Regulation of Caspases in the Nervous System: Implications for Functions in Health and Disease

Caspases, initially identified as a family of proteases regulating cell death, have been found to have nonapoptotic functions as well. Some family members are critical for mediating programmed cell death in development. After development, caspases are downregulated in the nervous system, but continue to perform important nonapoptotic functions relevant for neurogenesis and synaptic plasticity. In neurodegenerative diseases, where aberrant neuronal death is an outstanding feature, there is an increase in caspase activity. The specific caspase death pathways leading to dysfunction and death have still not been fully clarified, despite the plethora of scientific literature addressing these issues. In this chapter, we will present the current knowledge of caspase activation and activity pathways, the current tools for examining caspases, and functions of caspases in the nervous system in health and in disease. Alzheimer’s Disease, the most common neurodegenerative disorder, and cerebral ischemia, the most common cause of neurologic death, are used to illustrate our current understanding of death signaling in neurodegenerative diseases. A better understanding of how caspases function in health and disease would provide appropriate specific targets for the development of therapeutic interventions for these diseases.

Life and death are exquisitely regulated at the cellular level from development through maturity. During development, neuronal death is the major factor shaping the nervous system. This death is mainly caspase-mediated apoptosis. Once the waves of developmental death have passed (death occurs at different times in different parts of the nervous system), there is downregulation of the death machinery, as the postmitotic neurons should live for the life of the organism. Aberrant neuronal death is a major part of neurodegenerative disorders, but there is still no clear understanding of the processes leading to the phenotypes of the various diseases. Even the type of death that occurs continues to be debated, whether it is apoptotic, necrotic, or autophagic, or some combination of these death mechanisms. Here, we will discuss the role that the caspases play in neuronal function, dysfunction, and death. First, we will discuss the regulation of caspase activation and activity. We will examine the current understanding of caspase function in developmental neuronal death and then illustrate the role of caspases in neuronal death in disease employing two diseases of neuronal loss, Alzheimer’s Disease (AD), which is the most common chronic neurodegenerative disorder, and cerebral ischemia/stroke, the third most common cause of death in Western society, which is an acute neuronal disorder with chronic sequelae.

I. The Caspase Family of Death Proteases

Since the seminal work on death pathways in the model organism Caenorhabditis elegans, studies of apoptosis have increased exponentially. While it is agreed that caspases are critical mediators of apoptosis, many details of the death pathways remain to be explicated. Caspases are a highly conserved family, from worms to humans. One major difference, however, is the number of caspases: in worms, there is one caspase, while in mammals, there are 13 (Fig. 1). This difference raises the question of whether individual mammalian caspases have unique functions. Despite the plethora of studies, this is still an open question. In terms of disease intervention, specific nonredundant functions of individual caspases would provide therapeutic targets. As we increase our understanding of how caspase activation and activity are regulated, we can more critically evaluate current studies and reevaluate former studies.

Fig. 1. Mammalian caspases.

Mammalian caspases are schematically represented and grouped by activity. Yellow lines indicate cleavage sites.

In 1992, the interleukin-1β cleaving enzyme (ICE),^1,2 now known as caspase-1, was identified and within a few months, ced-3, an enzyme with significant homology to ICE, was identified in C. elegans and found to execute apoptosis.³ Many more mammalian homologs were identified in the ensuing years, and the term “caspase,” for cysteine dependent, aspartate-specific protease, was agreed upon for the mammalian family. The 13 different mammalian caspases can be divided either by structure or by putative actions. From a structural perspective, there are two general groups, caspases with short prodomains (caspases-3, -6, -7, and -14) and with long prodomains (caspases-1, -2, -4, -5, -8, -9, -10, -11, and -12). Caspase-14 is involved in keratinocyte differentiation and does not appear to have a function in the nervous system⁴ and will not be discussed further. The effectors of apoptosis are the caspases with short prodomains, caspases-3, -6, and -7. The caspases with long prodomains are further subdivided. Caspases-2, -8, -9, and -10 are initiators of apoptosis, although caspase-2 may act as both an initiator and effector. Caspases-1, -4, -5, and -11 process cytokines and contribute to inflammation: caspases-4 and -5 are found only in humans, while caspase-11 is present only in rodents. The inflammatory caspases can also lead to cell death by autocrine mechanisms. Thus, although these caspases may not be direct apoptosis initiators, their activity can lead to death. Caspase-12 is somewhat of an enigma. It may be involved in cytokine processing or in death, particularly in endoplasmic reticulum-associated death. Additionally, while the gene for caspase-12 is expressed in rodents and humans, the protein is expressed in rodents but only in a small number of humans.⁵ There are studies of caspase-12 that suggest that it acts as a dominant-negative regulator of caspase-1 activity and that the caspase-12 proteolytic activity results only in autoprocessing.⁶ Work also supports a role for caspase-12 in ER stress as an activator of matrix metalloprotease 3.⁷

A. Caspase Activation

Caspases are also grouped based on the mechanism of activation (Fig. 2). While the activation of effector caspases is dependent on cleavage at an aspartate residue in the intersubunit linker,^8,9 the activation of initiator caspases is more complex and new studies are refining the activation mechanisms. As a generalization, caspases with long prodomains undergo dimerization with a structural change that provides for the formation of an active site.^9–11 The longer intersubunit linker allows flexibility, leading to the formation of an active site. Most of the initiators exist as monomers in the zymogen form; caspase-2 zymogen is a dimer.¹² Interaction of the specific adaptor protein with the prodomain leads to dimerization and activation. Once the zymogen is activated, there is limited proteolysis of the interdomain. For caspase-9, proteolysis of the interdomain is not required: uncleavable mutants retain activity.¹³ However, recent work on caspase-8 shows that both dimerization and cleavage are required for optimal activity.¹⁴ This finding suggests that there are subtle differences in activation requirements that have implications for measurements of activation/activity of each caspase.

Fig. 2. Activation mechanisms of effector and initiator caspases.

Activation models for initiator and effector caspases are shown. Blue indicates inactive caspase while orange denotes activated caspase. Some of the activated caspases are subject to regulation by IAPs.

The long prodomain caspases have unique activation platforms. The intrinsic pathway of apoptosis centers on the mitochondrial regulation of death (Fig. 3). The apoptosome,¹⁵ the caspase-9 activation platform, was the first mammalian activation platform described. Release of cytochrome c from mitochondria leads to an ATP-dependent recruitment of caspase-9 by the adaptor, Apaf-1. Caspase-9 then undergoes limited autocleavage to a p35 fragment, which can be inhibited by XIAP, X-linked inhibitor of apoptosis protein. As noted above, cleavage is not required for activation of caspase-9 and, since the autocleavage allows inhibition of caspase-9 activity, it can actually lead to decreased activity. Caspase-3 can also cleave caspase-9 to a p37 fragment; this cleavage is not an activation step but potentiates activation by preventing XIAP inhibition of caspase-9,¹⁶ as described below.

Fig. 3. Caspase regulation: intrinsic pathway.

The intrinsic death pathway is activated by permeabilization of the mitochondria, leading to the release of cytochrome c and formation of the apoptosome, the caspase-9 activation platform. Once activated, caspase-9 can cleave and activate caspases-3 and -7. Caspases-3, -7, and -9 are subject to inhibition by XIAP. XIAP can be inhibited by Diablo/SMAC or HtrA2/Omi which are released from permeabilized mitochondria.

