The Aβ (amyloid-β peptide) has long been associated with Alzheimer's disease, originally in the form of extracellular plaques. However, in the present paper we review the growing evidence for the role of soluble intracellular Aβ in the disease progression, with particular reference to Aβ found within the mitochondria. Once inside the cell, Aβ is able to interact with a number of targets, including the mitochondrial proteins ABAD (amyloid-binding alcohol dehydrogenase) and CypD (cyclophilin D), which is a component of the mitochondrial permeability transition pore. Interference with the normal functions of these proteins results in disruption of cell homoeostasis and ultimately cell death. The present review explores the possible mechanisms by which cell death occurs, considering the evidence presented on a molecular, cellular and in vivo level.
AD (Alzheimer's disease) is the most common neurodegenerative disease of the elderly. AD is clinically characterized by progressive loss of declarative memory, which in turn impairs functions such as memory, language, perceptual skills, attention, orientation and problem-solving abilities. The loss of these brain functions eventually results in the patient's complete social dependence and inevitable death. The main feature of the disease progression is extensive death of neurons, starting in the entorhinal cortex and hippocampus, before proceeding to other parts of the brain cortex and subcortical grey matter. It is in these brain regions where extracellular amyloid plaques, mainly consisting of Aβ (amyloid-β peptide), and intracellular neurofibrillary tangles, caused by aggregation of the hyper-phosphorylated microtubule-associated protein tau, are found . Since the first description of the disease by Alois Alzheimer in the early 20th century, much work has been done to identify the molecular basis of the disease [2,3]. It is now clear that AD can either occur sporadically or be an inherited disease caused by mutations in genes encoding proteins involved in Aβ turnover. Sporadic AD occurs later in life, generally after the age of 65, whereas familial AD tends to have an earlier age of onset.
Aβ is produced from the transmembrane APP (amyloid precursor protein) by the sequential actions of the aspartate proteases β- and γ-secretase . β-Secretase sheds the N-terminal domain of APP, leaving a 99-residue fragment in the membrane. This fragment is further cleaved by γ-secretase at one of several sites within the transmembrane region to produce the Aβ peptide and the AICD (APP intracellular domain), which is released into the cytosol. The exact location of the γ-secretase cleavage determines the final size of the Aβ peptide, which is most commonly either 40 or 42 residues long, producing Aβ-(1–40) or Aβ-(1–42), respectively. An alternative non-amyloidogenic cleavage pathway exists in which α-secretase cleaves inside the Aβ region of APP thus preventing production of the Aβ peptide . The mutations observed in familial (hereditary) AD generally occur in one of the components of the amyloidogenic pathway (i.e. APP, β-secretase or γ-secretase) and lead to the increased production of Aβ, especially the more aggregation prone Aβ-(1–42) [5,6]. In contrast, tau-pathology results in the formation of intracellular neurofibrillary tangles and is considered by some to be a process associated with the later stages of AD caused directly or indirectly by the amyloid pathology [7,8].
A number of risk factors for the development of sporadic late-onset AD have been identified. These include carrying the ε4 allele of the ApoE (apolipoprotein E) gene  (recently reviewed in ), mutations in the gene encoding the membrane sorting receptor sortilin-1 [11,12] and increased levels of the non-proteinogenic amino acid homocysteine [13,14]. It has been suggested that these factors are indirectly associated with an increased production of Aβ in neuronal cells. Very recently two new genes, CLU [clusterin, also known as ApoJ (apolipoprotein J)] and PICALM (phosphatidylinositol-binding clathrin assembly lymphoid-myeloid leukaemia gene), have been linked to AD . However, the detailed mechanisms and intracellular processes resulting in the development of sporadic AD remain unclear.
There have been many excellent reviews of Aβ production and the different forms and structures of this unusual peptide [4,16–18]. Most of these reports have predominately been concerned with extracellular production and accumulation of Aβ and with how understanding this process is important in the prevention of plaque formation as a potential therapy. In the present paper we will discuss recent findings and implications of the accumulation of Aβ inside cells. Recently it has been realized that the presence of extracellular amyloid plaques is not a good indicator of disease state and that increased levels of intracellular Aβ, predominantly Aβ-(1–42), more accurately reflects the stage of neurodegeneration [6,19]. As a result of this finding, a hypothesis describing a ‘mitochondrial spiral’ of neurodegeneration in AD has been proposed  and subsequently much experimental work has substantiated the idea that mitochondria play a crucial part in the disease progression [21,22]. The importance of this new approach has been highlighted by the limited success of vaccination trials against Aβ . Although they have been shown to clear extracellular Aβ, by their nature these approaches do not clear the intracellular Aβ.
The present article summarizes the current understanding of the intracellular location of Aβ and reviews the evidence for its existence in mitochondria. We discuss how intracellular Aβ might exert its neurotoxic function, with a focus on its known intracellular binding partners. In particular, we will present recent evidence for the interaction of Aβ with two mitochondrial proteins, ABAD (amyloid-binding alcohol dehydrogenase) and CypD (cyclophilin D), both of which have been implicated in AD. The structure and function of ABAD and the consequences of its interaction with Aβ are discussed. Finally, we address how characterizing the interaction of Aβ with ABAD and CypD has highlighted potential therapeutic targets for the treatment of AD.
SOURCE OF INTRACELLULAR Aβ
The presence of APP and Aβ peptides within neuronal and non-neuronal cells has been reported by numerous researchers (reviewed in , see Figure 1). Aβ immunoreactivity has been located to the secretory pathways and cytosol of both neuronal and non-neuronal cells [24,25], the outer membrane of multivesicular bodies of cultured neurons [26,27], endosomes and lysosomes [28,29] and the mitochondria of neuronal cell cultures and in both the murine and human brain [30,31]. In all cases, the most studied sites of Aβ production are within membranes, such as in the plasma membrane [2,32], the secretory pathway  and in the endosomal compartment [34,35], where APP and all components of the cleavage machinery have been found. However, due to the orientation of APP and the secretases, it has been more difficult to explain the presence of Aβ peptides in the cytosol. Aβ can integrate into lipid membranes at high concentrations , possibly leading to the observed loss of the integrity of endosomes and lysosomes in the cell; subsequently Aβ can then leak out of these compartments . This possibility is substantiated by the known membrane-disordering capabilities of Aβ [37,38]. However, the mechanism that regulates the traffic of Aβ in and through membranes, as well as its subsequent significance for AD, still remains obscure.
APP and Aβ inside the cell
SPECIES OF INTRACELLULAR Aβ
Studies on human brains revealed that the majority of intracellular Aβ comprises the 42-residue peptide Aβ-(1–42) [39,40]. Immunohistochemical studies have shown that the levels of intracellular immunoreactivity against Aβ-(1–42) are reduced in more advanced stages of AD when plaque burden and cognitive dysfunction become more prominent, pointing towards a role for Aβ-(1–42) in the early stages of the disease . Accordingly, comparison of the effects of Aβ-(1–42) and Aβ-(1–40) in several studies has revealed that the activation of intracellular signalling events and ROS (reactive oxygen species) production is more pronounced with Aβ-(1–42) or when APPm (mutant APP), which produces higher levels of Aβ-(1–42), is expressed [41,42].
