Specific Compositions of Amyloid-β Peptides as the Determinant of Toxic β-Aggregation*

Alzheimer's disease (AD) may be caused by toxic aggregates formed from amyloid-β (Aβ) peptides. By using Thioflavin T, a dye that specifically binds to β-sheet structures, we found that highly toxic forms of Aβ-aggregates were formed at the initial stage of fibrillogenesis, which is consistent with recent reports on Aβ oligomers. Formation of such aggregates depends on factors that affect both nucleation and elongation. As reported previously, addition of Aβ42 systematically accelerated the nucleation of Aβ40, most likely because of the extra hydrophobic residues at the C terminus of Aβ42. At Aβ42-increased specific ratio (Aβ40: Aβ42 = 10: 1), on the other hand, not only accelerated nucleation but also induced elongation were observed, suggesting pathogenesis of early-onset AD. Because a larger proportion of Aβ40 than Aβ42 was still required for this phenomenon, we assumed that elongation does not depend only on hydrophobic interactions. Without any change in the C-terminal hydrophobic nature, elongation was effectively induced by mixing wild type Aβ40 with Italian variant Aβ40 (E22K) or Dutch variant (E22Q). We suggest that Aβ peptides in specific compositions that balance hydrophilic and hydrophobic interactions promote the formation of toxic β-aggregates. These results may introduce a new therapeutic approach through the disruption of this balance.

A␤ Sample Preparation-A stock solution of A␤ was prepared by dissolving powdered A␤ peptide in 100% HFIP to a final concentration of 1 mg/ml. After shaking for two hours at 4°C, the A␤ stock solution was aliquoted into siliconized tubes and stored at Ϫ80°C. Just prior to each experiment, aliquots were spin-vacuumed using an Integrated SpeedVac system (Savant). For experiments shown in figures except Fig. 1, aliquots of A␤40 were redissolved in 50% HFIP/14% NH3 solution and then spin-vacuumed. They were then dissolved in a HEPESbuffered solution.
Fluorescence Spectroscopy (ThT Assay)-The degree of ␤-aggregation was determined using the fluorescent dye, ThT, which specifically binds to fibrous structures (16). First, A␤ stock solution (see above) was diluted with 5 or 10 mM HEPES-NaOH and 0.9% NaCl. ThT was added to each test sample to a final concentration of 10 M. Each sample was prepared in 96-well Black Cliniplates (Labsystems) and shaken for 10 s prior to each measurement. Measurements were carried out every 20 min.
The relative degree of ␤-aggregation was assessed in terms of fluorescence intensity, which was measured at 37°C using a Fluoroskan Ascent FL spectrophotometer (Labsystems, Finland). Measurements were performed at an excitation wavelength of 444 nm and an emission of 485 nm, wavelengths that result in the optimum detection of bound ThT. To account for background fluorescence, the fluorescence intensity measured from each control solution without A␤ was subtracted from that of each solution containing A␤. By making three or six wells of each condition, their standard deviations were first calculated. They were then divided by the square root of n.
Cell Cultures-Human embryonic kidney (HEK) 293 cells were used to test the toxic effects of A␤-aggregation as assessed in the 3-[4,5dimethylthiazol-2-yl]-2,5-diphenyltetrazolium bromide (MTT) assay (see below). They were grown in Dulbecco's modified Eagle's medium (Sigma) containing 10% fetal bovine serum (HyClone) and incubated in a humidified chamber (85% humidity) containing 5% CO 2 at 37°C. On the morning of A␤ treatment, the cell culture medium was replaced with serum-free Dulbecco's modified Eagle's medium, and the cells were plated onto a 96-well, coated plate (Corning) at a final cell count of 20,000 cells/well. The cell viability was subsequently assessed using the MTT assay (see below).
