33).

Mitochondria were isolated from BALB/c hepatocytes (6–8 wk, female) or from different CEM-C7.H2 lines. Mitochondria were purified on a Percoll^® (Pharmacia, Uppsala, Sweden) gradient (34) and were stored on ice in CFS buffer for up to 4 h. Mitochondria were washed and resuspended in CFS supplemented with 2 mM ATP. Nuclei from HeLa cells were purified on a sucrose gradient and conserved in 50% glycerol (Sigma Chemical Co.) in HeLa nuclei buffer at −20°C for a maximum of 15 d, as described (35).

Purified hepatocyte mitochondria were treated with atractyloside (5 mM; Atr; Sigma Chemical Co.) to induce PT and liberation of AIF (9, 11). Supernatants (150,000 g, 1 h, 4°C) from these mitochondria were concentrated on Centricon 10 membranes (⩾10 kD; Amicon, Beverly, MA) and then injected into a FPLC column (MonoQ (HR5/5); Pharmacia) preequilibrated with protein-free CFS buffer (see below). Elution was performed on a linear gradient from 0 to 250 mM NaCl at 0.5 ml/min over 30 min, followed by elution at 1 M NaCl thereafter. All fractionation steps were carried out at 4°C to avoid loss of biological AIF activity. The active fraction (eluting at 110 mM NaCl; reference 11) was dialyzed against protein-free CFS buffer (4°C, overnight, 5,000× excess of CFS buffer), concentrated on Centricon 10 membranes, adjusted to a concentration of 30 μg/ml, and aliquoted to be snap frozen in liquid nitrogen and stored at −80°C.

For the induction of PT, mitochondria from different cell lines were incubated with cytosolic extracts from αFas–treated cells (standard dose of 30 μg protein/ml), purified recombinant ICE (50 μg/ml), the prooxidant ter-butylhydroperoxide (t-BHP; 30 μM), atractyloside (5 mM; Sigma Chemical Co.), the protonophore mClCCP (100 μl; Sigma Chemical Co.), bongkrekic acid (50 μM; provided by Dr. Duine, Delf University, Delf, The Netherlands), monochlorobiman (30 μM; Sigma Chemical Co.), and/or the calpain inhibitor N-benzyloxycarbonyl-l-leucyl-l-leucyl-l-tyrosine diazomethylketone (100 μM; Molecular Probes Inc., reference 9). Recombinant ICE was produced following standard procedures (36) and was allowed to partially (⩽5%) autoactivate by incubation at 20°C for 2 h, followed by storage on ice for a maximum of 4 h. Two different consequences of PT were assessed: (a) mitochondrial large amplitude swelling and (b) collapse of the ΔΨ_m. For determination of swelling, mitochondria were washed and resuspended in CFS buffer supplemented with 2 mM ATP at a concentration of 100 μg mitochondrial protein/10 μl buffer, followed by addition of 90 μM CFS containing 2 mM ATP and recording of adsorption at 540 nm in a spectrophotometer (DU 7400; Beckman Instrs., Carlsbad, CA), as described (9). After stabilization of the adsorption during a minimum interval of 30 s, the indicated substance was added in a volume of 5 μl. The loss of absorption induced by 5 mM atrectyloside within 5 min was considered 100% of the large amplitude swelling, as described (9). The ΔΨ_m was measured using DiOC₆ (3) (40 nM, 15 min at 37°C; Molecular Probes Inc.), after having added the indicated PT inducer (30 min, room temperature). Mitochondria were analyzed in an Elite cytofluorometer (Coulter Corp.). All ΔΨ_m determinations were performed at least three times in each experiment.

