Amyloid-β(1–42) Aggregation Initiates Its Cellular Uptake and Cytotoxicity *

The accumulation of amyloid β peptide(1–42) (Aβ(1–42)) in extracellular plaques is one of the pathological hallmarks of Alzheimer disease (AD). Several studies have suggested that cellular reuptake of Aβ(1–42) may be a crucial step in its cytotoxicity, but the uptake mechanism is not yet understood. Aβ may be present in an aggregated form prior to cellular uptake. Alternatively, monomeric peptide may enter the endocytic pathway and conditions in the endocytic compartments may induce the aggregation process. Our study aims to answer the question whether aggregate formation is a prerequisite or a consequence of Aβ endocytosis. We visualized aggregate formation of fluorescently labeled Aβ(1–42) and tracked its internalization by human neuroblastoma cells and neurons. β-Sheet-rich Aβ(1–42) aggregates entered the cells at low nanomolar concentration of Aβ(1–42). In contrast, monomer uptake faced a concentration threshold and occurred only at concentrations and time scales that allowed Aβ(1–42) aggregates to form. By uncoupling membrane binding from internalization, we found that Aβ(1–42) monomers bound rapidly to the plasma membrane and formed aggregates there. These structures were subsequently taken up and accumulated in endocytic vesicles. This process correlated with metabolic inhibition. Our data therefore imply that the formation of β-sheet-rich aggregates is a prerequisite for Aβ(1–42) uptake and cytotoxicity.

One of the pathological hallmarks of Alzheimer disease (AD) 2 is the presence of extracellular plaques composed mainly of 42-amino acid amyloid ␤ peptide (A␤ ) (1). The small hydrophobic A␤  peptide, which is generated by proteolytic cleavage of the amyloid precursor protein, is released as a monomer from the plasma membrane into extracellular space, and tends to aggregate spontaneously into oligomeric, protofibrillar, and fibrillar assemblies (2)(3)(4). Oligomeric species of A␤  are tightly linked to AD pathogenesis and are presumed to be the cause of neuronal damage (5). Several studies have suggested that the reuptake of extracellular A␤  into neurons may lead to the formation of intracellular aggregates, resulting in neuronal damage and neurotoxicity (6 -8). Endocytosis of misfolded proteins has also been observed in cell models of the tau protein, ␣-synuclein and huntingtin (9,10), and recent evidence suggests that it may be the initial step in the replication of the misfolded protein structures by prion mechanisms (10 -14). Several possible endocytic pathways, such as macropinocytosis and receptor-mediated endocytosis, have been discussed for A␤ and other misfolded protein aggregates (15)(16)(17)(18)(19). However, our understanding of the connection between aggregation and cytotoxicity is still limited. It has not been conclusively determined how and when the A␤(1-42) peptide becomes toxic, whether A␤ aggregates prior to internalization or during the internalization process and, if so, in which intracellular compartments the aggregates form. Elucidating the connection between aggregation and internalization of A␤  peptide may be thus vital in understanding its toxicity.
Here, we examine the relationship between the aggregation state of extracellular A␤  and the efficiency of its internalization. We aimed to determine whether the formation of aggregates and particularly ␤-sheet-rich structures, as reported by thioflavin dyes (20) and conformation-specific antibodies is a prerequisite for its neuronal uptake.
We found that cultured human neuroblastoma cells (SH-EP cells) can efficiently internalize ␤-sheet-rich aggregates of A␤  at nanomolar concentrations, and that cellular uptake was associated with metabolic inhibition. In contrast, monomeric A␤  only entered cells inefficiently and at high nanomolar concentrations. Internalization occurred after aggregation on the plasma membrane, which suggests that the membrane is an environment that facilitates aggregation of A␤  monomers.

