J. Neurochem. (2012) 120, 869–880.
β-Site APP-cleaving enzyme (BACE1) cleaves the amyloid precursor protein (APP) at the β-secretase site to initiate the production of Aβ peptides. These accumulate to form toxic oligomers and the amyloid plaques associated with Alzheimer’s disease (AD). An increase of BACE1 levels in the brain of AD patients has been mostly attributed to alterations of its intracellular trafficking. Golgi-associated adaptor proteins, GGA sort BACE1 for export to the endosomal compartment, which is the major cellular site of BACE1 activity. BACE1 undergoes recycling between endosome, trans-Golgi network (TGN), and the plasma membrane, from where it is endocytosed and either further recycled or retrieved to the endosome. Phosphorylation of Ser498 facilitates BACE1 recognition by GGA1 for retrieval to the endosome. Ubiquitination of BACE1 C-terminal Lys501 signals GGA3 for exporting BACE1 to the lysosome for degradation. In addition, the retromer mediates the retrograde transport of BACE1 from endosome to TGN. Decreased levels of GGA proteins and increased levels of retromer-associated sortilin have been associated with AD. Both would promote the co-localization of BACE1 and the amyloid precursor protein in the TGN and endosomes. Decreased levels of GGA3 also impair BACE1 degradation. Further understanding of BACE1 trafficking and its regulation may offer new therapeutic approaches for the treatment of Alzheimer’s disease.
APP intracellular domain
amyloid precursor protein
soluble APP N-terminal fragment
sortin nexin 6
‘vps, Hrs, and STAM’
Alzheimer’s disease (AD) is the major cause of dementia in the elderly and constitutes a major socio-economic burden to society with the increased aging of the population. Compelling genetic and experimental evidence implicate the accumulation of Aβ in the brain and the toxicity of Aβ oligomers in the etiology of AD (Masters and Beyreuther 2006). Aβ was initially identified as the major component of the amyloid plaques, which have defined the pathology of AD. The purification and sequencing of Aβ peptides isolated from amyloid plaques led to cloning of the amyloid precursor protein (APP) and finding that Aβ was a fragment excised from its precursor (Kang et al. 1987). APP is a type I integral membrane protein, which resembles a cell-surface receptor and can be alternatively processed through an amyloidogenic or a non-amyloidogenic pathway (reviewed in Evin and Weidemann 2002). In the amyloidogenic pathway leading to Aβ production, cleavage by β-secretase, or β-site APP-cleaving enzyme (BACE)1, 28 amino acid N-terminal distal from the transmembrane domain releases a large soluble APP N-terminal fragment (sAPPβ), with concomitant production of a 99-amino acid C-terminal fragment (C99) that contains Aβ and remains associated with the membrane. Processing of C99 by the membrane-embedded γ-secretase complex results in the release of 39–43 amino acid Aβ peptides on the luminal side of the membrane, and in the cytosolic release of the APP intracellular domain (AICD) that allows its translocation to the nucleus with transcriptionally active binding partners. Whereas a putative function for Aβ has been proposed and remains under investigation (Grosgen et al. 2010; Soscia et al. 2010), it has now been clearly established that the overproduction of Aβ, in particular overproduction of its longest and most aggregating forms, results in the formation of toxic fibrils causing neurodegeneration (Palop and Mucke 2010). Extensive cellular studies have proven the neurotoxicity of Aβ, and narrowed the toxic species to pre-amyloid, soluble Aβ oligomers (Shankar and Walsh 2009). Consistent with these experimental studies, clinical data have shown that the abundance of Aβ soluble oligomers correlates with cognitive decline (Lue et al. 1999; McLean et al. 1999; Naslund et al. 2000). Imaging of Aβ by positron emission tomography has revealed an increase in Aβ deposition in patients with mild cognitive impairment compared to age-matched controls, and this increase has been associated with a greater risk of developing AD (Villemagne and Rowe 2011).
The cleavage of APP by BACE1 constitutes the rate-limiting step in Aβ production as APP can also be processed by a non-amyloidogenic pathway that involves α-secretase enzymes such as ADAM-10 and ADAM-17 (reviewed in Evin and Weidemann 2002). In the non-amyloidogenic pathway, α-secretase cleaves APP within the Aβ sequence, 16 amino acid distal to the N-terminal end of the transmembrane domain, to release sAPPα, and produce an 83 amino acid, membrane-bound C-terminal fragment (C83). This fragment can be further processed by γ-secretase to release a 3 kDa N-terminal fragment (p3) and AICD, although a recent study indicates that the amyloidogenic pathway is the preferred route for the release of AICD (Belyaev et al. 2010). Whether APP becomes processed by BACE1 or by α-secretase will depend on its co-localization with either enzyme, which in turn will depend on the control of the cellular distribution of each secretase and APP. This review will focus on BACE1, its cellular trafficking, and consequent implications to AD pathogenesis.