Nuclei from mouse liver cells were purified on a sucrose gradient (35), washed two times (1,000 g, 10 min, 4°C), and resuspended in CFS buffer. In standard conditions, nuclei (10³ nuclei/μl) were cultured in the presence of mitochondrial preparations for 90 min at 37°C. Nuclei were stained with propidium iodide (10 μg/ml; Sigma Chemical Co.) and the lipophilic dye 5-methyl-bodipy-3-dodecanoic acid (100 nM; Molecular Probes Inc., reference 37), followed by cytofluorometric analysis in an analyzer (EPICS Profile II Analyzer; Coulter Corp.). Only membrane surrounded (5-methyl-bodipy-3-dodecanoic acid-labeled) particles were gated. A good correlation between the frequency of nuclei exhibiting chromatin condensation with 4′-6-diamidino-2-phenylindole dihydrochloride (10 μM; Molecular Probes, Leiden, The Netherlands; reference 38) and hypoploidy with PI was obtained (Susin, S.A., and G. Kroemer, manuscript in preparation). DNA fragmentation was determined by horizontal agarose gel electrophoresis and ethidium bromide staining as described (9). Electron microscopy of osmium tetroxyde-fixed nuclei was performed as described (13).

A cDNA encoding human CPP32 was generated by PCR using the plasmid pKSII-CPP32 (gift from Guy Salvesen, Burnham Institute, La Jolla, CA) as a template and the primers 5′-GGAATTCCATATGGAGAACACTGAAAACTCAGTG-3′ (forward) and 5′-CCGCTCGAGGTGATAAAAATAGAGTTCTTTTG-3′ (reverse). After digestion with Xho1 and Nde1, this PCR-generated cDNA was subcloned into pET21b at the Nde1/Xho1 sites to produce CPP32 protein with six histidine resides at its COOH terminus. Recombinant purified CPP32 was purified as described (39). New Zealand white female rabbits were injected subcutaneously with 200 μg of purified CPP32-His₆ fusion protein mixed (1:1 vol/vol) with Freund's complete adjuvant and then boosted seven times with 200 μg of protein in Freund's incomplete adjuvant before collecting blood and obtaining immune serum.

AIF-mediated cleavage of nuclear substrates was determined by the comparative analysis of SDS-PAGE of HeLa nuclei (5 × 10⁶/lane) cultured in the presence or absence of supernatant from Atr-treated mitochondria (10 μg protein/ml, 90 min, 37°C) in the presence or absence of Z-VAD.fmk, as described (11). Western blots of these nuclei were tested for degradation of poly (ADP-ribose) polymerase (PARP) using the monoclonal antibody C2-10 (purchased from Guy Poirier, Montreal University, Canada; reference 38). Cleavage of CPP32 in cells (8 × 10⁵ cells/lane) or in vitro (10 ng recombinant CPP32+ 10 μg protein of mitochondrial supernatant in 50 μl CFS buffer ± 100 μM Z-VAD.fmk, 15 min at 37°C) was determined by using a polyclonal rabbit antiserum recognizing both CPP32 and the p17 fragment of proteolytically activated CPP32 (39). Enzymatic activation of CPP32 (100 ng CPP32 + variable amounts of mitochondrial supernatant in 100 μl CFS buffer ± 100 μM Z-VAD.fmk, 15 min at 37°C) was detected by adding 1 μM Ac-DEVD-amino-4-methylcoumarin (30 min, 37°C), as described above. In one control experiment, Z-VAD.fmk (100 μM) was added together with Ac-DEVD-amino-4-methylcoumarin.