Results
Preparation and Characterization of Fluorescently Labeled A␤(1-42)-To investigate the relationship between the aggregation state of A␤(1-42), its internalization and cytotoxicity, we examined whether the presence of the secondary structure determines the behavior of A␤  toward the cells and membranes. Aggregated A␤, which may contain oligomers, protofibrils, and fibrils, was classified in two populations: a state without ␤-sheet structure and a state in which ␤-sheets have been formed. Concentrations of aggregated A␤  peptide are expressed in terms of their amount of A␤ monomers as "equivalent monomer concentration." To track the peptide and to observe aggregation of monomeric A␤  in cultured cells, we generated A␤  peptides that were fluorescently labeled at the N terminus with Atto565-maleimide (A␤   565 ), Atto488-maleimide (A␤   488 ), or Atto633-maleimide (A␤   633 ) as described below. Labeled A␤  was separated by size exclusion chromatography (SEC, Fig. 1A). A␤ eluted in two populations at 0.86 (protofibrils, PF) and 1.39 ml (monomer, M).
We used unlabeled monomers and aggregation-incompetent sc-A␤   565 as controls on SEC, both of which eluted similar to the A␤(1-42) 565 monomer fraction (M, Fig. 1A). Those monomers could not be detected by atomic force microscopy (AFM, Fig. 1B). In contrast, AFM identified the PF fraction of A␤   565 to be protofibrils with an average height of 4 nm (Fig. 1B). When incubated with ThT, the protofibrils (35 M monomer equivalent) bound the dye immediately and ThT fluorescence changed little over the course of 1 h, whereas ThT fluorescence of the monomeric A␤ (35 M) displayed sigmoid kinetics typical of nucleated polymerization (Fig. 1C).
We used labeled monomers (M) and protofibrils (PF, Fig. 1B) to compare their respective cellular uptake. Mature fibrils (Fig.  1E) were used to compare their uptake to that of protofibrils and monomers. For kinetic analysis of cellular uptake mixtures of 90% unlabeled A␤(1-42) and 10% A␤(1-42) 565 were used.
Efficient Uptake of A␤  Protofibrils-We treated cultured human neuroblastoma (SH-EP) cells with A␤(1-42) 565 monomers or protofibrils to determine whether both A␤ species are internalized differently. Cells were imaged by confocal microscopy after incubation with A␤(1-42) 565 at 37°C for 24 h (Fig. 2, A-E). A soluble membrane-impermeable fluorescent dye, calcein (20 M), was added to the extracellular medium to mark the uptake of extracellular medium during vesicular trafficking from the plasma membrane into the cells in endocytic vesicles (Fig. 3A). Intracellular A␤(1-42) 565 could be detected after treatment with protofibrils (150 nM, Fig. 2A) or with mono-mers at a higher concentration (500 nM, Fig. 2B). Treatment with low concentrations of monomeric A␤(1-42) 565 (150 nM, Fig. 2C) or with an aggregation-incompetent scrambled peptide, sc-A␤(1-42) 565 (500 nM, Fig. 2D), did not lead to detectable A␤   565 inclusions, as was the case for untreated cells (Fig. 2E, n.c.). Quantitative analysis of A␤(1-42) 565 fluorescence in the endocytic vesicles and in the cytosol revealed that a low level cytosolic A␤(1-42) 565 signal could be detected after treatment with protofibrils (150 nM) or high concentration (500 nM) monomeric A␤(1-42) but not at low monomer concentrations (150 nM). The A␤(1-42) 565 signal in the surrounding cytosol was about 30-fold lower than in the vesicles (Fig. 2F). The lack of uptake of sc-A␤(1-42) 565 demonstrated that internalization was specific for the A␤(1-42) sequence and was not a result of the fluorescent label (Fig. 2D).
We then verified that internalization of A␤(1-42) 565 into endocytic vesicles was also observed in primary hippocampal neurons (Fig. 3, A-C). After incubating neurons with aggregated A␤    3A) and for monomeric A␤  at 500 nM ( Fig. 3B), but not for monomeric A␤  at 150 nM (Fig. 3C), supporting our results from the neuroblastoma cell model. On the images, internalized A␤   565 shown in neurons (Fig. 3, A and B) and SH-EP cells (Fig. 2, A and B) seem different in quantity and distribution, which may be the result of the isolation of glia-free hippocampal neurons, or may reflect a general higher endocytic activity of neurons under the same experimental conditions.
To test whether exogenous fluorescence represented A␤ species, we co-stained cells with the anti-␤-amyloid antibody 6E10, which confirmed that Atto565 fluorescence indeed indicated the presence of A␤ (Fig. 3, D and E). Conversely, 6E10 staining detected only endogenous amyloid precursor protein in the absence of A␤(1-42) 565 (Fig. 3F). These results suggest that protofibrillar A␤  is taken up at lower concentrations than the monomers, and that an effective uptake of monomer may require A␤ to form an ordered aggregate structure prior to cellular uptake.
