Received 6/15/97 Accepted 6/28/97

Our molecular understanding of apoptosis has advanced profoundly since its original description by Wyllie and colleagues (1). While the first hallmarks of apoptotic cell death - membrane blebbing, chromatin condensation, and cellular fragmentation into 'apoptotic bodies' - were purely morphological, biochemical hallmarks have superseded these. Apoptosis is now studied as a cascade of proteases and endonucleases, where oligonucleosomal DNA laddering and cleavage of a variety of substrates by cysteine proteases have become the modern "gold-standards." Our knowledge of the executioner as well as some of its key modulators stems in large part from genetic studies of the nematode C. elegans.

Apoptosis is a developmentally programmed process in C. elegans whereby the death of individual cells is genetically determined and reproducibly observed. Mutants in which this pattern of programmed death is disturbed were described in the Horvitz laboratory at M.I.T., which led to the discovery of the cell death (ced) genes. Nematode mutants in the ced genes displayed a variety of phenotypes ranging from defects in engulfment by neighboring cells to excess death or survival (2, 3). In particular, the ced-3 and ced-4 genes were found to be required for the death of the normally occuring 131 programmed cell death events during C. elegans development. Gain-of-function mutations in the ced-9 gene suggested it may function to modulate the activity of ced-3 and ced-4 (4). Cloning of Interleukin 1-beta Converting Enzyme (ICE) protease as a mammalian homologue for CED-3 provided the first indication that proteases may play a critical role in apoptosis (5, 6). Following this observation, numerous ICE-family proteases have been identified in mammalian cells and are thought to constitute the core of the apoptosis executioner.

The ICE proteases all belong to the cysteine protease subfamily characterized by a cysteine residue at the active site. ICE/ced-3 homologues contain the conserved QACRG sequence surrounding the catalytic cysteine and show a preference for cleaving substrates after aspartate residues. Phylogenetic relationships among the proteases has led to their subdivision into three families (7). Typically, these enzymes can autocatalytically cleave and activate themselves as well as other ICE-family proteases. This may lead to amplification and diversification of available substrates during the execution phase of apoptosis in cells. The specific amino-acid sequence N-terminal to the target aspartate determines substrate preference among proteases of the ICE family: CPP32-like proteases cleave following DEVD, while ICE-like proteases cleave following YVAD. Specific peptide inhibitors which mimic cellular substrates containing these sequences have proved useful in studying the role and timing of activation of these proteases both in cells and in vitro. In the attempt to simplify the nomenclature for this ever-expanding family of enzymes, apoptosis proteases have recently been renamed 'caspases' (or cysteine proteases which cleave after aspartic acid), and numbered in function of the chronology of their discovery (8).

The ICE-family proteases known to date (over 10 members in mammals), strongly suggest that, unlike in C. elegans or other lower organisms, there may be considerable functional overlap within the death machinery of higher species. Alternatively, the wide variety of target substrates discovered so far in addition to in vivo observations afforded by the generation of knock-out mice, suggest that some proteases may have organ/tissue-specific expression patterns and/or cleave specific substrates within the cell. Mice deficient for CPP32, for example, display a brain-restricted apoptosis defect reminiscent of the neuron-specific phenotype of nematodes lacking ced-3 (9). Mice defective for ICE - the prototype of the apoptosis cysteine proteases - show no significant apoptosis defects during development or in response to apoptosis triggers such as ionizing radiation or dexamethazone, but are resistant to Fas-induced death (10, 11). While both these genetic studies point to the restricted tissue/pathway function of individual proteases in apoptosis, more extensive knock-out studies and crosses should reveal whether this trend holds true for other ICE-homologues.

