γ-Secretase-Regulated Mechanisms Similar to Notch Signaling May Play a Role in Signaling Events, Including APP Signaling, Which Leads to Alzheimer’s Disease

Abstract

Although γ-secretase was first identified as a protease that cleaves amyloid precursor protein (APP) within the transmembrane domain, thus producing Aβ peptides that are thought to be pathogenic in Alzheimer’s disease (AD), its physiological functions have not been fully elucidated. In the canonical Notch signaling pathway, intramembrane cleavage by γ-secretase serves to release an intracellular domain of Notch that shows activity in the nucleus through binding to transcription factors. Many type 1 transmembrane proteins, including Notch, Delta, and APP, have recently been shown to be substrates for γ-secretase, and their intracellular domains are released from the cell membrane following cleavage by γ-secretase. The common enzyme γ-secretase modulates proteolysis and the turnover of possible signaling molecules, which has led to the attractive hypothesis that mechanisms similar to Notch signaling contribute widely to proteolysis-regulated signaling pathways. APP is also likely to have a signaling mechanism, although the physiological functions of APP have not been elucidated. Indeed, we have shown that the intracellular domain of APP alters gene expression and induces neuron-specific apoptosis. These results suggest that APP signaling responds to the onset of AD. Here, we review the possibility of γ-secretase-regulated signaling, including APP signaling, which leads to AD.

Introduction

Notch receptors and their ligands are evolutionarily conserved single transmembrane spanning proteins (type 1 transmembrane proteins; the amino terminus is extracellular and the carboxy terminus is cytoplasmic) that mediate the fates of numerous cells in both invertebrates and vertebrates (Artavanis-Tsakonas et al. 1995; Lewis 1998). Notch signaling mediates somitogenesis, lymphoid cell differentiation, and neural development, and dysregulation of Notch signaling leads to developmental defects or cancer in mammals (Bolos et al. 2007). For example, cells expressing the major ligand Delta inhibit the neural determination of neighboring Notch-expressing cells during neurogenesis in Drosophila, and similar inhibitory effects on neurogenesis has also been reported in vertebrates (Artavanis-Tsakonas et al. 1995; Lewis 1998). Notch signaling has attracted interest because of its relation to Alzheimer’s disease (AD). The Caenorhabditis elegans (C. elegans) gene sel-12 which appears to facilitate the reception of signals mediated by lin-12 (C. elegans Notch), was identified by screening for a suppressor of lin-12 gain-of-function mutation (Levitan and Greenwald 1995). As sel-12 is thought to be a counterpart of human presenilin (PS), which is a catalytic component of γ-secretase and has been implicated in AD, the Notch signaling pathway may be related to AD. Indeed, as described below, the Notch signaling pathway has been shown to be regulated by γ-secretase-mediated proteolysis.

The Notch signaling pathway is unique in that it is controlled by proteolytic cleavage reactions (Artavanis-Tsakonas et al. 1999; Justice and Jan, 2002). In the canonical Notch signaling pathway, ligands bind to the extracellular domain of Notch expressed on neighboring cells, and trigger sequential proteolytic cleavage. Finally, the intracellular domain (ICD) of Notch (NICD) is released from the cell membrane by γ-secretase and translocates to the nucleus to modulate gene expression through binding to transcription factors. Therefore, γ-secretase plays a central regulatory role in Notch signaling.

γ-Secretase was first identified as a protease that cleaves amyloid precursor protein (APP) within the transmembrane domain and produces Aβ peptides (Haass and Selkoe 1993), which are thought to play pathogenic roles in AD (Hardy 1997; Selkoe 2001). However, the physiological functions of γ-secretase have not been determined (Kopan and Ilagan 2004; Selkoe and Wolfe 2007). Recently, it was demonstrated that more than five dozen type 1 transmembrane proteins are substrates for γ-secretase (McCarthy et al. 2009) and their ICDs are also released from the cell membrane. These observations indicating that the common enzyme, γ-secretase, modulates proteolysis and the turnover of putative signaling molecules have led to the attractive hypothesis that mechanisms similar to the Notch signaling pathway may contribute widely to γ-secretase-regulated signaling pathways (Koo and Kopan 2004; Nakayama et al. 2008a). Indeed, it has been reported that the intracellular domain of Dll1 (Dll1IC) is also released from the cell membrane by γ-secretase and translocates to the nucleus (Ikeuchi and Sisodia 2003; LaVoie and Selkoe 2003; Six et al. 2003). Moreover, we have shown that Dll1IC binds to Smads, which are transcription factors involved in the TGF-β/Activin signaling pathway, and enhances transcription of specific genes leading to neuronal differentiation (Hiratochi et al. 2007).

Interestingly, ICD of APP (AICD), which is released from the cell membrane by γ-secretase, has also been reported to undergo translocation to the nucleus (Cupers et al. 2001; Gao and Pimplikar 2001; Kimberly et al. 2001) and may function as a transcriptional regulator (Cao and Sudhof 2001; Guenette 2002). Several AICD-interacting proteins, which are thought to regulate AICD stability and cellular localization, have been identified (Zheng and Koo 2006; Muller et al. 2008) and some models of APP signaling have also been proposed. For example, AICD recruits Fe65 protein and translocates into the nucleus where the AICD-Fe65-Tip60 ternary complex may control transcription of target genes (Slomnicki and Lesniak 2008). As AICD has been shown to have apoptotic potential (Zheng and Koo 2006; Muller et al. 2008), it is likely that APP signaling induces cell death, which leads to AD.

