The secretases: enzymes with therapeutic potential in Alzheimer disease

Biological activity

Outside the CNS, amyloid precursor protein (APP) is preferentially cleaved by α-secretase (Figure 1a).^5–7 The α-secretase activity is mediated by a series of membrane-bound proteases (Figure 2a), which are members of the ADAM (a disintegrin and metalloprotease) family. The α-secretases cleave APP within the Aβ sequence itself, between amino acids 16 and 17 (Figure 1c), generating a soluble APPsα ectodomain and a membrane-bound carboxy-terminal fragment, APP-CTFα (Figure 1a). The latter fragment is degraded in lysosomes,^8,9 or is further processed by γ-secretase,¹⁰ yielding a series of short hydrophobic peptides including Aβ_17–40 and Aβ_17–42, which are collectively called p3 fragments (Figure 1a,c).¹¹ Processing of APP by α-secretases is postulated to be protective in the context of AD because the enzymes cleave within the Aβ sequence, thereby preventing the production of Aβ. Such a postulate is contingent, however, on multiple unknown aspects of APP and Aβ biology. For example, increasing α-secretase activity would be neuroprotective via the increased shedding of growth-promoting soluble APPα ectodomains only if the p3 fragment produced by α-secretase activity is not pathogenic, and if increasing α-secretase activity actually lowers Aβ production. Several studies have indicated that increased α-secretase-mediated processing of APP reduces the processing of APP by β-secretase and decreases Aβ production;^12,13 however, this finding has not been universally replicated.¹⁴ Furthermore, the p3 peptide is very hydrophobic and has been reported to be present in diffuse amyloid plaques.^15,16 Increasing α-secretase activity, as mentioned above, increases the production of APPsα, and has been reported to be neuroprotective and growth promoting,¹⁷ but the consequences of chronically upregulating α-secretase-mediated cleavage of other substrates remains unknown.

From a drug discovery point of view, α-secretase is a peculiar target to choose for the development of AD drug therapies, because it represents an activity that is ascribed to multiple proteases. Furthermore, the exact identity and the relative contribution of each α-secretase to the cleavage of APP in the brain has not been fully elucidated. Transgenic approaches that overexpress the metalloprotease ADAM10 in mice result in increased APPsα generation and are protective against amyloidosis in the human APP_V717I transgenic mouse model.^18,19 Unfortunately, the extent to which ADAM10 contributes to endogenous α-secretase activity in the human brain is unclear. Furthermore, other proteases of the same family (ADAM9, ADAM17 and ADAM19),^18,20–23 as well as β-site APP cleaving enzyme 2 (BACE2),^24,25 which shares considerable amino acid sequence homology with β-secretase, have also been shown to cleave APP at or close to the α-cleavage site in APP. Gene deletion studies have also been performed to ascertain the involvement of ADAM9, ADAM10 and ADAM17 in APP processing, but no clear understanding of how these proteins affect APP processing in the brain has been established.^26–28 In fact a study has shown that ADAM9 is actually not involved in the non-stimulated secretion of APPsα in neurons.²⁶ Further studies are clearly needed to establish the involvement of these proteases in APP processing. Unfortunately, progress in this area has been hampered by a lack of selective α-secretase inhibitors and the lethality often observed on ablation of the genes that code for these proteases.^27,29,30 Deletion of the Adam10 gene, for example, produces a lethal phenotype in mid-gestation. This phenotype is probably caused by vascular defects,²⁷ which in turn are probably a consequence of deficient N-cadherin, E-cadherin and Notch processing. Development of selective inhibitors of the ADAM proteases and conditional transgenic animal models should help to clarify which of the enzymes with α-secretase activity are most important for APP processing.

Drug development

Current evidence suggests that the identified α-secretases demonstrate a high degree of redundancy, and which α-secretases are responsible for APP cleavage in neurons and other brain cells is unclear.^27,31 Until this ambiguity is overcome, optimizing the development of compounds that directly activate the α-secretases will be difficult. Stimulating one or more of the signal transduction pathways involved in the regulation of α-secretase activity might be an alternative and indirect method of promoting α-secretase-mediated cleavage of APP. Protein kinase C, mitogen-activated protein kinases, tyrosine kinases and calcium-mediated pathways are all known to be involved in regulating α-secretase activity, and developing compounds that stimulate α-secretase via these pathways is clearly possible.³² Retinoic acid derivatives have been proposed to increase transcription of ADAM10 and could, therefore, also be used to indirectly stimulate α-secretase-mediated cleavage of APP.³³