Uptake of Monomer Requires a Minimum A␤ Concentration-To quantitatively analyze the dependence of monomer uptake on the concentration of A␤(1-42), we titrated the cells either with A␤(1-42) monomer or protofibrils at 15-1500 nM (equivalent monomer concentration), and quantified intracellular A␤(1-42) as a function of A␤(1-42) concentration by the fluorescence of labeled A␤(1-42) using HCS microscopy (Fig. 4A).
We tracked the amount of internalized A␤ as a function of time and found maximal intracellular A␤    A␤  was observed at concentrations of 150 nM and below (Fig. 4, C and D). In contrast, no such threshold concentration existed for the internalization of preformed A␤(1-42) 565 aggregates (Fig. 4, E and F). The observed threshold concentration for monomer uptake is similar to the critical concentration (22) that was reported for the aggregation of A␤(1-42) monomers, which suggests that aggregation may be a prerequisite for uptake.
Alternatively, differences in the degradation of A␤(1-42) by proteases, either extracellular prior to uptake or intracellular, could cause the higher amount of A␤  in the cells. To test this alternative hypothesis, we monitored concentrations of A␤  in the cell culture media and in cell lysates as a function of time by immunoblotting (Fig. 4, G and H). We observed a pronounced increase of A␤  in the lysates of cells incubated with aggregated A␤(1-42) (500 nM) and a smaller increase in cells incubated with monomer (500 nM) after 24 h, mirroring the results from HCS microscopy (Fig. 4B). We then monitored the A␤(1-42) concentration in the culture media, either in the presence of cells or in control samples that were incubated without cells for 24 h. When monitoring the A␤(1-42) concentration in the culture media, no decrease of A␤(1-42) over time was observed for monomeric A␤  in the presence of cells or for either form of A␤  in the control samples without cells. These data indicate that monomeric A␤(1-42) is not degraded by proteases in the media, which would prevent its uptake. Rather, only the concentrations of aggregated A␤(1-42) were reduced substantially over time, likely reflecting its more efficient binding to, and uptake into cells. Taken together, these data strongly support our initial hypothesis that uptake of A␤(1-42) does indeed depend on its aggregation state. If so, it would be important to determine the type of aggregates that are efficiently taken up by the cells.
Efficient Uptake Correlates with Formation of ␤-Sheet-rich Aggregates-To determine which A␤ species could be efficiently taken up, we combined in vitro aggregation kinetics and cellular uptake experiments. A␤(1-42) amyloid formation was monitored in vitro to collect A␤(1-42) species at different  A␤(1-42) (15 M, 10% A␤(1-42) 565 ) aggregation kinetics were monitored by ThT fluorescence (Fig. 5A). Samples were collected at four time points that corresponded to different phases of peptide aggregation: initiation (t0), lag-phase (t1), growth phase (t2), and plateau phase (t3). The samples were characterized by CD spectroscopy (Fig. 5B). Within the lag phase (t0 and t1), A␤(1-42) partially lost its disordered state. It adopted a ␤-sheet structure during the growth phase (t2 and t3) as indicated by a minimum at 218 nm in the CD spectra. When centrifuged at 200,000 ϫ g, soluble A␤(1-42) was lost during its growth phase (t2-t3, Fig. 5C). To determine which of these aggregation states are efficiently taken up, the SH-EP cells were incubated with A␤ species (t0 -t3) at 37°C at 150 nM equivalent monomer concentration, which is below the threshold concentration of A␤(1-42) 565 monomer internalization (Fig. 4D). After 24 h incubation, intracellular A␤(1-42) aggregates were quantified by HCS microscopy (Fig. 5D). Intracellular A␤(1-42) 565 aggregates were detected in cells that had been treated with growth (t2) and the steady-state phase (t3) aggregates. In contrast, no uptake of A␤(1-42) peptides was observed with samples from the initiation (t0) and lag phase (t1) of the poly-merization reaction. This suggests that ␤-sheet-rich A␤  aggregates are efficiently internalized, whereas unstructured monomers (t0) or small aggregates/oligomers (t1) without ␤-sheet structure are not.