To explore APP signaling, we established embryonic carcinoma P19 cell lines overexpressing AICD (Nakayama et al. 2008b). Although neurons were differentiated from these cell lines by treatment with all-trans-retinoic acid (RA), AICD expression induced neuron-specific apoptosis. The effects of AICD were restricted to neurons, and no effects were observed on non-neural cells. Furthermore, we evaluated changes in gene expression induced by AICD through this process of neuron-specific cell death by DNA microarray analysis (Ohkawara et al. 2011). The results indicated that AICD induces dynamic changes in the gene expression profile. Therefore, it is likely that APP also has a signaling mechanism and that AICD may play a role in APP signaling, which leads to AD.

In this review, we focus on the possibility that γ-secretase-regulated mechanisms similar to the Notch signaling pathway may play roles in signaling events involving type 1 transmembrane proteins. We also discuss the possibility that APP signaling induces dynamic changes in gene expression, which may be closely correlated with AICD-induced neuron-specific apoptosis, leading to AD.

Notch Signaling Pathway

The typical Notch gene encodes a type 1 transmembrane protein with a mass of about 300 kDa (Wharton et al. 1985). The extracellular domains of Drosophila Notch and some vertebrate homologs contain 36 tandem epidermal growth factor (EGF)-like repeats. The 11th and 12th EGF-like repeats of the Notch protein extracellular domain are necessary and sufficient for binding to its ligands (Rebay et al. 1991). NICD has six tandem ankyrin-like (CDC10) repeats, a nuclear localization signal, and a PEST sequence (Wharton et al. 1985). While Drosophila has only one Notch isoform, four Notch isoforms (Notch 1 to 4) have been identified in mammals. TAN1 (Notch1), which was the first mammalian homolog of Notch to be identified, was cloned as the gene responsible for human T cell acute lymphoblastic leukemia (T-ALL) (Ellisen et al. 1991). Notch2 was also identified as an oncogene responsible for feline thymic lymphoma (Rohn et al. 1996). CADASIL (cerebral autosomal dominant arteriopathy with subcortical infarcts and leukoencephalopathy), the main symptom of which is cerebral vascular disorder, is caused by a mutation in the Notch3 gene (Joutel et al. 1996). In addition, Notch4 is the cellular counterpart of the oncogene of mouse mammary tumor virus (int3) (Sarkar et al. 1994).

Drosophila has two different Notch ligands, i.e., Delta (Kopczynski et al. 1988) and Serrate (Fleming et al. 1990). However, two families of ligand, Delta-like protein (Dll 1, 3, and 4) (Bettenhausen et al. 1995; Dunwoodie et al. 1997; Shutter et al. 2000) and Jagged (Jagged 1 and 2) (Lindsell et al. 1995; Shawber et al. 1996), have been identified in mammals. The extracellular domains of all of these ligands also contain EGF-like repeats. All of these ligands also have a cysteine-rich conservative motif called the DSL (Delta: Serrate: Lag-2) domain (Tax et al. 1994), which plays an essential role in binding to Notch (Henderson et al. 1997).

In the canonical Notch signaling pathway, ligands bind to the extracellular domain of Notch on neighboring cells and trigger sequential proteolytic cleavage reactions (Fig. 1). This process is called the regulated intramembrane proteolysis (RIP) mechanism (Brown et al. 2000), and requires sequential cleavage steps within the juxtamembrane (JM) and transmembrane (TM) domains; these cleavage steps are carried out by the metalloproteases and γ-secretase, respectively (Selkoe and Kopan 2003). RIP serves to release NICD from the cell membrane and NICD is then translocated to the nucleus where it expresses its activity through binding to transcription factors. Members of the CSL family (CBF1/RBP-jκ in mammals, Su(H) in Drosophila, and Lag-1 in C. elegans) are major downstream transcription factors of Notch signaling (Artavanis-Tsakonas et al. 1995, 1999). NICD binds to CSL transcription factors, and six tandem ankyrin-like repeats in NICD have been shown to be essential but not sufficient for binding to CSL transcription factors (Roehl et al. 1996). As NICD also binds to Mastermind-like proteins (MAML family in mammals) (Wu et al. 2000), the CSL–NICD–MAML complex is formed. As a result of forming these complexes, co-repressors are dispersed from CSL and co-activators such as P/CAF and P300 are recruited by these complexes (Wu et al. 2000; Wallberg et al. 2002). Through this process, the CSL complexes are converted from transcriptional repressors to activators. Finally, CSL complexes bind to the cis-acting DNA sequences of target genes and promote their transcriptional activity.

Fig. 1.

Notch signaling pathway. Notch proteins are expressed on the cell surface as heterodimers after cleavage at the S1 site by furin. The binding of Notch to the ligand triggers sequential proteolytic cleavage of RIP. When Notch binds to the ligand, Notch is cleaved at the S2 site in the juxtamembrane region by TACE or ADAM protease. Next, the remaining protein stub is further cleaved by γ-secretase at the S3 and S4 sites within the transmembrane domain and NICD is released from the membrane. Then, NICD translocates into the nucleus and binds to the CSL together with MAML. The resultant CSL–NICD–MAML complex removes co-repressors from CSL transcription factor and recruits a co-activator, resulting in conversion from repressor to activator. Finally, the complexes of CSL–NICD–MAML-Co-activators promote transcription of the target genes