The discovery of BACE1 as β-secretase
Over a decade ago, five research groups independently identified the enzyme responsible for β-secretase activity as a novel aspartyl protease. Using an expression cloning strategy that enabled identification of genes that up-regulate Aβ levels, Vassar et al. (1999) identified a membrane-anchored aspartyl protease they termed BACE. Sinha et al. (1999) successfully isolated the same protease from human brain following a protein purification strategy including an inhibitor-affinity final step. Two other groups searched for β-secretase by screening express sequence tag databases with the assumption that the enzyme was an aspartyl protease (Hussain et al. 1999; Yan et al. 1999), and then showed the novel aspartyl protease Asp2 to possess the expected specificity of β-secretase. Tang’s group had earlier entered the sequence of the membrane-anchored aspartyl protease memapsin 2 before demonstrating its function as β-secretase (Lin et al. 2000). The five groups had identified the same protease and a consensus has been reached to name it BACE, or BACE1 to distinguish it from its homologue BACE2 that is expressed at very low levels in the brain and was thus ruled out as a β-secretase candidate (Bennett et al. 2000a; Farzan et al. 2000; Hussain et al. 2000).
Over-expression of BACE1 in cell lines stably transfected with APP, or APP carrying the Swedish mutation [AD causative KM→NL double mutation at residues [595–596] of APP695, which favours processing by β-secretase (Citron et al. 1992)] was shown to increase Aβ secretion and production of C99 (Hussain et al. 1999; Sinha et al. 1999; Vassar et al. 1999; Yan et al. 1999; Lin et al. 2000). Co-transfection of BACE1 with APP constructs also resulted in increased levels of sAPPβ and Aβ, with concurrent reduction of sAPPα levels (Sinha et al. 1999). Sequencing and mass spectrometry analysis of the APP products derived from BACE1 cleavage confirmed that BACE1 had the correct cleavage specificity to be β-secretase (Sinha et al. 1999; Vassar et al. 1999) and revealed a major secondary cleavage at Glu +11, producing a C89 fragment. The specificity of BACE1 was further demonstrated by showing that a synthetic APP substrate encompassing the Swedish mutation was cleaved at a higher rate than the corresponding wild type sequence (Vassar et al. 1999; Lin et al. 2000).
Gene interference experiments further established that BACE1 was the major β-secretase. Treatment of cultured cells with BACE1 antisense oligonucleotides resulted in a significant decrease of both sAPPβ and Aβ levels, with a concomitant rise in sAPPα level (Vassar et al. 1999; Yan et al. 1999). Gene knockout studies in animals showed that cortical neurons from BACE1-/- mouse embryos produced no detectable amounts of Aβ40 and Aβ42 or C99 fragment (Cai et al. 2001; Roberds et al. 2001). Remarkably, the mice were viable, fertile, and showed no overt phenotype, supporting the value of BACE1 inhibition for AD therapy.
Besides cleaving APP, BACE1 has been reported to process a subset of other substrates, including the APP homologues, APLP1 and APLP2 (Eggert et al. 2004; Li and Südhof 2004), alpha-2,6-sialyltransferase (Kitazume et al. 2005), the cell adhesion P-selectin glycoprotein ligand-1 (Lichtenthaler et al. 2003), the interleukin-1 receptor type II (Kuhn et al. 2007), the low density lipoprotein receptor-related protein (von Arnim et al. 2005), neuregulin 1 type III β1 (Hu et al. 2006; Willem et al. 2006), and the β2 subunit of voltage-gated sodium channel (Wong et al. 2005; Kim et al. 2011). Therefore, BACE1 inhibitor therapy will have to be monitored carefully so as to minimise potential side-effects.
BACE1 distribution in human brain
BACE1 tissue distribution is consistent with that expected for β-secretase. Although BACE1 is expressed in a wide variety of tissues (Vassar et al. 1999), its highest expression occurs in the brain and pancreas (Vassar et al. 1999; Lin et al. 2000; Marcinkiewicz and Seidah 2000), and despite the abundant level of BACE1 mRNA in the pancreas, the form expressed in this tissue is an alternative splice-variant that cleaves poorly APP (Bodendorf et al. 2001). BACE1 mRNA has been detected in various regions of the brain, the highest levels being found in the cortex (Vassar et al. 1999). Accordingly, transgenic mice over-expressing BACE1 showed particularly high levels of expression in cortical regions (Chiocco et al. 2004). Crossing these mice with APP transgenics resulted in specific alterations of APP processing and increased Aβ deposition in the cortex relative to other brain regions, in a pattern reminiscent of AD pathology.
