Register for Mechanobiology 2016!

Building γ-secretase – the bits and pieces
Dragana Spasic, Wim Annaert


γ-Secretase is a promiscuous aspartyl protease responsible for the final intramembrane cleavage of various type I transmembrane proteins after their large ectodomains are shed. The vast functional diversity of its substrates, which are involved in cell fate decisions, adhesion, neurite outgrowth and synapse formation, highlights the important role γ-secretase plays in development and neurogenesis. The most renowned substrates are the amyloid precursor protein and Notch, from which γ-secretase liberates amyloid β peptides and induces downstream signalling, respectively. γ-Secretase is a multiprotein complex containing presenilin (which harbours the catalytic site), nicastrin, APH1 and PEN2. Its assembly occurs under tight control of ER-Golgi recycling regulators, which allows defined quantities of complexes to reach post-Golgi compartments, where γ-secretase activity is regulated by multiple other factors. 3D-EM rendering reveals a complex with a translucent inner space, suggesting the presence of a water-filled cavity required for intramembrane proteolysis. Despite huge efforts, we are now only beginning to unravel the assembly, stoichiometry, activation and subcellular location of γ-secretase.


Regulated intramembrane proteolysis (RIP) entails cleavage of a peptide bond within the hydrophobic environment of a lipid bilayer. This unusual proteolysis is carried out by four known classes of intramembrane cleaving proteases (I-CLiPs): S2P, the signal peptide peptidases (SPPs), the rhomboids and the presenilins (PSs). Each can recognise numerous substrates, and RIP has therefore been implicated in a wide range of cellular responses (Annaert and De Strooper, 2002; Dillen and Annaert, 2006; Wolfe and Kopan, 2004). For example, both sterol-regulatory-element-binding protein (SREBP) and ATF6 undergo cleavage by the zinc protease S2P (following a luminal cleavage by S1P), which results in upregulation of genes involved in cholesterol biosynthesis and the unfolded protein response (UPR), respectively (Brown et al., 2000; Ye et al., 2000). SPPs, also known as presenilin-like homologues (PSHs), are aspartyl proteases that catalyse the proteolysis of remnant signal peptides after they have been cleaved from their precursor by signal peptidases. This activity is required to generate, for example, antigen-E epitopes in human lymphocytes that are subsequently recognised by the immune system. It is also exploited by hepatitis C virus to process core proteins and used in bacteria during processing of the prepilin leader peptide prior to secretion (LaPointe and Taylor, 2000). Rhomboids, a large family of serine proteases, provide another example of how a conserved RIP mechanism functions in various processes ranging from bacteria to man. For instance, in parasites and bacteria, their roles are related to host cell invasion and quorum sensing; in Drosophila, rhomboid-1 regulates EGFR signalling through cleavage of ligands such as Spitz (Urban, 2006); in mammals, the mitochondrial rhomboid PARL participates in cristae modelling and apoptosis through OPA1 processing (Cipolat et al., 2006).

The substrates are most diverse for presenilins. To date, PS1 and PS2 are known to target >30 type-I transmembrane proteins. The first substrate identified was the amyloid precursor protein (APP), in which γ-secretase is responsible for the final cleavage that liberates amyloid β peptides (Aβ), a major component of senile plaques in Alzheimer disease (AD) (Wolfe and Guenette, 2007). PSs were first recognised by geneticists as the protein products of genes mutated in rare cases of familial AD (FAD) (Levy-Lahad et al., 1995; Rogaev et al., 1995). Since then, over 150 mutations have been described in PS1 and about ten in PS2 ( FAD-PS mutations are now seen as loss-of-function mutations that result in incomplete C-terminal trimming of Aβ peptides and consequently an increase in the amount of the longer, more neurotoxic and aggregation-prone Aβ42 peptide (De Strooper, 2007; Shen and Kelleher, 3rd, 2007). However, the therapeutic value of targeting γ-secretase in AD is compromised by the fact that another substrate, Notch, is required for many cell fate decisions during development and through adulthood. Further complications arise from the fact that the catalytic PS has additional functions independent of γ-secretase, including roles in organelle transport and turnover (Esselens et al., 2004), in Akt-ERK signalling (Kang et al., 2005), in cytoskeletal dynamics (Khandelwal et al., 2007) and as Ca2+ leak channels in the endoplasmic reticulum (ER) (Tu et al., 2006).