One of the central questions remains how proteolysis leads to the demise of the cell and its stereotypical es. It is still unclear which substrates known to date (if any) are instrumental in the death pathway and which are simply biochemical markers of the process. Poly ADP-ribose polymerase (PARP) was one of the first proteins reported to be cleaved during apoptosis (12, 13), and is a target of the Yama/CPP32 protease, caspase-3 (14). PARP's role in recognizing DNA breaks along with the fact that CPP32-mediated cleavage separates its DNA-binding domain from its catalytic domain, has led to the notion that PARP inactivation may be functionally important for efficient DNA fragmentation to occur during late phases of apoptosis (15). However, mice lacking PARP show no obvious defects in apoptosis (16), suggesting that it is not required for apoptosis to occur, although functional homologues could in principle be rescuing PARP.

A wealth of substrates continue to be identified as potentially important signaling intermediates. These include, among others, the nuclear lamins (17-19), the 70kD component of the U1 snRNP (20, 21), DNA-PK (22), Gas2 (23), D4-GDI (24), PKC-delta (25), SREBPs (26), Huntingtin (27), PAK2 (28) and RB (29,30). Biochemical isolation of other relevant substrates based on their activity to signal apoptosis using in vitro systems should provide a clearer picture of the nuclear events downstream of the cysteine proteases. A number of cell-free systems like chicken S/M extracts (13, 31), HeLa cell extracts (32), Xenopus extracts (33, 34), as well as extracts of other cells triggered to die in response to a variety of apoptotic triggers (35-37) have already been used to reconstitute portions of the apoptosis pathway in vitro and characterize novel apoptosis-inducing activities. One such activity was recently isolated from HeLa cell extracts and termed the DNA fragmentation factor or DFF (38). DFF is cleaved both in vitro (following addition of recombinant CPP32 to extracts) and in vivo (in U937 cells treated with Staurospaurine), and is required for chromatin condensation and DNA laddering to occur. Although the mechanism by which DFF induces DNA fragmentation remains unknown, this new protein is one of the first ICE-family protease substrate to date with a demonstrated role in signaling downstream events of apoptosis.

Bcl-2 is the mammalian homologue of ced-9 which, in C. elegans, is required to protect cells that normally survive from undergoing programmed cell death (4, 39). It has been the focus of intense study ever since its demonstrated ability to rescue pre-B cells from apoptosis in response to IL-3 withdrawal (40). Originally identified as a result of the t(14;18) translocation in B-cell follicular lymphoma in which its juxtaposition with the IgH enhancer leads to dysregulated overexpression, Bcl-2's causal role in tumorigenesis was directly confirmed in studies where mice expressing a Bcl-2 transgene in lymphoid cells developed B-cell malignancies (41-43). Bcl-2 is now the prototype of a rapidly growing family of interacting proteins which share its ability to modulate apoptosis. A prevalent model has emerged by which a cell's threshold to apoptosis is determined by the levels of pro- and anti-apoptotic members which, through dimerization, act as a survival "rheostat switch" (44). The biochemistry through which Bcl-2 and its 'partners-in-crime' mediate their effects has only recently started to unfold.

Bcl-2 family proteins share several homology regions (Bcl-2-homology/BH domains) crucial for both their dimerization and apoptosis-modulatory functions (44). The BH domains along with the use of the yeast two-hybrid and other methods have been instrumental in the identification of novel interacting proteins such as Bax (45), Bcl-x (46), Bad (47), Bag-1 (48), Bak (49-51), Bik (52), Hrk/harakiri (53) and others. The pro-apoptotic Bax protein, cloned through its ability to co-immunoprecipitate with Bcl-2, was originally found to form homodimers as well as heterodimers with Bcl-2 via the BH1 and BH2 domains (45). Site-directed mutagenesis of these domains in Bcl-2 prevent heterodimer formation with Bax (or Bak) and abrogate its death-repressor activity (54). Interestingly, Bcl-2 mutations permissive for homodimerization, but not heterodimerization with Bax, also abolished Bcl-2 activity suggesting that Bcl-2 requires Bax to exert its death-repressor activity.

