Secretion and Intracellular Generation of Truncated Aβ in β-Site Amyloid-β Precursor Protein-cleaving Enzyme Expressing Human Neurons*

Insoluble pools of the amyloid-β peptide (Aβ) in brains of Alzheimer's disease patients exhibit considerable N- and C-terminal heterogeneity. Mounting evidence suggests that both C-terminal extensions and N-terminal truncations help precipitate amyloid plaque formation. Although mechanisms underlying the increased generation of C-terminally extended peptides have been extensively studied, relatively little is known about the cellular mechanisms underlying production of N-terminally truncated Aβ. Thus, we used human NT2N neurons to investigate the production of Aβ11–40/42 from amyloid-β precursor protein (APP) by β-site APP-cleaving enzyme (BACE). When comparing undifferentiated human embryonal carcinoma NT2− cells and differentiated NT2N neurons, the secretion of sAPP and Aβ correlated with BACE expression. To study the effects of BACE expression on endogenous APP metabolism in human cells, we overexpressed BACE in undifferentiated NT2− cells and NT2N neurons. Whereas NT2N neurons produced both full-length and truncated Aβ as a result of normal processing of endogenous APP, BACE overexpression increased the secretion of Aβ1–40/42 and Aβ11–40/42 in both NT2− cells and NT2N neurons. Furthermore, BACE overexpression resulted in increased intracellular Aβ1–40/42 and Aβ11–40/42. Therefore, we conclude that Aβ11–40/42 is generated prior to deposition in senile plaques and that N-terminally truncated Aβ peptides may contribute to the downstream effects of amyloid accumulation in Alzheimer's disease.

tion of aggregated A␤ peptides in senile plaques and vascular deposits. A␤ has classically been described as a 4-kDa peptide, derived from proteolytic processing of APP, varying in length from 40 to 42 amino acids due to C-terminal heterogeneity. The generation of the longer A␤1-42 peptide is specifically increased by several mutations linked to familial Alzheimer's disease (1). Furthermore, A␤1-42 is more fibrillogenic than A␤1-40 in vitro (2), consistent with the finding that both diffuse and senile plaques are composed of primarily A␤ peptides that terminate at position 42 (3). However, several N-terminal truncations are also found in A␤ peptides derived from AD brains, demonstrated as early as the first biochemical isolation of A␤ peptides from senile plaques (4). The relative importance of N-terminally truncated A␤ peptides in the pathogenesis of AD is unknown. Interestingly, some individuals with sporadic AD (5), familial AD (5,6), and Down's syndrome (7) preferentially accumulate N-terminally truncated A␤ species. Also, overexpression of APP harboring disease-associated mutations within the A␤ domain (i.e. A692G and E693G) results in increased secretion of A␤11-40/42 (8,9).
Whereas some truncations may be due to the partial degradation of full-length A␤ after secretion of peptides into the extracellular milieu, A␤ peptides beginning with glutamine at position 11 are derived from the membrane-bound ␤-site APPcleaving enzyme (BACE) (10 -12). To generate full-length A␤, BACE cleaves APP between the methionine and aspartate at position 1 (Asp-1) of the N terminus of A␤ (␤-cleavage), resulting in the secretion of a large N-terminal ectodomain, sAPP␤, and the retention of a 99-amino acid C-terminal fragment, C99. To generate N-terminally truncated A␤, BACE cleaves APP between tyrosine and glutamate at position 11 (Glu-11) within the A␤ domain (␤Ј-cleavage), resulting in the secretion of a slightly larger N-terminal ectodomain, sAPP␤Ј, and the retention of an 89-amino acid C-terminal fragment, C89. C89 can also be produced by proteolysis of C99 by BACE at Glu-11 (11). Alternatively, APP may also be cleaved at position 16 by ␣-secretase (13,14), resulting in secretion of the N-terminal sAPP␣ along with the retention of an 83-amino acid C-terminal fragment, C83. Membrane-bound C-terminal fragments are subjected to further proteolysis within the transmembrane do-main by ␥-secretase, with cleavage typically occurring at either position 40 or 42 within the A␤ region. Whereas all of the components of ␥-secretase have not been identified, the presenilin proteins are necessary for secretion of A␤ peptides and have been postulated to contain the active site of ␥-secretase (1). However, multiple ␥-secretases may exist as presenilin is not needed for A␤1-42 production early in the secretory pathway (15).
Untransfected non-neuronal cells and murine cells are not well suited toward the study of BACE-derived N-terminally truncated A␤. Many non-neuronal cells preferentially use the ␣-secretase pathway at the expense of ␤-cleavage of APP (13,14), although the secretion of full-length and N-terminally truncated A␤ can be increased upon overexpression of either APP or BACE (11,12,16,17). Furthermore, due to low BACE activity and low APP expression, intracellular A␤ is difficult to detect in non-neuronal cells (16,18). In contrast, neuronal cells cleave a larger proportion of APP by BACE (19 -21), although overexpression of human APP in rodent cells does not lead to the generation of A␤11-40/42 from human APP (22), consistent with evidence that ␤Ј-cleavage is species-specific (23). Furthermore, rodent neuronal cells preferentially cleave endogenous APP at position 11, whereas the presence of ␤Ј-cleavage in human neuronal cells has not been adequately addressed. Therefore, rodent cell culture models may not accurately reflect the proteolytic processing of APP in human neurons.