The extrinsic death pathway is activated by ligand binding to death receptors (Fig. 4). The caspase-8 activation platform is the death-inducing signaling complex (DISC). Ligand binding to a cell surface death receptor leads to the recruitment of an adaptor protein, which recruits caspase-8. As an example, Fas ligand binds to the fas receptor to recruit FADD and subsequently caspase-8. Binding to FADD dimerizes caspase-8 and then autocleavage provides for complete activation.^14,17 Cleavage, without dimerization, by another enzyme, an effector caspase or granzyme B, does not produce activation of caspase-8.¹⁴ Recent work suggests that TNF-α induces two different caspase-8 activation pathways, a RIPK1-dependent and a RIPK1-independent pathway.¹⁸ There is also phosphorylation-mediated regulation of caspase-8 activity.¹⁹

Fig. 4. Caspase regulation: extrinsic pathway.

The extrinsic pathway is activated when a ligand binds to a death receptor; binding of FasL to the Fas receptor is shown. This leads to recruitment of an adaptor protein (FADD), which recruits caspase-8, forming the DISC, leading to dimerization and activation of caspase-8. Caspase-8 then cleaves and activates effector caspases. FLIP inhibits the formation of active caspase-8 dimers.

The activation platform for caspase-2 is proposed to be the PIDDosome, a complex of PIDD, RAIDD, and caspase-2.^20,21 However, two different lines of PIDD null mice appear to undergo caspase-2-dependent death, although caspase-2 activity was not measured in those studies.^22,23 Studies of the PIDD complex in tumor cell lines showed that overexpression of PIDD induced RAIDD-dependent cell death²⁴ and that PIDD function in those cells depends on the cleavage state of PIDD.²⁵ PIDD autocleaves to PIDD-C and PIDD-CC. Partially cleaved PIDD, PIDD-C, complexes with NEMO and leads to NFκB activation and survival signaling. Further cleavage of PIDD produces a fragment, PIDD-CC, that, in the overexpression system, complexes with RAIDD and leads to cell death. There are three isoforms of PIDD²⁶: isoform 1 contains full-length PIDD and can be processed to PIDD-C and PIDD-CC; isoform 2 has a deletion of 146 amino acids at the N-terminus and an 11 amino acid deletion at position aa580; isoform 3 has a 17 amino acid deletion at aa705. Isoform 2 cannot form the PIDD-CC fragment. Expression of these isoforms is tissue- and cell-type specific: isoform 2 is present mainly in transformed cell lines and in normal liver, pancreas, and leukocytes, while isoforms 1 and 3 are found in normal tissue with the exception of the skeletal muscle. Isoform 1 is also expressed in transformed cell lines and isoforms 1 and 2 are found in several tumors. Since isoform 2 cannot form PIDD-CC, it should not interact with RAIDD. It is possible that PIDD function is different in cells with isoform 2 rather than the isoforms that can form PIDD-CC. This hypothesis would have interesting implications for the regulation of caspase-2 function in the different cell types. Caspase-2 activity is also regulated by phosphorylation.^27–29

Caspase-1 is activated in the inflammasome.^30–32 Several inflammasomes have been identified, the IPAF (ICE-protease activating factor), the NLRP1 (NACHT-, LRR-, and pyrin-domain containing proteins), and the NLRP2/3 inflammasomes.^31–34 The basic assembly involves the aforementioned proteins (IPAF or NLRP), which contain pyrin domains, caspase-1, which contains a CARD domain, and an adaptor protein with a CARD domain, such as ASC (apoptosis-associated speck-like protein containing a CARD) or CARDINAL (CARD inhibitor of NFkB-activating ligands). Caspase-1 is incorporated by ASC into the inflammasome through its CARD domain.^35,36 The inflammasomes are activated by receptor-mediated signaling through TLRs (toll-like receptors) or in the cytosol by the NLR family in response to stress signals.

B. Endogenous Inhibitors of Caspases, the IAP Family

The Inhibitor of Apoptosis Protein family is classified by the presence of BIR (baculovirus IAP repeat) domains. In mammals, there are eight IAPs (see Ref. 37 for a review). Of these, cIAP1, cIAP2, and XIAP have been shown to bind to caspases, but XIAP is the only one which inactivates caspases.³⁸ Other IAPs do inhibit cell death but not by directly inhibiting caspase activity. XIAP binds to and inactivates active caspases-3 and -7 via the BIR2-linker region and binds to and inactivates caspase-9 via the BIR3 domain. cIAP1 and cIAP2 can also bind to active caspases-3 and -7 but lack the region necessary for inactivating the caspases.³⁹ XIAP-BIR3 binds to the neoepitope of caspase-9 that is generated by autocleavage of caspase-9 at aa315. XIAP cannot bind to the capase-9 neoepitope generated by caspase-3 cleavage of caspase-9 at aa330. Thus, caspase-3 cleavage promotes activity of caspase-9 by preventing the inhibition of caspase-9 by XIAP. In a similar fashion, the p20 fragment of caspase-3 can be inhibited by XIAP, but the p17 fragment cannot be inhibited. The BIR3 domains of cIAP1 and cIAP2 contain only one of the four residues required to inactivate caspase-9.³⁸ Caspase-8 also has an endogenous inhibitor, c-FLIP, which is caspase-8 without the active site.⁴⁰ FLIP dimerizes with caspase-8, preventing its activation rather than inhibiting active caspase-8.

C. Endogenous Inhibitors of IAPs

There are several endogenous inhibitors of IAPs. Smac/Diablo^41,42 and HtrA2/Omi⁴³ are found in the mitochondria of healthy cells, and the N-terminal amino acids of these proteins form the IAP-binding motif (IBM). Mitochondrial permeabilization leads to release of these proteins, which then interfere with the IAP-inhibition of caspases, resulting in potentiation of cell death. Another endogenous inhibitor of IAPs is XAF1.⁴⁴ In nonneuronal tissues, XAF1 functions as a tumor suppressor. Originally identified as a specific inhibitor of XIAP, it is now reported that XAF1 can also bind to cIAP1 and cIAP2.⁴⁵ As the function of cIAP1/2 is not quite clear, it is also not clear what the function of XAF1 binding to these molecules would be.

D. Regulation of Mitochondria in Apoptosis

Mitochondrial outer membrane permeabilization (MOMP) allows for selective release of molecules that participate in death. These molecules include cytochrome c, Smac/Diablo, HtrA2/Omi, and AIF. The Bcl2 family of proteins participates in the regulation of MOMP.¹¹ This multimembered family is identified by the presence of BH domains and has both antiapoptotic members, which contain 4 BH domains, and proapoptotic members, which contain either 3 BH domains or only a BH3 domain. Bcl-2 family proteins reside either in the mitochondrial outer membrane or in the cytoplasm of healthy cells.

E. Measurement of Caspase Activation/Activity

Much of the literature regarding caspases should be reevaluated with regard to what the measures of caspases used in a particular study actually showed. Multiple methods have been used to measure caspase activity and activation, including Western blotting, immunocytochemistry, synthetic peptide substrate/inhibitor assays, and fluorescent indicators. Western blotting with specific antibodies can be used to quantify changes in zymogen expression and reveal the appearance of cleavage fragments. The relevance of cleavage to activation state varies among the caspases, as noted above. Proteolytic processing of the zymogen is required for effector caspases (3, 6, and 7) and appearance of the p17/19 and p10/12 fragments indicates that these caspases have been activated. For the initiator caspases, dimerization is required and there is a variable requirement for autocleavage (see above), but appearance of cleaved fragments is not a reliable read-out for activation since it will not distinguish between a cleavage event in the process of activation and a cleavage event that occurred without prior dimerization.