HCS data from Fig. 4, C and E, show that neither form, monomer or small aggregates, had a strong effect on the number of living cells, i.e. internalized A␤(1-42) 565 did not cause cell death within 24 h. We therefore analyzed cell metabolic activities by 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) reduction, which is a well established marker of early mitochondrial toxicity (23).
A␤  aggregates formed during the growth phase (t2 and t3) were significantly more toxic than monomers (t0) and samples collected during the lag-phase (t1) and inhibited mitochondrial metabolic activity (Fig. 5E). Under our experimental conditions A␤(1-42) monomers became cytotoxic only at concentrations above 150 nM (Fig. 5F), a point at which they also underwent internalization (Fig. 4, C and D).
It should be noted that the dilution to 150 nM might change the structure of A␤  in the medium from the structure recorded by CD at 15 M (Fig. 5B). Nevertheless, at concentrations of 150 nM and below only A␤(1-42) that had previously formed ␤-sheet-rich structures (t2, t3) was efficiently internal- ized, whereas unstructured monomers (t0) or small aggregates (t1) without ␤-sheet structure were not. This suggests that it was indeed the ␤-sheet aggregates that were taken up, rather than a different A␤(1-42) species. The question whether the intracellular A␤(1-42) still had ␤-sheet structure will be addressed in detail below.
These data support the hypothesis that the formation of a ␤-sheet structure may be a prerequisite for efficient uptake and subsequent cytotoxicity of A␤(1-42) aggregates in neuronal model cells. The results raise the question where A␤(1-42) aggregates form. Aggregation could occur in the medium or, alternatively, the cellular membrane may serve as a platform for the aggregation process.
Multimerization on Plasma Membrane Precedes Uptake of A␤(1-42)-To investigate the location of aggregate formation and to better track the dynamics of monomer uptake, we repeated the uptake experiment under conditions that allowed us to separate in time A␤(1-42) membrane interaction from its uptake into the cell. The cells were cooled to 4°C, which slows down both A␤(1-42) aggregation and cell metabolism, and then were treated with A␤ at 4°C for 45 min. The cells were then washed with ice-cold PBS to remove unbound A␤(1-42), and then restored at 37°C in fresh media (Fig. 6A).
Under these conditions (4°C, 60 min), even at a higher monomer concentration of 1 M, A␤(1-42) 565 was located only on the plasma membrane and no uptake could be observed (Fig.  6B). Calcein (20 M) uptake displayed that the cell still had some endocytic activity under these conditions. First, we tested whether monomeric A␤(1-42) assembles into larger species on the cell membrane. We incubated SH-EP cells with two types of fluorescently labeled monomeric A␤(1- We then tested if those membrane-bound A␤(1-42) aggregates did indeed form on the plasma membrane and could be taken up after their aggregation. To do so, SH-EP cells were treated successively with monomeric A␤   488 and A␤   633 , each at 500 nM concentration. SH-EP cells were first only incubated with monomeric A␤(1-42) 488 at 4°C for 30 min, which was then washed off. After removing the unbound A␤(1-42) 488 , the cells were treated with A␤(1-42) 633 at 4°C for another 30 min. After that, the cells were washed again to remove unbound A␤ , incubated in fresh medium at 37°C for 3 h to permit internalization of A␤ , and imaged by confocal microscopy (Fig. 6D). Both labeled A␤(1-42) species colocalized in the endocytic vesicles. The FRET signal indicates that both had formed di-or multimeric complexes. Because the washing step had removed all A␤(1-42) 488 from solution, we conclude that A␤(1-42) 488 interacted with A␤(1-42) 633 while located on the plasma membrane. The membranebound aggregates were then taken up by endocytosis after the cells were returned to 37°C.
To verify that the FRET signal is specific to physical interaction of A␤(1-42) monomers and is not an artifact of the label or of colocalization, we used transferrin (Tf) as a negative control. Tf binds to a membrane receptor and is internalized via clathrin-mediated endocytosis but does not aggregate on the membrane. After incubating the cells at 4°C with two types of fluorescent transferrin conjugates (Tf/A488 and Tf/A647), we observed colocalization of both Tf conjugates. However, the colocalization did not correspond to a FRET signal (Fig. 6E). This strongly suggests that the FRET between A␤(1-42) molecules results from a specific physical interaction.
