Membrane and surface interactions of Alzheimer’s Aβ peptide – insights into the mechanism of cytotoxicity


L. C. Serpell, School of Life Sciences, University of Sussex, Falmer, East Sussex BN1 9QG, UK
Fax: +44 (0)1273 678433
Tel: +44 (0)1273 877363
E-mail:; T. Williams, Department of Physics, Drexel University, 12-908, 3141 Chestnut Street, Philadelphia, PA 19104, USA
Fax: +1 215 895 5934
Tel: +1 215 895 1989


Alzheimer’s disease is the most common form of dementia and its pathological hallmarks include the loss of neurones through cell death, as well as the accumulation of amyloid fibres in the form of extracellular neuritic plaques. Amyloid fibrils are composed of the amyloid-β peptide (Aβ), which is known to assemble to form ‘toxic’ oligomers that may be central to disease pathology. Aβ is produced by cleavage from the amyloid precursor protein within the transmembrane region, and the cleaved peptide may retain some membrane affinity. It has been shown that Aβ is capable of specifically binding to phospholipid membranes with a relatively high affinity, and that modulation of the composition of the membrane can alter both membrane–amyloid interactions and toxicity. Various biomimetic membrane models have been used (e.g. lipid vesicles in solution and tethered lipid bilayers) to examine the binding and interactions between Aβ and the membrane surfaces, as well as the resulting permeation. Oligomeric Aβ has been observed to bind more avidly to membranes and cause greater permeation than fibrillar Aβ. We review some of the recent advances in studying Aβ–membrane interactions and discuss their implications with respect to understanding the causes of Alzheimer’s disease.


amyloid-β peptide


amyloid-β peptide1-40


amyloid-β peptide1-42


Alzheimer’s disease


atomic force microscopy


amyloid precursor protein


dimethyl sulfoxide




monosialoganglioside GM1




human islet amyloid polypeptide


large unilamellar vesicles


N-methyl d-aspartate






cellular, non-infectious form of prion protein.

Amyloid β assembly and fibrillization

Amyloid-β peptide (Aβ) accumulation in the brain is one of the pathological hallmarks of Alzheimer’s disease (AD). Aβ assembles to form amyloid fibrils that deposit in amyloid plaques in the neuropil [1]. However, the deposition of fibrillar Aβ in the brain does not necessarily correlate with the severity or progression of AD [2]. It has been suggested that Aβ exerts some of its deleterious effects on cells earlier on during the peptide self-assembly, which may trigger a cascade of processes [3]. Aβ is a secreted peptide cleaved from the transmembrane protein, amyloid precursor protein (APP), whereby cleavage at the C- and N-termini by γ- and β-secretase, respectively, results in the release of the Aβ peptide [4]. The predominant forms of Aβ peptide are those with 40 or 42 residues and the relative ratios of these forms are altered in Alzheimer’s patients [5]. It is well established that the 42-residue form (Aβ42) generally forms fibrils more quickly than the 40-residue form (Aβ40) and this may be as a result of the additional hydrophobic isoleucine and alanine at the C-terminus. Substitution of these two residues with hydrophilic amino acids results in a decrease in aggregation kinetics [6]. Two hydrophobic regions of Aβ42 at residues 17–21 and 31–42 are considered to be important for fibril structure [7]. Certain residues also play a key role in aggregation because substitution of the phenylalanine20 residue for the hydrophilic glutamic acid residue in the Aβ sequence reduces the aggregation propensity [8] and the toxic effect of the peptide [9]. The assembly may be driven in part by burial of the hydrophobic regions of the Aβ peptide.