PSs also distinguish themselves from other I-CLiPs in their absolute requirement for the co-factors nicastrin (NCT), anterior pharynx defective 1 (APH1) and PS enhancer 2 (PEN2). These four integral membrane proteins form a functional γ-secretase complex in which PSs provide the catalytic aspartyl residues (De Strooper et al., 1998; Wolfe et al., 1999). NCT was identified as a major protein co-immunoprecipitating with anti-PS1 antibodies (Yu et al., 2000), whereas APH1 (Goutte et al., 2002) and PEN2 (Francis et al., 2002) appeared in a genetic screen for new Notch modifiers in C. elegans. Here we discuss the functions of individual components in this complex as well as the regulation of its assembly and final stoichiometry.

Presenilin at the heart of γ-secretase

Mammalian PSs come in two flavours, PS1 and PS2, which are highly homologous. Both are polytopic membrane proteins and have ten hydrophobic domains, of which nine are proposed to span the membrane (Laudon et al., 2005; Oh and Turner, 2005; Spasic et al., 2006) (Fig. 1). PSs are initially translocated in the ER as full-length proteins but get converted by endoproteolysis within hydrophobic domain 7 to stable N- and C-terminal fragments (NTFs and CTFs, respectively) that associate to form a heterodimer. Although both fragments are part of the catalytic γ-secretase, endoproteolysis is not a requirement for activity (Li et al., 2000; Steiner et al., 1999) (reviewed in Dillen and Annaert, 2006).

Two highly conserved aspartate residues (Asp257 and Asp385 in human PS1) within transmembrane domain 6 (TMD6) and TMD7 constitute the core of the catalytic site. Mutation of either abolishes γ-secretase activity (Nyabi et al., 2003; Wolfe et al., 1999). Together with surrounding residues, they mark a highly conserved YD/GxGD consensus motif (Steiner et al., 2000). Tyr389 may also contribute to the catalytic site (Tolia et al., 2006). The conformation of the active site also depends on more remote sequences within PS1, such as the C-terminal PAL motif (Wang, J. et al., 2006) and Cys residues in TMD1 and TMD8 (Kornilova et al., 2006). Binding of an active-site-directed inhibitor prevents the disulphide cross-linking between TMD1 and TMD8, implicating their close proximity to or allosteric interaction with the catalytic site. Finally, catalytic activity must be sensitive to subtle changes in the overall TMD conformation, given the effects of the scattered FAD-linked mutations.

These interactions between remote parts of the molecule fit with the idea that PS1 adopts a ring-like topology (Annaert et al., 2001). This model is based on the identification of APP-binding regions in PS1 – TMD1 (extending to part of the first luminal loop domain) and the C-terminus – and led to the idea that the substrate must first bind to a remote docking site at the heterodimer interface prior to entering the active site. Indeed, subsequent imaging and biochemical studies have demonstrated that transition-state γ-secretase inhibitors that bind to PS1 do not affect the interaction between PS1 and the APP C-terminal fragment (Berezovska et al., 2003; Esler et al., 2002). Similarly, inhibitors based on the substrate TMD do not prevent labelling of the complex by active-site-directed inhibitors (Kornilova et al., 2003). Interestingly, these peptide inhibitors label only PS heterodimers and not the other γ-secretase components, which indicates that substrates directly dock on PS (Kornilova et al., 2005).

Fig. 1.

The membrane topologies of the four known components of the γ-secretase complex: PS1, PEN2, APH1 and NCT. The arrow indicates the endoproteolytic cleavage site in hydrophobic region 7 of PS1, which generates the PS1 NTF-CTF heterodimer that is the active form. Red asterisks denote the position of the catalytic aspartate residues in TMD6 and TMD7 of PS1. TMDs that are involved in intermolecular interactions between γ-secretase components are indicated with the mixed colours of the respective components.

