Cholesterol-dependent Formation of GM1 Ganglioside-bound Amyloid β-Protein, an Endogenous Seed for Alzheimer Amyloid*

GM1 ganglioside-bound amyloid β-protein (GM1/Aβ), found in brains exhibiting early pathological changes of Alzheimer's disease (AD) including diffuse plaques, has been suggested to be involved in the initiation of amyloid fibril formation in vivo by acting as a seed. To elucidate the molecular mechanism underlying GM1/Aβ formation, the effects of lipid composition on the binding of Aβ to GM1-containing lipid bilayers were examined in detail using fluorescent dye-labeled human Aβ-(1–40). Increases in not only GM1 but also cholesterol contents in the lipid bilayers facilitated the binding of Aβ to the membranes by altering the binding capacity but not the binding affinity. An increase in membrane-bound Aβ concentration triggered its conformational transition from helix-rich to β-sheet-rich structures. Excimer formation of fluorescent dye-labeled GM1 suggested that Aβ recognizes a GM1 “cluster” in membranes, the formation of which is facilitated by cholesterol. The results of the present study strongly suggested that increases in intramembrane cholesterol content, which are likely to occur during aging, appear to be a risk factor for amyloid fibril formation.

The critical step in the development of Alzheimer's disease (AD) 1 is the conversion of soluble, nontoxic amyloid ␤-protein (A␤) to aggregated, toxic A␤ rich in ␤-sheet structures (1). A␤ has been shown to form amyloid fibrils, but this requires concentrations of A␤ (Ͼ10 Ϫ4 to 10 Ϫ5 M) (2-4) much higher than the physiological concentration (10 Ϫ9 M). Therefore, it has been hypothesized that aggregation of soluble A␤ in vivo involves seeded polymerization (5,6).
Yanagisawa et al. (7) discovered GM1 ganglioside-bound A␤ (GM1/A␤) in brains exhibiting early pathological changes of AD and suggested that GM1/A␤ may serve as a seed for toxic, amyloid fibril formation. Indeed, immunochemical (8,9) and spectroscopic (10 -13) studies demonstrated that GM1/A␤ has a conformation different from that of soluble A␤ and accelerates the rate of amyloid fibril formation of soluble A␤ in vitro (12,14). Interestingly, however, GM1/A␤ is never found in the normal brain despite the fact that neuronal membranes are abundant in GM1 and physiological metabolism of amyloid precursor protein results in extracellular secretion of A␤. Thus, identification of factors that initiate formation of GM1/A␤ may be crucial for determination of the pathogenesis of AD and for development of preventive and curative treatment strategies. We have recently found that alterations in lipid composition of the host membrane can be such a factor (13); generation of GM1/A␤ is facilitated by the combination of cholesterol and sphingomyelin (SM) in membranes in proportions similar to the so-called detergent-insoluble glycolipid-rich domain (DIG) (15), suggesting that DIG is deeply involved in GM1/A␤ formation. This hypothesis is in agreement with the observation that A␤ is present in DIG in vivo (16,17).
In this study, the effects of GM1 and cholesterol contents in the membranes on the binding of A␤ to DIG-like lipid bilayers were examined in detail using fluorescent dye-labeled human A␤- . We report here that enrichment of cholesterol of the host membranes facilitated the generation of GM1/A␤ via formation of a GM1 "cluster" that acts as a binding site of A␤. A plausible mechanism of onset of AD will be discussed.

