Internalization of β-Amyloid Peptide by Primary Neurons in the Absence of Apolipoprotein E*

Extracellular accumulation of β-amyloid peptide (Aβ) has been linked to the development of Alzheimer disease. The importance of intraneuronal Aβ has been recognized more recently. Although considerable evidence indicates that extracellular Aβ contributes to the intracellular pool of Aβ, the mechanisms involved in Aβ uptake by neurons are poorly understood. We examined the molecular mechanisms involved in Aβ-(1–42) internalization by primary neurons in the absence of apolipoprotein E. We demonstrated that Aβ-(1–42) is more efficiently internalized by axons than by cell bodies of sympathetic neurons, suggesting that Aβ-(1–42) uptake might be mediated by proteins enriched in the axons. Although the acetylcholine receptor α7nAChR, previously suggested to be involved in Aβ internalization, is enriched in axons, our results indicate that it does not mediate Aβ-(1–42) internalization. Moreover, receptors of the low density lipoprotein receptor family are not essential for Aβ-(1–42) uptake in the absence of apolipoprotein E because receptor-associated protein had no effect on Aβ uptake. By expressing the inactive dynamin mutant dynK44A and the clathrin hub we found that Aβ-(1–42) internalization is independent of clathrin but dependent on dynamin, which suggests an endocytic pathway involving caveolae/lipid rafts. Confocal microscopy studies showing that Aβ did not co-localize with the early endosome marker EEA1 further support a clathrin-independent mechanism. The lack of co-localization of Aβ with caveolin in intracellular vesicles and the normal uptake of Aβ by neurons that do not express caveolin indicate that Aβ does not require caveolin either. Instead partial co-localization of Aβ-(1–42) with cholera toxin subunit B and sensitivity to reduction of cellular cholesterol and sphingolipid levels suggest a caveolae-independent, raft-mediated mechanism. Understanding the molecular events involved in neuronal Aβ internalization might identify potential therapeutic targets for Alzheimer disease.

In brains of individuals with Alzheimer disease (AD), 2 ␤-amyloid peptide (A␤) aggregates and accumulates as toxic fibrils in neuritic plaques and toxic soluble oligomers (1). A␤ is a 39 -43-amino acid peptide derived from the proteolytic cleavage of the amyloid precursor protein (APP) (2). The amyloid cascade hypothesis predicts that a gradual increase of A␤-(1-42) levels in brain interstitial fluid (3,4) may lead to A␤ oligomerization and eventually to A␤ fibrillization (5). Current evidence indicates that intraneuronal accumulation of A␤ is an early pathological biomarker for the onset of AD and may contribute to the cascade of neurodegenerative events (6). The observation that cortical neurons that accumulate A␤-  in brains of Down syndrome patients are apoptotic (7,8) provided additional indications of the importance of intracellular A␤ (7,8). Furthermore, microinjection of A␤-  or cDNA encoding A␤-  in cultured human neurons resulted in neurotoxicity (9). Besides, in the triple transgenic mouse model for AD there is a correlation between long term potentiation abnormalities and intraneuronal A␤ accumulation, all of which take place before the appearance of plaques or tangles (10). Intracellular accumulation of A␤-(1-42) results from one or a combination of the following processes: decreased A␤ degradation (disruption of the ubiquitin-proteasome system), increased intracellular generation of A␤, and increased uptake of A␤ from an external source (11). Both increased intracellular A␤ generation and uptake of extracellular A␤ are consistent with the endosomal/ lysosomal localization of intracellular A␤ (12,13).
Deposits of oligomeric A␤ are preferentially found on distal axons and synapses in mouse brains suggesting that these structures might represent the cellular entry points for A␤ (14). In support of the role of axons in A␤-induced neurotoxicity we have demonstrated previously that neurons exposed to A␤ exclusively in distal axons are more susceptible to apoptosis than neurons exposed only in cell bodies suggesting the possible involvement of axons in A␤ uptake (15). Nevertheless to our knowledge direct demonstration of A␤ uptake by axons is still missing.
In the brain, plasma, and cerebrospinal fluid as well as in senile plaques A␤ is associated to apolipoprotein E (apoE) (16 -18). Besides a pool of free A␤ likely exists. Internalization of the complex A␤-apoE has been extensively studied (19 -21); however, the mechanism that mediates uptake of A␤ that is not complexed to apoE is largely unknown. This pool likely increases considerably during AD development. Therefore, targeting free A␤ internalization might represent a way to regulate intracellular A␤ levels. In the present study, we examined the molecular mechanisms of A␤-  internalization by primary neurons in the absence of lipoproteins.
