Analysis of the Secondary Structure of β-Amyloid (Aβ42) Fibrils by Systematic Proline Replacement*

Amyloid fibrils in Alzheimer's disease mainly consist of 40- and 42-mer β-amyloid peptides (Aβ40 and Aβ42) that exhibit aggregative ability and neurotoxicity. Although the aggregates of Aβ peptides are rich in intermolecular β-sheet, the precise secondary structure of Aβ in the aggregates remains unclear. To identify the amino acid residues involved in the β-sheet formation, 34 proline-substituted mutants of Aβ42 were synthesized and their aggregative ability and neurotoxicity on PC12 cells were examined. Prolines are rarely present in β-sheet, whereas they are easily accommodated in β-turn as a Pro-X corner. Among the mutants at positions 15-32, only E22P-Aβ42 extensively aggregated with stronger neurotoxicity than wild-type Aβ42, suggesting that the residues at positions 15-21 and 24-32 are involved in the β-sheet and that the turn at positions 22 and 23 plays a crucial role in the aggregation and neurotoxicity of Aβ42. The C-terminal proline mutants (A42P-, I41P-, and V40P-Aβ42) hardly aggregated with extremely weak cytotoxicity, whereas the C-terminal threonine mutants (A42T- and I41T-Aβ42) aggregated potently with significant cytotoxicity. These results indicate that the hydrophobicity of the C-terminal two residues of Aβ42 is not related to its aggregative ability and neurotoxicity, rather the C-terminal three residues adopt the β-sheet. These results demonstrate well the large difference in aggregative ability and neurotoxicity between Aβ42 and Aβ40. In contrast, the proline mutants at the N-terminal 13 residues showed potent aggregative ability and neurotoxicity similar to those of wild-type Aβ42. The identification of the β-sheet region of Aβ42 is a basis for designing new aggregation inhibitors of Aβ peptides.

mainly consists of 40-and 42-mer peptides (A␤40 and A␤42) generated from amyloid precursor protein by two proteases, ␤and ␥-secretase (2,3). A␤42 plays a pivotal role in the pathogenesis of AD, because the aggregative ability and neurotoxicity of A␤42 are considerably higher than those of A␤40 (4). Because the aggregative ability of A␤ peptides is closely related to the neurotoxicity, precise structural information for amyloid fibrils is indispensable for understanding the molecular mechanisms of AD and related folding diseases and for developing new medicinal leads using the inhibitory activity of amyloid fibril formation.
Previous studies on A␤ fibrils showed that A␤ aggregates mainly consist of intermolecular parallel ␤-sheet (5-10). However, the technical barriers to using x-ray crystallography or solution NMR have hampered its structural determination in high resolution (10). Solid-state NMR spectroscopy is a fairly reliable approach to elucidating the structure of amyloid fibrils. In fact, solid-state NMR analysis on the A␤40 aggregates has been reported recently (7,9). However, there are few reports on the structure of A␤42 aggregates that are more important in AD, possibly because efficient synthesis of A␤42 with 14 hydrophobic and bulky amino acid residues at the C terminus is quite difficult (11). The weak point of solid-state NMR is that it requires a series of 13 Cand/or 15 N-labeled A␤ peptides at different positions in large quantity (20 -30 mg).
Systematic replacement with proline in peptides is a reliable and rapid method for predicting the secondary structure, especially ␤-sheet and turn (12). Prolines are rarely present in ␤-sheet, whereas they are easily accommodated in a variety of turns, for example, as a Pro-X corner (where X is a variable amino acid residue) (13). Quite recently, Williams et al. (14) have investigated a systematic proline replacement of A␤40, showing that the residues 15-21, 24 -28, and 31-36 are likely to include the ␤-sheet portions of the fibril and that the residues at positions 22,23,29, and 30 probably occupy turn positions among these ␤-sheet elements. This conclusion does not contradict our preliminary investigation using proline-substituted A␤42 at positions 19 -26 (15). However, such systematic proline replacement should be carried out with A␤42, the aggregative ability and neurotoxicity of which are especially high (4). Our continuous efforts to synthesize a series of A␤42 derivatives with proline replacement (Fig. 1) led to the proposal of a new structural model of A␤42 aggregates. This is a full report on the aggregative ability and neurotoxicity of a series of proline-substituted A␤42 mutants.
