Small Liposomes Accelerate the Fibrillation of Amyloid β (1–40)*

Background: Aggregation and fibrillation of amyloid β peptides (Aβ) in lipid bilayers lead to membrane disruption. Results: Small vesicles accelerated Aβ fibrillation, whereas large vesicles promoted amorphous aggregation. Conclusion: Moderate membrane-curvature-dependent interaction is an important factor for Aβ aggregation. Significance: Two types of membrane binding, one leading to amyloid nucleation and the other to non-productive binding, may be common to various amyloidogenic proteins. The deposition of amyloid β (Aβ) peptides is a pathological hallmark of Alzheimer disease. Aβ peptides were previously considered to interact specifically with ganglioside-containing membranes. Several studies have suggested that Aβ peptides also bind to phosphatidylcholine membranes, which lead to deformation of membranes and fibrillation of Aβ. Moreover, the role of membrane curvature, one type of deformation produced by binding of proteins to a membrane, in the binding and fibrillation of Aβ remains unclear. To clearly understand the relationship between the binding, consequent membrane deformation, and fibrillation of Aβ, we examined the amyloid fibrillation of Aβ-(1–40) in the presence of liposomes of various sizes. Membrane curvature increased with a decrease in the size of the liposomes. We used liposomes made of 1,2-dioleoyl-sn-glycero-3-phosphocholine to eliminate electrostatic effects. The results obtained showed that liposomes of smaller sizes (≤50 nm) significantly accelerated the nucleation step, thereby shortening the lag time of fibrillation. On the other hand, liposomes of larger sizes decreased the amount of fibrils but did not notably affect the lag time. The morphologies of fibrils, which were monitored by total internal reflection fluorescence microscopy, atomic force microscopy, and transmission electron microscopy, revealed that the length of Aβ-(1–40) fibrils became shorter and the amount of amorphous aggregates became larger as liposomes increased in size. These results suggest that the curvature of membranes coupled with an increase in water-accessible hydrophobic regions is important for binding and concentrating Aβ monomers, leading to amyloid nucleation. Furthermore, amyloid fibrillation on membranes may compete with non-productive binding to produce amorphous aggregates.

membranes are known to be involved in the fibrillation of ␣-synuclein and human islet amyloid polypeptide, which are associated with Parkinson disease and type 2 diabetes, respectively (24 -27). Various kinds of lipids that affect amyloid fibrillation have also been studied, and the findings obtained revealed that electrostatic interactions between proteins and membranes facilitated the binding of proteins to the membranes as well as their fibrillation (25,28,29). The size or radius of liposomes may also be an important factor for determining these interactions; ␣-synuclein binds to small unilamellar vesicles, and islet amyloid polypeptide interacts with high curvature membranes (30 -32).
However, the role of membrane properties on the fibrillation of A␤ peptides has not yet been elucidated in detail. Previous studies using giant unilamellar vesicles of 1,2-dioleoyl-snglycero-3-phosphocholine (DOPC) suggested that A␤ peptide binding caused membrane deformation (33,34). A simulation study revealed that the accumulation of A␤ peptides deformed membranes to produce highly curved membranes (35).
Thus, the deformation of membranes may be one of the key factors that accelerate fibrillation. Moreover, the presence of A␤ peptides in the synapses in vivo caused a dysfunction in synaptic vesicles with a diameter of ϳ40 nm (36). Because the binding of A␤ peptides to a membrane and the subsequent deformation are a linked phenomenon, the presence of deformed membranes may accelerate the binding and fibrillation of A␤. Membrane curvature, one possible type of membrane deformation, can be controlled by changing the radius of liposomes; therefore, studies are needed to verify the relationship between membrane deformation, binding, and fibrillation.
We herein examined the effects of liposomes of various diameters on the amyloid fibrillation of A␤-  to determine whether the fibrillation depends on the membrane curvature and consequent increase in water-accessible hydrophobic regions. We chose DOPC and 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) as the liposome components because A␤ binding was reported to bend phosphocholine membranes (35), and phosphocholine, a zwitterionic lipid, allowed us to focus on the effects of membrane curvature while minimizing the effects of electrostatic interactions (16,37). We previously observed that although A␤-  is relatively easy to keep in monomeric states, A␤-(1-42) forms nuclei for spontaneous fibrillation much faster than A␤-(1-40) (38). Consequent explosive fibrillation of A␤-  resulted in a large number of short fibrils, minimizing the interaction with the surface. Thus, to address the effects of membranes, we chose A␤- , although it will be important to compare A␤-  and A␤-  to obtain further insight into the interactions of A␤ with membranes.
