γ-Cleavage Is Dependent on ζ-Cleavage during the Proteolytic Processing of Amyloid Precursor Protein within Its Transmembrane Domain*

β-Amyloid precursor protein apparently undergoes at least three major cleavages, γ-, ϵ-, and the newly identified ζ-cleavage, within its transmembrane domain to produce secreted β-amyloid protein (Aβ). However, the roles of ϵ- and ζ-cleavages in the formation of secreted Aβ and the relationship among these three cleavages, namely ϵ-, ζ-, and γ-cleavages, remain elusive. We investigated these issues by attempting to determine the formation and turnover of the intermediate products generated by these cleavages, in the presence or absence of known γ-secretase inhibitors. By using a differential inhibition strategy, our data demonstrate that Aβ46 is an intermediate precursor of secreted Aβ. Our co-immunoprecipitation data also reveal that, as an intermediate, Aβ46 is tightly associated with presenilin in intact cells. Furthermore, we identified a long Aβ species that is most likely the long sought after intermediate product, Aβ49, generated by ϵ-cleavage, and this Aβ49 is further processed by ζ- and γ-cleavages to generate Aβ46 and ultimately the secreted Aβ40/42. More interestingly, our data demonstrate that γ-cleavage not only occurs last but also depends on ζ-cleavage occurring prior to it, indicating that ζ-cleavage is crucial for the formation of secreted Aβ. Thus, we conclude that the C terminus of secreted Aβ is most likely generated by a series of sequential cleavages, namely first ϵ-cleavage which is then followed by ζ- and γ-cleavages, and that Aβ46 produced by ζ-cleavage is the precursor of secreted Aβ40/42.

The mechanism of the formation of the ␤-amyloid protein (A␤) 2 is the central issue in Alzheimer disease research, not only because A␤ is the major constituent of senile plaques, one of the neuropathological hallmarks of Alzheimer disease, but also because A␤ formation may be a causative event in the disease (1). A␤ is proteolytically derived from a large single transmembrane protein, the ␤-amyloid precursor protein (APP), as a result of sequential cleavages by ␤and ␥-secretases (1). ␤-Secretase has been identified as a type I membrane aspartyl protease (2,3). Although the exact nature of ␥-secretase is still a matter of debate, accumulating evidence supports the idea that ␥-secretase is a multiple molecular complex composed of, at least, presenilins, nicastrin, Aph-1, and Pen-2 and that presenilin may function as the catalytic subunit (4).
In understanding the mechanism by which the C termini of secreted A␤ are generated during the processing of APP, three major intramembranous cleavages have been established. The first one is the cleavage now specifically referred to as ␥-cleavage (5), which produces the C termini of most of the secreted A␤ species that end at amino acids 40 (A␤40) or 42 (A␤42) of the A␤ sequence. The second one is the ⑀-cleavage occurring between A␤ residues 49 and 50, which produces the N terminus of most of the APP intracellular domain (AICD) (5)(6)(7)(8). The identification of this ⑀-cleavage site raises a question as to whether this ⑀-cleavage is obligatory for the generation of the C terminus of A␤, and this also raises a question as to the relationship between ⑀and ␥-cleavages, i.e. whether they are independent of each other or sequential. One of the obstacles in addressing these questions is that neither the intermediate A␤ peptide, which ends at the ⑀-cleavage site, nor the C-terminal fragment, which starts with an N terminus generated by ␥-cleavage, has ever been detected. In a recent study, we reported the identification of an intracellular long A␤ species, namely A␤ 46 , and this led to the discovery of the third major cleavage site, the -cleavage site at A␤ 46 between the known ␥and ⑀-cleavage sites (9). The presence of -cleavage site at A␤46 is further supported by a very recent study showing that A␤ 46 is the predominant form among the longer A␤ species detected intracellularly (10). However, the finding that the known ␥-secretase inhibitors, such as DAPT, DAPM, and compound E, inhibit the formation of secreted A␤ 40/42 , and on the other hand cause the accumulation of A␤ 46 , raises the question as to whether A␤ 40/42 and A␤ 46 are produced by the same enzyme or by different enzymes (9). Moreover, the roles of ⑀and -cleavages in the formation of secreted A␤ and the relationship among these three cleavages, namely ⑀-, -, and ␥-cleavages, also remain elusive. To address these key issues, the objectives of this study were focused on the following: (a) determining precursor and product relationship between A␤ 46 and A␤ 40/42 ; and (b) establishing the roles of ⑀and -cleavages in the formation of secreted A␤ 40/42 .
Cell Lines and Plasmids-N2a cells stably expressing either wild type presenilin 1 (PS1wt) alone or both PS1wt and Swedish mutant APP (APPsw) were kindly provided by Drs. Sangram S. Sisodia and Seong-Hun Kim (University of Chicago) and were maintained as described previously (11). The plasmid APPsw645, which expresses a C-terminal truncated APP ending at the ⑀-cleavage site A␤ 49 , was constructed using the site-directed mutagenesis kit (Stratagene). APPsw (12), kindly provided by Dr. Gopal Thinakaran (University of Chicago), was used as a template. A pair of oligonucleotides (E49, CGTCATCACCTTGTA-GATGCTGAAGAAG; E49-r, CTTCTTCAGCATCTACAAGGT-GATGACG), which are complementary to each other and contain a stop codon at position 50 of the A␤ sequence, were used as primers.