The feature that PS/γ-secretase of course shares with all I-CLiPs is the hydrolysis of a peptide bond within TMDs. This requires access of the catalytic site to water molecules, which is difficult to understand given the hydrophobic environment of the lipid bilayer. However, the recent success in crystallising GlpG, a rhomboid protease in E. coli, has resolved this biochemical conundrum. The reaction actually takes place in a V-shaped cavity separated from the lipid environment by six TMDs (Wang, Y. et al., 2006). Moreover, the catalytic His-Ser dyad is buried at the base of this cavity, which opens at the extracellular site. In the closed conformation, the cleft is capped by a flexible loop 5 (Wu et al., 2006). Lifting the capping loop and/or lateral displacement of helix 5 (Baker et al., 2007) is suggested to be induced by substrate binding. This not only allows water to access the cavity but destabilises hydrophobic side chain interactions, thereby opening a lateral gate for the substrate to reach the catalytic site (Wang and Ha, 2007). Although members exhibit different topologies, the principal mechanisms underlying water and substrate accessibility are conserved within the S2P family – as revealed by the structure of an archaebacterial S2P (Feng et al., 2007). Here, water enters via a cytosol-oriented V-shaped cavity that converges on the catalytic zinc atom ∼14Å from the lipid surface. In the open conformation, TMD1 and TMD5-TMD6 separate, generating a funnel-shaped groove that allows lateral substrate entry and probably accommodates unwinding of the substrate's transmembrane helix.

Whether these mechanisms apply to the aspartyl proteases remains to be explored. In any event, a lower-resolution 3D-EM reconstruction of purified active γ-secretase revealed a central cavity, with two lateral thin-density regions and two openings, which were suggested to be involved in substrate gating and release of proteolytic fragments, respectively (Lazarov et al., 2006) (Fig. 2). Recently two studies provided biochemical evidence for the contribution of TMD6 and TMD7 to a hydrophilic pore (Sato et al., 2006; Tolia et al., 2006). Based on the fact that residues in the cytosolic half of TMD6 and TMD7 are water-accessible from the intracellular, but not the extracellular side, Sato et al. (Sato et al., 2006) proposed that the hydrophilic space around the catalytic aspartate residues has a narrowed funnel-like shape that resembles the V-shaped cavity seen in rhomboids (Wang, Y. et al., 2006).

The 3D-EM study by Lazarov et al. also revealed that the NCT ectodomain covers the extracellular opening of the chamber (Lazarov et al., 2006). This is consistent with its role as a gatekeeper and substrate receptor in γ-secretase (Shah et al., 2005). A crucial Glu residue in the luminal DAP subdomain of the NCT ectodomain interacts with the free amino group of short substrate stubs generated by ectodomain shedding. This ensures that γ-secretase targets substrates whose ectodomain has been shed rather than relying on their primary sequence and allows it to cleave various substrates despite their lack of sequence similarity near the proteolysis site.

Fig. 2.

3D structure of the γ-secretase complex. (A) Surface rendering of the 3D reconstruction. The first row displays side views generated by rotation around a vertical axis, the second row shows tilted views generated by rotation around a horizontal axis. The rotation angles are shown within each view. Two openings at the top and bottom are labelled H1 and H2, respectively, where visible. The top density is labelled NCT because lectin labelling showed that the NCT ectodomain is located at this surface (Lazarov et al., 2006). (B) The potential transmembrane segment has a belt-like structure and is indicated by two parallel dashed lines, 60 Å apart. For size comparison, a typical transmembrane α-helix, taken from the rhodopsin structure (protein data bank ID code 1GZM), is shown to the left of the structure. (C) A cut-open view of the γ-secretase complex, revealing a large central chamber and one opening at the top (H1) and one at the bottom (H2). Two weak-density lateral regions are labelled with asterisks. Reproduced from Lazarov et al. PNAS, 103(18), 6889-6894, 2006 with permission [Copyright (2006) National Academy of Sciences, USA].