The results of the present study revealed that membrane curvature is an important factor for determining the amyloid fibrillation of A␤- . Liposomes of larger sizes decreased the length of amyloid fibrils, and this was accompanied by an increase in the amount of amorphous aggregates. These results suggest that amyloid fibrillation on membranes competes with amorphous aggregation, thereby providing an insight into the mechanism underlying synaptic vesicle impairments.
Preparation of Liposomes-Liposomes were prepared as previously described (39). Briefly, lipids (ϳ10 mM) were stored in methanol stock solutions. Lipid films were prepared by drying appropriate amounts of the stock solution in a stream of dry nitrogen followed by desiccation for at least 1 h to ensure removal of the organic solvent. Appropriate buffer solutions were added and vortexed to rehydrate the lipid films. After 10 freeze-thaw cycles, lipid suspensions were extruded 20 times through 50-, 100-, or 200-nm polycarbonate membranes (Whatman, Clifton, NJ) using a miniextruder (Avanti Polar Lipids, Inc., Alabaster, AL) just before use. Liposomes with a diameter of 30 nm were prepared by an ethanol injection method with slight modifications (40). Appropriate lipids (ϳ100 mM) in methanol were diluted to 10% (v/v). After vortexing, the lipid suspension in the tubes were set on a water bath-type ultrasonic transmitter with a temperature controller (ELESTEIN SP070-PG-M, Elekon Sci. Inc., Chiba, Japan) that applied repetitive ultrasonication pulses to samples from three directions (41). The temperature was set to 37°C, and ultrasonic pulses were applied with repeats of 1 min of ultrasonication and 9 min of quiescence for 3 h. After ultrasonication, this suspension was centrifuged to precipitate multilamellar liposomes or/and aggregations of lipids followed by the collection of the supernatant. As this supernatant may contain methanol, it was dialyzed against an appropriate buffer twice. After dialysis, the concentration of lipids was quantified. Liposome sizes were measured by dynamic light scattering using DynaPro Titan (Wyatt Technology Co., Goleta, CA). Liposomes containing the fluorescence lipid (1,2-dioleoyl-sn-glycero-3-phosphoethanolamine-N-(lissamine rhodamine B sulfonyl (Rho-DOPE)) (0.1% mol) or DOPS were prepared using the same procedure as described above.
Analytical Ultracentrifugation-The sedimentation velocity data were obtained by a Beckman-Coulter Optima XL-A analytical ultracentrifuge (Beckman Coulter, Miami, FL). The samples were first centrifuged at 3000 rpm (700 g) for 5 min to stabilize the absorbance and temperature. Next, centrifugation was performed at 60,000 rpm (250,000 ϫ g) and 4°C. Absorbance data at 220 nm were collected at an interval of 20 min at the radial increment of 0.003 cm in the continuous scanning mode. The absorbance data were analyzed by using sedfit to obtain the sedimentation coefficients. The parameters of solvent conditions (partial specific volume, density and viscosity) were calculated by using sednterp.

Total Internal Reflection Fluorescence Microscopy (TIRFM), Atomic Force Microscopy (AFM), and Transmission Electron
Microscopy (TEM)-The prism-type TIRFM system was used to observe individual amyloid fibrils. The details of this system were described previously (42,43). Briefly, an aliquot (14 l) of sample solution was deposited on a quartz glass slide, and a fibril image was obtained with TIRFM. We used an argon laser (532 nm) and a helium-cadmium laser (442 nm) to excite ThT and rhodamine, respectively. AFM images were obtained using a Digital Instruments Nanoscope IIIa scanning microscope (Veeco Instruments Inc., Plainview, NY) as reported previously (44). TEM images were obtained using a HITACHI H-7650 transmission microscope (Hitachi, Tokyo, Japan) at 20°C with a voltage of 80 kV and magnification of 30,000. The sample solution (5 l) was spotted onto a collodion-coated copper grid (Nisshin EM Co., Tokyo, Japan). After 1 min, the remaining solution was removed with filter paper, and 5 l 2% (w/w) uranyl acetate was spotted onto collodion-coated copper grids. After 1 min, the remaining solution was removed in the same manner.