It appears that the detectable form of intracellular Aβ reflects the state of neuropathology. In post-mortem human AD brains, intracellular Aβ has mainly been described as aggregates in the cell soma and perinuclear region, which do not stain with Congo Red or other β-sheet-selective stains . Zhang et al.  have tested the toxicity of Aβ-(1–42) in different aggregation states when microinjected into the cytosol. They found that the non-fibrillized (monomers, dimers and oligomers) and fibrillized Aβ-(1–42) are both toxic to neurons, whereas Aβ-(1–40) fibrils or non-fibrillized peptides do not cause neuronal cell death. Again, this result suggests that Aβ monomers and oligomers can be neurotoxic before intracellular fibrils are formed.
Park et al.  have shown that there is accumulation of GFP (green fluorescent protein)-tagged Aβ and the APP-C99 fragment in undefined perinuclear aggregates in both H4 neuroglioma cells and HEK (human embryonic kidney)-293 cells. They correlated the formation of these aggregates with the recruitment and attenuation of the proteosome and with apoptotic cell death. These findings agree with the results of a study by Bückig et al. , who observed perinuclear accumulation of Aβ which co-localized with ubiquitin in CHO (Chinese-hamster ovary) K1 cells. Yoon et al.  not only detected GFP-tagged Aβ in perinuclear aggregates following transfection into H4 neuroglioma cells, but also found SOD1 (superoxide dismutase 1) associated with the aggregates, which they reported to interact selectively with Aβ. Taken together, these results suggest a defined pathological role for monomers or low-number oligomers of Aβ inside the cell, whereas the observed perinuclear Aβ aggregates might be indicative of later stages of neurotoxicity.
Mitochondrial Aβ was first described in detail by Lustbader et al. in 2004 . The study also confirmed the interaction of Aβ with ABAD and showed, by immunoelectron microscopy, that the proteins co-localize inside mitochondria from the human AD brain. Subsequently, Caspersen et al.  also confirmed that Aβ accumulates in the mitochondria of APPm-expressing transgenic mice and in brains from AD patients, but to a lesser extent in non-transgenic mice and brains from non-demented subjects. Using immunoelectron microscopy and Western blot analysis, they detected Aβ-(1–42) and Aβ-(1–40) in AD-affected brain mitochondria co-localizing with the mitochondrial matrix chaperone Hsp60 (heat shock protein 60), although Aβ-(1–42) appeared to be the more abundant form . Although it was shown that Aβ peptides were in close proximity to Hsp60 by immunoelectron microscopy, the protease-protection assay on whole mitochondria used in the study did not discriminate between the matrix compartment and the intermembrane space or the inner mitochondrial membrane . Hence, the possibility of an association with the inner mitochondrial membrane, which has been found by other researchers using digitonin-treated mitochondrial preparations  or mitoplasts produced by osmotic shock , could not be ruled out. However, an inner mitochondrial membrane location seems to be incompatible with the interaction of Aβ with ABAD, which is located in the mitochondrial matrix . Thus it is currently unclear whether Aβ accumulates inside the mitochondrial matrix in the AD-affected brain or whether it is found exclusively in the membrane compartment.
A possible reason why Aβ has not been detected in the mitochondrial matrix might be its rapid degradation by mitochondrial proteases, such as PreP (prolyl endopeptidase). PreP is a thiol-sensitive metalloprotease that resides in the mitochondrial matrix fraction and is able to rapidly degrade different forms of Aβ . Its sensitivity to oxidative inactivation has been demonstrated in vitro , but so far deregulation or dysfunction of PreP in AD have not been reported.
Another Aβ-degrading enzyme, IDE (insulin-degrading enzyme), has been genetically linked to AD since 1998 [50,51]. Like PreP, IDE is an intra- and extracellular metalloprotease belonging to the pitrilysin family of peptidases . Inside cells, IDE is located in the cytosol and the mitochondria, where it degrades mitochondrial-targeting sequences cleaved by the mitochondrial processing peptidase . The relevance of IDE for AD was confirmed by the finding that IDE and Aβ can interact in vitro to form complexes that are stable to denaturing conditions, in both rat and human AD-affected brains . Moreover, nitrosative stress was recently reported to compromise its enzyme activity in a non-competitive manner by S-nitrosylation of essential cysteine residues . This post-translational modification has emerged as an important factor in neurodegenerative processes [55,56]. It is probable that IDE is able to degrade Aβ inside mitochondria and that it is affected by oxidative and nitrosative stress brought about by higher levels of Aβ during AD pathology. The possible compensatory increase of IDE protein levels in AD-affected brain regions seen in a transgenic mouse model  supports these ideas. The up-regulation of this protein, which was only observed after the occurrence of plaque pathology in an AD mouse model, and an up-regulation of the peptidase neprilysin , suggests that its contribution to very early events in disease pathogenesis is not significant. Nevertheless, mis-sorting or malfunction of IDE have to be taken into account as potential risk factors for the progression of AD [57,58] or other neurodegenerative diseases. Further investigation into the control of sorting and function of intracellular IDE will shed light on this aspect of AD and elucidate at which stage IDE is involved in the disease progression. In summary, it is possible that Aβ in the mitochondrial matrix has not been detected due to its rapid degradation.
HOW IS Aβ TRANSPORTED/PRODUCED INSIDE MITOCHONDRIA?
The origin of mitochondrial Aβ is still a matter of debate. There is experimental evidence for both the local production and/or the import of Aβ from the cytosol as possible explanations for its occurrence.
Localized Aβ synthesis
APP is targeted to mitochondria in cell-culture systems and in human brain due to its chimaeric N-terminal targeting sequence which causes the peptide to translocate to the ER (endoplasmic reticulum) and mitochondria [30,47]. In earlier studies, Anandatheerthavarada et al.  demonstrated that sorting of other proteins with chimaeric targeting sequences, to either the ER or mitochondria, can be regulated by phosphorylation, but no such phosphorylation has been described for the N-terminus of APP. Co-staining, immunoblotting and immunoprecipitation experiments only detected APP in a transmembrane-arrested form, in contact with the mitochondrial translocases of the outer and inner membrane ; APP was found in an N-in C-out orientation spanning the mitochondrial intermembrane space with a C-terminal 73 kDa fragment facing the cytosol, and the transport arrest was associated with mitochondrial dysfunction . The authors therefore proposed that the acidic domain, spanning amino acids 220–290 of APP-695, would hinder transfer . A correlation between the amount of membrane-arrested APP and the presence of the ApoE ε4 allele in AD patients has been observed .
The components of the γ-secretase complex [nicastrin, presenilin, APH-1 (anterior pharynx-defective homologue 1) and PEN-2 (presenilin-enhancer 2 homologue)] have been detected in mitochondria . Dual targeting of nicastrin was proposed, based on its amino acid sequence, and its targeting to mitochondria was demonstrated by immunoelectron microscopy. The other components of the γ-secretase complex were also shown to be present in mitochondria by electron microscopy . In summary, the components necessary for the production of Aβ in mitochondria, except the β-secretase, have been detected locally. However, it has also been reported that the topology of APP detected in mitochondria, with the Aβ region located in the intermembrane space , would not be suitable for cleavage by the γ-secretase complex, which is an intramembrane cleaving protease [32,61].
Interestingly, another enzyme, HtrA2 (HtrA serine peptidase 2), has been found to act on APP in the mitochondrial intermembrane space and sheds the N-terminal portion of the protein. HtrA2 is a serine protease which was known to interact with Aβ and APP-C100 fragments in cell cultures . This protease is able to cleave APP at amino acid 534 and produces a 161-residue long C-terminal fragment, which includes the Aβ region of APP, which is released into the cytosol of APP-transfected cells . The significance of this cleavage for APP metabolism and Aβ processing is not yet clear; it might represent a mechanism to clear membrane-arrested APP from the mitochondria, as indicated by the authors . However, despite this C-terminal domain cleavage, the N-terminal part of APP, with the acidic domain, would still be arrested in the inner mitochondrial membrane, disturbing mitochondrial protein transport and metabolism . The role of the two fragments and their impact on mitochondrial and cellular metabolism still remains to be elucidated.