Synthesis of A␤ Derivatives-Each A␤ derivative was synthesized in a stepwise fashion on 0.1 mmol of preloaded Fmoc-Val-PEG-PS (for A␤40 derivatives) or Fmoc-Ala-PEG-PS (for A␤42 derivatives) resin by Pioneer TM using the Fmoc method as reported previously (15, 19 -22). The coupling reaction was carried out using Fmoc amino acid (0.4 mmol), HATU (0.4 mmol), and DIPEA (0.8 mmol) in DMF for 30 min. After each coupling reaction, the N-terminal Fmoc group was deblocked with 20% piperidine in DMF.
After completion of the chain elongation, each peptide resin washed with DMF and CH 2 Cl 2 was treated with a mixture containing trifluoroacetic acid, m-cresol, ethanedithiol, and thioanisole for final deprotection and cleavage from the resin. After 2 h of shaking at room temperature, the crude peptide precipitated by diethyl ether was purified by HPLC under alkaline conditions as reported previously (15, 19 -22). Lyophilization gave a corresponding pure A␤ peptide, the purity of which was confirmed by HPLC (Ͼ98%). Each purified peptide exhibited satisfactory mass spectrometric data. The difference between the calculated and theoretical molecular mass was less than one mass unit.
Sedimentation Assay for Fibril Formation-Each A␤ derivative was dissolved in 0.02% NH 4 OH at 250 M. After a 10-fold dilution by 50 mM sodium phosphate containing 100 mM NaCl at pH 7.4, the resultant peptide solution (25 M) was incubated at 37°C for 4, 8, 16, 24, or 48 h. After centrifugation at 15,000 rpm in an Eppendorf microcentrifuge at 4°C for 10 min, 25 l of the supernatant was then analyzed by HPLC as reported previously (15,21,22). The area of the absorption at 220 nm was integrated and expressed as a percentage of the control. Molar concentration of soluble A␤ peptides present at equilibrium (C r ) was determined by Micro-BCA protein assay (Pierce).
Th-T Fluorescence Assay-Each A␤ derivative was dissolved in 0.02% NH 4 OH at 250 M. The peptide solution (25 M) diluted with the phosphate buffer solution described above (pH 7.4) was incubated at 37°C for 4, 8, 16, 24, 48, or 72 h. Three microliters of each A␤ solution was added to 300 l of 5 M Th-T in 50 mM Gly-NaOH (pH 8.5). Fluorescence intensity was measured at 450-nm excitation and 482-nm emission as reported previously (23).
Cell Culture of PC12 Cells-Rat pheochromocytoma PC12 cells were obtained from Riken Cell Bank and were cultured as reported previously (15,21,22). For experimental purposes, near-confluent cultures of the cells were plated at ϳ10 4 cells/100 l/well fresh culture medium in a 96-well tissue culture plate coated with collagen and incubated at 37°C under 5% CO 2 overnight before the experiments.
MTT Assay Using PC12 Cells-The reduction of MTT by mitochondrial reductase was carried out by protocols based on a previous report (24) with slight modifications. Each solution of A␤ derivatives (0.1% NH 4 OH) sterilized by a filter (0.22 m) was diluted with 0.1% NH 4 OH at concentrations ranging from 0.12 to 120 M. 10 l of the resultant solution and 10 l of 50 mM sodium phosphate (pH 7.4) containing 100 mM NaCl were, respectively, added to the above-mentioned 100-l cell culture, which was incubated at 37°C under 5% CO 2 for 48 h. After removal of 30 l of the medium, 10 l of 5 mg/ml MTT in the phosphate buffer solution described above was added to the cell culture, which was incubated at 37°C under 5% CO 2 for 4 h. After evacuation of the culture medium, the cell lysis buffer (100 l/well; 10% SDS, 0.01 M NH 4 Cl) was subsequently added to the cells and the cell lysate was incubated overnight in the dark at room temperature. The colorimetric determination of MTT was made at 600 nm. The absorbance obtained by the addition of vehicle was taken as 100%.