Amyloid Fibrillation and ThT Fluorescence Assay-The formation of A␤-(1-40) fibrils was examined at 10 M in 10 mM sodium phosphate buffer (pH 7.5) containing 5 M ThT, 100 mM NaCl, and DOPC or POPC liposomes at various concentrations. These solutions (200 l) were added to the 96-well microplates (Greiner-Bio-One, Tokyo, Japan), and a seal was affixed (PowerSeal CRISTAl VIEW, Greiner-Bio-One). The plates were set on a microplate reader (SH-9000, Corona Electric Co., Ibaraki, Japan), and amyloid fibrillation was monitored by ThT fluorescence. It is noted that, although the morphology of A␤ fibrils is likely to depend on agitation of solution (45), we used the fibrillation conditions without agitation. The fluorescence intensity of ThT at 490 nm was measured with an excitation wavelength of 450 nm at 37°C. The lag time of fibrillation was defined as the time at which ThT fluorescence reached 10% of the maximum.
ANS Assay to Evaluate Hydrophobicity-ANS fluorescence was measured using the Hitachi fluorescence spectrophotometer F4500 with the excitation wavelengths at 350 nm. The fluorescence of ANS was monitored by adding 2 M ANS to the solution. The concentrations of liposomes and amyloid fibrils were 2 M.
Isothermal Titration Calorimetry-Isothermal titration calorimetry (ITC) of A␤-(1-40) monomers to the two different sizes of DOPC liposomes (30 nm and 100 nm) in 10 mM phosphate buffer (pH 7.5) containing 100 mM NaCl was performed by using a VP-ITC instrument (GE Healthcare) at 37°C. The concentrations of A␤-  in the syringe and liposomes in the cell were 100 M and 5 mM, respectively. The reference power for the base line was set to 10 cal/s, and the initial delay was 1800 s. Titration of A␤-(1-40) was comprised of 23 injections with the spacing time of 900 s or 300 s and the filter period of 2 s. The injection volume was 12 l for each, and the stirring speed was 242 rpm.
The heat of dilution (and mixing) of A␤-(1-40) titrated to the same buffer was also measured. The apparent values of the change in enthalpy ( app K a ) and the association constant ( app K a ) were obtained by the fit of the binding isotherms to the one-site binding model incorporated in Origin 7.1. By using the relationships app ⌬G bind ϭ ϪRTln( app K a ) ϭ app K a Ϫ T app ⌬S bind , the apparent thermodynamic values of the change in Gibbs free energy ( app ⌬G bind ) and entropy ( app ⌬S bind ) were further calculated. The apparent dissociation constant ( app K d ) was obtained based on the relation of app K a ϭ 1/ app K a . It should be noted that app K a values are more susceptible to the fit than app K a values.

RESULTS
Preparation of Liposomes and A␤ Solution-To investigate the effects of the liposome size on the fibrillation of A␤-(1-40), we prepared liposomes of four sizes (ϳ200, 100, 50, and 30 nm) and analyzed the kinetics of fibrillation monitored by ThT fluorescence. We first confirmed the size of DOPC liposomes with TEM ( Fig. 1, A-D) and dynamic light scattering (Fig. 1, E-H).
The results obtained indicated that we could prepare liposome samples of distinct sizes with relatively small distributions. We call the respective preparations 30-, 50-, 100-, and 200-nm liposomes.
To check whether A␤-(1-40) samples before the fibrillation experiments were monomeric, we conducted a sedimentation velocity measurement (Fig. 1, I and J). The sedimentation coefficient distribution for A␤-(1-40) solution revealed a sharp peak at ϳ0.3 S. Using this value, we calculated the molecular weight to be ϳ4700, consistent with that of A␤-(1-40) monomer, 4300.
The absence of oligomeric aggregates was further confirmed by native-PAGE followed by Western blotting with antibodies to A␤-(1-40) monomers (data not shown). In addition, we observed the kinetics of A␤-(1-40) fibrillation with samples with and without ultracentrifugation to remove possible aggregates, where we also checked the effects of 30-and 100-nm liposomes. We did not observe significant differences between A␤-(1-40) samples with and without ultracentrifugation (data not shown). Therefore, we used A␤-(1-40) preparations without ultracentrifugation assuming that we can address fibrillation of A␤-(1-40) monomers.