MTT Assay-The toxic effects of spontaneous ␤-aggregation of A␤ on cell cultures were assessed. Solutions containing A␤ were prepared as described for the ThT assay, and then the samples were transferred immediately into wells containing the HEK293 cell (see "Cell Cultures"). After certain periods, MTT was added to each well, and the plate was kept in a CO 2 incubator for an additional 2 h. The cells were then lysed by the addition of a lysis solution (50% dimethylformamide, 20% SDS, pH 4.7) and were incubated overnight. The degree of MTT reduction in each sample was subsequently assessed by measuring absorption at 570 nm at 37°C using a Bio Kinetics Reader EL340 (Bio-Tek Instruments). Background absorbance values, as assessed from cell-free wells, were subtracted from the absorption values of each test sample. The absorbances measured from three or six wells were averaged, and the percentage MTT reduction was calculated by dividing this average by the absorbance measured from a control sample without A␤. By making three or six wells of each condition, their standard deviations were first calculated. They were then divided by the square root of n.
Immunoblotting Assay-A␤ sample was incubated for each time indicated at 37°C as described above (see "ThT Assay"). Samples were collected and mixed with buffer containing glycerol but no reducing agents such as SDS or 2-mercaptoethanol. The formation of intermediate ␤-aggregates was assessed by 16.5% Tris-Tricine SDS-PAGE and immunoblotting using monoclonal antibody 4G8 (Signet Laboratories, Inc.), which recognizes residues 17-24 of A␤.
Electron Microscopy-The ultrastructural characteristics of A␤-aggregates were examined as follows. A␤ sample was incubated for indicated periods at 37°C as described above (see "ThT Assay") and was mixed for 10 s every 20 min. 3 l of each sample was applied to 300-mesh copper grids with formvar supporting membrane, blotted with filter paper, and stained with 1% (w/v) neutralized tungstophosphoric acid (Nacalai tesque, Inc., Kyoto, Japan). The specimens were examined on a LEO 912AB electron microscope (LEO, Oberkochen, Germany), at an accelerating voltage of 100 kV.

RESULTS
We first assessed the baseline kinetics of ␤-aggregation. The time course of ␤-aggregation was determined by incubating A␤40 with ThT, a compound that fluoresces when specifically bound to ␤-sheet structures (16) (Fig. 1). Fluorometric measurements were carried out every 20 min for ϳ72 h at 37°C. As expected (11)(12)(13), the ␤-aggregation kinetics was sigmoidal over the first 5 h of incubation (Fig. 1A). With increasing incubation, however, ␤-aggregation decreased slowly (Fig. 1B), decaying to an approximate asymptotic level of less than half the peak aggregation.
Because there are several competing hypotheses concerning the factor(s) associated with AD pathogenesis, it was of interest to analyze the toxicity, biochemical constituents, and structure of the aggregates during ␤-aggregation. At selected intervals during the first two days of incubation, A␤/ThT samples were collected, transferred to HEK293 cell cultures grown in serumfree medium, and the toxicity of the A␤/ThT solution was examined using the MTT assay, which assesses cell viability by measuring mitochondrial activity (17) (Fig. 1C). Even after only a short period of incubation, the ␤-aggregates that formed were highly toxic. With longer incubation, however, the ␤-aggregates were less toxic. We next determined the time course of the formation of different A␤ intermediates during the course of ␤-aggregation. The A␤/ThT solutions were collected at selected times during aggregation and were analyzed by Western blot. Aliquots were applied to SDS-polyacrylamide gels and immunoblotted with an anti-A␤ antibody (Fig. 1D). Samples collected immediately after being incubated separated out as a thick band the size of the A␤ monomer and a thinner, less distinct band the size of the A␤ dimer, indicating that some dimerization had occurred almost immediately upon solubilization. After 2 h of incubation, the A␤ monomer band was less distinct, and an immunoreactive smear appeared, suggesting the formation of A␤ polymers. As incubation progressed, the amount of monomeric A␤ decreased. After extended incubation, we noticed increased staining at the bottom of the loading wells that presumably represented large aggregates that could not be resolved by the gel. The macromolecular structures of these various intermediates were examined using electron microscopy ( Fig. 1, E-H). No distinguishable fiber-like structures were observed in samples collected soon after incubation began or in samples collected after 1 h of incubation. After 2 h of incubation, however, small oligomeric aggregates and long fibrils were observed (Fig. 1, E and F). The fibrils appeared to be comprised of two smaller filaments that were twisted around each other (Fig. 1F), but no further appreciable association between fibrils was observed. In contrast, after 4 h of incubation, the time point at which the greatest amount of ␤-aggregates formed (Fig. 1C), long fibers appeared to be gathered closely together in a net-like structure (Fig. 1G). After 27 h of incubation, A␤ fibers, aligned side by side and formed numerous fiber bundles (Fig. 1H). Interestingly, samples from this time period in the toxicity experiments were only moderately toxic; the most toxic A␤-aggregates were those from the initial growth stage of fibrillogenesis. These findings prompted us to search for factors that cause an acute increase in the initial stage of ␤-aggregation.