Culture of Neurons-Sympathetic neurons were isolated from superior cervical ganglia of newborn Harlan Sprague-Dawley rats and cultured in compartmentalized dishes as before (22). Dissociated neurons (0.3 ganglia/dish) were plated in the center compartment. Within 1-2 days, axons elongated along the tracks and entered the side compartments that contained medium supplemented with 50 ng/ml NGF. Medium supplied to the center compartment contained 2.5% rat serum, 1 mg/ml ascorbic acid, 10 -15 M cytosine arabinoside, and 10 ng/ml NGF. After 5-6 days cytosine arabinoside treatment was discontinued, and NGF was confined to the side compartments. For some experiments sympathetic neurons were cultured in 8-well chamber slides or 24-well dishes. Sympathetic neurons from caveolin-1 knock-out (Cav1 Ϫ/Ϫ ) and control ((B6 129 SF2/J) (Cav1 ϩ/ϩ )) mice (The Jackson Laboratory, Bar Harbor, ME) were isolated and cultured as above. Experiments started at day 7-8 in culture. Rat basal forebrain cholinergic neurons were generously provided by Dr. Jhamandas (University of Alberta) and were prepared as before (23). Experiments started at day 3 in culture. Before the treatments neurons were rinsed several times with medium without serum, and the experimental treatments were performed in the absence of serum.
A␤-  Preparations-Non-fluorescent and fluoresceinconjugated A␤-(1-42) peptide were prepared according to published protocols (24). Briefly A␤-(1-42) peptide was initially dissolved to 1 mM in hexafluoroisopropanol and separated into aliquots in sterile microcentrifuge tubes. Hexafluoroisopropanol was dried under a stream of N 2 , and the peptide film was desiccated at Ϫ20°C. The peptide was resuspended in Me 2 SO at a concentration of 5 mM. L15-CO 2 medium (phenol red-free, antibiotic-free, and serum-free) was added to bring the peptide to a final concentration of 100 M and incubated at 4°C for 24 h. To prepare biotinylated A␤ (bio-A␤-(1-42)), a solution of 100 M A␤-(1-42) was treated with a 1.5 mg/ml concentration of the membrane-impermeable derivative of biotin (sulfo-NHS-S-S-biotin) in Leibovitz L15-CO 2 medium (phenol red-free, antibiotics free, serum free) according to the manufacturer's specifications. The reaction was stopped by adding 100 l of 250 mM glycine. The biotinylated peptide was diluted in Leibovitz L15-CO 2 medium to the appropriate concentration. All A␤ preparations used were similar to those characterized previously (24) in that they contain monomers and oligomers of A␤ but do not contain fibrils (see supplemental Fig. 1, A and B). Fluo-A␤-(1-42) and bio-A␤-(1-42) induced neuronal apoptosis like the non-modified peptide (supplemental Fig. 1).
Immunoblotting-For the detection of A␤, proteins were separated by SDS-PAGE on 12% polyacrylamide gels containing 0.1% SDS using two buffers system (25). For all other experiments the concentration of polyacrylamide gels was 10% containing 0.1% SDS. Transfer of proteins to polyvinylidene difluoride membranes was performed overnight at 4°C in 25 mM Tris, 192 mM glycine, 16% methanol buffer, pH 8.3. Membranes were blocked for 1 h in Tris-buffered saline, 0.1% Tween 20 (TTBS) containing 5% nonfat milk (blocking buffer) and incubated overnight in the primary antibody solution prepared in TTBS containing 5% nonfat milk. Primary antibodies for A␤ (4G8) (1:1000) and anti-␣7nAChR (1:1000) were used. Membranes were washed two times with Tris-buffered saline, two times with TTBS, and two times with Tris-buffered saline and then incubated for 1 h with the secondary antibody (1:2000) in blocking buffer at room temperature with gentle agitation. Immunoreactivity was detected by ECL (Amersham Biosciences).
Receptor Binding Assays-Radioligand binding assay in cultured neuronal membranes was performed as described earlier (26). In brief, cultured neurons were scraped into ice-cold Tris-HCl (50 mM, pH 7.4) and then homogenized by passing the cell lysates several times through a Pasteur pipette. Membranes were centrifuged three times and then incubated for 2 h at 22°C in Tris-HCl buffer (50 mM, pH 7.4) containing 1 mM MgCl 2 , 120 mM NaCl, 5 mM KCl, 2 mM CaCl 2 , 1% bovine serum albumin, and 1 nM 125 I-␣-BTx (2000 Ci/mmol, Amersham Biosciences) as mentioned before (27). Nonspecific binding was determined in the presence of 1 M unlabeled ␣-BTx. Displacement of 125 I-␣-BTx binding sites was also simultaneously studied in the presence or absence of 100 M nicotine and 5 M A␤- . Incubations were terminated by rapid filtration using a cell harvester filtering apparatus, and then radioactivity was measured using a ␥ counter. Binding experiments were performed four times, each in triplicate, and data are represented as percentage of specific binding.
