Intracellular amyloid-beta 1-42, but not extracellular soluble amyloid-beta peptides, induces neuronal apoptosis.

Alzheimer disease (AD), the most frequent cause of dementia, is characterized by an important neuronal loss. A typical histological hallmark of AD is the extracellular deposition of β-amyloid peptide (Aβ), which is produced by the cleavage of the amyloid precursor protein (APP). Most of the gene mutations that segregate with the inherited forms of AD result in increasing the ratio of Aβ42/Aβ40 production. Aβ42 also accumulates in neurons of AD patients. Altogether, these data strongly suggest that the neuronal production of Aβ42 is a critical event in AD, but the intraneuronal Aβ42 toxicity has never been demonstrated. Here, we report that the long term expression of human APP in rat cortical neurons induces apoptosis. Although APP processing leads to production of extracellular Aβ1–40 and soluble APP, these extracellular derivatives do not induce neuronal death. On the contrary, neurons undergo apoptosis as soon as they accumulate intracellular Aβ1–42 following the expression of full-length APP or a C-terminal deleted APP isoform. The inhibition of intraneuronal Aβ1–42 production by a functional γ-secretase inhibitor increases neuronal survival. Therefore, the accumulation of intraneuronal Aβ1–42 is the key event in the neurodegenerative process that we observed.

␤-secretase (7) and ␥-secretase activities allow the release of A␤. Several identified mutations in the APP and the presenilins genes segregate with inherited forms of AD known as early onset familial Alzheimer disease or FAD (8,9). Most of these mutations result in an increased production of the A␤ ending at position 42 (10). In vitro studies have shown that A␤42 rapidly aggregates into fibrils and that extracellular fibrillar A␤ peptides induce apoptosis in cultured neurons (11). On the other hand, recent reports have demonstrated an intraneuronal accumulation of A␤42 in AD-vulnerable brain regions (12,13). Intraneuronal A␤42 accumulation has also been reported in transgenic mice expressing FAD proteins (14) as well as in transgenic mice showing accelerated neurodegeneration without extracellular amyloid deposition (15). Altogether, these data support the idea that A␤42 accumulation and neuronal loss are closely correlated. Nevertheless, the direct link between A␤ production by neurons and neuronal death has not been clearly established until now.
Here, we report that the long term expression of human APP in rat-cultured neurons induces apoptosis. To understand how APP expression and processing modify neuronal survival, we characterized the extracellular and intracellular A␤ isoforms produced by the processing of different APP constructs. We further demonstrated that APP-induced neuronal apoptosis depends on intraneuronal A␤1-42 accumulation.

EXPERIMENTAL PROCEDURES
Cell Cultures and Reagents-Primary cultures of cortical neurons were prepared from 17-day-old Wistar rat embryos as described previously (16). Cells were plated in 6-or 96-well culture dishes (4 ϫ 10 5 cells/cm 2 ) or glass coverslips (1.25 ϫ 10 5 cells/cm 2 ) pretreated with poly(L-lysine) (10 g/ml in phosphate-buffered saline) and cultured for 6 days in vitro in NEUROBASAL™ medium supplemented with 2% B-27 and 0.5 mM L-glutamine prior to infection with recombinant adenoviruses. Under these conditions, neuronal cultures (up to 98% of neurons) display high differentiation and survival rates (17). Transfected CHO cell lines expressing human APP695 (18) were cultured in F12 medium containing 10% fetal calf serum for 48 h before the culture medium was collected. DAPT, a functional ␥-secretase inhibitor (19), was kindly provided by Aventis Pharma.
Recombinant Adenoviruses and Neuronal Infection-The construction of recombinant adenoviruses encoding ␤-galactosidase (AdRSV␤gal) has been described previously (20,21). The pAdRSVAPP695 vector was generated by subcloning the EcoRV-SalI fragment of pHM-GAPP695 (22) in a pAdRSV vector (20). The deletion of the APP 695 intracellular domain was generated by PCR amplification of the APP 695 sequence using a 5Ј primer (5Ј-AACGAAGTTGAGCCTGTTGATG-3Ј) encoding residues 560 -567 of APP 695 and containing a BglII restriction site and a 3Ј primer (5Ј-GTCGACCTAGTACTGTTTCTTCT-TCAGC-3Ј) complementary to the sequence encoding residues 648 -653 followed by a stop codon and a SalI restriction site. pAdRSVAPP⌬C was generated by insertion of the BglII-SalI-digested PCR product in the BglII-SalI sites of pAdRSVAPP695. Production, propagation, and purification of adenoviruses (AdRSVAPP, AdRSVAPP⌬C, and AdRSV␤-gal) were performed as described previously (20). After 6 days in vitro, neuronal cultures were infected at the multiplicity of infection of 100 for 4 h in a minimal volume of culture medium. Infection medium was then replaced by fresh culture medium for 3-5 days. Under these conditions, at least 75% of neurons express the proteins encoded by recombinant adenoviruses (16).