EXPERIMENTAL PROCEDURES
Peptides-Human A␤-  labeled with the 7-diethylaminocoumarin-3-carbonyl group at its N terminus (DAC-A␤, Fig. 1) was custom synthesized by the Peptide Institute (Minou, Japan). The peptide was characterized by matrix-assisted laser desorption ionization mass spectroscopy (calculated, 4574.07; found, 4574.0) as well as amino acid analysis under two different hydrolysis conditions. The dye-labeled peptide was always handled in light-protected, capped tubes under a nitrogen atmosphere to avoid photodegradation. Unlabeled human A␤-(1-40) was also purchased from the Peptide Institute. The latter peptide was first dissolved in 1,1,1,3,3,3-hexafluoro-2-propanol (Wako, Osaka, Japan) to avoid self-aggregation. After removal of the solvent by nitrogen purging, the peptide was redissolved in pure water (Nanopure) at 30 M and then mixed with an equal volume of double concentrated buffer (20 mM Tris, 300 mM NaCl, 4 mM CaCl 2 , pH 7.4). The labeled peptide was found to be less stable in this organic solvent and therefore was directly dissolved in pure water. Physiological A␤ is present in a soluble form. To mimic this situation, we removed aggregates, if any, by ultracentrifugation in 500-l polyallomer tubes at 100,000 ϫ g at 4°C for 1 h. Indeed, the supernatant used adopted unordered structures (see Fig. 6) and did not react with thioflavin T (data not shown), which has been widely used for the detection of amyloid aggregation (18). The aggregational state of the peptide was further characterized by SDSpolyacrylamide gel electrophoresis using precast 15% polyacrylamide gel SPU-15S (Atto, Tokyo). Only a band corresponding to monomer was detected by silver staining (data not shown). The peptide concentration of the supernatant was determined in triplicate by Micro BCA protein assay (Pierce).
Lipid Vesicle Preparation-Large unilamellar vesicles (LUVs) for fluorescence experiments were prepared and characterized as described elsewhere (19). Briefly, lipids were mixed in a chloroform/methanol mixture. The solvent was removed by evaporation in a rotary evaporator. The residual lipid film, after drying under vacuum overnight, was hydrated with buffer (10 mM Tris, 150 mM NaCl, 2 mM CaCl 2 , pH 7.4) and vortex-mixed to produce multilamellar vesicles. A physiological concentration of Ca 2ϩ was included because this divalent ion is known to interact with gangliosides (20). For PS LUV preparation, CaCl 2 , which induces aggregation and fusion of vesicles, was omitted, and 1 mM EDTA was added. The suspension was subjected to five cycles of freezing and thawing and then extruded through polycarbonate filters (100-nm pore size filter, 31 times) using a Liposofast extruder (Avestin, Ottawa, Canada). The lipid concentration was determined in triplicate by phosphorus analysis (21).
Small unilamellar vesicles for CD experiments were prepared by sonication of multilamellar vesicles under a nitrogen atmosphere for 15 min (three times for 5 min each) using a probe-type sonicator. Metal debris from the titanium tip of the probe was removed by centrifugation.
Fluorescence-Fluorescence measurements were carried out on a Shimadzu RF-5000 or RF-5300 spectrofluorometer with a cuvette holder thermostatted at 30°C. After blank subtraction (and volume correction in titration experiments), the spectra were corrected using the spectrum correction attachment provided by the manufacturer.
Fluorescence Titration-DAC-A␤ solution (1 M, 2 ml) was titrated with aliquots of a concentrated LUV suspension in a quartz cuvette with gentle stirring. Fluorescence emission spectra were recorded at an excitation wavelength of 430 nm. The titration interval was 3 min, which was confirmed to be sufficient for the establishment of binding equilibrium.
For competitive binding experiments, GM1-rich cholesterol-rich LUVs (20 M) were mixed with a DAC-A␤ solution (0.16 M). Under this condition, the binding sites were almost saturated with DAC-A␤. The mixture (2 ml) was titrated with aliquots of an unlabeled A␤-(1-40) solution (280 M, monomeric confirmed by SDS-polyacrylamide gel electrophoresis), which was prepared by dissolving the peptide in 0.02% ammonia on ice followed by ultracentrifugation (100,000 ϫ g, 3 h, 4°C) (22). Fluorescence intensity at 470 nm (excitation at 430 nm) was monitored during titration. Five minutes were required to establish equilibrium.