Because individuals with familial Alzheimer's disease have larger proportions of A␤42 compared with normal individuals (5-10), we examined the hypothesis that elevated levels of A␤42 may play a role in the initial stage of ␤-aggregation. Thus, we measured the time course of ␤-aggregation formed by various ratios of A␤40 and A␤42 ( Fig. 2A). Consistent with previous reports (11)(12)(13), incubation of A␤42 alone showed signs of ␤-aggregation almost immediately after incubation began. After this initial increase, however, aggregation remained at a fairly constant, modest level without further significant increase. Pretreatment of A␤42 with either HFIP or HFIP/NH 3 did not significantly alter the time course of ␤-aggregation. On the other hand, although the lag time until the start of ␤-aggregation of HFIP/NH 3 -treated A␤40 was much longer, the formation of ␤-aggregates increased to higher levels compared with that for A␤42. Because longer lag time seemed more appropriate to examine the initial steps of aggregation in detail, HFIP/NH 3 -treated A␤40 was used in the rest of the experiments. When the proportion of A␤42 was increased to the A␤40:A␤42 ratio of 10:1, ␤-aggregation kinetics systematically shifted to the left. Thus, A␤42 accelerated the seeding or nucleation of A␤40. When the proportion of A␤40 to A␤42 mimicked the proportion reported in some familial Alzheimer's disease mutants (i.e. A␤40:A␤42 ratio of 10:1) (10, 18), ␤-aggregation was further accelerated and also increased to a much higher level. Moreover, these A␤42-induced aggregates were highly toxic when assessed with the MTT assay (Fig. 2C). At the end of the incubation period, the level of A␤42-associated aggregation was lower than that of A␤40 (Fig. 2B). Aggregates formed by A␤42 alone were less toxic than aggregates formed from A␤40 alone. The most toxic aggregates were in samples with the highest levels of aggregation (i.e. A␤40:A␤42 combined at a ratio of 10:1). Thus, it was of interest to examine the macromolecular structure of these aggregates (Fig. 2, D-F).
The approximate diameter of long fiber bundles formed by A␤40 varied from 8 to 16 nm. On the other hand, closely associated fibers of aggregates formed by A␤42 were shorter and had a maximum diameter of ϳ8 nm; thus, we provisionally called these fibers "protofibrils" (Fig. 2E). In samples containing A␤40 and A␤42 at a 10:1 ratio, aggregates formed fiber bundles that appeared shorter than extended A␤40 fibers (compare Fig. 2, F and D). It is possible that, via the rapid formation of protofibrillar nuclei or seeds, A␤42 enhanced ␤-aggregation as well as increased the toxicity of the A␤40:A␤42 mixture. However, our results thus far suggest that A␤42 alone does not cause large elongation. In the next set of experiments, we sought to determine whether certain other factors affect the elongation step of ␤-aggregate formation. As previously suggested, the hydrophobic nature of the A␤42 C terminus is critical for accelerating the seeding process (11)(12)(13). Thus, if the C terminus is related to seeding, we hypothesized that the N terminus of A␤ may be related to elongation.
