Regulated intramembrane proteolysis is a novel mechanism involving proteases that hydrolyze their substrates in a hydrophobic environment. Presenilin (PS) 1 and PS 2 are required for intramembrane cleavage of an increasing number of type I membrane proteins, including the amyloid precursor protein of Alzheimer's disease and the Notch receptor, which signals during differentiation and development. Mutagenesis, affinity labeling, biochemical isolation, and reconstitution in cells reveal that PS, in complex with co-factors nicastrin, APH-1 and PEN-2, apparently contains the active site ofγ-secretase, a novel membrane aspartyl protease. In addition, other related aspartyl proteases have been identified. These include members of the type-4 prepilin peptidase family in bacteria, which are known proteases and carry a GD motif conserved in PS. A group of multi-pass membrane proteins found in eukaryotes also contain YD and LGXGD motifs in two transmembrane domains that are conserved in PS and postulated to constitute an aspartyl protease active site. Among these is signal peptide peptidase (SPP), which cleaves remnant signal peptides derived from signal-peptidase-mediated ectodomain shedding. SPP cuts type II membrane proteins, illustrating that PS-like proteases play a key role in intramembrane proteolysis of single-pass membrane proteins oriented in either direction.
The deposition of the amyloid β protein (Aβ) in the form of neuritic plaques is a primary pathological lesion of Alzheimer's disease (AD). Mounting evidence points to the 38-43 residue Aβ as an initiator of pathogenesis and not just a molecular bystander. Strong evidence for this hypothesis comes from genetics: early-onset, dominant forms of familial AD(FAD) are caused by mutations in the amyloid precursor protein (APP) and in presenilin 1 (PS1) and presenilin 2 (PS2), and these mutations affect Aβproduction. The 42 residue form of Aβ (Aβ42) is the predominant Aβ species deposited in the limbic and association cortices of both sporadic and familial AD patients(Gravina et al., 1995; Iwatsubo et al., 1994; Lemere et al., 1996; Mann et al., 1996a; Mann et al., 1996b). Increased amounts of Aβ42 are produced from stably transfected cells and transgenic mice overexpressing mutant presenilin (PS) genes(Borchelt et al., 1996; Citron et al., 1997; Duff et al., 1996; Xia et al., 1997), and in the plasma of mutant gene carriers (Scheuner et al., 1996). Mutations in PS genes account for >50% of early-onset FAD (Levy-Lahad et al.,1995; Rogaev et al.,1995; Sherrington et al.,1995). Although FAD cases represent a minority of all AD cases,the mechanism by which PS mutations affect Aβ production and cause FAD should provide insight into the pathogenesis and treatment of sporadic AD as well.
The production of Aβ from APP involves sequential proteolysis byβ- and γ-secretases. Cleavage by γ-secretase is heterogeneous and generates C-terminal variations in Aβ. Because FAD mutations in PS increase Aβ42 levels in particular, PS must affect the selectivity of γ-secretase. Deletion of PS1 in mice is lethal, producing a major disruption of somite segmentation(Shen et al., 1997; Wong et al., 1997). Neurogenesis is impaired, and massive neuronal loss is observed in specific regions (Shen et al., 1997). The developmental abnormalities and embryonic lethality can be rescued by both wild-type and FAD-mutant PS1 (Davis et al.,1998; Qian et al.,1998), which indicates that disease-causing PS mutants are at least partially functional. Cleavage of APP by γ-secretase in cultured PS1-knockout neurons is markedly inhibited(De Strooper et al., 1998; Herreman et al., 2000; Zhang et al., 2000), and PS-/- PS2-/- neurons produce undetectable levels of secreted Aβ. This clearly demonstrates that PS is required forγ-secretase activity and Aβ generation.
Studies with inhibitors have shown that γ-secretase has characteristics of an aspartyl protease(Shearman et al., 2000; Wolfe et al., 1999a). Identification of two conserved aspartate residues in transmembrane (TM)domains 6 and 7 of PS critical for γ-secretase activity (D257 and D385 in PS1) provided the conceptual basis for the novel hypothesis that PS is the aspartyl γ-secretase (Wolfe et al.,1999b). When either aspartate residue in PS1(Wolfe et al., 1999b) or PS2(Kimberly et al., 2000; Steiner et al., 1999) is mutated, γ-secretase activity is blocked, and Aβ levels are significantly reduced in cultured cells. This was also observed in transgenic mice overexpressing aspartate D257A mutant PS1(Xia et al., 2001). Further evidence summarized below suggests that not only is PS an aspartyl protease but an entire class of PS-like proteases is involved in intramembrane proteolysis of both type I and II membrane proteins.
