Inhibition of Receptor-mediated Endocytosis Demonstrates Generation of Amyloid β-Protein at the Cell Surface*

Sequential cleavages of the amyloid β-protein precursor (APP) by the β- and γ-secretases generate the amyloid β-protein (Aβ), which plays a central role in Alzheimer's disease. Previous work provided evidence for involvement of both the secretory and endocytic pathways in Aβ generation. Here, we used HeLa cells stably expressing a tetracycline-regulated dominant-negative dynamin I (dyn K44A), which selectively inhibits receptor-mediated endocytosis, and analyzed the effects on the processing of endogenous APP. Upon induction of dyn K44A, levels of mature APP rose at the cell surface, consistent with retention of APP on the plasma membrane. The α-secretase cleavage products of APP were increased by dyn K44A, in that α-APPs in medium and the C83 C-terminal stub in the membrane both rose. The β-secretase cleavage of APP, C99, also increased modestly. The use of specific γ-secretase inhibitors to study the accumulation of α- and β-cleavage products independent of their processing by γ-secretase confirmed that retention of APP on the plasma membrane results in increased processing by both α- and β-secretases. Unexpectedly, endogenous Aβ secretion was significantly increased by dyn K44A, as detected by three distinct methods: metabolic labeling, immunoprecipitation/Western blotting, and enzyme-linked immunosorbent assay. Levels of p3 (generated by sequential α- and γ-cleavage) also rose. We conclude that endogenous Aβ can be produced directly at the plasma membrane and that alterations in the degree of APP endocytosis may help regulate its production. Our findings are consistent with a role for the γ-secretase complex in the processing of numerous single-transmembrane receptors at the cell surface.

Genetic, neuropathological, biochemical, and animal modeling studies all suggest that the progressive accumulation of the amyloid ␤-protein (A␤) 1 in limbic and association cortices ini-tiates Alzheimer's disease. Such evidence has recently led to the development of therapeutic agents aimed at decreasing the production or enhancing the clearance of A␤. Despite this progress, a central unresolved question remains: precisely where in the cell is A␤ generated from the membrane-anchored amyloid ␤-protein precursor (APP)? This question has taken on greater interest with emerging evidence that APP appears to function as a cell surface receptor that can undergo regulated intramembrane proteolysis (RIP) (1,2). Sequential cleavage of APP by the aspartyl proteases BACE/ ␤-secretase and presenilin/␥-secretase releases A␤ into the lumen/extracellular space and liberates the APP intracellular domain into the cytoplasm in a process strikingly similar to the cleavage of the Notch receptors that enables their signaling in myriad cell fate decisions (2)(3)(4). An increasing number of cell surface proteins, including nectin-1␣ (5), CD44 (6), LRP (7), E-cadherin (8), andErbB4 (9), has been discovered to undergo RIP, with some of these releasing intracellular domains that may function as transcriptional regulators. Thus, the processing of APP by RIP, rather than being an anomaly related to the pathogenesis of Alzheimer's disease, appears to represent a conserved mechanism employed by cells to communicate information from events at the cell surface to the nucleus.
Because the initiation of receptor endoproteolysis and signaling often requires the binding of a cognate ligand located on an adjacent cell (e.g. Delta or Jagged in the case of Notch) or else intercellular adhesion (e.g. in the case of E-cadherin), it is reasonable to hypothesize that APP also undergoes processing at or near the cell surface. Indeed, deletion or mutation of a cytoplasmic YENPTY motif important for the internalization of surface-inserted APP has been reported to sharply reduce A␤ generation (10 -12). Moreover, selective labeling of surface APP molecules in living cells leads to the release of labeled A␤ into the medium in a time course consistent with the internalization and endocytic recycling of APP (12). Furthermore, trypsinization of surface proteins on human brain cells ex vivo leads to a decrease in secreted A␤ (13). Other studies suggest a distinct pathway for generating intracellular A␤. Immunocytochemistry has revealed A␤ in the trans-Golgi network (TGN) and ER (14,15). Preventing APP from reaching the cell surface by treating NT2N (differentiated NT2 cells) neurons with brefeldin A abrogated A␤ secretion but resulted in its intracellular accumulation (16). Similar results have been obtained in non-neural cells (17). Therefore, available evidence suggests that A␤ can be generated and accumulate in more than one subcellular locus. Whether the principal site of its proteolytic generation relates to the putative function of its precursor as a cell surface receptor is unclear.
An approach that could help clarify this issue but has been little exploited in research on APP is to take advantage of dominant-negative mutants of proteins known to mediate endocytosis. Here, we have systematically characterized the effects on APP processing of a well characterized dominantnegative mutation in dynamin I (dyn), K44A (18), that inhibits the internalization of transferrin receptor and certain other surface proteins, which undergo receptor-mediated endocytosis (18,19). Using a tetracycline-regulated expression system, we find that induction of dyn K44A results in increased retention of endogenous, full-length APP on the plasma membrane and a consequent increase in its processing by ␣-secretase. Surprisingly, this inhibition of APP endocytosis was associated with more, rather than less, secretion of A␤ into the medium. The increase in the proteolytic turnover of APP held at the plasma membrane occurred in the absence of detectable changes in the levels of ␣-secretase or presenilin/␥-secretase. We conclude that, in addition to evidence that A␤ can be generated during endosomal recycling, it can be produced from APP directly on the plasma membrane. Our findings are consistent with a critical role for the presenilin/␥-secretase complex in the processing of numerous single-transmembrane proteins at the cell surface. Moreover, our results suggest that ␤or ␥-secretase inhibitors intended to lower A␤ production may be able to operate principally at the plasma membrane, without the need for full cell penetration.