Adenoviral Infection-Recombinant adenoviruses expressing the tetracycline-regulated chimeric transcription activator (tTA), hemagglutinin-tagged wild type dynamin, hemagglutinin-tagged dynK44A mutant, and T7-tagged clathrin hub were prepared by Dr. Altschuler (28) and obtained from Dr. Nabi (University of British Columbia). To enhance the infection, viral stocks were preincubated with polylysine (29). Infection of sympathetic neurons with serial dilutions of adenovirus constructs was used to determine the minimum amount of virus necessary to infect the majority of the neuronal population (Ͼ80%). Titration was performed using QuickTiter Adenovirus Titer Immunoassay kit (Cell Biolabs, Inc.-Cedarlane Laboratories Ltd., Ontario, Canada). Neurons cultured in compartmentalized dishes were infected with a multiplicity of infection of 100 in 50 l of L15-CO 2 medium containing 50 ng/ml NGF, 2.5% rat serum, and 1 mg/ml ascorbic acid added to the cell body-containing compartment for 24 h. The next day the medium was replaced with fresh medium. The experiments were initiated after 48 -72 h of viral protein expression. To examine the expression of the recombinant proteins, cellular material from control and infected cultures was collected directly in modified Laemmli sample buffer and analyzed by immunoblot.
Assessment of A␤ Internalization by Immunofluorescence and Confocal Microscopy-Fluo-A␤-  was provided to sympathetic neurons cultured in three-compartment cultures or 8-well slides. The internalization of fluo-A␤-(1-42) in living neurons was examined using a Nikon TE300 inverted fluorescence microscope equipped with a Nikon digital camera, DXM-1200 (Nikon, Toronto, Ontario, Canada). Images were taken and analyzed using Northern Eclipse Version 7.0 image capture and analysis software (Empix Imaging, Mississauga, Ontario, Canada). For confocal microscopy studies, after the indicated treatments in each experiment, culture medium was removed, and cells were fixed in 4% paraformaldehyde for 20 min, permeabilized with 0.1% Triton X-100 in phosphate-buffered saline for 15 min, and then blocked using 0.1% gelatin in phosphate-buffered saline (blocking buffer) for 1 h. For detection of caveolin, primary rabbit anti-caveolin (1:250) and secondary Alexa594-labeled goat anti-rabbit antibody prepared in blocking buffer (1:1000) were used. Pictures were taken using a Zeiss LSM 510 confocal laser-scanning microscope equipped with an S-Fluor 40ϫ/1.3 oil objective using appropriate filter sets and excitation wavelengths. Co-localization was quantified using Metamorph software. For both channels, the best fit lower threshold value to remove most background signal was determined using the threshold tool.
Assessment of A␤ Internalization Using Biotinylated A␤-Biotinylated A␤ was added to neurons, and internalization was allowed to occur by incubation at 37°C. At the end of the experiment, the biotin moiety from A␤ that remained outside the neurons was cleaved by reducing its disulfide linkage with glu-tathione buffer (two times for 15 min at 4°C). Excess glutathione was inactivated using 5 mg/ml iodoacetamide in phosphate-buffered saline (two times for 5 min at 4°C) (30). Finally neurons were lysed using RIPA buffer (50 mM Tris-HCl, pH 8.0, 150 mM NaCl, 1% (v/v) Nonidet P-40, 2 mM EDTA, pH 8.0, and 2 mM MgCl 2 ) containing protease inhibitor mixture tablets (Complete Mini, Roche Diagnostics). The internalized bio-A␤-(1-42) protected from the cleavage buffer was pulled down with ImmunoPure immobilized streptavidin beads. The samples were eluted from the beads with boiling Laemmli sample buffer (40 mM Tris-HCl, pH 6.8, 1% SDS, 4% 2-mercaptoethanol, 10% glycerol, and 0.002% bromphenol blue), resolved by SDS-PAGE (25), and immunoblotted with 4G8 antibody against A␤ as above.
Assessment of A␤ Internalization by Enzyme-linked Immunosorbent Assay-A␤-  was provided to sympathetic neurons cultured in three-compartment cultures. Cellular material from the cell body-containing compartment of three dishes of the same treatment was harvested in 0.05 N NaOH. The mass of A␤-(1-42) in cell bodies was quantified using sandwich enzyme-linked immunosorbent assay according to the manufacturer's specifications. Incubation with chromogen substrate produced a colorimetric response (450 nm) measured using a SpectroMax (Molecular Devices, Sunnyvale, CA) microtiter plate reader. Protein content was determined by BCA assay (Pierce). The data are presented as pg of A␤/g of protein.
