Alternative Selection of β-Site APP-Cleaving Enzyme 1 (BACE1) Cleavage Sites in Amyloid β-Protein Precursor (APP) Harboring Protective and Pathogenic Mutations within the Aβ Sequence*

β-Site APP-cleaving enzyme 1 (BACE1) cleaves amyloid β-protein precursor (APP) at the bond between Met671 and Asp672 (β-site) to generate the carboxyl-terminal fragment (CTFβ/C99). BACE1 also cleaves APP at another bond between Thr681 and Gln682 (β′-site), yielding CTFβ′/C89. Cleavage of CTFβ/C99 by γ-secretase generates Aβ(1-XX), whereas cleavage of CTFβ′/C89 generates Aβ(11-XX). Thus, β′-site cleavage by BACE1 is amyloidolytic rather than amyloidogenic. β′ cleavage of mouse APP is more common than the corresponding cleavage of human APP. We found that the H684R substitution within human Aβ, which replaces the histidine in the human protein with the arginine found at the corresponding position in mouse, facilitated β′ cleavage irrespective of the species origin of BACE1, thereby significantly increasing the level of Aβ(11-XX) and decreasing the level of Aβ(1-XX). Thus, amino acid substitutions within the Aβ sequence influenced the selectivity of alternative β- or β′-site cleavage of APP by BACE1. In familial Alzheimer's disease (FAD), the APP gene harbors pathogenic variations such as the Swedish (K670N/M671L), Leuven (E682K), and A673V mutations, all of which decrease Aβ(11–40) generation, whereas the protective Icelandic mutation (A673T) increases generation of Aβ(11–40). Thus, A673T promotes β′ cleavage of APP and protects subjects against AD. In addition, CTFβ/C99 was cleaved by excess BACE1 activity to generate CTFβ′/C89, followed by Aβ(11–40), even if APP harbored pathogenic mutations. The resultant Aβ(11–40) was more metabolically labile in vivo than Aβ(1–40). Our analysis suggests that some FAD mutations in APP are amyloidogenic and/or amyloidolytic via selection of alternative BACE1 cleavage sites.

␤-Site APP-cleaving enzyme 1 (BACE1), 3 a type I transmembrane aspartic protease, was identified as the ␤-secretase that cleaves amyloid ␤-protein precursor (APP) to generate neurotoxic amyloid ␤ (A␤) (1)(2)(3)(4). BACE1 cleaves APP at the peptide bond between Met 671 and Asp 672 (␤-site; sequence numbering refers to the APP770 isoform) (5,6). This primary cleavage of APP generates the secreted form of the amino-terminal large fragment (sAPP␤) and the membrane-associated carboxyl-terminal fragment (CTF␤/C99) (reviewed in Ref. 7). Because CTF␤/C99 includes the complete amino acid sequence of the A␤ region, and subsequent cleavage of CTF␤/C99 by ␥-secretase generates the A␤(1-XX) peptides (A␤1 indicates the position of Asp 672 ), the ␤-site cleavage is referred to as amyloidogenic processing of APP (reviewed in Ref. 8). Alternatively, APP can also be cleaved by ␣-secretase (mainly ADAM10/17), at the peptide bond between Lys 687 and Leu 688 , generating sAPP␣ and CTF␣/C83 including the A␤(17-XX) region; accordingly, this cleavage is referred to as amyloidolytic processing of APP (9 -11). BACE1 also cleaves APP at another peptide bond between Tyr 681 and Gln 682 (␤Ј-site), resulting in generation of sAPP␤Ј and CTF␤Ј/C89 (12,13). Cleavage of CTF␤Ј/C89 by ␥-secretase yields A␤ , which lacks the first 10 amino acids of the A␤ domain. This ␤Ј-site cleavage of APP is also thought to be amyloidolytic, because the structural analysis suggests that A␤  and A␤  are less toxic than A␤  and A␤   (14), and the A␤  showed reduced neurotoxicity compared with A␤(1-42) and A␤  in a study with transgenic fly (15), although the neurotoxicity of A␤  in human brain is controversial (16).
A␤ is the major protein component of senile plaques observed in the brain of Alzheimer's disease (AD) subjects, and soluble A␤ oligomer(s) are thought to impair synaptic functions prior to A␤ deposition in the brain (17)(18)(19). Therefore, to understand AD pathogenesis and develop effective AD ther-apies, it is important to elucidate the molecular mechanisms of A␤ generation and degradation. Regulation of BACE1 activity represents a promising therapeutic option for decreasing A␤ production (reviewed in Ref. 20).