Under normal conditions, BACE1 cellular distribution in the brain, is mostly confined to neuronal cells (Hussain et al. 1999). This is consistent with early reports that β-secretase activity is low in astrocytes and that glial cells do not significantly contribute to Aβ levels in the brain (Zhao et al. 1996). However, studies with Tg2576 mice, which over-express the human APPSwe mutant, have shown that the astrocytic expression of BACE1 increases with age (Rossner et al. 2001). Immunolabelling of BACE1 and Aβ in these aged transgenic mice, after development of plaque pathology, revealed that BACE1 was selectively expressed in activated astrocytes surrounding the amyloid deposits and may have been triggered by Aβ toxicity. Chronic gliosis may trigger BACE1 expression in astrocytes (Hartlage-Rubsamen et al. 2003), suggesting that BACE1 expression may be a characteristic of astrocyte activation in response to various insults of the central nervous system. An experimental rat model of traumatic brain injury also showed high levels of BACE1 expression in both neurons and astrocytes (Blasko et al. 2004), and increased levels of sAPPβ in brain homogenates 48 h post-injury. Together, these findings showed that BACE1 expression occurs in neuronal cells, but that it can also occur in astrocytes following stress or trauma, and thereby exacerbate AD pathology.
BACE1 expression in Alzheimer’s disease
Post-mortem studies have revealed that the levels of BACE1 protein are increased in the brain of AD sufferers (Holsinger et al. 2002, 2004; Yang et al. 2003; Li et al. 2004; Johnston et al. 2005; Tesco et al. 2007; Hébert et al. 2008; Santosa et al. 2011). Increases in BACE1 proteolytic products (Holsinger et al. 2002) and enzymatic activity were also reported (Fukumoto et al. 2002; Yang et al. 2003; Li et al. 2004; Borghi et al. 2007) in frontal and temporal cortices of AD patients. In 2002, we reported that BACE1 protein is increased by over 2-fold in the frontal cortex of AD patients and we have recently confirmed this result using a new sample cohort. Our new study (Santosa et al. 2011) and that by Hébert et al. (2008) indicate that AD patients are divided in two subgroups, with high or normal BACE1 levels, those with high BACE1 levels representing about half of all AD cases. Li et al. (2004) have reported that both BACE1 mRNA and BACE1 protein are elevated in AD brain. We and others could not demonstrate a change in BACE1 mRNA expression (Holsinger et al. 2002; Johnston et al. 2005). An increase in BACE1 protein expression in AD has been attributed to alterations in gene translation regulation due to decreased levels of miRNA-29a and -29b (Hébert et al. 2008; Wang et al. 2008). An increase in a non-coding antisense transcript that stabilizes BACE1 mRNA has also been correlated with increased levels of Aβ42 in AD brain (Faghihi et al. 2008). Other studies propose that high BACE1 levels in AD cases are due to alterations of its metabolism as a result of abnormal cellular trafficking, as discussed in detail later. Therefore, BACE1 protein increase, as observed in about half of AD cases, may be associated with various post-translational and post-transcriptional events rather than a change in gene expression.
BACE1 structural features and post-translational modifications
BACE1 structure and biochemical properties
The BACE1 gene codes for a protein of 501 amino acids (Hussain et al. 1999; Sinha et al. 1999; Vassar et al. 1999; Yan et al. 1999; Lin et al. 2000) consisting of an N-terminal signal peptide [residues 1–21], followed by a pro-domain [residues 22–45], a protease domain [residues 46–460] that contains two consensus motifs characteristic of an aspartyl protease active site (DTGS, at residues 93–96, and DSGT, at residues 289–292), a single transmembrane domain [residues 461–477], and a short cytosolic domain [residues 478–501] (Fig. 1). The protease domain shows significant homology to other members of the pepsin family (30% identity to cathepsin E; 30–37% homology to renin and cathepsin D), but differs notably in the arrangement of the disulphide bonds (Haniu et al. 2000). The crystal structure of BACE1 in complex with an inhibitor has revealed that the enzyme’s active site is larger and less hydrophobic than that of other human aspartic proteases (Hong et al. 2000), making it a difficult target for designing inhibitors (for recent review see Evin et al. 2011). A distinctive feature of BACE1 is its anchoring in the membrane through a single transmembrane domain. This restricts its spatial distribution and constrains the placement of its catalytic domain in the same orientation as APP (Creemers et al. 2001). With an optimal pH of 4.0–5.5, BACE1 is expected to operate in acidic intracellular compartments such as the trans-Golgi network (TGN), endosomes and lysosomes (Sinha et al. 1999; Vassar et al. 1999; Lin et al. 2000).
