Membrane disruption by Alzheimer beta-amyloid peptides mediated through specific binding to either phospholipids or gangliosides. Implications for neurotoxicity.

Increasing evidence implicates Abeta peptides as neurotoxic agents in Alzheimer's disease. We investigated one possible mechanism of neurotoxicity, namely Abeta-membrane lipid interactions. We find that Abeta disrupts membranes containing acidic phospholipids. This disruption is greater at slightly acidic pH (characteristic of endosomes) than at neutral pH (characteristic of the extracellular space). This pH dependence suggests that Abeta has the capacity to disrupt endosomal and plasma membranes, and this disruption could account, at least in part, for the observed neurotoxic effects of the peptide. We also find that gangliosides induce Abeta to adopt a novel alpha/beta conformation at neutral pH.

). Possible mechanisms for A␤ neurotoxicity include (a) alterations in Ca 2ϩ homeostasis, (b) activation of specific receptors affecting cellular homeostasis, (c) direct disruption of membrane integrity, or (d) a combination of two or more of the above mechanisms.
Several studies indicate that A␤ neurotoxicity may be mediated, at least in part, by direct interactions between A␤ and membrane lipids. Arispe et al. (16) demonstrated that A␤40 forms cation-selective channels in membranes and have speculated that these channels disrupt ion homeostasis and thus cause toxicity. Terzi et al. (17,18) investigated the interaction of phosphatidylcholine (PC)/phosphatidylglycerol (PG) vesicles with A␤ peptides using biophysical techniques. Both A␤25-35 and A␤40 were shown to form ␤-sheet structures upon addition of PC/PG vesicles at pH 7.0; however, supraphysiological concentrations of A␤, 200 M, were required for the conformational transition. Terzi and co-workers (18) also state that the techniques they employed could not determine whether the A␤ peptides penetrated into the lipid membrane or aggregated on the membrane surface. Recently, a novel ganglioside-bound A␤ species was isolated from AD brain homogenates (19). This form was identified as A␤42 and speculated to be localized on the cell surface where it can act as a seed for amyloid fibril formation.
The experimental studies summarized above provide evidence suggesting that A␤ neurotoxicity stems from interactions of A␤ with membrane proteins or lipids. In the present study, we investigate the nature of the interactions of A␤ with phospholipids and gangliosides. Some of the questions we have addressed include the following. Does A␤ aggregate into fibrils and then partition into the membrane or is the interaction mediated through some other mechanism? Do membrane interactions differ between A␤40 and A␤42? And finally, do A␤membrane interactions at slightly acidic pH (characteristic of endosomes) differ from those at neutral pH (characteristic of the extracellular space)?

EXPERIMENTAL PROCEDURES
Handling of A␤ Peptides and Fibril Formation-A␤40, A␤42, and A␤25-35 were purchased from Bachem Biosciences Inc. (King of Prussia, PA). Peptides were initially dissolved in 0.5 ml of 100% trifluoroacetic acid (Aldrich) diluted in distilled H 2 O and immediately lyophilized. The lyophilized peptide was then dissolved in 40% (v/v) trifluoroethanol (Aldrich) in H 2 O and stored at Ϫ20°C until use. A␤9 -25 was synthesized on a Milligen 9050 Pepsynthesizer (Bedford, MA) and purified by high performance liquid chromatography. Purity of peptide was confirmed by fast atom bombardment-mass spectrometry and amino acid analysis. Nucleation seeds were removed from peptide stock solutions by ultracentrifugation at 100,000 ϫ g for 20 min on a Beckman Airfuge (Beckman, Mississauga, Ontario, Canada). The supernatant was diluted immediately into phosphate-buffered saline (PBS) pH 6.0 or 7.0. Samples were incubated with continual stirring for the indicated periods. Fibril formation was followed at 400 nm by 90°l ight scattering on a Photon Technology International Fluorimeter (South