In familial AD (FAD), several amino acid mutations have been reported in the APP gene, especially in or around the A␤ sequence. Some of these mutations are pathogenic, e.g. the Swedish (K670N/M671L), Leuven (E682K), and A673V mutations increase A␤ generation by promoting APP ␤-site cleavage by BACE1 (21)(22)(23). On the other hand, at least one protective mutation exists: the Icelandic mutation, A673T, decreases A␤ generation (24,25). Although the effect of the Swedish mutation, which is located outside the amino terminus of the A␤ sequence, is obvious, the molecular mechanisms by which amino acid alterations within A␤ affect the level of A␤ in the brain remain controversial. Because BACE1 cleaves APP either at the amyloidogenic ␤-site or the amyloidolytic ␤Ј-site, we investigated the effect of alternative BACE1 cleavage sites in APP on generation and degradation of A␤.

Results
Alternative Cleavage of APP by BACE1 Depends on the Amino Acid Sequence within the A␤ Region, but Not on Speciesspecific Differences in BACE1-APP, a membrane protein, is highly conserved among animals. The cytoplasmic region, with which many conserved cytoplasmic proteins interact to regulate the metabolism and/or functions of APP, is completely conserved in human and mouse, and even in the electric ray (reviewed in Ref. 26; 27, 28). However, the amino acid sequence of the A␤ region differs slightly between mouse and human (Fig. 1A). In mouse APP (mAPP), Arg at position 676 in human APP (hAPP) is Gly (R676G), Tyr 681 is Phe (Y681F), and His 684 is Arg (H684R). These three amino acid substitutions in mouse A␤ are located near the ␤and/or ␤Ј-sites of BACE1 cleavage. Rodent primary cultured neurons generate more A␤(11-XX) than A␤(1-XX) from endogenous APP into conditioned media, which were identified immunoprecipitation MS and radiosequencing analyses (29). When human APP was exogenously expressed in rodent primary cultured neurons, human A␤(1-40) was predominant (30). Furthermore, analysis of A␤42 species in human cerebrospinal fluid (CSF) with a surface-enhanced laser desorption/ionization technology revealed A␤(1-42) as a major form, whereas A␤  was less in quantity (31). This was thought to be a consequence of the species differences in the specificity of BACE1 (30). However, the molecular mechanism by which the ␤or ␤Ј-site is selected by BACE1 remains controversial.
To explore this issue, we first compared the generation of A␤ from mAPP and hAPP (Fig. 1B). To this end, we transiently expressed mAPP-(695) and hAPP-(695) in mouse N2a and human SHSY5Y neuroblastoma cells, and then analyzed the secreted A␤ forms by MALDI-TOF/MS following immunoprecipitation with pan-A␤ antibody 4G8, which recognizes an epitope within the A␤(17-24) region. We identified several A␤ species in which the amino-terminal amino acid was Asp (A␤(1-XX), a product of ␤-site cleavage) or Glu (A␤(11-XX), a product of ␤Ј-site cleavage). For convenience, we focused our investigation of alternative APP cleavage by BACE1 on A␤(1-40) and A␤ . In both human and mouse cells, much more A␤  was generated from hAPP than A␤ . By contrast, A␤  was the predominant form generated from mAPP in both types of cells (Fig. 1C, left). The preferential cleavage of mAPP at the ␤Ј-site was also observed for A␤  and A␤  in mouse cells (Fig. 1C, right), but these were not quantified in human cells because of less amounts. Nontransfected cells did not yield detectable A␤ signals in these MS spectra. These results indicate that mAPP is subject to amyloidolytic rather than amyloidogenic cleavage by both human and mouse BACE1. In other words, the difference between hAPP and mAPP is not due to species-specific differences in BACE1. Instead, the alternative cleavage of APP by BACE1, i.e. the selection of either the ␤or ␤Ј-site, depends on the amino acid sequence within the A␤ region.
This was further confirmed by a in vitro ␤-secretase assay with substrate peptides of human and mouse amino acid sequences (Fig. 2). Human and mouse APP-(662-691) peptides ( Fig. 2A) were incubated with a recombinant human BACE1, and the generated peptides cleaved at the ␤Ј-site (APP-(662-681)) and ␤-site (APP-(672-691)) were analyzed and quantified. In vitro assay, the ␤-site cleavage was predominant in both human and mouse substrates (Fig. 2B, left). However, the ratio of ␤Ј-cleavage to ␤-cleavage (␤Ј/␤) was significantly high in mouse substrate compared with human ( Fig. 2B, right). The result supports the conclusion that the mouse APP sequence is preferentially cleaved at the ␤Ј-site by BACE1 in the cell (Fig. 1), although the magnitude of preferential ␤Ј-site cleavage is slightly different in an in vitro study from a cell study, and we cannot exclude an affect of second cleavage of the first cleaved product.