The best-established target genes for Notch signaling are Hes (Hairy/Enhancer of split in Drosophila) genes, which encode basic helix-loop-helix (bHLH) transcription factors (Kageyama et al. 2007). Seven mammalian Hes, designated Hes1 to Hes7, have been identified to date, although the mouse does not have Hes4. Hes is known to act as a repressor for tissue-specific gene transcription (Kageyama et al. 2007). Hes1 and Hes5 have been shown to bind to their target DNA sequence, called the N box (CACNAG), by forming homodimers or heterodimers with Hey (Hes-related with YRPW motif) 1 or Hey2, and to recruit histone deacetylase (HDAC) activity by association with Groucho, resulting in transcriptional repression (Akazawa et al. 1992; Leimeister et al. 1999; Iso et al. 2001). Furthermore, Hes proteins associate with ubiquitously expressed bHLH factors (E proteins) and prevent the formation of complexes between tissue-specific bHLH factors, such as Mash1, and E protein (Kageyama et al. 2007). In this manner, Notch represses the differentiation of cells to specific lineages. In addition, the expression of Delta is induced by proneural genes that encode bHLH transcriptional factors, while multiple POU-binding factors are also important for Delta expression (Nakayama et al. 2004). Thus, Notch signaling strongly inhibits Delta expression.

APP

APP was first identified as a cDNA cloned using a partial amino acid sequence of the Aβ fragment from the amyloid plaque of AD brains (Kang et al. 1987). This cDNA encodes a type 1 membrane protein expressed in many tissues, which is concentrated in the synapses of neurons. In humans, the APP gene has a total length of 240 kb containing at least 18 exons (Yoshikai et al. 1990), and several alternative splicing isoforms of APP have been observed ranging in length from 365 to 770 amino acids, differing mainly in the absence (APP-695, predominantly expressed in neurons) or presence (APP-751 and APP-770) of the Kunitz protease inhibitor (KPI) domain located toward the N-terminus of the protein (Sisodia et al. 1993). Counterparts of this protein have been identified in other organisms, including Drosophila (Rosen et al. 1989; Luo et al. 1990), C. elegans (Daigle and Li 1993), and mammals (Coulson et al. 2000). As mentioned below, APP undergoes sequential proteolytic cleavage reactions to yield the extracellular fragment, intracellular fragment (AICD), and Aβ fragment located in the membrane-spanning domain, which is thought to be the component responsible for the onset of AD.

Although APP plays central roles in AD (Hardy 1997; Selkoe 2001), the physiological functions of this protein have yet to be fully elucidated (Zheng and Koo 2006). While APP has been reported to act as a cell adhesion molecule for cell–cell interaction (Soba et al. 2005), and as a neurotrophic and/or synaptogenic factor (Hung et al. 1992; Bibel et al. 2004; Leyssen et al. 2005), several lines of evidence support the suggestion that APP may act as a cell-surface receptor; e.g., Aβ can bind to APP and may, therefore, be a candidate ligand for APP (Lorenzo et al. 2000). It has also been reported that F-spondin (Ho and Sudhof 2004) and Nogo-66 receptor (Park et al. 2006) can bind to the extracellular domain of APP and regulate the production of Aβ. In addition, the extracellular domain of APP can potentially interact in trans suggesting that APP molecules can bind to each other in a homophilic manner (Wang and Ha 2004).

AICD has been reported to undergo translocation to the nucleus and may function as a transcriptional regulator (Muller et al. 2008). Indeed, APP homologs show significant sequence conservation in ICD (Nakayama et al. 2008b), which may reflect the functional importance of AICD. Moreover, AICD is thought to form complexes with the nuclear adaptor protein Fe65 and the histone acetyltransferase Tip60 (Cao and Sudhof 2001). These complexes can bind to the cis-acting DNA sequence of the tetraspanin protein KAI1 gene and regulate transcriptional activity (Baek et al. 2002). In addition, the Aβ region of APP is not well conserved across species.

γ-Secretase

Although γ-secretase was first identified as a protease that cleaves APP within the TM domain and produces Aβ peptides (Haass and Selkoe 1993), which are thought to be pathogenic in AD, the physiological functions of γ-secretase have also yet to be fully elucidated (Kopan and Ilagan 2004; Selkoe and Wolfe 2007). The γ-secretase complex is composed of PS, Nicastrin, anterior pharynx defective-1 (Aph-1), and PS enhancer-2 protein (Pen-2) (Iwatsubo 2004; Kopan and Ilagan 2004; Selkoe and Wolfe 2007). PS is recognized as a catalytic component of the γ-secretase complex, and the two PSs, PS1 (Sherrington et al. 1995) and PS2 (Rogaev et al. 1995; Levy-Lahad et al. 1995), were identified by genetic linkage analyses as the genes responsible for several forms of early-onset familial AD. Nicastrin, a single-pass membrane protein, is thought to be important for recognizing the substrate proteins of γ-secretase (Yu et al. 2000). The extracellular domain of Nicastrin resembles an aminopeptidase, but lacks catalytic residues, and can interact with the N-terminal stubs of γ-secretase substrates after ectodomain shedding (Shah et al. 2005). Thus, shedding of type 1 transmembrane proteins may be essential to produce free N-termini of these proteins retained in the membrane, which can then be recognized by Nicastrin. Aph-1 is thought to act as a scaffold during the γ-secretase complex assembly process and Pen-2 is thought to act as a trigger for the proteolytic cleavage of PS to regulate PS activity (Kopan and Ilagan 2004; Selkoe and Wolfe 2007).

