Address correspondence and reprint requests to Jean-Noël Octave, Universite catholique de Louvain, Institute of Neuroscience (IoNS), Avenue Hippocrate 53, IONS/5306, B-1200 Brussels, Belgium. E-mail: firstname.lastname@example.org or Anthony J. Turner, School of Molecular and Cellular Biology,Faculty of Biological Sciences,University of Leeds, Leeds LS2 9JT, UK. E-mail: email@example.com
Despite intensive studies of the secretase-mediated processing of the amyloid precursor protein (APP) to form the amyloid β-peptide (Aβ), in relation to Alzheimer's disease (AD), no new therapeutic agents have reached the clinics based on reducing Aβ levels through the use of secretase inhibitors or immunotherapy. Furthermore, the normal neuronal functions of APP and its various metabolites still remain under-investigated and unclear. Here, we highlight emerging areas of APP function that may provide new insights into synaptic development, cognition, and gene regulation. By modulating expression levels of endogenous APP in primary cortical neurons, the frequency and amplitude of calcium oscillations is modified, implying a key role for APP in maintaining neuronal calcium homeostasis essential for synaptic transmission. Disruption of this homeostatic mechanism predisposes to aging and AD. Synaptic spine loss is a feature of neurogeneration resulting in learning and memory deficits, and emerging evidence indicates a role for APP, probably mediated via one or more of its metabolites, in spine structure and functions. The intracellular domain of APP (AICD) has also emerged as a key epigenetic regulator of gene expression controlling a diverse range of genes, including APP itself, the amyloid-degrading enzyme neprilysin, and aquaporin-1. A fuller understanding of the physiological and pathological actions of APP and its metabolic network could provide new opportunities for therapeutic intervention in AD.
Alzheimer disease (AD) is characterized by the presence of two types of lesions in the brain, namely: intra-neuronal neurofibrillary tangles containing a hyper-phosphorylated form of the tau protein (Brion et al. 1985; Grundke-Iqbal et al. 1986) and extracellular senile plaques containing amyloid peptide, Aβ (Glenner and Wong 1984; Masters et al. 1985) derived from the amyloid precursor protein (APP). Since the discovery of APP in 1987 (Kang et al. 1987), processing of the protein has been widely studied in numerous cell lines, both neuronal and non-neuronal in origin (Kienlen-Campard et al. 2000; Petit et al. 2002; Belyaev et al. 2010) as well as in animal models (Morrissette et al. 2009). It is now well documented that APP is processed in two different catabolic pathways (Fig. 1): a minor amyloidogenic pathway in which APP is cleaved by β- and γ-secretases releasing Aβ peptide and a predominant (> 90%) non-amyloidogenic pathway in which the protein is successively cleaved by α- and γ-secretases precluding production of Aβ. Soluble, extracellular domains of APP are also released following the actions of α-secretase (sAPPα) and β-secretase (sAPPβ). The various metabolites of APP may all have distinct physiological roles, including Aβ, which has been suggested to play roles as diverse as ion channel regulation, control of haemostasis and transcriptional activation (Ohyagi et al. 2005; Pearson and Peers 2006; Hardy 2007; Bailey et al. 2011). In both catabolic pathways, the γ-secretase-mediated intramembrane cleavage of APP, after α- or β-secretase action, releases the APP intracellular domain (AICD), which is able to regulate transcription of several genes, including APP itself, the β-secretase BACE-1 and the Aβ-degrading enzyme neprilysin (NEP) (Cao and Sudhof 2001; von Rotz et al. 2004; Pardossi-Piquard et al. 2005; Müller et al. 2007; Belyaev et al. 2009, 2010).