For a more comprehensive analysis of the membrane-assisted self-assembly and subsequent uptake of A␤(1-42) monomers, the cells were treated with monomeric A␤(1-42) 565 at various concentrations (0 -1600 nM) at 4°C for 45 min. Subsequently the temperature was raised to 37°C to permit internalization of A␤(1-42) and the cells were incubated in fresh medium for 24 h. Intracellular A␤(1-42) 565 was then quantified by HCS microscopy (Fig. 6F). Intracellular A␤(1-42) 565 aggregate signals could be detected at A␤(1-42) 565 monomer concentrations above 200 nM, similar to the threshold concentration we previously observed for monomer uptake at 37°C (Fig. 4, B and C). These data demonstrate that A␤(1-42) binds to the plasma membrane and forms aggregates prior to cellular uptake, that the cell membrane assisted A␤(1-42) self-assembly, and that uptake of monomeric A␤(1-42) requires a critical A␤(1-42) concentration for membrane-assisted uptake.
Conceivably, small A␤(1-42) aggregates that are competent for uptake could be forming in solution and then bind to the plasma membrane. However, a comparison of conditions and time scales of A␤(1-42) aggregation in solution with those of the membrane binding experiment makes this hypothesis less likely. Incubation of cells with A␤(1-42) monomer for 45 min at 500 nM was sufficient to initiate uptake. In contrast, ThT positive A␤(1-42) species did not form in solution under the same conditions within 3 h even at monomer concentrations of 1500 nM (Fig. 6G).
Finally, we probed the complex formation of A␤(1-42) 488 and A␤(1-42) 633 (150 nM each) by FRET as a function of time (Fig. 6H). Little FRET signal was observed within the first 3 h, supporting our conclusion that membrane binding coincides with, or precedes interaction at this peptide concentration. Conversely, no FRET signal was observed after the co-incubation of scrambled sc-A␤(1-42) 488 and A␤(1-42) 633 (150 nM each), reinforcing that interaction depends on the A␤ peptide sequence and not on the fluorescent label. Formation of FRETpositive species also coincided with the time course of A␤ aggregation in vitro (Fig. 6, G and H), suggesting that the FRET signal results from multimeric aggregated A␤ species.
We therefore conclude that binding to the plasma membrane occurred within the lag-phase of ThT aggregation kinetics, well before the formation of ␤-sheet-rich aggregates. Because our previous experiment demonstrated that formation of ␤-sheet Aggregation Initiates A␤(1-42) Uptake SEPTEMBER 9, 2016 • VOLUME 291 • NUMBER 37

JOURNAL OF BIOLOGICAL CHEMISTRY 19597
structures in vitro correlated with efficient uptake (Fig. 5, B and C), these data suggest that A␤(1-42) peptides may convert into ␤-sheet aggregates on the membrane prior to uptake after soluble A␤(1-42) 565 peptides were removed. We therefore analyzed, whether intracellular A␤(1-42) contained ␤-sheet structures.
Internalized A␤  Contains ␤-Sheet-rich Structures-To identify the secondary structure of internalized A␤  in the cultured cells, we stained cells with the amyloidophilic dye ThS and the anti-amyloid fibril antibody LOC after A␤(1-42) uptake. ThS is widely used to visualize ␤-sheet-rich amyloid structures in histology (20,24). It displays enhanced fluorescence emission on ␤-sheet binding, and has higher affinity and less pH sensitivity than ThT (20). When pre-aggregated A␤(1-42) (t1-t3) is taken up, both ThS and LOC staining colocalize with A␤(1-42) 633 fluorescence confirming the presence of fibrillar aggregates (Fig. 7, A and B). However, ThS and LOC signals after incubation with monomeric A␤(1-42) do not clearly demonstrate the presence of ␤-sheet rich fibrils (Fig. 7,  A and B). This leaves open the possibility that the A␤(1-42) species detected by FRET have not yet formed ␤-sheet structures.