What the low-resolution 3D-EM does not reveal is how and where NCT, APH1 and PEN2 associate in the complex relative to PS or the contribution of TMDs bordering the translucent inner space. The lateral thin-density regions may for instance represent interfaces between different components, PS heterodimers [TMD1-TMD8 (Kornilova et al., 2006)] [TMD6-TMD7 (Sato et al., 2006; Tolia et al., 2006)] or even two PS molecules. Indeed, some studies indicate that the catalytic core consists of a PS dimer, one PS providing the substrate-binding site and the other the catalytic cleavage site (Cervantes et al., 2004; Schroeter et al., 2003). Also, we cannot exclude the possibility that, next to the NCT ectodomain, other regions of PS1, such as the long extracellular loop domain or adjacent TMDs, act as a `plug' that regulates the opening of the pore/cavity, as suggested for GlpG (Wu et al., 2006) and the SecY translocation channel (Van den Berg et al., 2004). This will require higher-resolution studies in combination with smart tagging of components or crystal structures. These could also substantiate the biochemical evidence supporting a 1:1:1:1 stoichiometry of the four components (Sato et al., 2007).

Mapping γ-secretase component interactions

At least 19 TMDs contribute to the hydrophobicity of γ-secretase, of which nine are from PS and seven are from APH1. PEN2 has a hairpin-like topology containing two TMDs, whereas NCT is the only type I transmembrane glycoprotein of the complex (Fig. 1). Genetic ablation of only one component results in mislocalisation, incomplete maturation and destabilisation of the remaining components, clearly indicating that inter- and intramolecular interactions are crucial in the course of assembly and activation of the γ-secretase complex (Fig. 3). Assembly is most likely to be initiated through the formation of an NCT-APH1 subcomplex to which the other components, either single or in subcomplexes, are added. The many pair-wise interaction studies have provided important information that helps us to understand this stepwise assembly of γ-secretase.

The importance of NCT in the complex is not restricted to APP recognition (Shah et al., 2005) and docking (Berezovska et al., 2003). The assembly and activity of the γ-secretase complex requires the integrity of the complete NCT ectodomain (Shirotani et al., 2003; Yu et al., 2000). The DAP domain of NCT must play a crucial role, because deletions (Yu et al., 2000) and single point mutations (Shirotani et al., 2004) significantly reduce or even abolish the interaction with PS or PEN2, without influencing binding of NCT to APH1a. The interaction with APH1, by contrast, involves the TMD of NCT (Fig. 3). An N-terminal sequence of four – mostly polar residues – is crucial (Capell et al., 2003). Interestingly, those residues are equally important for the interaction of NCT with retrieval to the endoplasmic reticulum 1 protein (Rer1p), a retrieval receptor controlling ER-Golgi trafficking of NCT (see below) (Spasic et al., 2007). Conserved residues in the juxtamembrane region of the NCT ectodomain – Ser632 and Trp648 – may stabilise the NCT-APH1 interaction (Walker et al., 2006). The C-terminus of PS1 is crucial for the interaction with NCT TMD, because deleting the last residue is sufficient to block γ-secretase assembly, NCT maturation and, to a lesser extent, PEN2 stabilisation (Kaether et al., 2004). The reason for this differential effect might be the fact that PEN2 interacts with the PS1 NTF and not the CTF (Fig. 3). More precisely, the first TMD of PEN2 (Kim and Sisodia, 2005b) binds to an `NF' motif in TMD4 of PS1 (Kim and Sisodia, 2005a; Watanabe et al., 2005) (Fig. 3), and this interaction is independent of the interaction with the NCT-APH1 subcomplex (Watanabe et al., 2005).

In the absence of PEN2, PS can be stabilised as a holoprotein by binding to NCT-APH1 (Takasugi et al., 2003). Hence, it has been suggested that PEN2 TMD1 is involved in PS endoproteolysis – it could induce a conformational change rendering the cleavage site more accessible or serve as a co-factor. Binding of PEN2 alone to PS1, however, is not sufficient for endoproteolysis (Kim and Sisodia, 2005a). After endoproteolysis, PS1 NTF and CTF are stabilised by interactions with the PEN2 C-terminus (Kim and Sisodia, 2005b). Interestingly, altering the length or integrity of either the luminal N- or C-terminus of PEN2 affects binding to other components and γ-secretase activity (Crystal et al., 2003; Hasegawa et al., 2004; Isoo et al., 2007; Prokop et al., 2005). PEN2 thus seems to use its `hairpin' topology literally, hooking up and stabilizing the final γ-secretase complex.