Dependence of Fibrillation on the Liposome Concentration-We monitored the fibrillation of 10 M A␤-(1-40) in 10 mM phosphate buffer (pH 7.5) containing 100 mM NaCl and 5 M ThT at 37°C using amyloid-specific ThT fluorescence (46). It has been reported that A␤-(1-40) fibrils with 2-fold symmetry with twisted morphology predominate when fibrils are grown in vitro with agitation of the A␤-(1-40) solution, whereas fibrils with 3-fold symmetry with straight ribbon morphology predominate when fibrils are grown without agitation (45). Throughout the paper, we prepared the fibrils without agitation because agitation might affect the morphologies of liposomes.
Spontaneous fibrillation without agitation occurred with a lag time of 12 h ( Fig. 2A). The straight and long morphology of mature fibrils was confirmed by TIRFM, AFM, and TEM (see below). These results were consistent with the findings of our previous studies (38,47).
To examine the effects of liposomes on the fibrillation of 10 M A␤-(1-40), we added 30-or 100-nm liposomes at 10, 50, 100, 250, 500, 750, and 1000 M (Fig. 2, A and D). We observed a significant acceleration in fibrillation in the presence of 30-nm liposomes at 10 M as revealed by a decrease in the lag  time from ϳ12 h to ϳ7.6 h ( Fig. 2A, inset): the relative lag time in 30-nm liposomes at 10 M was 0.6 ( Fig. 2B). Further increases in the liposome concentration slightly decreased the lag time. We compared the amount of fibrils by the ThT intensities at the end of the measurements (i.e. final ThT intensities at about 40 h). When the ThT intensity decreased after the maximum, we used the maximal ThT intensity. The final ThT intensity was similar to that in the absence of liposomes at all the liposome concentrations examined (Fig. 2C). In contrast, the presence of 100-nm liposomes at various concentrations slightly decreased the lag time (Fig. 2, D, inset, and E). The final ThT intensity in the presence of 100-nm liposomes was independent of the liposome concentration within an acceptable level of error (Fig. 2F). These results suggested that 10 M liposomes were sufficient to examine the effects of liposomes on the fibrillation of 10 M A␤- . The binding equilibrium between A␤-(1-40) and liposomes that affected amyloid fibrillation may have been established even at low concentrations of liposomes. Thus, we used 10 M DOPC liposomes to examine the effects of liposomes of various sizes on the fibrillation of A␤- .
Dependence of Fibrillation on the Liposome Size-The kinetics of A␤-(1-40) fibrillation was examined in the presence of 10 M DOPC liposomes of various diameters (i.e. 30, 50, 100, and 200 nm) (Fig. 3A). The size of DOPC liposomes notably affected both the lag time and final ThT intensity (Fig. 3, C and D). The lag time was the shortest for the 30-nm liposomes and increased with the size of liposomes, returning to the value in the absence of liposomes at 100-or 200-nm liposomes (Fig. 3, A, inset, and C). On the other hand, the maximal ThT intensities observed in the presence of 30-or 50-nm liposomes were similar to that in the absence of liposomes and were larger than those in the presence of 100-or 200-nm liposomes (Fig. 3D). The slope of the elongation process after the lag time was slightly less steep in the presence of 100-or 200-nm DOPC liposomes (Fig. 3A). These results suggested that small liposomes accelerated fibrillation without changing the final amount of fibrils, whereas the larger liposomes decreased the amount of fibrils without an apparent decrease in the fibrillation rate.
We performed the same experiments with POPC liposomes at 10 M (Fig. 3B). Both the lag time and final ThT intensity revealed similar dependences on the size of liposomes to those observed for DOPC liposomes (Fig. 3, B, inset, C, and D), although the final ThT intensities of POPC liposomes were smaller than those of DOPC liposomes. The smaller final ThT intensities of POPC liposomes might be caused by the difference of the membrane surface properties due to the difference of acyl chains or of the temperature of gel-liquid crystalline phase transition; transition temperatures are Ϫ17 and Ϫ2°C

Membrane-dependent Fibrillation of Amyloid ␤
for DOPC and POPC membranes, respectively. Although there were some differences between DOPC and POPC, we could observe a similar effect of liposome size on the kinetics of A␤-(1-40) fibrillation. These results suggested that the sizedependent effects of liposomes may be common to various liposomes. We then attempted to address why 30-and 50-nm liposomes significantly promoted the nucleation reaction, whereas larger liposomes reduced the maximal ThT value without notable effects on the lag time.