Human CEM-C7.H2 lymphoma cells can be induced to undergo apoptosis by cross-linking of Fas. As shown in Fig. 1,A, cells manifest a rapid activation of protease(s) capable of cleaving a fluorogenic substrate containing the tetrapeptide YVAD. As described (22, 25), activation of ICE-like proteases is a rapid process that peaks 15–30 min after Fas cross-linking. It thus precedes the Fas-induced ΔΨ_m disruption, as quantified by means of the ΔΨ_m-sensitive dye CMXRos. This ΔΨ_m collapse affects only a minor fraction of the cells beginning at 30 min after Fas ligation. An important fraction of cells (∼ 40%) exhibits a disrupted ΔΨ_m about 2 h after Fas cross-linking, when DEVDase activity is also significantly augmented. To further investigate the relationship between Fas-induced ΔΨ_m disruption and activation of CPP32, CEM-C7.H2 cells were stimulated during 2 h by Fas cross-linking, followed by staining with the ΔΨ_m-sensitive dye DiOC₆(3) as well as Annexin V (which measures the aberrant phosphatidyl serine exposure on the outer plasma membrane leaflet) and cytofluorometric purification of cells with a still normal ΔΨ_m (DiOC₆(3)^high) as well as cells with a disrupted ΔΨ_m (DiOC₆(3)^low) that are either in an early stage of the apoptotic process (Annexin V⁻) or in an advanced stage (Annexin V⁺) (Fig. 1,B). Only ΔΨ_m^low cells have cleaved the CPP32 precursor to yield CPP32 fragments (p21 and p17) and exhibit DEVDase activity (Fig. 1,B). This is observed for both ΔΨ_m^low Annexin V⁻ and ΔΨ_m^low Annexin V⁺ cells, indicating that CPP32/DEVDase activation occurs concomitant with (or shortly after) the ΔΨ_m disruption. In contrast, ΔΨ_m^high cells behave like unstimulated control cells and lack any detectable CPP32 cleavage or DEVDase activation (Fig. 1,B). Thus, CPP32 is only activated in cells whose ΔΨ_m is disrupted. Similar results have been obtained in other models of apoptosis induction, including ceramide-induced cell death (not shown). As in other models of apoptosis induction (4, 6, 9, 12, 31), the Fas-induced ΔΨ_m collapse precedes nuclear chromatinolysis as identified with the TUNEL technique (Fig. 1,A). Thus, cells that have disrupted their ΔΨ_m (CMXRos^low cells) can be subdivided into TUNEL⁺ and TUNEL⁻ populations, whereas TUNEL⁺ cells uniformly possess a ΔΨ_m^low (CMXRos^low) phenotype (Fig. 1 C). These findings place ΔΨ_m disruption upstream of nuclear apoptosis.

ICE-like proteases are involved in both the ΔΨ_m disruption and chromatinolysis, because cells treated with the ICE-specific inhibitor Ac-YVAD.cmk or cells stably transfected with the ICE inhibitor crmA fail to demonstrate mitochondrial or nuclear manifestations of apoptosis in response to Fas cross-linking (Fig. 1,C). Thus, in Fas-induced apoptosis, ICE (or ICE-like proteases) function upstream of mitochondria. Transfection-enforced hyperexpression of Bcl-2 does not interfere with Fas-triggered apoptotic changes, although it does prevent both the ΔΨ_m dissipation and DNA loss induced by the apoptosis-inducing second messenger ceramide (Fig. 1 C). This finding confirms previous observations that Bcl-2 does not prevent Fas-induced apoptosis, at least in certain experimental systems (26–28).

To determine the mechanism by which activation of ICE or ICE-like proteases causes ΔΨ_m disruption, we incubated isolated hepatocyte mitochondria with cytosolic extracts from αFas–treated cells. Cytosols from αFas-treated cells, but not cytosols from sham-treated control cells, were found to induce large amplitude swelling of isolated mitochondria (Fig. 2,A), a sign of PT. In addition, mitochondria treated with cytosols from αFas-treated cells manifest ΔΨ_m disruption, another sign of PT (Fig. 2,B). This PT-inducing activity was maximal in cytosols obtained from cells subjected to Fas cross-linking for 30 min (not shown), coinciding with the maximum activity of ICE (-like) proteases (Fig. 1,A). As expected based on the results in intact cells (Fig. 1,C), transfection with crmA impeded the cytosolic accumulation of such a PT-inducing activity (Fig. 2). In addition, the ICE-specific inhibitor Ac-YVAD. cmk and another inhibitor of ICE, Ac-YVAD.CHO (not shown), prevented the formation of the PT-inducing activity in cytosols from αFas-treated cells. Ac-YVAD.cmk also prevented the action of cytosols that already contained the PT-inducing activity on mitochondria, when added to the cytosol derived from αFas-treated cells in vitro (Fig. 2). In contrast, the peptide Ac-DEVD.CHO, an inhibitor of CPP32, was ineffective (not shown). This suggests that an ICE-like protease participates in the induction of mitochondrial PT, both in cells (Fig. 1,C) and in a cell-free system (Fig. 2).