Excimer Fluorescence-LUVs containing 5 or 10 mol % BODIPY-GM1 (Molecular Probes, Inc., Eugene, OR) were placed in a quartz cuvette. Fluorescence emission spectra were recorded at an excitation wavelength of 480 nm. Fluorescence anisotropy (r) was determined at an emission wavelength of 520 nm (monomer peak) using polarizers placed in both the excitation and emission light paths (23).
Fluorescence intensity was denoted by I, and the suffixes indicate polarization (in degrees) of the excitation-emission beams. Fluorescence intensity of the corresponding blank sample without BODIPY-GM1 was negligible. CD-Native human A␤-(1-40) (15 M) in buffer (10 mM Tris, 150 mM NaCl, 2 mM CaCl 2 ) was used for CD measurements. CD spectra were measured on a Jasco J-720 apparatus interfaced to an NEC PC9801 microcomputer, using a 1-mm path length quartz cell to minimize the absorbance due to buffer components. The instrumental outputs were calibrated with nonhygroscopic ammonium d-camphor-10-sulfonate (24). Eight scans were averaged for each sample. The averaged blank spectra (vesicle suspension or buffer) were subtracted.

RESULTS
Lipid Specificity of A␤ Binding-Binding of DAC-A␤ to LUVs of various lipid compositions was estimated on the basis of DAC fluorescence. The fluorophore was practically nonfluorescent in aqueous environments ( Fig. 2A, trace 1). The addition of GM1-rich cholesterol-rich LUVs at a lipid/peptide ratio (L/P) of 80 induced a large increase in fluorescence intensity accompanied by a blue shift in the emission maximum from 483 to 470 nm, indicating that the peptide was bound to the membrane with the N-terminal DAC moiety embedded in a hydrophobic environment. Fluorescence spectra of DAC-A␤ were also measured in various dioxane/water mixtures. The maximal wavelength of 470 nm in the presence of the membrane corresponded to that in a dioxane/water (3/1, v/v) mixture with a dielectric constant of ϳ20, suggesting that the DAC moiety was located at the interfacial region of the membrane (25).
To confirm that DAC-A␤ behaves similarly to native A␤, competitive binding experiments were carried out. The binding sites of GM1-rich cholesterol-rich LUVs were almost saturated with DAC-A␤. Unlabeled A␤ was then added, and a decrease in fluorescence was monitored as a function of unlabeled A␤-to-DAC-A␤ ratio (Fig. 2B, closed circles). About 40-fold unlabeled peptide was needed to replace 50% of the labeled peptide. Conversely, the pretreatment of the vesicles with excess unlabeled peptide decreased DAC-A␤ binding (Fig. 2B, open circle). These data suggest that both peptides competitively and reversibly bind to the common binding site with DAC-A␤ having a 40-fold larger affinity.
Thus, DAC fluorescence can be utilized to assess the binding affinity of A␤ for various membranes. As a quantitative measure, relative fluorescence enhancement (R) is defined as follows.
Fluorescence intensities at 466 nm (peak of raw, uncorrected spectra) in the presence and absence of LUVs are denoted by F and F 0 , respectively. Fig. 3  Cholesterol-induced A␤ Binding-The effects of lipid composition on peptide binding to GM1-containing DIG-mimicking membranes were examined in detail. Fig. 4A shows the R value as a function of GM1/DAC-A␤ ratio instead of L/P, because GM1 seems to constitute a "binding site" for the peptide. The binding was strongly dependent on GM1 as well as cholesterol contents. For the cholesterol-rich systems, a decrease in GM1 content from 40 to 20% significantly reduced DAC-A␤ binding (Fig. 4, open circles versus open triangles). Surprisingly, in the case of the GM1-poor systems, a decrease in the cholesterol/SM ratio from 1 to 0.25 markedly reduced DAC-A␤ binding (Fig. 4, open versus closed triangles).