Brunswick, NJ). * This work was supported by a grant from Alzheimer Association (U. S. A.). The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18  Vesicle Preparation-Pure phospholipids, mixtures of bovine brain PG, PS, PA, PC (Avanti Polar Lipids, Alabaster, AL), the mixed ganglioside fraction of bovine brain, and pure gangliosides (Sigmas) were dissolved in either chloroform/methanol (1:1 v/v) (50 g/ml). Mixed ganglioside fractions are composed of predominantly G M1 , and the remainder consisting of G M2 , G D1a , G D1b , and G T series. All mixtures of lipids were at a 1:1 ratio by weight. All lipid aliquots were dried under a stream of nitrogen, lyophilized overnight, and suspended in PBS, pH 7.0, in the presence or absence of a saturated solution of 5-(and -6)carboxy-2Ј,7Ј-dichlorofluorescein (Molecular Probes Inc., Eugene, OR). Lipid suspensions were carried through 10 cycles of freeze-thaw in an acetone:dry ice bath. Small unilamellar vesicles were then prepared by sonication for 20 min in a bath sonicator (Branson Ultrasonic Corp., Danbury, CT). Large unilamellar vesicles were prepared by passage through 100-nm polycarbonate membranes with the Liposofast system (Avestin Inc., Ottawa, Ontario, Canada). Column chromatography on Sepharose 4B (Sigmas) (1.5 ϫ 12 cm) was used to separate free dye from dye-loaded vesicles. Concentration of the phospholipid vesicles was determined using the Bartlett assay (20).
Dye Release Assay-Dye-loaded vesicles were diluted in PBS, pH 6.0 or pH 7.0, to a final lipid concentration of 20 M. After 50 s incubation (with continual stirring), A␤ peptide samples, bee venom mellitin (Sigmas), or solvent controls were added at a 1:20 peptide:lipid ratio. The final peptide concentration was 1.0 M. To monitor dye leakage, we used a Photon Technology International Fluorimeter. Excitation was at 506 nm and emission was monitored at 524 nm with a 4 nM bandwidth. Data were collected at a rate of one data point every 5 s. At the end of the experiment, 10% Triton X-100 was added to a final concentration of 0.3% to obtain complete dye release and measure total fluorescence. Vesicles were not used unless the total fluorescence was greater than 200% of the initial fluorescence. The percentage of total dye release was defined as: To control for artifacts arising from light scattering, the dye release assay was also performed with vesicles free of dye. Circular Dichroism-CD spectra were recorded on an Aviv Circular Dichroism Spectrometer Model 62DS (Lakewood, NJ) at 25°C. Spectra were obtained from 200 to 260 nm, with a 0.5-nm step, 1-nm bandwidth, and 20-s collection time per step. Peptide:lipid ratios were maintained at 1:20 with a final peptide concentration of 10 M. The effect of various lipids on peptide conformation was determined by adding an aliquot of stock peptide solutions to lipid vesicles suspended in PBS with continual stirring. The contribution of lipid vesicles to the CD signal was removed by subtracting the CD spectra of pure lipid vesicles from the CD spectra of peptide-lipid suspensions. Stock peptide conformations were determined in TFE/H 2 O under the same conditions.

RESULTS
Dye Release Assay-The ability of A␤ to disrupt lipid membranes was assessed using a dye release assay. This assay involves encapsulating a self-quenching fluorescent dye in lipid vesicles and monitoring the increase in fluorescence upon addition of a membrane disrupting molecule. We used 5-(and 6-) carboxy-2Ј,7Ј-dichlorofluorescein as the dye because its fluorescence is independent of pH above 5.7. We found identical re-sults when vesicles were prepared as either large or small unilamellar vesicles (data not shown). The integrity of the vesicles was determined by monitoring spontaneous diffusion of dye over time (Table I). If the spontaneous dye release exceeded 10% of the total release vesicles or if bee venom mellitin, a 26-amino acid cytotoxic peptide (21), did not induce 100% dye release, then the vesicles were not used in this study (Table I).