Hydrophobic Surface of Membranes Exposed to a Solvent-To clarify the mechanism underlying the promotion of nucleation by small liposomes, we assumed that A␤-(1-40) interacted preferentially with the parts of membranes where water molecules could penetrate. The penetration depth of water representing water-accessible hydrophobic regions may depend on the size of the liposomes and may be larger (deeper) for small size liposomes. Smaller liposomes (Յ50 nm) may have more of these regions than larger liposomes. A previous study reported that the insertion of the C terminus of A␤ peptides that contained a series of hydrophobic amino acids into membranes was crucial to the structural conversion of monomeric A␤ peptides to amyloid fibrils (48). Moreover, the interface between the head groups and acyl chains of lipids was shown to be responsible for the binding and conformational conversion of A␤ peptides (49,50). Hence, a more detailed characterization of the relationship between the kinetics of A␤ fibrillation and hydrophobicity of liposomes will be important.
To examine this hypothesis, we assessed the water-accessible hydrophobic surface of liposomes by measuring ANS fluorescence at 2 M in the presence of various sizes of DOPC lipo-somes at 2 M (Fig. 4A). ANS is a hydrophobic dye, the fluorescence of which increases markedly in a hydrophobic environment, and is often used to detect water-accessible hydrophobic regions in the intermediates of protein folding (51). Using this assay, we speculated that we may be able to detect the exposure of the hydrophobic groups (i.e. acyl chains) of membranes.
The fluorescence intensity of ANS was low at a maximum of 520 nm in the absence of liposomes (Fig. 4A). The fluorescence of ANS increased after the addition of 30-nm liposomes at 2 M, with a blue shift observed in the maximum to 500 nm that suggested the partial burial of the dye in a hydrophobic environment. The increase observed in ANS fluorescence and the blue shift was less for larger liposomes. (Fig. 4, A and B). These results indicated that the water-accessible hydrophobic regions (or solvent penetration) depended on the size of liposomes; smaller liposomes exposed more hydrophobic regions to the solvent than larger liposomes.
To address the dependence of ANS fluorescence on the liposome size, we calculated the distance between two head groups of lipids, d. As shown in Fig. 4C, we assumed a simplified model in which the inner leaflet was ignored, and the lipid of the outer leaflet was represented by a rectangle. According to this model, the inside circumference was (R-2l), and the number of rectangles was represented by n ϭ (R-2l)/A 0.5 , where R, l, and A are the diameter of a liposome, the height of a lipid, and occupied area for each lipid molecule, respectively. Consequently, d was calculated by d ϭ R/n Ϫ A 0.5 . The estimated value of d was plotted against R (Fig. 4C), which showed that d was inversely proportional to R. The dependence of d on the diameter of liposomes was consistent with that of the ANS fluores- cence of DOPC liposomes with a linear relationship between the ANS fluorescence and d, thereby arguing the critical role of membrane curvature (Fig. 4D). Importantly, the lag time of amyloid fibrillation showed a linear relationship with d, which indicated that membrane curvature and consequently wateraccessible hydrophobic surfaces were critical for determining the lag time (Fig. 4E).
On the other hand, the fluorescence of ANS increased in the presence of 2 M A␤-(1-40) fibrils, with a slight blue shift observed in the maximum to 510 nm (Fig. 4A). Thus, the fluorescence of ANS may also be useful for monitoring conformational changes in A␤-(1-40) as well as those of lipids. A further blue shift in the fluorescence of ANS to 480 nm occurred after the addition of 30-nm liposomes at 2 M, and this was accompanied by an increase in the intensity of fluorescence. The subtracted fluorescence spectrum suggested that the binding of A␤-(1-40) fibrils to membranes led to the additional binding of ANS (or a more hydrophobic environment on the prebound ANS molecules). The extent of the blue shift and increase in fluorescence intensity were less after the addition of 100-nm liposomes at 2 M, suggesting that additional ANS binding may be less for larger liposomes.
Fibril Morphology-We examined the morphologies of fibrils prepared in the presence of DOPC liposomes with various diameters using TIRFM, AFM, and TEM (Fig. 5). We used liposomes in which rhodamine-labeled lipids were included for TIRFM so that we could also monitor the location of the liposomes. Liposomes were deformed by interaction with surfaces; therefore, the exact morphology of liposomes could not be detected by our measurements. The three methodologies covered a wide range of dimensions from nm to m.