Cell-free Assay-In vitro turnover of A␤ 46 by ␥-secretase activity was assayed in a cell-free assay system established previously (13), following the procedure described previously (7) with minor modifications. Briefly, N2a cells were cultured in the presence of DAPM for 12 h and harvested in 9 volumes of homogenization buffer (10 mM MOPS, pH 7.0, 10 mM KCl) containing protease inhibitors (Complete, Roche Applied Science) and homogenized by passing through a 20-gauge needle 30 times. After removal of unbroken cells and nuclei by centrifugation at 800 ϫ g at 4°C for 10 min, membranes were pelleted by centrifugation at 20,000 ϫ g at 4°C for 30 min. The membranes were washed once with homogenization buffer and resuspended in assay buffer (150 mM sodium citrate pH 6.4, protease inhibitor mixture). Aliquots of equal amounts of membranes were then incubated at either 0 or 37°C. After 1 h of incubation, aliquots (25 l) were removed for Western blotting, and the remaining reaction mixtures were subjected to centrifugation at 20,000 ϫ g for 30 min at 4°C to yield the supernatant and pellet fractions. After addition of an equal volume of IP buffer (50 mM Tris/HCl, pH 7.4, 150 mM NaCl, 0.5% Nonidet P-40, 5 mM EDTA, and protease inhibitor mixture), the supernatant was subjected to immunoprecipitation using 6E10. The pellet fraction was solubilized with 1% Nonidet P-40 in IP buffer. After centrifugation at 20,000 ϫ g at 4°C for 15 min, the supernatant was diluted with equal amounts of IP buffer to lower the concentration of Nonidet P-40 to 0.5% and then subjected to immunoprecipitation using 6E10. The intracellular A␤ species were immunoprecipitated using 6E10. Both immunoprecipitates were analyzed by 10% Bicine/urea-SDS-PAGE, followed by Western blot analysis using 6E10 as described below.
Detection of A␤ 49 -Note that in all of the experiments throughout this study, A␤ 46 was determined by directly analyzing the cell lysates without immunoprecipitation. To determine the presence of the possible A␤ 49 , cells were cultured in the absence of any inhibitors and lysed with 1% Nonidet P-40 in IP buffer. After centrifugation at 20,000 ϫ g at 4°C for 15 min, the supernatant was diluted with an equal amount of IP buffer, and the A␤ 49 and other intracellular A␤ species were immunoprecipitated using 6E10 in the presence or absence of DAPT. Of note, based on our previous data (9) and unpublished data, 3 it was found that all the tested nontransition state inhibitors, such as DAPT, DAPM, and compound E, cause intracellular accumulation of A␤ 46 in the same fashion. However, in comparison with DAPM, the inhibitory effects of DAPT and compound E last longer, and this is probably because the enzyme binding activity of the latter two is stronger than that of DAPM. Therefore, DAPT and compound E were used in the in vitro assay and during the immunoprecipitation procedure. DAPM was used to cause the accumulation of A␤ 46 in cells that would be used for determining the turnover of A␤ 46 either in intact cells or in a cell-free system in which the inhibitor used for causing the accumulation of A␤ 46 needs to be removed.
Immunoprecipitation and Western Blotting-Immunoprecipitation and Western blotting were carried out as described previously (9) with the exception that in some cases the immunoprecipitation was carried out in the presence of 500 nM DAPT as indicated in the figure legends. Briefly, 24 h after splitting, cells were treated with inhibitors at various concentrations or with vehicle only as a control. Eight hours after treatment, cells were harvested and lysed in Western blot lysis buffer (50 mM Tris-HCl, pH 6.8, 8 M urea, 5% ␤-mercaptoethanol, 2% SDS, and protease inhibitors). Secreted A␤ was immunoprecipitated from conditioned media using a monoclonal A␤-specific antibody 6E10 (Senetek). The immunoprecipitates were analyzed by 10% Bicine/urea-SDS-PAGE and transferred to a polyvinylidene fluoride membrane (Immobilon-P, Millipore). The membranes were then probed with 6E10, and the immunoreactivity bands were visualized using ECL-Plus (Amersham Biosciences).
Fractionation and Co-immunoprecipitation-In order to determine the formation of the possible complex of A␤ 46 and presenilin, the following procedure, which was originally described in a previous study (14), was employed with slight modification. Briefly, N2a cells expressing APPsw695/PS1wt cultured in the presence of 3 nM compound E (or 500 nM L-685,458; see Fig. 4B) for 10 -12 h were harvested and then homogenized in homogenization buffer A (20 mM HEPES, pH 7.4, 50 mM KCl, 2 mM EGTA, 10% glycerol, protease inhibitor mixture (Roche Applied Science)) containing 10 nM compound E (or 2.5 M L-685,458) by passing through a 20-gauge needle 30 times. The homogenized samples were subjected to centrifugation at 800 ϫ g for 10 min to remove the unbroken cells and nuclei. The postnuclear supernatant was further centrifuged at 20,000 ϫ g for 1 h resulting in the supernatant and the pellet fractions. The resultant pellet, which contains both A␤ 46 and PS1 (Fig. 4A), was solubilized in buffer B (50 mM PIPES, pH 7.0, 150 mM KCl, 5 mM MgCl 2 , 5 mM CaCl 2 , and protease inhibitor mixture)) (15) containing 1% CHAPSO and 10 nM compound E (or 2.5 M L-685,458), for 1 h at 4°C and then subjected to centrifugation again at 20,000 ϫ g for 25 min to remove the insoluble materials. The supernatant was diluted with an equal volume of solubilization buffer B to adjust CHAPSO to a final concentration of 0.5%. After pre-clearing with protein A-Sepharose beads for 3 h, the supernatant was incubated with anti-PS1N, a rabbit polyclonal antibody raised against the N terminus of PS1 (9) in the presence of compound E (or L-685,458) with rotation at 4°C for 3-4 h, and then an appropriate amount of protein A-Sepharose beads was added and incubated overnight. After washing twice with solubilization buffer B containing 0.5% CHAPSO and ␥-secretase inhibitors, and then twice with PBS, the immunocomplex was eluted with SDS-PAGE Sample loading buffer and separated by 10 -18% SDS-PAGE followed by Western blotting using 6E10 to detect the co-immunoprecipitated A␤ 46 and CTF␤.