Fig. 3.

Intra- and intermolecular interactions in the γ-secretase complex. Given the hydrophobic nature of γ-secretase, most prominent and primary interactions are likely to be governed through or include their TMDs and hydrophobic domains. Here we present a bird's eye view of the TMDs of PS1 (yellow), NCT (green), APH1 (blue) and PEN2 (red), including the reported intra- (grey arrows) and intermolecular (black arrows) interactions. Suggested interactions are shown by dotted arrows, such as intramolecular interactions in APH1 (via GxxG motifs in TMD4) and the as-yet-uncharacterised interaction domains for APH1 in PS1 and for NCT in APH1. Ectodomain interactions of NCT with APP-CTF are indicated with a green arrow (see text for details). The red sparkle denotes the catalytic aspartate dyad.

Interactions involving APH1 are less well characterised. For instance, the APH1 TMD(s) interacting with NCT has not been determined but may reside in the second half of APH1 (Fortna et al., 2004) (Fig. 3). Mutations in a conserved GxxxG motif in the fourth TMD of APH1 abolish binding to PS without affecting NCT-APH1 subcomplex assembly. The additional fact that PS1 also binds to the NCT TMD suggests that the helix-helix interface between the NCT TMD and TMDs of APH1 provides a broad interaction surface for PS. The GxxG motifs are also required for the stability of APH1 itself – perhaps engaging in intramolecular interactions with GxxG-like motifs in other TMDs (Niimura et al., 2005) (Fig. 3).

Although the interactions are not fully mapped, it is obvious that assembly of γ-secretase is complex and requires both inter- and intramolecular interactions of individual components. Note also that interfering with these interactions often affects localisation of other components, which suggests that complex assembly and activation is intrinsically coupled to transport. It also follows that the component that exhibits the lowest abundance in a cell acts as the limiting factor for γ-secretase assembly and activity (Kimberly et al., 2003).

Building γ-secretase

Although all components colocalise initially in the ER, their assembly is not a random process but occurs sequentially and stoichiometrically and is superimposed on transport regulation that ensures cell- and tissue-specific levels of γ-secretase activity. Our knowledge about the sequence of these events is based on interaction studies, step- and pair-wise overexpression of various components and the identification of high-molecular-weight complexes by (blue) native-gel electrophoresis (BN-PAGE). All studies agree on the initial formation of an NCT-APH1 subcomplex as the first step. This subcomplex is stable even in the absence of PS and PEN2 (Shirotani et al., 2004), and is most resistant to detergent dissociation compared with other subcomplexes (Fraering et al., 2004). The following steps in γ-secretase assembly are less clear. There are basically two hypotheses (Fig. 4). Full-length PS has been shown to migrate in a high-molecular-weight complex with both NCT and APH1, which suggests that PEN2 joins a pre-existing trimeric intermediate and subsequently causes PS endoproteolysis (LaVoie et al., 2003; Niimura et al., 2005). The alternative hypothesis is based on identification of two additional intermediate complexes, NCT-APH1-PS1 CTF and PEN2-PS1 NTF. The fact that PEN2 can bind full-length PS independently of NCT and APH1 supports this view (Fraering et al., 2004). The PEN2-PS intermediate should be formed before or during the endoproteolysis of PS. In some alternative views, stabilisation of full-length PS by NCT-APH1 leads ultimately to the removal and degradation of APH1 from the maturing complex (Hu and Fortini, 2003) but this has not been confirmed.