Detection of Nuclear Apoptosis-Apoptotic cell death was identified by nuclear staining with Hoechst 33258 as described previously (15). Nuclei were visualized using a Nikon TE300 inverted fluorescence microscope equipped with a Nikon digital camera, DXM-1200. Images were analyzed using Northern Eclipse Version 7.0 image capture and analysis software (Empix Imaging). Five hundred to 1000 neurons per treatment were counted by an observer "blinded" to the neuronal treatment.

A␤-(1-42) Is Preferentially Internalized by Axons, Is
Retrogradely Transported, and Accumulates in Cell Bodies-We examined the ability of neurons to internalize A␤ in the absence of ligands (apoE and ␣ 2 -microglobulin) that bind to known surface receptors. Basal forebrain and sympathetic neurons, two neuronal types that respond similarly to A␤-(1-42) (15), were able to internalize fluo-A␤-(1-42) (Fig. 1A). Due to its hydrophobic nature, a significant portion of A␤ remained attached to the cell surface, and therefore it was difficult to differentiate plasma membrane-bound A␤ from internalized A␤. Therefore, we used a biotin internalization assay to confirm A␤ uptake. In this assay the biotin moiety of A␤ that remains bound to the extracellular aspect of the plasma membrane is cleaved. Hence when streptavidin is used to separate bio-A␤-(1-42) only internalized A␤, which is protected from cleavage, is recovered. In this way, we confirmed that A␤-(1-42) is internalized by neurons in a time-dependent manner (Fig. 1B). To make sure that apoE was absent in our studies, in addition to performing the experiments in the absence of serum we confirmed that neurons were not making apoE even when treated with A␤ (supplemental Fig. 2).
Previous studies in mouse brain using specific immunodetection indicated that A␤ localizes to cell processes, especially to axon terminals, suggesting the involvement of axons in A␤ uptake (31,32). To investigate A␤ internalization by cell bodies and axons separately, we chose a compartmentalized culture system (33), which allows the selective treatment of distal axons or cell bodies with A␤ (15). When fluo-A␤-(1-42) was added to the distal axon-containing compartment (DAX) only, we observed significant A␤ internalization and retrograde transport to the cell body/proximal axon-containing compartment (CB) (Fig. 1C, a-d). Moreover, we detected a dramatic increase of the mass of A␤ in the cell bodies, confirming A␤ uptake and transport (Fig. 1D). In contrast, when fluo-A␤-(1-42) was added exclusively to CB, there was significantly less A␤ internalization, and A␤ anterograde transport was almost negligible (Fig. 1C, e-h). Unfortunately we could not quantify the mass of A␤ when it was provided to cell bodies because the small amount of A␤ associated to the surface of cell bodies was sufficient to interfere with the determination. To confirm the preferential A␤ uptake by axons, we treated sympathetic neurons in DAX or CB with bio-A␤-  and examined A␤ internalization after cleavage of extracellular A␤ as indicated above. We were able to detect bio-A␤-(1-42) only in neurons given A␤ in the distal axons (Fig. 1E). The fact that A␤ internalization occurs preferentially through axons suggests that the mechanism that mediates A␤ uptake is regulated and likely involves proteins that are enriched in the axons.
␣-Bungarotoxin and Receptor-associated Protein Do Not Affect A␤ Internalization-A receptor previously implicated in A␤ internalization is the ␣7nAChR (34). Hence we investigated whether ␣7nAChR was enriched in distal axons of neurons cultured in three-compartment dishes. In parallel, we examined the expression of ␣7nAChR in basal forebrain neurons and sympathetic neurons cultured in regular dishes. As expected, ␣7nAChR was highly expressed in basal forebrain neurons and in sympathetic neurons cultured under cholinergic (with leukemia inhibitory factor) as well as adrenergic (without leukemia inhibitory factor) conditions ( Fig.  2A). Importantly we found a dramatic enrichment of this receptor in distal axons compared with cell bodies/proximal axons, making ␣7nAChR a good candidate to mediate A␤ uptake. However, competition binding experiments showed that 5 M A␤-(1-42) was unable to significantly compete with 125 I-␣-BTx receptor binding sites in membranes prepared from sympathetic neurons (Fig. 2B). In contrast, 1 M unlabeled ␣-BTx (i.e. nonspecific binding) and 10 M nicotine potently displaced 125 I-␣-BTx binding, thus indicating that the sites recognized by this ligand represent a subset of nicotinic receptors. These data, in agreement with a recent report (35), suggest that A␤ does not compete with the ␣7nAhR binding site. Accordingly in neurons given ␣-BTx together with fluo-A␤-(1-42) there was no decrease in A␤ internalization (Fig. 2C). All together our data suggest that ␣7nAChR does not mediate A␤ uptake at least in neurons under our experimental conditions. A second group of receptors that distribute differently to cell bodies and axons (36) and that have been linked to A␤ internalization is the LDL receptor (LDLr) family. Internalization through this receptor is blocked by the chaperone protein receptor-associated protein (RAP) (37)(38)(39). RAP, at a concentration that has been proven to inhibit ligand interaction of all LDLr family members (40), was unable to inhibit fluo-A␤-(1-42) uptake and transport (Fig. 2D). To prove that RAP effectively inhibits LDLr in sympathetic neurons under our experimental conditions, we examined uptake and transport of LDL particles labeled with DiI, which are natural ligands of these receptors. As expected RAP caused a dramatic decrease in the internalization of LDL (Fig. 2D). The lack of effect of RAP on A␤ uptake and transport suggests that LDLr family members are not required for A␤ internalization in the absence of apoE.