Development of a direct activator of α-secretase as a drug treatment for AD seems unlikely, at least in the short term. Drug treatments that indirectly act as α-secretase stimulators, however, are already being tested in clinical trials. In fact, evidence that indirect activation of α-secretase is successfully achieved by α-7-nicotinic acetylcholine receptor agonists, a 5-hydroxytryptamine receptor 4 agonist, and a γ-aminobutyric acid A receptor modulator has been used to support the clinical development of these agents (Table 1). In many cases, a paucity of data is available to show that these compounds increase α-secretase-mediated cleavage of APP and reduce Aβ levels in vivo. Despite such concerns, the strategy itself is attractive, at least from a theoretical point of view, in that the drug could be approved solely on the basis of symptomatic benefit and later be more extensively studied with respect to its potential for disease modification.^34–36 We note, however, that attenuation of AD symptoms by these drugs has not been demonstrated in patients.

Biological activity

The β-secretase enzyme might be a prime therapeutic target for AD, because inhibition of β-secretase should, in theory, decrease production of all forms of Aβ, including the pathogenic 42 amino acid form of Aβ (Aβ₄₂) (Figure 1b,c). In 1999, five different research groups independently reported the molecular cloning of β-secretase and, consequently, this enzyme has been given various names, including β-site APP cleaving enzyme (BACE), aspartyl protease 2 and membrane-associated aspartic protease 2.^37–41 The groups used differing isolation methods (for example, expression cloning, protein purification or genomics), yet all identified the same enzyme and concurred that it possessed all the known characteristics of β-secretase.⁴² For the purposes of this Review, this β-secretase will henceforth be referred to as BACE1.

BACE1 is a type 1 transmembrane aspartic protease (Figure 2b) that is related to the pepsin and retroviral aspartic protease families. BACE1 activity has a low optimum pH and the enzyme is predominantly localized in acidic intracellular compartments (for example, endosomes and the trans-Golgi network). Soon after BACE1 was discovered, a homologous protein, BACE2, was identified. BACE1 and BACE2 share 64% amino acid sequence homology, raising the possibility that BACE2 is also a β-secretase. Unlike BACE1, however, which is highly expressed in neurons of the brain, BACE2 is expressed at low levels in the brain and does not have the same cleavage specificity for APP that BACE1 does, thus indicating that it is a poor β-secretase candidate.

To unequivocally prove that BACE1 was identified correctly as the endogenous β-secretase, knock-out mice, in which the gene encoding BACE1 was deleted, were generated by several groups.^43–46 Initial reports indicated that Bace1^−/− mice were viable and fertile, suggesting that therapeutic inhibition of BACE1 might produce few mechanism-based adverse effects. Recent studies, however, have shown that Bace1^−/− mice exhibit some abnormal phenotypes that are related to the physiological functions of BACE1 (see below). This finding raises concerns that inhibition of BACE1 could produce serious adverse effects in humans. Nevertheless, Aβ generation, amyloid pathology, electrophysiological dysfunction, and cognitive deficits are all reported to be abrogated when Bace1^−/− mice are bred to APP transgenics (mice that overexpress the mutant form of APP) to generate Bace1^−/−;APP bigenic mice.^43,47,48 Bace1^−/− mice are devoid of cerebral Aβ, indicating that BACE1 is the main, if not the only, β-secretase enzyme in the brain.^49,50 This idea is further supported by reports that lentiviral delivery of Bace1 small interfering RNA attenuates both Aβ amyloidosis and cognitive deficits in APP transgenic mice.^48,51 In addition, the rescue of memory deficits in Bace1^−/−;APP bigenic mice suggests that therapeutic BACE1 inhibition should improve Aβ-dependent cognitive impairment in patients with AD. Taken together, the BACE1 characterization and validation studies have unequivocally demonstrated that BACE1 is the endogenous β-secretase in the brain and that it is a promising therapeutic target for lowering cerebral Aβ levels in patients with AD.