Given the topological problems of Aβ production in the mitochondrial membrane, one possible mechanism by which it could accumulate is via cleavage in MAMs (mitochondria-associated membranes). These membrane compartments represent close contact points between the ER and mitochondria , where lipids and membrane proteins are thought to be exchanged directly between the organelles . Presenilin 1 and 2 have been detected in this compartment , so it is possible that APP can be cleaved while residing in the ER membrane and that Aβ would subsequently be transported into mitochondria . Another possibility is the transfer of APP or β-site-cleaved APP from the ER to the mitochondrial outer membrane, alongside lipids that are also transferred through MAMs . The APP could subsequently be cleaved by the γ-secretase in the mitochondrial membrane [60,66], releasing Aβ on the mitochondrial side. The mitochondrial proteins DLP1 (dynamin-like protein, also known as Drp1) and Mfn1/2 (mitofusin 1/2) are known to be involved in mitochondrial fission and fusion . These proteins are thought to modulate the ER–mitochondrial functional link at MAMs  and have been reported to be down-regulated in an AD cell-culture model  and in post-mortem human AD brains . This down-regulation was shown by immunohistochemistry to result in the redistribution of mitochondria away from axons to the cell soma in human AD hippocampus and to negatively affect dendritic spine formation in primary mouse hippocampal neurons . Further investigations are clearly needed to address whether Aβ can be produced in the mitochondrial membrane, whether cleavage might take place in the MAM compartment and what the resulting implications are for AD.
Direct import of Aβ
Aβ itself can be imported into mitochondria via the translocase system and is consequently found within the mitochondrial cristae associated with the inner mitochondrial membrane . Hansson Petersen et al.  demonstrated mitochondrial localization of Aβ in mitochondrial preparations from in vivo human brain biopsies from non-demented patients. Significantly, this result indicates that Aβ is indeed present in mitochondria of the non-demented human brain, albeit at lower levels. Subsequent import studies on a human neuroblastoma cell line further revealed that Aβ can be imported into mitochondria from outside the cell and that import is independent of the mitochondrial membrane potential and involves the TOM (translocase of outer mitochondrial membrane) transporters TOM20, TOM40 and TOM70 .
Recent genetic studies on AD and non-demented human brain samples revealed another interesting link between the mitochondrial import machinery and AD [69,70]. In these studies, SNPs (single nucleotide polymorphisms) in the TOM40 gene on chromosome 19, which is directly proximal to the ApoE gene, were linked to an increased risk for AD. In an initial study looking at SNP haplotypes, the polymorphisms were not linked to a risk of AD independently from the ApoE ε4 allele . However, in a later study comparing diplotypes of these polymorphisms, one of these SNPs was found to contribute to a significantly increased risk for AD . Bekris et al.  also observed that these SNPs were associated with the level of ApoE in the cerebrospinal fluid. It is therefore possible that a SNP in the TOM40 gene might play a role in increasing the risk of developing AD, and further research is needed to clarify if this represents an independent risk factor and how this risk is realized at the molecular level.
BINDING OF Aβ TO INTRACELLULAR PROTEINS
Having established that Aβ is present within cells, it is of interest to consider its resulting intracellular action. The build-up of Aβ within cells has been found to affect the expression or activation of several signalling proteins within cells, such as the stress kinases of the JNK (c-jun N-terminal kinase) pathway , NF-κB (nuclear factor-κB) , the Ca2+-dependent metalloprotease calpain , the pro-apoptotic Bcl-2 protein family member Bim (Bcl-2-interacting mediator of cell death) , Akt/PI3K (phosphoinositide 3-kinase)  and CREB (cAMP-response-element-binding protein) . Intracellular Aβ has also been found to impair cellular metabolism by disturbing mitochondrial respiration [46,78]. Some of the intracellular effects, for example the activation of the p65 subunit of NF-κB, are indirect and are believed to be a result of the production of ROS . Other studies suggest direct interactions of mono- or oligomeric Aβ with intracellular proteins. These specific interactions are probably critical events in AD (see below); they might represent early steps in the development of this disorder.
PDK (phosphoinositide-dependent kinase)
In a recent study, Lee et al.  investigated the effect of Aβ on the expression and activity of PDK and its target, Akt, both in vitro and in vivo. Involvement of the PDK/Akt kinase pathway in AD had been suggested previously, when it was observed that Akt expression levels were reduced in APPm-expressing cell cultures and in lymphoblast cells from familial AD patients . Additionally, it was known that overexpression of Akt in PC12 cell cultures attenuated the apoptotic effect of extracellular Aβ . Evidence also existed that the PI3K/Akt kinase pathway can regulate levels of IDE , which might play a role in the degradation of Aβ inside mitochondria  and in the extracellular space , as described above.
Lee at al.  found decreased activation of Akt in human AD brain compared with control aged brain when they assessed the level of PDK-mediated phosphorylation. The in vitro activity of Akt from Aβ-expressing myotube or neuroblastoma cell cultures on a GSK (glycogen synthase kinase)-3 fusion protein was selectively reduced when compared with controls. PDK-dependent activation of Akt kinase activity in vitro was also diminished in the presence of Aβ. However, no direct binding studies with PDK, Akt and Aβ have been performed. Co-immunoprecipitation experiments have demonstrated the increased association of Aβ with Akt and PDK by Western blotting. However, dissociation of Akt from PDK was observed in preparations from AD-affected brain. The authors concluded that Aβ selectively interferes with the interaction between PDK and Akt, and therefore, Akt phosphorylation . A detailed description of the interaction between Aβ, PDK and Akt on the biochemical level has so far not been provided, but this will be an important step in understanding its molecular causes and consequences. It is clear that Akt inhibition would inevitably result in a lack of pro-survival signalling in the cell and contribute to neurodegeneration. However, the question remains whether these processes are specific for late stages of AD or also occur in early phases of the disease.
Yoon et al.  have demonstrated an interaction between Aβ and Cu–Zn SOD1 in a human neuroglioma cell line by co-immunoprecipitaion and co-localization. This interaction resulted in the inhibition of in vitro SOD1 catalytic activity and this effect was even stronger with a G93A mutant of SOD1 , which is implicated in familial cases of amyotrophic lateral sclerosis [81a]. At the same time, proteins involved in other neurodegenerative diseases, such as α-synuclein or its fragment NAC (non-Aβ component of AD amyloid) did not co-immunoprecipitate with SOD1. Co-localization studies revealed that GFP-tagged Aβ and SOD1 are eventually found in aggregates in the perinuclear region of cells . Similar Aβ aggregates have recently been reported to attenuate the actions of the proteasome and lead to mitochondrially induced apoptosis in neuronal cells . It is logical to conclude that inhibition of a key antioxidant enzyme in the cell will also lead to increased oxidative stress in the cell, which is known to cause neurodegeneration and memory deficits [82,83]. However, at present, there has been no evidence for a direct in vivo interaction of Aβ with SOD1. Despite this, the significance of SOD1 in AD is supported by an earlier study, which showed that the β-site cleavage enzyme BACE1 (β-site APP-cleaving enzyme 1) is able to bind Cu2+ and interact with CCS (Cu2+ chaperone of SOD1) in cell cultures and normal rat brain. The authors suggested that BACE1 levels could therefore control SOD1 activity by competing for the limited pool of CCS in cells . Additionally, these findings link an enzyme involved in Aβ production with SOD1 both in vitro and in vivo, and support the view that SOD1 could play an active role in the pathogenesis of AD.