To resolve whether this conversion occurs, we performed a competitive uptake experiment between monomer at a subcritical concentration (A␤(1-42) 565 , 150 nM), which was co-incubated with ␤-sheet aggregates (A␤(1-42) 633 , 900 nM). Both species were labeled with different fluorophores to track in which compartment both monomers and aggregates were found. In addition, ␤-sheet-rich structures were stained by ThS. According to our previous results, we would expect monomers to be taken up only if they converted into ␤-sheet-rich structures via co-aggregation with pre-formed aggregates and that internalized A␤(1-42) co-stains with ThS.
SH-EP cells were treated with A␤(1-42) 565 monomer (150 nM) at 4°C for 20 min. Monomeric A␤ was removed, and then cells were treated with A␤(1-42) 633 aggregates at 900 nM at 4°C for 20 min, washed with ice-cold cell culture medium, and stained with ThS at 4°C for 10 min. After that, the cells were further incubated in fresh cell culture medium with calcein (20 M) at 37°C for 1 h. After removing the calcein-rich medium, the cells were incubated in calcein-free medium at 37°C, and live cell images were taken by spinning disk confocal microscopy (Fig. 7C).
We quantitatively analyzed internalized and surface-bound species of monomeric A␤ from confocal live cell imaging (Fig.  7E). All monomeric A␤ found inside the cell both colocalized with preaggregated A␤ fibrils and stained strongly with ThS, whereas ϳ 1 ⁄ 3 of surface-bound monomeric A␤ did not colocalize with fibrillar A␤ or stained with ThS. We did not find any monomeric A␤ that stained with ThS without colocalizing with fibrillar A␤ either inside the cells or on the cell surface under these experimental conditions. This means that subthreshold concentrations of A␤(1-42) monomers were capable of binding to the plasma membrane. However, co-aggregation into ␤-sheet-rich structures was needed for their efficient uptake into the cells.
Finally, we monitored the incorporation of A␤ monomers into ␤-sheet-rich aggregates on the surface of SH-EP cells by live cell FRET imaging (Fig. 7F, supplemental Fig. S1). Monomeric A␤ coaggregated with A␤ fibrils within 30 min into structures that also bound ThS. Taken together, these data strongly support our hypothesis, that aggregation into ␤-sheet structures facilitates the efficient uptake of A␤(1-42) by neuroblastoma cells via an endocytic mechanism.
A␤  Aggregate Uptake Is Clathrin Independent-The location of A␤  aggregates in endocytic vesicles raises the question by which endocytic pathway ␤-sheet-rich aggregates are internalized. Here we tested whether A␤  aggregates are internalized by clathrin-mediated endocytosis (CME). The CME pathway was blocked by lowering the temperature to 4°C   25). Transferrin binds to the transferrin receptor on the plasma membrane and then can be internalized via coated pits. Fluorescent transferrin conjugates are therefore broadly used as markers for investigating the early phase of CME (25)(26)(27)(28). At 4°C, transferrin only binds to its receptor on the plasma membrane, and will not be taken up (25). Therefore, transferrin was used as specifically reported for inhibition of the CME pathway. However, it should be noted that at this temperature the fluidity of membrane is limited, which may also influence other endocytic mechanisms or the aggregation of A␤(1-42) (29,30).
Following incubation with monomeric and pre-aggregated A␤(1-42) 565 (150 nM, respectively) calcein (20 M) and transferrin (10 ng/ml) at 4°C for 45 min, SH-EP cells were fixed and imaged by confocal microscopy (Fig. 8, A and B). Transferrin was located exclusively at the cell membrane, demonstrating that CME was efficiently inhibited. Under these conditions A␤(1-42) monomer was only observed on the plasma membrane and was not internalized. However, the A␤(1-42) on the membrane did not colocalize with transferrin. The uptake of pre-aggregated A␤(1-42) species was reduced but not totally inhibited under these conditions (Fig. 8A). However, A␤  found in endocytic vesicles also did not colocalize with transferrin. These data suggest that uptake of A␤(1-42) is independent of CME.