Binding isotherms were obtained from Fig. 4A as follows. R max values were estimated by linear extrapolation of R versus DAC-A␤/GM1 plots (DAC-A␤/GM1 3 0) and are summarized in Table I. In the case of the GM1-poor cholesterol-poor bilayers, the R max value was assumed to be the same as that of the GM1-poor cholesterol-rich system because data close to saturation could not be obtained (Fig. 4A). This assumption was reasonable because the R max values for the other systems were similar. The R/R max ratio gives the bound fraction of the peptide at each data point. Fig. 4B shows binding isotherms, i.e. bound DAC-A␤ per exofacial GM1 (x) versus free DAC-A␤ concentration (c f ) plots. GM1 molecules on the outer leaflets (50% of total GM1) were assumed to be available for DAC-A␤ binding. The two isotherms of the GM1-rich systems were sigmoidal, implying cooperative binding. Therefore, the curves were analyzed by Equation 4 (Fowler's equation) instead of the conventional Langmuir equation (26).
The  Table I. Interestingly, the binding affinities (K) were very similar (2-3 ϫ 10 6 M Ϫ1 ), whereas the binding capacities (x max ) were highly dependent on GM1 as well as cholesterol contents. Detection of GM1 Cluster-Excimer formation of BODIPY-GM1 was utilized to detect the GM1 cluster. The fluorophore BODIPY is known to form an excimer that emits red-shifted fluorescence (ϳ630 nm) compared with monomer (ϳ520 nm) (27). The excimer formation, which occurs upon collision of two dye molecules (one is in the excited state), is facilitated by higher local concentration of the dye as well as lower membrane rigidity. Fig. 5 shows the fluorescence emission spectra of BODIPY-GM1-labeled liposomes. The spectra are normalized to the monomer peaks because the excimer/monomer fluorescence ratio is directly proportional to local dye concentration (28). BODIPY-GM1/GM1/cholesterol/SM (10:30:30:30) liposomes corresponding to the GM1-rich cholesterol-rich system exhibited a large excimer fluorescence (trace 1). In contrast, BODIPY-GM1/PC (10:90) liposomes with the identical dye content showed much weaker excimer fluorescence (trace 2). The anisotropy value of monomer fluorescence (r) as a measure of membrane rigidity of the latter membrane (0.046) was significantly smaller than that of the former (0.104). Therefore, BODIPY-GM1 forms an excimer much more easily in the former DIG-like environment despite its higher rigidity (higher anisotropy), strongly indicating that GM1 is present in a locally concentrated state, i.e. in a cluster.
A reduction in total GM1 concentration from 40 to 20% markedly decreased excimer fluorescence (BODIPY-GM1/ GM1/cholesterol/SM ϭ 5:15:40:40, corresponding to the GM1-  Table I. poor cholesterol-rich system, trace 3), although the intensity was much greater than that of the control BODIPY-GM1/PC (5:95) system (trace 5). The r value of the former (0.119) was again greater than that of the latter (0.041). It should be noted that the BODIPY-GM1/GM1 ratio remained constant at 1:3 in both GM1 40 and 20% systems. Therefore, if all GM1 molecules had been involved in the cluster formation, the same excimer fluorescence would have been observed. The weaker excimer fluorescence in the GM1-poor system indicated that the extent of clustering was smaller. BODIPY-GM1/GM1/cholesterol/SM (5:15:16:64) membranes corresponding to the GM1-poor cholesterol-poor system (r ϭ 0.115, trace 4) exhibited further weaker excimer fluorescence compared with trace 3. Thus, the extent of clustering was in the order GM1-rich cholesterol-rich Ͼ Ͼ GM1poor cholesterol-rich Ͼ GM1-poor cholesterol-poor, consistent with the order of the x max values ( Table I). The excimer formation in BODIPY-GM1/GM1/cholesterol/SM (10:30:12:48) bilayers corresponding to the GM1-rich cholesterol-poor system was also examined. The excimer fluorescence was even larger than that of trace 1 (data not shown), but the r value was much smaller (0.064). Therefore, this system could not be directly compared with the other systems.