BACE1 is involved in the processing of numerous other proteins in addition to APP. Identifying these proteins will be necessary for evaluating the possibility of potential mechanism-based toxicity arising from inhibition of BACE1. The inability of soluble BACE1 to efficiently process full-length APP suggests that all BACE1 substrates will probably be membrane-bound proteins.⁵² In fact, all reported BACE1 substrates, such as Golgi-localized membrane-bound α2,6-sialyltransferase,⁵³ P-selectin glycoprotein ligand 1,⁵⁴ amyloid-like protein 1 and amyloid-like protein 2,^55–57 LDL receptor–related protein (LRP),⁵⁸ the voltage-gated sodium channel (Na_v1) β2 subunit (Na_vβ₂), ^59,60 and neuregulin-1 (NRG1) and neuregulin-3 (NRG3),⁶¹ are transmembrane proteins. Abrogating the cleavage of NRG1 in BACE1^−/− mice reduces myelin sheath thickness of axons of both peripheral sciatic nerves and central optic nerves.^62,63 Deficient NRG1 processing in BACE1^−/− mice also impairs remyelination of injured sciatic nerves.⁶¹ These deleterious effects of attenuated processing of NRG1 by BACE1, coupled with the diverse characteristics of the other BACE1 substrates, suggests that the potential for drug treatments designed to inhibit BACE1 activity to display substantial mechanism-based toxicity is high. Concerns over treatment-linked mechanism-based toxicities has prompted research into whether partial inhibition of BACE1 activity could provide therapeutic benefits for patients with AD while limiting treatment-associated adverse effects. Decreased Aβ deposition was observed in the brains of 12 month-old APPswe;PS1DE9;Bace1^+/− mice, in comparison to APPswe;PS1DE9;Bace1^+/+ mice (Box 1); however, no substantial differences were observed between these two strains at 20 months.⁴⁸ Why the older mice in this study did not have reduced amyloidosis is unclear. In a similar study, however, a substantial reduction in Aβ burden was observed in the brains of 13 month-old and 18 month-old PDAPP;Bace1^+/− mice (Box 1).⁴⁹ Although the results of the two transgenic studies differ somewhat, they both suggest that partial inhibition of BACE1 might be sufficient to reduce Aβ burden.

Drug development

Since the discovery of BACE1, numerous attempts have been made to develop small-molecule BACE1 inhibitors. First-generation BACE1 inhibitors were peptide-based mimetics of the APP β-cleavage site, which had the APP β-site scissile amide bond replaced with a non-hydrolyzable transition state analog, such as statine.⁴⁰ More recently, nonpeptidergic compounds with high affinities for BACE1 have been generated.^64,65

Although initial drug development efforts with peptidomimetic BACE1 inhibitors were encouraging, BACE1 has since proved to be a challenging medicinal chemistry target, for several reasons. The fact that BACE1 has a large substrate binding site is the most obvious one, as this makes inhibiting the enzyme with small nonpeptidergic compounds extremely difficult. Ideally, BACE1 inhibitors should be <500 kDa, orally bioavailable, metabolically stable, intrinsically potent, and highly selective for BACE1 over BACE2 and other aspartic proteases. Potential BACE1 inhibitors must also, in principle, be able to penetrate both the plasma and endosomal membranes to gain access to the intracellular compartments where endogenous BACE1 is localized. An alternative method of delivering BACE1 inhibitors to the intracellular compartments could be achieved by designing BACE1 inhibitors that target and bind to the small fraction of BACE1 expressed at the cell surface. By binding to BACE1 on the cell surface, the inhibitor could be delivered together with BACE1 to the endosomes via endocytosis.⁶⁶ Efficacious BACE1 drugs would also need to efficiently cross the blood–brain barrier.

Despite these challenges, potent nonpeptidergic small-molecule BACE1 inhibitors have been reported that meet the necessary criteria and show efficacy in lowering cerebral Aβ levels in animal models of AD.⁶⁵ Moreover, the pharmaceutical company CoMentis recently announced the completion of the first phase I clinical trial of a candidate BACE1 inhibitor drug, and other candidate BACE1 inhibitors might soon be entering clinical trials (Table 1). Reports of successful development of potential nonpeptidergic small-molecule BACE1 inhibitors are encouraging, and suggest that therapeutic approaches involving BACE inhibition for the treatment or prevention of AD might be a reality in the future. Given that BACE1 seems to be involved in other important physiological roles, careful titration of BACE1 drug dosage might be necessary to minimize mechanism-based adverse effects.

Biological activity

Despite long-standing knowledge that Aβ was generated through the processing of APP by secretases, the identity of the γ-secretase remained unresolved for many years. Through a convergence of genetic, pharmacological, protein and cell biology studies, it is now clear that γ-secretase is a multi-subunit aspartyl protease that cleaves APP and many other type 1 transmembrane proteins within their transmembrane domains.^67,68 Presenilin 1 and presenilin 2 (PS1 and PS2) form the catalytic core of γ-secretase, and three accessory proteins, anterior pharynx-defective 1 (APH-1), nicastrin, and presenilin enhancer protein 2 (PEN2) are also required to complete the γ-secretase complex (Figure 2c).^67,69,70 The three accessory proteins seem to be involved in the maturation and stability of the complex. Nicastrin has also been reported to be a substrate receptor;⁷¹ however, the data supporting this function is inconclusive and other studies suggest that, like APH-1 and PEN2, nicastrin is required for complex maturation.⁷²