It remains unclear, however, who among the pro-and anti-apoptotic members are the key effectors in the apoptosis pathway. Indeed, unlike Bcl-2, not all members appear to require heterodimerization to carry out their protective effect. The Bcl-x transcript is alternatively spliced into a short and long form to yield both pro- and anti-apoptotic factors (Bcl-xS & Bcl-xL respectively) (46). While the long form resembles Bcl-2 in its dimerization preferences and anti-apoptotic capacity, Bcl-xL mutants which prevent its interaction with Bax or Bak can still rescue Sindbis virus-induced apoptosis (55). The short splice form of Bcl-x, although lacking BH1 and BH2 dimerization regions, appears to mediate its death-inducing function by antagonizing Bcl-2 and Bcl-xL (46). A simple 'rheostat switch' model whereby both pro- and anti-death molecules require co-interaction to mediate their effects on apoptosis is not sufficient to explain the above observations. Alternatively, death-promoting and death-suppressing Bcl-2 family proteins may operate in mechanistically different ways. The phenotype of Bax knock-out mice which display both an excess and lack of cell death in different tissues supports this possibility (56).

A third region of Bcl-2 homology - BH3 - has been demonstrated to be essential for the activity of the pro-apoptotic proteins. Indeed, although all three BH1, 2, & 3 regions are involved in dimer interface as revealed by structural studies (57), expression of the BH3 domain of Bak alone is sufficient to induce apoptosis (58). A crystal structure of Bcl-xL bound to a Bak peptide derived from the BH3 region demonstrates the importance of this domain in mediating protein-protein interaction (59). Bcl-xL BH1, BH2, and BH3 domains form an elongated hydrophobic cleft as revealed by NMR and X-ray structures (57). When bound with Bcl-xL, the BH3 region of Bak adopts an amphipathic alpha helix and mediates both hydrophobic and electrostatic interactions with the Bcl-xL hydrophobic cleft. Mutations in residues observed to form specific contacts within this cleft abolish interaction of the two proteins (59).

Structure determination of Bcl-xL, in addition to revealing the overall architecture of the BH dimerization domains, has also revealed a somewhat unexpected structural similarity of this protein to the membrane insertion domain of the diphtheria toxin and related colicins (60). Indeed, comparing Bcl-xL (as well as other members like Bcl-2) with the diphtheria toxin reveals that both proteins have two central helices with apolar residues (alpha5 and alpha6 in Bcl-xL) surrounded by three amphipathic helices (alpha1, alpha3, and , alpha4 in Bcl-xL)(57). The two central hydrophobic helices alpha5 and alpha6 of Bcl-xL are long enough to span the length of the membrane, and, like the insertional domain of the diphtheria toxin, may require dimerization to form a pH-dependent membrane pore (61). By analogy with the functional role of the insertional domain in toxins, these observations raise the possibility that proteins of the Bcl-2 family may control the passage of ions or other components across the mitochondrial membrane. Alternatively, Bcl-2 family proteins may interact with and regulate the activity of other mitochondrial transport proteins via these helical domains. A recent report that Bcl-xL forms pH-sensitive ion-conducting channels in synthetic lipid membranes has reinforced this hypothesis (62).

Until recently, Bcl-2's ability to regulate the apoptosis pathway remained a mechanistic enigma. Based on observations linking Bcl-2-mediated rescue of apoptosis with lower levels of oxygen free radicals as well as Bcl-2's localization to nuclear, ER and mitochondrial membranes, early studies advanced the possibility that Bcl-2 could function in an antioxidant pathway (63, 64). Bcl-2 and Bcl-xL can prevent hypoxia-induced cell death, however, suggesting that they exert their anti-apoptotic activity by a mechanism other than modulation of oxygen free radicals (65, 66). Based on cellular fractionation studies as well as electron microscopic observation, others hypothesized that Bcl-2 may be involved in nuclear transport (67). The more recent discovery of the extended family of Bcl-2 dimerization partners led to a model whereby ratios of these molecules may determine a cell's susceptibility to apoptosis in response to a variety of apoptotic stimuli (44). While these observations suggest how regulation may occur within the family, they give no hint as to how these proteins are regulated by exterior signals as well as a possible connection with the central apoptosis machinery. In addition to their putative role as channel-forming assemblies, Bcl-2 family proteins have now been found to recruit and interact with several signaling proteins which may regulate their activities.