Lysosomal/endosomal processing of APP liberates A␤ that is initially monomeric. To simulate this process, we prepared concentrated stock solutions of peptides in 40% trifluoroethanol (TFE), conditions that maintain A␤ in monomeric helical form (22), and then diluted the stock solutions into assay medium containing dye-loaded vesicles. The final concentration of TFE in a typical assay was 0.4%. This amount of TFE did not compromise the integrity of the vesicles (Table I). All experiments were done at physiological salt concentrations and at 1 M peptide concentration unless otherwise stated. Although still supraphysiological, this peptide concentration approaches physiological levels more closely than previous studies. Experiments were performed at pH 6.0 and pH 7.0 to simulate the endosomal and extracellular environments, respectively.
Interaction of A␤ Peptides with Phospholipids-The ability of A␤40 ( Fig. 1) and A␤42 (Fig. 2) to disrupt phospholipid vesicles composed of either PG or PC was examined. When A␤40 was added to PG vesicles at pH 6.0, a dramatic release of dye was detected (Fig. 1A), indicating that A␤40 disrupts the bilayer, allowing approximately 80% of the encapsulated dye to escape. A␤40 induced a more modest disruption of vesicles composed of phosphatidic acid (PA) and phosphatidylserine (PS) at pH 6.0. 2 A␤40 also disrupted PG vesicles at pH 7.0; however, the extent was decreased to 20% release of the encapsulated dye (Fig. 1B). Unlike the effect we observed with acidic phospholipids (PG, PA, and PS), A␤40 did not disrupt PC vesicles at either pH 6.0 ( Fig. 1A) or pH 7.0 (Fig. 1B). Thus, the membrane disrupting activity of A␤40 displays a specificity for acidic phospholipids. Membrane disrupting activity and phospholipid specificity of A␤42 ( Fig. 2A) were similar to A␤40 at pH 6.0; however, at pH 7.0 A␤42 was inactive (Fig. 2B). The error associated with light scattering, from A␤-induced aggregation of lipid vesicles, was found to be approximately 4% of the total fluorescence from the dye release assay and therefore negligible to the interpretation of our results.
Although the physiological concentration of A␤ is not strictly known, it is present under normal conditions in the cerebrospinal fluid at 0.5 nM and in cell culture at 1 nM concentrations (23). Decreasing the concentration of peptide decreases the percentage dye release from acidic phospholipid vesicles. Our data, at 100 nM concentrations, indicate that A␤40 is still able to release dye from PG vesicles above the 40% TFE control (  a 30 l of A␤9 -25 and A␤25-35 (dissolved in 40% TFE) were added to lipid vesicles in a 1:20 peptide to lipid ratio with a final peptide concentration of 1 M. 30 l of 40% TFE was used as the solvent control. Spontaneous dye release of the vesicles was determined by monitoring the increase in fluorescence of vesicles on their own. Data points were collected continuously for 10 min, and the data are reported as the maximal % dye release. S.D. were calculated from at least five experiments for each lipid system. All other values are the mean of two experiments. MG, mixed gangliosides; ND, not determined.
3). A␤42 is unable to induce dye leakage at these low concentrations.
To locate the region of A␤40/42 which is responsible for disrupting membranes, we examined the A␤ fragments, A␤9 -25 and A␤25-35; neither fragment disrupted PG nor PC vesicles (Fig. 1, A and B; Table I). These results imply that at low micromolar concentrations the entire A␤40/42 sequence is required to disrupt phospholipid membranes and allow passage of dye. In addition, these peptides serve as negative controls for the dye release assay (Table I).