To confirm this conclusion, we next investigated whether amino acid substitutions within the A␤ sequence would influence alternative selection of BACE1 cleavage sites. We focused on three amino acid positions within mouse A␤, which differs from human A␤ by R676G, Y681F, and H684R (Fig. 1A). hAPP harboring the single mutation R676G, Y681F, or H684R, or the triple mutation R676G/Y681F/ H684R (3mut, i.e. fully recapitulating the mouse sequence) were expressed in N2a cells along with innate hAPP and mAPP, and the secreted forms of A␤ in the medium were analyzed (Fig. 3A). The A␤(11-40)/A␤(1-40) ratio was compared with the ratio of A␤ secreted from cells expressing innate hAPP and mAPP (Fig. 3B). Among the single-substitution mutants, hAPP H684R had a significantly elevated A␤(11-40)/A␤(1-40) ratio; however, the ratio was lower than those of hAPP-3mut and innate mAPP. hAPP R676G showed a tendency an elevated A␤(11-40)/A␤(1-40) ratio, but did not have a significantly higher A␤(11-40)/A␤  ratio. The result indicates that the amino acid at position 684 plays an important role in the alternative selection of BACE1 cleavage sites of hAPP, although we cannot rule out a slight contribution of amino acid at position 676. Changes in amino acids within the A␤ region can increase the balance between amyloidogenic and amyloidolytic cleavage, in other words, the amino acid sequence within the A␤ region can determine the alternative of cleavage sites by BACE1.
The Protective A673T Mutation, but Not the A673V Mutation, within the A␤ Sequence Promotes ␤Ј-Site Cleavage of hAPP by BACE1-FAD-associated mutations of the APP gene include some variations within the A␤ sequence. Substitution of Val for Ala 673 (A673V) is pathogenic, resulting in elevated production of A␤, whereas substitution of Thr at the same position (A673T, the Icelandic mutation) is protective (23,25). However, the molecular mechanism underlying altered A␤ generation remains controversial, especially regarding the protective A673T mutation.
To investigate the role of the A673T mutation, we examined A␤ generation in N2a cells expressing hAPP harboring A673T or A673V along with the wild-type hAPP (WT) and hAPP harboring the Swedish mutation (K670N/M671L), which is predominantly cleaved at the ␤ site (Fig. 4). Levels of APP were identical in cells expressing each variant except for the Swedish mutation, which is susceptible to ␤-site cleavage (Fig. 4A). The CTF␤/C99 level of hAPP A673T was very similar to that of hAPP (WT); however, as reported (25), generation of A␤(1-40) decreased significantly, and the level of A␤(1-42) tended to decrease (Fig. 4B). The Swedish mutation, used as a positive pathogenic control, markedly promoted formation of both CTF␤/C99 and A␤(1-40) and A␤(1-42) compared with hAPP (WT), consistent with earlier work (32-34). As expected from a previous report (35), the A673V mutation was pathogenic, with elevated CTF␤/C99 and A␤(1-40) generation, although the slight increase in A␤(1-42) generation was not significant.
Next, we analyzed the alternative cleavage of hAPP harboring A673T and A673V by BACE1 (Fig. 4C). Interestingly, the ␤Ј-site cleavage product A␤  was present at significantly higher levels in cells expressing hAPP A673T: A␤(11-40) production was 2.5-fold higher, and the A␤(11-40)/A␤(1-40) ratio was 3-fold higher, than in cells expressing hAPP (WT) (Fig. 4D). By contrast, cells expressing hAPP A673V significantly decreased A␤  production, as did those expressing the Swedish mutation; in both cases, the A␤(11-40)/A␤(1-40) ratio was significantly lower than that of hAPP (WT). This result clearly shows that the protective effect of hAPP A673T is due to elevated ␤Ј-site cleavage of hAPP, resulting in higher levels of A␤(11-XX), rather than to the metabolic lability of A␤(1-XX) containing the A673T mutation. A673V is pathogenic, but the detailed mechanism underlying elevated A␤ generation was unclear (23). In this analysis, we clarified that the pathogenic effect of A673V is mediated by a decrease in ␤Ј-site cleavage of hAPP along with an increase in ␤-site cleavage, resulting in significantly higher levels of A␤(1-40) and a slight increase in the level of A␤ .