In accordance with the data obtained with cytosolic extracts, recombinant purified ICE suffices to induce both large amplitude swelling and dissipation of the ΔΨ_m in isolated mitochondria in vitro (Fig. 3, A and B). This effect of ICE is rapid (<30 s) and can be neutralized by Ac-YVAD.cmk and Ac-YVAD.CHO, but not by AcDEVD.CHO, indicating that it relies on the enzymatic activity of ICE. In contrast, Ac-YVAD.cmk does not interfere with t-BHP-induced PT, thus excluding that this modified tetrapeptide might prevent PT in a nonspecific fashion (Fig. 3, A and B). The use of additional inhibitors underscores the different mechanisms involved in ICE– and t-BHP–triggered PT. For example, cyclosporin A, bongkrekic acid, monochlorobiman (9), and the calpain inhibitor Cbz-LLT. CHN₂ (17) all inhibited the t-BHP– but not the ICE–induced PT in vitro (Table 1). Other proteases besides ICE, such as trypsin and proteinase K, also induce PT in isolated mitochondria (not shown), in accord with previous observations that microinjection of such proteases induces apoptosis in cells (40).

Thus, ICE (or ICE-like) protease(s) is/are necessary and sufficient to mediate the ΔΨ_m disruption in cells subjected to Fas ligation. The mechanism of ICE-induced PT differs from that of prooxidant-induced PT, suggesting a direct proteolytic effect of ICE on mitochondria.

We have previously shown that PT is accompanied by the release of an apoptogenic protein that induces isolated nuclei to undergo chromatin condensation and oligonucleosomal DNA fragmentation (9, 11). Accordingly, the ICE-induced PT is accompanied by the release of such an apoptogenic protein (AIF), which causes purified HeLa nuclei to manifest DNA condensation and loss of chromosomal DNA (subdiploidy). The ICE inhibitor Ac-YVAD.cmk prevents the ICE-induced mitochondrial release of AIF (Fig. 3 C), yet does not interfere with the activity of AIF itself in the cell-free system of nuclear apoptosis induction (9, 11, and see below), consistent with the fact that ICE by itself is insufficient to induce apoptosis in isolated nuclei (38). In conclusion, ICE-induced PT is accompanied by the release of a mitochondrial apoptogenic factor.

As outlined in the Introduction, Bcl-2 is incapable of suppressing the Fas-induced apoptosis in a number of different models (26–28). This applies also to CEM-C7-H2 lymphoma cells (Fig. 1,C, see above). Since mitochondrial hyperexpression of Bcl-2 prevents the induction of PT by different substances, including prooxidants (9, 11; Fig. 4), we tested whether it would also interfere with ICE-induced PT. Mitochondria isolated from Bcl-2–transfected cells manifest large amplitude swelling when treated with recombinant ICE, exactly as do control mitochondria from cells not hyperexpressing Bcl-2. In addition, Bcl-2 hyperexpression does not prevent the mitochondrial release of AIF induced by ICE, although it does suppress the t-BHP–induced PT and release of AIF (Fig. 4). This dichotomy in the Bcl-2–mediated regulation of PT, inhibition of prooxidant-induced PT and failure to prevent ICE-induced PT, was observed in human CEM-C7-H2 cells transfected with tetracycline-repressable bcl-2 construct (Fig. 4), as well as in murine 2B4.11 T cell hybridoma cell lines stably transfected with the human bcl-2 gene (not shown). Thus, Bcl-2 fails to neutralize the effects of ICE on mitochondria in vitro, consistent with its inability to prevent ICE-dependent apoptosis in cells. Since Bcl-2 prevents ceramide-induced apoptosis and ΔΨ_m disruption (Fig. 1,C), we investigated the AIF release of Bcl-2 hyperexpressing mitochondria treated with cytosolic extracts from cells that have been treated during a short interval (30 min) with either ceramide or anti-Fas. Control mitochondria readily release AIF upon incubation with such cytosols (Fig. 5). Bcl-2–hyperexpressing mitochondria demonstrate a selective protection against yet unidentified ceramide-elicited PT inducers, yet release AIF upon incubation with ICE-containing cytosols from anti-Fas–treated cells (Fig. 5). These results are compatible with the hypothesis that Bcl-2 prevents ceramide-induced apoptosis at the level of mitochondria.