MATERIALS AND METHODS
Cell Culture-An HeLa cell line expressing a dominant-negative dynamin I K44A protein under the control of a tetracycline transactivator was obtained from Dr. S. Schmid (Scripps Institute, La Jolla, CA). Cells were grown in Dulbecco's modified Eagle's medium containing 10% fetal bovine serum, 100 g/ml penicillin, 100 g/ml streptomycin, 200 g/ml G418, 200 ng/ml puromycin, and 1 g/ml tetracycline. Cells were split 1 day prior to dyn K44A induction. On the day of induction, the medium was changed to medium either containing or lacking tetracycline. The medium (with or without tetracycline) was changed again after 30 min. Ten milliliters of fresh medium (again with or without tetracycline) was added at 24 h post-induction, and cells and conditioned media were collected at 48 h. For the ␥-secretase inhibitor studies, 1 M Compound E was applied from 24 to 48 h of induction, and 1 M DAPT was applied from 48 to 52 h of induction.
Immunoprecipitation and Western Blotting-Cells were lysed in 1% Nonidet P-40 lysis buffer containing Complete protease inhibitor mixture (Roche Applied Science, Indianapolis, IN). Nuclei were pelleted in an Eppendorf centrifuge (Brinkman Instruments, Inc., Westbury, NY) for 2 min at 14,000 ϫ g. Equal protein amounts, determined using the DC Protein Assay (Bio-Rad, Hercules, CA), were analyzed by Western blotting. 10 -20% Tris-Tricine gels (Invitrogen, Carlsbad, CA) were used to separate the APP C-terminal fragments, C83 and C99. 10% Trisglycine gels (Invitrogen) were used to separate full-length APP species. HA-tagged dyn K44A protein was detected with 12CA5 (Roche Molecular Biochemicals). Full-length APP was blotted with the N-terminalspecific monoclonal antibody, 22C11 (Roche Molecular Biochemicals). C83 was blotted with the C-terminal-specific polyclonal antibody C8, or by C8 immunoprecipitation followed by blotting with monoclonal antibody 13G8 (gift of D. Schenk, Elan Pharmaceuticals, San Francisco, CA). All immunoprecipitated samples were first pre-cleared with protein A-Sepharose (PAS) (Sigma-Aldrich, St. Louis, MO). In some experiments, C7, an affinity-purified form of C8, was used. The epitopes for C8 and 13G8 lie in the last 20 amino acids of the APP cytoplasmic domain. C99 was immunoprecipitated with C8 and detected with 6E10 (Signet Laboratories, Inc., Dedham, MA). ␣-APPs was detected using the polyclonal antibody, 1736, which specifically recognizes the ␣but not the ␤-secretase-generated ectodomain fragment. Tubulin and ribophorin were blotted with antibodies from Santa Cruz Biotechnology, Inc. (Santa Cruz, CA). A␤ was immunoprecipitated with polyclonal antibody 1282 and then blotted with 6E10. Biotinylated transferrin (tfn-btn, Sigma-Aldrich) was detected using neutravidin-horseradish peroxidase (Pierce Biotechnology, Rockford, IL). Surface biotinylated proteins were immunoprecipitated with streptavidin-agarose (Molecu-lar Probes, Eugene, OR) and blotted with 22C11 and a ribophorin antibody. Western blots were developed with Pierce Femto SuperSignal (Pierce). All Western blots shown for APP and its proteolytic products were repeated in at least five independent experiments.
Quantitative Western Blotting-Mature and immature APP, C99, and tfn-btn levels were quantified by densitometry using ImageQuaNT software (Amersham Biosciences, Piscataway, NJ). Standard curves for APP and C99 were generated by transiently transfecting plasmids encoding APP695 or C99 (Effectene, Qiagen, Valencia, CA) and then Western blotting 2-fold serial dilutions of these overexpressed proteins in parallel with experimental samples. Surface-bound and internalized tfn-btn were quantified using a standard curve generated from Western blots of serial 2-fold dilutions of tfn-biotin, in parallel with experimental samples A␤ ELISA.
Media were conditioned and collected from 24 to 48 h of dynamin induction or non-induction. 1,10-Phenanthroline (2 mM, Sigma-Aldrich) and Complete protease inhibitor mix were added to the media and spun at 3,000 rpm for 10 min to remove debris in a Marathon 8K centrifuge (Fisher Scientific, Pittsburgh, PA). ELISA for A␤ X-40 was performed as described by Johnson-Wood et al. (20) with the following modifications. 96-well plates were coated with 3 g/ml 2G3, specific for the A␤40 C terminus (gift of P. Seubert and D. Schenk, Athena Neurosciences, San Francisco, CA). Standards of A␤ 1-40 were diluted in unconditioned medium. Bound A␤ was detected with biotinylated 266 antibody (0.25 g/ml), directed to the A␤13-28 region (gift of P. Seubert and D. Schenk, Elan Pharmaceuticals, San Francisco, CA). Plates were washed two times between antibody incubations with Tris-buffered saline/0.05% Tween 20 for 1 min each, and then 2 min each before developing with tetramethylbenzidine-ELISA (Pierce).