Transmission Electron Micrographs of Negatively Stained Preparations of the Fibrils Formed by A␤42 Derivatives-The fibril formation of the A␤42 derivatives was detected by electron microscope. Each A␤42 derivative (25 M) was incubated in 50 mM phosphate buffer (pH 7.4) containing 100 mM NaCl for 48 h at 37°C. After centrifugation, the supernatant was removed from the pellets. Aggregates were then suspended in water by gentle vortex mixing. These suspensions were applied to a 400-mesh collodion-coated copper grid (Nissin EM, Tokyo, Japan) and allowed to dry in air before being negatively stained for 2 min with 2% uranyl acetate. Fibrils were examined with the Hitachi H-7500 electron microscope.

Synthesis of Proline-substituted A␤42
Mutants-It is difficult to synthesize A␤42 with 14 hydrophobic and/or bulky amino acid residues at the C terminus in a highly pure form, because it easily aggregates even under weakly acidic and neutral conditions (11). We have recently established a highly efficient method for synthesizing long peptides of over 50 amino acid residues with a continuous flow-type peptide synthesizer (Pioneer TM ) using HATU (18) as an effective coupling reagent for Fmoc chemistry. After final deprotection and cleavage from the resin followed by purification using HPLC in the alkaline condition (CH 3 CN-0.1% NH 4 OH), each A␤ derivative was successfully taken in a highly pure form as reported previously (15, 19 -22). The total yields of the A␤ derivatives synthesized in this study were consequently between 2 and 36%, indicating that the average coupling yield of each condensation step was 95-97.5%. Their molecular weights were confirmed by MALDI-TOF MS, and their purity was determined by HPLC analysis (Ͼ98%).
Aggregative Ability and Neurotoxicity of the Proline-substituted A␤42 Mutants at Positions 15-32-We focused at first on the residue 22 that frequently mutates in cerebral amyloid angiopathy such as E22Q (Dutch) (25) and E22K (Italian) (26) and investigated the proline-substituted A␤42 mutants at positions 15-32. Their aggregative ability was estimated by the two methods, the sedimentation assay (HPLC analysis after centrifugation of the A␤ solution) and the Th-T fluorescence assay. Since Williams et al. (14) used thermodynamic stabilities of the A␤40 fibrils to obtain its structural information, we examined the thermodynamic stabilities of several A␤42 mutants that were reported as a preliminary communication (15). Fig. 2C shows the critical concentration (C r ) defined as the molar concentration of soluble A␤42 peptides present at equilibrium along with the aggregative ability estimated by the sedimentation assay ( Fig. 2A) and the Th-T fluorescence assay (Fig. 2B), which were generally correlated with each other. Fairly good correlation was observed between the kinetics and thermodynamics data. E22P-A␤42 with rapid aggregation kinetics showed a lower C r value (0.42 M) com-pared with wild-type A␤42 (0.85 M). On the other hand, V18P-, F19P-, and F21P-A␤42 with slower aggregation kinetics exhibited higher C r values (15,17, and 17 M, respectively). It is noteworthy that the C r value of wild-type A␤42 (0.85 M) determined by us is almost equal to that of wildtype A␤40 (0.90 M) reported by Williams et al. (14). Since A␤42 aggregated far more rapidly than 〈␤40 as shown in Fig.  2, A and B, aggregation kinetics rather than thermodynamics seems to be more important in A␤42. Thus, we adopted the kinetics data to deduce the secondary structure of A␤42 in its aggregates.
As shown in Fig. 3A, all of the proline-substituted A␤42 mutants at positions 15-32 with the exception of E22P-A␤42 hardly aggregated after 8 h of incubation. Only the data of the sedimentation assay are shown, because the HPLC data are more reliable than the Th-T fluorescence data as reported previously (22). Similar results were obtained after 16 and 24 h of incubation with the exception of V18P-, F19P-, G25P-, and A30P-A␤42 that showed slow but significant aggregation kinetics (data not shown). It is quite noticeable that only E22P-A␤42 aggregated faster than wild-type A␤42.