An additional subtle point that is often missed is that caspase activation and activity are not equivalent: IAPs can inhibit the activity of activated caspases. This distinction is critically relevant from a functional perspective. Only an active caspase will enact death. Caspases-3, -7, or -9 can be inhibited by XIAP in situ. Lysis of the cells for Western blotting disrupts IAP-caspase complexes and the Western blot shows cleaved caspases. Thus, activated but inhibited caspases cannot be distinguished from active caspases using Western blotting of cell lysates. For initiator caspases, interpretation of Western blotting data is even more problematic. Both caspase-8 and caspase-9 undergo autoproteolysis when incorporated into their respective activation complexes, although the full-length caspases have activity as well.^16,46,47 As stated before, caspase-8 requires further processing for optimal activity, while additional processing of caspase-9 does not substantially increase activity.¹⁴ Caspase-2 also undergoes autoproteolysis,¹² but it has not yet been clear how much this proteolysis contributes to total caspase-2 activity.

Another frequently used technique is immunocytochemistry, using antibodies which detect either the zymogen and cleaved fragments or are specific for neoepitopes generated during cleavage of zymogens. This technique is excellent for studying changes in the subcellular localization of a caspase, and for appearance of neoepitopes of effector caspases, with the same caveats noted above for Western blotting: appearance of the neoepitope does not necessarily indicate that there is an active caspase. Many studies inaccurately use the appearance of cleaved caspase-3 as a measure of apoptosis but, without another measure, cleaved caspase-3 alone does not mean that the cell is dying, as the caspase can be inhibited by XIAP.

Synthetic peptide pseudosubstrate caspase inhibitors/substrates, such as DEVD or IETD, have been and continue to be used by many groups as a method of distinguishing the role of individual caspases, despite studies published over a decade ago showing that these reagents are not specific for individual caspases.^48,49 More recently, a comprehensive study evaluated the specificity of these reagents and concluded that all of the so-called specific caspase inhibitors/substrates are best at detecting caspase-3 activity, and some do not even detect their putative targeted caspase.⁵⁰ To evaluate data, it is also important to understand how these peptides inhibit caspases: the peptide binds to the active caspase and inhibits downstream events. For in vivo measures of caspase activity, fluorescent versions of these peptides are being utilized as substrates that fluoresce when bound to the active caspase. These peptides bind irreversibly to the active site and cause inhibition of downstream events. Thus, once the fluorescent peptide is added, caspase activity is inhibited. Any cellular events that occur subsequent to the addition of the fluorescent peptide are not the result of caspase activation and could be the result of caspase inhibition. Thus, data acquired using these reagents should be reevaluated.

There are a few methods available to show that a specific caspase is active. One of these utilizes a broad-spectrum caspase substrate/irreversible inhibitor linked to biotin, bVADfmk. This active site affinity ligand binds irreversibly to the active caspase in the cell and inhibits the caspase. Preloading cells with bVADfmk before initiation of death captures initiator caspases and addition of bVADfmk at the time of harvesting cells undergoing death detects active effector caspases.⁵¹ The biotinylated affinity ligand–caspase complex is isolated with streptavidin, run on SDS-PAGE, and the specific caspase bound is detected with Western blotting for individual caspases. Studies using this method show that the active initiator caspases isolated by the affinity ligand are uncleaved, supporting the proximity-induced dimerization model of activation (no cleavage necessary initially) for initiator caspases. Adaptation of this method to animal studies could enable the identification of which caspases are active in vivo. Groups are also developing specific fluorescent read-outs of caspase activity. For caspase-2, such a system has been used in cultured cells to show dimerization and activation of caspase-2 after heat shock.⁵² This system could be adapted for other initiator caspases.

Functional relevance for individual caspases can be obtained using molecular knockdown/knockout of individual caspases. Abrogation of death shows that specific caspases are critical for execution of the death pathway under study. Studies have used antisense and siRNA knockdown of individual caspases. The ability to use siRNA to knockdown caspase-2 has been questioned because of the lack of reproducibility of published work using siRNA.⁵³ However, it appears that it may be the sequence used in the studies in question that is at fault, rather than the ability of siRNA to ablate expression of caspase-2, as siRNA have been successfully used to knockdown other caspases.

Caspase null mice have been valuable tools for studying caspases with a role in development, as discussed below. However, caspase null mice without profound phenotypes suggest either that these caspases do not have a major role in development or that there are compensatory changes in other caspases, which obscure the function of the targeted caspase. Use of conditional knockouts could address these issues.

F. Targets of Caspases

Caspases can cleave many proteins; cleavage is limited (cleavage at one or two sites) as illustrated by the activation of effector caspases by initiator caspases. There have been multiple reports of substrate cleavage changing the activity of the substrate: that is, converting a proapoptotic protein to an antiapoptotic one, or vice versa. Such events serve to amplify the signaling pathways leading to death or survival. Although predictions have been made, careful time courses of substrate cleavage events, combined with specific blockade of these events, will be required to determine the function of such cleavage in individual death pathways.

II. Caspases and Developmental Neuronal Death

During development, programmed cell death, which is mediated by caspases, is the major force that shapes the organism. This process is essential to the formation of a normal nervous system. Two general populations die during neuronal development: neuronal precursors and postmitotic neurons. Caspase-9 and the caspase-9 adapter, Apaf1, are essential for death of the neuronal precursors. Mice lacking either of these proteins have severe malformations of the nervous system.^54–57 The initial report of the caspase-3 null mice also showed a severe neuronal developmental phenotype, very similar to the caspase-9 null mice, with death of the mice soon after birth.⁵⁸ Care must be taken when drawing conclusions, because changing the genetic background of the caspase-3 null mice from a mixed background to a C57/Black6 background modified the phenotype drastically⁵⁹—on this background, neuronal development is grossly normal, and the mice are able to grow to adulthood and to breed. Thus, the dependence of neuronal developmental death on caspase-3 is unclear. It is also not clear that all neuronal developmental death is dependent only on the caspase-9 pathway. The timing of death varies among regions of the brain, from the early embryonic to the early postnatal periods.

Mice lacking many of the other caspases, such as caspase-1, -2, -6, -7, -11, or -12, do not have an obvious neuronal phenotype.^59–63 These caspases may play a role in regional pruning of neurons or in the plasticity of the nervous system, and be important in the maintenance of the mature nervous system. There is also an issue of compensatory changes in other caspases or caspase regulators, which can confound the simple interpretation of results from mouse knockout studies. This issue is a significant problem with apoptotic genes, where phenotypic selection in developing embryos can severely alter normal expression patterns, as has been shown in caspase-2, -3, and -9 null mice.^63,64 Compensations have been shown at the level of expression of other caspases⁶³ or of regulators of caspase activity.⁶⁴ The relative expression of the caspases and their regulators is a major factor in the choice of execution pathway. It would be wise to keep in mind this potential complexity when evaluating data on caspase involvement in neurodegenerative diseases.

III. Nonapoptotic Caspase Function in the Nervous System

There is increasing evidence that there are nonapoptotic functions of caspases in the nervous system. These data are most compelling for a function for caspases in neurogenesis and synaptic activity.⁶⁵ Studies of neurogenesis indicate that caspase-3 activity in neuronal progenitors facilitates neurogenesis.⁶⁶ Inhibition of caspase activity also blocked neurite extension, suggesting a role for caspases in the morphologic changes that occur during neurogenesis. Differentiation of mouse neural stem cells was inhibited by inhibition of caspase activity, and caspase inhibition also blocked p53 phosphorylation and transcriptional activation.⁶⁷ These data suggest a critical, nonapoptotic function of caspase-3 in neurogenesis.