The underlying causes and precise disease mechanisms remain elusive for the majority of AD cases, representing the late-onset or sporadic AD, for which aging is the greatest risk factor and which involves the ‘wild type’ form of APP. In less than 5% of the cases, AD is inherited as an autosomal dominant trait, and results from mutations in the APP gene or in the presenilin (PS) genes (Tanzi et al. 1996; Hardy 1997; Tanzi and Bertram 2005) encoding PS1 or PS2, the catalytic subunits of the γ-secretase multiprotein complex (De Strooper 2003; Edbauer et al. 2003). In these inherited AD cases, Aβ production is substantially enhanced and plays a key role in the etiology of the disease forming the starting-point for the amyloid cascade hypothesis (Hardy and Selkoe 2002; Goate and Hardy 2012) in which accumulations of Aβ, especially an increased ratio of Aβ42 : Aβ40, are the initial triggers for the disease process. Because the neuropathological signatures of inherited and sporadic AD cases are similar, Aβ was hence considered as an important therapeutic target for the treatment of AD, and numerous approaches were developed to inhibit Aβ production or enhance its clearance, using inhibitors of β- or γ-secretase (Wolfe 2012), as well as to remove amyloid deposits from the brain using active or passive immunotherapy (Agadjanyan and Cribbs 2009; Delrieu et al. 2012). When tested in clinical trials, however, inhibitors of γ-secretases induced severe side effects (Golde et al. 2012). Clinical trials using immunotherapy are still proceeding, show efficacy in removal of brain amyloid deposits (Delrieu et al. 2012), but do not improve AD patients very much, if at all (Karran 2012), suggesting that amyloid deposits alone are not intrinsically neurotoxic or that much earlier treatment is required. Given the possible physiological roles of Aβ, however, complete elimination of Aβ is probably not therapeutically desirable (Pearson and Peers 2006). Furthermore, not all of the pathological effects of APP appear to be dependent on γ-secretase action, since the initially formed β-secretase fragment, termed C99, can contribute to neurodegeneration and cognitive deficits in animal models (Lauritzen et al. 2012; Mitani et al. 2012). Also, the patterns of APP processing in autosomal dominant AD may differ from those seen in late-onset disease (Pera et al. 2013). Yet, the widely used animal models of AD incorporate one or more of the familial mutations and therefore do not adequately reflect the situation in late-onset disease involving wild-type APP. While enhanced Aβ production is clearly responsible for early onset disease, a failure of Aβ clearance mechanisms may predominate in sporadic AD (Hama and Saido 2005), including reductions in levels of Aβ-degrading proteases such as NEP (see e.g. Nalivaeva et al. 2012a, b for review).
The failure of the amyloid hypothesis to lead to the delivery of any clinically successful drugs to date has hence put pressure on amyloid-based therapeutic approaches, but the underlying genetics and pathology still strongly place Aβ as a favored target. Indeed, a very recent report identifying the first human mutation in APP (A673T, constituting the second residue of the Aβ peptide) that protects against AD, as well as age-related cognitive decline, provides an independent proof of principle for the amyloid hypothesis (Jonsson et al. 2012). This mutation, adjacent to the β-secretase cleavage site, causes a significant shift in APP processing away from the amyloidogenic pathway and reduced Aβ levels in a cell-based model (Jonsson et al. 2012). Conversely, the pathogenic A673V mutant causes increased amyloidogenic processing of APP (Di Fede et al. 2009). Furthermore, recent reappraisals of the amyloid hypothesis (Hardy 2009; Pimplikar et al. 2010) have highlighted that APP metabolites in addition to Aβ contribute to, or can protect against, AD pathogenesis and may provide new opportunities for therapeutic intervention. The nature of the toxic species has also shifted to a focus on oligomeric forms of Aβ (Larson and Lesné 2012), as well as truncated and modified forms of the peptide, especially the pyroglutamylated Aβ species (Schilling et al. 2008). Indeed, this latter species may provide a missing link between oligomers, tau toxicity, and disease propagation (Nussbaum et al. 2012).
Despite intensive research over two decades, the normal neuronal functions of APP remain unclear. APP knockout mice have a normal phenotype with only subtle defects, as a result of functional compensation by APP-like proteins 1 and 2 (Heber et al. 2000) suggesting that the APP family members serve essential and overlapping functions. Deletion of the APLP2 gene and one other member of the gene family, however, results in perinatal lethality implying a key neuronal role for APLP2 protein (Heber et al. 2000). The soluble, secreted forms of APP have long been thought to play a role in neuritogenesis and to be neuroprotective and accumulating data reveal a distinct role, especially for sAPPβ, in driving the neural differentiation of human embryonic stem cells (Freude et al. 2011; Chasseigneaux and Allinquant 2012). Conversely, further cleavage of sAPPβ, generating a major 35 kDa fragment referred to as N-APP, provides a ligand for the death receptor DR6 triggering axonal pruning and neuronal death via the action of caspase-6 (Nikolaev et al. 2009). The metalloprotease meprin-β can also generate a number of N-terminal APP fragments although they appear not to be neurotoxic (Jefferson et al. 2011).