Secondary Structure-The conformations of A␤-(1-40) were estimated from CD spectra. Fig. 6 shows data of the GM1-rich cholesterol-rich system. The spectrum in buffer had a minimum at 197 nm, characteristic of unordered structures (trace 1). At lower GM1/A␤ ratios, the spectra exhibited shallow minima around 218 nm, reminiscent of ␤-sheets (Fig. 6, traces 2  and 3). In contrast, the peptide adopted helical structures at higher GM1/A␤ ratios, as suggested by double minima around 209 and 222 nm (Fig. 6, traces 4 and 5). As estimated from the ellipticity at 222 nm (Ϫ13,000 degrees cm 2 dmol Ϫ1 ), the helicity at the largest GM1/A␤ value investigated was ϳ40% (29). The absence of an isodichroic point indicated that the helix-to-sheet transition is not a simple two-state process. The GM1-rich cholesterol-poor system showed very similar CD spectra (data not shown).

DISCUSSION
Dye Labeling-A␤ peptides with slight chemical modifications, such as 125 I (30 -32) and Trp labeling (14), have been widely used in many studies and provided valuable information. The DAC moiety employed in this study was as small as a single aromatic amino acid and was attached to the N terminus of A␤. Indeed, DAC-A␤ behaved very similarly to the native peptide. First, DAC-A␤ shared the common binding site with native A␤ (Fig. 2B). Second, the interfacial location of the DAC moiety in membranes is fully compatible with the observation that native A␤ lies on the surface of GM1-containing membranes (13). Third, DAC-A␤ showed lipid specificity identical to that of the native peptide, as described below.
Lipid Specificity-DAC-A␤ showed no affinity for phospholipids PC or PS but high affinities for GM1-containing membranes at physiological ionic strength (Fig. 3), consistent with previous studies using native (11,13) and Trp-labeled A␤ (14). Our study clearly indicated that A␤ does not bind cholesterol, because (i) DAC-A␤ was not bound to PC/cholesterol (2:1) liposomes (Fig. 3) and (ii) The cholesterol content in DIG-mimicking membranes did not correlate with A␤-binding activity (Fig.  4). Wood's group (33) reported that aggregated but not freshly dissolved A␤ binds cholesterol, in accordance with our results.
Binding Isotherms-Binding of peptides to lipid bilayers has often been analyzed by a simple partition model that does not include the concept of binding sites, because in most cases the driving force of peptide binding is not specific molecular recognition but simple electrostatic and hydrophobic interactions (34 -36). However, A␤ obviously recognizes GM1 or more accurately gangliosides (10 -14). Therefore, the binding isotherms were analyzed by the cooperative binding model (Equation 4). The binding affinities, K, were practically the same (2-3 ϫ 10 6 M Ϫ1 ) for the four DIG-mimicking systems investigated (Table  I). The competitive binding experiments (Fig. 2B) suggested that the K value of native A␤ is 40-fold smaller (ϳ 6 ϫ 10 4 M Ϫ1 ), which corresponds to a difference in a Gibbs free energy of 2 kcal/mol. This value is a reasonable one for transfer of the DAC moiety from water to membrane interface (25). A binding af- c Assumed to be the same as that of the GM1-poor cholesterol-rich system. d The large error is due to the absence of data points close to the origin (Fig. 4B). finity of 7 ϫ 10 5 M -1 was reported for the Y10W-A␤-(1-40)-GM1 system (14). The Tyr-to-Trp substitution enhances membrane binding by ϳ1 kcal/mol (25), which corresponds to a 5-fold increase in affinity. Therefore, taken together with the present estimation, the affinity of native A␤ for GM1 is estimated to be ϳ10 5 M Ϫ1 . Even in the absence of specific molecular recognition, binding isotherms are fitted by the Langmuir equation.