γ-Secretase shows sequence-independent cleavage of a diverse set of type 1 transmembrane proteins that have shed their extracellular domains (Struhl and Adachi 2000). As reflected by the flexible sequence specificity of γ-secretase activity, more than five dozen type 1 transmembrane proteins have been reported to act as substrates of γ-secretase (McCarthy et al. 2009). As shown in Table 1, these substrates also have a wide range of functions, including cell fate determination (Notch, Delta, and Jagged), cell–cell adhesion (N-cadherin, E-cadherin, and CD44), synaptic adhesion (Nectin-1), ion conductance regulation (β2 subunit of the voltage-gated sodium channel), axon guidance and tumor suppression (DCC), neurotrophin receptor (P75NTR), and its homolog (NRADD), lipoprotein receptor (ApoER2), and growth factor-dependent receptor tyrosine kinase (ERBB4).

Table 1.

Substrates for γ-secretase

Substrate	Function	PS or ICD function
ApoER2	Lipoprotein receptor, neuronal migration	Activate nuclear reporter
APP	Precursor to Aβ, adhesion, trophic properties, axonal transport?	Aβ generation, release of ICD, complex with Fe65/TIP60, Cell death?
APLP1/2	Cell adhesion?	Form complex with Fe65 and TIP60
E-cadherin	Cell adhesion	Promote disassembly of adhesion complex
N-cadherin	Cell adhesion	Promote CBP degradation
β-catenin	Transduce Wnt signals, stabilize adherens junctions	Facilitate phosphorylation
CD43	Signal transduction	Signaling molecule?
CD44	Cell adhesion	Activate TRE-mediated nuclear transcription
CSF1 receptor	Protein tyrosine kinase	Unknown
CXCL16 & CX3CL1	Membrane chemokine ligands	Unknown
DCC	Axon guidance, tumor suppressor	Activate nuclear reporter
Delta	Notch ligand	Transcriptional regulation
ERBB4	Receptor tyrosine kinase	Regulate heregulin-induced growth inhibition
HLA-A2	MHC class I molecule	Unknown
IFN-αR2	Subunit of type I IFN-α receptor	Transcriptional regulation
IL-1RI	Cytokine receptor	Unknown
IL-1RII	Cytokine receptor	Unknown
Jagged	Notch ligand	Modulate AP-1 mediated transcription
LRP	Scavenger and signaling receptor	Activate nuclear reporter
Na channel β-subunit	Cell adhesion, an auxiliary subunit of voltage-gated Na channel	Alter cell adhesion and migration
Nectin-1α	Adherens junction, synapse receptor	Remodeling of cell junctions?
Notch1-4	Signaling receptor	Transcriptional regulation
NRADD	Apoptosis in neuronal cells	Modulate glycosylation/maturation of NRADD
P75NTR	Neurotrophin co-receptor, dependence receptor	Modulate p75-TrkA complex?
		Nuclear signaling?
γ-protocadherin	Cell adhesion, neuronal differentiation	Regulation of gene transcription?
Syndecan-3	Cell-surface proteoglycan co-receptor	Regulation of membrane-targeting of CASK
Telencephalin	Cell adhesion	Turnover of telencephalin
Tyrosinase, Tyrosinase-related protein 1/2	Pigment synthesis	Intracellular transport of Post-Golgi Tyr-containing vesicles

PS presenilin, ICD intracellular domain, APLP amyloid precursor-like protein, CBP CREB (cAMP-responsive element binding protein)-binding protein, TRE TPA (12-o-tetradecanoylphorbol 13-acetate)-responsive element, AP-1 activator protein-1, CASK calmodulin-dependent serine kinase, Tyr tyrosinase

It has been reported that several γ-secretase substrates follow the RIP mechanism and their ICDs are released from the cell membrane (Koo and Kopan 2004; Selkoe and Wolfe 2007). As shown in Fig. 2, the process of sequential proteolytic cleavage of APP is similar to that of Notch and follows the RIP mechanism. Cleavage of APP by α-secretase (Esch et al. 1990) or β-secretase (Vassar et al. 1999) at the α-site or β-site, respectively, within the JM region results in the shedding of almost the entire extracellular domain and generates membrane-tethered α- or β-carboxy terminal fragments (CTFs). Several zinc metalloproteinases, such as TACE/ADAM protease (Buxbaum et al. 1998; Lammich et al. 1999), and the aspartyl protease BACE2 can cleave APP at the α-site (Farzan et al. 2000), while BACE1 (β-site APP cleaving enzyme) cleaves APP at the β-site (Vassar et al. 1999). After shedding the extracellular domain, the remaining stub is further cleaved at least twice by γ-secretase within the TM domain at γ- and ε-sites, producing either p3 peptide (in combination with α-secretase) or Aβ (in combination with BACE1), respectively, and AICD (Kopan and Ilagan 2004; Selkoe and Wolfe 2007). In addition, AICD was shown to be a substrate of caspase and is cleaved at the group III caspase consensus sequence (16 aa from the membrane border within AICD).

Fig. 2.