Neuronal circuits are regulated by selective pruning of synapses, axons, and dendrites. In Drosophila, the initiator caspase Dronc is required for pruning.⁶⁸ Studies of trophic factor deprivation-mediated death of cultured dorsal root ganglion neurons suggest that there is selective activation of caspase-6 in axons, which leads to axonal pruning but not to death.⁶⁹ Caspases have also been implicated in neuronal plasticity. In vivo studies in the zebra finch showed that caspase-3 activity is required for the birds to learn a new song.⁷⁰ The cleaved caspase-3 is in the synapses and is associated with XIAP, suggesting that XIAP modulates the activity of caspase-3, so that synapse remodeling occurs but not death. Studies using DEVDfmk in rats support a role for caspase activity in long-term potentiation⁷¹ and in active avoidance learning.⁷² It has been recently shown that caspase-3 activation is required for long-term depression. LTD was inhibited in caspase-3 null hippocampal slices and XIAP-BIR3, a specific inhibitor of caspase-9, also inhibited LTD.⁷³ These nonapoptotic caspase functions also indicate why it is not appropriate to connote caspase-3 cleavage with apoptosis.

These nonapoptotic functions of caspases need to be better understood for the design of appropriate therapeutic interventions to treat diseases with aberrant caspase activation and neuronal death.

IV. Caspase Function in Neurodegenerative Diseases

The study of diseases requires the use of models of the disease. We will present two diseases that are among the most highly prevalent human neurological disorders of aging. We will consider what the appropriate models are and what the current understanding of the caspase pathways is in each disease.

A. Alzheimer’s Disease

Over the past century, the life span in industrialized countries has increased an average of 25 years. This increased life expectancy has shifted the average age of the population and more people are living beyond 80 years of age. With an increase in the aged population, there is an increase in the prevalence of diseases associated with aging. At age 65, about 2–3% of people have AD, but the incidence of AD doubles for every 5 years of age afterward, and approximately 50% of people over the age of 85 have AD. There are about 5.3 million cases of AD in the United States.⁷⁴ As a result of increasing life expectancy and expanding population, the number of AD cases should triple over the next 40 years.⁷⁴ AD is the most common form of dementia among the elderly population and represents the fourth leading cause of death in industrialized countries.^75,76

AD was characterized by Alois Alzheimer in 1907 as a progressive impairment of cognitive functions.⁷⁷ The most frequent symptom is gradual loss of episodic memory, which is memory of recent events. With disease progression, impairment is found in language, cognition, reason, and temporal-spatial orientation, together with other changes in mood and personality. After the initial clinical symptoms, there is continuous and progressive decline leading to mutism, vegetative state, inanition, and finally death. Survival ranges from 5 to 20 years from detection of the first symptoms of the disease to death. The definitive diagnosis of AD is pathologic but biopsies are rarely done and tissue is usually obtained postmortem. Thus, the clinical diagnosis is of possible or probable AD. The two main histopathological hallmarks required for the definitive diagnosis of AD are senile plaques, extracellular congophyllic deposits composed of Aβ peptide, and neurofibrillary tangles (NFTs), intraneuronal filamentous aggregates composed of hyperphosphorylated tau proteins and paired helical filaments (PHF).⁷⁸

Multiple factors have contributed to the lack of full understanding of the order of events in the development and progression of AD; and these factors include the insidious onset of clinical symptoms, the lack of a definitive clinical diagnosis, and the chronic nature of the disease. The progressive loss of synapses and neurons in limbic and cortical areas, translated functionally as a disconnection between different brain areas, is postulated to lead to the clinical symptoms.^79,80 This initial disconnection does not seem to have repercussions on behavior for several decades, since the gradual loss of synapses and neurons seems to start 20–40 years before the manifestation of the first symptoms.^81,82 Distribution of NFT correlates better than amyloid plaque burden with the severity of dementia and neuronal death.^83–87 But, as no studies of progressive pathologic analysis of single patients over time exist, all of the human data should be considered as correlative rather than definitive with regards to molecular causality. The use of imaging of amyloid plaques in AD patients may provide information about the progression of the disease, when correlated with symptoms.

Three genes have been linked to familial AD: amyloid precursor protein (APP),⁸⁸ located on chromosome 21, and presenilin-1 (PS1)⁸⁹ and presenilin-2 (PS2),⁹⁰ located on chromosomes 14 and 1, respectively. About 85% of EOAD cases correspond to mutations in the PS1 gene, and mutations in PS2 or APP are much less frequent.^91,92 There have been 157 mutations in PS1 and 10 in PS2 described worldwide. There are families with a clear genetic component that have not been associated with mutations in any of these three genes, which suggests that other genes may be involved.

1. Mouse Models in AD

Knowledge of the genetic mutations in familial AD enabled the development of mouse models expressing these human mutations in APP and PSEN1/2 as tools to study AD. The mouse models do exhibit some of the pathological hallmarks of AD, including amyloid deposits, abnormal tau hyperphosphorylation, and gliosis, and also display impairments of memory and learning.^93–99 The mice do not have NFTs or neuronal loss. Additional models were generated containing mutant APP and PS1 or both of these and mutant tau, in an attempt to provide a phenocopy for AD.^100–110 Since some of these models overexpress mutant genes that are not even found in AD cases, the results extracted from these models should be taken with caution.

Although these mice do not exhibit every single hallmark of the disease, they are a valuable tool for understanding the interaction between genetic factors able to modify the in vivo production and deposition of Aβ, identifying molecular targets, and evaluating new therapies designed to stop or slow down the pathological and clinical manifestations of this devastating disease. Some of the mouse models may represent good models of the early, preclinical disease. The models have been exploited to show how plaques develop in real time using 2-photon microscopy on live animals. Surprisingly, these studies revealed that individual plaques develop in several hours and then remain constant over months.¹¹¹ This technique promises to be useful for examining pathways and structural changes.

2. In Vitro Models of AD

In vitro models utilizing primary neuronal cultures provide an opportunity to decipher molecular mechanisms of AD. These models have been instrumental in our understanding of how familial AD gene mutations alter Aβ metabolism, secretion, and/or degradation.¹¹² Work using these models has shown that alterations in the γ-secretase complex produce an enhancement of APP processing^113,114 and lead to an increase in the intracellular concentration of Aβ species.¹¹⁵ These modifications in the processing of APP alter the Aβ₄₀/Aβ₄₂ ratio, increasing the relative amount of Aβ₄₂.^116–118 In vitro models have provided critical insights into where in the cell APP is processed and how Aβ is generated and transported within the cell.¹¹⁹

Arguably, the most relevant use of in vitro models is to identify the neurotoxic species triggering the progression of the disease. The lack of correlation between the amyloid plaque burden and the clinical progression of the disease has led to the ongoing debate about the role of amyoid in the etiology of AD. However, studies of soluble oligomeric Aβ species show that early memory impairment and synaptic dysfunction correlate better with the levels of oligomeric Aβ than with the fibrillar species.¹²⁰ It is important to keep in mind that Aβ assemblies are highly dynamic and are in active equilibrium with the environment. This property makes it very difficult to ascribe specific toxic characteristics to a certain type of aggregation state.