The γ-secretase complex seems to display a high degree of heterogeneity. Humans have two presenilin genes and two APH1 isoforms—APH-1A, which can be alternatively spliced to long and short forms, and APH-1B—and six different functional γ-secretase complexes have been identified.^73,74 Rodents have an extra Aph1C gene, which probably arose through duplication of Aph1B.^75,76 Rodent γ-secretases might, therefore, display a higher degree of heterogeneity than human γ-secretase. Little is known about the physiological roles of the diverse γ-secretase complexes and how they might process different substrates. One report has shown that in cell free assays, γ-secretase complexes containing APH-1A produce different ratios of intracellular Aβ species from γ-secretase complexes containing APH-1B. In cell culture however, γ-secretase complexes with diverse APH-1 isoforms produce similar ratios of secreted Aβ.⁷⁷ Deletion of the gene encoding APH-1A is embryonically lethal, almost certainly owing to inhibition of Notch signaling. By contrast, transgenic mice with the genes encoding APH-1B and APH-1C deleted develop normally and have no gross phenotype.

The subunit heterogeneity of the γ-secretase complex suggests that selective targeting of one particular subunit might be an effective alternative treatment strategy to nonselective γ-secretase inhibition. Elective removal of APH-1B and APH-1C in a mouse model of AD has already been shown to decrease Aβ plaque formation and improve behavioral deficits, without adverse effects attributable to impaired Notch signaling being reported.⁷⁷ This study provides proof of concept that specific inhibition of γ-secretase complexes containing APH-1B in the brain could be a useful strategy for lowering Aβ without inducing Notch-related adverse effects.

γ-Secretase is not only heterogeneous with respect to its protein composition and expression, but it also exhibits a heterogeneous protease activity by cleaving numerous and diverse transmembrane proteins. Two critical asparate residues that are present in opposing trans-membrane domains are believed to catalyze the hydrolysis of the peptide bond in a manner analogous to traditional aspartyl proteases;⁷⁸ however, this mechanism of catalysis has not been formally proven. Electron microscopy and biochemical studies suggest that γ-secretase forms a pore-like structure in the membrane that allows water to access the active site, although the electron microscopy studies lack the resolution to provide clear insight into the catalytic mechanism.^79–82 Furthermore, detailed structural data from nuclear magnetic resonance spectroscopy or X-ray crystallography seem unlikely to emerge in the short term given the complexity of analyzing a molecular complex with 19 transmembrane domains.

Particularly intriguing from a therapeutic perspective is that γ-secretase preferentially cleaves the membrane ‘stub’ of the substrate following ectodomain ‘shedding’, and seems to cleave the substrate with an initial cleavage within the transmembrane domain close to the cytoplasmic border of the membrane, a cleavage often referred to as the ε-cleavage (Figure 1). The ε-cleavage releases the intracellular domain from the membrane, enabling this domain to carry out an intracellular signaling function. For example, following γ-secretase processing, the intracellular domain of Notch translocates to the nucleus, where it binds to various intracellular factors and thereby regulates transcription.⁸³ After the initial cleavage at the ε-cleavage site, γ-secretase cleaves the substrate again (in fact, several times) within the transmembrane domain, a process that ultimately results in the secretion of the amino-terminal portion of the substrate’s membrane stub.⁸⁴ This sequential cleavage of APP by γ-secretase produces Aβ and seems to be a key determining feature that can increase an individual’s risk of developing AD. In fact, all AD-associated mutations in the presenilin genes and many other mutations in APP increase the proportion of Aβ₄₂ that is produced.⁸⁵ Aβ₄₂ aggregates more readily than the shorter Aβ peptides and seems to be required for Aβ aggregation and accumulation in vivo.^86,87 Moreover, several studies support the concept that shorter Aβ peptides, including the predominant Aβ₄₀ species, inhibit Aβ aggregation and deposition.^88,89