Elevated BACE1 Activity Cleaves CTF␤/C99, the Precursor of A␤(1-XX), to Generate A␤(11-XX)-BACE1 activity generates the amyloidogenic fragment CTF␤/C99 from hAPP. This ␤-site cleavage of hAPP is facilitated by pathogenic mutations that attenuate ␤Ј cleavage of hAPP. Because BACE1 can use CTF␤/ C99 as well as APP as substrates (12), we investigated whether elevated BACE1 activity can cleave CTF␤/C99 at the ␤Ј-site subsequent to the primary cleavage. BACE1 overexpression was previously reported to decrease brain A␤ levels in vivo (36), but the mechanism remains controversial.
For these experiments, we expressed hAPP and CTF␤/C99 in N2a cells with or without BACE1. Protein levels of hAPP, CTF␤/C99, and BACE1 were confirmed by Western blotting (Fig. 5A). hAPP can exist in both mature (N-and O-glycosylated) and immature (N-glycosylated) forms (reviewed in Ref. 26). Overexpression of BACE1 increases formation of CTF␤/ C99 and decreases the proportion of mature hAPP (Fig. 5B).

FIGURE 2. Cleavage of peptides including the human and mouse A␤ sequence by human BACE1 in vitro.
A, amino acid sequences of substrate human peptide and peptides cleaved at ␤and ␤Ј-site by BACE1. When substrate APP-(662-691) is cleaved at ␤-site or ␤Ј-site, the generated APP-(672-691) and APP-(662-681) are shown, respectively. B, representative MS spectra of products cleaved at ␤and ␤Ј-sites of human and mouse substrates as indicated in panel A. Spectra of human (upper) and mouse (lower) substrates were shown (␤, peptide cleaved at ␤-site; ␤Ј, peptide cleaved at ␤Ј-site). Levels of the peptide cleaved at the ␤Ј-site are indicated as the ratio to that of the peptide cleaved at ␤-site, a ␤Ј/␤ ratio, and compared between human and mouse substrate. Data are shown as mean Ϯ S.E. Statistical significance was examined by Student's t test (n ϭ 4), and p values are indicated (**, p Ͻ 0.01).
Because only mature APP is a bona fide substrate of APP secretases (37), almost all mature hAPP is subject to cleavage upon overexpression of BACE1, along with less cleavage by endogenous APP ␣-secretase. Overexpression of BACE1 decreases the level of mature APP by disturbing axonal transport in the brain in vivo (36); however, based on our observations in undifferentiated N2a cells (Fig. 5A), it is reasonable to speculate that the decrease in the level of mature hAPP is due to proteolysis by overexpressed BACE1 along with elevated generation of CTF␤/ C99 and CTF␤Ј/C89.
As expected based on a previous report (12), overexpression of BACE1 significantly increased the level of C89/CTF␤Ј (Fig. 5,  A and B). We speculated that CTF␤/C99 might be further cleaved at the ␤Ј-site due to the elevated activity of overexpressed BACE1. We tested this possibility in cells expressing C99/CTF␤ with BACE1. Cells overexpressing BACE1 contained lower levels of CTF␤/C99 but dramatically higher levels of CTF␤Ј/C89 (Fig. 5C), indicating that CTF␤/C99 was cleaved again at the ␤Ј-site by elevated BACE1 activity, as reported (12). To confirm this conclusion, we analyzed A␤ species secreted from cells that did or did not overexpress BACE1 (Fig. 5D). In medium from cells expressing hAPP without BACE1 overexpression, A␤(1-40) was again the major form, accompanied by a small amount of A␤ . By contrast, in cells overexpressing BACE1, A␤(1-40) generation was remarkably reduced and a great deal of A␤  was detected, consistent with the results of Western blot analysis (Fig. 5D). In the medium of cells expressing CTF␤/C99, overexpression of BACE1 remarkably decreased A␤(1-40) generation, but increased the generation of A␤   (Fig. 5D). The elevated A␤(11-40)/A␤(1-40) ratio in cells with BACE1 overexpression was statistically significant in cells expressed by either hAPP or CTF␤/C99 (Fig.  5D, right). These results clearly showed that CTF␤/C99, the product of hAPP primarily cleaved at the ␤-site by BACE1, can be further cleaved at the ␤Ј-site by excess BACE1 activity. These data suggest that excess BACE1 activity functions amyloidolytic instead of amyloidogenic, at least under our experimental conditions, rather than by decreasing the levels of its substrates (mature APP and CTF␤/C99) by altering the intracellular APP transport system in in vivo neurons (36).