Interaction of A␤ Peptides with Gangliosides-Gangliosides are sialic acid containing glycosphingolipids that are important in biosignaling and memory function. Recently, a gangliosidebound form of A␤ was isolated from AD brain (19). Pure gangliosides do not form stable vesicles, and they are commonly mixed with PC for vesicle studies. We have examined the ability of A␤40 (Fig. 1) and A␤42 (Fig. 2) to disrupt vesicles composed of a 1:1 mixture (by weight) of PC and gangliosides from bovine brain. We have shown above that neither A␤40 (Fig. 1) nor A␤42 (Fig. 2) interacts with PC vesicles at pH 6.0 or 7.0; consequently, any effects observed with mixed ganglioside/PC vesicles must result from A␤-ganglioside interactions alone. Of all the lipid vesicles tested here, the greatest disrupting activity was seen when either A␤40 (Fig. 1C) or A␤42 ( Fig.  2A) was added to mixed ganglioside/PC vesicles at pH 6.0. In addition, the percent dye release, 90%, of A␤40/42 from mixed ganglioside/PC vesicles at pH 6.0 approached that of bee venom mellitin ( Fig. 1 and Fig. 2B). At pH 7.0, on the other hand, A␤40 released 48% dye from mixed ganglioside-containing vesicles (Fig. 1D), whereas A␤42 possessed no activity (Fig. 2B).
We examined the specificity of the ganglioside-disrupting activity of A␤40 and A␤42 using monosialoganglioside G M1 (the major component of brain mixed gangliosides) and asialoganglioside. At pH 6.0, G M1 /PC dye release was less than that observed with mixed ganglioside/PC vesicles for both A␤40 (Fig. 1C) and A␤42 ( Fig. 2A), 35 and 26%, respectively. At pH 7.0, A␤40 released approximately 23% of the encapsulated dye from G M1 /PC vesicles (Fig. 1D), whereas A␤42 was inactive (Fig. 2B). These data indicate that G M1 alone cannot account for the total disrupting activity observed with mixed gangliosides. A␤40 disrupted asialo-G M1 /PC vesicles at pH 6.0 only marginally and was inactive at pH 7.0 ( Fig. 1). A␤42 was inactive at both pH 6.0 and pH 7.0 (Fig. 2). Thus, sialic acid appears to be critical for A␤-ganglioside disruption. The differences in the ability of A␤40/42 to release dye from acidic phospholipids versus acidic gangliosides may involve the presence of the complex sugar moieties that are part of the ganglioside headgroup.
Neither A␤9 -25 nor A␤25-35 disrupted mixed ganglioside/PC vesicles (Fig. 1, C and D; Table I). Thus, as found with A␤-phospholipid interactions, it appears that the entire A␤40/42 sequence is also required for disruption of ganglioside membranes.
Concentration Dependence of A␤40/42-induced Dye Release from Phospholipids and Ganglioside-containing Vesicles-The dye release curves in Figs. 1 and 2 plateau before reaching 100% dye release. This may be the result of multiple vesicle types present in our system, one of which is accessible to A␤40/42 disruption and another which is inaccessible. Alternatively, A␤40/42 may have a strong interaction with a limited number of lipid molecules where it becomes "trapped," thereby leaving a number of vesicles untouched. To investigate this, the dependence of dye release from lipid vesicles on peptide concentration was examined using A␤40/42 and PG vesicles as a model for acidic phospholipids. Increasing concentrations of A␤40/42 resulted in a monotonic increase in dye release. The addition of 5 M A␤40 induces virtually 100% dye release from the PG vesicles at pH 6.0 (Fig. 3A). A␤42, on the other hand, did not reach 100% dye release even at 5 M concentration at pH 6.0. Thus under the conditions tested, the entire intravesicular pool of dye is susceptible to release by A␤40 but not by A␤42. The dye release curves for G M1 /PC vesicles demonstrate a lower and more pronounced plateau at pH 7.0; the effect of increasing peptide concentrations was also investigated in this system. Increasing concentrations of A␤40 produced a monotonic increase in the percent dye release from G M1 /PC vesicles; at the highest concentration tested, 10 M, the percent dye released was 45% (Fig. 3B). A␤42 was unable to induce dye release from G M1 /PC vesicles even at 10 M concentrations. The concentration dependence of A␤40/42-induced dye release suggests that when A␤ peptides are added at low concentrations to dye-loaded vesicles, they bind and disrupt the lipid vesicle; however, after disruption they remain associated with the vesicle and are unavailable for further disruption. Mellitin, on the other hand, induces 100% dye release at 5.8 nM concentrations for all lipid systems tested. The ability of mellitin to induce dye release has been well characterized and has been shown to be dependent on both peptide concentration and lipid composition (21). These results provide evidence that the differences in A␤40/42 induced dye release caused by specific peptide-lipid interactions rather than variability in the vesicle types present in the different preparations.