It is also important to note that the manganese-dependent mitochondrial superoxide dismutase, SOD2, has been implicated in AD. Li et al.  found elevated Aβ levels and an increased plaque burden in APPm-expressing mice when they lacked one allele of the SOD2 gene. Two recent studies have also shown that overexpression of SOD2 reduces oxidative stress and memory deficits in a transgenic mouse model for AD [86,87]. However, no direct interaction between Aβ and Mn–SOD2 has been reported.
Another protein that was reported to directly interact with Aβ peptides is the hydrogen peroxide-degrading enzyme, catalase. There is in vitro evidence for an interaction between isolated catalase and biotinylated Aβ, which inhibits the catalase's enzymatic activity [88,89]. Milton  characterized the binding of Aβ to catalase and found that Aβ-(1–42) and Aβ-(25–35) bound to catalase with a Kd of 3.3±0.02 nM, whereas Aβ-(12–28) did not bind. The author suggested that the inhibition of catalase activity was caused by oxidation of the enzyme, as the effect of Aβ could be relieved in the presence of ethanol or NADPH . Moreover, the cytotoxic effect of extracellular Aβ was enhanced when catalase was inhibited by 3-aminotriazole. . Applying an antisense peptide approach, Milton et al.  later identified that amino acids 400–409 of the human catalase protein can interact with Aβ. When testing the Aβ-binding affinity of a peptide representing this region, a Kd of 1.2±0.1 nM was determined, which is similar to the Kd identified for binding of Aβ to catalase. The peptide was also able to protect from some of the adverse effects of Aβ on a myeloma cell line . Despite these results, it is not clear whether inhibition of catalase by Aβ is relevant for physiological and pathological processes in AD. Notably these studies used concentrations of up to 20 μM Aβ and the aggregation state of the different amyloid peptides is undefined. This concentration is important, as it is known that different Aβ states (i.e. monomers, oligomers or fibrils) can have different effects on cells and this also turns out to be true for different concentrations of Aβ [73,77]. It would be interesting to see whether catalase also interacts with Aβ peptides in a cellular environment and at lower concentrations and if binding of Aβ to antioxidant proteins, like SOD and catalase, turns out to be a common theme.
To date, the most characterized intracellular Aβ-binding protein has been ABAD. ABAD was first identified as an Aβ-binding protein in 1997 using a yeast two-hybrid screen  and later, in a separate study, as the human analogue of a newly discovered bovine hydroxyacyl-CoA dehydrogenase type II . As ABAD was originally identified within the ER [92,94], it was initially termed ERAB (ER-associated amyloid-binding protein) . However, it was identified later within mitochondria [92,94–96] and it has been suggested that the distribution of ABAD in cells may be cell-line-dependent . ABAD is also known by a number of other names, including SCHAD (human brain short chain L-3-hydroxyacyl-CoA dehydrogenase) [97,98], HSD10 (17β-hydroxysteroid dehydrogenase) [99,100] and HADH II (human hydroxyacyl-CoA dehydrogenase type II) [101,102]. This protein is expressed in all tissue types, particularly in the heart and liver, and was also found to be expressed in all regions of the brain .
Structure and function of ABAD
ABAD is a multifunctional enzyme catalysing the reduction of aldehydes and ketones and oxidation of alcohols (1) and as such it is known to act on a broad range of structurally diverse substrates, including simple alcohols [97,101,102], steroids , hydroxysteroids [95,97,101] and 3-hydroxyacetyl-CoA derivatives, such as acetoacetyl-CoA [95,101,102] and D-β-hydroxybutyrate .
Reduction and oxidation of alcohols and ketones by ABAD
Table 1 lists the experimentally determined enzymatic parameters for a range of these substrates. Unsurprisingly, different substrates have different reaction rates with the enzyme, indicating that although ABAD is able to catalyse reactions on a number of different substrates, some have a higher turnover than others. It can also be seen that widely ranging values for enzyme activity have been reported for the same substrate, often varying by several orders of magnitude. However, comparisons between values obtained may be complicated due to the range in conditions used during assays used to study activity. Despite this, the results are still an indication of the potentially wide range of roles that the enzyme is able to perform within the cell. It is, however, important to note that an enzyme's ability to metabolize a particular substrate in vitro does not necessarily guarantee that it does so in an in vivo environment. Given the mitochondrial location of ABAD, one of its main functions is thought to be in energy production and metabolic homoeostasis, notably the third step of β-oxidation of fatty acids, utilizing its function as an L-3-hydroxyacyl-CoA dehydrogenase [93,102]. This role may be especially important in glucose-deficient environments, where other energy sources become more significant. For example, it has been found that the overexpression of ABAD in COS cells increases the ability of the cell to utilize ketones, such as D-β-hydroxybutyrate, in the absence of other energy sources . Similarly, transgenic mice overexpressing ABAD showed increased utilization of D-β-hydroxybutyrate compared with non-transgenic animals, indicating their better adaptability to metabolic challenges .
|Substrate||Reference||Co-factor||Specific activity (μmol·min−1·mg−1)||Vmax (μmol·min−1·mg−1)||Km (μM)||kcat (s−1)|
|Substrate||Reference||Co-factor||Specific activity (μmol·min−1·mg−1)||Vmax (μmol·min−1·mg−1)||Km (μM)||kcat (s−1)|
A proposed alternative role to energy homoeostasis is in the metabolism of hydroxysteroids, such as oestradiol . The role of ABAD in metabolizing sex steroids could be significant, as it is documented that women are more likely to suffer from AD than men , and that postmenopausal hormone replacement therapy can prove beneficial in delaying the onset of the disease . ABAD has also been shown to oxidize steroid modulators of the GABAA receptor (GABA is γ-aminobutyric acid) to give inactive metabolites, and as such it has been suggested that these compounds are better substrates for ABAD than the sex steroids  (see Table 1).
ABAD is known to play a role in the degradation pathway of isoleucine. In clinical cases of MHBD (2-methyl-3-hydroxybutyryl-CoA dehydrogenase) deficiency, i.e. deficiency of the enzyme catabolizing the penultimate step in isoleucine degradation, two missense mutations within ABAD were identified in patients presenting with MHBD deficiency; Arg130 was mutated to a cysteine residue in four patients and was found to cause neurological deficits, loss of mental and motor skills and psychomotor retardation, whereas a Leu122 to valine substitution, identified in a single case, presented with only psychomotor retardation . The mutations were shown to either fully (R130C) or greatly (L122V) inactivate the enzyme; in addition, the R130C mutation was also thought to reduce the enzyme's stability, causing the lower protein levels observed in these patients .
It has been proposed that in the absence of Aβ, ABAD is able to play a cytoprotective role during periods of stress. For example, in mouse models of ischaemic stress (stroke), ABAD expression was found to be increased in both ABAD-overexpressing and non-transgenic mice following 45 minutes of transient middle cerebral artery occlusion. However, the transgenic animals showed fewer effects of the stroke, including fewer neurological deficits and increased ATP levels  and were hence thought to be protected to some degree by the elevated levels of ABAD. Conversely, ABAD levels were shown to be decreased in the ventral midbrain of PD (Parkinson's disease) patients, as well as in the ventral midbrain of MPTP (1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine)-treated mice, used as a model of PD. However, MPTP-treated mice overexpressing ABAD were protected against apoptosis and the loss of dopaminergic neurons in this brain region, suggesting that this enzyme can protect against neurodegeneration .