Although in many cases levels of pro-and anti-apoptotic Bcl-2 family proteins determine a cell's susceptibility to apoptosis, there exist several demonstrated instances where expression levels do not correlate with apoptosis propensity (68, 69). Post-translational modification has been put forth as a possible mechanism for modulating the death activities of these proteins. Recent evidence suggests that trophic factor-induced phosphorylation of both Bcl-2 and Bad may regulate their apoptotic functions. For both Bcl-2 and Bad, phosphorylation has been suggested to have an inhibitory effect. Bcl-2 phosphorylation on serine occurs in response to a variety of stimuli and often correlates with decreased anti-apoptotic potential (70, 71). Bcl-2 has also been proposed to recruit the Raf-1 kinase to outer mitochondrial membranes and induce phosphorylation of Bad (72). This phosphorylation triggers Bad's dissociation from Bcl-xL and cytoplasmic association with the 14-3-3 protein (73). While the effect of these modifications on the apoptosis functions of Bcl-xL or Bcl-2 are not yet fully understood, they may provide a link between growth factor pathways and regulation of the downstream apoptosis machinery.

An interaction between Bcl-2 family proteins, members of the ICE-family cysteine proteases, and the nematode protein CED-4 has recently been demonstrated by a number of groups (74-76). Considering the epistatic relationship of these genes in C. elegans whereby ced-9 rescues cells from apoptosis by antagonizing the activity of ced-3 and ced-4 (4), a direct physical interaction between these proteins was tested. CED-9 appears capable of binding the death-inducing splice form of CED-4 (77) using the yeast two-hybrid system and in vitro binding studies (75, 76). Mutations in CED-9 which eliminate its anti-apoptotic activity were found to disrupt its ability to interact with CED-4. These studies point to the importance of CED-4 binding for CED-9 function, and raise the possibility that the CED-4 interaction domain of CED-9 could potentially be used as a bait for identifying mammalian CED-4 homologues.

A key observation which allowed for a CED-4/Bcl-xL/ICE interaction to be tested in a mammalian tissue culture system was that the nematode CED-4 protein is sufficient to induce apoptosis when overexpressed in mammalian cells (74). In addition, CED-4-induced apoptosis can be inhibited with Bcl-xL or caspase inhibitors suggesting that in mammalian cells, CED-4, under control of the Bcl-2 family, induces death via ICE-family protease activation. Once this was established, Chinnaiyan et al. took advantage of mutant proteins with the binding but not the cytotoxic properties of wild-type molecules in order to overexpress and immunoprecipitate CED-4 without killing cells. They observed that wild-type CED-9 (or Bcl-xL) coimmunoprecipitates with CED-4 when cotransfected into 293 cells. In agreement with the yeast two-hybrid studies, immunoprecipitation of CED-4 failed to bring down a CED-9 mutant (lacking BH1 and BH2 domains) which does not block CED-4-induced apoptosis. Similarly, the ability of Bcl-xL mutants to inhibit apoptosis correlated with their capacity to interact with CED-4. Coexpression of pro-apoptotic Bcl-2-family members such as BAX, BAK or BIK disrupted the CED-4/Bcl-xL interaction.

In addition to its ability to interact with certain anti-apoptotic Bcl-2 family proteins, CED-4 was also observed to immunoprecipitate with ICE family-proteases which contain a large pro-domain (such as ICE or FLICE). Although the functional significance of this binding as well as the mechanism through which CED-4/ICE interaction may activate the protease, these results place CED-4 as a central regulatory switch between the Bcl-2 family and the downstream executioner machinery. The importance of CED-4 as an adaptor between these two families is manifest in the fact that a CED-4 mutant which interacts with CED-3 (but not CED-9) reduces Bcl-xL's ability to coimmunoprecipitate with ICE or FLICE. The fact that CED-4 has no known mammalian homologue explains why the link between these two central apoptosis families has remained elusive until now. The key was to test the activity of the nematode protein in the context of a mammalian cell culture system. Given its central role, CED-4 may present a new therapeutic target for intervention in cancer, for example, where some upstream signaling components of the apoptosis machinery are often disabled.