APP function, isoforms, and epigenetic regulation
There are three major isoforms of APP (APP695, APP751, APP770) generated as a result of alternative splicing of exons 7 and 8. Compared with APP695, the 751 isoform contains an additional Kunitz-type protease inhibitor (KPI) domain and the 770 isoform also contains the 19-amino acid, OX-2 domain, which shares identity with the OX-2 antigen of thymus-derived lymphoid cells. In brain, APP695 is principally neuronal and is expressed at relatively high levels compared with the other two isoforms, although there are regional differences, and the balance between the KPI- and non-KPI isoforms may influence Aβ deposition. In AD brain, the KPI isoforms of APP are significantly raised (Moir et al. 1998) and the various isoforms show different temporal- and disease-specific expression implying they exert distinct functional and metabolic roles. Until recently, no clear-cut functional differences have been ascribed to the different APP isoforms apart from the protease-inhibitory role of the KPI domain. Although all three APP isoforms are potentially amyloidogenic, it has now been shown that, in neuronal cell lines, sAPPβ, Aβ and AICD are preferentially formed from the neuronal APP695 isoform (Belyaev et al. 2010). Consistent with this observation is the much earlier report that, in both human brain and cerebrospinal fluid, Aβ may be generated in vivo in humans specifically from APP695 because the KPI-containing isoforms were preferentially subjected to α-secretase cleavage (Kametani et al. 1993). In addition only APP695, when expressed in neuronal cell lines, increased nuclear AICD levels and NEP expression (Belyaev et al. 2010). We have hence emphasized that it is nuclear, rather than total cellular, AICD levels that reflect AICD functional in gene regulation (Belyaev et al. 2010). The APP isoforms themselves can exist in homodimeric forms with the KPI and transmembrane regions being involved in dimerization, which causes APP751 to be more efficiently processed through the non-amyloidogenic pathway than APP695, probably by selective regulation of APP trafficking (Ben Khalifa et al. 2012a, b).
Although AICD can potentially be formed via both the amyloidogenic and non-amyloidogenic pathways (see Fig. 1), several groups have now unequivocally established that nuclear location of AICD and consequent transcriptional regulation is dependent on APP processing through the amyloidogenic (β-secretase) pathway (Belyaev et al. 2010; Goodger et al. 2009; Flammang et al. 2012). The cohort of genes regulated by AICD has been a controversial issue (discussed critically in Beckett et al. 2012; Pardossi-Piquard and Checler 2012; Bórquez and González-Billault 2012), which likely reflects the cell specificity of AICD regulation and its dependence also on cell density, aging, and other factors (Belyaev et al. 2010; Bauer et al. 2011; Xu et al. 2011; Hong et al. 2012). However, it has been unequivocally demonstrated by chromatin immunoprecipitation studies that AICD does indeed occupy the promoter region of putatively regulated genes (e.g. NEP) displacing HDAC1 (Belyaev et al. 2009, 2010), that it is localized in specific nuclear transcription factories (Konietzko et al. 2010) and that it forms part of a transcriptional complex involving the nuclear mediator component MED12 along with the histone acetyltransferase Tip60 and Fe65 protein (Turner et al. 2011; Xu et al. 2011; Müller et al. 2013). Hence, APP mediates epigenetic control of a number of genes at least in part through displacement of histone deacetylases (HDAC) (Belyaev et al. 2009, 2010; Huysseune et al. 2009). It is not only the AICD C-terminal fragment of APP that controls gene expression and other APP metabolites (e.g. sAPPβ) may exert these effects in some cases (Li et al. 2010). Thus, gene regulation by APP and its metabolome is a complex and inadequately understood process and such epigenetic mechanisms may underlie a number of the changes seen in AD (Balazs et al. 2011). Supporting this concept are the observations that increased histone acetylation and hence transcriptional activity are linked with recovery of learning and memory (Fischer et al. 2007; Nalivaeva et al. 2012c) whereas deacetylation is associated with memory impairment in aged mice (Peleg et al. 2010). The HDAC inhibitor sodium valproate has been shown to increase the number of dendritic spines and improve memory in rats with cognitive impairments caused by prenatal hypoxia (Zhuravin et al. 2011). However, valproate inhibits various classes of HDACs and affects expression of up to 4% of genes. Hence, highly selective HDAC inhibitors and other chromatin modifying agents, which are currently under development, may provide novel therapeutic options in neurodegenerative disease, including AD (Karagiannis and Ververis 2012).