PS: the active site of γ-secretase
PS1 and PS2 are 467 and 448-residue polypeptides, respectively, and share∼60% sequence similarity. Full-length PS undergoes endoproteolysis to form stable N-terminal (NTF) and C-terminal (CTF) fragments, which remain associated (Borchelt et al.,1996; Capell et al.,1998; Thinakaran et al.,1996). Transition state analogue γ-secretase inhibitors that were designed to target the active site of the protease and efficiently decrease Aβ generation in vivo and in vitro bind to both members of the PS NTF-CTF heterodimeric complex, which suggests that the active site lies between the two associated fragments(Esler et al., 2000; Li et al., 2000). Although the aspartyl protease inhibitor pepstatin A binds to both full-length PS and NTF-CTF complexes (Evin et al.,2001), more potent γ-secretase inhibitors bind only to associated NTF-CTF but not to full-length PS(Esler et al., 2000; Li et al., 2000; Seiffert et al., 2000). The same inhibitors bind to PS1ΔE9, a PS1 variant that lacks the processing site but is nevertheless functional (Li et al., 2000). These findings suggest that a conformational change in PS caused by endoproteolysis or deletion of this processing site produces the active protease that is susceptible to exogenous inhibitors.
Besides endoproteolysis of PS to form functional NTF-CTF, its association with other components is apparently also critical for γ-secretase activity. Investigation of the PS-containing high-molecular-weight (HMW)complex indicates that additional cofactors intimately associate with PS to form the active γ-secretase complex. Nicastrin, which was initially identified from PS1 co-immunoprecipitates(Yu et al., 2000), binds specifically to an immobilized transition state analogue γ-secretase inhibitor (Esler et al.,2002). Nicastrin and PS1 NTF-CTF are capable of binding to this immobilized γ-secretase inhibitor under conditions that also maintain the γ-secretase activity; conditions disrupting the association of these proteins with the inhibitor also render the complex inactive. In common with anti-PS1 antibodies, anti-nicastrin antibodies can precipitate the functionalγ-secretase complex (Esler et al.,2002). Although glycosylation of nicastrin is not absolutely required for γ-secretase activity(Herreman et al., 2003),mainly the mature form of nicastrin was identified in the HMWγ-secretase complex, and levels of nicastrin in cells closely correlate with PS levels (Arawaka et al.,2002; Kimberly et al.,2002; Leem et al.,2002; Tomita et al.,2002; Yang et al.,2002). Reduction of PS levels leads to a concomitant reduction in nicastrin levels, and downregulation of nicastrin expression decreases the levels of stabilized PS molecules. Thus, reduction in the amount of either protein decreases Aβ generation(Edbauer et al., 2002).
Presenilin is not only involved in the intramembranous processing of APP but also in the processing of and signaling from the Notch receptor(Kopan and Goate, 2002). Proper Notch signaling is critical to a wide variety of cell fate determinations during embryonic development and adulthood. Notch is a type I integral membrane protein that has a large extracellular domain, a single transmembrane domain and an intracellular domain. After translation, the receptor is proteolyzed in the trans-Golgi as part of its maturation into a heterodimeric cell surface receptor. Notch then suffers a second proteolysis as a result of ligand activation, leading to shedding of the extracellular domain of the receptor. The remaining membrane-bound C-terminal stub is subsequently cleaved within its transmembrane domain to release the Notch intracellular domain, which translocates to the nucleus where it regulates gene expression. This final intramembrane proteolysis is mediated by the multi-component γ-secretase complex(Fortini, 2002).