Finally, an additional signaling activity has recently been attributed to the Bcl-2 family, and involves the release from mitochondria of proteins which may be instrumental in downstream apoptosis events. Several studies have shown that cytochrome c is released from mitochondrial stores upon induction of apoptosis, and that Bcl-2 overexpression is able to prevent this release (78, 79). Interestingly, cytochrome c was previously implicated in signaling apoptosis through its ability, in combination with dATP, to trigger PARP cleavage in naive HeLa cell extracts (32). Research also indicates that Bcl-2 may exert its anti-apoptotic activity by blocking the release of other death signaling molecules from mitochondria such as an as-of-yet unidentified apoptogenic protease (80). Whatever normally causes the release of either cytochrome c or other signaling entities from mitochondrial stores upon induction of apoptosis, as well as the mechanisms through which these molecules activate the executioner remain to be discovered.

In conclusion, current data point in several directions for a functional role of the Bcl-2 family in apoptosis. Anti-apoptotic members of the bcl-2 family may exert their death-suppressing function through their ability to bind and sequester a CED-4 homologue in mammals. Binding of CED-4 by these proteins presumably interferes with its ability to bind and activate CED-3 homologues like ICE or FLICE. Through dimerization with Bcl-2 or Bcl-xL, pro-apoptotic Bcl-2 proteins may, in turn, disrupt their ability to bind CED-4, thereby freeing it for interaction with executioner proteases. Such a model offers an attractive signaling pathway which provides a direct biochemical link between the Bcl-2 family of apoptosis modulators with the cysteine proteases, and fits well with the epistatic relationships described in nematodes. Nevertheless, while the interaction with CED-4, recruitment of Raf-1, formation of ion-channels, and ability to release cytochrome c (and perhaps other apoptosis effectors) from mitochondria are all reported activities of Bcl-2 family proteins, it remains unclear how these diverse activities converge to modulate apoptosis.

Cytokine-induced apoptosis by Fas or tumor necrosis factor (TNF) is now one of the best studied and understood biochemical pathways to cell death which directly links ligand-receptor interactions at the plasma membrane with components of the executioner machinery. During development of the immune system, over 90% of immature thymocytes are deleted as a result of unproductive T cell receptor rearrangements or because they become self-reactive as a result of this rearrangement. Selection can occur in the thymus, in the peripheral circulation, or following the immune response (in the case of mature T cells). Based on the phenotypes of lpr (lymphoproliferation) and gld (generalized lymphoproliferative disease) mice which harbor defects in the Fas and Fas ligand (FasL) genes respectively (81), Fas-induced apoptosis is thought to be involved in at least two of these processes: peripheral clonal deletion and downregulation following immune reaction (82, 83). In addition, the Fas pathway has been linked to several types of autoimmune disease such as ALPS (autoimmune lymphoproliferative syndrome) and autoimmune diabetes (84-87).

FasL is a member of the tumor necrosis factor (TNF) family of membrane and secreted proteins which is displayed on T cells following T cell receptor signaling (88-90). Fas (also referred to as APO-1 or CD95) is a receptor of the TNFR family of cell surface proteins (91, 92). While Fas and TNF receptors are abundantly expressed in a wide variety of tissues (thymus, liver, heart, kidney), FasL and TNF expression patterns (mostly in natural killer cells and activated lymphocytes) appear to be a more relevant indication of where Fas-mediated cell death occurs. Both FasL and TNF undergo metalloprotease-mediated cleavage from the membrane and therefore also exist as soluble forms. It remains unclear whether this cleavage extends their signaling range or whether the shedding of these molecules is a means of downregulating their activity (82). Until recently, the signaling mechanisms by which ligand/receptor interaction resulted in apoptosis remained unknown. The identification of the death domain dimerization motif in the cytoplasmic tails of Fas and TNFR has led to the rapid isolation of a number of crucial intracellular intermediates in the Fas pathway.