RESULTS
L-685,458 Inhibits the Formation of A␤ 46 -In our recent study, we have shown that treatment of cells with nontransition state ␥-secretase inhibitors, such as DAPT, DAPM, and compound E, caused an increase in the accumulation of intracellular A␤ 46 , indicating that these inhibitors have no effect, or little effect, on the newly identified -cleavage. On the other hand, when the cells were cultured in the presence of transition state analogs, such as L-685,458 and 31C, A␤ 46 was not detectable, strongly suggesting that these inhibitors inhibit the -cleavage and block the formation of A␤ 46 (9). However, it cannot be ruled out that the absence of A␤ 46 in cells treated with L-685,458 may be due to the inability of this inhibitor to block the turnover of A␤ 46 . To address these issues, N2a cells expressing both PS1wt and APPsw were treated with DAPT, compound E, and L-685,458, either individually or in combination. Both the cell lysate and the secreted A␤ 40/42 immunoprecipitated from conditioned medium (CM) were analyzed by 10% Bicine/urea-SDS-PAGE as described previously (16), followed by Western blotting using 6E10. As shown in Fig. 1 16 -19, shows the dose-dependent effects of compound E and DAPM on the reduction of secreted A␤ 40/42 and the concomitant accumulation of intracellular A␤ 46 , respectively. Of note, by directly analyzing the cell lysate, A␤ 46 can also be detected in cells not treated with any inhibitor after prolonged exposure of the Western blot ( Fig. 1, lane 11), as has been shown in our recent study (9). On the other hand, treatment with L-685,458, at a range of concentrations from 0.5 to 2.5 M, completely abolished the formation of secreted A␤ 40/42 in the CM (Fig. 1, lanes 4 -6, lower panel), whereas it did not cause the accumulation of intracellular A␤ 46 (upper panel). To determine whether the absence of A␤ 46 was a result of the failure of L-685,458 to block the turnover of A␤ 46 , cells were treated with L-685,458 plus 0.5 M DAPT or plus 5 nM compound E, both of which have been shown to block the turnover of A␤ 46 (9), see also Fig. 1, lanes 2 and 3. As shown in Fig. 1, the addition of DAPT (lanes 7 and 8) or compound E (lanes 9 and 10) did not lead to the accumulation of A␤ 46 in the presence of L-685,458. This result clearly indicates that the absence of A␤ 46 , in cells treated with L-685,458, is due solely to its inhibition of the formation of A␤ 46 , rather than its failure to block the turnover of A␤ 46 .
A␤ 46 Is Processed into A␤ 40/42 in Vitro-As reported in our recent study (9), at a low range of concentrations, DAPM, DAPT, and compound E cause a dose-dependent decrease in secreted A␤ 40/42 and a concomitant increase in intracellular A␤ 46 (see also Fig. 1), suggesting a possible precursor-product relationship between A␤ 46 and A␤ 40/42 . This hypothesis is also supported by the facts that A␤ 46 contains the ␥-cleavage site at A␤40/42 and that A␤ 46 is detectable in living cells in the absence of any inhibitors (Fig. 1, lanes 1 and 11), which suggests that -cleavage occurs prior to ␥-cleavage, otherwise the -cleavage product A␤ 46 would not have had a chance of being formed. To explore the possible precursor-product relationship between A␤ 46 and A␤ 40/42 , we first determined whether A␤ 46 is processed into A␤ 40/42 . To address this issue, a system that contains pre-existing A␤ 46 is required. For this purpose, a cell-free system, which has been established and used in many previous studies to assay the in vitro ␥-secretase activity (7, 13), was employed. Cells were cultured in the presence of 100 nM DAPM, which has been shown to cause the accumulation of A␤ 46 (Fig. 1), and the membranes were prepared as described under "Materials and Methods." As shown in Fig  Membrane preparation and the cell-free assay were performed as described under "Materials and Methods." After incubation in the absence or presence of inhibitors, the reaction mixtures were subjected to the centrifugation, and the resulting supernatants (lanes 1-4) and pellets (lanes 5-8) were then subjected to immunoprecipitation using 6E10 in IP buffer. The immunoprecipitates were analyzed by 10% urea-SDS-PAGE followed by Western blotting using 6E10. Lanes 1 and 5 are immunoprecipitates from the supernatant and pellet, respectively, obtained from membranes incubated at 0°C for 1 h. Lanes 2 and 6 are immunoprecipitates from the supernatant and pellet, respectively, obtained from membranes incubated at 37°C for 1 h. Lanes 3 and 7 are immunoprecipitates from the supernatant and pellet, respectively, obtained from membranes incubated at 37°C for 1 h in the presence of DAPT. Lanes 4 and 8 are immunoprecipitates from the supernatant and pellet, respectively, obtained from membranes incubated at 37°C for 1 h in the presence of L-685,458 (L685). Lane 9 is the mixture of synthetic A␤ 40/42 , and lane 10 is the synthetic A␤ 46 standard. Since 6E10 was used for both immunoprecipitation and Western blotting, the bands between the fAPP and CTF␤ bands are the heavy chain (HC) and light chain (LC) of mouse IgG. The middle panel is the light exposure of the Western blot for observing the changes in the amount of CTF␤. Bottom panel, prior to centrifugation, aliquots of the reaction mixture were separated by 10 -18% regular SDS-PAGE and probed with C15, an APP C-terminal specific antibody (9), to detect AICD. As discussed in Fig. 3, the two CTF␣ bands correspond to the Myc-tagged recombinant CTF␣ and the endogenous CTF␣. Similarly, the two AICD bands correspond to the Myctagged recombinant AICD and the endogenous AICD. indicates that L-685,458 has no effect on the turnover of A␤ 46 . It was noted that only a trace amount of detectable A␤ 40/42 was immunoprecipitated from the supernatants (Fig. 2, lanes 2 and 4, upper panel). One possibility is that the secretion of A␤ 40/42 from the living cells might be an energy-dependent procedure. Therefore, the secretion of A␤ 40/42 from the membrane of a cell-free system is not as efficient as in the living cells. Notably, small amounts of A␤ 46 , CTF␤, and full-length APP were also detected in these supernatants. It is possible that the A␤ 40/42 , as well as other APP derivatives detected in the supernatants, may be associated with the low density membranes contained in the supernatant fraction. The other possibility is that once the A␤ 40/42 is released from the membrane, or more precisely from the ␥-secretase complex, it might be rapidly degraded by the proteases released, during resuspension and incubation, from the membrane, which contains many kinds of protease-containing vesicles. In contrast, the A␤ 40/42 , which still remains in the membrane or, more precisely, before being released from the ␥-secretase complex, was protected from the degradation. In this regard, it has been reported that the association of CTF␤ with PS1 protects CTF␤ from random degradation (17). This degradation of A␤ 40/42 released from the membrane may also account for the fact that the amount of A␤ 40/42 detected is smaller than expected, compared with the decrease in precursor A␤ 46 .