The crystal structure of ABAD is well documented, with several structures of the enzyme published. These include structures in complex with its co-factor NAD+ , a human mutant version complexed with an inhibitor  and the human protein in complex with Aβ . From these structures, information on the catalytic mechanism and its interaction with Aβ has been deduced.
ABAD was found to form a homotetramer, both in solution and in the crystal form, which was made up of four identical single domain monomers of 27 kDa each. Tetramerization has been shown, by molecular modelling, to stabilize the binding interface region . The conserved catalytic triad of Ser155, Tyr168 and Lys172 is also found in the active site of other short-chain dehydrogenase reductase enzymes [45,99]; mutation of these residues to glycine inactivates the enzyme . In the reduction of a ketone to an alcohol, the hydrogen atom of Tyr168 is thought to co-ordinate to the carbonyl of the ketone substrate, increasing the electrophilicity of the carbonyl carbon atom. It is proposed that the ammonium group of Lys172 interacts with the hydroxy group of Tyr168, increasing the acidity of this residue. The hydride that effects reduction is donated to the activated carbonyl by the NADH co-factor, leading, simultaneously, to deprotonation of Tyr168 by the newly formed hydroxy group. The hydroxy group of Ser155 is able to form a hydrogen bond with the deprotonated tyrosine, stabilizing the resulting negative charge (Figure 2).
Representation of the catalytic core of ABAD
Structures of rat ABAD with either 3-ketobutyrate or 17β-oestradiol showed that the two substrates bound in similar positions within the active site . These structures confirm the close proximity of the presumed catalytic residues to the substrate molecule. Liu et al.  have investigated the mechanism of the related enzyme HAD (3-hydroxyacyl-CoA dehydrogenase) using fluorinated substrates and have determined that the oxidation of hydroxyacyl-CoA-linked substrates proceeds through an enolate intermediate, stabilized by Asn208 and Ser137, and that this enolate formation is essential for the reaction to occur. Given the similarity of the active-site residues, it is possible that oxidation of similar substrates by ABAD might occur through a similar mechanism.
The crystal structure of rat ABAD indicated the presence of an ‘active-site loop’ from residues 202–220, which encloses the active site in the presence of substrate, preventing further access by solvent molecules. Thr208 of this loop would therefore be able to form a strong hydrogen bond with the substrate . Molecular modelling suggested that this loop is in an open conformation in the absence of substrates, whereas the presence of NAD+ induces the conformational change to a ‘closed’ position .
Compared with other HAD type I short-chain enzymes, both rat  and human  ABAD were found to have two significant insertions between residues 100–110 and 140–150. A model of the expected binding of a CoA-linked substrate to the active site suggests an interaction between the negatively charged phosphate groups of CoA and a group of positively charged residues within the 100–110 region (namely Lys99, His102, Lys104 and Lys105) [99,102]. CoA-linked substrates were found to be more efficiently oxidized than their non-CoA analogues . The region connecting the proposed CoA-binding site and the active site is lined with hydrophobic residues, which is consistent with the binding of aliphatic chains. In contrast, few interactions were observed outside the active site with 17β-oestradiol as the substrate , corroborating the idea that sex steroids are not the main substrates for ABAD.
The initial identification of ABAD as an intracellular binding partner of Aβ was based on a yeast two-hybrid screen , which identified four positive clones (one from human brain and three from HeLa cells), all of which had the same cDNA sequence. Radiolabelled ligand-binding studies confirmed the interaction of ABAD and Aβ and a Kd of 88 nM was determined . Subsequently, a number of techniques have been employed to demonstrate the interaction between ABAD and Aβ, including ELISA , crystallography , SPR (surface plasmon resonance) [45,112], co-immunoprecipitation [45,92, 95] and immunocytochemistry followed by confocal microscopy .
Both Aβ-(1–40) and Aβ-(1–42) were found to inhibit the activity of purified ABAD, with Ki values of 1.2–1.6 μM for the reduction of acetoacetyl-CoA, [94,101], 2.6 μM for the oxidation of octanol  and 3.2 μM for the reduction of 17β-oestradiol . Studies by Oppermann et al.  showed that residues 13–22 of Aβ were critical for inhibiting ABAD activity, a region that is also characterized by its fibril-forming properties (residues 16–20). It is interesting to note that inhibition of ABAD requires micromolar concentrations of Aβ, whereas binding has been shown to occur in the nanomolar range [92,112]. This result implies that Aβ monomers alone are not sufficient to induce inhibition. Perhaps further aggregation of Aβ is necessary to alter the conformation of the enzyme to an extent that its active site is distorted or, more simply, that an aggregation of Aβ on the surface sterically hinders access of substrate to the active site.
The crystal structure of ABAD in complex with Aβ lends support to the first hypothesis that the enzyme shape is distorted upon binding to Aβ . Compared with other published ABAD structures [99,102], the active site and NAD+-binding site were shown to be highly distorted in the presence of Aβ and no bound NAD+ co-factor was seen. Further studies using SPR confirmed binding of ABAD and Aβ at nanomolar concentrations and showed that a conformational change in ABAD occurs upon binding of Aβ . Saturation transfer difference NMR was used to show that the presence of Aβ inhibited the binding of NAD+ to ABAD in a concentration-dependent manner. Similarly, the ability of Aβ to bind ABAD was reduced in the presence of NAD+, suggesting that the binding of Aβ and NAD+ to ABAD are competing . These observations provide strong evidence that Aβ has an influence on the physical structure of ABAD, disrupting its activity.
Although no electron density for Aβ was observed in the ABAD–Aβ crystal, SDS/PAGE confirmed its presence in the complex, indicating that the Aβ present within the crystal is in a disordered state. The region comprising residues 100–110 was also highly disordered , despite it being well ordered in other known structures [99,102]. These observations led to the hypothesis that this region, referred to as loop D, may be the binding site for Aβ. It is possible that the lack of order in the Aβ component of the complex is either due to high levels of flexibility within the amino acid chain of Aβ or due to a disordered aggregation of the peptide, which would fit with the aggregation hypothesis. Mutagenesis studies showed that replacing residues within the loop D region prevented the binding of Aβ to ABAD. In particular, two groups of residues were highlighted as being significant: the first group consisted of Ser98, Lys99, Thr100 and Tyr101, and the second of Thr108, His109 and Thr110. Point mutations to replace these residues with alanine residues, either individually or in combination, resulted in the loss of ABAD–Aβ binding . The activity of related enzymes, such as the bacterial 3β/17β dehydrogenase and type I HADH, which have a similar mechanism, but do not contain the insertion of loop D seen in ABAD, is unaffected by Aβ , again providing support for this region of the enzyme being the Aβ-binding region. In addition, the energetics of the binding between ABAD and Aβ were broken down into their enthalpic and entropic components, revealing a large increase in entropy upon binding, overcoming an unfavourable enthalpy change. Yan et al.  proposed that this large increase in entropy is probably due to the displacement of highly ordered water molecules from the protein surface upon binding of Aβ to ABAD.
The interaction between ABAD and Aβ has also been demonstrated in vivo using a number of techniques. Analysis of human cerebral cortex samples using immunoprecipitation revealed the enrichment of the ABAD–Aβ complex in AD brains, and similar results were seen in mitochondria isolated from the cerebral cortex of mice APPm-expressing or Tg-APPm/ABAD (transgenic expression of both APPm and ABAD) mice . These results indicate that the interaction of ABAD and Aβ does occur in physiologically relevant environments. Immunocytochemistry was also used to show co-localization of ABAD and Aβ and to verify the localization of ABAD in mitochondria by co-localization with VDAC (voltage-dependent anion channel) .