Genetic screening for proteins that cause Notch-like defects and interact with presenilin and nicastrin orthologues in C. elegans(SEL-12 and APH-2, respectively) identified two multipass transmembrane proteins: APH-1 (anterior pharynx defective) and PEN-2 (presenilin enhancer)(Francis et al., 2002; Goutte et al., 2002). C. elegans APH-1 and its human orthologues are predicted to contain seven membrane-spanning regions, and they are closely associated with SEL-12/PS and APH-2/nicastrin in C. elegans/humans. APH-1 apparently facilitates the localization of APH-2 to the cell surface of C. elegans embryos:most APH-2 remains in ER-like compartments close to the nucleus in APH-1 mutant embryos (Goutte et al.,2002). The phenotype of APH-1-mutant embryos is similar to that of embryos possessing defective SEL-12, indicating that APH-1 and SEL-12 (PS) may act together for the proper function of APH-2 (nicastrin)(Goutte et al., 2002).
Further genetic screening for Notch pathway components in C. elegans not only confirmed APH-1 as a regulator of γ-secretase but also revealed the additional component PEN-2(Francis et al., 2002). PEN-2 has two predicted TM domains. It does not contain any known protease motif and does not carry a signal peptide. Interestingly, its chromosomal location(chromosome 19) is close to the ApoE gene. The ApoEϵ4 allele is the major risk factor associated with AD. Inactivation of APH-1 or PEN-2 in cultured Drosophila cells significantly reduces the γ-secretase cleavage of APP and Notch, and PS NTF-CTF levels are also reduced (Francis et al., 2002). In mammalian cells, PEN-2 protein levels are significantly reduced in the absence of PS; downregulation of nicastrin synthesis similarly causes a reduction in PEN-2 levels. Downregulation of PEN-2 synthesis likewise decreases the levels of stabilized PS molecules and mature nicastrin, with a concomitant reduction in γ-secretase activity(Steiner et al., 2002). Reduction of APH-1 levels in mammalian cells also reduces γ-secretase activity (Lee et al., 2002b). Recent studies have demonstrated that APH-1, PEN-2 and nicastrin directly associate with PS (Gu et al.,2003; Luo et al.,2003; Kimberly et al.,2003) and overexpression of these four proteins enhancesγ-secretase activity (Edbauer et al., 2003; Kimberly et al.,2003; Takasugi et al.,2003). Increased levels of APH-1, PEN-2 and PS1 facilitate PS heterodimer formation, as well as glycosylation of nicastrin(Edbauer et al., 2003; Kimberly et al., 2003; Takasugi et al., 2003). Thus,nicastrin, APH-1 and PEN-2 regulate one another and are indispensable forγ-secretase activity. Although reconstitution of PS, nicastrin, APH-1 and PEN-2 in yeast leads to PS endoproteolysis and Aβ/AICD production(Edbauer et al., 2003),definitive evidence for PS as a protease requires in vitro reconstitution assays using purified PS and substrate.
From bacterial proteases to PS: a conserved active site motif
The catalytic apparatus of the classical aspartyl protease consists of two aspartic acid residues. The two aspartate residues coordinate a water molecule, activating it for cleavage of peptide bonds. Unlike serine or cysteine proteases, aspartyl proteases do not use nucleophilic attack as part of their mechanism. Thus, there is no covalent intermediate formed between the enzyme and the substrate (Beynon and Salvesen, 2001). Because PS lacks the classic D(T/S)G motif of an aspartyl protease, the conserved D257 and D385 residues may either be critical for the aspartyl protease activity of the HMW γ-secretase complex or themselves constitute the active site of a novel aspartyl protease that does not carry the classic D(T/S)G motif. Studies of bacterial aspartyl proteases have shown that the multi-pass type-4 prepilin peptidases (TFPP) likewise lack the D(T/S)G motif but have a conserved GD motif that is also conserved in presenilins and includes the second critical aspartate residue (D385 of PS1). D385A, G384P and G384K mutations in PS1 block proteolysis of the APP and Notch TM region. A FAD-causing mutation at this position in PS1 (G384A) instead causes a six-fold increase in Aβ42 generation compared with wild-type PS1-expressing cells, but it does not prevent proteolysis of the Notch TM domain (Steiner et al.,2000). The conserved GD motif lies in TM7 of PS1. This motif is likewise found on the C-terminal side of the active site of bacterial TFPP,and mutation of the aspartate residue of the GD motif blocks its ability to remove leader peptides from certain substrates(LaPointe and Taylor, 2000). Although the primary sequences of PS and TFPP are not highly similar overall,the similar requirement for the aspartate-containing motifs for proteolysis from bacteria to humans suggests that TFPP and PS are novel polytopic membrane aspartyl proteases (Steiner et al.,2000).