Metabolic Labeling-HeLa cells were cultured in medium with or without tetracycline. At 24 h of induction, cells were incubated for 30 min with Cys/Met-free and serum-free medium. Cells were then conditioned in medium with or without tetracycline containing 1 mCi of [ 35 S]Cys, [ 35 S]Met for 24 h. Media were collected and prepared as for ELISA (above). Samples were pre-cleared with PAS. p3 and A␤ were immunoprecipitated with 1282 and PAS. Samples were run on 10 -20% Tris-Tricine gels and detected using a E425 PhosphorImager (Amersham Biosciences).
The time course of APP biosynthesis and maturation was determined by pulse-chase experiments on HeLa cells in 10-cm dishes. After tetracycline induction (or no induction) for 48 h, cells were serum-starved for 30 min and then pulsed with 2 mCi of [ 35 S]Cys plus [ 35 S]Met for 7 min. Cells were washed once briefly with Dulbecco's phosphate-buffered saline (Invitrogen) to terminate the pulse, and 5 ml of medium with or without tetracycline was added for the chase. Cell lysates in 1% Nonidet P-40 lysis buffer containing Complete protease inhibitor mix were collected after cold chase times of 0, 10, and 20 min or 0, 30, 60, and 90 min. Nuclei were pelleted at 14,000 rpm for 2 min. 25 l of lysate from each sample was precipitated with 10% trichloroacetic acid for 20 min at 4°C, spun at 14,000 rpm for 10 min and resolubilized in 1 ml of 100 mM Tris-HCl, pH 8, containing 1% SDS. Incorporation of radiolabel into proteins was measured by adding 40 l of resolubilized proteins to 4 ml of EcoScint H (National Diagnostics, Atlanta, GA) and counted in a 1212 Rackbeta liquid scintillation counter (LKB Wallac, Gaithersburg, MD). Samples corresponding to equal radioactive counts were diluted to a 1-ml volume in 1% Nonidet P-40 lysis buffer, pre-cleared with PAS, and immunoprecipitated with C8. Samples were then run on a 4 -20% Tris-glycine gel and analyzed using an Amersham Biosciences E425 PhosphorImager. Each radiolabeled blot was repeated twice.
Transferrin Uptake Assays-tfn-btn (8 g/ml) was used to determine the kinetics of tfn uptake. Briefly, HeLa cells induced or non-induced for 24 h in six-well plates were washed three times in Hanks' balanced salt solution (HBSS) (Invitrogen), incubated with tfn-btn for 7 min in culture medium, then placed at 4°C. Remaining surface-bound transferrin-biotin was removed with three brief acid washes in HBSS at pH 4. Surface-bound tfn was determined by incubating cells with tfn-btn at 4°C for 30 min, then washing three times with HBSS. Cells were lysed in 1%Nonidet P-40 lysis buffer containing Complete protease inhibitor mix, and nuclei pelleted with at 14,000 rpm for 2 min. Samples were Western blotted as described above for both qualitative and quantitative analysis.
Surface Biotinylation-Confluent 10-cm dishes of HeLa cells induced or not induced for 48 h were washed three times with HBSS. Cells were biotinylated with 0.5 mg/ml Sulfo-NHS-LC-Biotin (Pierce) for 45 min at 4°C. Biotinylation was quenched with a 100 mM glycine/HBSS wash followed by two HBSS washes. Immunoprecipitation/Western analysis of biotinylated materials was carried out as described above. Surface biotinylated Western blots shown are representative of three independent experiments.
Surface Trypsinization-10-cm dishes of HeLa cells were treated as above with or without tetracycline for 44 h. Cells were then washed twice at 4°C with 1 mM EDTA (Sigma) in HBSS without Ca 2ϩ or Mg 2ϩ (Invitrogen) buffered with 10 mM HEPES (Invitrogen) (1 mM EDTA/ HBSS/HEPES). Cells were treated with or without 1 ml of 0.25% trypsin (Sigma-Aldrich) in 1 mM EDTA/HBSS/HEPES for 30 min at 4°C, then quenched with the addition of 1 ml of 0.125% soybean trypsin inhibitor (Sigma-Aldrich) and 8 ml of culture medium containing fetal bovine serum. Cells were washed three times with HBSS and lysed in 1% Nonidet P-40 lysis buffer containing Complete protease inhibitor mixture. Equal protein was loaded on a 4 -20% Tris-glycine gel for tubulin Western blotting. C99 and full-length APP were immunoprecipitated from protein-normalized samples with C8 and detected with 6E10. Surface trypsinization experiments were repeated three times.

Induction of Dynamin K44A I Potently Blocks
Receptor-mediated Endocytosis-We investigated the role of receptor-mediated endocytosis in the generation of A␤ from APP by utilizing a HeLa cell line stably transfected with a dominantnegative dynamin construct, dyn K44A, under the control of a tetracycline-responsive element. Removing tetracycline from the culture medium for 48 h induces the dominant-negative dynamin and potently inhibits receptor-mediated endocytosis (18).
We began by verifying the effects of dyn K44A on the internalization of transferrin, which is brought into cells by transferrin receptors through clathrin-mediated endocytosis. The effects of dyn K44A on transferrin uptake were quantified as internalized transferrin/surface-bound transferrin/unit time. We allowed cells to internalize biotinylated transferrin for 7 min at 37°C and then detached the residual surface-bound (non-internalized) transferrin-biotin by repeated acid washes, thus measuring the internalized transferrin. Surface-bound transferrin was measured by binding transferrin-biotin to the plasma membrane at 4°C and then repeatedly washing with HBSS to remove non-specifically bound transferrin-biotin. As expected, the absence of tetracycline in the culture medium for 48 h led to a strong induction of dyn K44A (Fig. 1A). Following this induction, surface binding of transferrin was increased by ϳ68% (Fig. 1B), reflecting an increased retention of transferrin receptor on the plasma membrane. Accordingly, the total bulk uptake of transferrin was decreased by ϳ58% when dyn K44A was induced (Fig. 1C). Overall, clathrin-mediated endocytosis (internalized transferrin/surface-bound transferrin/unit time) decreased by 75% when dominant-negative dynamin was induced for 48 h (Fig. 1D). Our results are entirely consistent with previously published findings using the same cell line under similar conditions (18), and they confirm the potent inhibition of receptor-mediated endocytosis by induction of dyn K44A.