The fibrillogenesis of Aβ is assumed to occur through various pathways and, for Aβ42, it has been suggested that pentamer or hexamer paranuclei (Fig. 1A,B) form the basic subunits of the protofibril (Fig. 1C), leading to a beaded chain-like structure as a result of self-association of the paraneuclei [10]. Oligomer size distributions have been determined both experimentally and using computer simulations, with both yielding similar frequency distributions. Initially, computer simulations show that both Aβ40 and Aβ42 have similar size distributions peaking at monomers. As assembly progresses, the mean occurrence probability for Aβ40 peaks at dimers and monotonically decreases. Whereas, Aβ42 shows a greater frequency distribution centred around trimers, followed by a significant decrease in tetramers and another peak at pentamers, which monotonically decreases after pentamers [11]. The computer simulations show that Aβ40 oligomers form a more compact confirmation compared to Aβ42 oligomers because of the additional conformational freedom associated with the additional isoleucine and alanine amino acids [11]. Photo-induced cross-linking of unmodified proteins, covalently stabilizes the interacting polypeptide chains without any pre facto chemical modifications, and also has been used experimentally to show the oligomer distributions of Aβ40 and Aβ42. Bitan et al. [12] reported oligomer frequency distributions similar to those found in simulations and concluded that Aβ40 and Aβ42 oligomerize through distinct pathways. Electron microscopy reveals a difference in the rate of assembly of Aβ40 and Aβ42 peptides through oligomers, to protofibrils and subsequently to fibrils (Fig. 2A–F). Structural investigations of small ‘neurotoxic’ Aβ40 assemblies have indicated that they contain a high level of β-sheet conformation, similar to fibrils [13].

Figure 1.

 Discrete molecular dynamics simulations using a four-bead protein model [88]. (A) Aβ42 monomer, (B) hexamer and (C) protofibril-like assembly comprising 28 Aβ42 peptides. N-terminal Asp1 is colour-coded red to highlight the distribution of N-termini in the assembly. Figure adapted from [88], with thanks to Dr Brigita Urbanc.

Figure 2.

 Transmission electron micrographs showing the growth of statically incubated amyloid fibrils. Fibrillogenesis of 100 μm Aβ42 (A–C) and Aβ40 (D–F) with time, from freshly dissolved (A, D) to 24 h of incubation (B, E) and, finally, after 72 h of incubation (C, F). Aβ42 forms long straight fibrils more rapidly than Aβ40 at the same starting concentration of peptide. Peptides were incubated at room temperature without agitation. Comparison of fibrils grown with and without lipids: (G) fibrils formed by 100 μm Aβ42 alone after 72 h of incubation; (H) 1 mg·mL−1 LUVs alone; and (I) freshly dissolved Aβ42 (10 mm) incubated with 1 mg·mL−1 LUVs for 72 h. The images show that Aβ42 assembles in the presence of LUVs to form long straight amyloid-like fibrils that appear to associate with the membranes. The LUVs remain intact despite the observed leakage of self-quenching dye induced by freshly dissolved Aβ42 [31].

Amyloid fibrils can be formed in vitro from a broad range of proteins and peptides, and these fibrils share a core structure consisting of a cross-β architecture [14,15]. It has been suggested that fibrillization to form amyloid could be a common feature of all peptides and proteins [16] because, under certain denaturing conditions, typically soluble proteins such as insulin can form amyloid-like fibres [17]. Inter- and intramolecular forces have been shown to influence fibrillization and the assembly of amyloid fibres, as well as act to stabilize the fibrils. The hydrophobicity and net charge, as well as a sequence propensity to form secondary structures, have been shown to modulate amyloidogenicity [18,19]. Short peptides are able to form amyloid-like fibrils in vitro and provide an ideal model system for structural studies. X-ray fibre diffraction of amyloid fibrils formed by a central region of the Aβ peptide (11–25) revealed a cross-β arrangement of extended β-strands [20]. Crystal structures of amyloid-like fibrils formed by the peptides with the sequences GNNQQNY and NNQQNY show that the cross-β spine formed in the amyloid fibres consists of a pair of β-sheets, whereby the residue side chains interdigitate to form a steric zipper along the length of the fibre [21]. Structural studies of full-length (40 or 42) residue Aβ have been possible using hydrogen/deuterium exchange [7], solid-state NMR [22] and electron microscopy [23]. It has been established that Aβ assembles to form a parallel, in-register structure in which Aβ forms two β-strands connected by a β-bend that stack up via hydrogen bonding to form a pair of β-sheets [7,24]. Electron microscopy shows that the fibrils are composed of several ‘protofilaments’, which twist around one another to form the mature fibril [25] (Fig. 2). Many of the inter- and intramolecular interactions involved in the assembly of amyloidogenic peptides may also be involved in the interactions between peptides and surfaces, including membranes [26].