PS homologues with putative protease activity
Database searches for PS homologues led to the identification of another group of multipass transmembrane proteins that contain aspartyl-protease-like domains similar to those proposed for PS(Ponting et al., 2002). Except for the TM1 domain of PS, the sequences of these PS homologues, termed PSHs,loosely align with the PS sequence. These proteins thus appear to have topologies similar to that of PS. The YD motif in TM6 and the LGXGD motif in TM7, containing the two key aspartate residues, are conserved in almost all reported PSHs. In addition, all known PS proteins contain a proline-alanine-leucine-proline (PALP) motif starting at P433 (amino acid numbering based on human PS1) at the C-terminus, and mutation of the first proline of the PALP domain of PS leads to destabilization of the PS high-molecular-weight complex (Tomita et al., 2001). Interestingly, this PALP domain is also conserved in PSHs. Several animal and plant PSH's also contain N-terminal protease-associated (PA) domains that often co-occur with peptidase domains(Mahon and Bateman, 2000; Ponting et al., 2002).
Rogaev and colleagues independently found the same group of PS-like proteins, which they collectively name IMPAS (for intramembrane-protease-associated activity) or IMPs(Grigorenko et al., 2002). For example, IMP1 is identical to PSH3, which is encoded by a gene located on chromosome 20. Furthermore, the sequence of IMP1/PSH3 is identical to that of signal peptide peptidase (SPP), the protease involved in cleaving signal peptide remnants (Grigorenko et al.,2002; Ponting et al.,2002; Weihofen et al.,2002) (see below).
Signal peptide peptidase: PS-like aspartyl protease
SPP has a membrane topology like that of PS(Weihofen et al., 2002). After signal peptides are cleaved from pre-proteins by signal peptidase, the remnant signal peptides anchored in the ER membrane become substrates for SPP. Human SPP is predicted to have seven TM domains, its N-terminus facing the ER and its C-terminus facing the cytoplasm (Fig. 1). The active site motif YD is located in the center of TM 4,whereas the other active site motif, LGXGD, is located in the center of TM 5. When the cDNA encoding human SPP is expressed in yeast, SPP activity in microsomal vesicles can be solubilized with CHAPS(Weihofen et al., 2002), a detergent capable of solubilizing γ-secretase activity(Esler et al., 2002). In vitro, SPP mediated cleavage of its substrate has clearly demonstrated the critical role of the conserved aspartates in SPP: mutation at D265 in the LGXGD motif prevents cleavage of the substrate. Addition of a specific inhibitor of SPP, TBL4K, which reversibly binds to the active site of the protease, completely blocks in vitro SPP proteolytic activity. Indeed,Weihofen et al., have used a photocrosslinkable version of this inhibitor to purify and identify SPP (Weihofen et al.,2002). Therefore, SPP represents a novel class of PS-like aspartyl protease.