Coupling assembly to transport

From the moment the individual components are co-translationally inserted into the ER until they function in an active complex in distal, mostly post-Golgi compartments (see Kaether et al., 2006 and references therein), γ-secretase-complex assembly encounters many hurdles. Since only a relatively small pool of endogenous PS1 is associated with active complexes, mechanisms controlling cellular levels of γ-secretase activity must exist. The mutual dependency of the stability, maturation and transport of the components pins γ-secretase assembly down to early biosynthetic compartments (Fig. 5). Full assembly might even occur in the ER (Capell et al., 2005; Kim et al., 2007). Although we cannot exclude this, several findings indicate that ER-Golgi recycling mechanisms play an active role in the stepwise assembly of γ-secretase. For example, endogenous PS1 is enriched in coat protein complex I (COPI)-coated areas of the intermediate compartment (IC) (Annaert et al., 1999; Rechards et al., 2003), and mature glycosylation of NCT displays slow kinetics (Herreman et al., 2003). Furthermore, in vitro reconstitution of ER exit has revealed prevalent exit of full-length PS1 in COPII-coated vesicles, indicating that endoproteolysis and maybe PEN2 association occurs later (Kim et al., 2005). In the same experimental set-up, the NCT-APH1 subcomplex readily exits the ER in the absence of PS, which demonstrates that this subcomplex is already present in the ER and that complete assembly of γ-secretase is not a prerequisite for individual components or subcomplexes to leave the ER (Katleen Dillen and W.A., unpublished data).

Fig. 4.

Temporal sequence of interactions leading to γ-secretase complex formation. The earliest step in complex assembly is probably the association of NCT (green) with APH1 (blue) in the ER. From then on two `pathways' are proposed, depending on the experimental approaches used. (A) Full-length PS1 (yellow) might bind to NCT-APH1, this is followed by recruitment of PEN2 (red), which promotes endoproteolysis and stabilisation of the full complex. (B) Alternatively, full-length PS1 might first bind to PEN2, leading to endoproteolysis and binding to the pre-existing NCT-APH1 subcomplex.

Fig. 5.

Percolation-model that integrates γ-secretase assembly into ER-Golgi transport regulation. Individual and newly synthetised γ-secretase components are co-translationally inserted in the ER. Given that NCT interacts very early with APH1 and that Rer1p preferentially binds to immature NCT, we suggest that immature NCT `percolates' between the ER and Golgi compartments (a). It can be captured by Rer1p in the IC (a′) or cis-Golgi (a″) and retrieved via COPI-coated organelles to the ER. By competing with the NCT TMD, APH1 can displace Rer1p to form an NCT-APH1 subcomplex. This can occur in the ER (b) as well as IC compartments (b′). Once NCT-APH1 is formed, PS1 and PEN2 (either sequentially or together, see Fig. 4) join it to form a fully assembled complex. Again, this event does not need to be restricted to the ER (c) but can also take place in the IC (c′). Full-length PS1 can exit the ER (Kim et al., 2005), which suggests that endoproteolysis (and PEN2 interaction) occurs later and maybe during additional retrieval events from Golgi/IC to ER. The concentration of endogenous PS1 in COPI-coated membranes (Rechards et al., 2003) indicates that PS1 percolates through early biosynthetic compartments but potential retrieval mechanisms have not yet been identified. Hence, the ER-Golgi quality control system ensures that monomeric γ-secretase components are selectively retrieved from the cis-Golgi to the ER through interaction with cargo-retrieval receptors. Conversely, full complex assembly leads to a masking of the interactions with these retrieval receptors, allowing escape from ER-Golgi recycling (d) and transport of assembled complexes to their final destination in distal compartments, including the cell surface and endosomes. APP behaves differently because it has a short residence time in ER-Golgi compartments and gets maturely glycosylated quickly. This supports the finding that APP trafficking from the ER is uncoupled from PS1 (Kim et al., 2005) (e).