The preparation of amyloidogenic peptides, especially Aβ, can result in experimentally observed differences. It has been shown that the mode of preparation can significantly affect the secondary structure of the peptide, and therefore may alter its fibrillization and assembly characteristics. The pretreatment of amyloidogenic peptides has been investigated extensively, and harsh solvents and treatment methods have been employed to render the peptide homogenous, disaggregated and unstructured. The addition of trifluoroethanol and, more recently, hexafluoroisopropanol (HFIP), has been used to pretreat Aβ, aiming to dissagregate the peptide and render it α-helical; however, it has also been shown to promote other intramolecular hydrogen bonds, including turns and β-hairpins [27,28]. HFIP contamination has been implicated in inducing membrane leakage and cell toxicity [29]. Dimethylsulfoxide (DMSO) is commonly used in the preparation of Aβ and other amyloid-forming peptides. However, dissolution of β2-microglobulin in the polar solvent, DMSO, has been shown to cause destruction of the hydrogen bond networks in amyloid fibril aggregates by acting as a strong proton acceptor [30]. Therefore, recent protocols have included a critical step that removes residual solvents [31].

Aβ interactions with solid surfaces

Solid surfaces influence the assembly of amyloid-forming peptides, and the adsorption of the peptide to the surface can promote fibrillization. Hydrophobicity, surface charge and surface roughness have all been shown to play a role in influencing fibre assembly because the surface may act to increase the local concentration and modulate the conformation of the peptide, leading to alterations in the propensity for association [32]. Hydrophobic surfaces such as negatively-charged Teflon have been used to mimic the nonpolar plane of membranes. At physiological pH, Teflon and Aβ are both negatively charged, and therefore electrostatic interactions would suggest partial repulsion between the surface and the peptide. However, at pH 7, it has been shown that Aβ40 adsorbs to the nonpolar substrate [33] as a result of protein dehydration effects contributing to the adsorption of peptides to hydrophobic surfaces [26]. The adsorption of Aβ40 and Aβ42 to hydrophobic Teflon particles at pH 7 also promotes aggregation and fibrillization of the peptide [32], and adsorption of Aβ42 to hydrophobic graphite leads to nucleation-controlled growth of fibrils [34]. Conversely, the adsorption of Aβ to hydrophilic silica, which has been used to mimic the polar, charged membrane surface, only occurs when the peptide is positively charged at pH 4 and 7 [35]. This suggests that Aβ adsorption to hydrophilic surfaces is mainly driven by electrostatic interactions. The adsorption of Aβ42 on hydrophilic mica occurred quickly; however, the aggregation was slow and gradual coalescence was observed [34]. Similar behaviours between surfaces and other amyloid-forming peptides have been observed. Fibrillization of the recombinant amyloidogenic light chain variable domain was observed on negatively-charged mica, although no fibres were apparent on positively-charged Teflon despite adsorption of the peptide on both surfaces, suggesting the significant involvement of electrostatic interactions in assembly [36].

The influence of surfaces on amyloid fibril formation can also affect fibril morphology. The fibrillization of Aβ on hydrophilic mica surfaces lead to the formation of particulate, pseudomicellar aggregates, whereas, at higher concentrations, linear protofibrillar assemblies were formed [34]. Moreover, Aβ assembly on highly-ordered pyrolytic graphite results in uniform, elongated sheets, with fibre formation being observed in three directions orientated 120o to each other, which is suggested to result from hydrophobic interactions that maximize contact between the carbons in the graphite and the hydrophobic residues within the Aβ chain [34]. The assembly of amyloid fibrils in solution and on surfaces is reviewed in more detail elsewhere [37].

Aβ interactions with lipids

The proteolytic cleavage of the APP from its transmembrane location results in the release of the Aβ peptide, and therefore the soluble peptide may retain affinity for the cellular membrane or certain features of the membrane. Phospholipids are composed of two hydrophobic fatty acids bound to carbon atoms in the glycerol, which in turn is joined to the hydrophobic polar headgroup via a negatively-charged phosphate group. Therefore, phospholipids are amphipathic. The phospholipid bilayer of cellular membranes provides an extensive surface for amyloid interactions and comprises one of the primary cellular structures that Aβ comes into contact with. Plasma membranes have been suggested as a possible target for the cytotoxicity associated with Aβ and therefore in the pathology of AD. The interactions between lipid bilayers and Aβ have been studied using a variety of biophysical techniques, including CD, fluorescence spectroscopy, surface plasmon resonance and atomic force microscopy (AFM). Biomimetic unilamellar vesicles provide a simple system for studying amyloid–membrane interactions because bilayer composition can be stringently regulated, and it is possible to encapsulate solute molecules within the aqueous space [31].