AIF is a preformed ∼50-kD intermembrane protein that is released from mitochondria undergoing PT (11). Isolated nuclei exposed to AIF exhibit a step-wise alteration in the morphology of the nucleus which consists in a first step of chromatin condensation (15 min), followed by disruption of the nuclear envelope and an associated loss of electron-dense material (60 min). At this latter stage, nuclei frequently demonstrate two homogeneous zones that differ in their electron density and resemble nuclei from cells at an advanced stage of apoptosis (Fig. 6,A). In addition, isolated nuclei exposed to AIF display two biochemical hallmarks of apoptosis: (a) loss of total nuclear DNA (hypoploidy) (11; and Fig. 6,B) and (b) oligonucleosomal DNA fragmentation (11; and Fig. 6,C). The apoptogenic effect of AIF liberated from ICE-treated mitochondria is neutralized by the broad spectrum inhibitor of ICE-like proteases Z-VAD. fmk, but not by the more specific ICE inhibitor Ac-YVAD.cmk (9, 11; and Fig. 6, B and C), in accord with the fact that ICE itself is not apoptogenic (38), whereas ICE-like proteases do have the capacity to provoke nuclear apoptosis in vitro (41). Since AIF apparently has an ICE-like catalytic activity (11), we tested whether AIF would share with ICE the PT-inducing potential. Indeed, the supernatant of ICE-treated mitochondria provokes PT of freshly isolated mitochondria (Fig. 6,D). This activity is not neutralized by Ac-YVAD.cmk, yet is completely abolished by Z-VAD.fmk, suggesting that it is mediated by AIF and not by residual ICE. Accordingly, AIF enriched on an anion exchange FPLC column (11) can induce PT in a Z-VAD.fmk-inhibitable fashion (Fig. 6 D). These data suggest that AIF possesses a biological spectrum of activities that overlaps, but is not identical, with that of ICE.

The proteolytical activation of CPP32 has been previously shown to be inhibited by Z-VAD.fmk (42), which also inhibits the effects of AIF (11; and Fig. 6). Stimulated by this fact, we have tested whether mitochondrial intermembrane fractions containing AIF would activate purified recombinant CPP32. As shown in Fig. 7,A, supernatants of mitochondria induced to undergo PT in response to Atr (9, 11) indeed activate the cleavage of a fluorogeneic peptide substrate containing the CPP32 cleavage site DEVD. This effect is inhibited by Z-VAD.fmk. Moreover, the same effects were observed with FPLC-purified AIF (Fig. 7,A). In addition, AIF (which is by itself incapable of cleaving the “death substrate” PARP) activates CPP32 to digest PARP (Fig. 7,B). As expected by the finding that AIF activates CPP32, CPP32 digestion by AIF-containing preparations yields a 21 precursor and a canonical p17 fragment (Fig. 7,C) that may associate with the p12 fragment to yield a biologically active heterotetramer (33, 43). This activation does not require the autocatalytic processing of CPP32, since it is not inhibited by Ac-DEVD.CHO (Fig. 7 C). In conclusion, AIF is endowed with the capacity of activating one of the signature processes of apoptosis, CPP32.

The data presented in this work provide multiple novel connections between proteases and mitochondrial PT during the apoptotic effector phase. These interactions are bidirectional. On the one hand, ICE can provoke mitochondrial PT, and, on the other hand, PT entails the mitochondrial release of a CPP32-activating protease.

As outlined in the Introduction, ΔΨ_m disruption constitutes an early event of apoptosis that precedes nuclear apoptosis. The apoptotic ΔΨ_m disruption involves opening PT pores on the inner mitochondrial membrane, based on the observation that PT pore antagonists such as bongkrekic acid inhibit the apoptotic ΔΨ_m loss (4, 9, 12). Abundant literature (for review see reference 16) indicates that numerous physiological effectors regulate PT: concentrations of divalent cations and protons, the redox status of mitochondrial thiols (in equilibrium with the redox status of glutathione), the redox status of the pyridine nucleotide pool (NADH₂/NAD + NADPH₂/NADP; reference 44), concentrations of adenine nucleotides (ADP, ATP), specific peptides, lipid acid oxidation products (16, 45), and proteases from the calpain family (17). Here we show that ICE (or ICE-like proteases) contained in the cytosol of Fas-activated cells, as well as recombinant purified ICE, are capable of inducing a PT-like effect in isolated mitochondria. ICE induces all three hallmarks of PT: (a) colloidosmotic swelling (Fig. 3,A), (b) disruption of the ΔΨ_m (Fig. 3,B), and (c) release of AIF (Fig. 3,C), which is self-sufficient to provoke nuclear apoptosis in a cell-free, cytosol-free system. In contrast with other methods of PT induction, ICE-mediated PT is not regulated by various pharmacological inhibitors of PT (e.g., monochlorobiman, bongkrekic acid; Table 1) and it is not inhibited by overexpression of Bcl-2 in the mitochondrial membrane (Fig. 4). Thus, it appears that the direct proteolytic effect of ICE on unidentified mitochondrial substrates provoke PT and a consequent ΔΨ_m collapse that disrupts mitochondrial functions. The ICE-induced PT is accompanied by AIF release from mitochondria, similar to PT induced by other compounds including calcium, pro-oxidants, or the thiol–cross-linking agent diamide (9).