In addition to the direct interaction between ABAD and Aβ, it has been shown that Aβ can influence the expression of ABAD. Indeed, transgenic mice overexpressing APPm, including the triple transgenic model (expressing AD-linked mutations in APP, presenilin 1 and tau), show increased ABAD expression in the hippocampus when compared with non-transgenic littermates [113,114]. This increased expression has also been observed in human AD brains, where comparison with age-matched controls revealed elevated expression of ABAD in the temporal lobes of AD brains, which was localized to neuronal cells . Significantly, when not associated with elevated Aβ, overexpression of ABAD can have a positive outcome with regard to other stresses (see above) [103,108]. However, under conditions when both ABAD and Aβ are elevated the consequences of their interaction are numerous and occur at the molecular, cellular and in vivo levels.
Consequences of ABAD–Aβ interaction
At the molecular level, as described above, the binding of Aβ causes the inhibition of ABAD activity. However, overexpressing a catalytically inactive form of ABAD in the presence of Aβ does not enhance cytotoxicity in cell cultures compared with Aβ alone . This observation indicates that these cellular effects are not simply based on the inactivation of ABAD, but are due to other downstream effects mediated by the active enzyme once bound with Aβ. This view is supported by the observed discrepancies in the binding constants of Aβ and ABAD (in the nanomolar range) and the Ki values determined for the inhibition of ABAD by Aβ (in the micromolar range) .
Another direct effect on the enzyme was reported when the localization of ABAD was seen to change in the presence of Aβ. Under normal conditions, ABAD was observed both in the ER and in mitochondria. However, in the presence of Aβ (which was either applied externally or produced from a transfected APP plasmid), redistribution from the ER to the inner surface of the plasma membrane was reported [92,101].
At the cellular level, it has been shown that overexpression of ABAD in the presence of elevated Aβ has deleterious effects on cell function and survival. Aβ was found to suppress the reduction of MTT (dimethylthiazolyl diphenyltetrazolium) bromide and increase DNA fragmentation and apoptosis in the neuroblastoma SK-N-SH cell line, and these effects were minimized by the addition of anti-ABAD antibodies . Apoptosis and DNA fragmentation were greatly enhanced in COS cells co-transfected with plasmids expressing Aβ and ABAD , or APPm and ABAD , compared with those transfected with Aβ or ABAD alone. Conversely, this enhancement of toxicity was not seen in cells transfected with mutant (inactive) ABAD in the presence of Aβ, despite Aβ having a similar binding affinity for the mutant and wild-type forms (Kd of 64.5 nM and 38.9 nM respectively) . Again, these results indicate that it is the combination of increased ABAD activity and Aβ expression that is necessary for toxicity.
It has been suggested that the interaction of ABAD with Aβ may induce a toxic effect through the build-up of toxic aldehydes. Increases in both HNE (4-hydroxynonenal) and MDA (malondialdehyde) have been correlated with AD [115,116] and a cytoprotective action of ABAD against these toxic aldehydes has been demonstrated . SH-SY5Y cells transfected with ABAD and then treated with HNE for 24 hours showed an improved survival compared with control cells. Similarly, HeLa cells transfected with ABAD were able to catabolize HNE better than control cells, an effect lost in the presence of Aβ-(1–42) . However, neuroblastoma cell cultures overexpressing both ABAD and APP (either wild-type or APPm forms) were found to exhibit enhanced production of MDA and HNE, compared with those expressing ABAD or APPm alone. Transfection with mutant (inactive) ABAD and APPm together did not produce this response . It has therefore been proposed that a native function of ABAD may be to remove toxic aldehydes such as HNE and MDA and that this function is impaired in the presence of Aβ, resulting in the disruption of normal cellular processes .
The toxic effect of overexpressing ABAD together with Aβ has also been confirmed in an AD mouse model. Compared with neurons from non-transgenic mice and mice overexpressing ABAD or APPm alone, E18 cortical neurons cultured from Tg-APPm/ABAD mice were found to exhibit higher levels of hydrogen peroxide, decreased mitochondrial function and increased cell death . Mitochondrial dysfunction was observed in vivo in Tg-APPm/ABAD mice, which had decreased glucose utilization and ATP production at 9 months of age . These mice were also found to have deficits in spatial and temporal memory compared with non-transgenic mice, with impaired performance in the radial-arm water maze as early as 4–5 months of age . These results again emphasize that it is the combination of ABAD and Aβ that is necessary for effects to be seen and that these effects occur early in the disease process.
Proteomic analysis of brains from Tg-APPm/ABAD mice has identified increased expression of other proteins, hinting at other downstream effects of the interaction of ABAD and Aβ. Expression of Prx-II (peroxiredoxin II), an antioxidant enzyme, was found to be increased in mice overexpressing APPm and in Tg-APPm/ABAD mice. Further analysis of Tg-APPm/ABAD mice brains and human AD brains by Western blotting and immunocytochemistry showed increased expression of Prx-II in the cerebral cortex. Transfection of cortical neurons with Prx-II was found to reduce Aβ toxicity, suggesting that its overexpression in AD is playing a protective role . Interestingly, Prx-II has also been linked with PD, where it was found to be phosphorylated by Cdk5 (cyclin-dependent kinase 5), and hence inactivated in MPTP-induced models of the disease . Similarly, increased phosphorylation of Prx-II was seen in the nigral neurons of human PD brains, whereas the overall expression levels of Prx-II remained unchanged. As in AD models, overexpression of Prx-II in the MPTP-induced in vitro and in vivo models of PD was found to protect against neuronal loss . In light of these findings, the consequences of Prx-II up-regulation in AD, its association with ABAD and potential protective role still need to be investigated in more detail.
A second protein, Ep-I [endophilin I, also referred to as SH3GL2 (SH3-domain GRB-like 2)], was also identified as being up-regulated in Tg-APPm/ABAD mice compared with mice expressing ABAD alone and non-transgenic littermates. Further analysis showed that there was up-regulation of Ep-I in the hippocampus and cortex of Tg-APPm/ABAD mice and in the temporal cortex of human AD brains . Ep-I is a presynaptic protein known for its involvement in synaptic vesicle biogenesis  and it has previously been shown that its expression could also activate JNK . Increases in JNK activation have been observed in AD patients  and AD mouse models , as well as in Aβ-expressing cell culture models [125–127], although this had been thought to be solely due to increases in ROS production. Ren et al.  showed that an increase in Ep-I expression could increase JNK activity with the subsequent death of primary neuronal cell cultures. However, when neuronal cultures were transfected with truncated Ep-I, lacking its SH3 domain, the activation of JNK by Aβ was blocked and cell viability increased . Thus the increase in Ep-I expression shown in the AD brain could be another mechanism for the activation of the JNK signalling pathway. Notably, the reported increases in both Prx-II and Ep-I were shown to be directly due to the binding of ABAD and Aβ, as interfering with this binding in living organisms resulted in the expression of these two proteins returning to normal (see below), therefore emphasizing the importance of this interaction in vivo.
From these studies, it can be seen that the interaction of Aβ with ABAD has multiple effects at the molecular, cellular and whole animal level. Indeed, when Aβ binds to ABAD the net effect is to inhibit the ABAD enzyme activity. However, the precise molecular mechanisms of how this occurs are yet to be deciphered; what is known is that in the living brain this results in the activation of genes which in AD appear to shift the balance of events, with increasing Prx-II expression promoting neuronal survival and increasing Ep-I expression promoting neuronal death.