Regulation of synaptic activity by APP
Contrary to most of the animal models of AD, which do not accurately recapitulate the human disease (Hardy 2009; Pimplikar et al. 2010), AD is characterized by an important neuronal loss to which abnormal processing of APP could contribute. Therefore, it could well be critical to remove excess brain amyloid deposits before appearance of massive neurodegeneration. Because current animal models are inappropriate, overexpression or downregulation of endogenous APP expression in primary cultures of cortical neurons has recently allowed investigation of some of the neuronal functions of APP (Santos et al. 2009). In vivo, cortical neurons form oscillating networks of various sizes involved in temporal representation and long-term consolidation of information (Buzsaki and Draguhn 2004), which are profoundly affected in AD. Primary cultures of cortical neurons following a few days of differentiation develop mature networks in which glutamate-mediated excitatory synaptic inputs generate spontaneous synchronous calcium oscillations detectable by single-cell calcium imaging (Fig. 2a). However, adenoviral expression of APP even in about 20% of these neurons is sufficient to completely abolish spontaneous synchronous calcium oscillations in the entire network (Fig. 2a). Although whole-cell calcium current analysis in patch-clamp experiments demonstrated that the maximal amplitude of total calcium currents is similar in APP- and GFP-expressing neurons (Fig. 2b), treatment of the cells with nimodipine, a selective antagonist of L-type calcium channels, leads to a significant increase in L-type calcium currents in response to maximal stimulation only in APP-overexpressing cells (Fig. 2b). The resulting calcium influx stimulates calcium-activated K+ channels (SK channels), leading to an increase in medium afterhyperpolarization (mAHP) peak amplitude in APP-expressing neurons (Fig. 2b). Treatment of the cells with the bee venom neurotoxin, apamin, which inhibits SK channels mediating mAHP (Blatz and Magleby 1986) results in restoration of spontaneous calcium oscillations in these cells (Fig. 2a).
The proposed molecular mechanisms for the above are that expression of APP increases the activity of L-type calcium channels, thereby stimulating apamin-sensitive SK channels responsible for mAHP. Increased mAHP lowers excitability of APP-expressing neurons (Santos et al. 2009) which results in inhibition of synaptically propagated synchronous calcium oscillations in the entire network. On the contrary, silencing endogenous APP expression, using shRNA, increases the frequency and decreases the amplitude of calcium oscillations (Fig. 2a).
We can therefore conclude that an important function of APP is to maintain neuronal calcium homeostasis, which is essential for synaptic transmission (Berridge 1998). Disruption of calcium homeostasis is considered as the common pathway for aging and AD (Khachaturian 1989; Toescu and Verkhratsky 2003) as it results in neurodegeneration and disturbed neuronal communication observed both in animal models and AD patients (Emilsson et al. 2006; Stutzmann 2007; Bezprozvanny and Mattson 2008). AD is a complex neurodegenerative disorder in which hippocampal synaptic dysfunction is an early feature of the degenerative process that is clearly linked to memory impairment. Although the relationship between spine density and memory formation is not fully elucidated, there is evidence that spine loss often contributes to deficits in learning and memory (Segal and Andersen 2000; Fiala et al. 2002; Alvarez and Sabatini 2007). APP can differently affect formation of spines at the early stages of the disease depending on the model used and the region of the brain studied (Mucke et al. 2000; Priller et al. 2006; Perez-Cruz et al. 2011). Although controversial, the data on the effects of APP overexpression or downregulation on spine density (Jung and Herms 2012) support the role of APP in regulation of spine structure and functions. On the other hand, the lack of sAPPα is considered as a major contributor to spine loss since conditioned medium from APP+/+ cultured hippocampal neurons restores spine number and integrity in neurons from APP−/− mice (Tyan et al. 2012). In contrast, APLP2 deletion does not affect hippocampal dendritic structure, spine density, or synaptic function (Midthune et al. 2012). The accepted neurophysiological events underlying learning and memory [long-term potentiation (LTP) and long-term depression (LTD)] were shown to be profoundly modified under APP excess or deficiency in aged animals, and play a key role in AD pathogenesis (reviewed in Palop and Mucke 2010).
Although processing of APP has been widely studied, the normal functions of the protein and its various metabolites remain less clear, but a growing body of evidence points toward a key role in the control of gene expression and of synaptic activity. Most past and current efforts to develop AD therapies are based on the amyloid cascade hypothesis and agents have been developed to reduce Aβ production or eliminate Aβ from the brain. However, since all amyloid-based approaches have failed to provide benefit to patients in clinical trials there is clearly a need to look beyond Aβ to develop effective AD therapies based not only on APP processing but also on APP function. The identification of functional roles for APP metabolites (the APP metabolome), of novel APP-interacting partners [the APP interactome (Perreau et al. 2010)], of epigenetically regulated genes, and of specific roles in synaptic spine activity provides new directions for therapeutic approaches.
We thank the UK Medical Research Council, Alzheimer's Research UK, Programme RAS “Fundamental Sciences for Medicine” (AJT and NNN) and the Belgian Fonds pour la Recherche Scientifique (FNRS-FRS), the Interuniversity Attraction Poles Programme-Belgian State-Belgian Science Policy, and the Programme d'excellence « Marshall » Diane convention (JNO, SFS and NP).