It thus emerges that mitochondria function as a cellular sensor of stress including changes in redox potentials, direct oxidative effects, and protease activation. These data support PT as a candidate for the “central apoptotic executioner” that has been postulated by several groups (1–5) and that would allow for the convergence of very different apoptosis induction pathways into one event downstream of which would follow the final common pathway of apoptosis.

Irrespective of the PT-triggering stimulus, PT results in the mitochondrial release of an apoptogenic protease that we have termed AIF (9, 11). Although the molecular cloning of cDNA encoding AIF is still in progress, functional tests performed on purified AIF indicate that it possesses three unique features. First, AIF is the first protease that has been shown to suffice in inducing apoptotic changes in isolated nuclei (9, 11; and Fig. 6, A–C). Thus, at difference with another mitochondrial product, cytochrome c (10), AIF does not appear to require the presence of further cytosolic factors to induce nuclear apoptosis. We cannot exclude the possibility, however, that purified nuclei are associated with factors derived from cytosol that are necessary for AIF function. Second, AIF shares at least one biological effect of ICE, namely the capacity to trigger PT (Fig. 6,D). Thus, AIF liberated from mitochondria undergoing PT may engage in a self-amplifying apoptotic switch, and thus aid to lock the cell in an irreversible stage of apoptosis, beyond the point of no return. Since the effects of AIF are inhibited by a degenerate tripeptidic inhibitor of ICE and ICE-like proteases, Z-VAD.fmk (which acts as universal inhibitor of nuclear apoptosis in mammalian cells, perhaps with the exception of blastomeres; references 46, 47), this may explain why Z-VAD.fmk can inhibit ΔΨ_m disruption, at least in some systems of apoptosis induction (13). Third, AIF induces cleavage and activation of CPP32 in vitro (Fig. 7). This is a rapid process, with detectable CPP32 cleavage in as little as 5 min (not shown). CPP32 activation appears to be a consistent concomitant of the apoptotic process that may contribute to the apoptotic degradation of different cellular and nuclear substrates (33, 41, 43), including those that are not cleaved by AIF such as PARP (Fig. 7 B). This finding is in accord with the fact that the apoptosis-associated activation of CPP32 is inhibited by Z-VAD-fmk in vivo (39, 47). In conclusion, AIF has biological properties which render it a firm candidate to act as a central molecule of the apoptotic effector phase.

When integrated with the current literature, the data reported in this work suggest the following scenario for Fas-mediated apoptosis (Fig. 8). After Fas cross-linking, the Fas receptor complex rapidly (within seconds) recruits and causes the proteolytic activation of a protease (pro-FLICE/MACH1/ Mch5/caspase 8; references 20, 21, 48), which indirectly facilitates the activation of pro-ICE to activate ICE (peak: ∼30 min). Thereafter, ICE (or possibly ICE-like proteases) would cause mitochondrial PT (beginning at ∼60 min), which in turn would provoke the liberation of AIF from the mitochondrial intermembrane space. AIF then acts as an effector protease or protease activator and activates other downstream enzymes including CPP32. In this scheme, FLICE/MACH1 and ICE would act as “initiator” and “amplifier” proteases (49), within the private initiation phase of Fas-induced apoptosis. ICE and perhaps other ICE-like proteases would then induce mitochondrial PT, a process that causes the release of the “effector” protease/protease activator AIF from mitochondria, which in turn would contribute to further induction of mitochondrial PT. In cells in which Fas-induced apoptosis relies on ICE rather than on other pathways (e.g., ceramide), Bcl-2 would fail to impede the ICE-dependent induction of PT. Thus, PT and associated AIF release would constitute the first event of the common pathway of apoptosis and the central executioner of the effector phase. AIF release then would activate the “machinery” protease (49) CPP32 and perhaps other Ced-3–like proteases, which may participate in the degradation phase of apoptosis.