ProBACE1 and removal of the prodomain
The purification of BACE1 from human brain revealed that 10% of the protein started at tyrosine residue 22, and 90% at glutamate 46 (Sinha et al. 1999), suggesting that the N-terminal region of BACE1 was processed. Sequence analysis confirmed the presence of a prodomain homologous to that of other aspartyl protease precursors (Koelsch et al. 1994). Treatment of cells with brefeldin A, which blocks protein transport from ER to Golgi, resulted in the accumulation of proBACE1 and demonstrated that the removal of BACE1 prodomain takes place in the Golgi (Bennett et al. 2000b). The role of the prodomain would be to assist in the proper folding of the protein rather than inhibiting BACE1 enzymatic activity (Capell et al. 2000; Haniu et al. 2000) as significantly less activity was recovered when recombinant BACE1 was produced without a prodomain (Shi et al. 2001). Cellular studies have shown that the prodomain can be shed by proprotein convertases (PC). These are Ca2+-dependent subtilisin-like serine proteases that reside in the Golgi and cleave protein sequences specifically after a K/R-(X)n-K/R basic motif, where n = 2, 4 or 6 residues and X is any amino acid except cysteine or proline (Seidah and Chretien 1999). The prodomain of BACE1 bears the PC recognition motif R-X-X-R (at residues 42–45), and data from BACE1 N-terminal sequencing are consistent with cleavage of its prodomain occurring after this motif (Bennett et al. 2000b). Co-transfection of BACE1 with the furin inhibitor, α1-antitrypsin (α1-PDX) resulted in the accumulation of proBACE1 (Bennett et al. 2000b), thereby demonstrating the role played by furin in the conversion of proBACE1. PC7 and PC5/6 can also process BACE1 prodomain (Pinnix et al. 2001). Preventing proBACE1 removal by mutating the arginines within the PC recognition site severely affect BACE1 maturation (Capell et al. 2000; Benjannet et al. 2001), suggesting that the prodomain hinders interaction with the Golgi post-translational machinery, and that BACE1 prodomain removal occurs upon reaching the Golgi. Some studies proposed that proBACE1 can be processed by autocatalysis and cleavage near the end of the prodomain (Lin et al. 2000; Beckman et al. 2006) and this can be facilitated by the binding of heparin and glycosaminoglycans (GAG) to proBACE1 (Klaver et al. 2010).
BACE1 full maturation involves various post-translational modifications. Analysis of its primary sequence revealed four sites of N-glycosylation within the protease domain (Asn residues 153, 172, 223 and 354), which have been confirmed by site-directed mutagenesis (Haniu et al. 2000; Charlwood et al. 2001). BACE1 undergoes co-translational N-glycosylation in the ER, as demonstrated by treatment with tunicamycin, which inhibits the first step of glycoprotein synthesis. Further complex glycosylation is achieved as BACE1 transits through the Golgi (Capell et al. 2000). Once BACE1 full glycosylation is complete, it becomes insensitive to endoglycosidase H (Capell et al. 2000; Creemers et al. 2001; Pinnix et al. 2001). There is evidence that sulfation can also occur on the N-glycosylation sites as part of BACE1 maturation (Benjannet et al. 2001). Depending on experimental conditions and individual cell lines’ glycosylation machinery, the molecular weight of BACE1 has been reported to be between 70 and 75 kDa for the mature protein (Capell et al. 2000; Haniu et al. 2000; Creemers et al. 2001; Pinnix et al. 2001), and between 60 and 70 kDa for its immature forms (Capell et al. 2000; Haniu et al. 2000; Creemers et al. 2001; Pinnix et al. 2001). Complete deglycosylation by treatment with endoglycosidase F results in a molecular weight of about 50 kDa, which is consistent with BACE1 primary sequence. Thus, the carbohydrate content of BACE1 accounts for about 30% of its molecular weight, similar to other endosomal/ lysosomal resident proteins.
Other post-translational modifications
Prior to its full maturation in the Golgi, BACE1 undergoes reversible acetylation in the endoplasmic/Golgi intermediate compartment (Costantini et al. 2007; Jonas et al. 2008; Ko and Puglielli 2009). Nascent BACE1 becomes transiently acetylated in a cluster of seven lysine residues, thereby increasing the stability of the protein and allowing its traffic through the secretory pathway. Two lysine acetyltransferases involved with this process have been characterized and shown to be up-regulated by ceramide, a sphingolipid intermediate associated with stress response and apoptosis (Ko and Puglielli 2009).
The cytoplasmic tail of BACE1 is susceptible to several post-translational modifications. A cluster of three cysteine residues (at positions 478, 482 and 485) can undergo palmitoylation (Benjannet et al. 2001). This modification regulates the targeting of proteins to lipid rafts and favours their oligomerization (Levental et al. 2010). Mutation of BACE1 cytoplasmic cysteines enhanced the levels of protein recovered in cell culture media, indicating that palmitoylation stabilize BACE1 anchoring at the plasma membrane and limits its putative shedding (Benjannet et al. 2001).
BACE1 C-terminal tail can also undergo phosphorylation at serine 498 and ubiquitination at lysine 501, two modifications important for its cellular targeting, as outlined later.