The neurotoxic effects on the PC12 cells of these A␤42 mutants were examined by MTT assay to confirm whether their aggregation reflects the pathological aggregation of wild-type A␤42. The MTT assay consists of the conversion of MTT to colored formazan by mitochondrial reductase and serves as an indirect measurement of cell proliferation and viability. After a 48-h incubation with each A␤42 derivative, formazan formation was measured at 600 nm in a concentration range of 0.01-10 M. Because wild-type A␤42 inhibited ϳ50% formazan formation at 10 Ϫ6.5 M, the formazan formation in the presence of 10 Ϫ6.5 M each proline-substituted A␤42 peptide was measured simultaneously to estimate precisely the relative neurotoxicity (Fig. 3A). The results showed that only E22P-A␤42 potently inhibited the formazan formation at 10 Ϫ6.5 M. The IC 50 value of inhibition of the formazan formation by E22P-A␤42 was 0.084 Ϯ 0.011 M, whereas that of wild-type A␤42 was 0.97 Ϯ 0.18 M. Although only D23P-A␤42 showed significant neurotoxicity at 10 Ϫ6.5 M (IC 50 ϭ 1.3 Ϯ 0.26 M), other A␤42 mutants were almost inactive at 10 Ϫ6.5 M.
Effects of the C-terminal Residues of A␤42 on Its Aggregative Ability and Neurotoxicity-It is obvious that the C-terminal two residues of A␤42 play a critical role in its aggregative ability and neurotoxicity. Weinreb et al. (27) proposed the "hypothesis of hydrophobic cluster," stating that hydrophobic interaction among the side chains at the C terminus induces aggregation (Fig. 4A). In this hypothesis, Ile-41 is incorporated in the hydrophobic core formed by Leu-34 and Met-35. To confirm the role of the hydrophobic side chains at the C terminus of A␤42, the hydrophilic threonine mutants at positions 41 or 42 (I41T-and A42T-A␤42) were prepared and examined for their aggregative ability and neurotoxicity (Fig. 4B). Both I41T-and A42T-A␤42 aggregated rapidly similar to wild-type A␤42. Substitution with Thr did not abolish their cytotoxic effects. The IC 50 values of the cytotoxicity on PC12 cells were 1.1 Ϯ 0.11, 0.70 Ϯ 0.10, and 0.97 Ϯ 0.18 M for I41T-, A42T-, and wild-type A␤42, respectively, suggesting that hydrophobicity of the side chains at positions 41 and 42 is not requisite for the aggregative ability and neurotoxicity of A␤42.
Because it is conceivable that the C-terminal residues participate in the ␤-sheet formation, a series of the proline-substituted A␤42 mutants at positions 33-42 were synthesized and their aggregative ability and neurotoxicity were tested. As shown in Fig. 3B, the aggregative ability of L34P-and G38P-A␤42 as estimated by the sedimentation assay was significantly higher than that of wild-type A␤42. Both aggregated FIG. 2. Aggregation kinetics was estimated by the sedimentation assay (A), which was estimated by the Th-T fluorescence assay (B) and the molar concentration of soluble peptide present at the equilibrium (C r ) of several proline-substituted A␤42 mutants (C). •, wild-type A␤42; OE, wild-type A␤40; ࡗ, E22P-A␤42; Ⅺ, V18P-A␤42; ‚, F19P-A␤42; and E, A21P-A␤42. Because the Th-T fluorescence of wild-type A␤40 did not reach a plateau value even after a 72-h incubation in our assay condition, the C r value of A␤40 was not determined. almost completely after a 4-h incubation (only the data after an 8-h incubation is shown in Fig. 3B). G33P-and V39P-A␤42 also showed significant aggregative ability. Although G38Pand V39P-A␤42 showed potent neurotoxicity comparable with wild-type A␤42, the neurotoxicity of G33P-and L34P-A␤42 was weak regardless of its significant aggregative ability (Fig. 3B). M35P-, V36P-, G37P-, V40P-, I41P-, and A42P-A␤42 did not aggregate, even after a 24-h incubation (data not shown). They were almost inactive in the neurotoxicity assay (Fig. 3B).

FIG. 3. Aggregation kinetics was estimated by the sedimentation assay after an 8-h incubation and neurotoxicity in PC12 cells estimated by the MTT assay of the proline-substituted A␤42 mutants at positions 15-32 (A), 33-42 (B), and 3-13 (C).
In the aggregation assay, similar results were obtained after a 16-or 24-h incubation except in V18P-, F19P-, G25P-, A30P-, and G33P-A␤42, which showed moderate aggregation after a 24-h incubation (data not shown). In the neurotoxicity assay, the concentration of the proline-substituted A␤ mutants was 10 Ϫ6.5 M, which is close to the IC 50 value of wild-type (WT) A␤42. All of the peptides in each group were tested simultaneously to compare relative cytotoxicity.