NT2N neurons are a post-mitotic, terminally differentiated human neuronal cell culture model derived from retinoic acid treatment of the human embryonal carcinoma cell line NTera2/ c1.D1 (NT2Ϫ) (24 -27). Their high APP expression and ␤-secretase activity make them amenable to biochemical analysis of both neuronal specific and human-specific characteristics of APP processing (18,21). We have shown that NT2N neurons generate A␤1-40/42 intracellularly prior to secretion and that endogenous secretion of A␤1-40/42 from NT2N neurons increases with age in culture (19,20). Furthermore, a detergentinsoluble pool of intracellular A␤ accumulates with time in NT2N neurons (28). In this study, we found that sAPP and A␤ secretion increase upon neuronal differentiation of NT2N neurons, correlating with BACE expression. Given a recent report (29) that BACE protein expression and activity are increased in AD, we sought to refine further our understanding of APP metabolism in NT2N neurons by examining the effect of exogenous expression of BACE in undifferentiated NT2Ϫ cells and differentiated NT2N neurons. We found that the increase in A␤ production due to BACE overexpression was more pronounced in NT2N neurons than in non-neuronal NT2Ϫ cells. Furthermore, we found that A␤11-40/42 is produced endogenously by NT2N neurons and that BACE overexpression increases the secretion of both full-length and truncated A␤ peptides. We further demonstrate that A␤11-40/42 is generated intracellularly, indicating that A␤11-40/42 is produced prior to deposition in senile plaques. The effect of BACE expression on the generation of both full-length and N-terminally truncated A␤ underscores the role of BACE activity in the generation of A␤ peptides and indicates that N-terminally truncated A␤ peptides may contribute to the pathogenesis of AD.
Western Blot Analysis-Cell lysates were collected in RIPA buffer (0.5% sodium deoxycholate, 0.1% SDS, 1% Nonidet P-40, 5 mM EDTA in TBS, pH 8.0) in the presence of protease inhibitors (1 g/ml each of pepstatin A, leupeptin, L-1-tosylamido-2-phenylethyl chloromethyl ketone, 1-chloro-3-tosylamido-7-amino-2-heptanone, soybean trypsin inhibitor, and 0.5 mM phenylmethylsulfonyl fluoride) and briefly sonicated. Protease inhibitors were also added to conditioned media samples. Samples were centrifuged at 100,000 ϫ g for 20 min at 4°C, electrophoresed on 7.5% Tris-glycine acrylamide gels, and transferred to nitrocellulose. When indicated, samples were immunoprecipitated with Karen, a goat polyclonal antibody raised against sAPP, prior to electrophoresis. APP and total sAPP were probed with Karen. sAPP␣ and sAPP␤Ј were probed with Ban50, a mouse monoclonal antibody recognizing A␤ residues 1-10, or with NAB228, a mouse monoclonal antibody recognizing A␤ residues 1-11. 2 sAPP␤ was specifically probed with C5A4/2, a rabbit polyclonal antibody raised against a synthetic peptide (CSEVKM) corresponding to the C terminus of sAPP␤ (11). sAPP␤Ј was specifically probed with C10A4, a rabbit polyclonal antibody raised against a synthetic peptide (CHDSGY) corresponding to the C terminus of sAPP␤Ј. The specificity of C5A4 and C10A4 was determined by their lack of immunoreactivity with full-length APP or Cterminal APP fragments, and by blocking experiments in which only peptides with the corresponding free C terminus are able to block immunoreactivity (data not shown). Immunoblots were visualized by enhanced chemiluminescence (PerkinElmer Life Sciences) after application of species-specific horseradish peroxidase-conjugated anti-IgG antibodies (Santa Cruz Biotechnology, Santa Cruz, CA). However, for quantification of APP and sAPP, Karen immunoblots were labeled with 125 I-protein A (PerkinElmer Life Sciences) after application of a rabbit anti-goat IgG linker. Radiolabeled APP and sAPP were quantified using PhosphorImager analysis (Amersham Biosciences). For detection of BACE, crude membrane fractions were prepared by first collecting cells in hypotonic buffer (10 mM NaCl, 10 mM Tris, 1 mM EDTA, pH 7, protease inhibitors) followed by centrifugation at 100,000 ϫ g for 20 min at 4°C. After an additional wash with hypotonic buffer, membrane proteins were extracted by sonication of the pellet in 1 M NaCl, 40 mM Tris, 4 mM EDTA, protease inhibitors, 0.5% Triton X-100, pH 7. The fraction was cleared by another round of centrifugation, and the BACEcontaining supernatants were electrophoresed as above and immunoblotted with a rabbit polyclonal anti-BACE (CT) antibody (ProSci, Poway, CA).
Metabolic Labeling and Immunoprecipitation-Cells were incubated in methionine-free DMEM (Invitrogen) for 30 min, labeled with [ 35 S]methionine (250 Ci/ml in methionine-free DMEM supplemented with 5% dialyzed FBS; PerkinElmer Life Sciences) for 90 min, and chased for 1 h. Cells were treated with 10 M PMA in Me 2 SO (Sigma) during the chase period and/or with 10 M TAPI in Me 2 SO (Peptides International, Louisville, KY) 30 min prior to and throughout the chase period. Protease inhibitors were added to conditioned media, and sAPP was immunoprecipitated with Ban50 prior to electrophoresis on 7.5% Tris-glycine acrylamide gels. For C-terminal fragment analysis, cells were labeled for 2 h with [ 35 S]methionine in the presence of 200 M MG132 (Peptides International), rinsed, and lysed in 1000 l of RIPA buffer containing protease inhibitors for immunoprecipitation. Lysates were briefly sonicated, and both lysates and media were cleared by centrifugation at 100,000 ϫ g for 20 min at 4°C. C-terminal fragments were immunoprecipitated with 2493, a rabbit polyclonal antibody recognizing the C-terminal region of APP, and resolved on 10/16.5% step gradient Tris-Tricine gels. Gels were fixed in 50% methanol, 5% glycerol, dried, and exposed to PhosphorImager plates for visualization. Finally, to detect both secreted and intracellular A␤, two 10-cm dishes of replate 2 NT2N neurons were infected with recombinant Semliki Forest virus encoding wild type APP695, prepared as described previously (18,30). Cells were infected in serum-free medium for 1 h, cultured in complete growth medium for 14 h, and then labeled with [ 35 S]methionine (500 Ci/ml) for 8 h. Conditioned media and RIPA cell lysates were collected, cleared by centrifugation, and immunoprecipitated with 4G8 (Senetek, Maryland Heights, MO) prior to electrophoresis on 10/16.5% step gradient Tris-Tricine gels.