In the absence of liposomes, A␤-(1-40) fibrils were straight with a length up to several M. They were well dispersed without clustering. Fibrils were still long and dispersed in the presence of 30-or 50-nm liposomes, although some clusters and fibril clumps were observed. In the TIRFM images, the locations of liposomes revealed by rhodamine fluorescence always overlapped with amyloid fibrils, which indicated that the liposomes were the sites of amyloid nucleation. The sizes of liposomes were markedly smaller than those of the fibrils.
AFM images indicated that fibrils became shorter and amorphous aggregates accumulated, with an increase in the size of liposomes, as was evident from AFM images obtained in the presence of 100-and 200-nm liposomes. TEM images in the presence of 100-and 200-nm liposomes revealed that fibrils associated with the ruptured liposomes. TIRFM images in the presence of 100-and 200-nm liposomes showed the significant clustering of fibrils. TEM images in the presence of large liposomes showed thick and short fibrils. Taken together, the morphologies of A␤-(1-40) fibrils observed by TIRFM, AFM, and TEM depended on the size of the coexisting liposomes, which indicated that fibrils became shorter and clustered with an increase in the size of the liposomes, and the amount of amorphous aggregates increased.
Binding of A␤-  to Liposomes Characterized by ITC-To examine the interactions between A␤-(1-40) and DOPC liposomes, we performed ITC measurements (Fig. 6). Liposomes with the sizes of 30 nm and 100 nm were chosen to address the effects of high and low curvatures, respectively, on A␤-(1-40) fibrillation. Consecutive titrations of A␤-(1-40) to 30-nm liposomes in the ITC cell showed the positive ITC peaks with gradual decreases in the intensity (Fig. 6A). The heats of dilution and mixing of A␤-(1-40) to buffer were much smaller than those of A␤-(1-40) titrations to liposomes (Fig. 6A, inset).
To obtain further insights into interactions between A␤-(1-40) and liposomes, we evaluated apparent thermodynamic parameters based on analyses of binding isotherms. The values of app K d of A␤-(1-40) for 30-nm and 100-nm liposomes were 67.1 M and 36.4 M, respectively. These values were smaller than those of ␣-synuclein binding to the liposomes, which mimicked presynaptic membranes ( app K d ϭ 261 M), or the liposomes, which consisted of 1-palmitoyl-2-oleoyl-sn-glycero-3-phospho-(1Ј-rac-glycerol) ( app K d ϭ 93 M) (52). Accordingly, app ⌬G bind (Ϫ6.3 kcal/mol) of A␤ to 100-nm liposome was smaller than that of A␤ to the 30-nm liposomes (Ϫ5.9 kcal/ mol). These indicated that apparent binding affinity of A␤-(1-40) to DOPC liposomes was slightly stronger than those of ␣-synuclein to model liposomes and that affinity of A␤ to 100-nm liposome was slightly stronger than that to 30-nm liposome.
Interestingly, the sign and amplitude of app K a and app ⌬S bind were different depending on the size of liposomes. The A␤-30-nm liposome binding was driven by the positive entropy change (T app ⌬S bind ϭ 14.3 kcal/mol) to overcome the unfavorable endothermic reaction ( app K a ϭ 8.4 kcal/mol). On the contrary, the A␤-100-nm liposome binding was favored by the negative enthalpy change ( app K a ϭ Ϫ62.0 kcal/mol) to overcome the entropy loss (T app ⌬S bind ϭ Ϫ55.7 kcal/mol). The large value of app K a in the A␤-100-nm liposome binding may imply the complex mechanism in comparison with the A␤-30-nm liposome binding. Similar complicated ITC results were observed for the ␣-synuclein binding to liposomes consist of 1,2-dioleoyl-sn-glycero-3-phospho-(1Ј-rac-glycerol) (53).
Moreover, biphasic patterns of each ITC peak in titration of A␤ to 100-nm liposomes suggested apparent two processes; the fast process with a sharp and intense peak was followed by the slow process with a broad and low intensity peak (Fig. 6B). These indicated that binding of A␤ to 100-nm liposomes was not a simple intermolecular interaction, although further studies are required to clarify the details of the liposome size-dependent binding of A␤.