BACE1 has been primarily detected as a monomer in detergent-solubilised cell lysates (Huse et al. 2000); however, several reports have described the existence of BACE1 dimers (Schmechel et al. 2004; Westmeyer et al. 2004; Jin et al. 2010). Native-gel electtrophoresis of homogenates from human cells and mouse brains has characterized BACE1 as a 140 kDa dimer. Co-immunoprecipitation studies from cells co-expressing alternatively tagged BACE1 constructs revealed a homodimeric interaction (Westmeyer et al. 2004). As mutation of the active site Asp289 lowered, but did not abolish β-secretase activity, it has been proposed that two N-terminal aspartates (Asp93) brought into close proximity upon homodimerization would help retain enzymatic activity. BACE1 homodimerization is further corroborated by a recent study showing that simultaneous expression of two BACE1 constructs, each mutated at a single and different catalytic aspartate, can partially rescue BACE1 activity in BACE1-/- mouse embryo fibroblasts (Jin et al. 2010). The dimeric form of BACE1 is more active in cleaving a synthetic APP substrate than monomeric soluble BACE1 lacking both the transmembrane and cytosolic domains (Westmeyer et al. 2004), suggesting that this is the optimal form of the enzyme in vivo.
BACE1 cellular trafficking and metabolism
BACE1 is a relatively stable protein with a half-life of about 16 h (Huse et al. 2000). The cytosolic domain contains specific residues that determine its trafficking between organelles (Fig. 1). BACE1 contains a di-leucine (LL) motif at positions Leu499–Leu500. The LL motif is a signal for sorting proteins at the TGN and increases retrieval from the plasma membrane for targeting to the endosomal/lysosomal compartment (Sandoval and Bakke 1994). Accordingly, deletion of the LL motif causes BACE1 to accumulate at the plasma membrane and prevents its sorting to endosomes as well as its endocytosis (Huse et al. 2000; Pastorino et al. 2002). It also prevents BACE1 from being degraded in the lysosome (Koh et al. 2005). Phosphorylation of serine residue 498 also contributes to orientating BACE1 trafficking (Walter et al. 2001; Pastorino et al. 2002). Mutagenesis of Ser498 produces mutants that accumulate in early endosomes, in contrast to wild type proteins that also localise to perinuclear structures corresponding to Golgi and endosomes. Thus, the phosphorylation of Ser498 promotes BACE1 retrieval to the endosomal compartment and recycling to the TGN. A list of the major proteins that have been uncovered as key players in the cellular trafficking of BACE1 is given in Table 1, and these are reviewed below.
|Protein name||Function||Interaction with BACE||Effect on BACE trafficking||Association with AD||References|
|GGA1||Adaptor sorting protein||[496–500] DXXLL motif |
DXXLL motif with phosphorylated Ser498
|Sorting at Golgi for endosome targeting |
Recycling from plasma membrane and endosomes to TGN
|Levels may be decreased, thus impair BACE trafficking||He et al. 2002; |
Wahle et al. 2006;
Santosa et al. 2011
|GGA3||Adaptor sorting protein||[496–500] DXXLL motif |
|Sorting at Golgi for endosome targeting |
Sorting from endosome to lysosome for degradation
|Decreased levels coincide with impaired BACE trafficking and turnover||He et al. 2002; |
Tesco et al. 2007;
Santosa et al. 2011
|Seladin (DHCR24 reductase)||3-β-hydroxysteroid-Δ-24 reductase. Involved in cholesterol synthesis||Indirect||Protects neurons against oxidative stress. Controls raft formation and segregation of APP from BACE. Prevents casp-3 cleavage of GGA3||Decreased levels contribute to GGA3 depletion and increased BACE levels||Greeve et al. 2000; |
Crameri et al. 2006;
Sarajarvi et al. 2009
|E3 ligase Fbx2||Neuron-specific ubiquitin ligase||Lys501||Mediates BACE ubiquitination in early endosomes and its targeting to lysosomes||No evidence yet||Gong et al. 2010|
|GTPase ADP ribosylation factor 6 (ARF6)||Plasma membrane/endosomal protein regulating endocytic trafficking||[499–500] LL motif||Modulator of BACE endocytosis and sorting from plasma membrane to early endosomes through a different route to APP||No evidence yet||Sannerud et al. 2011|
|SorLa (SORL1/LR11)||Receptor sorting molecule. Retains APP in Golgi.||Co-immunoprecipitation with BACE and APP||Prevents APP/BACE interaction in the Golgi and controls BACE cleavage of APP||Susceptibility gene for AD. Decreased levels of SorLa in AD brain. Would favour cleavage of APP by BACE||Scherzer et al. 2004; |