Substrates: type I and II membrane proteins
PS is predicted to have eight TM domains, the N-terminus, TM6-TM7 loop and C-terminus all oriented towards the cytoplasm(Fig. 1)(Doan et al., 1996; Li and Greenwald, 1998). Thus,the orientation of TM6, which contains the YD motif, is lumen to cytoplasm,and the orientation of TM7, which contains the LGXGD motif, is cytoplasm to lumen. This contrasts with the orientations of TM regions containing the YD and LGXGD motifs in SPP but is consistent with the orientations of their respective substrates (Fig. 1):SPP substrates (signal peptides), are type II membrane proteins whose N-termini face the cytoplasm, whereas γ-secretase substrates are type I membrane proteins whose C-termini facing the cytoplasm. All of the knownγ-secretase substrates (APP, Notch, ErbB-4, E-cadherin, CD44, LRP and nectin1α) are type I membrane proteins and are cleaved in their TM domains in a PS-dependent manner (Fig. 2). Some of the cleavage sites have been fully characterized,others remain to be determined. Both products of γ-secretase-mediated cleavage have been characterized in the case of APP and Notch. These substrates are cut in the middle of their TM domains and at a residue close to the interface of the membrane and cytoplasm, and both cleavages can be inhibited by γ-secretase inhibitors and are fully dependent on PS(De Strooper et al., 1999; Okochi et al., 2002). In the cases of other γ-secretase substrates, only one product at best has been fully characterized. ErbB-4 undergoes γ-secretase cleavage between residue A672 and the conserved residue V673, several residues from the membrane-cytoplasm boundary (Lee et al.,2002a; Ni et al.,2001). Cleavage of E-cadherin occurs between residues L731 and R732, apparently right at the membrane-cytoplasm interface(Marambaud et al., 2002). CD44 is cleaved in a fashion similar to APP: a cleavage occurs in the middle of the TM domain (between residues A278 and L279), releasing an Aβ-like molecule(CD44-β; and another cleavage releases the CD44 intracellular domain(CD44-ICD) (Lammich et al.,2002). Ectodomain shedding of full-length low-density lipoprotein receptor-related protein (LRP) (May et al., 2002) or nectin 1α(Kim et al., 2002) in each case leads to the formation of a membrane-bound C-terminal fragment and is followed by PS-dependent γ-secretase-like cleavage in the TM domain to release the intracellular domain. Although both LRP and nectin 1α have a conserved valine residue close to the membrane-cytoplasm boundary, the actual cleavage sites of LRP and nectin 1α have yet to be determined. None of these known PS/γ-secretase substrates appears to be cleaved by PSHs,since γ-secretase cleavage of these substrates does not occur in PS1-/- PS2-/- cells.
Studies of SPP substrates (signal peptides) indicate that the valine residue close to the interface between membrane and cytoplasm is not required for cleavage (Lemberg and Martoglio,2002). Although SPP can cleave signal peptides from the hormone prolactin, human polymorphic MHC class I molecules, calreticulin, and the viral proteins vesicular stomatitis virus G protein (VSVG), it fails to cleave signal peptides from human cytomegalovirus glycoprotein UL40 and RNase A. Sequence comparison of these substrates suggests that residues having a tendency to disrupt the helix of the TM domain are required for SPP cleavage. Importantly, signal peptidase cleavage of pre-proteins to form a signal peptide is a prerequisite for SPP cleavage, which is consistent with the requirement for ectodomain shedding in γ-secretase substrates. Both type I and II membrane proteins thus undergo ectodomain shedding as a prerequisite for γ-secretase/PS- or SPP-mediated intramembrane proteolysis.
Although the majority of experimental results support the idea that PS is an aspartyl protease, definitive evidence from in vitro reconstitution assays using purified PS and substrate is needed. This approach is complicated owing to the existence of at least three necessary co-factors for protease activity:nicastrin, APH-1 and PEN-2. Although the functions of these co-factors are under extensive investigation, database searches for protease motifs have not revealed any candidates for protease domains in these co-factors. By contrast,relatives of PS have emerged as potential proteases carrying the unique putative active site motifs in two TM domains, and these proteins appear to function without the need for other co-factors. The TFPPs, for example, are known proteases carrying the conserved GD motif of PS, but the primary sequences of TFPPs share no other similarity with PS. The multipass PSH(IMPAS) TM proteins share some sequence similarity with PS, possessing both YD and LGXGD motifs of the postulated active site. SPP (PSH3/IMP1) has a PS-like membrane topology and has protease activity. Although ectodomain shedding by substrates of PS/γ-secretase/SPP is a necessary prerequisite for subsequent proteolysis, the detailed molecular events involved in proteolysis of TM domains are not clear. As more γ-secretase substrates are discovered, related molecules will be examined for possible cleavage by PS-like proteases. Searches for YD and LGXGD motifs in genome sequences may also reveal additional PS-like proteases similar to SPP. Further biochemical characterization of these proteases will help us to elucidate the molecular mechanism of intramembrane proteolysis.
Some studies discussed in this review were supported in part by the Alzheimer's Association, NIH NS 41355, NIH AG 17574 (to M.S.W.), and NIH AG 17593 (to W.X.).