Presenilin and γ‐secretase: structure meets function

During the search for the major genes responsible for autosomal-dominant, familial forms of Alzheimer's disease (AD), many thought the encoded proteins would reveal at least one, if not both, of the proteases that cut the amyloid-β peptide (Aβ) out of the amyloid-β precursor protein (APP) (Fig. 1). Amyloid-β peptide is the major protein component of the characteristic cerebral plaques of AD (Glenner and Wong 1984), and APP itself had already been identified in 1991 as a locus for familial AD (FAD), with disease-causing missense mutations found immediately near the Aβ boundaries (Chartier-Harlin et al. 1991; Goate et al. 1991; Murrell et al. 1991). However, when the search identified presenilin-1 (PS1) and presenilin-2 (PS2) in 1995 (Levy-Lahad et al. 1995; Sherrington et al. 1995), it was far from clear given the primary structure what the normal function of these proteins might be and why mutant forms cause disease. Ironically, over the past few years the argument has come full circle, and it now appears that presenilins are indeed responsible for the final proteolysis that produces Aβ. Topology and proteolytic processing of APP and Notch. APP is cleaved by either α-secretase to form C83 or β-secretase to form C99. In either event, the APP ectodomain is shed (α- and β-APPs). C83 and C99 are further processed by γ-secretase within the transmembrane domain to produce p3 and Aβ, respectively. Stars in APP represent sites of missense mutation near β- and γ-secretase cleavage sites that cause familial Alzheimer's disease. Notch is processed in the Golgi by a furin-like convertase and then transported to the cell surface as a heterodimer. Upon ligand binding, the ectodomain is shed by TACE, and the resulting membrane-associated fragment (NEXT) is cleaved within its transmembrane domain by a γ-secretase-like protease, releasing the Notch intracellular domain (NICD). APP is processed by several different proteases called secretases (Selkoe 1994). Cleavage by β-secretase, a recently identified membrane-bound aspartyl protease (De Strooper and Konig 1999), takes place just outside the membrane, releasing soluble APP (β-APPs) and leaving behind a 99-residue membrane-associated C-terminal fragment, C99 (Fig. 1). Alternatively, APP is cut by α-secretase, a membrane-bound metalloprotease (Buxbaum et al. 1998; Lammich et al. 1999), to produce α-APPs and an 83-residue membrane-associated C-terminal fragment, C83. Both C99 and C83 are substrates for γ-secretase, a mysterious enzyme that carries out an unusual proteolysis in the middle of the transmembrane domain of APP (Selkoe and Wolfe 2000), resulting in the formation of the 4-kDa Aβ from C99 and p3, a 3-kDa N-terminally truncated form of Aβ, from C83. Proteolysis by γ-secretase results in considerable C-terminal heterogeneity in Aβ, with 40- and 42-residue forms (Aβ40 and Aβ42) being the major variants. Remarkably, over 70 FAD-causing presenilin mutations have now been identified (see http://www.alzforum.org/members/resources/pres_mutations/index.html), and all are missense mutations with the exception of one that results in alternate splicing with deletion of exon 9 (PS1 ΔE9). These mutations are found in many regions of the linear sequence, although they tend to cluster in transmembrane domains (Fig. 2). A major clue to the role of presenilins in the pathogenesis of FAD was the discovery that disease-causing mutations increase production of the highly fibrillogenic Aβ42 over Aβ40 in transfected cells, transgenic mice and plasma of mutant carriers (Citron et al. 1992; Cai et al. 1993; Citron et al. 1994; Haass et al. 1994; Suzuki et al. 1994; Citron et al. 1997; Johnson-Wood et al. 1997). Thus, these presenilin mutations alter the specificity of γ-secretase processing of APP. Another milestone was the finding that knockout of PS1 in mice caused a dramatic reduction in γ-secretase processing of APP without apparent affect on APP maturation or subcellular distribution (De Strooper et al. 1998), suggesting that presenilins mediate most, if not all, proteolysis by γ-secretase. Topology and proteolytic processing of presenilins. Presenilins are cleaved within the hydrophobic region of the large cytosolic loop between TM6 and TM7, resulting in the formation of a heterodimeric complex composed of the N-terminal fragment (NTF) and the C-terminal fragment (CTF). Stars represent sites of missense mutation that cause familial Alzheimer's disease. The two conserved aspartates required for presenilin endoproteolysis and γ-secretase processing of APP and Notch are predicted to be within TM6 and TM7. The region of the protein encoded by exon 9 is denoted; natural deletion of this exon leads to a noncleavable but functional presenilin. Meanwhile, pharmacological characterization of γ-secretase pointed to an aspartyl protease mechanism: the activity was blocked by substrate-based peptide analogues that mimic an intermediate in aspartyl protease catalysis (Wolfe et al. 1999a). This finding led to the specific question of whether presenilins might be novel aspartyl proteases. These proteins bear essentially no sequence resemblance to aspartyl proteases. Provocatively, though, presenilins contain two aspartates that are completely conserved from worms to man (Li and Greenwald 1997) [and even in a recently identified plant presenilin (Lin et al. 1999)], and all aspartyl proteases