As can be seen in the ITC thermograms (Fig. 6), the binding reactions between A␤-(1-40) and liposomes did not saturate, which brought uncertainty of the fitting. Thus, the error values of K a were large. Calculation of other thermodynamic parameters by using K a with the large error further increased uncertainty. To obtain a saturated curve, which can decrease errors, it was required to increase the concentrations of A␤-(1-40) in the syringe. However, the concentration of A␤-(1-40) (100 M) used was already high, and it was not practical to further increase the A␤-(1-40) concentration due to the formation of nonspecific aggregations. Although decreasing temperature to delay nonspecific aggregation might be an alternative possibility at higher A␤-(1-40) concentrations, there were concerns about changing the binding mode and/or aggregation pathway compared with those at 37°C. Thus, we conclude that although the errors were large, the ITC results clearly showed distinct bindings of A␤-(1-40) to the two kinds of liposomes and suggested that the binding mode and aggregation pathway of A␤-(1-40) depend on the degree of membrane curvature.
Effects of Anionic Phospholipids on the Fibrillation-To address the role of electrostatic interactions between A␤-(1-40) and membranes in fibrillation, we examined the effects of liposomes composed of DOPC and DOPS at a ratio of DOPC: DOPS ϭ 1:1 (i.e. 50% DOPS liposomes). DOPS is one of anionic phospholipids reported to accelerate fibrillation of A␤-(1-40), and it has been suggested that phosphatidylserine groups of DOPS liposomes interact with positive-charged Lys residue at the C terminus of A␤-(1-40) (37). However, the dependence on the size of liposomes has not been examined. We monitored the fibrillation A␤-  in the presence of 50% DOPS lipo- somes of various sizes (Fig. 7A). Only the 30-nm liposomes with 50% DOPS shortened the lag time (Fig. 7B).
We then prepared the 30-nm liposomes composed of 100% DOPC, 50% DOPC and 50% DOPS, or 100% DOPS and monitored the fibrillation of A␤-  in the presence of these liposomes (Fig. 7, C and D). The liposomes containing 50% DOPS promoted the nucleation reaction as well as 100% DOPC liposomes. However, the degree of promotion of DOPS liposomes was small.
Thus, although the DOPS liposomes were reported to accelerate the A␤-(1-40) fibrillation (37), we saw no clear acceleration under our conditions. On the contrary, we observed slightly adverse effects to suppress the accelerating effects produced by small DOPC liposomes. It is possible that, under our experimental conditions, the electrostatic interaction between the negative-charged head group of PS and positive-charged amino acid residues of A␤-(1-40) tightly trapped the A␤-(1-40) molecules on liposomes so that further nucleation became difficult.

DISCUSSION
We herein showed that DOPC liposomes markedly accelerated the fibrillation of A␤-(1-40) and that these effects depended on the size of liposomes even at the same total concentration of phospholipids; smaller liposomes more strongly accelerated fibrillation. Acceleration was suppressed by an increase in the liposome size. The amount of fibrils decreased, and that of amorphous aggregates increased.
To understand these distinct effects of liposomes on fibrillation, we assumed two types of interactions between A␤-(1-40) and liposomes; one was productive binding leading to nucleation and fibrillation, and the other was non-productive binding that formed amorphous aggregates (Fig. 8, A and B). An analysis of the water-accessible hydrophobicity of liposomes probed by ANS binding and estimation of curvature suggested that small liposomes with higher surface curvature had more water-accessible hydrophobic regions and were effective for producing amyloid nucleation. Small liposomes with larger water-accessible hydrophobic regions interact with A␤ monomers weakly and concentrate them adequately to initiate nucleation. Although the binding of A␤-  to larger liposomes could also occur, it may be non-productive for amyloid nucleation because the local concentration was not sufficiently high to break supersaturation. Alternatively, although the local concentration was high, bound A␤-(1-40) molecules were tightly trapped on the surface of membranes and, thus, were unable to diffuse and transform to seed competent conformations. The ITC measurements showed that the affinity of A␤ to the 100-nm liposome was slightly stronger than that to the 30-nm liposome, consistent with the tighter binding and trapping of the A␤-(1-40) on the larger liposomes. These two types of interactions may propagate beyond liposome surfaces, leading to amyloid fibrillation or amorphous aggregation in bulk solvent.