The death domain (DD) was originally recognized by homology between Fas and TNFR1. Mutagenesis studies of these receptors revealed this region of approximately 80 residues to be crucial for FasL/TNF-induced apoptosis (93, 94). Several groups identified proteins interacting with Fas and TNFR using the DD as a yeast two-hybrid bait: FADD/MORT1 (Fas-associated protein with death domain), TRADD (TNFR1-associated death domain protein), and RIP (receptor interacting protein) (95-99). The recent NMR structure of the DD reveals the biochemical basis for its ability to oligomerize and mediate signaling (100). In fact, the DD may be somewhat of a misnomer since it is found in a number of receptors not all of which appear to be involved in signaling apoptosis (101). In addition to this domain, the recognition of a death effector domain (DED) in FADD led to the isolation of the FLICE (FADD-like ICE)/MACH (MORT1-associated CED-3 homologue) protein (102, 103). Activation of Fas or TNFR1 is thought to result in the formation of a 'death-inducing signaling complex' (DISC) which leads to interaction between these proteins based on DD and DED interactions. In addition, the recently identified DR3/WSL-1/Apo-3 receptor homologous to TNFR1 and capable of inducing apoptosis via its interaction with TRADD/RIP raises the possibility that other receptors also mediate cytokine-induced cell death (104-106).

A current model for Fas-induced apoptosis involves trimerization of the receptor upon binding a FasL trimer followed by FADD/MORT1 recruitment to the receptor complex via death domain interaction. Binding of the FLICE/MACH protease to FADD/MORT1 via its death effector domain leads to its activation which, in turn, is likely followed by proteolytic activation of other ICE family members. The recent discovery of a new family of viral FLICE-inhibitory proteins (v-FLIPs) found in several gamma-herpesvirus and the tumorigenic human molluscipoxvirus points to the importance of FLICE as a central target for inhibiting the Fas/TNFR1 apoptosis pathway (107). By interacting with FADD via their DED, these viral proteins prevent the FADD-dependent recruitment and activation of FLICE, and completely block apoptosis induced via Fas, TRAMP or the as-of-yet unidentified TRAIL receptor (108, 109).

In addition to its role in apoptosis, TNF also induces activation of the NF-kappaB transcription factor. TNFR1 signaling is not as clear and well understood as that of Fas. Following TNF receptor trimerization, both TRADD and FADD/MORT1 can be recruited. It is at this level that the pathways of apoptosis and NF-kappaB activation bifurcate, since TRADD has been shown to have both apoptosis and NF-kappaB signaling properties (97). TRADD interacts with the RIP serine/threonine kinase to induce apoptosis, as well as with members of the TRAF (TNF receptor-associated factor) family to mediate NF-kappaB activation (110). While RIP does not contain a DED, it is its DD which is required to induce apoptosis suggesting an interaction with another signaling molecule. Such a candidate molecule was recently identified as RAIDD (RIP-associated ICH-1/CED-3-homologous protein with a death domain), which in addition to interacting with RIP through its DD, contains a sequence homologous to the 'pro' domain of the ICH-1/CED-3 proteases (111). This region of homology mediates RAIDD's interaction with these proteases and likely leads to their activation. Therefore, at least two potential signaling links have been made between Fas/TNFR1 and the downstream executioner machinery: one via FADD/MORT1 and involving the FLICE/MACH protease, and the other via TRADD, RIP, and RAIDD to activate ICH-1.

Several important questions regarding the Fas/TNFR signaling remain unanswered. How is the decision between FADD/TRADD-mediated death and TRADD-mediated NF-kappaB activation made given that certain TRAFs (TRAF-2 for example) interact with RIP as well as TRADD (99)? Why have cells evolved ways of controlling these two pathways via the same receptor complex? Studies suggest that NF-kappaB activation likely results in the expression of a 'survival gene' since inactivation of this pathway (with IkappaB phosphorylation-defective mutants for example) leads to increased sensitivity to Fas and TNF (112, 113). The potential for modulating apoptosis in this fashion may extend to cancer therapy (114). The ability of NF-kappaB-induced gene expression to antagonize apoptosis induced by other triggers remains to be fully investigated.