FIG. 4. Role of the C-terminal two residues.
A, the hypothesis of hydrophobic cluster in A␤ fibril formation (27). B, aggregation kinetics estimated by the sedimentation assay after an 8-h incubation and neurotoxicity in PC12 cells estimated by the MTT assay of I41T-and A42T-A␤42. In the neurotoxicity assay, the concentration of the proline-substituted A␤ mutants was 10 Ϫ6.5 M, which is close to the IC 50 value of wild-type A␤42. These peptides were tested simultaneously to compare relative cytotoxicity.

Effects of the N-terminal Residues of A␤42 on Its Aggregative
Ability and Neurotoxicity-Because the C-terminal three residues play a critical role in the aggregative ability and neurotoxicity of A␤42, the contribution of the N-terminal residues to the activities was also investigated. A␤3-42 lacking the Nterminal two residues of A␤42 aggregated potently with velocity quite similar to that of wild-type A␤42 (data not shown) and showed significant neurotoxic effects. The IC 50 values of A␤3-42 and wild-type A␤42 were 0.043 Ϯ 0.010 and 0.97 Ϯ 0.18 M, respectively. This indicates that the N-terminal two residues are not necessary for the aggregative ability and neurotoxicity of A␤42. To identify the ␤-sheet-forming region at the N-terminal portion, proline-substituted A␤42 mutants at positions 3-13 were synthesized and examined for their aggregative ability and neurotoxicity (Fig. 3C). All of the mutants at positions 3-13 aggregated potently and exhibited significant neurotoxic effects.
Transmission Electron Micrographs of Negatively Stained Preparations of the Fibrils Formed from the Proline-substituted A␤42 Mutants-Fibril formation of the proline-substituted A␤42 mutants with potent aggregative ability was evaluated by transmission electron microscopy after a 48-h incubation at 37°C. N-terminal proline-substituted A␤42 mutants at positions 3-13 and E22P-, L34P-, G38P-, V39P-, I41T-, and A42T-A␤42 exhibited typical fibril formation, several examples of which are shown in Fig. 5. The fibrils of E22P-A␤42 have been reported previously (22). The morphologies of these fibrils resembled each other well. DISCUSSION The aggregation of A␤ peptides is significantly related to the pathogenesis of neuronal degeneration in AD. Despite many previous studies on the structural analysis of A␤ aggregates, the precise mechanism has not yet been clarified. To obtain information on the structure of A␤42 fibrils, we adopted the proline-scanning method proposed by Wood et al. (12). Thirtyfour proline-substituted A␤42 mutants were synthesized in high purity using the method recently established by us (19 -22) and were subjected to measurements of their aggregative ability and neurotoxicity.
The aggregative ability of these mutants was estimated by the sedimentation and the Th-T fluorescence assay, which are not always correlated with each other (22), because the Th-T fluorescence can vary depending on the structure and morphology of the fibrils. However, these two assays were generally in good correlation. Only the data from the sedimentation assay are shown in Fig. 3. Among the proline-substituted A␤42 mutants at positions 15-32, only E22P-A␤42 aggregated more rapidly than wild-type A␤42, whereas other proline-substituted mutants at positions 15-21 and 24 -32 hardly aggregated after an 8-h incubation. Because proline has a propensity to form a ␤-turn structure as a Pro-X corner (13), the data strongly suggest that turn at positions 22 and 23 is a critical secondary structure in the A␤42 fibrils. This finding demonstrates very well the high aggregative ability of the A␤42 mutants in cerebral amyloid angiopathy (21,22), E22Q-A␤42 (Dutch), and E22K-A␤42 (Italian), because Gln-Asp (Dutch) and Lys-Asp (Italian) sequences at positions 22 and 23 are more frequently found in the two-residue ␤-turn (13) than in Glu-Asp (wild type). Recent investigations using solid-state NMR (7-9) have indicated a parallel organization of ␤-sheets in A␤40 fibrils, because the ␤-carbons of Ala-21 and Ala-30 are located within 5.5 Å, respectively. Thus, the present results obtained from the proline mutagenesis indicate that turn formation at positions 22 and 23 followed by intermolecular parallel ␤-sheet formation between positions 15-21 and 24 -32, respectively, leads to the organization of A␤42 fibrils (Fig. 6A).