Similarities in the proteolytic processes among Notch, APP, and CD44. a In response to ligand binding, Notch undergoes shedding due to metalloprotease cleavage at the S2 site within the JM domain. After shedding the extracellular domain, the remaining Notch stub is further cleaved by γ-secretase at S3 and S4 sites within the TM domain. This sequential proteolysis produces NICD and Nβ fragment. b Cleavage of APP by α-secretase or β-secretase at the α-site or β-site, respectively, within the JM domain results in shedding of almost the entire extracellular domain and generates membrane-tethered α- or β-carboxy terminal fragments (CTFs). Several zinc metalloproteinases and BACE2 can cleave APP at the α-site, while BACE1 cleaves APP at the β-site. After shedding the extracellular domain, the remaining stub is further cleaved at least twice within the TM domain at γ- and ε-sites by γ-secretase, producing either p3 peptide (in combination with α-secretase) or Aβ (in combination with BACE1), respectively, and AICD. c Several stimuli, such as PKC activation and Ca²⁺ influx, trigger ectodomain cleavage of CD44 by a metalloprotease at the site within the JM domain, resulting in the secretion of soluble CD44 (sCD44). After shedding the extracellular domain, the remaining CD44 stub is further cleaved by γ-secretase at two sites within the TM domain. This sequential proteolysis produces the CD44 ICD and CD44β, an Aβ-like peptide

As shown in Fig. 2, the process of sequential proteolytic cleavage of CD44, which is important for immune system function, is very similar to those of Notch and APP (Nagase et al. 2010). In addition, ICD of this protein (CD44ICD) also translocates to the nucleus.

As mentioned above, similar to Notch, several γ-secretase substrates follow the RIP mechanism and their ICDs are released from the cell membrane, and these ICDs translocate to the nucleus. Therefore, the observations that the common enzyme, γ-secretase, modulates proteolysis and the turnover of possible signaling molecules led to the attractive hypothesis that mechanisms similar to the Notch signaling pathway may contribute widely to γ-secretase-regulated signaling pathways (Koo and Kopan 2004; Nakayama et al. 2008a).

Indeed, several groups have shown that Dll1 is cleaved sequentially by proteases, probably including ADAM and γ-secretase, and the intracellular domain of Dll1 (Dll1IC) is released from the cell membrane and undergoes translocation to the nucleus (Ikeuchi and Sisodia 2003; LaVoie and Selkoe 2003). We have also shown that Dll1IC binds to Smads, which are transcription factors for the TGF-β/Activin signaling pathway, and enhances transcription of specific genes leading to neuronal differentiation (Hiratochi et al. 2007). Therefore, these results suggest that Dll1 also has a signaling mechanism similar to Notch signaling.

What is the function of γ-secretase?

Contrary to the hypothesis that γ-secretase regulates the signaling pathways of certain membrane proteins, Kopan and Ilagan (2004) proposed that γ-secretase is a proteasome of the membrane. They presented clear evidence suggesting that ICDs of Syndecan-3, Nectin-1α, p75, and DCC, all of which are released by γ-secretase, are rapidly degraded. Furthermore, ectodomain shedding seems to be constitutive for some substrates; only intramembrane cleavages of Notch (Schroeter et al. 1998), Delta (Hiratochi et al. 2007), Syndecan-3 (Schulz et al. 2003), and ERBB4 (Ni et al. 2001) are enhanced by ligand binding. As AICD is also degraded rapidly (Cupers et al. 2001; Edbauer et al. 2002), it is usually difficult to detect ICDs of these substrates by western blotting in vivo. In addition, Kopan and Ilagan (2004) noted that considerable evidence supporting the signaling hypothesis was obtained in overexpression assays, which do not quite correspond to physiological conditions. Based on these observations, they proposed the proteasome hypothesis that the primary function of γ-secretase is to facilitate the selective disposal of type 1 membrane proteins (Kopan and Ilagan 2004).

While the proteasome hypothesis of γ-secretase is persuasive, the signaling pathways of some substrates, such as Notch (Artavanis-Tsakonas et al. 1999; Selkoe and Kopan 2003; Koo and Kopan 2004), have also been established. Therefore, it is possible that γ-secretases are not uniform complexes and different γ-secretase complexes exist in different combinations with various components, such as Aph1, Pen2, and/or presenilin isoforms, and these different complexes may have different cellular functions, for e.g., roles in signaling or degradation (Kopan and Ilagan, 2004). From the viewpoint of the signaling mechanism, it is not yet clear whether a large proportion of ICDs of these substrates that are released by γ-secretase are required for signaling events. In general, substrates of γ-secretase, such as APP, are considerably more abundant proteins than transcription factors, which are usually rare molecules. Therefore, although a large proportion of ICDs of these substrates are rapidly degraded, it seems likely that small amounts of remaining ICDs may fulfill its signaling function with small quantities of transcription factors. Thus, the majority of ICDs of these substrates may be degraded and only a small proportion may play roles in signaling.

In relation to this issue, Buoso et al. (2010) proposed an interesting model in which a specific stimulus triggers Fe65 binding to AICD to prevent AICD degradation by the proteasome. AICD/Fe65 complexes, alone or with Tip60, translocate to the nucleus, where they control the expression of certain genes in association with Tip60. On the other hand, different stimuli induce phosphorylation of AICD, which strongly inhibits binding to Fe65 and translocation to the nucleus. Phosphorylated AICD left in the cytosol is rapidly degraded, probably by insulin-degrading enzyme (IDE) (Edbauer et al. 2002), which is ubiquitously expressed with the highest levels of expression in the liver, testes, muscle, and brain. Indeed, it has been shown that when phosphorylated at Thr⁶⁶⁸ in the APP-695 isoform, AICD cannot bind to Fe65 (Kimberly et al. 2005).