Survival Assays and Nuclear Staining-Neuronal survival was measured by the colorimetric MTT assay as described previously (23). Neurons grown in 96-well culture dishes were incubated after infection for 2 h at 37°C in fresh culture medium containing 0.5 mg/ml MTT. Medium was removed, and dark blue crystals formed were dissolved by adding 100 l/well of lysis solution (isopropyl alcohol/0.04 N HCl). Outer diameter was measured on a microplate reader (492 nm). For nuclear staining, cells were fixed (0.37% formaldehyde/0.2% glutaraldehyde in phosphate-buffered saline) and incubated for 30 min in the Hoechst 33342 dye (1 g/ml). Nuclear morphology was analyzed under fluorescence microscopy at excitation/emission wavelengths of 350/450 nm.
Protein Analysis by Western Blot-Cell culture medium and cell lysates were analyzed by Western blot as described previously (16). Cell lysates (20 g of protein) and culture medium (15 l) were subjected to 10% SDS-PAGE and blotted onto nitrocellulose membrane, incubated overnight at 4°C with human APP-specific primary WO2 antibody at 1 g/ml (24), washed, and incubated with 1/10,000 anti-mouse Ig horseradish peroxidase-conjugated secondary antibody followed by ECL revelation.
Immunoprecipitation and Quantification of A␤ Production-A␤ production was monitored by immunoprecipitation of cell culture medium. The quantification of A␤1-40 and A␤1-42 isoforms was performed by ELISA. Culture medium was collected, treated with protease inhibitors (1 g/ml pepstatin, 10 g/ml leupeptin, 1 mM phenylmethylsulfonyl fluoride), and cleared by centrifugation (16,000 ϫ g, 5 min, 4°C). One hundred l of the supernatant was used for A␤ quantification by fluorescent sandwich ELISA according to the manufacturer's instructions (BIOSOURCE, Camarillo, CA). Previous experiments showed that there is no cross-reaction between A␤1-40 and A␤1-42 recognition. Fluorescence emission was measured at excitation/emission wavelengths of 485 nm/535 nm. Immunoprecipitation was performed on 1.5 ml of the remaining culture medium with 15 l/ml anti-A␤ whole rabbit serum (16). The immunoprecipitate was analyzed by Western blot on a 4 -12% Nupage™ gel using the WO2 antibody. A␤ was extracted from cell lysates by a modification of the protocol described previously (25). Neurons (ϳ10 7 cells) were scraped and pelleted in cold phosphatebuffered saline. Cell pellets were solubilized in 300 l of formic acid (70%). Formic acid-solubilized cell pellets were cleared (16,000 ϫ g, 5 min, 4°C) to remove cell debris, and supernatants were centrifuged at 21,000 ϫ g, 4°C for 20 min. The supernatants were vacuum-dried, and the resulting pellet was resuspended in 1 ml of alkaline carbonate buffer (2% Na 2 CO 3 , 0.1 N NaOH) and centrifuged (16,000 ϫ g, 3 min, 4°C). Protein concentration was measured on 50 l of the resulting supernatant by using the BCA protein assay (Pierce). For immunoprecipitation, 800 l of the supernatant was neutralized with 800 l of 1 M Tris-HCl, pH 6.8, and diluted 1:3 in H 2 O. Immunoprecipitation was performed as described above. For ELISA, 100 l of the remaining supernatant was neutralized with 100 l of 1 M Tris-HCl, pH 6.8, and diluted 1:3 in H 2 O containing 10% fetal calf serum, 0.5% Triton X-100, and 0.5% Nonidet P-40 (final concentrations). A␤1-40 and A␤1-42 concentrations were measured by ELISA on 100 l of neutralized extract.
Statistical Analysis-The number of samples (n) in each experimental condition is indicated in the figure legends. Statistical analysis was performed by one-way analysis of variance (ANOVA) followed by Bonferroni's multiple comparison post-test.