Induction of ␤-Sheet Conformation by Acidic Phospholipids-The above studies indicate that A␤40/42 possess a pH-dependent membrane-disrupting activity that is specific for acidic phospholipids. To determine whether there is a structural basis for this activity, we analyzed A␤40 (Fig. 4) and A␤42 (Fig. 5) in the presence of various phospholipid vesicles by circular dichroism (CD) spectroscopy. While CD measurements were performed on peptide-lipid suspensions, control measurements us-ing pure lipid suspensions were used to eliminate the contribution of the lipids to the total CD signal. Thus the data in Figs. 4 and 5 represent the CD signal originating from the peptides only. A␤40/42 were stored in 40% TFE, under these conditions their CD spectra exhibited a strong minima at 208 nm and a weaker minima at 222 nm (Figs. 4A and 5A). These spectra are indicative of partially ␣-helical structures (24) and are consistent with previous NMR studies on A␤40 in TFE (22). Upon dilution into PBS at pH 6.0 or 7.0, both A␤40 (Fig. 4A) and A␤42 (Fig. 5A) became unstructured. In the presence of PG vesicles at pH 6.0, however, A␤40/42 both possessed a strong CD minima at 218 nm which is indicative of ␤-structure (Figs. 4B and 5B). At pH 7.0, the extent of ␤-structure formation was reduced (Figs. 4B and 5B). Unlike the effects seen with PG vesicles, A␤40/42 remained unstructured in the presence of PC vesicles (Figs. 4C and 5C).
Combining the results of the dye release assay with the CD data, a structure-function relationship emerges. A␤ appears to be unstructured in aqueous solutions at low micromolar concentrations. However, if acidic phospholipids are present and the pH is slightly acidic, A␤ disrupts with the bilayer in a ␤-structured conformation.
Induction of a Novel ␣/␤ Conformation by Gangliosides-At pH 7.0, A␤40/42 disrupted ganglioside vesicles to a greater extent than phospholipid vesicles (Figs. 1 and 2), thus the structure-function relationship in the presence of ganglioside vesicles might differ from that in the presence of phospholipid vesicles. To investigate these possible differences, we examined the secondary structure of A␤40 (Fig. 4) and A␤42 (Fig. 5) in the presence of ganglioside/PC vesicles. As with the phospho- lipid studies, the contributions of the gangliosides were removed from the CD data. Similar to that seen for phospholipids at pH 6.0, mixed ganglioside/PC vesicles induced significant ␤-structure for A␤40 (Fig. 4D) and A␤42 (Fig. 5D). At pH 7.0, on the other hand, the CD spectra of both A␤40 (Fig. 4D) and A␤42 (Fig. 5D) exhibited a strong CD minimum at 220 nm and a weaker one at 208 nm; these spectra are suggestive of a novel ␣/␤ conformation (24). Deconvolution of these spectra using the method of Yang et al. (25) yielded the following results: A␤40 was comprised of 10% ␣-helix, 16% ␤-sheet, and 74% non-␣/␤ structure; A␤42 contained a higher percentage of ␤-sheet structure, 20%, and a concomitant decrease in ␣-helix content, 6%. Similar results were seen for A␤40 (Fig. 4E) and A␤42 (Fig. 5E) in the presence of G M1 /PC vesicles. Asialo-G M1 (Figs. 4F and 5F) could not induce the unique ␣/␤ conformation; therefore, the sialic acid moiety of gangliosides is critical for formation of this structure. The CD spectra of A␤40/42 in the presence of ganglioside/PC vesicles were significantly different from the spectra in 40% TFE, with respect to the positions and magnitudes of the CD minima. Analysis of the CD spectra of A␤40/42 in 40% TFE reveals that A␤40 is 22% ␣-helical and 78% non-␣/␤ structure, whereas A␤42 is 16% ␣-helical. Comparison of the CD spectra of A␤40 and A␤42 in the presence and absence of TFE reveals an isodichroic point at 201 nm, this is indicative of a mixture of ␣-helical and random structures only (26). The same analysis of A␤40/42 in the presence and absence of gangliosides does not reveal this isodichroic point, therefore, indicating the presence of non-␣-helical structure (26). Thus, the three-dimensional structure of A␤40/42 in the presence of gangliosides is different from the structure in 40% TFE.