However, several studies have examined the relationship between toxicity and aggregation. In 1998, the characterization of soluble, nonfibrillar oligomeric species (ADDL: Aβ derived diffusible ligands) and demonstration of their neurotoxic properties¹²¹ suggested that oligomers, not fibrils, were the toxic species. ADDLs caused neuronal death in primary hippocampal neuron cultures and inhibited long-term potentiation in cultured slices from rat brains.^121,122 Another modification of the in vitro systems employs naturally produced, cell-derived Aβ oligomers^123,124 that are toxic in cultures and inhibit LTP when injected into rat brain.¹²⁴ The relevant size of the oligomers is a topic of debate, and oligomers as large as dodecamers have been isolated from human AD brains¹²⁵ and from transgenic mice.¹²⁶ These species are also toxic in primary neuronal cultures and impair LTP in hippocampal slices.¹²⁷

3. Caspases in AD

a. Caspases in AD brains.

Neuronal loss is an outstanding feature of AD. A role for apoptosis in this neuronal loss is suggested by the increased expression of caspases and cleaved caspase substrates in postmortem AD brains.^128,129 mRNA expression of caspases-1, -2, -3, -5, -6, -7, -8, and -9 is increased in the brain of AD patients compared to controls.¹³⁰ Pyramidal neurons from vulnerable regions involved in the disease showed an increase in activated caspase-3 and -6.^131,132 Negative correlations have been found between caspase-8 levels and age at disease onset and age at patient death, suggesting a role in disease regulation.¹³³ Postmortem AD brain tissue showed extensive expression of cleaved caspase-8 (p18) in neurons and reactive astrocytes in the hippocampus and the entorhinal cortex¹³⁴; the immunoreactivity of cleaved caspase-8 is colocalized in tangle-bearing and plaque-associated neurons. As noted above, it is not clear whether this is active caspase-8 or effector-processed (nonactive) caspase-8.

Synaptosomes prepared from AD brain frontal cortices showed an enrichment in caspase-9 compared to nondemented controls.¹³⁵ While there is a great deal of correlative data showing that caspases are somehow involved in AD, there is not a great deal of specific data for individual caspases. Much of the data either measures changes in expression, which does not necessarily translate into a functional role for proteins that require activation, or utilizes the nonspecific peptide inhibitor/substrates to imply the involvement of specific caspases. The peptide inhibitor/substrate data can be viewed as indications of caspase function, but these data certainly cannot be construed as implicating individual caspases in AD dysfunction pathways.

b. Caspase cleavage of APP.

APP contains caspase cleavage sites and it has been described that caspases-3, -6, -7, and -8 can cleave APP.^129,131,136 Cultures of neuronal cells deprived of serum showed activation of caspase-6 and subsequent processing of APP by caspase-6, generating a 6.5-kDa fragment that contains Aβ.¹³¹ In this context, caspases may play an active role in Aβ-induced neurotoxicity. Amyloid plaques are enriched in caspase-cleaved APP.¹²⁹

In vitro systems have been instrumental in defining the cytotoxic properties of APP cleavage fragments, particularly the C-terminal fragments (CTF) derived from γ- and β-secretase cleavage of APP. C99 is toxic to neurons¹³⁷ and expression of C99 in rodent brains induces Aβ deposition, neurodegeneration, alterations in behavior, and synaptic deficits.¹³⁸ C99 can be further cleaved by caspases to generate C31 and it may be the C31 that leads to the CTF toxicity.^138,139 The specific caspase that cleaves C99 has not been adequately identified, although caspases-8 and -9 have been implicated.¹⁴⁰ Expression of C31 alone is cytotoxic, suggesting that the cleavage of CTFs to generate C31 may potentiate susceptibility to apoptosis.¹³⁵ Mutations in APP to abolish the caspase cleavage site significantly attenuate the cytotoxic effects of C99.¹³⁹ PDAPP mice carrying the cleavage mutation [PDAPP(D664A) mice] deposit Aβ but do not develop behavioral deficits, supporting a role for C31 in the potentiation of AD deficits.¹⁴¹ However, a recent study using these mice showed that there is no effect of expression of APPD664A on the development of behavioral defects.¹⁴² Further work is needed to clarify the functional relevance of C31.

c. Caspase processing of tau.

Caspases do not only process APP; caspase cleavage of tau enhances tau filament polymerization in vitro.^143,144 Cleaved caspase-6 in the cortex and hippocampus of human AD brains is colocalized with plaques and NFTs.^145,146 There is also an increase in caspase-6 cleaved tau fragments, detected with an antibody to the neoepitope generated when caspase-6 cleaves tau.¹⁴⁶ A study of cultured dorsal-root ganglion neurons suggests that NGF-withdrawal leads to APP cleavage, which then induces caspase-6 activation.⁶⁹ Taken together, these studies suggest that cleaved APP activates caspase-6, which may induce NFT formation. In support of a role for caspases in NFT formation are 2-photon studies. Live imaging from mice expressing mutant tau shows that there is activation of caspases prior to development of tangles. However, this study used a fluorescent caspase substrate/inhibitor (FLICA), which binds irreversibly to active sites, to follow caspase activation, so the causal link between caspase activation and tangle development cannot be inferred, despite the conclusions drawn from this study.¹⁴⁷ This work illustrates that caspase activation may be an early event in the death pathway. This is supported by studies discussed below showing that Aβ induces caspase activation in primary neurons within 2 h of addition to the cultures.¹⁴⁸

d. Caspases and inflammation in AD.

Inflammation is believed to contribute to the progression of AD.¹⁴⁹ AD brains have increased mRNA and protein expression of the inflammatory cytokines IL-1β and IL-18 in both neurons and glia that are colocalized with Aβ and tau.^150,151 Microglia isolated from autopsies of AD patients showed stronger IL-1β response to Aβ than did microglia isolated from control brains.¹⁵² In the Tg2576 AD mouse model, astrocytes and microglia surround amyloid plaques, and IL-1β expression is upregulated in these astrocytes, suggesting that Aβ induces reactive astrocytes, leading to an increase in the inflammatory cytokine.¹⁵³

While caspases may not be directly involved in regulating the expression of cytokines, they do regulate their release. Caspase-1 can regulate Aβ-mediated release of cytokines. Aβ treatment of mouse microglia causes increased IL-1β release.¹⁵⁴ Caspase-1 activation was assessed in this study by measuring its cleavage profile, which does not indicate whether the cleaved caspase-1 is active (see above), but the increase in IL-1β is strong support for activation of caspase-1.

What is the role of the NLRP caspase-1 inflammasome in the context of AD? No direct study has been done linking AD and caspase-1 activation in the human brain, but NLRP1 protein is found in neurons and oligodendrocytes.¹⁵⁵ It has been shown in cell lines and primary mouse macrophages that NLRP3 is activated following intracellular K+ withdrawal.¹⁵⁶ In spinal cord neurons treated with valinomycin, a K+ depleting agent, there is activation of the NLRP1 caspase-1 inflammasome.¹⁵⁷ Moreover, cerebellar neurons grown in K+ free medium exhibited higher NLRP1 protein levels.¹⁵⁸ A link between AD and K+ channel dysfunction has been reported. Aβ has been implicated in inducing a decrease in intracellular K+ levels.^159,160 Based on these studies, it can be inferred that increased Aβ in the brain may activate the NLRP1 caspase-1 inflammasome through intracellular K+ depletion.

e. Caspases and Aβ toxicity.

Caspase-2 has been shown to be required for Aβ-mediated death of primary neurons using antisense knockdown; neurons from caspase-2 null mice are resistant to Aβ toxicity.¹⁶¹ Most recently, caspase-2 activity has been measured using the bVAD affinity ligand method in neurons treated with Aβ; measurable activity is induced within 2 h of treatment.¹⁴⁸ Other studies of Aβ induction of caspase-2 have employed peptide substrates to measure activity; these measures are not specific for caspase-2 (see above).