Long Term Expression of Human APP in Rat Cortical Neurons Induces
Apoptosis-When rat cortical neurons are infected by AdRSVAPP or AdRSV␤-gal, a maximal and stable production of APP and ␤-gal is observed at day 3 postinfection (16). APP expression and processing have been analyzed by using the human-specific WO2 antibody (24). Five days after infection by AdRSVAPP, high levels of soluble human APP (s␣APP) and A␤ are detected in the culture medium (Fig. 1A). This indicates that human APP is efficiently processed through both non-amyloidogenic and amyloidogenic pathways in rat neurons. The extracellular A␤ was quantified in the culture medium of neurons expressing human APP (Fig. 1B). These neurons secrete about 100 pg/ml of A␤ 1-40, but there is no detectable extracellular A␤1-42 (not shown). In the same ex-perimental conditions, the MTT survival assay shows that human APP expression induces neuronal death, whereas ␤-gal expression has no effect ( Fig. 1C). Taken together, these results demonstrate that a 5-day expression and processing of human APP in rat cortical neurons exert strong neurotoxic effects.
To further study the mechanism of the neuronal death triggered by human APP expression, neurons were stained with the Hoechst 3342 nuclear dye. This morphological analysis of the nuclei allows us to discriminate between the surviving and the apoptotic cells that display high nuclear condensation or fragmentation (26). Nuclear morphology of neurons was analyzed 5 days after infection by AdRSV␤-gal or AdRSVAPP (Fig.  1D), and the results were quantified to compare the proportion of apoptotic neurons in each condition (Fig. 1E). There is a 2-fold increase of apoptotic nuclei in neurons expressing APP as compared with the non-infected or ␤-gal-expressing neurons. This establishes that a 5-day expression of human APP induces apoptosis in rat-cultured neurons.
APP-induced Neuronal Apoptosis Does Not Involve Extracellular APP and A␤-We next analyzed whether the extracellular secretion of APP and A␤ could be responsible for the neurotoxic effects observed. To that end, neurons were incubated in the culture medium of a transfected CHO cell line or neuronal cultures expressing human APP695. The analysis of these two conditioned media indicates that they contain similar amounts of s␣APP but different amounts of A␤ ( Fig. 2A). The quantification of A␤ shows that the CHO-conditioned medium contains almost 100-fold more A␤1-40 than the neuronal-conditioned medium. In addition, A␤1-42 is present in the CHO culture medium, whereas it is undetectable in the neuronal culture medium (Fig. 2B). The treatment of neurons with these conditioned media does not significantly modify neuronal survival (Fig. 2C). Altogether, these results demonstrate that, in our model, the neuronal apoptosis induced by human APP is not triggered by any APP derivative, including A␤1-40 and A␤1-42, secreted in the culture medium. This raises the hypothesis that intracellular APP derivatives could be responsible for its neurotoxic effects.
APP-induced Neuronal Apoptosis Does Not Involve the Intracellular C-terminal Domain of APP-A possible origin of APPinduced neuronal death could be related to the intracellular C-terminal domain of the protein, which has been demonstrated to induce apoptosis in other cellular models (27,28). To test this hypothesis, neurons were infected with a recombinant adenovirus (AdRSVAPP⌬C) encoding a human APP isoform deleted in the intracellular C terminus of the protein. Five days after infection by AdRSVAPP⌬C, s␣APP and A␤ were detected in the neuronal culture medium (Fig. 3A). Neurons expressing human APP⌬C secrete about 50 pg/ml of A␤ 1-40 (Fig. 3B). This corresponds to half of the concentration of extracellular A␤ produced following the expression of full-length APP (Fig.  1B). In the same experimental conditions, the MTT survival assay shows that human APP⌬C expression induces neuronal death (Fig. 3C). Hoechst staining indicates that APP⌬C, like APP, triggers neuronal apoptosis (not shown). Taken together, these results demonstrate that the intracellular C-terminal domain of human APP is not involved in the neuronal death observed.