possess two active site aspartates responsible for catalysis. The two PS aspartates are predicted from hydropathic analysis and topology studies to align in the membrane with each other and with the γ-secretase cleavage site in APP [most topology studies support an eight-transmembrane-domain (TM) topology for the presenilins (Doan et al. 1996; Hardy 1997; Li and Greenwald 1996, 1998), although six- and seven-TM topologies have also been suggested (Dewj and Singer 1997; Lehmann et al. 1997)]. Indeed, each of these two aspartates in PS was found to be critical for γ-secretase processing of APP: overexpression of Asp-mutant PS1 or PS2 resulted in a dominant negative effect, lowering Aβ production and raising levels of γ-secretase substrates C83 and C99 (i.e. effects similar to knocking out PS1) (Leimer et al. 1999; Steiner et al. 1999a; Steiner et al. 1999c; Wolfe et al. 1999b; Kimberly et al. 2000; Yu et al. 2000). Overexpression of wild-type presenilin, however, does not lead to increased γ-secretase activity (Citron et al. 1997). How then could it be the protease? Presenilins themselves undergo proteolytic processing in the hydrophobic loop between transmembrane regions 6 and 7 to form stable heterodimeric complexes (Fig. 2) (Thinakaran et al. 1996; Capell et al. 1998). However, only limiting levels of presenilin heterodimers are produced even upon overexpression, suggesting that stoichiometric cellular factors gate their formation (Thinakaran et al. 1996; Podlisny et al. 1997; Ratovitski et al. 1997; Thinakaran et al. 1997; Steiner et al. 1998). Moreover, full-length presenilin is rapidly degraded, whereas the heterodimer is quite stable. The tight regulation of heterodimer formation and the stabilization of this species together suggest that the heterodimer is the bio-active form of presenilins. Thus, heterodimeric presenilin could be the catalytic component of γ-secretase, which would explain the observation that presenilin overexpression does not augment the activity of this protease. Interestingly, the two conserved aspartates are required not only for γ-secretase activity but also for conversion of full-length presenilins to heterodimers. These roles, however, can be separated. The site of presenilin endoproteolysis is absent in PS1 ΔE9, and this presenilin variant is functional and partially stabilized as a holoprotein (Levitan et al. 1996; Ratovitski et al. 1997; Zhang et al. 1998; Steiner et al. 1999b). Asp mutation of PS1 ΔE9 still results in a dominant negative effect on γ-secretase processing of APP (Wolfe et al. 1999b). Thus, the presenilin aspartates are critical for cleavage of APP by γ-secretase, even in a PS variant that does not require heterodimer formation for activity. The hydrophobic portion of the cytosolic loop encoded by exon 9 may be a pro-domain that inhibits protease activity. Cleavage (wild-type PS), deletion (PS1 ΔE9) or mutation (Steiner et al. 1999c) of this domain allows activity. Furthermore, the requirement of the two aspartates for presenilin heterodimer formation suggests that, after interacting with limiting factors, presenilins may be activated by autoproteolysis. More study is needed, however, to determine whether the exon 9 region truly encodes a pro-domain and whether PS in fact undergoes auto-activation. The Notch receptor, involved in cell-fate determinations during embryonic development, is proteolytically processed in a manner strikingly similar to APP (Fig. 1). After translation in the ER, Notch is processed by a furin-like protease, resulting in a heterodimeric receptor that is shuttled to the cell surface (Logeat et al. 1998). Upon interaction with a cognate ligand, the ectodomain of Notch is shed by a metalloprotease apparently identical to tumor-necrosis-factor-α converting enzyme (TACE; Brou et al. 2000; Mumm et al. 2000). Interestingly, metalloproteases such as TACE and a disintegrin and metalloprotease (ADAM-10) are among the identified α-secretases that shed the APP ectodomain (Buxbaum et al. 1998; Lammich et al. 1999). The membrane-associated C-terminus is then cut within the postulated transmembrane domain to release the Notch intracellular domain (NICD), which then translocates to the nucleus where it interacts with and activates the CSL family of transcription factors (Schroeter et al. 1998). NICD formation is absolutely required for signaling from the Notch receptor: knock-in of a single point mutation near the transmembrane cleavage site in Notch1 results in an embryonic lethal phenotype in mice virtually identical to that observed upon knockout of the entire Notch1 protein (Huppert et al. 2000). Presenilins are absolutely required for the transmembrane cleavage of both APP and Notch (Herreman et al. 2000; Zhang et al. 2000), and γ-secretase inhibitors designed from the APP cleavage site also block the transmembrane cleavage of Notch (De Strooper et al. 1999). Moreover, the two conserved aspartates are also required for Notch TM proteolysis (Ray et al. 1999; Steiner et al. 1999a; Berezovska et al. 2000; Capell et al. 2000). Thus, both APP and Notch are cleaved within their TM domains by a γ-secretase-like activity that required presenilins. At minimum, presenilin heterodimers are essential components of a larger γ-secretase complex. These heterodimers copurify with γ-secretase activity after solubilization of isolated microsomes and size-exclusion chromatography (Li et al. 