Expression of Dominant-negative Dynamin Selectively Increases the Level of Mature APP-When dyn K44A was induced, the total cellular levels of mature, N-plus O-glycosylated APP increased robustly, whereas the levels of the immature, N-glycosylated APP were minimally elevated ( Fig.  2A). This result suggested that inhibition of dynamin function affected principally the cell surface form of APP, which is fully glycosylated. When we quantified the increase in holoAPP levels, we found that mature APP increased by ϳ500% (p Ͻ 0.0005), whereas immature APP increased by only 47% (p Ͻ 0.0003) (Fig. 2B). As negative controls, we examined the levels of the cytosolic protein, tubulin, and the transmembrane protein, ribophorin, which is retained in the ER. Tubulin and ribophorin levels were unchanged in the presence versus absence of dyn K44A (Fig. 2C); quantitation of the tubulin signal showed no detectable difference (not shown). When we induced the overexpression of dynamin in a tet-off HeLa cell line stably expressing wild-type (WT) dynamin, we found no difference in the levels of full-length APP (data not shown). Because overexpression of WT dynamin exerts a minimal influence on rates of endocytosis (18), we conclude that neither the presence of tetracycline nor the overexpression of WT dynamin increases mature APP levels. These data indicate that the effects of dyn K44A are specific for the dominant-negative dynamin. Furthermore, the dramatic rise in the level of mature APP, ϳ10-fold greater than that of immature APP, is consistent with dyn K44A principally affecting APP molecules after they have passed through the secretory pathway, completed glycosylation, and reached the cell surface.
Expression of Dominant-negative Dynamin Results in Accumulation of APP on the Plasma Membrane-To confirm that the increase in cellular APP (Fig. 2, A and B) was principally due to retention of APP at the plasma membrane, we biotinylated surface proteins with sulfo-NHS-LC-biotin, which is cellimpermeant, for 45 min on ice following the 48-h induction/no induction of dyn K44A. Samples were washed thoroughly with HBSS containing 100 mM glycine as a quenching reagent, precipitated with streptavidin-agarose beads, and Western blotted. As controls for APP migration, straight cell lysates were loaded adjacent to the streptavidin-precipitated samples. The top portion of the membrane was immunoblotted with an antibody against the APP ectodomain (22C11), and the bottom portion with an antibody against ribophorin, a resident ER membrane protein, as a control. We found that after dominantnegative dynamin expression, the mature form of APP was selectively biotinylated, as confirmed by co-migration with mature APP in the straight lysate (Fig. 2D). The levels of ribophorin were again unchanged by dyn K44A induction, and no biotinylation of ribophorin occurred, as expected (Fig. 2E). The increase in plasma membrane holoAPP could be caused by increased surface retention due to the documented inhibition of receptor-mediated endocytosis (Fig. 1) and/or by decreased proteolytic processing of holoAPP due to inhibition of ␣and/or ␤-secretase activity. As shown below, the former is the explanation.
Biosynthesis and Maturation of APP Are Not Affected by Dynamin Function-To prove that the effects of dynamin dysfunction on APP trafficking and processing were not due to an effect in the secretory pathway, we performed a pulse-chase analysis. Following a 7-min pulse with 2 mCi of [ 35 S]Cys, [ 35 S]Met, and chase times of 0, 10, and 20 min, we found that the generation of immature APP was not significantly different between dyn K44A induced and non-induced cells (Fig. 2F). In addition, we found that the appearance of fully mature endogenous APP, reflecting the trafficking of APP through the Golgi apparatus, was also unchanged at the earliest detection time of 20 min. We next performed a second pulse-chase experiment with chase times of 0, 30, 60, and 90 min to confirm that labeled APP accumulates at times consistent with its delivery to the cell surface. As expected, the cells with blocked endocytosis showed a selective rise in mature APP levels after the 60-and 90-min chases, time points consistent with the arrival of mature APP at the cell surface, whereas immature APP levels fell equally with or without dyn K44A induction.
␣-Secretase Cleavage of APP Increases When Dynamin Function Is Down-regulated-␣-Secretases cleave APP 12 amino acids in front of its transmembrane domain to release a soluble ectodomain fragment called ␣-APPs and leave a C-terminal membrane-anchored stub termed C83. Prior studies have implicated both the secretory pathway and the plasma membrane as sites of ␣-secretase activity. Both ADAM10 and ADAM17 (TACE) have been reported to serve as ␣-secretases for APP.
We first asked if the levels of ADAM10 and TACE were altered when dyn K44A was induced. The levels of mature ADAM10 and TACE were both unchanged (not shown). In addition, we established that the induction of dyn K44A did not alter levels of total cellular presenilin and nicastrin (not shown). Next, we determined the accumulation of secreted ␣-APPs from 24 to 48 h of dyn K44A induction, compared with uninduced control cultures. ␣-APPs was specifically detected by antibody 1736, which recognizes the last 16 residues of ␣-APPs that are not present in ␤-APPs. In the presence of dominant-negative dynamin, the quantity of ␣-APPs secreted into the medium was markedly elevated (Fig. 3A). These data support previous results that identified the plasma membrane as a principal site of ␣-secretase activity (21).