The coupling of γ-secretase assembly to transport is reminiscent of secondary ER quality control in the assembly of several heteromultimeric membrane proteins. MHC class II proteins and K+ channels provide examples of how retrieval from the Golgi, which depends on double arginine motifs in individual components, can keep improperly assembled subunits of a complex in pre-Golgi compartments. Assembly masks retention/retrieval signals, allowing only correctly formed complexes to pass the Golgi (Michelsen et al., 2005). Because no typical retention/retrieval motifs are present in individual γ-secretase components, besides an ER retention motif in the PS1 C-terminus (Kaether et al., 2004), recycling may be controlled by additional proteins bearing such motifs. At least one such protein was recently identified as Rer1p (Kaether et al., 2007; Spasic et al., 2007). In S. cerevisiae, Rer1p retrieves escaped ER proteins (e.g. Sec12p, Sec71p and Sec63p) or unassembled Fet3p subunits (Sato et al., 2004). Although two different γ-secretase components – NCT and PEN2 – were shown to interact with Rer1p (Kaether et al., 2007; Spasic et al., 2007), functional implications for γ-secretase assembly and activity were only demonstrated in the case of the Rer1p-NCT interaction (Spasic et al., 2007). Moreover, Rer1p recognises the same polar residues within the NCT TMD that are important for the interaction between NCT and APH1. Rer1p and APH1 thus compete for binding to NCT. Binding to Rer1p could therefore selectively retrieve immature NCT from the IC or cis-Golgi back to the ER, preventing it from premature escape and enhancing the probability of encountering APH1 (Fig. 5). Binding of APH1 would sterically mask the Rer1p interaction, allowing the NCT-APH1 subcomplex to escape the Rer1p-dependent `percolation' mechanism (Fig. 5). Subsequent binding of the PS1 C-terminus to the TMD of NCT could provide a molecular mechanism to `lock' the NCT-APH1 interaction into a maturing complex, preventing it from shifting back to the Rer1p interaction. Thus, Rer1p can be considered a novel limiting factor that negatively regulates the stepwise assembly of the γ-secretase complex. Finally, full assembly of the complex may subsequently also mask the PS1 ER-retention motif, providing a dual `hide and run' mechanism for γ-secretase complexes to reach distal compartments.

We cannot exclude the possibility that Rer1p targets additional components or that other ER-Golgi retrieval receptors are involved in γ-secretase-complex assembly. This is likely to occur given the enrichment of PS1 in COPI-coated areas in early compartments and – maybe more importantly – its many γ-secretase-independent functions. Mechanisms excluding PS1 from being incorporated into γ-secretase and fostering these alternative functions remain to be explored.

γ-Secretase interactors: activity or transport modulators?

Numerous other proteins interact with PS/γ-secretase, and many of these interactions have not been explored in great detail. It is therefore difficult to judge whether they affect the activity or instead modulate trafficking of γ-secretase and/or its substrates. For instance, phospholipase D1 (PLD1) binds specifically to PS1 and also negatively regulates the generation of Aβ (Cai et al., 2006). The finding that PS1 recruits PLD1 at the Golgi–trans-Golgi network suggests that this interaction involves the assembled γ-secretase complex. Moreover, because PLD1 overexpression results in decreased association of γ-secretase components, the PLD1 interaction may alternatively function in complex disassembly. Surprisingly, the interaction of transmembrane emp24-like trafficking protein 10 [TMED10, also known as transmembrane protein of 21 kDa (TMP21) or p23] – a member of the p24 family of cargo retrieval receptors in ER-Golgi transport – with γ-secretase does not affect the trafficking/assembly of γ-secretase. Instead, it specifically modulates γ-secretase activity towards the γ site but not other cleavage sites within the APP TMD, such as the ϵ-site (Chen et al., 2006).

Two other factors are potential negative regulators of γ-secretase activity but have no influence on complex assembly. One are the kinases ERK1/2, which reduce the activity of the complex by phosphorylating NCT (Kim et al., 2006). The other is CD147, a membrane glycoprotein claimed to be an integral regulatory subunit of the native γ-secretase complex whose downregulation increases Aβ production (Zhou et al., 2005). Interestingly, CD147 transcription is activated by the transcription factor SCP2 (sterol carrier protein 2). High levels of SCP2 in neuronal cells could be therefore an important regulator of γ-secretase activity in the brain (Ko and Puglielli, 2007).