Spoelgen et al. 2006
|Sortin nexin 6 (SNX6)||Retromer subunit involved in retrograde trafficking of sortilin and other proteins from endosomes to TGN||Indirect?||Negative regulator of BACE retrograde transport Down-regulation increases endogenous levels of BACE and APP processing by BACE||Retromer defects in AD may increase BACE retrograde transport and favour co-localization of BACE with APP in TGN||Okada et al. 2010|
|Sortilin||Retromer-mediated retrograde transport||Indirect?||Positive regulator of BACE recycling. Down-regulation causes decreased processing of APP by BACE||Increased levels in AD brain. Favours BACE and APP co-localization in TGN||Finan et al. 2011|
The role of GGA in mediating BACE1 trafficking
The trafficking of BACE1 is tightly controlled by Golgi-localised γ-adaptin ear-containing ADP ribosylation factor-binding proteins (GGA), a family of monomeric clathrin-adaptor proteins that facilitate the sorting of selected membrane proteins between TGN and endosomes (Boman 2001). There exist four human GGA homologues, which all comprise the three following structural domains: a ‘vps, Hrs, and STAM’ (VHS) domain that binds to the acidic-cluster-dileucine motif (DXXLL) of cargo proteins; a ‘GGA and TOM’ (GAT) domain that mediates association with Golgi ADP ribosylation factor 6-GTP receptors; a γ-adaptin ear domain, involved in recruiting adaptin and other proteins for the packaging of transport vesicles. The GAT and GGA domains are linked by a hinge domain that binds to clathrin. The BACE1 DXXLL consensus sequence, that includes Ser498 and the Leu499–Leu500 motif, mediates an interaction with the VHS domain of GGA proteins (He et al. 2002). Immunofluorescence studies have demonstrated the colocalisation of BACE1 and GGA in the TGN and endosomes (Shiba et al. 2004). Shiba and colleagues also showed that phosphorylation of Ser498 enhanced BACE1 binding affinity for the VHS domain, suggesting that BACE1 interaction with GGA proteins is phosphoregulated. RNA interference to abolish GGA expression also resulted in the accumulation of BACE1 in the endosomes and confirmed the important role that GGA proteins play in regulating BACE1 trafficking (He et al. 2005). BACE1 can potentially bind to the four GGA homologues, but the best-characterized interactions involve GGA1 and GGA3. Experimental evidence indicates that GGA1 facilitates the retrograde transport of phosphorylated BACE1 from endosomes to TGN (He et al. 2004; Wahle et al. 2005), whereas GGA3 mediates targeting of BACE1 to the lysosomes for degradation. GGA3, which contains two binding sites for ubiquitin in its hinge region, is known to play a role in targeting ubiquitinated cargoes for lysosomal degradation (Puertollano and Bonifacino 2004). Down-regulating GGA3 expression by siRNA increased BACE1 cellular levels and thus highlighted the involvement of GGA3 in regulating BACE1 turnover (Tesco et al. 2007). Ubiquitination at C-terminal Lys501, is a GGA3 recognition signal independent from the LL motif, and targets BACE1 to the lysosome for degradation (Qing et al. 2004; Tesco et al. 2007; Kang et al. 2010). BACE1 ubiquitination is mediated by the neuron-specific Skp1-Cullin1-F-box 2 (Fbx2)-Roc1 ubiquitin ligase complex, SCFfbx2-E3 (Gong et al. 2010; Liang et al. 2010). Over-expressing of the E3 ubiquitin ligase subunit Fbx2 in neuronal cells of APP transgenic Tg2576 mice enhanced BACE1 degradation and reduced Aβ production. The administration of exogenous viral expression of Fbx2 into the brain of Tg2576 mice attenuated amyloid deposition and improved synaptic function. These studies demonstrate the importance of ubiquitination for BACE1 turnover.
A loss of GGA3 contributes to BACE1 accumulation in the endosomes
Tesco et al. have shown that, under apoptosis, GGA3 is cleaved by caspase 3, and this impedes the targeting of BACE1 from endosomes to lysosomes (Tesco et al. 2007). Down-regulation of the neuroprotective protein, seladin-1, as observed in AD cortex (Greeve et al. 2000), exacerbates GGA3 depletion (Sarajarvi et al. 2009). Indeed, manipulation of seladin-1 expression in human neuroblastoma cells by siRNA altered the cellular response to apoptosis, with enhanced caspase 3 activity and depletion of GGA3. Consequently, increased BACE1 protein levels and activity were observed, and more APP was processed through the amyloidogenic pathway.