In accord with the marked rise in ␣-APPs, the levels of C83 also increased when dynamin function was suppressed (Fig.  3B). By immunoprecipitating and blotting C83 with highly sensitive antibodies (C7 and 13G8, respectively), we could detect a small amount of endogenous C83. When dyn K44A was induced, the levels of C83 rose, in parallel with the strong increase in ␣-APPs levels.
When ␥-secretase activity was inhibited by incubating the cells with a ␥-secretase inhibitor, Compound E, between 24 and 48 h of culture, we saw a dramatic increase in the levels of C83, as expected (data not shown). The large increase in C83 levels caused by dyn K44A was still preserved in the presence of Compound E (Fig. 3C). Thus, independent of ␥-secretase activity, the levels of C83 were elevated when dyn K44A was induced, consistent with an increase in the amount of surface APP molecules processed by ␣-secretase.
HeLa Cells Produce Endogenous C99, the Immediate Precursor to A␤, and Its Levels Rise Modestly When dyn K44A Is Induced-A prerequisite for the generation of A␤ is the presence of one or both of the ␤-secretases, known as BACE-1 and BACE-2. Recent work has implicated BACE-1 as the protease making by far the major contribution to the generation of A␤ species beginning at Asp-1. Because we wished to determine the fate of A␤ when endocytosis was inhibited, we ascertained whether the HeLa cells could generate endogenous A␤. HeLa

FIG. 2. Induction of dyn K44A for 48 h leads to a selective increase in the mature form of full-length APP.
A, endogenous levels of mature APP were increased following the induction of dyn K44A. B, quantification of the increase in immature and mature APP levels. Immature APP increased by 47% (p Ͻ 0.0003) while mature APP increased by 507% (p Ͻ 0.0005). C, the levels of a control cytosolic protein, tubulin, and a control transmembrane protein that is retained in the ER, ribophorin, were not affected by the induction of dyn K44A. D, surface biotinylation revealed a selective increase in mature APP at the plasma membrane when dyn K44A was induced. Cells were biotinylated for 45 min at 4°C. Samples were immunoprecipitated with streptavidin-agarose beads. Lanes 1 and 2 are surface-biotinylated material detected with APP antibody 22C11 in the presence and absence of tetracycline. Lanes 3 and 4 are straight lysates to confirm the migration of the mature and immature forms of APP in the presence and absence of tetracycline. E, surface biotinylation control. The Western transfer membrane shown in D was cut, and the lower portion probed for biotinylated ribophorin, which was not detected. F, pulse-chase analysis of APP. Induced and non-induced cells were pulse-labeled for 7 min with 2 mCi of [ 35 S]Met and chased for 0, 10, and 20 min or 0, 30, 60, and 90 min. No differences were seen in the rate of APP biosynthesis or maturation. At 60 and 90 min, mature APP was stabilized when dyn K44A was induced, relative to non-induced cultures. m, mature; im, immature. cells are known to contain low levels of BACE, and we could not unambiguously detect endogenous levels of the BACE-1 or BACE-2 enzymes themselves by Western blotting (data not shown). Therefore, we attempted to rely instead on detection of the specific BACE-1 cleavage product, C99. Immunoprecipitation of ϳ15 mg of total cell lysates with the APP antibody C8 was sufficient to detect endogenous C99 after blotting with 6E10, targeted to residues 1-16 of C99. Because this epitope ends just prior to the ␣-cleavage site, 6E10 cannot recognize C83.
We found that steady-state cellular levels of C99 were modestly elevated when dyn K44A was induced (Fig. 4A). When we quantified the change in C99 levels after dyn K44A was expressed, there was a small (30%) but highly reproducible and statistically significant (p Ͻ 0.03) increase. The C99 band was increased when cultures were treated with a ␥-secretase inhibitor, DAPT, for 4 h at the end of the 48-h dyn K44A induction period, as expected (Fig. 4B). In the DAPT-treated cultures, C99 levels still rose after dyn K44A induction, as compared with controls. Thus, our data indicate that ␤-cleavage of APP increased modestly when APP (and possibly BACE) was retained at the plasma membrane.
C99 Selectively Accumulates at the Cell Surface following Inhibition of Dynamin Function-Having established the existence of C99 in these cells and its modest elevation following induction of dyn K44A, we next investigated whether the elevation of C99 occurred selectively at the cell surface. C99 contains three putative cleavage sites for trypsin: arginine at position 5 and lysines at positions 16 and 28. Cleavage at any of these sites destroys recognition of C99 by 6E10. Following 44 h of dyn K44A induction or no induction, we treated cells with 1 M DAPT for 4 h to stabilize C99 levels. We then washed cells briefly with HBSS containing 1 mM EDTA and trypsinized cell surface proteins with 0.25% trypsin in 1 mM EDTA/HBSS for 30 min. We quenched the reaction by adding soybean trypsin inhibitor and excess culture medium. In control samples not treated with trypsin, we found that the mature form of fulllength APP was selectively elevated in dyn K44A-expressing cells, as already shown (Fig. 5). Following cleavage of cell surface proteins with trypsin, we found that the levels of mature APP dropped to the same level as in uninduced cultures. The levels of immature APP, which is not predicted to reside at the cell surface, were unaltered by trypsin treatment. In addition, levels of the cytoplasmic control protein, tubulin, were equal in all samples. Thus, trypsinization specifically removed proteins at the cell surface, revealing that the increase in cellular levels of mature APP induced by dyn K44A expression occurred exclusively at the cell surface.