Northern Analysis-Total RNA was extracted from NT2Ϫ cells and NT2N neurons with the Trizol Reagent (Invitrogen) as per manufacturer's protocol. RNA concentrations were determined by optical density readings, and equal amounts of RNA were separated on 1% aga-rose-formaldehyde gels. RNA was transferred to a nitrocellulose membrane (Amersham Biosciences) and hybridized with a radiolabeled probe generated by either PstI digestion or AccI-HincII double digestion of a BACE cDNA. The blot was washed and exposed to a PhosphorImager plate for visualization. Glyceraldehyde-3-phosphate dehydrogenase levels were obtained by using a glyceraldehyde-3-phosphate dehydrogenase probe purchased from Ambion (Austin, TX). 28 S and 18 S ribosomes were visualized by staining a duplicate gel with ethidium bromide.
Stable Transduction of NT2Ϫ Cells-NT2Ϫ cells were stably transduced with a vesicular stomatitis virus surface glycoprotein (VSV-G) pseudotyped self-inactivating lentiviral vector. 3 To generate the virus, QBI 293A cells were plated on poly-D-lysine-coated 10-cm dishes. Cells were transfected with pMD.G (containing the VSV-G envelope glycoprotein), pCMV⌬R8.2 (containing viral structural, enzymatic, and accessory genes), and SIN-EFp-GFP/SIN-EFp-BACE (containing minimal human immunodeficiency virus-based viral sequences, the elongation factor 1-␣ promoter, and either a GFP or BACE cDNA) using standard CaPO 4 techniques. Media conditioned with viral particles were harvested over 3 days, centrifuged at 1000 rpm for 5 min, and passaged through a 0.45-m filter to remove any cellular debris. Viral supernatants were added to NT2Ϫ cells, and transduced NT2Ϫ cultures were subcloned by limited dilution into 96-well plates. Uniform expression was verified either by direct fluorescence for GFP-expressing cells or indirect immunofluorescence with BaceN1, a rabbit polyclonal antibody raised against the N terminus of BACE (32), for BACE-expressing cells.
Immunoprecipitation/Mass Spectrometry-Media conditioned for 10 days were collected in the presence of protease inhibitors and cleared by centrifugation at 100,000 ϫ g for 20 min at 4°C. Media were immunoprecipitated with 4G8, and immunoprecipitated material was eluted with a saturated solution of ␣-cyano-4-hydroxycinnamic acid in 0.1% trifluoroacetic acid, 50% acetonitrile. Data were collected on an ABI/ Perspective (Framingham, MA) Voyager DE-PRO MALDI-TOF instrument in the positive-ion mode at the Protein Microchemistry/Mass Spectrometry Facility of the Wistar Institute (Philadelphia). Samples were spotted to a 100-well plate using ␣-cyano-4-cinnamic acid matrix (Sigma) at 10 mg/ml. Reflector mode with the accelerating potential at 20 kV was used. External calibration was performed on all samples.

RESULTS
sAPP and A␤ Production Correlates with Neuronal Differentiation-APP processing differs between non-neuronal and neuronal cells, underscoring the importance of using neuronal systems to study APP metabolism and A␤ generation (18 -20, 28). Prior to studying BACE activity and ␤Ј-cleavage in NT2N neurons, we quantified the differences in the expression and proteolytic processing of APP between NT2Ϫ cells and NT2N neurons. Whereas both NT2Ϫ cells and NT2N neurons express high levels of APP, the two cell types express different isoforms of APP (Fig. 1A). We have shown previously (19) that NT2Ϫ cells predominantly express the 751-and 770-amino acid isoforms of APP (APP751/770) that appear as a doublet corresponding to immature (ϳ110 kDa) and mature N-and Oglycosylated APP (ϳ125 kDa). NT2N neurons predominantly express the shorter 695-amino acid isoform of APP (APP695), the majority of which is immature APP695 (ϳ95 kDa) with relatively less mature APP695 (ϳ110 kDa). Despite the difference in isoform expression, densitometric quantification indicated that when normalized for total protein content, NT2Ϫ cells expressed APP at 98% Ϯ 5 (S.E.) compared with NT2N neurons. However, despite equivalent total APP expression, we found that proteolytic processing of APP was more efficient in NT2N neurons compared with NT2Ϫ cells. To measure total sAPP, derived from both ␣and ␤-secretase cleavage of APP, media conditioned by NT2Ϫ cells or NT2N neurons for 24 h were immunoprecipitated with a polyclonal antibody raised  lane). B, after normalization to intracellular fulllength APP expression levels, total sAPP was purified from media conditioned from either NT2Ϫ cells or replate 3 NT2N neurons for 24 h by immunoprecipitation with Karen, separated on a 7.5% Tris-glycine gel, and immunoblotted with Karen. C, total sAPP was quantified by Karen immunoblots of conditioned media, analyzed by PhosphorImager, corrected for intracellular APP expression, and shown as mean values relative to NT2Ϫ cells Ϯ S.E. One-way analysis of variance revealed p Ͻ 0.0001; post hoc analysis showed p Ͻ 0.01 (*) compared with NT2Ϫ cells. D, A␤1-40 levels were quantified by Ban50/BA-27 ELISA, normalized to intracellular APP expression levels, and shown as mean values relative to NT2Ϫ cells Ϯ S.E. One-way analysis of variance revealed p value of 0.0015. Post hoc analysis showed p Ͻ 0.01 (*) compared with NT2Ϫ cells.