Next, we performed similar studies on hAPP harboring pathogenic mutations (Fig. 6). Again, N2a cells expressed hAPP in the presence or absence of BACE1 overexpression. Expression of BACE1 decreased APP, irrespective of the presence of a FAD mutation (Fig. 6, A and B), as shown in Fig. 5. Cells expressing hAPP with FAD mutations expressed significantly (Swedish; K670N/M671L) or slightly without a statistically significance (Leuven; E682K) higher level of CTF␤/C99 compared with those expressing hAPP (WT) (Fig. 6C, left). Moreover, overexpression of BACE1 remarkably increased the level of CTF␤Ј/ C89 in cells expressing hAPP, irrespective of the presence of a FAD mutation (Fig. 6C, middle). We also examined A␤ species in the medium (Fig. 6D). When BACE1 was overexpressed, the ratio of A␤(1-40) to A␤  significantly decreased, whereas the A␤(11-40)/A␤(1-40) ratio increased significantly, in cells expressing hAPP (WT) and hAPP (Swedish) (Fig.  6D, right). In cells expressing hAPP (E682K), the level of A␤(1-40) was greatly decreased by BACE1 overexpression, as in the case of hAPP (WT) and hAPP (Swedish); unexpectedly, however, no increase in A␤  was detected (Fig. 6D). Position 11 of the A␤ sequence is Glu 682 , the site of the Leuven mutation. CTF␤Ј/C89 with substitution of Lys for Glu at A␤ position 11 (amino terminus of A␤(11-XX)) may be preferentially cleaved by ␣-secretase to generate CTF␣/C83. In other words, A␤  harboring the Leuven E682K mutation is more likely to be cleaved by ␣-secretase; therefore, no increase in the level of A␤  was detected in the medium of cells expressing hAPP E682K, in contrast to cells expressing hAPP (WT) and hAPP (Swedish), which have Glu at position 682 (Fig. 6D). The preferential cleavage of CTF␤Ј/C89 (including E682K) by ␣-secretase was reflected by the significantly higher level of CTF␣/C83 in cells overexpressing BACE1 (Fig. 6C, right). Overexpression of BACE1 decreased the level of CTF␣/C83, significantly in hAPP (Swedish) and with a trend of decrease in  hAPP (WT) (Fig. 6C, right), because mature APP was predominantly cleaved by overexpressed BACE1 but not by ␣-secretase (Figs. 5A and 6A). Subsequently, the resultant CTF␤Ј/C89 was further cleaved by ␥-secretase to yield A␤(11-XX). By contrast, the level of CTF␣/C83 derived from hAPP (E682K) was equivalent to that in cells overexpressing BACE1 (Fig. 6C, right). Based on these observations, it is reasonable to explain that the higher level of CTF␣/C83 derived from hAPP (E682K) in BACE1-overexpressing cells was due to ␣-site cleavage of CTF␤Ј/C89. BACE1 can cleave the peptide bond between hAPP Leu 705 (position 34 in A␤) and Met 706 (position 35 in A␤) (38). Accordingly, as shown in Figs. 5D and 6D, cells overexpressing BACE1 contained higher levels of A␤(1-34) and A␤  fragments. Overexpression of BACE1 promoted cleavage at this position to increase production of A␤(1-34) and A␤ . Again, A␤  was not detectable in medium of cells expressing hAPP E682K (Fig. 6D). We speculate that, like A␤ , A␤(11-34) harboring E682K was cleaved preferentially by ␣-secretase. Therefore, the excess activity of BACE1 can function amyloidolytic, at least in FIGURE 6. Elevated BACE1 activity cleaves hAPP at the ␤-site to decrease the level of A␤(1-XX), even if hAPP harbors FAD mutations. A, protein levels of APP, APP CTFs, and BACE1 in cells. N2a cells expressed hAPP with or without FAD mutation in the presence (ϩ) or absence (Ϫ) of BACE1-FLAG expression. Cell lysates (10 g of protein) were analyzed by Western blotting with anti-APP, anti-FLAG, and anti-actin antibodies. hAPP (first row), CTF␤/C99, CTF␤Ј/C89, and CTF␣/C83 (second row), BACE1 (third row), and actin (fourth row) are indicated. Numbers indicate molecular size markers (in kDa). B, the levels of APP are indicated as ratios relative to the levels of hAPP (WT) without BACE1 expression, which was assigned a reference value 1.0. Statistical analysis was performed by Student's t test. p values are provided for the comparison with the APP without (Ϫ) BACE1 expression (mean Ϯ S.E., n ϭ 6; **, p Ͻ 0.001; ****, p Ͻ 0.0001). C, the levels of APP CTFs are indicated as ratios relative to the levels of hAPP without BACE1 expression, which was assigned a reference value 1.0. Statistical analysis was performed by Student's t test. p values are provided for comparison with the CTF without (Ϫ) BACE1 expression (mean Ϯ S.E., n ϭ 6; *, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001). In another case, statistical analysis was performed using Dunnett's multiple comparison test. p values are provided for comparison with the hAPP wild-type (WT) (mean Ϯ S.E., n ϭ 6; ##, p Ͻ 0.001; ###, p Ͻ 0.001). D, representative MS spectra of A␤ with amino-terminal ␤and ␤Ј-cleavage sites generated from innate hAPP (WT) or hAPP harboring the Swedish (middle) or E682K FAD mutation. Culture media of cells expressing wild-type hAPP (upper), hAPP Swedish (middle), and hAPP E682K (lower) in the presence (right) and absence cell studies, even in APP harboring pathogenic variations such as the Swedish and Leuven mutations.