Studies in animal models further support a link between caspase 3 activation and increased BACE1 stability (Xiong et al. 2008; Zhang et al. 2010). The administration of caspase 3 inhibitor to rats subjected to transient ischemia attenuated BACE1 protein accumulation and reduced C99 levels (Xiong et al. 2008). Rat studies also demonstrated that caspase 3 activation during ischemia stimulates the generation of mutant ubiquitin UBB+1 (Zhang et al. 2010). The UBB+1 protein is an aberrant form of ubiquitin previously found in plaques and tangles in AD and Down syndrome patients (van Leeuwen et al. 1998). In the rat model, caspase 3 activation preceded an increase of UBB+1 levels, whereas caspase 3 inhibition reduced both UBB+1 and BACE1 levels (Zhang et al. 2010). Furthermore, UBB+1 can bind to BACE1, and this interaction is diminished upon caspase 3 inhibitor treatment. Elevated mutant ubiquitin UBB+1 behaves as a competitive inhibitor of proteasome function (van Tijn et al. 2007); therefore the above studies provide supporting evidence for the involvement of the ubiquitin-proteasome system in BACE1 degradation, as proposed by Qing et al. (2004). However, Kang et al. (2010) have argued that, under stress conditions, UBB+1 would as also impair BACE1 degradation by the lysosome.
The role of the retromer in BACE1 internalization and recycling
The retromer protein complex controls the recycling of selected receptors between endosomes and TGN. Silencing the expression of the retromer subunit VPS26 by gene interference caused BACE1 accumulation in early endosomes (He et al. 2005). A molecular association of BACE1 with the retromer subunit sortin nexin 6 (SNX6) has also been recently identified by applying cross-linking and tandem-affinity purification to BACE1 in human neuroblastoma SH-SY5Y cells (Okada et al. 2010). SNX6 is a component of a multiprotein cargo complex known to mediate retrograde trafficking of sortilin and the cation-independent mannose-6-phosphate receptor. Down-regulation of SNX6 expression increases BACE1 cellular levels, with a concomitant rise in sAPPβ and C99 levels (Okada et al. 2010). Immunofluorescence studies indicate that the depletion of SNX6 perturbs the transport of BACE1 in the endocytic pathway, resulting in its accumulation in uncharacterized perinuclear structures that are negative for TGN, ER and endosomal markers. Thus, SNX6 may act as a modulator of BACE1 trafficking from endosome to TGN in an opposite manner to GGA that promotes the sorting of BACE1 from TGN to endosomes.
Recently, the Vps10p domain-sorting receptor sortilin has also been implicated in the retrograde transport of BACE1 from endosomes to TGN (Finan et al. 2011). The cytoplasmic domain of sortilin is required for its interaction with the retromer and for its retrieval of BACE1 to the TGN. Indeed, over-expression of sortilin mutants with a truncated cytoplasmic tail caused the redistribution of BACE1 to the endosomes and impaired its retrograde transport to the TGN. The over-expression of sortilin increased the cleavage of APP by BACE1, presumably by promoting the interaction of these two proteins in the TGN. Conversely, down-regulating the expression of sortilin reduced the production of sAPPβ and Aβ.
Another recent study, which analysed in detail BACE1 internalization from the plasma membrane, reported that the small GTPase ADP ribosylation factor 6 selectively targeted BACE1 to RAB GTPase 5-positive early endosomes independently from clathrin-dependent endocytosis (Sannerud et al. 2011).
Alterations of BACE1 trafficking in AD promote Aβ production
The cellular trafficking of BACE 1 is summarised in Fig. 2. BACE1 transits through the secretory pathway and is sorted in the Golgi for transport to the endosome in clathrin-coated vesicles, under the control of GGA adaptor proteins. From there, it can recycle to the TGN and plasma membrane. BACE1 that has reached the plasma membrane is then endocytosed in into early endosomes from where it can follow two alternative routes. Following one path, it will be recycled to the TGN under the control of GGA1 for re-delivery to the plasma membrane, pending its phosphorylation at Ser498. In the alternate path, it will follow the default endocytic pathway to return to the endosome and, after ubiquitination, to be delivered to the lysosome for degradation under the control of GGA3. This complex trafficking may be disrupted at several levels.
Tesco and colleagues have reported that GGA3 protein levels are decreased in AD temporal cortex and are inversely correlated to the levels of BACE1, suggesting that GGA3 is directly responsible for the control of BACE1 levels in the brain (Tesco et al. 2007). Our own studies have also demonstrated a marked decrease of GGA3 and a trend for a decrease in GGA1 in AD frontal cortex (Santosa et al. 2011). We could not establish a direct correlation between BACE1 and GGA levels in AD frontal cortex, as it appears that BACE1 expression was heterogeneous in our sample cohort and perhaps dependent on additional factors, such as translational events. However, we showed that in AD cases with low GGA3 levels, BACE1 subcellular distribution was altered so that it was more extensively localized in membrane fractions rich in APP, providing a potential mechanism for increased amyloidogenic processing of APP and increased Aβ production in AD. Two other studies have also described the expression of GGA1 in human temporal cortex and presented data to support its decrease in AD (Wahle et al. 2006).
Tesco’s group has provided solid evidence for the involvement of the lysosomal system in BACE1 degradation (Koh et al. 2005). As the dysfunction of lysosomal system is an invariant feature of AD and other age-related pathologies (Nixon et al. 1992, 2000; Bahr and Bendiske 2002), this may contribute to the high levels of BACE1 protein reported in AD brain.