A␤ Uptake Is Dynamin-dependent but Clathrin-independent-Internalization of A␤ in a complex with LDL receptor ligands is mediated by coated pits. We aimed to identify whether clathrin was required for A␤ uptake in the absence of apoE. Clathrin-and caveolae/lipid raft-mediated endocytosis remain the two best characterized mechanisms of ligand-receptor internalization (41). Both mechanisms require the GTPase dynamin, but caveolae/lipid raftmediated endocytosis is independent of clathrin. Using adenoviral infection we expressed in neurons dominant-negative constructs of dynamin or overexpressed the clathrin hub peptide that inhibits clathrin-mediated endocytosis. For control we expressed wild type dynamin. Under these conditions we investigated the internalization route involved in A␤ uptake.
Overexpression of wild type dynamin caused non-significant changes in A␤ internalization (Fig.  3, f and h versus b and d), but expression of the mutant dynamin K44A significantly reduced A␤ internalization and transport (Fig. 3, j versus f). In addition, overexpression of clathrin hub did not significantly affect A␤ internalization (Fig. 3, n  versus b). These results suggest that A␤ is internalized by a mechanism that requires dynamin but is independent of clathrin. Importantly overexpression of clathrin hub in our experiments did block clathrindependent endocytosis as demonstrated by the dramatic decrease of LDL internalization when clathrin hub was overexpressed (Fig. 3, p versus d). Together these results indicate that dynamin is a central player in A␤ uptake and that clathrin is not involved in the process.
Caveolin Is Not Required for A␤ Endocytosis-Clathrin-mediated endocytosis targets proteins to the early endosomes (characterized by the presence of the protein EEA1) where proteins are sorted for either recycling or trafficking to late endosomes and lysosomes (42,43). In contrast, caveolin-positive vesicles arise from caveolae and represent a non-clathrin internalization pathway (44). Co-localization with EEA1 and caveolin-1 has been used as indication of clathrin-and caveolae-mediated endocytosis, respectively (45).
Sympathetic neurons were given fluo-A␤-(1-42) for 30 min, and the internalization compartments were analyzed by threecolor immunofluorescence to simultaneously detect EEA1 (a marker of early endosomes), caveolin (a marker of the caveolae pathway), and A␤-(1-42) (Fig. 4). Our data indicate that the majority of internalized A␤-(1-42) did not co-localize with EEA1 or caveolin. As expected significant co-localization of EEA1 and caveolin was not observed either. These results corroborated that A␤ is internalized independently of clathrin and imply that A␤-  internalization is independent of caveolin as well.
To further investigate the requirement of caveolin, we examined A␤-(1-42) internalization in neurons isolated from caveolin-1 knock-out (Cav1 Ϫ/Ϫ ) and from control ((B6 129 SF2/J) (Cav1 ϩ/ϩ )) mice. Neurons that did not express caveolin-1 were able to internalize A␤ as effectively as neurons isolated from wild type mice (Fig. 5). This finding strongly indicates that caveolin is not necessary for A␤ internalization.