If individual GM1 molecules constitute binding sites for A␤, the binding isotherms (Fig. 4B) would be independent of lipid composition. However, the binding capacity increased quadratically with intramembrane GM1 content in the cholesterolrich matrix (Fig. 4B and Table I), suggesting that some cooperative interactions between GM1 molecules generate the binding site. In accordance with this view, the membrane binding of native A␤-(1-40) also requires a threshold GM1 content dependent on the lipid composition of the host matrix (12,13). The most straightforward interpretation is that A␤ recognizes not monomeric but clustered GM1 or that in a GM1-enriched microdomain (39), the formation of which is regulated by cholesterol content. Indeed, a correlation was observed between binding capacity (x max in Table I) and excimer formation of BODIPY-GM1 as a measure of clustering (Fig. 5). However, the relationship between x max and excimer fluorescence was semiquantitative. The x max value of the GM1-poor cholesterol-poor system was very small, whereas significant excimer fluorescence was observed (Fig. 5, trace 4), probably because the introduction of relatively high amounts of BODIPY-GM 1 (25% of total GM1) slightly affected domain formation. The excimer experiments also suggested that DIG-like environments play a crucial role in GM1 clustering. Ferraretto et al. reported that GM1 as well as cholesterol form GM1-and cholesterol-enriched domains in SM bilayers (39). It is therefore plausible that at lower GM1 contents (e.g. 20%) an increase in cholesterol content, through segregation of cholesterol molecules by the cholesterol-rich domain formation, enhances local GM1 concentration, which leads to GM1 clustering (Fig. 4). In contrast, a GM1 content of 40% appears to be sufficient for effective formation of GM1-rich domains, and thus cholesterol content would have no further effect on GM1 clustering. In the PC matrix, a high level of excimer fluorescence was never observed despite its lower rigidity (Fig. 5, traces 2 and 5). An electron microscopic study also indicated that GM1 is molecularly dispersed in fluid phosphatidylcholine bilayers at lower GM1 contents (40).
Cooperativity was observed for peptide binding to the GM1rich membranes with larger capacities. This may be related to conformational transition from ␣-helix-rich conformations at lower x values to ␤-sheet-rich conformations at higher x values (Fig. 6). The formation of the latter structures can involve interpeptide interactions. Terzi et al. (41) reported similar structural transition in phosphatidylglycerol bilayers at low ionic strength. The presence of ␣/␤ structures was also found in ganglioside-containing membranes (10). The conformational transition of the N-terminal region (residues 10 -24) from ␣-helix to ␤-strand was reported to facilitate amyloid formation (42).
Pathological Implications-Recently, a great deal of attention has been focused on the pathological implications of altered cholesterol metabolism, which is likely to occur with aging or the expression of apolipoprotein E, in the development of AD. There is accumulating evidence that the metabolism of amyloid precursor protein, including A␤ generation, is significantly modulated by the content of cellular cholesterol (43)(44)(45). It is particularly of interest that a recent study indicated that a unique A␤ species with seeding ability was generated by cultured cells in a cholesterol-dependent manner (46). Even with this information we are still far from understanding how cholesterol is involved in the development of AD, especially in the initiation of amyloid fibril formation.
The results of the present study indicated, for the first time, that increases in the content of cholesterol in the membrane induce the formation of GM1/A␤, one of the candidates as an endogenous seed for Alzheimer amyloid. With regard to the alteration of cholesterol in neuronal membranes, recent studies by Wood and co-workers (47,48) are very informative. They reported that the content of cholesterol in the exofacial leaflets of synaptic plasma membrane is increased during aging (47) and by apolipoprotein E deficiency (48). Taken together with the results of present study, these observations strongly suggest that alterations in the content of cholesterol in neuronal membranes underlie abnormal aggregation of A␤ in the AD brain.
Finally, if A␤ forms amyloid fibrils via seeded polymerization in the brain with AD, then the seed could be a target for therapeutic and preventive treatment regimens for AD. Indeed, we have recently found that GM1/A␤ formed in DIG-like membranes works as a seed for fibril formation. 2 To generate a compound that specifically recognizes GM1/A␤ and inhibits its seeding ability, it will be necessary to clarify the molecular processes underlying alterations of the secondary structures of A␤ via binding to and accumulation in GM1 "clusters."