Changes in BACE1 recycling pathway can also contribute to increasing its levels in AD. Sortilin acts as a positive regulator BACE1 retrograde trafficking and favours the amyloidogenic processing of APP (Finan et al. 2011). Interestingly, it has been reported that the levels of neuronal sortilin are increased by about 25% in AD post-mortem temporal cortex (Finan et al. 2011), suggesting that this contributes to increasing BACE1 and amyloid levels in AD brain. It would be worth investigating a possible link between BACE1 expression and a retromer deficiency observed in AD (Muhammad et al. 2008).
A wealth of experimental evidence supports that the cellular trafficking of APP is distinct from that of BACE1 (Ranganathan et al. 2011). Therefore, the encounter of the two proteins is expected to be limited. APP follows the secretory pathway and can be shed by α-secretases such as ADAM-10 or ADAM-17 upon reaching the plasma membrane, or it can be endocytosed through clathrin-coated pits. The NPXY motif in the cytosolic tail of APP is a recognition motif for the targeting of cargoes to clathrin-coated vesicles and regulates the endocytosis of APP together with its binding partners. APP trafficking is controlled by SorLA (or LR11), which sequesters APP in the Golgi to control its delivery to the endocytic pathway (Andersen et al. 2005; Spoelgen et al. 2006). The interaction with SorLA also blocks APP from accessing BACE1 (Andersen et al. 2006). Therefore, a significant reduction of SorLA, as was observed in AD brain, may contribute to increasing Aβ levels (Scherzer et al. 2004). This would supplement the alterations in BACE1 trafficking and degradation caused by either a decrease of GGA, a defect in ubiquitination, an increase of sortilin levels, a dysfunction of the lysosomal system, or alterations in cellular membrane lipid composition and fluidity.
Experimental evidence also supports that Aβ production occurs in lipid rafts [reviewed by (Cheng et al. 2007)]. Lipid rafts are membrane microdomains that are enriched in cholesterol and sphingolipids, and that are primarily localized to Golgi and plasma membranes. Palmitoylation at cysteine residues within BACE1 cytoplasmic domain is a targeting signal for segregation into lipid rafts (Benjannet et al. 2001), and cellular studies indicate that indeed a significant proportion of BACE1 is associated with lipid rafts (Riddell et al. 2001; Tun et al. 2002; Marlow et al. 2003). Substituting BACE1 transmembrane and cytoplasmic domains with a GPI-anchor was shown to up-regulate sAPPβ and Aβ production in human neuroblastoma cells, supporting that APP cleavage by BACE1 occurs in lipid rafts (Cordy et al. 2003). Other studies with neuroblastoma cells indicate that APP and BACE1 are mostly located in separate lipid rafts that will fuse during endocytosis (Ehehalt et al. 2003). Co-patching experiments aiming to induce coalescence, at the plasma membrane, of the rafts containing APP and those containing BACE1 result in increasing Aβ production in a cholesterol-dependent manner (Ehehalt et al. 2003). Life-time imaging of APP and BACE1 co-transfected neurons showed the close vicinity of BACE1 and APP at the plasma membrane, and their increased co-localization following short exposure to cholesterol (Marquer et al. 2011). The role of cholesterol in the β-secretase processing of APP is further corroborated by a recent study indicating that the pluripotent lipid mediator, sphingosine-1-phosphate, is a positive regulator of BACE1 activity (Takasugi et al. 2011).
Because of technical limitations, only a few studies have been conducted on neuronal cells, although these represent the best model to analyse the amyloidogenic processing of APP. In contrast to alternate cell types, the neuronal cells express high levels of BACE1 and of a particular variant of APP, the APP695 isoform. The complexity and the tight regulation of BACE1 and APP cellular trafficking that restrict their co-localization would suggest that APP processing by BACE1 fulfils a physiological role in the brain. Considering that the deactivation of GGA function by caspase cleavage and the up-regulation of sortilin occur in response to stress and changes in cholesterol levels, and that both of these events favour the amyloidogenic processing of APP, it may be hypothesized that BACE1 cleavage of APP contributes to a repair or a damage-control mechanism for aging neurons. This will eventually become overwhelmed as Aβ overproduction and accumulation will cause neurotoxicity and trigger further co-localization of BACE1 with APP and Aβ production. Further understanding of BACE1 trafficking and of its regulation may offer new therapeutic avenues for AD treatment.
We thank Dr Janetta G. Culvenor for critical reading of the manuscript. This work is supported in part by the National Health and Medical Research Council of Australia (NHMRC project grant 566520) and by the Judith Jane Mason and Harold Stannett Williams Memorial Foundation (ANZ Mason Foundation). The authors have no conflict of interests to declare.