In Mouse, A␤  Is Less Abundant Than A␤  in Vivo-We showed above that mAPP is predominantly cleaved at the ␤Ј-site in cells, resulting in preferential generation of A␤(11-XX) rather than A␤(1-XX) as observed typically in A␤  and A␤(1-40) species (Fig. 1). To extend this finding, we asked whether A␤  is also predominantly generated from endogenous APP in mouse neurons in vivo (Fig. 7). Using IP-MS, we analyzed the A␤ forms present in the medium of mouse primary mixed cultured neurons (cerebrum cortex plus hippocampus) and in the adult mouse CSF (mixture of 3 individuals 6 months old), as well as those in the medium of N2a cells overexpressing mAPP. As shown in Fig. 1, N2a cells again generated more A␤(11-40) than A␤(1-40). Mouse neurons also expressed A␤  as the major form, as reported previously (29). Unexpectedly, mouse CSF included A␤(1-40) as the major form, and the level of A␤  was remarkably lower (Fig. 7A). The A␤(11-40)/A␤(1-40) ratio was clearly lower in CSF when compared with those generated N2a cells and primary cultured neurons. Although the analysis is a duplicate study without statistical analysis, and we cannot rule out a possibility that the A␤ species generated from adult neurons may differ from those secreted from embryonic cultured neurons, these data suggest that mouse neurons predominantly generate A␤(11-XX) rather than A␤(1-XX), but that A␤(11-XX) is metabolically labile in vivo and subject to degradation or faster clearance than A␤(1-XX).
To explore this possibility that A␤  is less stable than A␤  in vivo, A␤(1-40) and A␤  were incubated in vitro with neprilysin, which is a major extracellular A␤-degrading enzyme (39), and lability of A␤ peptides was analyzed (Fig.  8). In the presence of neprilysin, the level of A␤(11-40) lowered quickly and significantly (by 40% at 1 h and 20% at 2 h), whereas the levels of A␤(1-40) remained over 50% during incubation time. Although the amounts of both peptides tended to decrease in the absence of enzyme, the results were standardized with values of reaction in the absence of neprilysin.
Although it may be difficult to exclude other possibilities of the lower level of A␤  in CSF (Fig. 7), this in vitro study supports that A␤  is less stable rather than A␤ , at least in degradation by neprilysin. In other words, the A␤ species generated by ␤Ј-site cleavage are subject to more rapid clearance than those generated by ␤-site cleavage.

Discussion
It is widely accepted that ␤-secretase BACE1 is the primary APP-cleaving enzyme responsible for generation of the A␤(1-  XX) species, and that the combined cleavage of APP by BACE1 and the ␥-secretase complex generates multiple types of A␤ species with distinct carboxyl termini, e.g. A␤(1-40) and A␤(1-42) (reviewed in Refs. 8 and 40). Previous work showed that BACE1 also cleaves APP at the ␤Ј-site, but the biological significance of this reaction remained unclear because the ␤Ј-cleaved products of hAPP, CTF␤Ј/C89, and A␤(11-XX) are scarce relative to CTF␤/C99 and A␤(1-XX) (12,41). Along with previously reported data, our results confirm that mAPP is preferentially cleaved at the ␤Ј-site within the A␤ sequence and predominantly generates A␤(11-XX); by contrast, hAPP mostly generates A␤(1-XX) in vivo and in vitro (29 -31).