To test this hypothesis, we assessed the ␤-aggregation kinetics of several A␤40 peptides with pathogenic mutations in the N-terminal domain, Italian variant (E22K) and Dutch variant (E22Q). Previously, such substitutions were suggested to affect the intersheet stacking between ␤-sheets (19 -21). We compared the ␤-aggregation time courses of wild type A␤40 (E22), Italian variant (E22K), Dutch variant (E22Q), and their mixtures (Fig. 3A) (20,21). None of the variants by themselves showed significantly higher levels of ␤-aggregation compared with that of the wild type. Indeed, Italian variant (E22K) showed undetectable ␤-aggregation. However, only when the Italian or Dutch variants were mixed with wild type A␤40 in a 1:1 ratio did ␤-aggregation increase significantly during the early stage of elongation. Although the Dutch variant alone formed ␤-aggregates faster than wild type A␤40, the mixing of the two induced and enhanced elongation. Although ␤-aggregation by the Italian variant and wild type A␤40 mixture was faster and more robust compared with that of the wild type alone, ␤-aggregation by the Italian variant alone was undetectable. Clearly, the interactions between hydrophilic residues in the N-terminal domain play an important role in elongation under our experimental conditions.
As before, we assessed the impact of A␤42 on ␤-aggregation, this time in ratios reported for wild type A␤40 and Italian variant A␤40 in vitro (22) (Fig. 3B). When wild type or variant A␤42 was added to the wild type A␤40/Italian A␤40 sample, an even larger elongation was observed compared with that when A␤42 was absent (Fig. 3C). This combination was very toxic as well (Fig. 3D). The most probable reason to explain the difference between the time courses of aggregation of a mixture containing wild type A␤40 and Italian A␤40 in Fig. 3, A and B would be the difference between two lots of Italian A␤40 peptides, one lot from American Peptides (Fig. 3A) and the other from RIKEN BSI (Fig. 3, B-D and Fig. 4B). Nonetheless, one is tempted to draw conclusions about phenotypes from these in vitro results. The differences in nucleation and elongation of ␤-aggregation resulting from different A␤ combinations do not completely explain differences in the pathological features of these mutations that show altered A␤40:〈␤42 ratios and A␤ sequence mutations. However, because conditions that mimic these pathogenic mutations both induce and accelerate ␤-ag-
Our results thus far clearly hint that particular proportions of A␤40 and A␤42 either enhance or retard nucleation and elongation. Our findings are also consistent with previous stud- ies showing that the accelerating effect of A␤42 resides in the hydrophobicity of its C terminus (11)(12)(13), and the hydrophilic side chains in the N terminus most probably have an effect on intersheet stacking (19). In this last set of experiments, because overall ␤-aggregation depends on the relationship of factors that affect nucleation and elongation, we took a closer look at the respective role(s) of such factors, such as the A␤42 C terminus and A␤40 N terminus, in ␤-aggregation. To determine whether disproportionately greater concentrations of A␤42 affect nucleation, we first assessed the aggregation kinetics of samples containing broader ratios of A␤40:A␤42. As seen in Fig. 2, greater proportions of wild type A␤42 accelerated the seeding of A␤40 (Fig. 4A). In terms of the elongation process, however, the proportion of A␤42 appeared to have an inverse effect. In samples containing larger proportions of A␤42 than A␤40, the overall level of ␤-aggregation was lower (Fig. 4A). On the other hand, although Italian variant A␤40 induced the seeding of wild type A␤40 less effectively compared with that by wild type A␤42, it effectively induced the elongation of aggregates formed by wild type A␤40 (Fig. 4B).