A␤ 46 Is Processed into A␤ 40/42 in Living Cells in the Presence of L-685,458-As shown in Fig. 2, A␤46 was indeed processed into A␤ 40/42 in a cell-free system. Data presented in Fig. 2 also clearly demonstrate that L-685,458, the transition state analog, did not block the turnover of A␤ 46 into A␤ 40/42 in a cell-free system, indicating that L-685,458 has no detectable effect on ␥-cleavage that produces A␤ 40/42 from A␤ 46 Fig. 1, at the specified concentrations, DAPM completely blocked the formation of secreted A␤ 40/42 and caused marked accumulation of A␤ 46 (Fig. 1, lane 18), and L-685,458 completely blocked the formation of A␤ 40/42 and A␤ 46 (Fig. 1, lane 4). After 12 h of incubation, L-685,458 (0.5 M) was added to the cells in Fig. 3A, lane 2. L-685,458 was also added to the cells in Fig. 3A, lanes 5 and 7, in addition to the existing DAPM, and was continuously incubated for 40 min to completely stop the generation of new A␤ 46 in these cells, because at this concentration, L-685,458 blocked the formation of A␤ 46 (Fig. 1, lane 4). Since L-685,458 has no effect on the turnover of A␤ 46 (Fig. 2), this treatment also allows the complete turnover of the A␤ 46 possibly existing in the cells of lane 2 of Fig. 3A. As a control, cells in Fig. 3A, lane 1, were cultured in the presence of Me 2 SO throughout Note, CTF␤ and CTF␣ generated from exogenous APP, which is expressed with a Myc tag fused to its C terminus, were designated as CTF␤⅐myc and CTF␣⅐myc, respectively; CTF␣ generated from endogenous APP were designated as CTF␣(end), as described in a previous study (27) ␥-Cleavage Depends on -Cleavage the course of the experiment. All cells were then washed twice with fresh medium containing the appropriate inhibitor, which was to be used in the next incubation step, and cultured for an additional 2 h either in the presence or absence of inhibitors as indicated. A␤ 40/42 was immunoprecipitated using 6E10 from CM of the last 2-h cultures. Both cell lysates and A␤ 40/42 immunoprecipitated from CM were analyzed by 10% Bicine/urea-SDS-PAGE followed by Western blotting using 6E10.
As shown in Fig. 3A, 2nd panel, secreted A␤ 40/42 was detected in cells cultured in the absence of any inhibitors throughout the course of the experiment (lane 1, 2nd panel). Secreted A␤ 40/42 was also detected in cells cultured in the absence of inhibitors during the last 2-h incubation period, with a concomitant decrease in both CTF␤ and A␤ 46 (Fig. 3A,  lane 4). As expected, secreted A␤ 40/42 was not detected in cells cultured in the presence of L-685,458 either throughout the course of the experiment (Fig. 3A, lane 6) or during the last two incubation periods (40 min and 2 h) (lane 2). Also, secreted A␤ 40/42 was not detected in cells cultured in the presence of DAPM throughout the course of the experiment (Fig. 3A, lane 3). However, when the DAPM was replaced by L-685,458 during the last 2-h incubation period, secreted A␤ 40/42 was detected in the media, and concomitantly, the pre-accumulated A␤ 46 disappeared (Fig. 3A, (Fig. 3A, lane 6, bottom panel) or during the last two incubation periods (40 min and 2 h) (lanes 2 and 5, bottom panel), indicating that L-685,458 prevented CTF␤ from turnover. Therefore, the A␤ 40/42 detected in Fig. 3A, lanes 5 and 7, should have been produced solely from the pre-accumulated A␤ 46 by DAPM during the prior 12 h of culture in the presence of DAPM. This is also supported by the fact that without pre-accumulation of A␤ 46 during the first 12 h of culture, A␤ 40/42 was not detected in cells (Fig. 3A, lane 2) cultured in the presence of L-685,458 during the last two incubation periods (40 min and 2 h). Since both A␤ 46 and CTF␤ decreased, the secreted A␤ 40/42 detected in Fig. 3A, lane 4, is apparently the sum of the A␤ 40/42 produced from both pre-accumulated A␤ 46 and CTF␤, and the CTF␤ was most likely first converted to A␤ 46 , and the resulting A␤ 46 was further processed to A␤ 40/42 . A small amount of AICD was also detected in cells cultured in the presence of DAPM throughout the course of the experiment (Fig. 3A, lane 3, bottom panel). This result further confirmed that the nontransition state inhibitor DAPM has less effect on the turnover of CTF␤ by ⑀and -cleavages. As discussed below, the accumulation of CTF␤ in the presence of DAPM is possibly the result of the partial inhibitory effect of DAPM on ⑀-cleavage or, alternatively, results from the accumulation of A␤ 46 , which remains tightly associated with PS1 (Fig. 4) and which prevents CTF␤ from accessing the ␥-secretase. It was noted that the amount of AICD detected was smaller than expected, compared with the decrease in CTF␤ and CTF␣ in Fig. 4, lanes 1 and 4. This is likely because of the rapid degradation of this CTF fragment in living cells, as reported previously (18). Nevertheless, the detection of AICD in DAPM-treated cells clearly indicates that at the tested concentration DAPM has less effect on ⑀-cleavage, even though at the same concentration DAPM completely blocked the formation of secreted A␤ 40/42 . In contrast, L-685,458 completely inhibits the formation of AICD and this is in agreement with a previous report (5).