Recently, a second major Aβ–protein interaction has been found within mitochondria. CypD, a peptidylprolyl isomerase F, is found in the mitochondrial matrix and translocates to the inner mitochondrial membrane during the opening of the mPTP (mitochondrial permeability transition pore) in times of oxidative stress . There are many excellent reviews on the mPTP [129,130], but in brief, the mPTP plays a central role in both necrotic and apoptotic neuronal cell death. Opening of the mPTP collapses the membrane potential and possibly amplifies apoptotic mechanisms by releasing proteins with apoptogenic potential from the inner membrane space . The mPTP is thought to involve, at least, ANT (adenine nucleotide translocase) in the inner membrane, VDAC in the outer membrane (although note that recent studies have suggested the involvement of the mitochondrial phosphate carrier [129,132]) and CypD in the mitochondrial matrix [129,131]. CypD associates with ANT and potentially other targets on the inner mitochondrial membrane, contributing to the opening of the mPTP. This association leads to colloidosmotic swelling of the mitochondrial matrix, dissipation of the inner membrane potential (ΔΨm) and/or generation of ROS. Therefore CypD is considered to be part of the mPTP complex as summarized in Figure 3. Oxidative and other cellular stresses promote CypD translocation to the inner membrane [128,133,134] and other studies provided substantial evidence that a genetic deficiency in CypD protects against Ca2+- and oxidative stress-induced cell death [135,136]. In addition to providing a pivotal regulatory role in the mPTP opening , CypD has also been shown to be involved in protein folding [138,139].
Components of the mPTP
The observations that Aβ progressively accumulates in brain mitochondria of AD patients led Du et al.  to further investigate the mechanism underlying Aβ-mediated mitochondrial dysfunction. In these studies, it was established by SPR that CypD can bind Aβ . Elevated CypD levels were reported in human AD brains, as well as in an APPm-expressing mouse model for AD . The Kd for the interaction of CypD and monomeric Aβ-(1–40) and Aβ-(1–42) was 1.7 μm and 164 nM respectively, whereas interactions with oligomeric Aβ-(1–40) and Aβ-(1–42) had a Kd of 227 nM and 4 nM, thus indicating that the oligomeric forms of Aβ appear to have a greater affinity for CypD and that Aβ-(1–42) has a greater affinity than Aβ-(1–40). Co-localization of CypD and Aβ in the mitochondria was also observed by confocal microscopy in the cerebral cortex of both mice overexpressing APPm and in human AD brains, and immunoprecipitation confirmed the enriched presence of CypD–Aβ complexes in AD brains .
As with the ABAD and Aβ interaction, at present the exact contact sites of the interaction between CypD and Aβ are unknown, though recent molecular-docking experiments have attempted to produce a model . Singh et al.  also predicted an interaction between ANT and Aβ, which together with that of CypD has a possible functional impact on the mPTP . Whether these predictions prove true will only be known when they are experimentally tested. To date, the crystal structure of CypD has been published [142,143], but only in the presence of DMSO  or CsA (cyclosporin A), an inhibitor of CypD . Notably, in both cases a truncated mutant form of CypD was used, starting at Cys29 and containing a single point mutation (with Lys133 replaced by an isoleucine residue). However, at present it is unknown whether these regions are an important region for the interaction.
Consequences of CypD–Aβ interaction
CypD levels of expression are elevated in the aging human brain and in an Aβ-rich environment . The reported consequence of the binding of CypD and Aβ is elevated levels of ROS, which in turn induces mPTP opening and cell death [140,141]. Additionally it is thought that this interaction enhances the translocation of CypD from the matrix to the inner mitochondrial membrane where CypD will interact with the mPTP, resulting in its opening . The interaction would lead to a build-up of Aβ in the inner mitochondrial membrane, which in itself can cause changes in the mitochondrial membrane potential, leading to cell death .
Much of these potential consequences of the interaction of Aβ and CypD have been determined in studies of CypD-deficient animals (Ppif−/−). Notably, the cortical mitochondria isolated from a AD mouse model lacking CypD are resistant to both Aβ and Ca2+-induced mitochondrial swelling and opening of the mPTP, and display an increased calcium buffering capacity and an attenuation of the generation of mitochondrial ROS. Furthermore, CypD-deficient neurons are protected against Aβ- and oxidative stress-induced cell death. Importantly, deficiency of CypD greatly improved the learning and memory of a transgenic APPm-expressing AD mouse model [140,144]. These animals exhibited increased spatial and memory learning and alleviated Aβ-mediated reduction of long-term potentiation at 12  and 24 months, by which age the APPm-expressing mice are known to display AD-like symptoms and synaptic dysfunction . Thus the CypD/Aβ-dependent activation of the mPTP directly links to the cellular and synaptic perturbation relevant to the pathogenesis of AD .
ARE INTRACELLULAR Aβ-BINDING SITES POSSIBLE THERAPEUTIC TARGETS IN ALZHEIMER'S DISEASE?
Having described a number of intracellular binding partners for Aβ, there is the obvious question of whether these sites are relevant for therapeutic intervention in AD. At present there is evidence that both ABAD and CypD are potential drug target sites.
ABAD as a therapeutic target
It has been established that the interaction between ABAD and Aβ can lead to harmful effects on cell viability along with subsequent deleterious effects on the cognitive performance in transgenic AD mouse models, thus reflecting the importance of these cellular effects on disease progression. Therefore these studies indicate that the ability to block this interaction could provide a potential target for the treatment of AD.
Mutagenesis studies have shown that ABAD contains two groups of residues that are particularly important for the binding of Aβ (i.e. the Ser98, Lys99, Thr100 and Tyr101 group and the Thr108, His109 and Thr110 group; both are found in Loop D as described above). It was further reported the development of a ‘decoy peptide’, ABAD-DP, which spanned these important amino acids (residues 92–120), prevented the binding of Aβ to ABAD . The use of this region as a discrete synthetic peptide in competitive SPR-binding studies was able to prevent the binding of Aβ to ABAD at micromolar concentrations, with a Ki of 4.9 and 1.7 μM for Aβ-(1–40) and Aβ-(1–42) respectively, and also attenuated cytotoxicity of Aβ towards primary neurons in a cell-based assay .
Notably, the same region of ABAD had been independently identified previously using an antisense peptide approach . In this approach the antisense DNA strand was translated to give the antisense peptide sequence, which was believed to contain complementary binding surfaces to the peptide produced from the coding DNA strand. This antisense sequence was then compared with the Aβ sequence in order to reveal potential Aβ-binding sites. As a result, a region similar to Aβ residues 16–20 was identified within ABAD residues 99–108, which contained a Leu-Val-Phe-Phe motif. This region of ABAD, again synthesized as a discrete peptide, was found to bind biotinylated Aβ in an ELISA binding-assay with a Kd of 107 nM and to increase the level of neuronal cell survival in the presence of Aβ . This Kd value is in the same range as the Kd value observed by Yan et al.  for the interaction of Aβ with the whole enzyme (88 nM) but is much lower than the Ki observed for the inhibition of the ABAD–Aβ interaction found by Lustbader et al. , supporting the hypothesis mentioned previously that oligomeric forms of Aβ are the species interacting with and inhibiting the enzyme.