From the viewpoint of the signaling mechanism, it is likely that AICD has no effect in the normal brain, because almost all AICD is rapidly degraded. However, both AICD and Aβ are overproduced in the AD brain. Therefore, although the majority of AICD is degraded, it is possible that a small amount of the remaining AICD may fulfill its function and cause neuron-specific cell death in the AD brain.

AICD Induces Neuron-Specific Apoptosis

There is accumulating evidence in support of the concept that AICD and its interacting proteins contribute to AD through APP signaling. For example, transgenic mice overexpressing AICD/Fe65 showed abnormal activity of glycogen synthase kinase 3 beta (Gsk3b protein) (Ryan and Pimplikar 2005), leading to hyperphosphorylation and aggregation of TAU, resulting in microtubule destabilization and reduction of nuclear β-catenin levels causing a loss of cell–cell contact mechanisms that may contribute to neurotoxicity in AD. Subsequent neurodegeneration and working memory deficits were also observed in these transgenic mice (Ghosal et al. 2009). In other experiments, similar transgenic mice exhibited abnormal spiking events in their electroencephalograms as well as susceptibility to kainic acid-induced seizures independent of Aβ (Vogt et al. 2009). In addition, c-Abl kinase was also reported to play a role in the transcriptional regulation of AICD, and c-Abl was shown to modulate AICD-dependent cellular responses, transcriptional induction, and apoptotic responses (Vazquez et al. 2009). Interestingly, elevated AICD levels have also been reported in the AD brain (Ghosal et al. 2009). Therefore, it is likely that AICD-dependent differential gene expression, reflecting the involvement of APP signaling, induces cell death, which may lead to AD pathology.

To explore APP signaling, we established several AICD-overexpressing P19 cell lines (Nakayama et al. 2008b). Undifferentiated AICD-overexpressing P19 cells retained an epithelial cell-like morphology similar to that of control cells, while the differentiated cells became round and showed a bipolar morphology with neurite extension. In addition, undifferentiated cells were negative for the neuronal marker Tuj1, while almost all differentiated cells were Tuj1-positive. These results indicated that all of these cell lines could differentiate into neurons. Although neurons were differentiated from these cell lines after aggregation culture with all-trans-retinoic acid (RA) treatment, AICD expression induced neuron-specific cell death. Indeed, while neurons differentiated from control cells were healthy, almost all neurons derived from AICD-overexpressing P19 cells showed severe degeneration 4 days after induction of differentiation (Fig. 3). Furthermore, DNA fragmentation was detected, and terminal deoxynucleotidyl transferase (TdT)-mediated deoxyuridine triphosphate (dUTP)-biotin nick end-labeling (TUNEL)-positive cells were also Tuj1-positive neurons. Based on these results, we concluded that AICD can induce neuron-specific apoptosis (Nakayama et al. 2008b). The effects of AICD were restricted to neurons, with no effects observed on non-neural cells. Thus, although further studies are required, it is possible that AICD plays a role in APP signaling, which leads to AD.

Fig. 3.

Overexpression of AICD in P19 cells induced neuronal cell death. After aggregation culture with RA, AICD-overexpressing P19 (AICD/P19) and control P19 cells (pCDNA/P19) carrying vector alone were replated and cultured for the indicated periods on dishes and allowed to differentiate. Undifferentiated AICD/P19 cells retained epithelial cell-like morphology similar to control cells, while the differentiated cells became round and showed a bipolar morphology with neurite extension. Two days after replating (Day 2), all cell lines grew well and neurons with long neurites appeared. Four days after replating (Day 4), control cells still grew well as clusters and many neurons had differentiated from these cells. However, many AICD/P19 cells showed severe degeneration, becoming spherical with numerous vacuoles and detached from the culture dishes

AICD Induces Dynamic Changes in the Gene Expression Profile

If APP signaling occurs, AICD should alter the expression of certain genes. To test this hypothesis and identify the genes involved in this process of neuron-specific cell death, we cultured both AICD-overexpressing P19 cells and control P19 cells and evaluated AICD-induced changes in gene expression at three time points during culture: the undifferentiated state, after 4 days of aggregation with RA (aggregated state), and 2 days after replating (differentiated state) (Ohkawara et al. 2011). The levels of expression of 41256 transcripts and transcript variants of more than 20000 independent genes were monitored by DNA microarray analysis (Ohkawara et al. 2011). As shown in Fig. 4, the expression levels of 277 genes were upregulated by more than 10-fold in the presence of AICD, as estimated from the intensity of hybridization signals. Conversely, 341 genes showed downregulation of expression to less than one tenth of the original level (Fig. 4). Reverse transcription-polymerase chain reaction of 17 selected genes showed excellent agreement with the microarray results and served to validate the findings of DNA microarray analyses.

Fig. 4.

Venn diagrams showing the total numbers of genes upregulated by more than 10-fold in the presence of AICD (a) and genes downregulated to less than one tenth of their original level (b) at three states of neural differentiation in P19 cells: undifferentiated, aggregated, and differentiated

Several genes were strongly upregulated in the presence of AICD (Table 2) (Ohkawara et al. 2011). For example, protein tyrosine phosphatase receptor T (Ptprt) was undetectable at all sampling points in control P19 cells. However, AICD-overexpressing P19 cells showed strong Ptprt gene expression at all sampling points: i.e., 906-fold, 204-fold, and 116-fold upregulation in the undifferentiated, aggregated, and differentiated states, respectively, estimated from the intensity of hybridization signals. These upregulated genes show various biological functions: for e.g., protein amino acid dephosphorylation (Ptprt), proteolysis (Cpb1), and cytoskeletal protein (Myh1).