Simple, one-lipid species vesicles prepared from soybean phosphatidylcholine have been used to examine the effect of lipid-induced Aβ aggregation using an absorbance assay. It was demonstrated that, in the absence of lipid vesicles, the aggregation of Aβ40 followed typical lag-growth kinetics. In the presence of neutral phosphatidylcholine vesicles, the lag time was delayed and the half-aggregation time increased by 30%, dependent on lipid concentration [38]. Nucleation and elongation were also influenced by the presence of neutral lipid surfaces, and appeared to decrease when lipids were incubated with Aβ40. Secondary structural changes in Aβ conformation as a consequence of lipids have been demonstrated by CD [39]. When solubilized in the hydrophobic solvent, trifluoroethanol, Aβ40 and Aβ42 show characteristic α-helical structures and, upon dissolution in sodium phosphate, this resulted in the peptide becoming unstructured [39]. The presence of various lipids, including egg yolk phosphatidylglycerol, bovine brain phosphatidylserine and phosphatidylethanolamine, resulted in a strong 218 nm CD minima indicative of β-sheet structure [39]. It was suggested that the head-group charge of the phospholipids contributes to the association between Aβ and the membrane via electrostatic interactions. The affinity of DMSO-solubilized Aβ40 to 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) was found to be weaker than for 1-palmitoyl-2-oleoyl-sn-glycero-3-phospho-rac-(1-glycerol) (POPG) [40], supporting the view that the headgroups mediate binding.

Moreover, the mass adsorption between Aβ40 and POPG was between 50–100% greater than Aβ40 mass adsorption to POPC membranes and it appeared that, although POPC binding did not result in aggregation of the Aβ peptide, POPG liposomes markedly increased aggregation. Interactions between fresh Aβ40 and lipid vesicles of various compositions have shown that the surface charge and hydrophobicity of the membranes can modulate the binding of the Aβ to the membrane surface. It was reported that Aβ40 preferentially bound to negatively-charged phosphatidylglycerol membranes and composite membranes containing negatively-charged lipids compared to neutrally-charged membranes. The adsorption of 1 day fibrillized Aβ40 showed slight variations in the binding kinetics compared to fresh Aβ. The affinity between fibrillar Aβ and negatively-charged membranes is lower than the affinity between fibrillar Aβ and neutrally-charged membranes. The electrostatic forces were found to be more significant between fresh Aβ and membranes, although hydrophobic forces were considered to be more important between fibrillar Aβ and membranes [41].

AFM studies have been used to monitor the assembly of Aβ42 on lipid bilayers composed of brain total extract, which provides a physiologically relevant ratio of acidic and neutral phospholipids. In such a study, the peptide was prepared in trifluoroacetic acid followed by incubation in trifluoroethanol. Small aggregates, ∼ 5–7 nm in height, initially associated with the membrane and, after 7 h of in situ incubation, small aggregates ∼ 5–15 nm in height were observed [42]. Incubation of Aβ42 with the bilayers revealed very little membrane disruption. Interestingly, this preparation of Aβ42 did not appear to ‘grow’ on the membrane bilayer [42]. To investigate the possibility that Aβ42 is able to form ion channels, Aβ42 was incorporated into planar bilayers. AFM revealed multimeric complexes protruding above the lipid bilayer. The individual channels showed varying numbers of Aβ subunits, ranging from a two subunit arrangement to rectangular four subunit structures and hexagonal six subunit structures [43]. Electrophysiological records supported the view that the Aβ was creating channels that allowed the passage of calcium ions [43]. Therefore, it was suggested that the mechanism of membrane disruption may either be the result of the formation of defects within the lipid bilayer or the formation of membrane-spanning channels. We use the term ‘defects’ here to mean the penetration/permeation of the lipid bilayer by amyloidogenic peptides that results in a noncontinuous membrane surface and the emergence of defects/holes or deformations within the bilayer structure. By contrast to the conclusion that Aβ is able to form pores, Kayed et al. [44] showed that size exclusion chromatography-purified Aβ42 oligomers (eluting at ∼ 90–110 kDa) caused an increased conductance across mixed-lipid bilayers in a concentration-dependent manner. The results were interpreted as showing that conductance was nonselective for different ions and was inconsistent with specific pore formation [44]. Further studies indicated that Aβ42 caused increased conductance by interacting with the membrane surface, possibly by spreading the lipid headgroups apart, leading to membrane thinning, and thus a lowering of the membrane dielectric barrier and increased conductance [45]. Recently, however, these studies have been called into question by the results obtained in a study by Capone et al. [29] who suggested that membrane thinning was a result of HFIP contamination.