To confirm further the finding that A␤ 40/42 was produced solely from pre-accumulated A␤ 46 in the presence of L-685,458, a time course experiment was performed. As shown in Fig. 3B, and as described in  1, 2, 6, and 7) for 12 h. Then L-685,458 (0.5 M) was added to the cells in Fig. 3B, lanes 2 and 7. In addition to the existing DAPM, L-685,458 was also added to the cells in Fig. 3B, lanes 5 and 10, and was continuously incubated for 40 min to stop completely the generation of new A␤ 46 in these cells. As controls, cells in Fig. 3B, lanes 1 and 6, were cultured in the presence of Me 2 SO throughout the course of the experiment. All cells were then washed twice with fresh medium containing the appropriate inhibitor, which was to be used in the next incubation step, and were cultured for an additional 1 (Fig. 3B,  lanes 1-5) or 2 h (lanes 6 -10), either in the presence or absence of inhibitors as indicated. By using 6E10, A␤ 40/42 was immunoprecipitated from CM of the last 1-and 2-h cultures. Both cell lysates and A␤ 40/42 immunoprecipitated from CM were analyzed by 10% Bicine/urea-SDS-PAGE followed by Western blotting using 6E10. As shown in Fig. 3B, lanes 1 and 6 (lower panel), in the absence of inhibitors, A␤ 40/42 is apparently produced in a time-dependent manner during the last 1 (lane 1) and 2 h (lane 6) of culture. As shown in Fig. 3B, lanes 2 and 7, CTF␤ accumulated in a time-dependent manner, but neither secreted A␤ 40/42 nor intracellular A␤ 46 was detected in the cells treated with L-685,458 during the last two incubation periods (40 min and 1 or 2 h). As shown in Fig. 3B, lanes 3 and 8, no A␤ 40/42 was detected in the CM of cells treated with DAPM throughout the course of the experiment (lower panel). Instead, an accumulation of intracellular A␤ 46 and CTF␤ was observed in these cells (Fig. 3B, upper panel). In contrast, when DAPM was removed during the last 1 and 2 h of incubation, the accumulated A␤ 46 and CTF␤, with concomitant increase in secreted A␤ 40/42 , were decreased in a time-dependent manner (Fig. 3B, compare lane 9 with lane 4 of both upper and lower panels). More interestingly, when DAPM was replaced with L-685,458 during the last 1 and 2 h of incubation, the time-dependent decrease in pre-accumulated A␤ 46 (Fig. 3B, compare lane 10 with lane 5, upper panel) and the concomitant increase in secreted A␤ 40/42 (compare lane 10 with lane 5, lower panel) was also observed. It is notable that the accumulated CTF␤ remained unchanged (Fig. 3B, compare lane 10 with lane 5, upper panel) during this time course. These results clearly indicate that the secreted A␤ 40/42 detected in these cells was solely produced from the pre-accumulated A␤ 46 . This conclusion is also supported by the observation that without the pre-accumulating A␤ 46 , no secreted A␤ 40/42 was detected in CM of cells cultured in the presence of L-685,458 during the last 1 and 2 h (Fig. 3B, lanes 2 and 7). The secreted A␤ 40/42 detected in Fig. 3B, lane 9, is the sum of the A␤ 40/42 produced from both accumulated A␤ 46 and CTF␤.
The fact that the secreted A␤ 40/42 is produced from A␤ 46 in the presence of L-685,458, in both cell-free and living cell systems, clearly indicates that L-685,458 has no direct inhibitory effect on the ␥-cleavage. Therefore, the absence of secreted A␤ 40/42 in cells treated with L-685,458, which blocks the formation of A␤ 46 from CTF␤ by -cleavage, indicates that A␤ 40/42 cannot be generated directly from CTF␤ by ␥-cleavage. In other words, formation of A␤ 46 by -cleavage is an indispensable step during the course of ␥-secretase-mediated processing of CTF␤ to produce A␤ 40/42 . A␤ 46 Is Associated with PS1-A previous study has shown that as a substrate of ␥-secretase, CTF␤ forms a complex with PS1, which is the putative catalytic subunit of the ␥-secretase complex, at the sites of A␤ formation (19). If A␤ 46 is the precursor of A␤ 40/42 , then A␤ 46 , as an intermediate, may still be associated with PS1. To determine whether A␤ 46 is still associated with PS1, the co-immunoprecipitation experiment was performed. As described under "Materials and Methods," lysates of cells treated with compound E, which causes the accumulation of A␤ 46 , were first subjected to 800 ϫ g centrifugation to remove the unbroken cells and nuclei. The resulting postnuclear supernatant was subjected to further centrifugation at 20,000 ϫ g resulting in the supernatant, which contains the low density microsomal and cytosolic fractions (20), and the pellet, the crude membrane fraction containing the trans-Golgi network (TGN) and plasma membrane (21). As shown in the upper panel of Fig. 4A, A␤ 46 was detected in the whole cell lysate (lane 2) and the crude membrane fraction of 20,000 ϫ g (lane 3), but not in the supernatant fraction of 20,000 ϫ g (lane 4). Most interestingly, as shown in the lower panel of Fig. 4A, PS1 was also detected in the whole cell lysate (lane 2) and the fraction of pellet at 20,000 ϫ g (lane 3), but not in supernatant at 20,000 ϫ g (lane 4), indicating that A␤ 46 co-fractionates with PS1 into the crude membrane fraction. Therefore, as described under "Materials and Methods," after solubilization of the pellet fraction of 20,000 ϫ g, co-immunoprecipitation was carried out by using anti-PS1N, an antibody specific to the N terminus of PS1 (9). As shown in Fig.  4B, A␤ 46 was indeed co-immunoprecipitated with PS1 from the crude membranes prepared from cells treated with compound E (lane 3) but not in cells treated with L-685,458 (lane 5), which inhibits the formation of A␤ 46 from CTF␤ and causes accumulation of CTF␤ (Fig. 1). In agreement with the previous study (19), CTF␤ was co-immunoprecipitated with PS1 in the L-685,458-treated cells (Fig. 4B, lane 5, lower panel). As controls, neither A␤ 46 nor CTF␤ was immunoprecipitated by pre-immune rabbit IgG (Fig. 4B, lanes 2 and 4). The observation that A␤ 46 is tightly associated with PS1 in the TGN-containing membrane fraction is in agreement with the previous report that TGN is the major site for A␤ formation (22).