Several studies have also demonstrated the requirement for ceramide in TNF-mediated cell killing, suggesting that the signaling pathways to the executioner may not be as direct as anticipated. Ceramide is a lipid second messenger derived from the sphingomyelin cycle, and has been implicated in mediating apoptosis in response to a number of different triggers like radiation (115), chemotherapeutics (116, 117) as well as Fas and TNF (118, 119). Ceramide may nevertheless be a downstream signal in the Fas/TNF pathway since ICE-like activity appears to be necessary for ceramide generation in response to the Drosophila REAPER protein which, like Fas and TNFR, contains a death domain (101, 120). Similarly, an ICE-like activity has been suggested to be required for activation of the p38 stress activated kinase pathway by Fas (121).

Anther mystery is the mechanism through which FLICE becomes activated as well as its relevant substrates. Considering that other ICE-family proteases are typically activated following cleavage of their 'pro' domain (122), FLICE activation may similarly involve such a modification or cleavage reaction. The ICE protease cascade activated downstream of FLICE also remains poorly understood. The timing of CPP32-like and ICE-like protease activities following Fas ligation has been investigated using specific fluorescent substrates. These studies demonstrate that an ICE-like activity is transiently activated, whereas the CPP32-like activity accumulates during FasL/TNF-induced apoptosis (123). Nevertheless, it is likely that FLICE activation is still an early event in the pathway to irreversible death, given the ability of inhibitors such as Bcl-2 and Bcl-xL to block Fas-induced death both in vitro and in vivo (124-126). In keeping with this notion and the recent emphasis on mitochondrial events in apoptosis signaling, cytochrome c function appears to be impaired in Jurkat cells undergoing Fas-induced death (127). Taken together these observations suggest that Fas/TNFR1-induced apoptosis signals may converge with other classical apoptosis pathways on a conserved executioner machinery.

During development of the nervous system, excess neurons are produced only to subsequently die by competing for a limited supply of secreted or membrane-bound neurotrophic substances. The 'neurotrophic hypothesis' was validated in the 1950s with the discovery of nerve-growth factor (NGF) - a polypeptide shown to be important for the survival of sensory and sympathetic neurons in vivo - and, more recently, with the identification of other neurotrophins such as BDNF, CT-1, NT-3 and CNTF (128). In vitro studies showed that PC12 pheochromocytoma cells required NGF for both differentiation and survival. While Ras/Raf/Erk was required for differentiation, PI3-K was found to be crucial for the survival effect of NGF. Indeed, the ability of NGF to prevent apoptosis could be blocked by PI3-K inhibitors wortmannin and LY294002 (129). Treatment of cells ectopically expressing the PDGF-R with PDGF induced PI3-K and protected against apoptosis while PDGF receptor mutants unable to activate PI3-K had no effect. With the recent discovery of the Akt kinase as a dowstream PI3-K substrate with anti-apoptotic activity, we are now one step closer to understanding how PI3-K is linked to the apoptosis machinery.

The protooncogene Akt (also referred to as protein kinase B) is a ubiquitously expressed serine/threonine kinase whose catalytic domain is homologous to that of PKA and PKC family kinases, and which contains an N-terminal Pleckstrin homology (PH) domain. Akt kinase activity is upregulated in the presence of serum in a variety of cell lines in a manner dependent on PI3-K. This was originally shown by Franke et al. using a PDGFR-negative cell line into which PDGFR mutants lacking PI3-K binding sites (Y740F & Y751F) were transfected. Alternatively, treatment of cells with the PI3-K inhibitor wortmannin also blocked Akt kinase activity (130). In addition, Akt activation was abrogated by point mutations in the PH domain, suggesting the importance of this domain in regulation of the kinase. These results along with those showing that PI3-K's ability to rescue cells from serum deprivation-induced apoptosis was independent of p70S6K (131) pointed to Akt as a likely candidate. Shortly thereafter it was discovered that IGF-1 or insulin-mediated survival of cerebellar cultures in vitro is dependent on Akt (132). Cells transfected with dominant-negative constructs of the kinase - PH domain alone or kinase-dead versions of the protein - induce apoptosis of neurons in the presence of growth factors. Conversely, neurons into which wt-Akt was overexpressed were resistant to serum-deprivation suggesting the role of this kinase in cell survival.