Intracellular A␤1-42 Accumulation Induces Neuronal Apoptosis-Since the neurotoxic effect of APP does not involve the intracellular domain of the protein, we investigated whether the accumulation of intraneuronal A␤ could trigger apoptosis. An important fraction of intracellular A␤ has been shown to be insoluble (29,30). Therefore, cells were solubilized in 70% formic acid as described previously (25) to recover all the intraneuronal A␤ peptide. The analysis of formic acid-solubilized cell pellets after 3 days of infection by AdRSVAPP or AdRSVAPP⌬C reveals a similar expression pattern of the in-traneuronal human proteins, although APP⌬C is detected in lower amounts as compared with APP (Fig. 4A). In the same experimental conditions, immunoprecipitation of cellular extracts show that intraneuronal A␤ is undetectable (Fig. 4A). The survival assay shows that, after 3 days of infection, neither APP nor APP⌬C expression causes neurotoxic effects (Fig. 4C), indicating that the overexpression of different levels of APP or APP⌬C per se does not induce any neurotoxicity. After 5 days of infection, neurons still express different amounts of APP or APP⌬C, but they accumulate similar amounts of intraneuronal A␤ 1-42. (Fig. 4B) In all these experiments, intraneuronal A␤1-40 was not detectable (not shown). After 5 days, both APP and APP⌬C induce a massive neuronal death, as compared with non-infected or ␤-gal-expressing neurons (Fig. 4D). Taken together, these results clearly establish that, in our model, APP-induced neuronal death takes place only when intraneuronal A␤1-42 is detected.
To further demonstrate that intraneuronal A␤1-42 accumulation leads to neuronal apoptosis, neurons expressing human APP were treated with DAPT, a functional ␥-secretase inhibitor (19). In our experimental conditions, DAPT does not display significant neurotoxicity by itself (not shown). Although DAPT treatment does not modify the secretion of s␣APP in the culture medium, it reduces the extracellular A␤1Ϫ40 concentration to a non-detectable level (Fig. 5A). DAPT also strongly reduces (57%) the production of intraneuronal A␤1-42 without affecting the levels of APP expression (Fig. 5B). This reduction of intraneuronal A␤1-42 production is concomitant with a significant recovery (52%) of cell survival (Fig. 5C). Altogether, these results demonstrate that the neuronal apoptosis induced by human APP in our model is triggered by the production and accumulation of intraneuronal A␤1-42.

DISCUSSION
It is currently well admitted that APP plays a central role in AD, but less is known about the link existing between APP processing and the massive neuronal death that takes place in the disease. In the present study, we report that long term expression of human APP triggers apoptosis in rat neurons. This neuronal apoptosis is not related to the adenoviral-mediated overexpression of an exogenous protein since the adenoviral-mediated expression of ␤-galactosidase is without effect on neuronal survival. Our results are in line with previous studies showing that the adenoviral expression of human APP695 induces apoptosis in both rat hippocampal neurons (31) and rat brain (32). It is thus very important to understand how human APP could induce neuronal death.
We first investigated the role of secreted A␤ in APP-induced neurotoxicity. We utilized the culture medium of a CHO cell line or neurons expressing different levels of human APP as a source of extracellular A␤. The concentrations of A␤1-40 and A␤1-42 in the CHO culture medium are comparable with those measured in the cerebrospinal fluid of AD patients (24). These concentrations of extracellular A␤ do not induce any neurotoxicity, indicating that extracellular soluble A␤ peptides are not directly responsible for neurodegeneration. It has been demonstrated previously that extracellular A␤ must aggregate into fibrils to acquire neurotoxic properties (11). At micromolar concentrations, fibrillar A␤ provokes oxidative injuries followed by cell death in neuronal and glial cells (33,34). This extracellular A␤ toxicity could be mediated by the interaction of fibrillar A␤ with APP present at the neuronal membrane (35). In the present study, even when neurons express human APP at their cell surface, the extracellular A␤ produced fails to induce neuronal death.
The fact that APP-induced apoptosis occurs independently of secreted APP derivatives led us to investigate the role of the C-terminal domain of the protein in neuronal death. The APP C terminus is essential for the cell surface APP signaling function (36) and for the APP-dependent axonal anterograde transport (37,38). Here we show that the neuronal expression of either C-terminal deleted APP (APP⌬C) or full-length APP (APP695) induces apoptosis. In other cellular models, the Cterminal domain of APP has been shown to mediate cytotoxic effects. The cleavage of the intracellular domain of APP by caspases generates a C31 cytotoxic fragment in mouse N2a neuroblastoma cell lines (27). The interaction of the intracellular domain of the V642I APP mutant with G proteins leads to nucleosomal DNA fragmentation in F11 neuronal cell lines (28,39). Since the neuronal apoptosis observed in this study is not mediated by the C-terminal domain of APP, we conclude that important differences in the metabolism and function of APP may exist between neuronal primary cultures and cell lines.