Drug development

Unlike β-secretase, γ-secretase has proved to be a highly tractable target for AD drug treatment, at least with respect to the development of orally bioavailable, brain-penetrant γ-secretase inhibitors (GSIs).⁹⁰ GSIs decrease Aβ production in human and mouse brains, and chronic administration decreases Aβ deposition in APP mouse models.^91–94 Multiple orally bioavailable, brain-penetrant GSIs have been identified; however, target-based toxicity has been a major obstacle to the clinical development of these compounds. Target-based toxicity was anticipated on the basis of preclinical studies. The γ-secretase enzyme processes numerous other type 1 transmembrane proteins in addition to APP; for example, γ-secretase processing of Notch1 is crucial for Notch signaling.⁹⁵ Deletion of PS1 is embryonically lethal in mice, and these mice display a phenotype—prominent skeletal and brain abnormalities—that is also evident in Notch deficient mice.^96,97 Moreover, nonselective GSI treatment leads to a variety of toxicities related to Notch inhibition, including, but not limited to, gastrointestinal, skin and immune system abnormalities.^98–100 Several dozen γ-secretase substrates have now been identified; however, it is unclear to what extent these other substrates, apart from Notch, contribute to the phenotypes that mice with attenuated γ-secretase processing display. Furthermore, accumulation of APP carboxy-terminal fragments following GSI treatment has been suggested to be toxic, although mice treated with GSIs provided no evidence for this assertion.¹⁰¹

GSIs are now being tested in clinical trials (Table 1). LY450139 is a nonselective GSI, and is currently being tested in a phase III study. Through use of stable isotope labeling combined with cerebrospinal fluid (CSF) sampling to measure the synthesis rate of Aβ, a single oral dose of this compound has been shown to lower CSF Aβ production in healthy men.^91,102 Notably, in the same study, enzyme-linked immunosorbent assay analyses of Aβ showed that a substantial decrease in total Aβ was evident only in the treatment group receiving the highest dose of the compound.

As noted previously, the main concern regarding nonselective GSIs is target-mediated toxicity. A narrow therapeutic window exists in which nonselective GSIs substantially reduce Aβ levels in the brain in the absence of Notch-related adverse effects. To overcome these toxicity issues, pharmaceutical companies have been trying to develop ‘Notch-sparing’ GSIs. Such pharmacological selectivity has been shown to be possible;^103,104 however, the original compounds were not particularly potent. Two new Notch-sparing inhibitors have recently been described (Table 1), one of which (BMS-708,163) has been shown in phase I clinical trials to lower plasma and CSF Aβ levels.¹⁰⁵ No human data have yet been disclosed for the other novel Notch-sparing inhibitor, begacestat.¹⁰⁶ The early Notch-sparing GSIs were reported to bind to an adenosine triphosphate site on the γ-secretase complex;¹⁰³ however, the mechanism by which the second-generation Notch-sparing GSIs work is either not known or has not been disclosed. Some authors have proposed that these GSIs work by binding to the substrate docking sites on γ-secretase that are distinct between Notch and APP. Another possibility is that the compounds selectively target different γ-secretase complexes. At present, the extent to which Notch-sparing GSIs inhibit the processing of the other identified γ-secretase substrates is unclear, and safety concerns around this issue are largely theoretical.

Compounds known as γ-secretase modulators (GSMs) have been shown to alter the profile of the Aβ peptides produced by γ-secretase activity in vitro and in vivo.^107,108 Aβ₄₂-reducing GSMs not only decrease levels of Aβ₄₂, but also increase levels of shorter Aβ peptides. Inverse or Aβ₄₂-raising GSMs, on the other hand, increase levels of Aβ₄₂ and decrease the production of shorter Aβ peptides.^107,109 Although GSMs have been reported to have varying effects on substrates other than APP, they do not seem to inhibit signaling events mediated by these substrates;^107,110 thus, GSMs seem to avoid target-based toxicity. Moreover, given that AD-associated mutations in APP and the presenilin genes modify γ-secretase activity subtly to favor Aβ₄₂ production by as little as 30%, selectively reducing the levels of Aβ₄₂ with GSMs is an attractive alternative treatment strategy to GSI therapy. Tarenflurbil (R-flurbiprofen), the first GSM to be evaluated in humans, was shown to be safe in the long term but recently failed in phase III clinical trials. After 18 months of treatment, patients with AD who received either the active treatment or placebo had almost identical scores on the Alzheimer’s Disease Cooperative Study–Activities of Daily Living (ADCS-ADL) score and the Alzheimer’s Disease Assessment Scale Cognitive Subscale (ADAS-Cog). This failure was possibly due to low potency ascribed to the compound and poor CNS penetration.¹¹¹ The recent discovery that certain GSMs bind directly to APP instead of γ-secretase suggests that these compounds might act through a novel mechanism.¹¹² Several GSMs have been reported that are better AD drug candidates than the first-generation GSM compounds, but their mechanisms of action remain unknown.¹¹³