A␤ Internalization Depends on the Levels of Cellular Sphingolipids and Cholesterol-The results presented above and previous evidence from brains of AD patients indicating that A␤ is present in a complex with the lipid raft component GM1 (46) suggest that A␤-(1-42) is internalized by caveolin-independent, lipid raft-mediated endocytosis. If so, A␤-(1-42) should be sensitive to changes in the levels of cholesterol and/or sphingolipids (SPLs). To test this possibility we attempted to decrease membrane cholesterol. The most widely used way to achieve membrane cholesterol depletion is to perform short treatments (up to 30 min) with methyl-␤-cyclodextrin. However, our experiments involved long incubations (up to 18 h) with A␤ during which time membrane cholesterol levels will likely return to normal. We therefore decided to inhibit cholesterol synthesis using pravastatin, a well known inhibitor of 3-hydroxy-3-methylglutaryl-CoA reductase, which has already been tested in sympathetic neurons (47). The reduction of cellular SPLs was achieved by using the ceramide synthase inhibitor Fumonisin B1 (22). The inhibitors were given to neurons cultured in three-compartment dishes according to the scheme in Fig. 6A before adding fluo-A␤-  in the axon-containing compartment. Parallel cultures were subjected to the same treatments with inhibitors but were given CTxB, a marker of lipids rafts endocytosis. Fumonisin B1 was added to all three compartments to ensure complete inhibition of SLP synthesis in both cell bodies and axons. On the other hand, cholesterol is synthesized exclusively in cell bodies, and therefore pravastatin was given only to the center compartment as before. Both inhibitors are effective in sympathetic neurons and do not have deleterious effects on neuronal survival under the conditions used here (47,48).  and caveolin were detected using the primary antibodies indicated under "Experimental Procedures" followed by an Alexa647-conjugated anti-rabbit antibody and an Alexa594-conjugated anti-mouse antibody, respectively. Cells were observed by confocal laser-scanning microscopy. A␤, EEA1, and caveolin staining are colored green, red, and purple, respectively. Co-localization should appear yellow for A␤/EEA1and white for A␤/caveolin. Significant co-localization of A␤ with EEA1 or caveolin was not observed. DECEMBER 7, 2007 • VOLUME 282 • NUMBER 49

A␤ Internalization in Cultured Primary Neurons
Inhibition of SPL synthesis or cholesterol alone did not decrease A␤ internalization (Fig. 6, B (f and j versus b) and C). However, Fumonisin B1 and pravastatin together caused significant reduction (ϳ51%) in A␤ internalization (Fig. 6, B (n versus  b) and C). All together our data indicate the requirement of cholesterol and SPLs for A␤-(1-42) uptake. CTxB internalization was more sensitive to changes in cholesterol and SPLs alone than A␤-(1-42) uptake (Fig. 6, B (g, h, k, l, o, and p) and C).
Although pravastatin treatment caused a significant decrease in total cholesterol mass, which was similar to that obtained with methyl-␤-cyclodextrin (Fig. 6D), when we examined cholesterol by filipin staining (Fig. 6E) we found that pravastatin caused a preferential decrease of intracellular rather than plasma membrane cholesterol. On the other hand methyl-␤cyclodextrin-treated neurons showed a dramatic decrease of cholesterol at the plasma membrane (Fig. 6E). This difference in the nature of the pool of cholesterol affected could explain the lack of effect of pravastatin alone in A␤ uptake. We next examined the association of A␤ with CTxB, a marker of the lipid raft endocytic pathway (Fig. 7). We found that fluo-A␤-(1-42) only partially co-localized with CTxB. In addition, partial co-localization between CTxB and caveolin was observed. As before no co-localization between A␤-(1-42) and caveolin was present.

DISCUSSION
The molecular events involved in neuronal A␤ internalization in AD are mostly unknown, and any progress in this respect has the opportunity to identify potential therapeutic targets. The experiments reported herein demonstrated that, in the absence of apoE, A␤-(1-42) is taken up and internalized into primary neurons by a regulated mechanism that is enriched in the axons and requires cholesterol and sphingolipid synthesis. Previous studies in cortical neurons have detected A␤ internalization in the absence of serum (49); however, in these studies a different preparation of A␤ was used, and the mechanism of A␤ uptake was not investigated.
Extracellular accumulation of A␤ represents the foundation of the amyloid cascade hypothesis (50). The importance of intraneuronal A␤ accumulation in the pathogenesis of AD has emerged more recently (51)(52)(53). Intraneuronal A␤ is present in brains of AD (6,54,55) and Down syndrome patients (7,56,57). Studies in animal models of AD also support the role of intraneuronal A␤. In studies using transgenic mice harboring constructs that target A␤ either extracellularly or intracellularly, only the intracellular A␤-producing transgenic mice developed neurodegeneration (58). In APP and presenilin-1 double transgenic mice, intraneuronal A␤ accumulation precedes plaque formation (59), and in the triple transgenic model long term potentiation abnormalities and cognitive dysfunctions correlate with the appearance of intraneuronal A␤ prior to the occurrence of plaques or tangles (10,60).