Another possible origin of APP-induced apoptosis is the intracellular accumulation of A␤. Neurons have been shown previously to produce intracellular A␤42 (25,40). Here we report that neurons expressing human APP or APP⌬C accumulate very similar amounts of intraneuronal A␤1-42, whereas they produce different amounts of extracellular A␤1-40. The extracellular A␤ production by neurons expressing APP or APP⌬C is in agreement with previous observations in transfected cells (18,41). After 3 days of infection, the expression of membrane human APP or APP⌬C at different levels does not induce any neurotoxicity, indicating that the overexpression of APP per se is not toxic. In addition, the amount of intraneuronal A␤1-42 is probably too low to be detected, and there is no neuronal damage observed. After 5 days of infection, the intraneuronal accumulation of A␤1-42 in neurons expressing APP or APP⌬C provokes a massive neuronal apoptosis. To further demonstrate that intraneuronal A␤1-42 accumulation induces neuronal apoptosis, the intraneuronal production of A␤1-42 was inhibited by a functional ␥-secretase inhibitor, DAPT (19). DAPT was chosen among other functional ␥-secretase inhibitors described (40) because it was the only one that was not neurotoxic by itself in our experimental conditions. Moreover, DAPT specifically inhibits A␤ production without affecting APP expression and processing through the non-amyloidogenic pathway. DAPT was able to reduce extracellular A␤1-40 pro-   , n ϭ 3). B, analysis of intracellular APP expression by Western blot (top) and quantification of A␤ accumulation (bottom) in formic acid-solubilized cell pellets (mean ϮS.E., n ϭ 6). C, neuronal survival measured by MTT assay 5 days after infection. Results are given as the percentage of neuronal survival as compared with non-infected (NI) control cultures (***, p Ͻ 0.001, **, p Ͻ 0.01; n ϭ 12). duction to a non-detectable level and inhibited intraneuronal A␤1-42 production by 57%. The differential inhibition observed at A␤40 and A␤42 sites could be viewed as evidence that different ␥-secretases generate A␤1-40 and A␤1-42 or could result from the production of these two peptides in different cellular organelles to which ␥-secretase inhibitors have access with different efficiency (42). The production of intraneuronal APP and A␤ was studied in three different experimental conditions: (i) production of APP without detectable A␤1-42, (ii) production of APP and A␤1-42, (iii) production of APP and partial inhibition of the production of A␤1-42. Our results show that the DAPT-mediated inhibition of intraneuronal A␤1-42 accumulation (57%) is similar to the DAPT-mediated recovery of neuronal survival (52%). This clearly indicates that the neuronal apoptosis that we observed results from the accumulation of intraneuronal A␤1-42. The mechanisms by which intraneuronal A␤ accumulation triggers apoptosis are currently unknown. The highly amyloidogenic A␤1-42 is produced in the endoplasmic reticulum/intermediate compartment of neuronal cells (43). The intracellular accumulation of A␤1-42 may cause an overload of the endoplasmic reticulum, leading to neuronal cell injury (44).
Intraneuronal A␤ accumulation has been described in transgenic mice expressing FAD mutations. In double transgenic mice expressing human PS1 and APP mutants, intraneuronal A␤ accumulation precedes amyloid deposition (14). Intraneuronal A␤42 accumulation together with extensive neuronal loss occurs without amyloid deposition in transgenic mice expressing a human PS1 mutation (15). Neuronal loss has also been documented in transgenic mice expressing the Swedish APP mutation that leads to plaque formation (45).
Intraneuronal accumulation of A␤42 in AD brains has been recently reported (12). Although AD patients show a severe neuronal loss in specific brain regions, the involvement of apoptosis in AD neurodegeneration remains a matter of debate (46). Apoptotic features have been observed in brains of AD patients (47), but it may be very difficult to observe the transient apoptotic state of neurons when looking at the lesions several years after the onset of the disease.
In conclusion, our data, in agreement with other recent reports, strongly support the idea that the intraneuronal production and accumulation of A␤1-42 are key events in AD. Elucidating how intraneuronal A␤ triggers apoptosis should in turn allow a better understanding of the neurodegeneration occurring in the disease.