Over the course of this study, we made several discoveries regarding the role of BACE1 in APP metabolism. First, we showed that alternative selection of cleavage sites by BACE1 in both human and mouse is determined by the amino acid at position 684 (His in human, Arg in mouse). This suggests that some FAD-associated pathogenic and/or protective mutations within the A␤ region of the APP gene influence the selection of cleavage sites by BACE1. Second, we showed that the pathogenic mutation A673V decreased ␤Ј-site cleavage of hAPP, whereas the protective Icelandic mutation A673T increased ␤Ј-site cleavage. This observation strongly indicates that amino acid substitutions at position 673 influenced the alternative selectivity of the cleavage site by BACE1. In other words, A673V induces the pathogenic ␤-site cleavage of APP by BACE1, whereas A673T raises the protective ␤Ј-site cleavage of APP. This is the primary cause of FAD harboring APP A673V, and this is the primary effect to protect AD in subjects harboring APP A673T. Third, we demonstrated that elevation of BACE1 activity due to overexpression of the enzyme results in secondary cleavage of CTF␤/C99 at the ␤Ј-site, dramatically increasing A␤(11-XX) production. This is true, at least in the cell study, even for APP harboring the Swedish mutation, which predominantly produces A␤(1-XX), although a pilot study in a BACE-overexpressing transgenic mouse suggested that reduction in A␤ deposition is mediated by another mechanism (36). Finally, we showed that A␤(11-XX) may be more metabolically labile than A␤(1-XX) in vivo and in vitro.
Thus, contrary to the popular view, our findings may suggest that activation of BACE1 in AD subjects (including FAD patients), using procedures other than overexpression of BACE1, represents a promising target for AD therapies aimed at decreasing the level of A␤. Development of inhibitors of BACE1 and/or ␥-secretase activities has been pursued since these enzymes were identified (reviewed in Ref. 42). However, we believe that more attention should have been devoted to ␥-secretase inhibitors. It is reasonable to predict that attenuation of ␥-secretase activity by an inhibitor would decrease generation of neurotoxic A␤ generation. In contrast to this early idea, recent progress in understanding the molecular mechanism by which ␥-secretase cleaves APP suggested to us that attenuating ␥-secretase activity would result in reduced production of A␤  and elevated production of A␤ , which is a precursor of A␤  and is more neurotoxic than A␤  and A␤(1-40) (43) (reviewed in Ref. 44).
Our results suggest that ␤-secretase inhibitors may face some of the same problems as ␥-secretase inhibitors. Suppres-sion of APP cleavage using a BACE1 inhibitor is a potential therapeutic strategy for decreasing A␤ generation. However, many other substrates of BACE1 have been reported to date, suggesting that BACE1 inhibition could have considerable side effects. Nevertheless, suppression of BACE1 activity specifically in the brain may be a practical means for treating AD patients (reviewed in Ref. 42). Our results show that cleavage of APP by excess BACE1 activity can degrade A␤ by cleaving APP at the ␤Ј-site. Moreover, if CTF␤/C99 is abundant in FAD subjects, sufficient levels of BACE1 activity could cleave CTF␤/C99 again at the ␤Ј-site to generate A␤(11-XX), which is more metabolically labile. Consequently, the total amount of A␤ in the brain would decrease, as demonstrated in BACE1 transgenic mice (36). Because A␤(1-34) is degraded by neprilysin, an A␤-degrading enzyme, faster than A␤(1-40) (45), A␤(11-XX) may also be degraded more easily by such enzymes than A␤(1-XX) (46). At least we showed that A␤  was degraded quickly rather than A␤  in a study in vitro. Therefore, to treat AD patients (e.g. FAD subjects harboring the Swedish mutation), activation of BACE1 might be a more effective therapy than administration of a BACE1 inhibitor. Moreover, in contrast to development of ␥-secretase modulators that enhance the peptidase-like activity of ␥-secretase to reduce A␤(1-42) generation while promoting formation of A␤ , it may be difficult to develop a drug that modifies the selectivity of BACE1 because selection of a cleavage site (i.e. the ␤or ␤Ј-site) depends on the sequence of the substrate A␤ domain. However, we cannot rule out other possibilities that the intracellular environment of BACE1 can influence in the selectivity of cleavage site of APP by BACE1.