Studies using lipid vesicles encapsulating self-quenching fluorescent dyes such as calcein and fluorescein have been used to study the effect of Aβ assembly on membrane integrity. The addition of Aβ42 to 1,2-dimyristoyl-sn-glycero-3-phosphocholine large uni-lamellar lipid vesicles (LUVs) encapsulating calcein demonstrated that early oligomeric soluble Aβ42 causes permeation of the membranes and the release of the encapsulated fluorescent dye [31]. It was also demonstrated that, as the Aβ42 peptide assembles into fibres in solution, the propensity to cause membrane permeation decreases, and mature fibres show a lack of ability to cause permeation. Interestingly, early oligomeric Aβ added to the vesicles elongates to form amyloid fibrils that appear to be associated with the membranes on electron microscopy [31] (Fig. 2). This is similar to the damage to 1,2-dioleoyl-sn-glycerol-3-phosphocholine (DOPC)/1,2-dioleoyl-sn-glycerol-3-phosphoserine calcein-loaded vesicles caused by human islet amyloid polypeptide (hIAPP) [46]. This was suggested to be caused by mechanical disruption of the lipid membrane as a result of the associated growth of amyloid-like fibrils [46]. Confocal microscopy has been used to observe the release of different sized Alexa Fluor dyes from the aqueous space of giant unilamellar vesicles as a result of Aβ42-induced permeation. It was reported that the smaller Alexa546 (Mr ∼1300) dye leaked from the membranes before the diffusion of the larger Alexa488 (Mr 10 000) dye, whereas the overall shape of the vesicles was maintained [47]. It was suggested that Aβ is able to influence the cohesion of the components of the membranes [47]. Anisotropy has been used to monitor the effects of Aβ39 and Aβ40 on the membrane fluidity of POPC and POPG membranes containing 1,6-diphenyl-1,3,5-hexatriene fluorescent dye. Freshly-prepared Aβ39 and Aβ40 did not have any observable effects to membrane fluidity but, upon oligomerization of the peptides, membrane fluidity was decreased in a time- and concentration-dependent manner. This affect was more striking for those peptides assembled at pH 6 rather than neutral pH [48]. This is assumed to be the result of a conformational difference between aggregates of Aβ formed at the two different pHs that, in turn, influence the exposure of hydrophobic regions and association with the membrane [48].

Aβ interactions with sterols

Cholesterol is a vital component of eukaryotic cell membranes, and influences membrane fluidity, permeability and dielectric properties. Cholesterol causes the immobilization of the first few hydrocarbon groups of the phospholipid molecules, making the lipid bilayer less viscoelastic and therefore decreaseing the membrane permeability to small water-soluble molecules. Cholesterol also prevents the crystallization of the hydrocarbons and prevents phase shifts within the membrane. In AD, the cholesterol content of certain regions of the brain can be markedly different from the same regions in nondemented brains. In the grey matter of the superior temporal gyrus, the cholesterol to phospholipid mole ratio in AD brains is 0.46 ± 0.08 whereas the ratio in nondemented brains is 0.66 ± 0.05; however, the cholesterol content in the cerebellum is not significantly different [49]. This reduction of ∼ 33% between the cholesterol content of the temporal gyrus of AD brains and nondemented brains could significantly affect the fluidity of the neuronal membranes and render them more susceptible to Aβ-induced permeation. Moreover, the decrease in cholesterol/phospholipid ratio in AD brains may affect the cleavage of APP and cause an elevation in generated Aβ [50]. A 4 Å decrease in the membrane bilayer width (D-space) could ensure that the cleavage site is more accessible to secretase enzymes, which otherwise may be inaccessible [51]. Changes in cholesterol contents as a result of ageing may also play a role in reduced amyloid degradation because the amyloid degradation pathway may involve the activation of plasmin in cholesterol-rich domains and be inhibited when cholesterol levels are reduced [52]. However, treatment with statins, which lower neuronal cholesterol, has been shown to decrease the amount of Aβ secreted by neurons. This is probably because both γ- and β-secretases are found in cholesterol-rich domains within the membrane and therefore a lowering of cholesterol reduces secretase activity [53]. The cholesterol content of membranes may play a role in modulating Aβ penetration because > 20% w/w cholesterol induces a membrane transition from a fluid-disordered to fluid-ordered phase, and Aβ25–35 is unable to intercalate into the membrane bilayer [54,55]. However, using monolayer surface pressure measurements, it was shown that Aβ40 spontaneously inserts into monolayers containing a 30 mol% cholesterol to phospholipid ratio and, in this context, Aβ adopts an α-helical structure. Below this molar ratio, Aβ40 prefers to associate with the membrane surface region with a greater β-sheet content [56]. The involvement of cholesterol in Aβ secretion and processing appears to be very complex and its involvement in AD has been reviewed [52–56].