2000a). Moreover, immunoprecipitation of presenilin heterodimers with PS1-specific antibodies brings down γ-secretase activity (Li et al. 2000a). These findings argue strongly against a direct trafficking, chaperoning or folding role of presenilins in facilitating γ-secretase activity, although presenilins may have some indirect effect on these functions (Naruse et al. 1998). The strongest evidence in favor of heterodimeric presenilins comprising the active site of γ-secretase comes from recent affinity labeling experiments using transition-state analogue inhibitors. Li et al. identified a peptidomimetic that blocks γ-secretase activity with an IC50 of 0.3 nm in a solubilized protease assay (Li et al. 2000a) and contains a hydroxyethyl isostere, a transition-state mimicking moiety found in many aspartyl protease inhibitors. While the transition-state mimicking hydroxyethyl group directs the compound to aspartyl proteases, flanking substructures determine specificity. Photoactivatable versions of this compound bind covalently to presenilin subunits exclusively (Li et al. 2000b). Interestingly, installation of a photoreactive group on one end of the inhibitor led to labeling of the N-terminal presenilin subunit, while installation on the other end resulted in the tagging of the C-terminal subunit. Moreover, while these agents did not label wild-type PS1 holoprotein, they did tag PS1 ΔE9, which as described above is not processed to heterodimers but nevertheless active. Similarly, Esler et al. identified peptidomimetic inhibitors containing a difluoro alcohol group, another type of transition-state mimicking moiety for aspartyl proteases, and these compounds were developed starting from a substrate-based inhibitor designed from the γ-secretase cleavage site in APP (Esler et al. 2000). Conversion of one such analogue to a reactive bromoacetamide provided an affinity reagent that likewise bound covalently and specifically to PS1 subunits in cell lysates, isolated microsomes and whole cells. Either PS1 subunit so labeled could be brought down with antibodies to the other subunit under co-immunoprecipitation conditions, demonstrating that the inhibitor had bound to heterodimeric PS1. Because the reactive groups of this compound and that reported by Li et al. are located < 12 Å from the transition-state mimicking alcohol group, the tagged protein should be located no further from the active site aspartates and is highly likely to be the protease itself. Taken together, these results strongly suggest that heterodimeric presenilins contain the catalytic component of γ-secretase: the inhibitors are classical transition-state analogues targeted to the active site of the protease. Interestingly, the membrane-tethered aspartyl protease β-secretase was also identified using an inhibitor directed to the active site (Sinha et al. 1999). The active site of γ-secretase is apparently at the PS heterodimeric interface: both subunits are labeled by γ-secretase affinity reagents, and each contributes one critical aspartate. The fact that presenilins are not homologous to cytosolic aspartyl proteases is not a conceptual hurdle: the site-2 protease involved in cholesterol metabolism appears to be a novel polytopic membrane metalloprotease (Rawson et al. 1997; Zelenski et al. 1999), and the bacterial type-4 prepilin peptidases (TFPPs) are apparently unusual aspartyl proteases with eight TM domains (LaPointe and Taylor 2000). Polytopic membrane proteases are likely to have evolved from membrane proteins that acquired residues necessary for catalysis and not from their mechanistic cousins found in the cytosol. Gathering definitive proof that presenilins carry out proteolysis will be difficult. Indeed, proteolytic activity has not been formally demonstrated with purified forms of the site-2 protease or TFPPs. Even with identification of other members of the γ-secretase complex and careful reconstitution, determining which component(s) comprise the active site is likely to be challenging. Crystallographic studies of the reconstituted complex showing water or a transition-state analogue coordinated between the two conserved PS aspartates would be indisputable evidence; however, given the problems of crystallizing polytopic membrane proteins, such evidence is not likely to be forthcoming. Nevertheless, the labeling of presenilin heterodimers with aspartyl protease transition-state analogue inhibitors of γ-secretase, along with the identification of the two conserved aspartates as critical for protease activity, provides compelling evidence that presenilins are the business end of γ-secretase. The affinity labeling studies simultaneously identify the protease responsible for the transmembrane cleavage of Notch, because substrate-based γ-secretase inhibitors like those used by Esler et al. also block this Notch proteolysis (De Strooper et al. 1999). These findings considerably reinforce the amyloid hypothesis of AD pathogenesis: all FAD-causing mutations identified to date (accounting for ∼ 60% of all FAD cases) are either in the precursor protein leading to Aβ (APP) or are in the protease that catalyzes the final step in Aβ generation (γ-secretase/presenilin). For some time, γ-secretase has been an activity without a protein, and presenilins have been proteins without a known function. It is thus ironic that structure meets function: presenilin appears to be the catalytic component of γ-secretase.