Discussion
Taken together, our data allow us to map the first steps in A␤  internalization starting from the monomeric peptide. First, A␤(1-42) monomers or very early disordered oligomers bind rapidly to the plasma membrane. Their binding partner could be either the lipid bilayer itself or plasma mem-brane proteins. Numerous studies have demonstrated interaction of A␤ with lipid bilayers. Its role in A␤(1-42) uptake will be scrutinized in a future study. There, they aggregate either on the membrane surface itself or in a compartment that is very close to the membrane, from which they are taken up into endocytic vesicles. The highly efficient uptake of ␤-sheet-rich structures suggests that aggregation into, or co-aggregation with ␤-sheet aggregates occurs.
For A␤(1-42) monomers, the peptide concentration has to be sufficient to initiate the first two steps of the process, resulting in a threshold for internalization at the saturation concentration, or critical concentration, for A␤(1-42) aggregation (22). In contrast, no concentration threshold exists for the internalization of preformed A␤(1-42) aggregates with ␤-sheet structure.
This has several implications for the possible mechanism of A␤(1-42) toxicity. In our experiments metabolic inhibition was directly correlated with the formation and uptake of ␤-sheet-rich aggregates. Aggregated species have been found to play a central role in A␤(1-42) toxicity, not all of which are large ␤-sheet-rich structures (5). Although ␤-sheet-rich A␤ that was efficiently endocytosed and that inhibited mitochondrial activity could be pelleted at 200,000 ϫ g (Fig. 5), very large aggregate structures were unable to enter the cell. Our data therefore suggest that neurons preferably take up ␤-sheet-rich oligomeric and protofibrillar structures of intermediate size. It is tempting to speculate that small oligomers may have a higher affinity to the plasma membrane than the monomeric peptide, facilitating the conversion to ␤-sheet-rich structures and subsequent internalization. Our results demonstrate that, unlike monomers, preexisting A␤ aggregates are internalized at low nanomolar concentrations, which corresponds to previously observed binding of A␤ oligomers to neuronal plasma membranes at nanomolar concentration (31,32).
Our experiments did not provide evidence that uptake of aggregates proceeds via CME. A␤  can enter the cells via a non-clathrin-mediated pathway, and then locate in the endocytic vesicles. Costaining with caveolin suggests a possible uptake route via the caveolin endocytosis pathway. A␤  is also believed to be involved in cholesterol and caveolin trafficking (33).
A␤  aggregates may be taken up via receptor independent endocytosis, as had been observed previously (34). Other pathways for the internalization of amyloidogenic proteins have been discussed. Synthetic peptide aggregates of sizes Ͻ500 nm were taken up into HEK cells by nonspecific endocytosis, whereas larger aggregates were internalized by a mechanism similar to phagocytosis (35). Tau aggregates can be internalized via micropinocytosis, mediated by glycosaminoglycans (36). It is possible that A␤(1-42) aggregates enter the cell by the same pathway.
Second, we found that A␤(1-42) aggregates can form in a concentration-dependent manner through self-assembly of A␤(1-42) on the plasma membrane and that internalized A␤(1-42) aggregates have ␤-sheet structure. It has long been known that lipid interaction promotes A␤(1-40) transition to ␤-sheet structure (37) and our data support the interpretation that this process is central to A␤ uptake and toxicity. Our data suggest the binding of A␤(1-42) to the lipid bilayer or to membrane proteins may be the first step in the formation of cytotoxic A␤(1-42) aggregates de novo. Factors that increase partitioning of A␤ to the plasma membrane will therefore likely promote the formation of ␤-sheet-rich aggregates on the membrane. These include lipid peroxidation products, such as secocholesterol and 4-hydroxy-nonenal that facilitate A␤ membrane binding and aggregation (38 -40), divalent metal ions promoting A␤ membrane interaction (41), and interaction with membrane proteins (16). Our data support a central role of endocytosis in A␤ cytotoxicity (15) and provide strong evidence that aggregation precedes endocytosis, a question that had previously not been conclusively resolved (7,8). This experimental evidence may improve our understanding of AD pathology and may inform more focused therapeutic approaches targeting membrane binding and self-assembly of the A␤ peptide.  and A␤(1-42) with a single N-terminal cysteine residue (Institute for Medical Immunology, Charité, Berlin, Germany) were dissolved in hexafluoro-2-propanol and incubated at room temperature overnight. After flash freezing by liquid nitrogen, hexafluoro-2propanol was removed by lyophilization (Savant SpeedVac, Thermo), and the peptides were stored at Ϫ20°C until use.