In general, it is possible that both BACE1 and ␥-secretase activities may decrease with age, although at least one report showed that the BACE1 level is elevated in AD (47). Reduction of BACE1 activity may attenuate the amyloidolytic ␤Ј-site cleavage of APP relative to amyloidogenic ␤-site cleavage, resulting in the generation of A␤(1-XX). Attenuation of ␥-secretase activity, especially its carboxypeptidase-like activity, promotes production of A␤(XX-42) at the expense of A␤(XX-38). Therefore, the combination of altering and/or weakening the activities of both BACE1 and ␥-secretase increases production of the most neurotoxic species, A␤ , whereas decreasing production of A␤(11-XX) and A␤(XX-38) species. Based on our understanding of the mechanisms of APP cleavage by BACE1 and ␥-secretase, administration of compounds to regulate BACE1 and ␥-secretase activities to AD subjects should proceed with scrupulous caution.
Immunoprecipitation MALDI-TOF/MS (IP-MS) and -MS/MS Analysis and Quantitation of A␤ Species-A␤ peptides secreted into the medium (1 ml) and mixed mouse CSF (C57BL/6, 100 l) were recovered by immunoprecipitation with anti-pan A␤ antibody 4G8, which was raised to A␤ 17-24 epitope (BioLegend, San Diego, CA) plus Protein G-Sepharose 4B (GE Healthcare Bio-Sciences) following pre-clearing with the same beads without antibody. The beads were sequentially washed, and the bound proteins were eluted with trifluoroacetic acid/acetonitrile/water (1:20: 20) saturated with sinapinic acid as described (51). The dissolved samples were dried on a target plate, and matrix-assisted laser desorption ionization time-of-flight tandem mass spectrometry (MALDI-TOF/MS) analysis was performed using an UltraflexII TOF/TOF (Bruker Daltonics, Bremen, Germany). MS/MS analysis was performed to confirm the amino acid sequence of A␤  and A␤(1-40) (supplemental Fig. S1). The quantitative accuracy of mass spectrometric analysis was confirmed by studies with a mixture of synthetic human A␤(1-40) and A␤  peptides (supplemental Fig. S2) with a procedure described previously (51,52). Molecular masses were calibrated using a peptide calibration standard (Bruker Daltonics). Secretion of A␤40 and A␤42 into the medium was quantitated by sandwich ELISA (sELISA) as described previously, except that biotinylated 6E10 antibody (catalog number 803003; BioLegend), which was raised A␤(1-16), was used instead of the 2D1 antibody, which was raised to A␤(1-27) (37).
Studies with mice were conducted in compliance with the guidelines of the Animal Studies Committee of Hokkaido University. The C57BL/6 mice used in this study were housed in a specific pathogen-free environment with MicroVent units (Allentown Inc., Allentown, NJ) throughout the study period.
Mouse Primary Cultured Neurons-The primary culture of mixed mouse cortical and hippocampal neurons was performed with a modified version of the method described by Bartlet and Banker (54). C57BL/6 wild-type were used as described (55). In brief, the cortex and hippocampus of mice at embryonic day 15.5 were dissected, and neurons were spread in a buffer containing papain and cultured at 5 ϫ 10 4 cells/cm 2 in a medium composed of Neurobasal Medium (Life Technologies) containing 30% Nerve-Cell Culture Medium (DS Pharma Biomedical), 2% B-27 supplement (Thermo Fisher Scientific), Glutamax I (4 mM), 5% heat-inactivated horse serum (Thermo Fisher Scientific), and antibiotics (Thermo Fisher Scientific) on poly-D-lysine-treated plates for 5 days. All animal studies were conducted in compliance with the guidelines of the Animal Studies Committee of Hokkaido University. Mice were housed in a specific pathogen-free environment.
Degradation of A␤  and A␤  by Neprilysin in Vitro-A␤  and A␤(11-40) (15 ng; GeneScript, Piscataway, NJ) were incubated in a buffer (50 mM Tris-HCl, pH 7.5, including 0.4% bovine serum albumin) with or without 1.5 ng of recombinant Neprilysin (catalog number 1182-ZNC; R&D Systems) at 37°C for the indicated times. Peptides in the reaction tube were analyzed by Western blotting with anti-pan-A␤ antibody 4G8.
Statistical Analysis-Statistical analyses were performed with GraphPad Prism (GraphPad Software, San Diego, CA). The threshold for significance is indicated with p value.
Author Contributions-A. K., S. H., and T. S. participated in the design of study, and A. K. and S. H. carried out all studies. T. S. conceived the study, and T. S. and S. H. wrote the paper. All authors read and approved the final manuscript.