against the N-terminal domain of APP. We found that NT2N neurons secreted more total sAPP relative to NT2Ϫ cells (Fig.  1B). Densitometric quantification of total sAPP indicated that NT2N neurons secreted over 5-fold more sAPP than NT2Ϫ cells (Fig. 1C). Differentiation had an even larger effect on A␤ secretion, as NT2N neurons secreted over 20-fold more A␤ than NT2Ϫ cells (Fig. 1D), determined by sandwich ELISA for A␤1-40. Because differences in APP expression cannot explain the increase in A␤ secretion, differences in either ␤or ␥-secretase activity are likely to be responsible for the more efficient proteolysis of APP in NT2N neurons.
TAPI-insensitive sAPP Production in NT2N Neurons-The increased production of sAPP secretion by NT2N neurons indicated that BACE and/or ␣-secretase activity might be elevated in NT2N neurons upon neuronal differentiation of the NT2Ϫ cells. To distinguish between these possibilities, we first addressed the presence of ␣-secretase by testing the pharmacologic responses of NT2Ϫ cells and NT2N neurons upon ␣-secretase inhibition. ␣-Secretase has been attributed to members of a family of proteases that contain a disintegrin and a metalloprotease domain (ADAM), including ADAM10 and tumor necrosis factor-␣ converting enzyme. ␣-Cleavage is enhanced by phorbol ester-induced stimulation of tumor necrosis factor-␣ converting enzyme via protein kinase C and can be inhibited by metalloprotease inhibitors. Therefore, we treated metabolically labeled NT2Ϫ cells and NT2N neurons with either phorbol 12-myristate 13-acetate (PMA) or the specific metalloprotease inhibitor, TAPI. sAPP from media samples was immunoprecipitated using Ban50, a monoclonal antibody that recognizes the first 10 amino acids of A␤, thereby recognizing both sAPP␣ and sAPP␤Ј. In NT2Ϫ cells, PMA treatment increased sAPP secretion by 2.15 Ϯ 0.29-fold, whereas TAPI treatment inhibited sAPP production to 0.54 Ϯ 0.04-fold of untreated NT2Ϫ cells (Fig. 2, A and B). The magnitude of sAPP inhibition in NT2Ϫ cells was comparable with that reported for other non-neuronal cells (38). Thus we concluded that ␣-secretase activity is present in NT2Ϫ cells, and a large proportion of the base-line sAPP secreted by NT2Ϫ cells is derived from ␣-secretase. Whereas NT2N neurons also exhibited PMA-induced up-regulation of ␣-secretase activity (1.80 Ϯ 0.37), TAPI had no effect on sAPP production (0.95 Ϯ 0.14; Fig. 2, A and B). Furthermore, in both NT2Ϫ cells and NT2N neurons, TAPI was able to prevent PMA-induced sAPP␣ production, demon-strating that TAPI is able to inhibit PMA-induced ␣-secretase in both non-neuronal and neuronal cells. However, the inability of TAPI to decrease sAPP generation by NT2N neurons below base-line levels indicated that a relatively small proportion of APP is normally proteolyzed by ␣-secretase in NT2N neurons.
Because Ban50 recognizes both sAPP␣ and sAPP␤Ј, the ability of Ban50 to immunoprecipitate TAPI-insensitive sAPP from NT2N neurons suggested that NT2N neurons may produce sAPP␤Ј endogenously. Therefore, to demonstrate more directly the presence of BACE-derived sAPP fragments, we used a panel of antibodies that recognize different sAPP species to analyze media conditioned for 24 h by NT2Ϫ cells and NT2N neurons. To better compare the relative abundance of sAPP species, media samples were corrected for either lysate protein concentration or intracellular full-length APP levels prior to analysis. Despite preferential utilization of ␣-cleavage over ␤-cleavage in non-neuronal cells (13,14), we detected less sAPP with Ban50 from NT2Ϫ cells compared with NT2N neurons (Fig. 2C, top panel), consistent with the possibility that NT2N neurons produce sAPP␤Ј endogenously. This result was confirmed with a second monoclonal antibody, NAB228, that recognizes the first 11 amino acids of A␤ (Fig. 2C, 2nd panel). C5A4/2, a polyclonal antibody that specifically recognizes the C terminus of sAPP␤, showed the presence of sAPP␤ in media conditioned by NT2N neurons (Fig. 2C, 3rd panel). In contrast, the amount of sAPP␤ in media conditioned by NT2Ϫ cells was undetectable. Finally, a polyclonal antibody that specifically recognizes the C terminus of sAPP␤Ј demonstrated the presence of endogenous sAPP␤Ј produced by NT2N neurons (Fig.  2C, bottom panel). Therefore, not only is ␣-secretase activity relatively low in NT2N neurons, as determined pharmacologically, but BACE cleavage at both Asp-1 and Glu-11 is readily detected as a product of normal APP metabolism from NT2N neurons.