Detection of the Possible A␤ 49 -The data presented above clearly demonstrate that A␤ 46 is an intermediate precursor of secreted A␤ 40/42 . We next attempted to determine the possible presence of A␤ 49 generated by ⑀-cleavage. Lysates of cells cultured in the absence of inhibitor were subjected to immunoprecipitation followed by Western blotting FIGURE 5. A, detection of A␤ 49 . N2a cells stably expressing both PS1 and APPsw were cultured in the absence of inhibitor. Cell lysis, and immunoprecipitation were performed as described under "Materials and Methods." The immunoprecipitates (IP) were analyzed by 10 -18% regular SDS-PAGE followed by Western blotting using 6E10 (lanes 2 and 3). Note, IP* indicates that immunoprecipitation was carried out in the presence of DAPT. ␥-Cleavage Depends on -Cleavage using 6E10. As shown in Fig. 5A, A␤ 46 was immunoprecipitated from untreated cells (lane 2). Most interestingly, when the immunoprecipitation was carried out in the presence of DAPT, in addition to the band of A␤ 46 , a band with a slower migration rate was detected (Fig. 5A, lane  3). Possibly due to the lower concentration and the hydrophobicity, mass spectrometric analysis of this A␤ species was unsuccessful. To estimate its molecular size, we synthesized three A␤ peptides, A␤ 46 , A␤ 48 , and A␤ 49 . As the new A␤ species migrates at the same rate as that of the synthetic A␤ 49 (Fig. 5A, lane 4), it is most likely the long sought after intermediate, A␤ 49 , generated by ⑀-cleavage. This conclusion is also supported by the fact that the majority of AICD starts from A␤ 50 as reported by previous studies (5)(6)(7)(8). It should be noted that under normal conditions, A␤ 49 can only be detected after enrichment by immunoprecipitation carried out in the presence of DAPT, indicating its rapid turnover.
A␤ 49 Is the Precursor of A␤ 46 -Data presented in Fig. 3A clearly indicate that L-685,458 inhibits ⑀-cleavage that produces AICD. To determine further the effect of L-685,458 on the -cleavage, which produces A␤ 46 , and the relationship between A␤ 49 and A␤ 46 , we created a construct, APPsw645, that expresses a C-terminal truncated APPsw ending at the ⑀-cleavage site A␤ 49 . N2a cells, which stably express wild type PS1, were stably transfected with APPsw645. As shown in Fig. 5B, secreted A␤ 40/42 was detected in the medium of cells cultured in the absence of any inhibitors (lane 2, middle panel). However, when the cells were treated with compound E, no secreted A␤ 40/42 was detected in the medium with a concomitant accumulation of intracellular A␤ 46 (Fig.  5B, lane 3, top panel). This result clearly indicates that formation of secreted A␤ 40/42 from A␤ 49 is also mediated by the formation of the intermediate A␤ 46 . Most interestingly, when cells were treated with L-685,458, neither A␤ 40/42 nor A␤ 46 was detected (Fig. 5B, lane 4, top and middle panels), with concomitant accumulation of intracellular A␤ 49 (lane 4, top panel). Given the fact that L-685,458 has no effect on the ␥-cleavage that produces A␤ 40/42 (Figs. 2 and 3), this result indicates that the blockage of the formation of A␤ 40/42 from A␤ 49 , by L-685,458, is not due to inhibition of ␥-cleavage but rather due to inhibition of -cleavage that produces A␤ 46 , which can be further processed into A␤ 40/42 even in the presence of L-685,458. It was noted that a low amount of A␤ 49 was also detected in cells cultured in the absence of inhibitors (Fig. 5B, lane 2, top panel). This may be a result of the lower efficiency of ␥-secretase processing because of the lack of AICD, which may be required for APP to efficiently initiate the interaction with the ␥-secretase complex. This inefficient interaction of A␤ 49 with ␥-secretase complex may also account for the detection of the unprocessed A␤ 49 secreted into the medium (Fig. 5B, lane 2, middle and bottom panels). In this regard, it was also noted that in the presence of inhibitors, specifically in the presence of L-685,458, a significant amount of unprocessed A␤ 49 was detected in the media (Fig. 5B,  lanes 3 and 4, middle and bottom panels). One possibility is that in the presence of these inhibitors, the ␥-secretase-bound A␤ 49 or the intermediate A␤ 46 occupies the binding site of ␥-secretase complex and prevents other unprocessed A␤ 49 from binding to ␥-secretase complex, resulting in the secretion of these unprocessed A␤ 49 into the media.