Following surface trypsinization, we found that C99 levels in dyn K44A-expressing cells were also significantly lowered (Fig.  5). The vast majority of C99 present in dyn K44A-expressing cells was present at the cell surface and accessible to trypsin. Interestingly, the levels of C99 in uninduced cells were not significantly decreased by trypsin treatment. These data demonstrate that the expression of dyn K44A retains C99 at the cell surface, similar to its effect on mature APP. However, in cells not expressing dyn K44A, C99 is probably endocytosed rapidly following a brief presentation at the cell surface. ϪTet medium. Using an antibody specific for ␣-APPs, 1736, we detected a strong increase in the release of ␣-APPs from dyn K44A-expressing cells. B, cellular levels of C83, the C-terminal cleavage fragment from ␣-cleavage of APP, were also increased when dyn K44A was induced. Cell lysates were immunoprecipitated with polyclonal antibody C7, and probed with monoclonal antibody 13G8, both targeted against the last 20 amino acids of APP. The band at ϳ15 kDa is nonspecific and indicates equal protein loading of the samples. C, in the presence of a ␥-secretase inhibitor, compound E, from 24 -48 h of conditioning, levels of C83 increased dramatically and levels of C83 were still elevated when dyn K44A was induced.

FIG. 4. Endogenous ␤-cleavage of APP occurs in HeLa cells, and the ␤-cleavage product C99 from endogenous APP are increased when dyn K44A is induced.
A, using the sensitive and specific antibody 6E10 to recognize ␤-cleaved C-terminal fragments of APP, we saw an increase in C99 levels when dyn K44A was induced. Antibody 6E10 recognizes an epitope on C99 that corresponds to the A␤ region 1-16, which is absent in C83. B, levels of C99 increased after treatment with the ␥-secretase inhibitor, DAPT, and levels were further elevated in cells that were cultured in the absence of tetracycline.

The Secretion of the ␥-Secretase Cleavage Products A␤ and p3
Is Increased upon Inhibition of Dynamin Function-We next examined the secretion of the endogenous human A␤ and p3 peptides that result from ␥-secretase cleavage of C99 and C83, respectively. First, we conditioned media from 24 to 48 h of dyn K44A induction (during metabolic labeling with [ 35 S]Met) and compared the levels of secreted A␤ and p3 to those of uninduced controls. After immunoprecipitation of the labeled media with the strong A␤ antibody, 1282, we were able to detect endogenous A␤ and p3 secretion only when dyn K44A was induced (Fig. 6A). Because antibody 1282 was generated against the entire N terminus of A␤ and p3 begins at position 17 of the A␤ sequence, our results underestimate the amount of p3 secreted relative to A␤. As a parallel positive control, we also detected these peptides in conditioned medium of HEK293 cells stably expressing wild-type human APP (Fig. 6A).
Next, we used a sensitive sandwich ELISA to quantify the endogenous A␤ X-40 peptide. This revealed a substantial increase in A␤ levels when dyn K44A was induced (endogenous A␤ X-42 levels were below the limits of detection). A␤ X-40 levels, determined from 18 pairs of samples taken in three independent experiments, rose from a mean of 121 to 365 pg/ml (Fig. 6B). In addition, we performed time-course experiments by collecting conditioned media from cells expressing dyn K44A at 0 -24, 24 -48, and 48 -72 h of induction and from parallel plates of uninduced cultures. By immunoprecipitating with 1282 and blotting with 6E10, we observed a strong increase in A␤ accumulation at 24 -48 h and an even greater rise at 48 -72 h (Fig. 6C). In contrast, in the 0 -24 h induced cultures (when dyn K44A exerts little effect) and in all collections of control cultures, there was no endogenous A␤ signal detectable above background (Fig. 6C). Note that a control immunoprecipitation using only 1282 in the absence of conditioned medium revealed a faint background A␤ signal that is due to trace amounts of the human synthetic A␤ immunogen carried in the 1282 rabbit serum (Fig. 6C). The levels of A␤ in the 24 -48 and 48 -72 h dyn K44A media were far above this background level. We also attempted to measure intracellular endogenous A␤ but found that the levels were below our limits of detection (data not shown). This measurement would require an alternative cell model in which transfected APP could yield more intracellular A␤.
To confirm unequivocally that the increased A␤ secretion we observed after dyn K44A induction represented bona fide en-dogenous A␤, we showed that it was abolished in the presence of the potent ␥-secretase inhibitor, Compound E. With Compound E in the conditioning medium from 24 to 48 h of induction, there was no detectable A␤ generation, confirming the ␥-secretase-mediated generation of A␤ (Fig. 6D). To verify that the increased A␤ we detected was due to higher secretion and not decreased degradation, we carried out specific A␤ degradation assays (22). Radiolabeled A␤ at physiological levels (40 pM) was incubated on intact HeLa cells, and the release of counts into the trichloroacetic acid-soluble fraction of conditioned medium was quantified following a brief incubation at 37°C. No change in A␤ degradation was detected between cells expressing and not expressing dyn K44A. 2

DISCUSSION
Here, we show that inhibiting endocytosis by expressing a dominant-negative dynamin protein results in the retention of endogenous human APP and C99 at the cell surface and a striking increase in the secretion of endogenous A␤. Consistent with previous reports (10, 12), we found that retention of APP at the cell surface also led to increased ␣-secretase cleavage of surface APP and enhanced secretion of the p3 fragment, which is generated by sequential ␣and ␥-secretase cleavages. We conclude that the normal pathway of APP processing includes receptor-mediated endocytic recycling and that down-regulating this recycling enhances amyloidogenic processing of APP at the cell surface to yield secreted A␤. Taken together, these results show for the first time that robust A␤ generation can occur directly on the plasma membrane.