A possible critique against this sequence of events stems from the evidence that caspases can activate each other via direct interactions, at least in vitro (1–3, 5). Moreover, in some cell-free systems, proteases, in combination with yet unknown cytosolic factors, can provoke nuclear apoptosis (FLICE, ICE, CPP32; references 25, 50). Nonetheless, recombinant caspases including CPP32 do not induce nuclear DNA fragmentation in vitro on their own, in the absence of additional cytoplasmic extracts (33). Moreover, a putative direct caspase activation cascade fails to explain important facts such as the latency between YVADase and DEVDase activation (∼2 h) or the temporal sequence between YVADase activation, mitochondrial changes plus CPP32 activation, and late nuclear apoptosis, which is observed in intact cells and is mimicked by our cell-free system. In this context, it may be important to note that cytosolic extracts from cells which have been treated with apoptosis inducers (αFas, ceramide) for a short period (30 min) themselves are inefficient inducers to nuclear apoptosis in vitro, unless mitochondria are added into the system (Fig. 5). Thus, the cell-free system that we are using in this study confirms and extends the notion that mitochondrial products have a major, and perhaps essential, apoptogenic potential (7–11).

At first glance, our model may appear to be in contradiction with findings reported by Enari et al. (25) who attribute a decisive regulatory role to CPP32 in apoptosis regulation, based on the fact that preincubation of cells with the CPP32 inhibitor Ac-DEVD.CHO prevents Fas-induced apoptosis. However, addition of Ac-DEVD.CHO to cells after Fas cross-linking has no apoptosis-inhibitory effect (25), suggesting that Ac-DEVD.CHO acts on upstream proteases such as caspase 8, which cleaves the sequence motif DEVD (and thus is inhibited by Ac-DEVD. CHO; Reed, J.C., unpublished observation). Thus, the temporal and functional analysis of different proteases activated during Fas-induced apoptosis would suggest that CPP32 participates in the degradation, rather than in the execution, phase of apoptosis. Accordingly, addition of CPP32 inhibitors can suppress detectable DEVDase activation without affecting the mitochondrial phase of the apoptotic process (data not shown). As a caveat, this does not imply that CPP32 (and other closely related Ced-3 homologues) would only participate in the late phase of apoptosis. Indeed, it is conceivable that in response to other apoptosis inducers (e.g., developmentally programmed cell death), CPP32 may be involved in an earlier (private) step of the apoptotic cascade. It has been shown that CPP32 could activate unknown cytosolic factors (25) (which likely include mitochondrial products; reference 10) to become apoptogenic and thus to induce nuclear apoptosis in a cell-free system. These data suggest that CPP32 can activate other, soluble apoptogenic factors.

Irrespective of these possibilities, the results of this work suggest a unified view of protease-dependent and mitochondrial events of the apoptotic cascade. Proteases may have a major impact on mitochondrial function at the same time that mitochondria can release proteases and/or protease activators with apoptosis-inducing properties. The data reported here, therefore, provide new clues about the events that trigger the effector phase of apoptosis.

We thank Anita Diu-Hercend (Roussel Uclaf, Romainville, France) for critical comments and Dr. Maurice Geuskens (Université Libre de Bruxelles, Bruxelles, Belgium) for electron microscopy.

This work was supported by Agence Nationale pour la Recherche contre le SIDA, Association pour la Recherche contre le Cancer, Centre Nationale de la Recherche Scientifique, Fondation de France, Fondation pour la Recherche Médicale/Sidaction, Ligue Française contre le Cancer, Institute National de la Santé et de la Recherche Médicale, North Atlantic Treaty Organization, the French Ministry of Science (to G. Kroemer), and National Institutes of Health grant CA72994 (to J.C. Reed). S.A. Susin receives a postdoctoral fellowship from the European Commission.