DISCUSSION
By using the combination of L-685,458 with compound E or L-685,458 with DAPT, we clearly demonstrated that the absence of A␤ 46 in cells treated with L-685,458 is not due to its failure to block the turnover of A␤ 46 , but instead is due exclusively to its inhibition of the formation of A␤ 46 . Our data further demonstrate that in both the cellfree system and in living cells, L-685,458 has no detectable effect on the turnover of A␤ 46 under the current experimental conditions. A similar inhibition profile was also observed for 31C (data not shown), indicating that these inhibitors, known as transition state analogs, share the same inhibitory specificity, i.e. they specifically inhibit the formation of AICD by ⑀-cleavage and the formation of A␤ 46 by -cleavage, but have no effect on ␥-cleavage which produces secreted A␤ 40/42 from A␤ 46 . The observation that L-685,458 has no effect on the turnover of A␤ 46 is important because this made it possible to determine the following: 1) the precursor and product relationship between A␤ 46 and A␤ 40/42 ; 2) the key role of -cleavage in the formation of A␤; and 3) the sequential relationship among the three major intramembrane cleavages, namely the ␥-cleavage, the ⑀-cleavage, and the newly identified -cleavage.
By using the differential inhibition approach, our data presented in Figs. 2 and 3 clearly reveal the important findings that in the presence of L-685,458, A␤ 46 undergoes further ␥-cleavage to produce secreted A␤ 40/42 , both in a cell-free system and in living cells. These results indicate that L-685,458 does not directly inhibit ␥-cleavage. Therefore, the fact that inhibition of the formation of A␤ 46 by L-685,458 also blocks the formation of A␤ 40/42 from CTF␤ (Fig. 3) and A␤ 49 (Fig. 5B, lane 4) indicates that without the formation of the intermediate A␤ 46 by -cleavage, A␤ 40/42 cannot be directly generated from CTF␤ or A␤ 49 by ␥-secretase, i.e. A␤ 46 is the intermediate precursor of A␤ 40/42 . However, it cannot be totally ruled out that A␤ 40/42 can be generated directly from CTF␤ or A␤ 49 by a distinct ␥-cleavage, which is inhibited by L-685,458.
To confirm further the notion that A␤ 46 is the intermediate precursor of A␤ 40/42 or, in other words, A␤ 46 is the intermediate product of the ␥-secretase-mediated proteolytic processing of CTF␤, we performed co-immunoprecipitation experiments and found that, as an intermediate product during the intramembranous processing by ␥-secretase, A␤ 46 is indeed tightly associated with PS1. In agreement with the previous study (19), CTF␤ was also co-immunoprecipitated with PS1 (Fig.  4B, lanes 3 and 5). It was noted that in compound E-treated cells, only a small amount of CTF␤ was detected in the co-immunoprecipitate (Fig.  4B, lane 3). One possibility is that, in the presence of compound E, which prevents the turnover of A␤ 46 into A␤ 40/42 , the accumulated intermediate A␤ 46 occupies the binding site of the ␥-secretase complex and thus prevents the further binding of CTF␤ to the ␥-secretase complex and results in less CTF␤ co-immunoprecipitating with PS1. This possibility is also supported by the observation that most of the accumulated CTF␤ is not associated with PS1 but is detected in the subcellular fraction that is free of PS1 (Fig. 4A, lane 4). In contrast to the low amount of CTF␤ co-immunoprecipitating with PS1 in L-685,458treated cells, the observation that a significantly high amount of A␤ 46 was co-immunoprecipitating with PS1 in compound E-treated cells indicates that the complex formed between PS1 and A␤ 46 is more stable than that formed between PS1 and CTF␤. The tight association of A␤ 46 with PS1 in the TGN-containing membrane fraction, which has been reported to be the major site of A␤ generation (22), provides further strong support for the notion that A␤ 46 is an intermediate product formed during the ␥-secretase processing and that further turnover of A␤ 46 must be dependent on a PS1-containing enzyme, i.e. most likely the same ␥-secretase. Taken together, these results indicate that once bound to presenilin, the initial substrate, CTF␤, and, specifically, the intermediate product, A␤ 46 , are closely associated with presenilin until the release of the final product of secreted A␤ 40/42 . Thus, all the data presented support our hypotheses that A␤ 46 is the precursor of the secreted A␤ 40/42 and that the -cleavage, which produces A␤ 46 , plays a key role in A␤ formation.