The competitive mechanism between productive and nonproductive binding (Fig. 8A) was analogous to the competition between amyloid fibrillation and amorphous aggregation in the absence of liposomes, as represented by a phase diagram of amyloid fibrillation dependent on salt and protein concentrations. Amyloid fibrils have been shown to form in the supersaturated solutions of responsible proteins via a nucleation and growth mechanism (41, 54 -56). Supersaturated regions can be classified into two types depending on the stability of supersaturation (Fig. 8C) (41). Supersaturation was remarkably stable in the metastable region, at which the solute concentration was slightly above the solubility, and seeding was essential to break supersaturation and make fibrils. Spontaneous fibrillation could occur after a certain lag time in the labile region, in which the driving force of precipitation increased. Various kinds of agitations such as shaking (57), stirring (58), or ultrasonic irradiation (44,59,60) have been shown to effectively force spontaneous fibrillation under these supersaturated regions. One possible mechanism responsible for agitation-induced fibrillation is the concentration of amyloidogenic proteins at the hydrophobic air-water interface produced by agitation (41,61). Ultrasonication is known to produce cavitation bubbles with hydrophobic surfaces, and stirring or shaking will also increase the hydrophobic air-water interface. The hydrophobic air-wa-ter interface adsorbs proteins, concentrates them, and produces various aggregates including seed competent aggregates. Once amyloid fibrillation starts on the seeds, it even propagates to bulk regions without a hydrophobic surface.
Water-accessible hydrophobic surfaces produced by membrane curvature may play a similar role in membrane-assisted amyloid fibrillation to the air-water hydrophobic surface of cavitation bubbles induced by ultrasonic irradiation. A␤ monomers may be concentrated effectively by absorption to the hydrophobic surface due to membrane curvature. Thus, the membrane-assisted acceleration of fibrillation might be similar to the agitation-induced acceleration of fibrillation in the labile region (Fig. 8).
In the phase diagram excessively strong precipitation forces produced glassy amorphous aggregates in which molecules were trapped in non-amyloidogenic conformations. One possible mechanism underlying aggregation is that promiscuous interactions between molecules produce various conformational states. In a similar way, various types of interactions on the membrane may trap the monomers, preventing rearrangements to seed-competent conformations. The evenly dispersed charged head groups of phospholipids may contribute to preventing the bound A␤ monomers from associating to the seed-competent conformation. Thus, we considered that membrane-assisted amorphous aggregation corresponded to amorphous aggregation under strong forces of precipitation (Fig. 8).
We previously reported the amyloid fibrillation of A␤-  in real time at a single fibrillar level at pH 7.0 with TIRFM combined with the use of ThT (62). We used the surfaces of various chemically modified substrates. Extensive fibrillation was generally observed on surfaces with negative charges. In contrast, fibril growth was largely suppressed on positively charged or hydrophobic surfaces. A previous study reported In the presence of larger liposomes, A␤-(1-40) interacts with the surface of membrane more strongly than that of smaller liposomes. This stronger interaction makes it difficult to induce nucleation and leads some A␤ peptides to amorphous aggregates. Other A␤ peptides exist in medium, and they form amyloid fibrils or amorphous aggregates (B). C-E, schematic phase diagram of conformational states of A␤-  in the absence (C) or presence of small (D) or large vesicles (E). The locations of the protein solutions as studied here are indicated by dots.
that A␤ fibrillation was facilitated by negatively charged acidic phospholipids rather than neutral phospholipids (63). Because the pI value of A␤ is ϳ5, the net charge was slightly negative under neutral pH conditions. These results indicated that strong electrostatic attraction tightly trapped the A␤-(1-40) monomers on the surface so that subsequent lateral movement and nucleation were inhibited. The inhibitory role of a strongly hydrophobic surface can be also explained in a similar manner assuming that the hydrophobic surface interacts with the hydrophobic residues of A␤- . These results support the view that there are two types of membrane binding, one leading to amyloid fibrillation and the other to amorphous aggregation.
Finally, although lipid rafts have been considered to play an important role in A␤ fibrillation (13, 16 -19), the underlying mechanism remains unclear (21,22). It is possible that the lipid raft carries the site of productive binding, as proposed here more than the other lipid membranes. Our proposal that there are two types of membrane binding for amyloidogenic proteins, one leading to amyloid nucleation and the other to non-productive binding and amorphous trapping, may be common for various amyloidogenic proteins. The specific removal of amyloidogenic proteins may be possible by preparing liposomes with only non-productive binding.