BACE Expression in NT2Ϫ Cells and NT2N Neurons-To better understand the secretion and intracellular generation of BACE-derived A␤ peptides in human neurons, we characterized BACE expression in NT2Ϫ cells and NT2N neurons. Expression of BACE mRNA was determined by Northern analysis of NT2Ϫ cells and NT2N neurons. BACE mRNA expression was clearly present in NT2N neurons, as seen by the presence of 7.0-, 4.4-, and 2.6-kb bands (Fig. 3A). NT2Ϫ cells, however, showed markedly less BACE expression, most notably demon-  (39,40), was somewhat reduced in NT2Ϫ cells relative to NT2N neurons. Two different radiolabeled BACE cDNA fragment probes yielded the same results ( Fig. 3A and data not shown).
In agreement with the Northern analysis, BACE protein was undetectable in NT2Ϫ cell lysates, whereas NT2N neuron lysates exhibited faint immunoreactivity (Fig. 3B). This suggested that low BACE expression limits A␤ production in NT2Ϫ cells. To determine the effect of BACE expression on A␤ production from endogenous APP, NT2Ϫ cells were stably transduced using a pseudotyped self-inactivating lentiviral vector encoding BACE or GFP. This viral vector is non-toxic to NT2Ϫ cells and NT2N neurons and results in stable transgene expression in differentiated neurons for over 5 months in vivo. 3 Several subclones uniformly expressing GFP or BACE were isolated (data not shown). As a control, GFP-expressing NT2G7Ϫ cells were selected to ensure that the process of transduction and subcloning did not alter APP processing. Two subclones, NT2B30Ϫ and NT2B17Ϫ, were chosen to represent low and high BACE-expressing clones, respectively (Fig. 3B), and to control for transgene insertion effects. Subclones retained their ability to differentiate into neurons upon retinoic acid exposure, and transgene overexpression was maintained in differentiated neurons (Fig. 3B).
To assay BACE cleavage at Asp-1, NT2Ϫ and NT2N conditioned media were immunoprecipitated with polyclonal antisera to the APP ectodomain to collect total sAPP. The immunoprecipitate was then subjected to SDS-PAGE and immunoblotted with C5A4/2, a rabbit polyclonal antibody specific for the C terminus of sAPP␤ (Fig. 3C). NT2Ϫ cells did not secrete appreciable levels of sAPP␤, consistent with the fact that BACE expression is nearly absent in NT2Ϫ cells. Overexpression of BACE in NT2Ϫ cells, however, resulted in the detection of sAPP␤ in a dose-dependent manner. Unlike untransduced NT2Ϫ cells, sAPP␤ was detected from untrans-duced NT2N neuron-conditioned media. The shift in electrophoretic mobility between sAPP␤ recovered from BACEexpressing NT2Ϫ cells and NT2N neurons reflects the shorter isoform of APP expressed in NT2N neurons. Furthermore, sAPP␤ secretion by NT2B30N and NT2B17N neurons, which express exogenous BACE as a consequence of retroviral transduction, was increased relative to control NT2N neurons. In addition to increased cleavage at Asp-1, we determined that ␤Ј-cleavage at Glu-11 was also present in BACE-expressing cells by immunoprecipitating C-terminal APP fragments from metabolically labeled NT2Ϫ cells, NT2N neurons, and BACE subclones with 2493, a polyclonal antibody raised against the C terminus of APP. BACE overexpression resulted in increased C99 and C89 relative to control cultures in both undifferentiated cells and neurons, corresponding to increased proteolysis of endogenous APP at residues Asp-1 and Glu-11, respectively. Significantly, C89 levels increased in a dose-dependent manner in that higher BACE-expressing clones had higher C89 levels. GFP expression was found to have no effect on sAPP␤ secretion or C-terminal fragment production in both undifferentiated cells and neurons (data not shown).
Secretion of Full-length and N-terminally Truncated A␤-Given the differences in endogenous BACE expression between NT2Ϫ cells and NT2N neurons, we sought to determine the effect of overexpressing BACE on A␤ production in these cells and to demonstrate the presence of N-terminally truncated A␤ peptides derived from BACE cleavage of APP at Glu-11. Media conditioned for 24 h were subjected to sandwich ELISA analysis and normalized for APP expression. Two sandwich ELISA systems were used in which either Ban50 (anti-A␤1-10) or BNT77 (anti-A␤11-28) monoclonal antibodies were used to capture A␤. The Ban50 ELISA was used to detect full-length A␤, whereas the BNT77 ELISA was used to detect both fulllength and N-terminally truncated A␤ species. Both ELISA systems demonstrated the production of A␤ in NT2Ϫ cells upon the overexpression of BACE (Fig. 4A) at levels similar to that secreted by untransduced NT2N neurons. However, the effect of BACE overexpression was more pronounced in NT2N neurons, increasing A␤ secretion 5-8-fold over untransduced NT2N neurons. Furthermore, the differences in A␤ concentration as detected by Ban50 or BNT77 indicated that a large proportion of secreted A␤ peptides are N-terminally truncated. To identify N-terminally truncated A␤ peptides secreted by NT2N neurons, conditioned media were immunoprecipitated by 4G8 and subjected to MALDI-TOF mass spectrometry.  Fig. 4D), consistent with the ELISA data. Other peaks corresponding to C-terminally truncated A␤ A␤1-34, A␤1-37, A␤1-38, and A␤1-39 with masses of 3784.71, 4076.02, 4132.14, and 4230.10 Da, respectively) were also identified in the mass spectra of neuronal medium (Fig. 4, C and D). However, since these C-terminally truncated A␤ peptides contain intact N termini, the increased concentration detected by BNT77 is primarily due to the presence of A␤11-40/42 in culture media.
