Suicidal Tendencies: Apoptotic Cell Death by Caspase Family Proteinases*

Certain proteases are not merely degradative enzymes but are highly regulated signaling molecules that control critical biological processes via specific limited proteolysis. Caspase proteinases and their central role in apoptotic cell death provide a prime example of this concept. These cysteine proteinases exist as latent zymogens; however, once activated by apoptotic signals, they systematically dismantle and package the cell by cleaving key cellular proteins solely after aspartate residues. Here we review caspase proteinases with an emphasis on their structure, activation, and critical role in the apoptotic mechanism.

taining one or two aspartate cleavage sites separates the large and small subunits. Activation accompanies proteolysis of the interdomain linker and usually results in subsequent removal of the prodomain. The active enzymes function as tetramers, consisting of two large/small subunit heterodimers (17)(18)(19)(20). The heterodimers each contain an active site composed of residues from both the small and large subunits. Each active site contains a positively charged S 1 subsite that binds the substrate's negatively charged P 1 aspartate (17)(18)(19)(20). This S 1 binding site is highly conserved; therefore, all caspases cleave solely after aspartate residues.
The individual caspases have two major structural differences. First, the predicted S 2 -S 4 substrate binding sites vary significantly, resulting in varied substrate specificity in the P 2 -P 4 positions, despite an absolute requirement for aspartate in the P 1 position (4 -8, 17-20). Thornberry et al. (21) recently defined the optimal tetrapeptide substrate specificity for 10 caspases using a synthetic combinatorial peptide library. The sequence preferences generally correlate with caspase function as apoptotic initiators, apoptotic executioners, and cytokine processors ( Table I). Note that the tetrapeptide preferences listed in Table I are not absolute and that the preferences do not represent kinetic values that can be directly compared. For example, caspase-3 and caspase-7 both prefer DEXD-based peptides; however, the kinetics of the individual hydrolysis reactions may differ significantly.

Caspase Activation
Because caspases exist as latent zymogens, the question remains as to how the zymogens are activated. Current evidence suggests that activation may proceed by autoactivation, transactivation, or proteolysis by other proteinases.
Adapter molecules link apoptotic sensors such as death receptors and mitochondria to procaspases. To accomplish this, adapters generally contain one domain that couples the adapter to the sensor and another that binds to long prodomain procaspases. These domains include death domains (DDs), 2 death effector domains (DEDs), and caspase recruitment domains (CARDs) ( Table I). DDs, DEDs, and CARDs all contain six anti-parallel ␣-helices arranged in a similar three-dimensional fold and associate via like-like interactions (31)(32)(33). However, hydrophobic interactions are important for DED-DED interactions, whereas electrostatic interactions are critical for CARD-CARD interactions (31)(32)(33).
Mitochondria sense apoptotic signals and convey them to the activation adapter APAF-1 via the release of cytochrome c. Cytochrome c binds to APAF-1, and in the presence of adenine nucleotides, the APAF-1-cytochrome c complex promotes activation of procaspase-9 (38). Cytochrome c and adenine nucleotides likely induce a conformational change that exposes the APAF-1 CARD domain. The exposed APAF-1 CARD domain can in turn recruit procaspase-9 by a homophilic interaction involving CARDs. Subsequent procaspase-9 oligomerization then facilitates caspase autoactivation. Bcl-X L , an anti-apoptotic Bcl-2 family protein, may inhibit apoptosis by blocking these interactions (39). A second CARDcontaining adapter, RAIDD, couples procaspase-2 to death receptors via CARD-CARD interactions (40,41). Thus, adaptermediated protein-protein interactions are widespread among the apoptotic caspases.
Less is known about activation of pro-inflammatory caspases. CARDIAK, a CARD-containing kinase, promotes procaspase-1 activation in vitro via a CARD-CARD interaction, suggesting that CARD-mediated oligomerization could function in procaspase-1 activation (42). Ligation of the CD40 receptor also promotes procaspase-1 activation (43); however, whether CARDIAK facilitates this activation is unknown. Finally, caspase-11 does not directly process procaspase-1 but may facilitate zymogen activation by a non-proteolytic interaction (16).
FIG. 1. Caspase activation mechanisms. Sensors convey activation signals to adapters that facilitate oligomerization and subsequent autoactivation of long prodomain caspases (1). Once activated, caspases transactivate other procaspases (2). Direct proteolysis by non-caspase proteinases may also promote procaspase activation (3). See the text for details. Once activated, caspases transactivate other procaspases, providing the opportunity for cascade amplification and positive feedback. Caspase-8 for example efficiently activates procaspase-3 (k cat /K m ϭ 8.7 ϫ 10 5 M -1 s -1 ) (44), and active caspase-3 in turn may activate procaspase-8. Although this positive feedback loop is theoretically possible, it has not yet been demonstrated. Additionally, because caspases have varied substrate specificity, a single activated caspase may not directly activate all other family members. For example, caspase-9 activates procaspase-3 and procaspase-7 but cannot activate procaspase-6 (29). Propagation of a caspase cascade will thus depend on which caspases a cell expresses, the relative concentrations of each caspase, and the kinetic efficiency of the individual transactivation reactions.