The results with ganglioside-containing vesicles demonstrate that the structure-function relationship of A␤40/42 exhibits a strong pH dependence. At pH 6.0, this relationship resembles that seen with acidic phospholipid vesicles where A␤40/42 adopts a predominantly ␤-structure and disrupts the membrane. At pH 7.0, on the other hand, both A␤40/42 adopt a unique ␣/␤ structure; however, A␤40 disrupts ganglioside-containing membranes and A␤42 does not.
Peptide Aging, Fibril Formation, and Lipid Interactions-Cell culture assays of neurotoxicity have demonstrated that while addition of freshly dissolved A␤40/42 has minimal toxicity, aging of these peptides, accomplished by incubating the aqueous peptide solutions for 3-7 days at 37°C, induces significant toxicity (11,14,15). It has been shown that during the aging process, A␤ assembles into fibrils (14). Lansbury and co-workers (27) used light scattering measurements to show that A␤-fibrillogenesis is a nucleation-dependent process with a definite lag phase. These experiments require peptide stocks to be free of nucleation seeds (28).
We examined the effect of peptide aging on membrane disruption to determine whether the disrupting activity of A␤40/42 was dependent on fibril formation (Figs. 6 and 7). Prior to peptide aging, we removed potential seeds from stock solutions (ϳ0.18 mM peptide in 40% TFE) by ultracentrifugation and monitored fibril assembly by light scattering. Since membrane vesicles also scatter light, they were added after peptide aging.
When A␤40 (1.0 M) was aged at pH 6.0, fibril assembly exhibited a lag time of 30 -50 min, followed by a period of exponential growth that reached a plateau at approximately 5 h (data not shown). When aged under the same conditions, A␤42 did not exhibit a lag time but did equilibrate at 5 h (data not shown). These light scattering results are qualitatively similar to those performed at higher concentrations and neutral pH conditions (28). We measured the membrane disrupting activity at 0, 5, and 18 h of peptide aging; the light scattering data suggest that at these aging times there should be no fibrils, small fibrils, and large fibrils, respectively. While the percent dye release of A␤40 toward PG vesicles was identical at 0 and 5 h aging, after 18 h the release increased significantly (Fig. 6A). The percentage dye released by A␤42, on the other hand, increased significantly between 0 and 5 h aging and was unchanged at 18 h (Fig. 6B). In summary, we find that conditions that increase the number or size of A␤40/42 fibrils also increase the disrupting activity toward acidic phospholipid membranes. Furthermore, A␤42 acquires its maximal disrupting activity at a faster rate than A␤40, and these kinetic differences correlate with fibril assembly kinetics measured by light scattering. These observations parallel the peptide-aging effects seen in neurotoxicity studies (11,14,15).
The effect of aging on the interaction of A␤ peptides with phospholipids was also investigated at pH 7.0 (Fig. 7). Under our experimental conditions (1 M peptide concentration, physiological NaCl concentration, 25°C), fibril formation by A␤40 at pH 7.0 (assessed by light scattering) did not occur when nucleation seeds were removed from peptide stocks prior to aging; however, when A␤40 was aged without removal of nucleation seeds, fibril formation occurred spontaneously (data not shown). On the other hand, A␤42 formed fibrils spontaneously even when nucleation seeds were removed from peptide stocks (data not shown). We measured the membrane disrupting activity at 0 and 24 h of peptide aging; the light scattering data suggest that at these aging times there should be few, if any, fibrils at 0 h and many large fibrils at 24 h. While A␤40/42 disrupted PG vesicles minimally at 0 h, a significant increase in disruption was observed at 24 h (Fig. 7); however, at all aging times the disruption at pH 7.0 was lower than that at pH 6.0.