Intracellular Generation of Truncated A␤-Previous analysis of several cell lines indicated that both high expression of APP and high ␤-secretase activity were co-requisites for intracellular A␤ detection (18,28). We therefore tested whether BACE overexpression increases intracellular A␤ in NT2N neurons. Because the BNT77 ELISA did not have the sensitivity required for accurate intracellular A␤ quantification, a more FIG. 3. BACE expression in NT2؊ cells and NT2N neurons. A, mRNA from NT2Ϫ cells and replate 3 NT2N neurons were electrophoresed on formaldehyde-agarose gels and hybridized with a radiolabeled PstI BACE cDNA fragment and visualized by PhosphorImager. Equal mRNA loading was determined by glyceraldehyde-3-phosphate dehydrogenase (GAPDH) expression and by 28 S and 18 S ribosome levels. B, crude membrane fractions from stably transduced NT2Ϫ cells and replate 2 NT2N neurons expressing BACE (NT2Ϫ/NT2B30Ϫ/ NT2B17Ϫ, undifferentiated cells; NT2N/NT2B30N/NT2B17N, neurons) were separated on a 7.5% Tris-glycine gel and immunoblotted for BACE. C, total sAPP from media conditioned for 24 h from NT2Ϫ cells or replate 3 NT2N neurons was immunoprecipitated with Karen, electrophoresed on a 7.5% Tris-glycine gel, and immunoblotted for sAPP␤ with C5A4/2, showing sAPP␤ derived from either APP751/770 or APP695. D, C-terminal APP fragments were immunoprecipitated with 2493 from NT2Ϫ cells and replate 2 NT2N neurons that were metabolically labeled for 2 h, separated on a 10/16.5% discontinuous gradient Tris-Tricine gel, and exposed to a PhosphorImager screen. The bands corresponding to C99, C89, and C83 are labeled.
sensitive ELISA was utilized in which A␤ peptides from cell lysates were captured with either JRF/cA␤40 or JRF/cA␤42, specific for A␤40 and A␤42, respectively. Captured peptides were detected with either JRF/A␤N, recognizing A␤1-7 (to measure full-length A␤) or m266, recognizing A␤13-28 (to measure full-length and N-terminally truncated A␤). BACE overexpression resulted in the presence of intracellular A␤ from undifferentiated NT2B30Ϫ and NT2B17Ϫ cells, in contrast with NT2Ϫ and NT2G7Ϫ cells (Fig. 5). The amount of intracellular A␤ from BACE-expressing non-neuronal cells was comparable with the amount of intracellular A␤ from untransduced NT2N neurons. Similar to secreted A␤, BACE overexpression resulted in a marked increase in intracellular A␤ in NT2N neurons. Furthermore, N-terminally truncated A␤ peptides comprised a large proportion of intracellular A␤. The difference between A␤ concentration as determined by JRF/ A␤N and m266 indicated that N-terminally truncated A␤ is present intracellularly not only in BACE-overexpressing cells but in untransduced NT2N neurons as a result of normal metabolism of endogenous APP.
To demonstrate further the presence of intracellular N-terminally truncated A␤ species, NT2N, NT2B30N, and NT2B17N neurons were metabolically labeled, and both media and cell lysates were immunoprecipitated with 4G8. Due to the lower sensitivity of 4G8 immunoprecipitation, we overexpressed APP in NT2N neurons to increase the production of A␤. As shown in Fig. 6A, overexpression of APP in NT2N neurons resulted in a large increase in secreted A␤. However, by using both Ban50 and BNT77 ELISAs, we found that the relative amount of truncated A␤ secreted by NT2N neurons upon APP overexpression was reduced. That is, although 24.5% of A␤ produced from endogenous APP is truncated, only 8.7% of A␤ is truncated upon APP overexpression. Despite this relative de-crease in truncated A␤, immunoprecipitates from NT2N media revealed three bands (Fig. 6B) corresponding to A␤1-40/42 (upper 4-kDa band), A␤11-40/42 (middle 3.2-kDa band), and p3 (A␤17-40/42, lower 3-kDa band). As expected, A␤1-40/42 and A␤11-40/42 were increased in a dose-dependent manner upon BACE expression, consistent with both the ELISA and mass spectral analysis. Immunoprecipitates from neuronal lysates demonstrated that full-length A␤ increased as a result of BACE expression. Furthermore, A␤11-40/42 was also recovered from BACE-expressing NT2N cell lysates in a dose-dependent manner, as shown by the presence of a 3.2-kDa band that co-migrated with the A␤11-40/42 recovered from media. Importantly, p3 was not detected from NT2N neuron lysates, indicating that p3 does not accumulate intraneuronally and that lysates were not contaminated with media. Thus, although the relative amount of truncated A␤ recovered by 4G8 immunoprecipitation was altered by APP overexpression, these results nonetheless confirm that N-terminal heterogeneity of A␤ is part of the intracellular processing pathway of APP and not solely due to partial degradation of secreted A␤ peptides.