Specifically, the turn position of our A␤42 aggregation model was different from that of Petkova et al. (7), which is based on the solid-phase NMR of A␤40 aggregates in which two residues at positions 26 and 27 adopted a bend structure. This structure seems to be reasonable, because the Ser-Asn sequence is often found in the two-residue ␤-turn (13). However, the aggregative ability of S26P-A␤40 was very low (data not shown), similar to that of wild-type A␤40, whereas E22P-A␤40 aggregated more rapidly than wild-type A␤40 (15), suggesting that the structure of the A␤40 fibrils resembles that of the A␤42 fibrils. After the completion of this study, Williams et al. (14) have reported the systematic proline replacement of A␤40 and have shown that residues at positions 22 and 23 of A␤40 probably occupy turn positions, supporting our conclusion. The turn at position 22 of A␤ fibrils does not contradict the solid-phase NMR data, because no NMR data at position 22 of A␤40 were described in the study of Petkova et al. (7). Moreover, the chemical shifts for Asn-23 predicted non-␤-strand ⌽ and ⌿ angles.
Turn formation at position 22 of A␤ peptides is closely related to the cytotoxic effects on PC12 cells. In the model proposed by Petkova et al. (7), the turn at positions 26 and 27 is stabilized by an ionic interaction between the side chains of Asp-23 and Lys-28. This conformation seems to be non-malignant, because S26P-A␤42 and S26P-A␤40 did not show any cytotoxicity against PC12 cells (data not shown). Mutation at position 22 of A␤ peptides by the amino acid residues that prefer turn formation would change the turn position from 26 and 27 to positions 22 and 23. This conformational change might increase the intermolecular parallel ␤-sheet region to enhance the aggregative ability and neurotoxicity of A␤ peptides.
It is widely accepted that A␤42 aggregates far more potently and is cytotoxic to PC12 cells compared with A␤40. However, there has been no persuasive explanation regarding this issue until now. Hydrophobicity of the C-terminal two residues was considered to be a critical factor for the high aggregative ability of A␤42, and the "hypothesis of hydrophobic cluster" was proposed as mentioned above (Fig. 4A) (27). However, our present results using I41T-and A42T-A␤42 did not support this hypothesis because substitution at positions 41 and 42 of A␤42 with the hydrophilic threonine residue did not decrease aggregative ability and neurotoxicity.
Some investigators have considered that the C-terminal hydrophobic amino acid residues are involved in the ␤-sheet formation (9). We also considered that the C-terminal residues adopt a ␤-sheet structure and examined the aggregative ability and neurotoxicity of the proline-substituted A␤42 mutants at positions 33-42 (Fig. 3B). V40P-, I41P-, and A42P-A␤42 hardly aggregated and were almost inactive in the assay using PC12 cells, indicating that the C-terminal three residues significantly participate in ␤-sheet formation. It is not known whether the ␤-sheet at positions 40 -42 is intermolecular or intramolecular. However, this is a solid conclusion for explaining the potent aggregative ability and neurotoxicity of A␤42. According to the proline mutagenesis of A␤40 (14), the Cterminal residues at positions 37-40 of A␤40 were judged to be excluded from the ␤-sheet structure in the A␤40 fibril because these proline-substituted A␤40 mutants aggregated with a potency similar to that of wild-type A␤40. Previous spin-labeling studies using A␤40 (28) have also suggested that the C terminus of A␤40 is not packed in a rigid structure within the fibril. These data clearly indicate that the roles of the C terminus of A␤42 and that of A␤40 are quite different from each other. Tycko and co-workers (7,9) have recently proposed a structural model for A␤40 protofilaments consisting of the two cross-␤ units attached by the hydrophobic interaction between the side chains of the C-terminal amino acid residues. However, our present data suggested that the interaction at the C terminus between the two cross-␤ units is not a hydrophobic interaction but a parallel or anti-parallel ␤-sheet. This ␤-sheet formation would increase the rate of fibril formation of A␤42.