Dynamin is involved in certain other processes that are not directly linked to receptor-mediated endocytosis in some cell types. For example, recent work suggests that the GTPase activity of dynamin is required for some budding events from the TGN, as well as for actin rearrangement and apoptosis (23). However, the HeLa cells that we used in this study have been well characterized to show no effect on the secretory pathway as regards the maturation and delivery of the transferrin receptor and the occurrence of budding events from the TGN to both the plasma membrane and lysosomes (18,19). In accord, our pulse-chase analysis revealed no significant change in the trafficking and maturation of newly synthesized APP through the secretory pathway in the presence versus absence of dyn 2 W. Farris, J. H. Chyung, and D. J. Selkoe, data not shown.
FIG. 5. C99 accumulates at the cell surface following dyn K44A induction. Intact HeLa cells were treated with or without 0.25% trypsin for 30 min at 4°C and quenched with 0.125% soybean trypsin inhibitor and excess culture medium. Cell surface trypsinization selectively removed mature APP from the cell surface (top panel). Tubulin, a cytoplasmic control protein, was not altered by cell surface trypsin treatment (middle panel). C99 levels were also elevated following dyn K44A induction, but were greatly reduced following surface trypsinization (lower panel).
K44A. Surface biotinylation and surface trypsinization experiments also confirm that the effects of dyn K44A on APP and its proteolytic products occur specifically at the cell surface. Thus, although dynamin can function at sites other than the cell surface, we have shown that APP and C99 are specifically increased at the cell surface following dyn K44A induction.
Based on the findings reported here, we hypothesize that modulating APP trafficking represents a physiological mechanism for regulating amyloidogenic APP processing and that modulating dynamin function, in particular, alters APP trafficking and processing. In support of this interpretation is work that shows that enhanced trafficking of APP to the neuronal surface from intracellular TGN stores as a result of insulin signaling can increase A␤ secretion and concomitantly decrease intracellular A␤ levels (24). In a previous study, Gasparini et al. (24) describe insulin effects on A␤ levels that may involve both insulin receptor signaling and the competitive inhibition of insulin degrading enzyme, a known A␤-degrading protease. In this regard, a probable mechanism of action of insulin signaling that could lead to the observed alteration of APP trafficking and processing is the known ability of insulin to enhance the phosphorylation of dynamin (25). Dynamin phosphorylation inhibits dynamin function (25), thus retarding the endocytosis of a broad range of surface molecules, including APP.
The phosphorylation status of dynamin is believed to be important for regulating synaptic vesicle recycling (26). Nerve terminal depolarization results in calcineurin activation, which in turn allows dephosphorylation of dynamin I (27), which would lead to increased endocytosis. Immediately following depolarization and stimulation of endocytosis, dynamin is rephosphorylated by protein kinase C in response to increased Ca 2ϩ concentrations (26,28). Thus, dynamin normally undergoes cycles of rapid phosphorylation and dephosphorylation, and our data predict that this would have an effect on APP processing at synaptic termini. In turn, any disease-related alterations in synaptic vesicle recycling might be predicted to alter APP proteolysis via dynamin dysfunction. In this regard, a recent report finding decreased dynamin I transcripts in Alzheimer's disease brains (29) suggests that broad dynamin dysfunction could occur, at least in the later stages of AD and that this could contribute to a worsening of extracellular A␤ accumulation and thus the disease process.
Studies involving the mutation or deletion of the cytoplasmic YENPTY domain of APP have suggested that accumulating APP at the cell surface reduces A␤ generation (11,12). These data are in direct contrast to our results. We provide here two possible explanations for the differences in results. First, although the above-mentioned studies show that the YENPTY sequence is important in targeting APP for endocytosis, this sequence is also important in binding cytosolic adaptor proteins, such as Fe65, Mints, disabled, and JIP-1 (30). Binding of Fe65 and Mint proteins is known to affect APP processing, with Fe65 binding increasing A␤ secretion and Mint-1 binding decreasing A␤ secretion (31)(32)(33). Additionally, Fe65 affects APP trafficking, delivering more APP to the plasma membrane when overexpressed (33). Thus, deleting or mutating the YENPTY sequence would have effects on APP processing independent of its actions on endocytosis, including decreased delivery of APP to the plasma membrane and reduced A␤ generation.