The ability of PI3-K and Akt to promote survival is not limited to cells of neuronal origin, however, as it was found that either activated PI3-K or Akt protects Myc-overexpressing Rat1 cells from undergoing apoptosis upon growth factor withdrawal. Indeed, as serum deprivation is not only an apoptotic trigger for neurons but also for transformation-selective apoptosis in fibroblasts (133), the ability of PI3-K/Akt pathway to rescue factor-dependent survival in these cells was tested by several groups (134, 135). Transfection of a constitutively active mutant of the p110 catalytic subunit of PI3K was found to block apoptosis induced by serum starvation in Myc-Rat-1 cells. V12 Ras's ability to rescue cells was also tested as the p110 point-mutant used in these experiments (p110 K227E) is thought to mimic the conformational e induced by binding of PI3-K to activated Ras. V12 Ras, however, enhanced rather than suppressed serum-deprivation induced death suggesting that Ras could mediate both pro-and anti-apoptotic signals. Using Ras mutants which differ in their ability to activate downstream effectors like Raf versus PI-3K, Kauffmann-Zeh et al. demonstrated that interaction with Raf promotes apoptosis while interaction with PI3-K is necessary for survival (135). In addition, Akt was necessary to mediate this effect as shown by the ability of dominant-negative (kinase-dead) Akt constructs to promote apoptosis.

These results have been confirmed in careful studies by Kennedy et al. demonstrating the ability of dominant active forms of Akt to protect cells from serum withdrawal-induced death (134). In the attempt to understand how Akt promotes survival, levels of Bcl-2 and Bcl-xL proteins were tested and found to remain uned in Akt-overexpressing cells. Interestingly, however, while Bcl-2 can overcome apoptosis induced by Wortmannin, dominant negative (kinase-dead) Akt antagonizes the ability of Bcl-2 to rescue cells in low serum, suggesting an interplay between these proteins in preventing apoptosis. Perhaps not surprisingly, inhibition of PARP cleavage activity was shown to correlate with Akt-induced survival of these cells, establishing a role for Akt kinase upstream of apoptosis proteases with CPP32-like activity.

Additional pathways which could activate Akt in PI3-K-independent ways as well as Akt substrates important for preventing cell death remain to be discovered. So far, Akt is thought to be regulated via the action of at least two pathways - one involving PI3-K and the other dependent on an Akt-kinase (136). PI3-K regulates Akt by controlling phosphatidylinositol-3,4,5-P3 (PtdIns-3,4,5-P3) and phosphatidylinositol-3,4-P2 (PtdIns-3,4-P2) levels. Akt homo-oligomerizes upon binding PtdIns-3,4-P2 via its PH domain. This interaction, however, is not sufficient for full activation which requires additional phosphorylation in its catalytic loop (threonine 308) and on its C-terminal regulatory domain (serine 473). Since Akt is recruited to the plasma membrane through binding of its PH domain to PtdIns-3,4-P2, it may thus be brought into proximity of such a kinase. Although both MAPKAP-2 and the recently purified PDK1 kinase (activated by either PtdIns-3,4,5-P3 or PtdIns-3,4-P2) have been shown to phosphorylate Akt on its catalytic loop in vitro, it remains to be seen whether phosphorylation by these kinases plays a significant role in vivo (137). The role of the downstream GSK-3 in mediating Akt's apoptosis protective effects also remains unclear at the time. It will be interesting to see what links exist between the central apoptosis machinery and the Akt pathway, and whether Akt can protect cells from apoptosis induced by a variety of other triggers such as UV or ionizing radiation, although it has already been shown to be important in apoptosis upon loss of matrix adhesion (138).