The cholesterol to phospholipid ratio has been reported to affect the extent of high-affinity Aβ binding to synthetic lipid membranes because pure phospholipid mixtures showed very little peptide binding. It was suggested that this increase in peptide-membrane affinity was a result of the involvement of cholesterol in altering membrane fluidity and structure [57]. Oligomeric, but not monomeric, Aβ42 was shown to insert into POPC/cholesterol membranes. Oligomer insertion was also found to occur in negatively-charged monolayers but not in the absence of cholesterol [58]. By contrast, cell culture assays using PC-12 and SH-SY5Y cells showed an inverse relationship between cholesterol content and Aβ40 surface binding, where cholesterol-depleted cells demonstrated higher Aβ-cell surface binding [59]. It was suggested that this increase in Aβ binding may increase the internalization of Aβ to a greater extent because the decreased membrane cholesterol content affects the fluidity and permabilization of membranes [60]. The inclusion of < 30% cholesterol in POPC membranes gave rise to channel activity induced by the addition of Aβ40, whereas no channel activity was observed in POPC only membranes [61]. It was suggested that cholesterol-rich membranes prevented the fibrillization of Aβ40 by increasing the incorporation of the peptide within the membranes.

Aβ interactions with membrane receptors

The complex nature of biological membranes includes various membrane receptors such as glycolipid and glutamate receptors. Both ionotropic glutamate receptors such as N-methyl d-aspartate (NMDA) receptors and metabotropic glutamate receptors such as metabotropic glutamate receptors have been implicated in the alteration of synaptic activity. The binding of Aβ oligomers to synaptic plasma membranes alters the diffusion of metabotropic glutamate receptors and causes abnormal Ca2+ mobilization [62]. Moreover, the administration of Aβ42 to cortical neurons was shown to reduce the density of synaptic NMDA receptors on the cortical neuronal membranes, either by promotion of endocytosis of the cell surface receptors or prevention of the delivery of the NMDA receptors to the neuronal membranes, which has been shown to reduce both memory and learning [63]. Other receptors reported to bind to Aβ oligomers include the cellular, non-infectious form of prion protein (PrPc). At nanomolar concentrations, Aβ42 was able to bind to the PrPc receptor of mice engineered to express the PrPc protein, and caused significant synaptic dysfunction [64]. Using surface plasmon resonance, synthetic Aβ42 oligomers were shown to bind with a high affinity (Kd = 70 nm) to recombinant human prion protein, whereas Aβ42 monomers and mature fibres did not bind to human prion protein receptors [65]. Therefore, the existence of one specific receptor for Aβ membrane binding may not be a realistic suggestion; Aβ may possess varying degrees of affinity to a range of membrane receptors and certain receptors may be more significant in modulating Aβ induced toxicity.