An effect of the ABAD-DP was also observed in cell culture models. Addition of the Tat domain from HIV allowed the peptide to cross cell membranes, where it was shown to be effective at protecting cultured primary neurons (from wild-type, ABAD-expressing and Tg-ABAD/APPm animals) from Aβ-mediated toxicity. Mitochondrial stress resulting from Aβ treatment was alleviated, as shown by reductions in cytochrome c release, the production of ROS, DNA fragmentation and LDH (lactate dehydrogenase) release . A major problem with testing the effects of the small ABAD-DP in cellular systems is its rapid degradation by peptidases. In order to stabilize the peptide, in a separate study it was therefore fused with TRX (thioredoxin I), to give the ABAD-(92–120)–TRX peptide, and introduced into PC12 cells using a lentiviral system. There the fusion peptide was still found to protect cells against Aβ-induced toxicity . Moreover, TRX is a scavenger of ROS and known to assist with protein folding and stability [146,147], and so it was also noted to complement the decoy peptide's protective activity, probably by scavenging ROS produced as a result of Aβ toxicity. ABAD-(92–120)-TRX-transfected cells exhibited decreased apoptosis, decreased LDH release and increased cell viability in response to Aβ or hydrogen peroxide treatment compared with untransfected cells or those transfected with TRX alone .
The decoy peptide has also been effective in whole animal studies. Expression levels of protein biomarkers known to be elevated in AD brains, such as Prx-II  and Ep-I , have also shown a response to the disruption of the ABAD–Aβ interaction using the decoy peptide. In these studies, a peptide spanning residues 93–116 of ABAD was again modified, this time to contain a Tat sequence for transport across the cell membrane, along with a mitochondrial-targeting sequence to direct the peptide to the site of ABAD–Aβ complexes in the mitochondria. This Tat-mito-DP-(93–116) peptide was introduced into transgenic APPm-expressing mice by intraperitoneal injection, resulting in systemic application of the peptide. Transgenic APPm mice were shown to have increased Prx-II in the hippocampus, whereas mice treated with Tat-mito-DP showed a significant reduction in Prx-II, comparable with the level seen in non-transgenic mice . Similarly, Tg-APPm/ABAD mice, which exhibit elevated Ep-I expression compared with wild-type littermates, showed a significant reduction in Ep-I, which returned to basal levels . Therefore this indicated that disrupting the ABAD–Aβ interaction can prevent further downstream effects seen in this AD mouse model. It has been shown that by using these additional peptide sequences the decoy peptide can enter the brain and as such it can be used to reverse biochemical symptoms in mouse models of AD. Transgenic mice expressing both APPm and the decoy peptide ABAD-(91–119), as well as transgenic APPm-expressing mice systemically treated with the decoy peptide by intraperitoneal injection, showed improvements in the radial-arm water maze test compared with untreated transgenic APPm-expressing mice .
Taken together, these experimental results consistently underline the therapeutic value of interrupting the ABAD–Aβ interaction. However, whereas small peptides have been shown to inhibit the ABAD–Aβ interaction, the nature of peptides limits their application as drugs due to their low bioavailability and instability. There is therefore a need to develop alternative small molecule inhibitors of the ABAD–Aβ interaction. This has been initiated and screening of a commercially available fragment library consisting of compounds that interact with Aβ or have neuroprotective properties resulted in the development of a series of benzothiole urea compounds, which are capable of inhibiting the interaction at micromolar concentrations, as shown by ELISA . However, further studies into the cellular effects of these compounds and their pharmacological properties will be required for the development of new treatment strategies for AD.
CypD as a therapeutic target
The results showing that CypD deficiency is able to ameliorate Aβ toxicity in transgenic animals means that CypD can also be considered as a potential drug target for AD, as it has for other neurodegenerative disorders [140,149,150]. Indeed, it has been reported that in the presence of CsA, a known immunosuppressant and inhibitor of CypD , there is a decrease in mPTP formation and that a CsA–CypD complex is formed in mitochondria . More directly it was shown that CsA can block some of the Aβ-induced toxicity . Another recent study also indicates that inhibition of CypD is the basis for its neuroprotective properties in, for example, ischaemia/reperfusion injury . However, it has been noted previously that CsA is a large, bulky compound with poor solubility in water and relatively poor bioavailability, especially in the brain , limiting the use of CsA as a drug molecule for neurodegenerative disorders. It also has to be taken into account that CsA exerts its function via several intracellular routes in addition to the inhibition of CypD . Specifically, in the cytoplasm CsA binds to a complex of CypA (cyclophilin A) and calcineurin and thus blocks the peptidyl prolyl isomerase activity of CypA as well as the calcineurin phosphatase activity . It was also noted that CypA can activate peroxiredoxins  and CsA could therefore influence cellular peroxide levels. CsA also blocks the JNK and p38 stress kinase signalling pathways, independently from its inhibition of calcineurin signalling . CsA has also been found to inhibit Ca2+ entry into mitochondria by inhibition of the Ca2+-uniporter in the mitochondrial inner membrane .
Considering these multilayered intracellular effects of CsA and that it is currently used as an immunosuppressant, which in itself would not be beneficial for a disease of the elderly, more research is required to decipher the intracellular effects of CsA. There is also a need for the identification of new modulators of mPTP formation in order to clarify under what circumstances each of them can act neuroprotectively, and at which stage their administration might be useful in neurodegenerative diseases like AD.
In conclusion, many studies have identified numerous receptor proteins that can bind to extracellular Aβ in all its different forms of both size and aggregation state. However, it is becoming increasingly clear that before this build-up of extracellular Aβ there are events that are occurring within cells. Of those described in the present review, we have concentrated on two molecules which have both been proven to bind directly to Aβ in the AD brain, both of which are centred on the action of the mitochondria. These studies have resulted in the identification of new signalling events that are occurring both potentially in the early stages of the disease and all the way through to its final stages (Figure 4). Therefore it could be envisioned that modulators that interfere with the interactions of Aβ with either CypD or ABAD could be potential therapeutic targets of the disease; however this will require imaginative new approaches. Recent possibilities could include the use of new mitochondrial-targeting compounds  or the use of novel compounds that prevent the mitochondrial uptake of Aβ in neuronal cells . What is certain is that there are still many undiscovered events that are occurring due to the rise of intracellular Aβ levels, whether that is the Aβ-(1–40) or Aβ-(1–42) form, and the actual aggregation states at a particular binding site are still largely unknown. As such, new knowledge on these critical topics will result in the identification of potential new drug targets. Although the development of selective drugs for these targets may at first appear to be difficult, it is unlikely that it will prove intractable, and ingenious solutions to these problems will arise.
Summary of the consequences of intracellular Aβ accumulation
amyloid-binding alcohol dehydrogenase
ABAD decoy peptide
adenine nucleotide translocase
amyloid precursor protein
β-site APP-cleaving enzyme 1
Cu2+ chaperone of SOD1
cyclin-dependent kinase 5
green fluorescent protein
- HADH II
human type II hydroxyacyl-CoA dehydrogenase
heat-shock protein 60
HtrA serine peptidase 2
c-jun N-terminal kinase
mitochondrial permeability transition pore
phosophoinositide dependent kinase
reactive oxygen species
single nucleotide polymorphism
surface plasmon resonance
transgenic expression of both APPm and ABAD
translocase of outer mitochondrial membrane
voltage-dependent anion channel
The work of our laboratory is supported by the Alzheimer's Research Trust (a William Lindsay Scholarship to K. E. A. M.); the German Academic Exchange Service (to E. B.); and by the Biotechnology and Biological Sciences Research Council, U.K. S. J. C. thanks St Hugh's College, University of Oxford, Oxford, U.K. for research support.
These authors contributed equally to this work.