DISCUSSION
Insoluble amyloid deposits from AD brains are heterogeneous in morphology and composition. Although the seeding and maturation of senile plaques in vivo is not well understood, increased ␤-cleavage has been implicated in the pathogenesis of AD, either due to a pathogenic mutation (K595N/M596L, APPsw) at the ␤-cleavage site of APP in familial AD (41)(42)(43) or by increased BACE expression in sporadic AD (29). Alternatively, other familial AD-associated mutations in either APP or presenilin increase the production of A␤1-42 relative to A␤1-40 (1). Despite the predominance of A␤1-40 in cerebral spinal fluid (CSF) (44), A␤1-42 is more abundant in senile FIG. 4. BACE expression increases full-length and N-terminally truncated A␤ secretion. A, media conditioned for 24 h by NT2Ϫ cells, NT2N neurons (5-week-old replate 2 neurons) and stably transduced subclones (NT2G7Ϫ/NT2B30Ϫ/NT2B17Ϫ, undifferentiated cells; NT2G7N/ NT2B30N/NT2B17N, 5-week-old replate 2 neurons) were assayed for A␤ production by sandwich ELISA. Full-length A␤ was measured by Ban50 ELISA, whereas full-length and N-terminally truncated A␤ was measured by BNT77 ELISA, shown as mean values corrected for APP expression Ϯ S.E. from six cultures over three independent collections. A control mixture of A␤1-40 and A␤11-40 (B), NT2N conditioned media (C), and NT2B30N conditioned media (D) were immunoprecipitated with 4G8 and subjected to MALDI-TOF analysis to identify different secreted A␤ species. Several peaks corresponding to C-terminally truncated A␤ peptides, in addition to A␤1-40 and A␤11-40, are labeled.
plaques (3), consistent with its ability to aggregate more readily than A␤1-40 in vitro (2). A third class of familial AD mutations, located in the middle of the A␤ domain, has been postulated to increase the amyloidogenicity of A␤ peptides (9). Interestingly, these mutations also appear to increase the production of A␤11-40/42 (8,9). Similar to C-terminal extensions to A␤, N-terminal truncations have been shown to reduce solubility although increasing sedimentation and ␤-pleated sheet structure of A␤ peptides relative to full-length A␤ (45)(46)(47). Furthermore, cyclization of the N-terminal glutamate in A␤11-40/42 protects the peptide from degradation by most aminopeptidases (48). Therefore, N-terminally truncated A␤ peptides may accelerate the seeding and maturation of senile plaques and thus exacerbate the progression of Alzheimer's disease, particularly in some genetic settings.
Until recently, the mechanism whereby A␤ peptides are Nterminally truncated has been unclear. A␤ peptides beginning at Glu-11 were first identified from purification of A␤ peptides from human CSF (37) and are found in insoluble fractions from AD brain (5,12). The discovery of BACE led to the realization that two alternative cleavage sites are present in the N-terminal region of A␤ and that ␤Ј-cleavage is species-specific (10,23). Therefore, although rodent neuronal cells preferentially cleave endogenous APP at position 11 (22,49), overexpression of human APP in rodent cells does not result in the formation of human A␤11-40/42 (22). Many modified A␤ peptides accumulate in brains of tg2576 mice, a transgenic mouse model overexpressing APPsw, including isomerized Asp-1 (L-iso-Asp), stereoisomerized Asp-1 (rectus Asp), and pyroglutaminated Glu-3. However, pyroglutaminated Glu-11 is conspicuously absent from tg2576 brains (50). Therefore, although many of the mechanisms for N-terminal modification of human A␤ are present, the generation of A␤11-40/42 is currently missing from both rodent cell culture and transgenic models. Given the limitations of rodent models, we investigated the generation of A␤11-40/42 in human NT2N neurons. We found that the secretion of sAPP and A␤ correlates with the expression of BACE that occurs upon neuronal differentiation. The increased secretion of sAPP was predominantly due to the secretion of sAPP␤ and sAPP␤Ј. Additionally, NT2N neurons produce the N-terminally truncated A␤11-40/42 from normal metabolism of endogenous APP. Furthermore, exogenous BACE expression increased the secretion and intracellular generation of both A␤1-40/42 and A␤11-40/42. Interestingly, increasing APP expression decreased the relative amount of truncated A␤ produced by NT2N neurons. In contrast, non-neuronal cells with higher levels of BACE overexpression than reported here resulted in the preferential generation of N-terminally truncated A␤ over fulllength A␤ (11,12). Taken together, the ratio of APP to BACE expression may dictate the extent of ␤Ј-cleavage. Regardless, the intracellular generation of A␤11-40/42 from normal APP processing in NT2N neurons indicates that this N-terminally truncated peptide is generated prior to deposition into insoluble aggregates in AD.
Additional A␤ peptides were recovered from NT2N neuron medium with truncated C termini. These C-terminally truncated A␤ species are also found in human CSF (51) and AD brain homogenates (5,12), indicating that they may also contribute to amyloid formation. The close correlation between A␤ peptides found in NT2N neuronal medium and human CSF further validates the NT2N neuronal culture system as a useful model to study the generation of N-and C-terminally truncated A␤ peptides. Immunoprecipitation of intracellular A␤ from BACE-expressing NT2N neurons yielded a faint band slightly smaller than full-length A␤ (see Fig. 6). Although obscured somewhat by the intense signal derived from full-length A␤, this truncated A␤ peptide appeared to be present in NT2N media samples, indicating that it corresponds to one of the C-terminally truncated A␤ peptides identified by mass spectrometry. APP and presenilin mutations that are known to affect C-terminal ␥-secretase cleavage also result in the increased accumulation of N-terminally truncated A␤ peptides (5,6), indicating that ␤and ␥-secretase cleavage may influence each other. Importantly, the magnitude of the increase in A␤ production upon BACE expression indicates that the level of endogenous BACE expression in NT2Ϫ cells and NT2N neurons is rate-limiting in terms of A␤ generation. Although ␥-secretase activity was not addressed directly in these experiments, the modest effect of BACE expression in non-neuronal NT2Ϫ cells compared with the effect of BACE expression in NT2N neurons indicates that ␥-secretase cleavage is enhanced in neurons.