Cortical neurons from caspase-12 null mice are resistant to Aβ.^161,162 Caspase-12 has been shown to be involved in the signaling pathways associated with ER stress-induced apoptosis in rodent AD models.^162,163 However, due to a frame shift mutation in the human caspase-12 gene in most humans, the full-length protein is expressed in only a subset of people and there is no increase of AD in this population.^5,164 Hence, caspase-12 is not believed to play any role in human AD. Human caspase-4 shares a high homology with caspase-12, and it is also induced by ER stress.¹⁶⁵ Aβ increases caspase-4 cleavage in the human neuroblastoma cell line, SK-N-SH, within 24 h.¹⁶⁵

It has also been proposed that Aβ cross-links the cell surface death receptors Fas/TNFR. In neuronal PC12 cells, Aβ induces downregulation of FLIPs, the endogenous caspase-8 inhibitor, leading to cleavage of caspase-8.^166,167 Another study utilizing SH-SY5Y neuroblastoma cells revealed that caspase-8 is recruited to the APP complex following Aβ stimulation. Once in the complex, caspase-8 is cleaved to a p10 fragment. It is argued that the cleaved/activated caspase-8 cleaves APP to induce cell death.¹⁶⁸ These studies have not shown that removal of caspase-8 blocks Aβ effects.

Calcium and the calmodulin-dependent protein kinase II (CaMKII) could play a partial role in regulating Aβ-induced caspase activity.¹⁶⁹ This work points to a potential calcium dysregulation induced by Aβ to activate CaMKII, which is sufficient to elicit caspase activity, measured with peptide substrates. However, CamKII may also phosphorylate caspase-2 to inactivate the caspase.²⁷ The function of CamKII with regard to Aβ and caspase-2 needs to be studied further. The role of calcium in regulating caspase actions is further demonstrated in a study that utilized a human neuronal cell line (hNT), revealing that Aβ induces accumulation of a lipid metabolite, 1-o-hexadecyl-2-acetyl-sn-glycero-3-phosphocholine (C16:0PAF). This lipid product causes elevated intracellular calcium levels as a result of ER stress that leads to caspase activation. Peptide substrates were used to measure the caspase activity.¹⁷⁰ Furthermore, m-calpain was activated before detectable caspase activity using the peptide substrates, implicating this enzyme in the regulation of caspase activation.

f. Caspases and Aβ generation.

In addition to responding to Aβ toxicity, caspase-2 or other caspases may play a role in generating Aβ. BACE, the β-secretase that generates Aβ from APP, has been shown to be stabilized by caspase-3 cleavage of GGA3, an adaptor protein involved in BACE trafficking.¹⁷¹ As discussed below, it was shown that in a model of cerebral ischemia, this mechanism leads to an increase in Aβ.

Taken together, caspases are very closely linked to the degeneration process found in AD brains, but the specific caspases involved and the downstream effect of these caspases has not been adequately defined. Figure 5 proposes how caspases may act in AD pathogenesis. Table I summarizes the current evidence of a role for each caspase in AD.

Fig. 5. Proposed caspase actions in Alzheimer’s Disease.

The amyloid precursor protein (APP) can be proteolytically processed to release multiple products, including Aβ, C99, and C31. C99 and C31 are toxic to neurons. Aβ can form fibrils, which deposit as amyloid plaques and soluble oligomers and can lead to ER stress, caspase activation, tau hyperphosphorylation, neuronal dysfunction, and neuronal death.

TABLE I.

Caspases Implicated in Alzheimer’s Disease

Caspase	Paradigm	Measure	References
1	AD brain Aβ treated microglia	mRNA Casp1 cleavage In vitro Tau cleavage	128,143,154
2	AD brain Aβ treated hippocampal neurons Aβ production from H4-C99 cells	mRNA Casp2 null neurons siRNA depletion Casp2 cleavage z-VDVAD-fmk	128,161,235
3	AD brain Aβ production from H4-C99 cells Tg4519 mice	mRNA z-DEVD-fmk In vitro Tau cleavage cleaved Casp3 IHC Casp3 cleavage	128,143,235–238
4	Aβ treatment of SK-N-SH cells	Casp4 cleavage siRNA depletion	165,239
5	AD brain	mRNA	128
6	AD brain Tg4510 mice	mRNA In vitro Tau cleavage Casp6 cleavage	128,131,143,236
7	AD brain Tg4510 mice	mRNA In vitro Tau cleavage	128,143,236
8	AD brain Aβ production from H4-C99 cells	mRNA siRNA depletion z-IETD-fmk In vitro Tau cleavage cleaved Casp8 IHC	128,134,143,235,240
9	AD brain	mRNA cleaved Casp9 IHC	128,240
12	Aβ treated cortical neurons Hippocampal organotypic slices treated with Aβ	Antisense to Casp12 Casp12 null neurons Casp12 cleavage	162,241

B. Cerebral Ischemia

Stroke, an acute neurodegenerative disorder, is the third most common cause of death and a leading cause of disability in major industrialized countries.¹⁷² Approximately 85% of strokes are ischemic, resulting from a thrombus that occludes a major cerebral artery and leads to a focal loss of blood flow. Unless normal blood flow is recovered within a short period of time, under 3 h from onset of occlusion, there is massive cellular death within the ischemic territory. Strokes can also be a consequence of either cardiac arrest or intraparenchymal hemorrhage.

The series of events instigated by ischemia develops over many days. Inflammation and altered microvascular permeability produce tissue edema. Direct effects on cells trigger glutamatergic excitotoxicity, ionic imbalance, and free-radical reactions, among others. There are also dynamic interactions of the neurovascular unit (the complex of neurons, the microvessels that supply them, the supportive cells: astroglia and microglia, and other resident inflammatory cells).^173,174

Over the past decade, the mode of cell death induced by ischemia has been shown to be combined necrotic and apoptotic cell death. The center of the lesion, the core, has complete energy depletion, and there is a gradient of energy depletion from the core toward the bordering zone, the penumbra. From a morphologic perspective, the core has been considered mainly necrotic,^175,176 while the penumbra has been considered apoptotic.^177–179 However, it has been suggested that the initial events in the core are apoptotic, and secondary necrosis results from a rapid failure to fully develop the apoptotic program because of the maintained depletion of apoptosis-requiring energy stores in the core.¹⁸⁰

Neuronal death in the adult brain does not necessarily exhibit the classic morphological manifestations of apoptosis, as initially described by Kerr et al.,¹⁸¹ while in the neonatal brain, hypoxic-ischemic insults can cause typical morphological changes of apoptosis. In the adult brain, neurons are equipped with antiapoptotic molecules that raise the apoptotic threshold.¹⁸² There is a large body of evidence that suggests there is caspase-mediated cell death after ischemia,¹⁸³ and multiple studies implicate apoptotic pathways as critical mediators of ischemia.¹⁸⁴ For more than a decade, it has been suggested that cell death in the CNS following injury can coexist as apoptosis, necrosis, and hybrid forms along an apoptosis–necrosis continuum.¹⁸⁵

The damage associated with the sudden return of blood flow, or reperfusion, to the oxygen-starved, energetically compromised tissue is mechanistically and therapeutically important.¹⁸⁶ With timely reperfusion, either spontaneous or therapeutic, territory in the penumbra may be salvaged, but a delay in reperfusion will cause cell death. This phenomenon is termed “reperfusion injury.” It is held that much of this injury is produced during transient ischemia by inflammation, excitotoxicity, and apoptotic cell injury,¹⁷⁴ and can result in the disruption of the blood–brain barrier (BBB) and formation of brain edema.¹⁸⁷ An advantage of modeling stroke is that, unlike other neurodegenerative diseases, in stroke, the primary insult that triggers the pathology is known, and the models provide a highly reproducible, clinically relevant approach for studying this acute insult.