The effect of aging or fibril formation on the interaction of A␤ peptides with gangliosides was investigated at pH 7.0, characteristic of the extracellular space. The interaction of A␤40 with G M1 /PC vesicles decreased with fibril formation (Fig. 7C). A␤42 was unable to induce dye release from G M1 /PC vesicles even after fibril formation (Fig. 7D). These data suggest that at pH 7.0, A␤40 can only disrupt ganglioside vesicles when it is in the ␣/␤ conformation and not when it is present in a ␤-structured fibril state. DISCUSSION The experiments in this study were designed to simulate potential A␤-lipid interactions that could occur after A␤ is cleaved from APP in vivo. Our results can be summarized by the following thermodynamic model.
U is A␤40/42 in a soluble unstructured state; ␣/␤ is the novel ␣/␤ structure of A␤40/42 observed in the presence of gangliosides; and ␤-fibril is the fibrillar ␤-structured state. While the ␣/␤ state of A␤40 has weak disruptive activity toward ganglioside membranes, the ␣/␤ state of A␤42 is inert. At pH 7.0, the ␤-fibril state does not disrupt ganglioside membranes but does have weak disruptive activity toward acidic phospholipid membranes. At pH 6.0, the ␤-fibril state possesses strong disruptive activity toward acidic phospholipid and ganglioside membranes, and this activity approaches that of the cytotoxic peptide, mellitin. This is the simplest model consistent with all of our data; however, more complicated models could also apply. Superimposing this model on proposed pathways of in vivo A␤ production (7, 29) reveals certain implications for neurotoxicity in AD.
In vivo production of A␤40/42 is believed to occur through sequential cleavage of APP by ␤and ␥-secretases (7). While ␤-secretase activity appears to localize to endosomal vesicles, ␥-secretase activity is hypothesized to be restricted to regions near the external plasma membrane. Because of the spatial separation of ␤and ␥-secretase activities, A␤ generation may be concomitant with secretion, and as a consequence A␤ does not accumulate intracellularly (7). However, intracellular accumulation of A␤ has been observed with the human neuronal cell line, NT2N (30), and in COS cells transfected with APP cDNA bearing the Swedish mutation (29). The pathway for intracellular accumulation of A␤ appears to be distinct from that for A␤ secretion. Martin and colleagues (29) suggest that while A␤ is usually produced through the secretory pathway, under certain conditions, such as the presence of the Swedish mutation, the intracellular pathway is up-regulated.
The intracellular production of A␤ occurs in the slightly acidic environments of the endosomes (pH 5-6.5) and possibly also in lysosomes (pH 4 -5). If we assume that A␤ is present at micromolar or submicromolar concentrations inside endosomes/lysosomes, then our data suggest that it is initially soluble and unstructured; however, when it encounters acidic lipids (either phospholipids or gangliosides), it disrupts the bilayer in a ␤-structured conformation. Since this membrane disruptive activity of A␤40/42, at pH 6.0, is comparable to that of mellitin, we suggest that membrane disruption, in endosomal/lysosomal membranes by intravesicular A␤, may contribute to neurotoxicity by hampering protein sorting and degradation, as well as by interfering with vesicular transport. There are several other indications that endosomal/lysosomal activity is affected in AD. For example, there is a 2-8-fold increase in the number of endosomes and lysosomes in all areas of the brain that are at risk for developing AD neuropathology (31). Another indication of endosomal/lysosomal involvement comes from studies of familial AD. Most familial AD cases are caused by mutations in the presenilin 1 and 2 genes (32, 33). The presenilins are localized to the endoplasmic reticulum and Golgi apparatus of neuronal cells (34), and they are homologous to two Caenorhabditis elegans proteins that function in intracellular protein trafficking (32). The current hypothesis is that mutations in the presenilins affect the sorting of APP and ultimately increase the production of A␤42.