Although secretion of A␤ from neuronal cells is high relative to non-neuronal cells, the concentration of A␤ in human CSF is below the threshold for A␤ aggregation in vitro (2). The stability and insolubility of intraneuronal A␤ lead to the hypothesis FIG. 5. Intracellular accumulation of truncated A␤ peptides. A␤ from NT2Ϫ cells, NT2N neurons (5-week-old replate 2 neurons), and stably transduced subclones (NT2G7Ϫ/NT2B30Ϫ/NT2B17Ϫ, undifferentiated cells; NT2G7N/NT2B30N/NT2B17N, 5-week-old replate 2 neurons) was extracted with RIPA and assayed by sandwich ELISA for intracellular A␤. Full-length A␤ was measured by JRF/A␤N ELISA, whereas full-length and N-terminally truncated A␤ was measured by m266 ELISA, shown as mean values Ϯ S.E. from four to six cultures over three independent collections. Undifferentiated cultures were corrected for expression of APP, whereas neuronal cultures were corrected for expression of neuronal specific enolase.
that intracellular A␤ may be the source of A␤ aggregates that seed senile plaques. Indeed, insoluble A␤ accumulates in NT2N neurons with age in culture (28), and SDS-stable oligomeric A␤ is found intracellularly prior to secretion (52). Furthermore, prior to the presence of amyloid pathology, A␤ can be detected biochemically from tg2576 mice (50) and patients with early cognitive dysfunction (53). Intracellular A␤ has been found in affected brain regions in AD brains (54 -56) and in animal models of AD amyloid pathology (57-60). Finally, mRNA iso-lated from senile plaques is predominantly neuronal (61). These reports suggest that the nidus for senile plaque formation may be intraneuronal A␤. Interestingly, although various N-terminally truncated A␤ species, including A␤11-40/42, are readily detected from detergent-insoluble preparations from AD brain, p3 is not detected (5). p3 is a major component of diffuse plaques in AD (62,63) and in diffuse plaques in the cerebellum of Down's syndrome patients (64). However, cerebellar diffuse plaques of Down's syndrome patients do not progress to form neuritic senile plaques even though in vitro studies of the p3 peptide indicate that it is highly hydrophobic, capable of forming fibrils, and has the tinctoral properties of amyloid as determined by thioflavin T and Congo Red staining (64). We could not detect intracellular p3 from NT2N neurons, consistent with the generation of p3 at or near the plasma membrane (65,66). These observations are also consistent with the hypothesis that the intracellular environment is necessary to convert fibrillogenic A␤ peptides into a nidus for senile plaque formation.
Multiple subcellular sites are responsible for the production of different A␤ peptides. The trans-Golgi network produces predominantly A␤1-40 (67,68), although the endoplasmic reticulum/intermediate compartment produces A␤1-42 (28,31,68,69). Endoplasmic reticulum/intermediate compartment-derived A␤1-42 is not secreted but rather is retained intracellularly and contributes to the accumulation of a pool of insoluble A␤ that can be recovered with formic acid. Unfortunately, the sandwich ELISAs used in this study either do not have the sensitivity (BNT77) or are incompatible (JRF/A␤N and m266) with formic acid lysates. However, the increased production of intracellular A␤ upon BACE expression is expected to increase the accumulation of insoluble full-length and truncated A␤ peptides. Production of endoplasmic reticulum/intermediate compartment-derived A␤ is independent of presenilin, indicating that multiple ␥-secretases may responsible for ␥-secretase cleavage in different subcellular organelles (15). In contrast, BACE-deficient untransduced NT2Ϫ cells do not have appreciable levels of intracellular A␤, although BACE overexpression increases intracellular A␤. Therefore, BACE appears to be responsible for both secreted and intracellular A␤. The extent of ␤Ј-cleavage is also dependent on the subcellular localization of BACE and APP in 293 cells (12). The engineering of BACEoverexpressing NT2N neurons allows for future investigations into the subcellular site of A␤11-40/42 generation in neuronal cells. However, the downstream effect of A␤11-40/42 generation on plaque formation awaits the engineering of transgenic mice co-expressing human BACE and human APP.
Increased BACE expression has been implicated in the pathogenesis of AD (29). Interestingly, BACE expression had a more profound effect on A␤ generation in NT2N neurons than in non-neuronal NT2Ϫ cells. Therefore, even modest increases in BACE expression may precipitate amyloid formation due to overproduction of A␤. Conversely, mild inhibition of BACE activity may have a large effect on A␤ generation, underscoring the possibility of using BACE inhibitors as a therapy for AD. However, given the endogenous production of A␤11-40/42 by human NT2N neurons, the effect of BACE inhibitors on both full-length and N-terminally truncated A␤ peptides needs to be determined.  NT2N, NT2B30N, and NT2B17N neurons were transduced with Semliki Forest virus to overexpress APP695 and metabolically labeled for 8 h. Media and RIPA lysates were immunoprecipitated with 4G8, electrophoresed on a 10/16.5% discontinuous gradient Tris-Tricine gel, and exposed to a PhosphorImager screen. Fulllength APP, C-terminal APP fragments, full-length A␤1-40/42, truncated A␤11-40/42, and p3 are labeled. The contrast of the bottom of the gel has been enhanced to demonstrate the increase in A␤ production due to BACE overexpression. The contrast was further enhanced (bottom) to demonstrate the presence of N-terminally truncated A␤.