1. Animal Models of Stroke

Animal models are widely used to mimic human stroke. In contrast to other neurodegenerative diseases, there are excellent animal models of stroke that provide evidence for mechanisms, as well as prevention and treatment efficacy in vivo, that allow for control of collateral factors.

While cerebral ischemia can be modeled in large animals (pigs, baboons, etc.), rodent models are the most widely used. Although there are differences in anatomy and functionality between the rodent and the human brain, these models permit a careful dissection of mechanisms of injury and neuroprotection and allow examination of events that occur from onset of the insult to weeks or months after the ischemic event [for a review, see Ref. 188].

There are a variety of models that reflect the assortment of insults in humans. These models can be divided into three subgroups: global ischemia, focal ischemia, and hemorrhagic infarct. Global models, where the major blood vessels to the forebrain are occluded, are models of the consequences of cardiac arrest rather than stroke.¹⁸⁹ In focal models, a smaller arterial branch is occluded. Most models occlude the middle cerebral artery (MCA), as occlusion of this vessel is found in the majority of human ischemic strokes.¹⁸⁹ Global and focal occlusion models are either permanent or transient, allowing examination of the damage that results from both ischemia and reperfusion. Hemorrhagic stroke models involve the stereotactic injection of collagenase into the cerebral parenchyma.¹⁸⁹

An overriding question of much animal model research in the last decade has been the translational value of the models. The lack of efficacy of neuroprotectants against stroke in clinical trials raised the question of the validity of the animal models for preclinical research. A proof of confidence in the value of animal models is that changing several physiological parameters, like reperfusion, hyperglycemia, hyperthermia, or blood pressure, has similar effects on the outcome of stroke in humans and in animal models [for a review, see Ref. 189]. Nevertheless, although animal models mimic the human stroke pathology, small mammalian brains are quite different from human brains in function and morphology, and this may influence the efficacy of the therapies. This phenomenon was recently represented in an in silico model, which considered the differences in the percentage of white matter and glia between rodents and humans.¹⁹⁰ If this hypothesis is correct, the efficacy of the therapies should be apparent when animals with a more similar brain composition to humans are used for models.

From a practical, logistical perspective, it is difficult to treat patients at early time points; therefore, clinical investigators should focus on mechanisms that occur later in the ischemic cascade. However, there are also clinical settings where the risk of stroke is increased, such as cardiac surgery, where preventative treatments would be useful. Additionally, it is important to consider how much protection is achieved in preclinical trials, where a reduction of at least 50% of the ischemic damage has been the goal, and also to develop appropriate behavioral tests. There are multiple mechanisms occurring simultaneously and sequentially during stroke. Therefore, it is important to know the chronology and relevance of the different mechanisms that are occurring as a consequence of ischemia and reperfusion. The best treatment strategy might rely on a combination of neuroprotective agents, each geared toward a different death mechanism, administered at the appropriate time to interfere in the multiple ischemia-induced death pathways.

2. In Vitro Models of Ischemia

In vitro models have been utilized for over two decades in the study of cerebral ischemia.¹⁹¹ Much of our understanding of the pathogenesis of stroke derives from studies of cell lines, and primary and organotypic cultures exposed to hypoxia, anoxia, oxygen, and glucose deprivation. These models allow the dissection of pathways in a controlled setting. Of the models under study, the oxygen–glucose deprivation (OGD) model is believed to most resemble the in vivo ischemia setting.^192–194 Typical insults last for 15–90 min and are immediately followed by reperfusion, that is, addition of glucose to media and return to normoxic conditions.^192,193 Acute ischemic insults (10–20 min) are used for electrophysiological analysis in hippocampal neurons^195,196 as well as spiny/aspiny striatal neurons.^197,198 The discussion below gives a brief overview of work in hippocampal primary cultures or brain slices, unless otherwise stated.

Early evidence with the OGD model suggested that ischemic injury was mainly necrotic and mediated by glutamate excitotoxicity,^199–201 but subsequent work showed that apoptosis is masked by glutamate receptor activation in OGD in primary cultures.¹⁹³ Apoptosis in primary cultures is a delayed event (~24–36 h postinsult), while glutamate-mediated necrosis is observed as early as 2 h after the insult.¹⁹² As with some other insults, the extent of the insult, in this case OGD, determines how much necrosis occurs. Shorter periods of OGD (~15 min) result in most neurons dying by apoptosis, while longer periods of OGD (~75 min) do not alter apoptotic death, but increase the amount of necrosis.²⁰² The relative timing of apoptosis and necrosis are different in hippocampal brain slices exposed to OGD compared to primary cultures. In slices, apoptosis and glutamate-dependent necrosis both occur within 3 h after the ischemic episode.^203–205 These discrepancies may signify the importance of the neuron–glia context in studying ischemia in vitro.

The role of caspases has been studied in limited detail in OGD.^{202,204–207} These studies show that caspase-3 is activated by OGD and can be regulated by pan-caspase inhibitors. OGD cells treated with NMDA receptor antagonists undergo apoptotic death that can be inhibited by a pan-caspase inhibitor, z-VAD-fmk.^204,206 Another caspase inhibitor (Ac-YVAD-cmk) exhibited similar neuroprotection in OGD brain slices.²⁰⁵ These inhibitors are not specific for individual caspases,⁵⁰ and therefore, these studies do not clarify which caspases are important in promoting death in this ischemia model. Knockout of caspase-3 is also neuroprotective against OGD.²⁰⁸ Cleavage of caspases-3, -8, and -9 has been observed in microglia exposed to OGD,²⁰⁹ although the causality of the different caspases has not been determined. It has also been shown that cleaved caspase-7 and caspase-3 are found exclusively in microglia and neurons, and not astrocytes in OGD in mixed primary cultures,²⁰² again suggesting that the neuron–glia interactions are important in modeling ischemia.

As will be discussed below, caspase-1 and generation of interleukin-1β are found in animal models of ischemia. Activation of these pathways and subsequent neuronal death are found when SOD-1, the cytoplasmic superoxide dismutase, is downregulated.^210,211 This death is mediated by activation of caspase-1, secretion of IL-1β, and generation of peroxynitrite, all molecules that have been shown to be important in the progression of ischemia in vivo.^212–215 Thus, this model may provide a cellular model for investigation of the mechanism of neuronal death in ischemia.

3. Caspases in Stroke

In 1997, a function for caspases in mammalian neurodegeneration was first shown with the demonstration that a dominant-negative mutant of caspase-1 provided protection against tMCAo.²¹⁶ During the ensuing years, multiple studies have been published, and it is clear that caspases are key molecules in the death mechanisms induced during ischemia.^183,186 We will present data for caspases that are conserved in human and rodents.

a. Caspase-1.

As noted above, the first studies of caspase in neuronal disease showed that ablation of caspase-1 activity provided protection against tMCAo, but those studies used a dominant-negative caspase-1 that could block activity of other caspases in addition to caspase-1.²¹⁶ This work was followed with studies using caspase-1 null mice which exhibited reduced ischemic damage.²¹⁷ Indeed, several lines of evidence point to an important role for this caspase. The pro-inflammatory cytokine IL-1β, which is the main target of caspase-1 processing, is rapidly induced by focal/global ischemia.²¹⁸ Ablation of IL-1β expression provides more than 80% protection from ischemic damage.²¹⁴ Caspase-1 has been reported to be increased in the core of the stroke¹⁸⁰ and in the penumbra.²¹⁹ A recent review discusses the role of inflammation in acute neurodegeneration.³⁰

b. Caspase-2.