As mentioned, the induction effect of wild A␤42 is related to the hydrophobic nature of its C-terminal, resulting in ␤-sheet formation (11)(12)(13). On the other hand, the effect of Italian variant A␤40 is due to the substitution of a negatively charged Glu at position 22 to a positively charged Lys. This likely induces ␤-aggregation through electrostatic interaction of side chains, thus promoting intersheet stacking (19) and elongation. Because this is an interaction between paired ␤-sheets having oppositely charged side chains, Italian variant A␤40 does not aggregate by itself. This notion was further supported by experiments mixing four kinds of non-pathogenic mutants at position 22 with wild type A␤40 (Fig. 4, D-G). Although the degree of induction effect was slightly different, two mutant A␤40s having positively charged Arg and shortK (see "Experimental Procedures") induced the elongation of wild type A␤40 in a manner similar to the Italian variant (E22K) (Fig. 4, D, E,  and B). To examine the effect of size of side chains at position 22, we also tested two other mutants with negatively charged Asp and longE (see "Experimental Procedures"). Both mutants by themselves showed lower levels of ␤-aggregation than wild type A␤40 (Fig. 4, F and G). Moreover, as the proportion of E22D and E22longE was increased, the ␤-aggregation was decreased in a dose-dependent manner, suggesting the importance of size of side chains in ␤-aggregation. In contrast, although all the mutants with positively charged residue at 22 (E22K, E22R, and E22shortK) by themselves also had lower ␤-aggregation level than wild type, when their small proportions were mixed with wild type, ␤-aggregation was significantly elevated. These results suggest the important role of electrostatic interactions of side chains in the ␤aggregation process.
Substantial induction effects by both wild A␤42 and Italian variant A␤40 on the overall ␤-aggregation of wild A␤40 were observed when the proportion of inducers was smaller than wild A␤40. Because overall ␤-aggregation depends on both ␤-sheet formation (i.e. nucleation) and intersheet stacking (i.e. elongation), specific balance between these interactions determines the formation of fast and induced ␤-aggregation that leads to toxic activity. The importance of such balance is underscored by our finding that Italian variant A␤42 suppressed the aggregation of wild type A␤40 at all ratios tested (Fig. 4C). This is because the hydrophobic effect at the C terminus and electrostatic interactions between side chains are physically opposed. Only when they balance at the certain composition of A␤ peptides are toxic ␤-aggregates formed. DISCUSSION There is tremendous evidence that supports a single hypothesis for the pathogenesis of AD. This so-called "amyloid hypothesis" suggests that the process of A␤ deposition is intimately connected to the initiation of AD pathogenesis and that all other features of the disease are secondary to this initiation (23,24). A recent, novel suggestion relating to this hypothesis is that oligomeric aggregates of A␤ peptides are the forms that generate neurotoxicity (25,26). The aggregation kinetics presented in Fig. 1 show that initially formed aggregates are indeed more toxic than fiber bundles formed after a long incubation. However, one cannot conclude from these observations that deposited amyloids or bundles of fibers are innocuous but suggest that they are possibly less neurotoxic than initially formed A␤-aggregates.
Mixing A␤40 and A␤42 at specific ratios induced and accelerated ␤-aggregation. These results suggest that particular ratios of A␤ species are very important for the toxic ␤-aggregation process and perhaps in AD pathogenesis. However, elevated ratios of A␤42/40 have not been observed in the media collected from cells transfected with several intra-A␤ mutations, including the Dutch and Italian cases (22). The predominant increase of ␤-aggregation observed by mixing both wild type and Italian variant A␤ suggests that elongation is largely dependent on interactions between the N-terminal side chains of ␤-sheets, which are nucleated by hydrophobic interactions of C-terminal regions. Hydrophobic residues in the N-terminal domain may also induce nucleation or elongation by an entropically driven process, whereas other polar or charged side chains will interact more directly with each other. The latter type of interaction is hydrophilic and thus is generally known to be weaker in stabilizing proteins. Because these two types of interactions usually repel each other in a manner similar to oil and water, A␤ compositions that balance these interactions in a specific way determine whether the formation of toxic ␤-aggregates proceeds effectively. This is relevant to the possible mechanism of A␤ toxicity. One idea is that A␤-aggregates penetrate the membrane structure either extracellularly or intracellularly (27). ␤-sheeted A␤-aggregates formed under conditions in which polar-nonpolar interactions are balanced in a specific way would make an alignment of charged residues that perhaps disrupt functions of the cell membrane. This is consistent with a recent finding that non-fibrillar, ThT-positive aggregates (i.e. aggregates of ␤-sheets) of a wide range of misfolded proteins exhibit similar toxicity to that shown in the present study (28).