Table 2.

Expression levels of 7 upregulated and 7 downregulated genes, as well as 3 housekeeping genes

	Gene symbol	Gene name	Function	Relative expression levels (fold)
				Undifferentiated	Aggregated	Differentiated
Non-regulated (house keeping)	Actb	β-actin	Cytoskeleton protein	−1.2	1.2	1
	Sdha	Succinate dehydrogenase subunit A	Electron transporter in the TCA cycle and respiratory chain	−1.1	−1.6	−1.2
	Eef1a1	Eukaryotic translation elongation factor-1 alpha 1	Essential component for the elongation phase during protein translation	1	−1.1	1.2
Upregulated	Ptprt	Protein tyrosine phosphatase receptor T	Protein tyrosine phosphatase that regulates STAT3 activity	906	204	116
	Cpb1	Carboxypeptidase B1	Hydrolysis of C-terminal end of basic amino acid peptide bond	16	296	222
	Nr2e1	Tailless homolog	Transcription factor that is essential for neural stem cell proliferation and self-renewal	5.8	244	54
	Myh1	Myosin heavy chain 1	One of the components of the motor protein myosin	−4.2	259	−1.1
	Dnahc7c	Axonemal dynein heavy chain	Essential for motility of cilia and flagellae	133	41	43
	Alkbh3	Alkylation repair homolog 3	AlkB enzyme that repairs methylation damage in DNA and RNA	69	80	43
	Ctgf	Connective tissue growth factor	Skeletogenesis/vasculogenesis by modulating BMP, Wnt, and IGF-I signals	90	54	40
Downregulated	Hes5	Hairy and enhancer of split 5	Transcription factor that inhibits neurogenesis	−8.7	−3039	−2515
	Slc10a6	Sodium-dependent organic anion transporter	Transport of sulfoconjugated steroid hormones and bile acids	−145	−785	−1212
	Nid1	Nidogen-1	Extracellular matrix linker protein	−304	−165	−507
	LOC213332	Analog of Na+ dependent glucose transporter 1	Putative glucose transporter	−232	−325	−306
	Dtx1	Deltex1	Regulator of Notch signaling pathway	−30	−85	−691
	Rbp4	Retinol-binding protein 4	Retinol transporter from the liver to extrahepatic tissues	−525	−100	−24
	Col3a1	Collagen type III alpha 1	Extracellular matrix protein	4.1	−29	−234

Relative expression levels (-fold) were estimated from the intensity of hybridization signals. Housekeeping gene expression was unaltered in AICD-overexpressing P19 and control P19 cells in all states, suggesting that these genes may not be affected by AICD. These results also indicated that the observed differences in expression were not due to technical problems, such as uneven hybridization or poor RNA quality

In contrast to these upregulated genes, AICD strongly inhibited the expression of several genes (Table 2) (Ohkawara et al. 2011). For example, although hairy and enhancer of split 5 (Hes5) expression level was extremely low in undifferentiated control P19 cells, its level of expression was markedly increased through the process of neural differentiation: i.e., increases in expression level of almost 300-fold were observed as estimated from the intensity of hybridization signals in the aggregated and differentiated states. However, this extreme induction through the differentiation process in control P19 cells was not detected in AICD-overexpressing P19 cells (Table 2), indicating that AICD strongly inhibits this induction. Several genes, such as the sodium-dependent organic anion transporter (Slc10a6), nidogen 1 (Nid1), and an analog of Na⁺-dependent glucose transporter 1 (LOC213332), were also strongly inhibited by AICD. Therefore, AICD mediates both the upregulation and downregulation of many genes. Although further studies are required, it is likely that AICD plays an important role in APP signaling.

Gene ontology (GO) analysis was performed, and these upregulated and downregulated genes were classified according to GO terms at several hierarchical levels (Ohkawara et al. 2011). Although a few of the upregulated and downregulated genes were classified into GO terms related to cell death, many of these genes were unrelated to cell death. Furthermore, we evaluated AICD-induced changes in expression of genes thought to be involved in cell death in AD; however, we found no significant changes in expression of these genes. Therefore, it is likely that AICD does not directly induce the expression of genes involved in cell death. Although further studies are required to resolve these issues, it is likely that the extreme dynamic changes in gene expression described here disrupt the homeostasis of certain neurons and give rise to neuron-specific cell death. In addition, we also evaluated AICD-induced changes in expression of several genes encoding AICD-interacting proteins, which are thought to regulate AICD stability, cellular localization, and transcriptional activity; however, no significant changes in expression of these genes were found.

There are many Reported Discrepancies Regarding AICD Target Genes and AD-Related Genes

Several candidate target genes expression of which is affected by AICD have been reported. However, there is still controversy regarding putative target genes, with different groups reporting significant or no significant changes in expression of certain target genes. Such conflicting conclusions are thought to be due to the use of different cell culture systems or animal models in these studies. Reflecting these problems, we found no significant correlations between our microarray data and those reported previously.

We compared our results with several previous reports of AD-specific DNA microarray data. For example, Blalock et al. divided AD patients into three classes based on cognition level: incipient, moderate, and severe. They compared the hippocampal expression profiles of these classes with those of normal subjects and extracted genes showing differential expression patterns (Blalock et al. 2004). However, there were no significant correlations between our data and the results reported by Blalock et al. (2004).