The initial characterization of the caspase-2 null mice showed that the caspase-2 null mice were not protected from tMCAo.⁶¹ A more recent study of caspase-2 and PIDD in global ischemia shows a role for PIDD in this model of ischemia.²²⁰ However, this study does not directly measure caspase-2 activity and, since the PIDD null mice appear to undergo caspase-2 dependent death^22,23 (although neurons have not been studied), it is not clear that caspase-2 is part of the ischemic death mechanism. But it must be considered that, since the caspase-2 null mice have been shown to have an increased expression of caspase-9 and Smac/DIABLO in the brain,⁶⁴ the compensations in death molecules might mask a role for caspase-2 in ischemia. Thus, the role of caspase-2 and of PIDD in ischemia requires further study.

c. Caspase-3.

Caspase-3 is present in the ischemic penumbra^59,180 and its deletion renders mice more resistant to ischemic injury.²⁰⁸ However, although caspase activation is generally described in the penumbra of focal infarcts, immunohistochemical analysis in MCAo has detected neurons containing caspase-3 in the infarct core.^221,222 In humans, although there is limited information available, there is a procaspase-3 increase in expression within hours resulting from permanent arterial occlusion,²²³ and activated caspase-3 and cleaved PARP have been detected in some neurons several days after cardiac arrest with reperfusion.¹⁸³ Cleaved caspase-3 has also been detected in glial cells 24 h postinfarction.²²⁴

d. Caspase-6.

Little data exists with regard to a role for caspase-6 in stroke. Studies of mRNA (see below) showed an increase in caspase-6 expression in rats subjected to pMCAo.

e. Caspase-7.

Much of the data with regard to caspase-7 is overlapping with the data about caspase-3. A clear, distinct role for caspase-7 in stroke has not yet been described. Studies of mRNA (see below) showed an increase in caspase-7 expression in rats subjected to pMCAo.

f. Caspase-8.

Increased expression of caspase-8 has been seen in both the core and in the penumbra,¹⁸⁰ as has increased expression of FasL.²¹⁹

g. Caspase-9.

A limited study of human brains obtained from 4 h to 5 days postinfarction suggests an increase in caspase-9 protein expression peaking at 24 h postinfarct.²²⁵ In rodent models, caspase-9 is increased in the penumbra.¹⁸⁰

h. mRNA expression.

Harrison et al.²²⁶ reported a strict transcriptional regulation of caspase expression following pMCAo in rats. There is a pattern of expression with an increase of caspase-1, -3, -6, -7, -8, and -11 mRNA, and a decrease of caspase-9 was detected at different time points. Caspase-2 mRNA shows no changes. But, in this report, the cellular types expressing these proteases were not studied.

Figure 6 presents an overview of what is presently known about the localization of different caspases during tMCAo. The timing of appearance and cell type location for the different caspases still needs further study. Actual measures of specific caspase activity induced by ischemia are also needed. The knowledge of the pathways and molecules that interconnect these caspases is still elusive, and more research has to be done in this area. Table II summarizes the evidence for a role for each caspase in stroke.

Fig. 6. Caspase activation in cerebral ischemia.

Caspases are activated in the ischemic core early in stroke and then in the penumbra as the stroke progresses.

TABLE II.

Caspases Implicated in Ischemia

Caspase	Paradigm	Measure	References
1	tMCAo pMCAo	Dominant-negative Casp1 Casp1 null mice ac-YVAD-AFC mRNA	180,216,217,226,242
2	Global ischemia	Caspase-2 cleavage	220
3	tMCAo pMCAO	Casp3 null mice WB clC3 ac-DEVD-AFC/WB/IHC mRNA	180,208,221,226
6	pMCAo	mRNA levels	226
7	pMCAo	mRNA levels	226
8	pMCAo	mRNA levels ac-IETD-AFC/WB/IHC	180,221,226
9	human stroke tissue tMCAo rodent tMCAo canine	IHC C9 Release of C9 from mitochondria	219,225,243

C. Linking Stroke and AD

In this chapter, we have summarized the current knowledge of the roles of caspases in AD and stroke. Caspases can either participate in the development of the disease or in the progression of the disease. Interestingly, a number of studies have revealed a positive correlation between stroke and AD. Cross-epidemiological studies show that patients who have suffered a stroke have an increased likelihood of developing dementia.^227–229 In a rodent ischemic model, tau hyperphosphorylation is present in the cortices.²³⁰ In terms of the relationship between stroke and amyloid, studies point to caspases as key players in modulating between these two conditions. Rodents subjected to tMCAo showed increases in APP and Aβ expression in the ischemic area, corpus callosum, thalamus, striatum, caudate-putamen, and hippocampus within days of reperfusion.^231–233 The increased expression levels are observed in both neurons and astrocytes in short-term studies ranging from days to weeks. However, over longer survival time, the levels of APP and Aβ gradually decrease.²³¹ This change may be explained by a downregulation of APP and Aβ in astrocytes and/or clearance of these peptides by astrocytes and microglia. Although the expressions of APP and Aβ decrease over time, in the thalamus, the expression profile of Aβ changes from diffused to plaque-like deposits.²³² In sum, these rodent studies point to an initial upregulation in APP and Aβ in a diffused pattern in various brain regions following MCAO. Over time, the levels of these proteins are reduced. However, there remains residual damage in the form of plaque-like structures.

The mechanism for elevated APP and Aβ in brain regions following stroke has begun to be addressed by investigators. Administration of zDEVDfmk to rats prior to tMCAo blocked the appearance of cleaved caspase-3 and abrogated the increased expression of BACE1, the enzyme that produces Aβ from APP, and Aβ.^233,234 Similar effects were seen in a model of brain trauma that shows increased Aβ levels.^233,234 While these studies do not specifically examine caspase-3 function, they do implicate activation of caspases in the production of Aβ, providing a mechanism underlying the clinical observations that stroke or head trauma are risk factors for the development of AD.

V. Future Directions

There is an increasing appreciation that caspases have a variety of actions, both death and non-death-related. With this in mind, it is even more important that our understanding of the specific functions of individual caspases in health and disease is increased. In this way, appropriate disease interventions that do not interfere with the non-disease-related actions of caspases can be devised. In order to progress in our understanding of caspase functions, attention must be directed to the tools available to study caspases and the development of better tools for analysis. Multiple studies have now shown that many of the tools used to study caspases are inadequate, yet they continue to be used and to confound the literature. A search of the current literature will show how many studies still use the pseudopeptide inhibitors/substrates as evidence for the function of specific caspases. Many studies still equate caspase cleavage by Western blotting with activity, or caspase-3 cleavage by immunocytochemistry with apoptosis. As of today, the best tools available are molecular manipulation and affinity ligands.

Another issue in the study of human disease is the models available. In the two diseases that we have used to illustrate current knowledge of caspases in neuronal disease, there are different issues. The in vivo models of cerebral ischemia mimic the human disease quite accurately and provide excellent systems for studying death mechanisms. The cellular models of ischemia, while providing the advantages of cell-based models in the study of biochemical and molecular events, do not offer the complexity of the various elements clearly involved in the evolution of the ischemic event. For AD, the in vivo models available at best model preclinical disease. There is no adequate model to mimic the progression of the disease. The current models have little or no neuronal loss, making the study of death mechanisms in these animals difficult. Thus, at present, the cell models and the hippocampal slice systems offer the best models for studying death mechanisms in AD. Better models and more specific tools are in development and are eagerly awaited by the field.