Second, dyn K44A inhibits both caveolin-mediated and clathrin-mediated endocytosis, whereas YENPTY mutants are predicted to affect only APP entry into clathrin-coated pits. Lipid rafts are enriched in APP and its processing enzymes, ␤and ␥-secretase (34,35), and disrupting lipid raft formation by depleting cholesterol potently inhibits A␤ production (36,37). A recent report argues that ␣-processing of APP occurs princi-FIG. 6. A␤ secretion is increased when dyn K44A is induced. A, cells were labeled with [ 35 S]Met from 24 -48 h of culture in ϩ Tet or Ϫ Tet medium. A␤ and p3 were immunoprecipitated using polyclonal antibody 1282 overnight at 4°C. A␤ and p3 were increased when dyn K44A was induced, and the bands co-migrated with the positive control for A␤ and p3 from APP-overexpressing HEK cells. B, conditioned media from 24 -48 h of 18 paired cultures (in three independent experiments) were analyzed by an ELISA using 2G3, a C-terminal A␤ antibody specific for Ab40, and 266, which detects A␤ residues 13-28. HeLa cells secreted a mean of 121 pg/ml A␤ X-40, and A␤ X-40 secretion increased to a mean of 365 pg/ml when dyn K44A was induced (p Ͻ 0.0001). C, media were collected from 0 -24, 24 -48, and 48 -72 h of ϩ Tet or Ϫ Tet treatment. Media were immunoprecipitated with 1282 and detected with 6E10. A control was run using only 1282 antiserum on unconditioned medium. There was detectable A␤ signal from the 1282 antiserum alone (see text). Levels of A␤ well above this background were present at 24 -48 h and 48 -72 h of Ϫ Tet induction only. D, using a ␥-secretase inhibitor, Compound E, the increase in A␤ following dyn K44A induction seen by immunoprecipitation/ Western blotting was reduced to background levels. Antibodies used were as in panel C. pally in non-lipid raft compartments, whereas ␤-processing of APP requires lipid rafts (38). These authors found that antibody-induced cross-linking of APP or BACE results in increased co-patching in lipid rafts at the plasma membrane and increased A␤ secretion. In addition, they observed that inhibiting endocytosis using dominant-negative variants of dynamin II or Rab5 led to decreased secretion of newly generated A␤ in pulse-chase experiments. Their model is different from ours in that they use overexpressed Swedish mutant APP, BACE, and dynamin-2, and they do not see increases in either full-length APP or C83. Despite these differences, we believe their result regarding decreased secretion of newly generated A␤ may be consistent with our finding of increased total A␤ secretion when dynamin function is inhibited, in that pulse-labeling a pool of APP in the dynamin-inhibited cultures would result in a competition for ␤and ␥-cleavage of newly synthesized APP molecules with many more unlabeled APP molecules retained at the plasma membrane. Thus, the time to generate A␤ from a given APP molecule should be longer, and in a short pulsechase experiment there would be relatively less radiolabeled A␤ secretion measured (38). Moreover, other data of Ehehalt and coworkers (38) provide direct evidence for the cell surface generation of A␤. In the context of blocked endocytosis (i.e. dynamin inhibition), they found that antibodies that copatched BACE and APP at the surface, particularly in lipid rafts, markedly increased A␤ secretion (see Fig. 7B of Ehehalt et al. (38)). Although this finding was not emphasized, the data clearly implicate the lipid-rich domains at the plasma membrane as sites capable of generating A␤ from APP. Therefore, increasing A␤ secretion from the cell surface likely requires an inhibition of lipid-raft-mediated endocytosis. More specifically, it probably requires a dynamin-dependent lipid-raft-mediated endocytosis pathway.
Despite the difference in effect on A␤ secretion between our dynamin inhibition experiments and those utilizing YENPTY mutants in APP, it is important to note that secretion of p3 (which is generated by ␣followed by ␥-secretase cleavage) was increased both in cells expressing C-terminal deletions of APP (10,12) and in our cells with blocked endocytosis. Because ␣-secretase cleavage of APP is not thought to require lipid rafts, our dynamin mutant may act similarly to a YENPTY mutant in the effects it has on APP conversion to p3 during clathrin-mediated endocytosis. In this regard, the results of both approaches strongly suggest that the ␥-secretase complex can be active at the cell surface. In fact, presenilin heterodimers have been biotinylated at the cell surface (39), as has endogenous, fully mature nicastrin (40), indicating that there is an available pool of the mature ␥-secretase complex at the cell surface. A recent study also determined that APP and presenilin are closely associated at or very near the plasma membrane using fluorescence lifetime imaging microscopy (41). Other substrates of ␥-secretase are also cleaved sequentially, first requiring ectodomain release by a metalloprotease. These substrates, including nectin-1␣, ErbB4, E-cadherin, Notch, CD44 and LRP, are all known to be cell surface proteins involved in intercellular adhesion or receptor signaling. The metalloprotease-dependent generation of p3 from APP at the cell surface is consistent with the processing of other known substrates of the ␥-secretase complex.
The pathway of APP processing uniquely important for Alzheimer's disease is the sequential cleavage by ␤and ␥-secretases to generate A␤. At this writing, no other ␥-secretase substrate is known to undergo two alternative ectodomain cleavage pathways, i.e. ␣versus ␤-cleavage. Understanding the functional importance of the ␤-cleavage pathway in APP processing is critical to assessing the therapeutic potential of targeting BACE. In our study, we found that accumulation of APP at the cell surface allows ␤-cleavage to occur and substantially enhances A␤ secretion. Therefore, the cell surface is a site for ␤-cleavage, suggesting that proteolysis of APP by BACE may itself require induction by an extracellular signal. Furthermore, our demonstration that APP is processed at cell surface has therapeutic implications, because drugs targeting A␤ generation, whether ␤or ␥-secretase inhibitors, would not necessarily need to penetrate to the interior of the cell. Given the insidious and progressive nature of the disease, a small (20 -30%) decrease in A␤ generation at the cell surface could have a profound impact on clinical outcome by lowering A␤ levels below the critical concentration needed for the formation of potentially synaptotoxic oligomers (42).