Some evidence suggests that the intracellular pool of A␤ is derived from slow production from APP inside the cells (61)(62)(63)(64). On the other hand, work by D'Andrea et al. (6) favors a mechanism that involves A␤ endocytosis. Moreover, in organotypic hippocampal slice cultures, A␤-(1-42) gradually accumulates and is retained intact by field CA1 but not by other subdivisions, suggesting the existence of selective A␤ uptake by neurons at risk in AD (65). In addition, extracellular A␤-  can increase levels of intraneuronal A␤-(1-42) (66), and A␤ reuptake is important for A␤ accumulation in situ (67).
Current development of AD therapy has focused on the inhibition of A␤ production with one strategy being to inhibit ␥-secretase (68), the multienzyme complex responsible for APP intramembrane cleavage. Yet therapies that suppress this enzymatic activity might eliminate critical functions with dramatic deleterious consequences because other important proteins (Notch, CD-44, Erb4, and E-cadherin) are also substrates of ␥-secretase (52,69). Therefore, neuronal A␤ internalization represents an attractive process to be targeted for drug intervention in AD independently or in conjunction with more selective ␥-secretase inhibitors.
Previous studies have demonstrated internalization of synthetic A␤ peptides in neurons of hippocampal slices (65,70) and in neuronal models in culture (65,(71)(72)(73). Although the evidence supports the notion that endocytosis is involved in A␤ internalization, the molecular details of A␤ uptake were not addressed in those studies, and the precise A␤ points of entry were not investigated.
With respect to this last aspect, findings that oligomeric A␤ deposits are present preferentially on distal axons and synapses in Tg2576 mouse brain and in central nervous system-derived neuronal cells suggested that these structures are important sites of A␤ entry (14,32). The use of the three-compartment culture system in our studies allowed us to directly examine A␤ uptake by axons because neurons are plated in the center compartment of the dish, which is over 1 mm wide; therefore neurites that grow in the side compartments represent axons exclusively (74). We demonstrated that A␤-(1-42) is preferentially internalized through distal axons and retrogradely transported to cell bodies. To our knowledge, this is the first direct demonstration of more efficient uptake of A␤ by axons than by cell bodies. This result might have important implications in particular for basal forebrain neurons that differ from other neurons affected in AD in that their somas are located in nuclei that are physically distant from the areas of preferential A␤ accumulation in the cortex and hippocampus. In addition, we found that there is robust retrograde transport of A␤ to cell bodies, but the anterograde transport from cell bodies to axons was negligible. The limited A␤ internalization found when A␤ was added to the center compartment cannot be explained by the presence of less cellular membrane in this compartment because uptake of DiIlipoproteins by cell bodies was observed in parallel experiments (not shown). Our system, however, does not permit the study of uptake by dendrites exclusively because the center compartment also contains cell bodies and proximal axons.
The difference in A␤ internalization by cell bodies and axons indicated that A␤ uptake in the absence of apoE is a regulated mechanism possibly mediated by proteins enriched in axons. Among a few receptors previously involved in A␤ uptake (23,34,75,76) the ␣7nAChR seemed one of the most relevant. The most vulnerable neurons in AD appear to be those that abundantly express the ␣7nAChR such as neurons of hippocampus and cholinergic projection neurons from the basal forebrain (77,78). Besides there is evidence indicating that A␤-(1-42) interacts with ␣7nAChR with high affinity (79,80), and some recent studies have suggested the involvement of ␣7nAchR in the internalization of FIGURE 6. A␤ internalization depends on cellular sphingolipids and cholesterol levels. Rat sympathetic neurons cultured in three-compartment cultures were left untreated (control) or pretreated with 25 M Fumonisin B1 (FB1) for 2 days or/and 50 M pravastatin (Prav) for 1 day followed by treatment with fluo-A␤-  or CTxB according to the scheme shown in A. B, after 18 h, the fluorescence of the cell body-containing compartment was analyzed. Bar, 20 m. C, fluorescence intensity in cell bodies of 200 neurons treated with A␤ was quantified using Northern Eclipse software. Statistical significance of differences between untreated and treated neurons is indicated by * and was evaluated by analysis of variance (p Յ 0.001). The experiment was repeated twice with comparable results. D, sympathetic neurons cultured in 24-well plates were treated with 50 M pravastatin for 1 day or 10 mM methyl-␤-cyclodextrin (M␤CD) for 30 min at 37°C, and cholesterol mass was measured with the Amplex Red cholesterol determination kit. Data are expressed as means Ϯ S.D. of three independent experiments. E, neurons were treated with pravastatin or methyl-␤-cyclodextrin as in D, and cholesterol was labeled with 0.1 g/ml filipin.
A␤-(1-42) by neuronal cells lines (34). However, we found that A␤-  was unable to compete with ␣-BTx nicotinic receptor binding sites in neuronal membranes, and ␣-BTx did not affect A␤-  internalization. Our results are in agreement with recent evidence indicating that A␤-(1-42) does not directly interact with ␣7NAChR (35). The difference between our findings and those of Nagele et al. (34) might be due to the use of A␤ monomers and not oligomers and the presence of serum in their studies.