In the A␤-secretory pathway, A␤ production is concomitant with secretion into the neutral pH environment of the extracellular space (7). At neutral pH, fibril formation is less efficient (35) and requires the presence of nucleation seeds (26). Our experiments, at micromolar A␤ concentration, indicate that if A␤ is incubated in the presence of nucleation seeds, then fibril formation will ensue at neutral pH, and once formed these fibrils are capable of disrupting membranes containing acidic phospholipids, albeit to a lesser extent than at pH 6.0. Extrapolating these results to the in vivo situation, we suggest that if A␤ molecules in the extracellular environment are nucleated to form fibrils, then these fibrils can disrupt neuronal membranes that contain acidic phospholipids. This disrupting activity could account, at least in part, for the neurotoxic effects of A␤ observed in cell culture studies (11,14,15).
The formation of A␤ fibrils and their deposition in senile plaques is a manifestation of AD. In normal individuals, secreted A␤ molecules have been found in the cerebral spinal fluid (23), and histological examination of normal aged brains has revealed the presence of A␤ in diffuse amyloid (36). Recently, Yanagisawa and colleagues (19) isolated A␤42 bound to gangliosides and found that brain tissue with increased amounts of diffuse amyloid (e.g. from normal aging, Down's Syndrome, and AD) also contained increased amounts of ganglioside-bound A␤42. From their findings, they proposed that at least some of the A␤ in diffuse amyloid is bound to G M1 or its related member. Our results demonstrate that when A␤40/42 binds to gangliosides at neutral pH, it adopts a unique ␣/␤ conformation (Figs. 3 and 4) which is different from the ␤-sheet structure of A␤ fibrils isolated from senile plaques (37). Our results demonstrate that A␤40 was able to induce dye leakage of G M1 /PC vesicles, whereas A␤42 could not. This suggests that while A␤40 is able to penetrate the lipid bilayer in the ␣/␤ conformation, A␤42 adopts the ␣/␤ conformation and resides on the bilayer surface. One interesting possibility is that the ability of A␤40 to penetrate the bilayer (via interactions with gangliosides) represents a mechanism by which A␤40 is cleared from the extracellular space. Such clearance should not occur with A␤42 which is sequestered on the cell surface. A␤42 has the novel ␣/␤ conformation when associated with gangliosides; this novel conformation may be the key to the histological differences seen between diffuse and senile plaques (1). It may also account for the higher concentrations of A␤42 present in diffuse and senile plaques of normal aged (36), AD (2), and Down's Syndrome (38).
Cell culture studies have previously shown that A␤25-35 is neurotoxic (39,40), whereas our data demonstrate that A␤25-35 is unable to induce dye leakage from all lipid vesicles used. Pike and colleagues (40) have shown that neurotoxicity was dependent on both peptide aggregation and ␤-sheet formation. The inconsistency between our results and those previously reported may be explained by the differences in peptide concentration and structure. We have used at least 20-fold lower peptide concentration than those previously reported, and at our peptide concentrations, A␤25-35 adopts a random structure, even in the presence of acidic lipids (data not shown). The lack of ␤-sheet induction may explain the inability of A␤25-35 to disrupt lipid vesicles in our system.
In summary, we have demonstrated that A␤40/42 disrupts acidic lipid membranes, and this disruption is greater at pH 6.0 than at pH 7.0. Because membrane disruption is dramatic at the endosomal pH of 6.0, we suggest that A␤40/42 may cause neurotoxicity by disrupting endosomal membranes. This disruption could occur immediately following intracellular production of A␤. Although membrane disruption is less in more neutral pH environments, such as those in the extracellular space, it is still above background. This suggests that disruption of the plasma membrane by extracellular A␤ may account, at least in part, for the observed A␤ neurotoxicity in cell culture studies. At pH 7.0, gangliosides induce A␤40/42 to adopt a novel ␣/␤ conformation. We speculate that gangliosides could sequester A␤ and thereby prevent ␤-structured fibril formation; alternatively, gangliosides may be involved in normal A␤ functioning and/or clearance. These issues are presently being addressed in the laboratory.