Caspases and the Apoptotic Mechanism
Apoptotic cells die a stereotypical death, regardless of the initiating death signal (53). The cytoplasm shrinks, the plasma membrane blebs and vesiculates, and phosphatidylserine redistributes to the cell surface. Simultaneously, the nucleus shrinks, chromatin condenses, and DNA fragments into high molecular weight and oligonucleosomal pieces. Identification of caspase substrates has provided insight into how caspases promote these changes. Table II groups caspase substrates into five categories based on putative function and lists examples for each category. More extensive substrate lists are available elsewhere (7,8,54).
Pro-and anti-apoptotic proteins are probably the first caspase targets following an apoptotic stimulus (Table II) (7, 8, 54). These cleavage events participate in signal amplification and inhibitor inactivation. For example, transactivation of procaspases could generate sufficient proteolytic activity to overwhelm endogenous caspase inhibitors such as inhibitor of apoptosis proteins. Caspase-3 cleaves Bcl-2 and Bcl-X L , which destroys the anti-apoptotic function of these proteins and releases C-terminal fragments that are pro-apoptotic. Similarly, caspase-8 cleaves BID, a proapoptotic Bcl-2 family member, releasing a C-terminal fragment that induces release of mitochondrial cytochrome c (55,56). Caspase-8 also cleaves p28Bap31, a Bcl-2-associated protein present on the endoplasmic reticulum. This cleavage event may enhance apoptotic signaling because the N-terminal p28Bap31 fragment induces apoptosis. Thus, activated caspases have multiple avenues to enhance apoptotic signaling.
As the apoptotic signal propagates, caspases activate other components of the apoptotic machinery (Table II). These components likely include ICAD, gelsolin, PAK2, MEKK1, and PKC␦ (7,8,54). Proteolysis of these proteins directly impacts upon the apoptotic phenotype. For example, caspase-3 cleaves ICAD, prompting release of active CAD, which then cleaves DNA and promotes chromatin condensation. Caspase-3 also cleaves and activates gelsolin, a protein that regulates actin dynamics. Interestingly, activated gelsolin promotes both cytoplasmic and nuclear apoptosis, including DNA fragmentation. Caspase-dependent activation of kinases including PAK2, MEKK1, and PKC␦ also promotes cytoplasmic and nuclear apoptosis. Interestingly, MEKK1 activation enhances caspase activation, suggesting that kinases may amplify or initiate the caspase cascade.
Caspases also cleave structural proteins of the nucleus and cytoskeleton (Table II) (7,8,54,57). Proteolysis of lamins, NuMa, and SAF-A likely promotes nuclear dissolution and packaging. Similarly, caspases may disrupt cytoskeletal integrity by proteolysis of fodrin, Gas2, keratins, Rabaptin-5, and actin. Cleavage of ␤-catenin and FAK may interrupt cell-cell contacts and cell-matrix focal adhesions. Altogether, proteolysis of these caspase substrates may promote cellular packaging and subsequent engulfment by phagocytes. It should be noted, however, that the functions of these cleavage events are largely speculative and await a more rigorous evaluation.
Another class of caspase substrates includes proteins important for cellular signaling, cellular repair, and macromolecular synthesis. These proteins include kinases, other enzymes, and factors necessary for protein and nucleic acid synthesis (Table II) (7,8,54). Caspase-dependent proteolysis also promotes degradation of Akt-1 and Raf-1, kinases important for cell growth and survival (58). However, these proteins are not direct caspase substrates, suggesting that caspases may activate other proteinases that participate in apoptotic events (58). Overall, proteolysis of these proteins could disrupt cellular homeostasis and terminate survival signals; however, this has not yet been formally demonstrated.
Caspases also cleave presenilins, huntingtin, atrophin-1, and other proteins implicated in neurodegenerative disease (Table II) (7, 8, 54, 59 -61). Some of these proteins are cleaved during apoptosis; however, the significance of these proteolytic events is not clear. Wellington et al. (59), however, suggest that caspase zymogens may cleave neurodegenerative proteins at a low level, generating toxic fragments that initiate apoptosis and perhaps neurodegenerative diseases.

Conclusion
Numerous studies establish caspases as essential mediators of apoptosis. Because disordered apoptosis can promote human disease, these findings have broad implications. Insufficient apoptosis because of caspase inactivation may promote oncogenesis by allowing cell accumulation (62). Recent evidence supports this hypothesis, and careful study of the caspase knockout animals should provide more definitive answers to this question (12,63). On the other hand, caspase over-reactivity promotes cellular suicide, and this may be the basis for degenerative diseases such as Huntington's disease and Alzheimer's disease (59 -61). Net increases in apoptosis could result from a decreased apoptotic threshold, enhanced apoptotic stimulation, or a combination of these factors. Regardless of the inciting factor(s), enhanced caspase activity should result. On the molecular level, increased expression/activity of caspases, apoptotic sensors, or adapters or diminished expression/activity of caspase inhibitors could increase caspase activity. Similarly, loss of survival signals, mitochondrial dysfunction, or diminished stress responses might prompt caspase activation and apoptosis. Further study of the caspase system will prove or disprove these theories and will one day offer treatments for disorders of apoptosis.
Acknowledgments-Because of space limitations, it was not possible to include a comprehensive list of references for all the work discussed. We apologize to those authors whose important contributions could not be described or properly cited.
Note Added in Proof-Since this Minireview was accepted for publication, key studies concerning caspase activation and caspase involvement in human disease have been published. First, Stennicke et al. (64) established that caspase-9 activation does not require proteolytic processing of the zymogen and that the zymogen itself has significant activity in the presence of cytosolic cofactors (presumably APAF-1). Second, we demonstrated that calpain may regulate caspase-9 activation by removing the procaspase-9 CARD domain (65). Third, Yuan and colleagues (66) demonstrated that polyglutamine repeats induce neuronal cell death via caspase-8 and implicated caspase-8 in Huntington's disease, a polyglutamine repeat disease. Finally, Nicholson's group (67) implicated caspase-dependent processing of the amyloid-␤ precursor protein in the pathogenesis of Alzheimer's disease.