Preparation of Monomers and Fluorescent Labeling of A␤(1-42)-Synthetic human A␤
To prepare unlabeled monomer, lyophilized A␤(1-42) was dissolved in 10 mM NaOH, sonicated for 30 min in ice-cold water bath, and passed through a 0.22-m and a 30-kDa filter (Millipore). The monomers were kept on ice and used immediately or within 1 h.
Monitoring Aggregation Kinetics of A␤  by ThT and SDS-PAGE-A␤(1-42) monomer aggregation assays were performed in PBS or 50 mM sodium phosphate, pH 7.4, as indicated, containing 20 M ThT in a microplate reader (InfinitE M200, Tecan, Austria) at 37°C. ThT fluorescence (excitation wavelength of 440 nm and emission wavelength of 485 nm) was measured every 5 min after 5 s shaking. 50 l of A␤(1-42) was removed at different time points as indicated and centrifuged for 30 min at 200,000 ϫ g (Beckman TL-100). Supernatants were analyzed by denaturing SDS-PAGE.
Circular Dichroism Spectroscopy-A␤(1-42) samples (15 M) in PBS were measured in a 1-mm path length cuvette. Circular dichroism (CD) spectra were recorded between 200 and 260 nm with a step size of 1 nm in a CD spectrometer (J-720, Jasco, Japan).
Atomic Force Microscopy-10-l samples were loaded onto freshly cleaved mica (mica was glued on a glass slide) for 5 min, washed with freshly filtered deionized water (3 ϫ 100 l), and dried overnight. All the images were taken using intermittent contact mode on a Nanowizard II/Zeiss Axiovert setup (JPK).
Cell Culture and Cellular Uptake-SH-EP cells were maintained in culture medium (10% (v/v) fetal bovine serum, 4 mM L-glutamine, 110 mg/liter of sodium pyruvate, 100 units/ml of penicillin-streptomycin, 4.5 g/liters of D-glucose in 500 ml of Dulbecco's modified Eagle's medium (Gibco)) at 37°C and 5% CO 2 . Cells were seeded into 96-well plates or 35-mm cell cul-ture dishes (MatTek) and grown to about 80% confluence. For cellular uptake, the cells were incubated with A␤(1-42) or labeled A␤  at various concentrations, temperatures, and times. The cells were then washed twice with PBS and fixed with 3.7% paraformaldehyde for 10 min at room temperature. Primary rat hippocampal neurons were a gift of J. Meier, MDC Berlin, and prepared and cultured as previously described (42).
Dot Blot Quantification of A␤-SH-EP cells were cultured as described above and seeded in a 96-well plate (10,000 cells/well) for quantification of A␤  in supernatant and in a 12-well plate (80,000 cells/well) for quantification of A␤  in cell lysate. Cells were incubated with monomeric A␤  or A␤(1-42) that was pre-aggregated to the t3 time point (Fig. 5A) in sodium phosphate buffer (50 mM, pH 7.4) as indicated. Cells were washed in PBS, lysed in 100 l of Tris buffer (50 mM, pH 5.1) containing 2% SDS, and boiled for 10 min. 100 l of cell lysates or culture media was run through a nitrocellulose membrane (Bio-Rad) on a 96-well vacuum apparatus. Serial dilutions for monomeric and pre-aggregated A␤(1-42) were spotted in triplicate on the same membrane as the cell samples, and were used as standard reference for calculation of A␤  concentrations. The membrane was incubated with blocking buffer (5% BSA in PBS) for 1 h, was washed twice with PBS-T buffer (0.1% Tween 20 in PBS) for 10 min, and then incubated for 1 h with primary antibody, anti-␤-amyloid mouse monoclonal antibody (1:500, 6E10, Covance), followed by secondary antibody, IRDyeGAM-800CW (1:15,000, Li-Cor), in PBS-T buffer containing 0.5% BSA. Fluorescence of three replicate wells was evaluated through densitometry. A␤ concentrations were calculated from signals of reference dilution samples by linear regression analysis.
For live cell imaging, cells were incubated with ThS at 10 g/ml in cell culture medium at 37°C for 10 min. Then the cells were washed five times with PBS and kept in cell culture medium until microscopy.