The fact that L-685,458 has no effect on the turnover of A␤ 46 (Figs. 2 and 3) and that L-685,458 blocks the formation of A␤ 46 from A␤ 49 by -cleavage (Fig. 5B), and also blocks the formation of AICD by ⑀-cleavage (Fig. 3), indicates that the transition state analog L-685,458 specifically inhibits both ⑀and -cleavages but has no effect on ␥-cleavage. Moreover, the finding that L-685,458 does not directly inhibit ␥-cleavage strongly supports an important notion that this inhibitor inhibits the formation of A␤ 40/42 in living cells by a mechanism other than directly inhibiting the ␥-cleavage, namely by indirectly inhibiting the formation of A␤ 46 by -cleavage. This idea is supported by the fact that A␤ 46 is detectable in cells cultured in the absence of any inhibitor, indicating that -cleavage must occur prior to ␥-cleavage, i.e. -cleavage is upstream of ␥-cleavage. Therefore, the finding that inhibition of upstream -cleavage by L-685,458, which does not directly inhibit ␥-cleavage, completely prevents the downstream ␥-cleavage from taking place strongly suggests that in living cells ␥-cleavage not only occurs secondarily but is also dependent on -cleavage occurring first. In this regard, the fact that the putative A␤ 49 , which contains the -cleavage site at A␤46, is detectable in living cells (Fig. 5A, lane 3) strongly suggests a possibility that ⑀-cleavage occurs prior to -cleavage, otherwise the ⑀-cleavage product A␤ 49 would not have had a chance of being formed. Moreover, our data clearly demonstrate that A␤ 49 cannot be processed directly into A␤ 40/42 by ␥-cleavage. It has to be first processed into A␤ 46 by -cleavage and then the A␤ 46 undergoes further processing by ␥-cleavage to produce A␤ 40/42 (Fig. 5B). Taken together, as illustrated in Fig. 6, our data strongly suggest the possibility that under normal conditions, after ␤or ␣-cleavage of APP, the resulting CTF␤ and CTF␣ first undergo ⑀-cleavage, followed by a sequential but rapid -cleavage, and then by a ␥-cleavage, commencing at the site closest to the membrane boundary and proceeding toward the site in the middle of the transmembrane domain of APP. Support for this sequential action model also comes from the notion that water molecules play an important role in the peptide bond hydrolysis catalyzed by a protease, and ␥-secretase has been proposed to be an aspartyl protease (4). According to the catalytic mechanism of aspartyl protease, in order to hydrolyze the peptide bond of the substrate, one of the two aspartate residues in the enzyme active site, disposed on opposite faces of the peptide bond to be cleaved, needs to first act as general base to activate the water molecule. The activated water molecule then attacks and breaks the peptide bond, in cooperation with the second aspartate, which acts as a general acid to protonate the departing amine product. The ⑀-cleavage site is close to the membrane boundary and is easily accessed by water mole-cules in the cytosol. The initial ⑀-cleavage may not only release the AICD but may also create a path for the water molecule to have access to the next cleavage site, namely -cleavage site and then ␥-cleavage site. Accordingly, without removal of the three C-terminal residues from A␤ 49 by -cleavage, water molecules may not be able to access the ␥-cleavage site, resulting in the prevention of ␥-cleavage from taking place. Thus, the blockage of water access may account, at least in part, for the fact that ␥-cleavage depends on -cleavage occurring first.
Regarding the relationship between ⑀and -cleavages, as discussed above, one possibility is that ⑀-cleavage occurs before -cleavage. However, since L-685,458 blocks both ⑀and -cleavages and the inhibitor, which specifically inhibits ⑀-cleavage or -cleavage, has not yet been identified, it cannot be ruled out that ⑀-cleavage and -cleavage may also occur simultaneously. Nevertheless, even though it is currently not clear whether -cleavage is dependent on ⑀-cleavage, the finding that the generation of A␤ 40/42 from A␤ 49 has to be mediated by the formation of A␤ 46 by -cleavage indicates that once ⑀-cleavage occurs it has to be followed by -cleavage to produce A␤ 46 , which is then further processed into A␤ 40/42 by ␥-cleavage.
Regarding the catalytic mechanism of the ␥-secretase, the finding that ⑀-/-cleavages and ␥-cleavage can be differentially inhibited by transition state analogs and nontransition state inhibitors, respectively, suggests several possibilities. First, these cleavages may be catalyzed by two enzymes, and second, these cleavages may be catalyzed by one enzyme that has two inhibitor binding sites, one for the transition state analogs, such as L-685,458, and the other for the nontransition state inhibitors, such as compound E, as suggested by a recent inhibitor binding kinetic study (23). The sequential relationship of these cleavages and specifically the finding that ␥-cleavage is dependent on ⑀and -cleavages occurring first, suggest that ␥-, -, and ⑀-cleavages are catalyzed by a single enzyme. The single enzyme model is further strongly supported by the fact that the intermediate A␤ 46 is tightly associated with PS1, the putative catalytic subunit of the ␥-secretase complex. The one enzyme model is also supported by the fact that both groups of the inhibitors have been shown to bind to presenilins (24 -26). According to this model and the hypothesis that transition state analogs and nontransition state inhibitors bind to different sites (23), the transition state analogs may inhibit the initial cleavage, namely the ⑀-cleavage, by directly binding to the catalytic site. As a result, the downstreamand ␥-cleavages are also prevented from taking place. On the other hand, the nontransition state inhibitors may bind to a remote site of the enzyme and induce conformational changes in the enzyme, resulting in the preferential inhibition of ␥-cleavage, with less effect on the ⑀and -cleavages, by preventing the ␥-cleavage site from having access to the catalytic site of the enzyme. However, the one catalytic site model fails to account for the fact that A␤ 46 is still processed to A␤ 40/42 by ␥-cleavage in the presence of the transition state analog (Figs. 2 and 3), which is assumed to bind the catalytic site (23). Therefore, the third possibility is likely that these sequential cleavages may be catalyzed by an enzyme that has two catalytic sites, one engaged in carrying out the ⑀and -cleavages and the other engaged in carrying out the ␥-cleavage. Regardless of whether there is one or two catalytic sites, according to the one enzyme model, at high concentrations the nontransition state inhibitors, which preferentially inhibit ␥-cleavage, may also inhibit ⑀and -cleavages by an allosteric mechanism, i.e. these compounds may induce conformational changes into the ␥-secretase complex, resulting in partial inhibition of ⑀and -cleavages. The other possibility that may account for the accumulation of CTF␤ in cells treated with nontransition state inhibitors is that in the presence of nontransition state inhibitors, which inhibit the turnover of A␤ 46 into A␤ 40/42 , the accumulated intermediate A␤ 46 FIGURE 6. Schematic illustration of APP processing. L-685,458 specifically inhibits ⑀and -cleavages (-); DAPT strongly and preferentially inhibits ␥-cleavage (O) and also inhibits ⑀and -cleavages at higher concentration (----). ⑀P3 and P3 are the putative P3-related fragments produced from CTF␣ by ⑀and -cleavages, respectively.

␥-Cleavage Depends on -Cleavage
occupies the binding site of the ␥-secretase complex and prevents the further binding of CTF␤ to the ␥-secretase complex, resulting in accumulation of unprocessed CTF␤.