It was an unexpected result that L34P-and G38P-A␤42 aggregated faster than the wild-type A␤42. To examine the thermodynamic stabilities of the fibrils formed by these A␤42 mutants, the molar concentration of soluble peptide present at equilibrium (C r ) was measured. The C r values of L34P-and G38P-A␤42 were 1.0 Ϯ 0.010 and 1.4 Ϯ 0.087 M, respectively, which were almost equal to those of wild-type A␤42 (0.85 M). Moreover, G33P-and V39P-A␤42 also showed relatively high aggregation velocity. These results suggested the turn formation at positions 33 and 34 as well as 38 and 39 and the ␤-sheet formation at positions 35-37. To confirm these turn positions, the triple A␤42 mutant (P3-A␤42) substituted with proline residues at the three possible turn positions (22,34, and 38) was prepared. P3-A␤42 aggregated in the velocity equal to that of wild-type A␤42 with a C r value of 1.0 Ϯ 0.022 M, supporting the three turns in the A␤42 aggregates.
It is noteworthy that the cytotoxicity to PC12 cells of G33Pand L34P-A␤42 was significantly weaker than that of wild-type A␤42, unlike that of G38P-and V39P-A␤42. We speculated that the residue at positions 33 and 34 plays a critical role in cytotoxicity rather than aggregation. The critical amino acid residue for expressing cytotoxicity of A␤ peptides is considered to be Met-35. According to Butterfield and colleagues (29), radical formation at the sulfur atom of Met-35 is a requisite condition for A␤42 to damage cells. The flexibility of this region would make A␤42 a more cytotoxic species by radical transfer from Met-35 to Gly-33 as proposed.
In contrast, the N-terminal two residues were not necessary for the aggregative ability and neurotoxicity of A␤42. Moreover, all of the proline-substituted mutants at positions 3-13 showed potent aggregative ability and neurotoxicity comparable with those of wild-type A␤42 (Fig. 3C). This finding suggests that the N-terminal 13 residues do not adopt any solid structure such as ␤-sheet and ␣-helix. This conclusion is in good agreement with those previously reported using solidstate NMR analysis (7) and electron spin resonance spectra (28) in which the secondary structure of the N-terminal 10 -15 residues of A␤40 is not defined.
The electron microscope measurements of the proline-substituted A␤42 mutants with potent aggregative ability clearly showed fibrillar materials (Fig. 5). Although the resolution at this level cannot distinguish fine structural differences, it is certain that the aggregates in Fig. 5 are not amorphous but are fibrils. The fibrils of the proline-substituted A␤42 mutants were quite similar to those of wild-type A␤42. This was also supported by the Th-T data. These mutants showed significant Th-T fluorescence, characteristic of amyloid fibrils (data not shown). These data indicated that the fibrils of the prolinesubstituted A␤42 mutants adopt a tertiary structure similar to that of wild-type A␤42.
Finally, we proposed a new aggregation model of A␤42 on the basis of the systematic proline replacement along with a requirement for the parallel ␤-sheet at positions 21 and 30 (5), as shown in Fig. 6A. The most important structural feature of A␤42 fibrils is the turn at positions 22 and 23 and the two intermolecular ␤-sheets on both sides (positions 15-21 and 24 -32). This structure resembles several models recently pro-posed (7,10,28) with the exception of the turn position. As mentioned above, the turn formation at position 22 of A␤42 can explain most reasonably the pathogenesis of cerebral amyloid angiopathy (Dutch and Italian mutation). The proline replacement of A␤40 gave a similar conclusion regarding the turn position at position 22 (Fig. 6B) (14). However, the C-terminal structure in the A␤40 aggregation model is quite different from that of A␤42. Our proline mutagenesis data indicated that the C-terminal three residues adopt a ␤-sheet structure. Although it is not clear whether this ␤-sheet is intermolecular or intramolecular, we believe that this is an intermolecular ␤-sheet for binding each A␤42 unit tightly to form fibrils because the N-terminal 13 residues are not involved in the ␤-sheet formation. The lack of the C-terminal intermolecular ␤-sheet of A␤40 shows its fairly slow aggregative kinetics compared with that of A␤42. The existence of turns at positions 33, 34, 38, and 39 seems to be necessary to stabilize the A␤42 by the formation of a circular structure. The aggregation model proposed in this study (Fig. 6A) gives unique opportunities to design reasonably novel inhibitors for A␤42 aggregation.