Because the complex A␤-apoE is internalized by LDLr-related protein and related receptors we investigated the role of these receptors in A␤ internalization in the absence of lipoproteins. Members of the LDLr family are differentially distributed in cell bodies and distal axons of sympathetic neurons (36), and ligands of the LDLr family members (apoE and activated ␣ 2 -microglobulin) associate with A␤ resulting in the internalization of the complex A␤-ligand (75,(81)(82)(83). To our knowledge the role of the LDLr family in the uptake of A␤ in the absence of LDLr ligands has not been investigated. Our studies showed that RAP, a chaperone that universally blocks ligand binding to LDLr, had no effect on A␤ uptake, whereas it completely blocked LDL internalization under the same experimental conditions. This result indicates that when A␤ is not forming a complex with apoE or other ligands its internalization is not mediated by members of the LDLr family. A␤ uptake is not even mediated by clathrin-dependent mechanisms as indicated by A␤ uptake in neurons in which clathrin-mediated endocytosis had been abrogated by overexpression of the clathrin hub. Moreover, internalized A␤ did not co-localize with EEA1, a marker for clathrin-mediated endocytosis. Taken together these results suggest that although LDLr and ␣7nAChR have been implicated in neuronal internalization other mechanism(s) operate in axons for A␤ uptake. 1% bovine serum albumin for 30 min at 37°C. After fixation caveolin was detected using the primary antibody indicated under "Experimental Procedures" followed by an Alexa647-conjugated anti-rabbit antibody. Cells were observed by confocal laser-scanning microscopy. A␤, CTxB, and caveolin staining are colored green, red, and purple, respectively. Co-localization should appear yellow for A␤/CTxB and white for A␤/caveolin and for CTxB/caveolin. Limited co-localization of A␤ with CTxB was observed. However, our data do indicate that A␤ uptake entails endocytosis because neurons expressing the inactive mutant of dynamin exhibited reduced A␤ internalization. Caveolae/lipid raft-mediated endocytosis is the most important dynamin-dependent, clathrin-independent endocytic process (41). By confocal studies we demonstrated, however, that A␤ does not co-localize with caveolin but partially co-localizes with CTxB, suggesting that A␤-(1-42) internalization involves non-caveolae, GM1-containing rafts. Normal A␤ uptake by neurons isolated from caveolin-1 null mice further indicates that caveolin is not required for A␤ internalization. Although caveolins are expressed in neurons, the existing evidence suggests that neurons lack caveolae, and therefore caveolins might participate in neuronal processes different from endocytosis (84). Nevertheless lipid raft caveolae-independent endocytosis also occurs in cells with caveolae (41).
Plasma membrane cholesterol and sphingolipid levels regulate raft-mediated endocytosis. Cholesterol has been linked to several processes that take place in AD including A␤ aggregation (85)(86)(87)(88) and modulation of APP cleavage and A␤ production (89 -91). On the other hand, very little information is available on the role of cholesterol in A␤ internalization. Here we provide evidence that A␤ internalization is dependent on the cellular levels of sphingolipids and cholesterol. We demonstrated that reduction of cellular cholesterol or sphingolipids levels alone is not sufficient to block A␤ internalization, but the simultaneous decrease in cholesterol and sphingolipids effectively reduced A␤ uptake and trafficking. Our experiments also indicated that independent reduction of cholesterol or sphingolipids results in decreased CTxB uptake demonstrating that the drug treatments were effective. Previous work indicated that cholesterol content regulates A␤ binding to the cell surface. Arispe and Doh (92) and Yip et al. (93) reported that reducing membrane cholesterol in PC12 cells results in increased incorporation of the peptide into the membrane. Conversely another group found that cholesterol depletion abolishes the binding of A␤-(1-40) to PC12 cells (86). None of these studies investigated the role of cholesterol in A␤ internalization. However, our data for the requirement of functional dynamin for A␤ internalization and trafficking suggest that penetration of A␤ into the membrane bilayer is not required for A␤ uptake.
We propose a model in which free A␤ and A␤ bound to apoE and other ligands enter the neurons by different mechanisms (Fig. 8). The proportion of each pool of A␤ will be determined by the levels of A␤ in the extracellular space and by the efficiency of A␤ binding to apoE. In turn A␤ binding to apoE is regulated by the apoE isoform present. The model also predicts that at very high levels of A␤ the pool of free A␤ will become more significant than the pool of A␤-apoE. The complex interplay of A␤ pools might explain the conflicting findings on the apoE isoform effects.