Liang et al. studied the expression profiles of six regions of the brain (entorhinal cortex, hippocampus, medial temporal gyrus, posterior cingulate, superior frontal gyrus, and primary visual cortex) and compared the profiles of AD brains with those of normal brains (Liang et al. 2008). We also compared our data with these results, and found no significant correlations.

As we found no significant correlations between our results and those reported in previous studies of AD brains, we compared the brain expression data reported by Blalock et al. (2004) with those reported by Liang et al. (2008); surprisingly, we also found no correlations between these two sets of data. These discrepancies are likely due to variable quality of RNA isolated from the AD brain, because it is technically difficult to isolate high-quality RNA from dying cells such as AD neurons. Therefore, AICD-expressing P19 cells described in this paper may be a more useful model not only for the study of APP signaling but also for cellular and molecular studies of AD.

Is Aβ Sufficient to Clarify all Aspects of the Onset of AD?

Autosomal dominant mutations in and around the Aβ region of APP are responsible for hereditary early-onset AD (Goate et al. 1991), probably due to acceleration of proteolytic processing. In humans, the APP gene is located on chromosome 21 (Kang et al. 1987), which is abnormally present in an extra copy in individuals with Down’s syndrome (trisomy 21). Almost all people with Down’s syndrome also develop AD by 40 years old, probably due to this gene dosage effect (Lott and Head 2005). In addition, both PS1 and PS2 are catalytic components of the γ-secretase complex (Iwatsubo 2004) and were identified by genetic linkage analyses as the genes responsible for familial AD (Sherrington et al. 1995; Rogaev et al. 1995; Levy-Lahad et al. 1995). Therefore, both APP itself and its proteolytic processing may also be responsible for the onset of AD.

The amyloid hypothesis is widely accepted as the mechanism of the onset of AD. The traditional amyloid hypothesis is that overproduced Aβ forms insoluble amyloid plaques, which are commonly observed in the AD brain and are believed to be the toxic form of APP responsible for neurodegeneration (Hardy and Selkoe 2002).

However, several questions have recently been raised regarding this hypothesis. One of the most significant arguments against the amyloid hypothesis is the presence of high levels of Aβ deposition in non-demented elderly people (Terry et al. 1999), suggesting that Aβ amyloid plaques are not toxic. Indeed, transgenic mice overproducing Aβ mimic the amyloid deposition seen in AD brains but do not show neurodegeneration (McGowan et al. 2003), although it has been reported that the soluble form of Aβ oligomer is toxic (Klein et al. 2001; Selkoe 2002). Furthermore, several anti-amyloid drugs have failed in phase III clinical trials (Abbott 2008). Therefore, some researchers have suggested that AD may be caused by an APP-derived fragment, but not necessarily Aβ-amyloid (Schnabel 2009). Indeed, both extracellular fragment and AICD are generated at the same time as Aβ is being generated. From this viewpoint, Nikolaev et al. (2009) reported important results indicating that APP is a ligand of Death receptor 6 (DR6), which mediates cell death and is expressed at high levels in the human brain regions most affected by AD. APP is cleaved by β-secretase, releasing the extracellular domain (sAPPβ), which is further cleaved by an as yet unknown mechanism to release a 35-kDa N-terminal fragment (N-APP). This N-APP fragment binds DR6 to trigger degeneration through caspase 6 in axons and caspase 3 in cell bodies. Therefore, it was suggested that N-APP may be involved in the onset of AD (Nikolaev et al. 2009).

As AICD induces neuron-specific apoptosis, AICD may also be involved in the onset of AD. Indeed, Ghosal et al. reported AD-like pathological features in transgenic mice expressing AICD (Ghosal et al. 2009), although conflicting conclusions have also been reported (Giliberto et al. 2010). Therefore, it is possible that AD is caused by multiple mechanisms mediated by multiple APP-derived fragments, not only Aβ but also N-APP and AICD.

Conclusions

Although γ-secretase was first identified as a protease that cleaves APP within the transmembrane domain and produces Aβ peptides, which are thought to be pathogenic in AD, the physiological functions of γ-secretase have not been fully elucidated. As reviewed here, the Notch signaling pathway is mediated by γ-secretase; intramembrane cleavage of Notch by γ-secretase serves to release ICD that has activity in the nucleus through binding to transcription factors. Recently, it was demonstrated that many type 1 transmembrane proteins, including Notch, Delta, and APP, are substrates for γ-secretase and release ICDs of these proteins from cell membranes. The observations that the common enzyme, γ-secretase, modulates proteolysis and the turnover of possible signaling molecules led to the attractive hypothesis that mechanisms similar to the Notch signaling pathway may contribute widely to γ-secretase-regulated signaling pathways, including APP signaling, which leads to AD. Indeed, as reviewed here, we have shown that AICD alters gene expression and induces neuron-specific apoptosis. Here, we discussed the possibility that APP may play a role in signaling events and that γ-secretase-regulated APP signaling may be responsible for the onset of AD.

Abbreviations

AD

Alzheimer’s disease

APP

Amyloid precursor protein

AICD

The intracellular domain of APP

bHLH

Basic helix-loop-helix

C. elegans

Caenorhabditis elegans

Dll

Delta-like protein

Dll1IC

The intracellular domain of Dll1

EGF

Epidermal growth factor

GO

Gene ontology

Hes

Hairy/enhancer of split

JM

Juxtamembrane

PS

Presenilin

ICD

Intracellular domain

NICD

The intracellular domain of Notch

RA

All-trans-retinoic acid

RIP

The regulated intramembrane proteolysis

TM

Transmembrane