Fraering and co-workers have shown that ATP activates γ-secretase in vitro by directly binding to a nucleotide-binding site in the PS1 CTF. ATP promotes γ-secretase-mediated production of Aβ and the APP intracellular domain (AICD) but not release of the Notch intracellular domain (Fraering et al., 2005). Given this selective effect, exploring the use of tyrosine kinase inhibitors – which target ATP-binding sites – may identify novel drug strategies for selective inhibition of APP RIP (Eisele et al., 2007; Netzer et al., 2003).

At first glance, it is striking that many interacting proteins share a common role as negative regulators of γ-secretase assembly or activity. However, the cell- and tissue-specific expression of such proteins provides means to adapt γ-secretase levels to the specific needs of a given cell or tissue to process different substrates. Other cellular factors may also influence γ-secretase activity, such as the local lipidic microenvironment. γ-Secretase complexes are recovered in detergent-resistant membranes (DRMs) isolated from post-Golgi compartments (Vetrivel et al., 2004). This, however, does not conclusively indicate they are components of lipid rafts. Moreover, lipid rafts are microdomains rich in highly ordered cholesterol and sphingolipids, and it is conceptually difficult to fit in an 8-10 nm (Lazarov et al., 2006) γ-secretase complex that includes 19 TMDs. To overcome this and explain the DRM association, γ-secretase might instead be dynamically targeted to cholesterol-rich nanodomains [liquid-ordered or Lo domains in Hancock (Hancock, 2006)] that collide to form larger domains or `lipid-shells' surrounding membrane proteins (Anderson and Jacobson, 2002). If so, one or more γ-secretase components might directly interact with cholesterol, but this remains to be demonstrated. So far, the relation of γ-secretase with cholesterol and sphingomyelin metabolism appears rather indirect and mediated through Aβ40/Aβ42 levels (Grimm et al., 2005).

Since γ-secretase is active in different compartments ranging from the trans-Golgi network to endosomes and the cell surface, its localisation defines to an extent the likelihood of different substrates becoming processed. Environmental factors that affect (directly or indirectly) its localisation should also affect proteolysis and downstream signalling. Note that activation of certain G-protein-coupled receptors (e.g. β-adrenergic and δ-opioid receptors) induces their internalisation along with cell-surface-localised γ-secretase, which results in elevated Aβ production and amyloid plaque formation in an AD mouse model (Ni et al., 2006).

Concluding remarks

Despite the tremendous efforts worldwide, many aspects of the cell biology of γ-secretase are still puzzling. Currently, we have only a glimpse of the complicated inter- and intra-molecular interactions involved. The best available 3D-EM study and the few biochemical data seem to support the idea of a water-filled cavity required for intramembrane proteolysis. However, we will probably have to wait for the γ-secretase complex to be crystallised to reveal all its secrets. Once this is achieved, the stoichiometry of the complex will be clear (including whether it is dimeric), we will be able to identify which components or subdomains contribute to the central cavity and the docking/gating of different substrates. This knowledge will be important if we are to develop novel compounds specifically interfering with APP processing.

Other questions also urgently await further investigation. Although specific roles are attributed to PS and NCT, the exact contributions of APH1 and PEN2 to assembly and activation are far from understood. PS1 and PS2 belong to mutually exclusive γ-secretase complexes and the same holds true for APH1a [including the short and long splice variants (APH1aS and APH1aL, respectively), APH1b and APH1c (Hebert et al., 2004)]. Several differently composed complexes may therefore co-exist in the same cell. But are their assemblies regulated in an identical manner or do they differ? Do different active complexes localise to different compartments or subdomains/subpopulations within the same compartment? And, if so, does this provide a basis for a differential selectivity for different cleavage sites within one substrate or between different substrates?

These numerous unanswered questions not only leave γ-secretase in the game as a potential drug target but further stress how important it is to understand its fine cell biological and structural details before we can comprehend the full spectrum of its impact in health and disease.


The W.A. laboratory is indebted to KULeuven (GOA2004/12; IUAP-VI 2007-11 P6/43), VIB, Stichting Alzheimer Onderzoek (SAO-FRMA 2006) for strong financial support. D.S. holds a PhD fellowship of the KULeuven (OE/07/26).

  • Accepted December 13, 2007.


View Abstract