Gangliosides are a group of glycosphingolipids composed of a hydrophilic sialic acid terminal sugar exposed to the external environment and a hydrophobic ceramide moiety that is embedded within the membrane [66]. Gangliosides have been reported to serve a variety of functions, including as cell type-specific markers, as differentiation and developmental markers, and as receptors and mediators of cell adhesion [67], and they comprise 5–10% of the outer membrane leaflet [68]. The affinity between gangliosides and Aβ can vary, with a reported Kd of 1.2 × 10−6 m between Aβ42 and disialoganglioside GD1a and 7.7 × 10−7 m between Aβ42 and disialoganglioside GT1b, with the highest affinity between Aβ42 and monosialoganglioside GM1 (GM1) with a Kd of 5.2 × 10−7 m [69]. The expression of gangliosides within the membrane bilayer induces structural transitions in Aβ. Upon dilution of Aβ in NaCl/Pi, the peptide initially adopts a random structure but, subsequent to the interaction with GM1, the Aβ transitions into a α/β conformation at pH 7 and into a β conformation at pH 6 [70]. In the same study, the significance of the ganglioside moieties was studied. It was shown that neither the ceramide, nor sialic acid moieties could induce the structural transitions of Aβ40 and Aβ42 alone, implying that it is critical for sialic acid to be associated with the carbohydrate backbone for Aβ structural transitions to occur [70]. A small structural transition of Aβ42 at pH 6 was observed with ceramide-containing lipid vesicles, and it was suggested that the presence of a hydrogen acceptor in the form of the amide carbonyl and a hydrogen donor in the form of a hydroxyl group on the ceramide was responsible for this structural transition. Fibrillization of Aβ40 in the presence and absence of GM1 in the membrane showed that GM1 causes increased Aβ fibrillization. Additionally, GM1 decreased the fibrillization lag of Aβ40 from 5–6 days to 1 day, with an observed structural transition from a random conformation to β-sheet within 3 h [71]. Electrostatic interactions may play a pivotal role in mediating Aβ–GM1 interactions (Aβ is slightly positively charged and the GM1 headgroup is negatively charged); the pressure exerted on the membrane as a result of Aβ40 inserting into GM1-containing membranes in a low ionic-strength aqueous environment at pH 5.5 is greater than when the ionic strength surrounding the membrane is increased [68]. Increasing the pH to 7.2 (where the Aβ is negatively charged) leads to a decrease in Aβ insertion pressure because the interaction between the negatively-charged Aβ and negatively-charged GM1 becomes repulsive.

Gangliosides have also been shown to modulate the permeation of biological membranes and, using dye release assays, the inclusion of GM1 within the bilayers resulted in a significantly greater permeation by Aβ compared to membranes without GM1 [31]. Exclusion of GM1 from lipid vesicles containing calcein showed a 57% reduction in Aβ42-induced membrane permeation compared to when GM1 was present within the vesicle membrane [31]. This increase in permeation of GM1-containing membranes was attributed to hydrophobic interactions between the solvent-exposed aromatic residue stacks on the glycolipid sugar rings of GM1. This stacking interaction is driven by the net positive charge of the sugar ring in close proximity to the π-electron cloud of the amino acid aromatic ring, where the polar moiety of the GM1 provides a complementary surface for the polar amino acids of the Aβ to form hydrogen bonds [72]. The permeation of calcein-loaded LUVs has also been shown to be Aβ concentration- and pH-dependent, where increasing concentrations of Aβ40 resulted in a monotonic increase in the permeation of dye-filled LUVs [70,73] and a 42% reduction in permeation of pH 7 vesicles compared to pH 6 vesicles [73].

Aβ–membrane interactions and permeation: pore formation or detergent effects

The mechanism associated with Aβ cytotoxicity has not been determined definitively, although the involvement of the membrane surface has been implicated as a possible source of Aβ-induced toxicity. The solvent exposure of the N-terminal region of Aβ has been suggested to play a critical role in mediating toxicity. Assembly of Aβ42 in the presence of certain C-terminal fragments causes the Aβ42 N-terminus to be less solvent exposed and significantly reduces cell toxicity, and it was suggested that the N-terminus may play a critical role in mediating Aβ-cellular interactions [74]. The membrane may act as an extensive surface for Aβ to aggregate and fibrillize, and membranes serve as a primary barrier to Aβ entry into the cytosol of cells. Therefore, it is not unexpected that Aβ may preferentially bind to certain domains or receptors within these membranes. The pre-processing location of APP from the transmembrane region also suggests that Aβ may retain certain affinities for the membrane. Therefore, a potential cytotoxic mechanism involving the cytoplasmic membranes maybe justified. Aβ42 has been shown to tightly insert into membranes, with a small portion being peripherally associated with lysosomal membranes, leading to destabilization of lysosomal membranes and the induction of toxicity [75,76]. The association was shown to be pH-dependent and neutralization of the lysosomal pH in differentiated PC12 cells decreased Aβ membrane insertion [76]. The amphipathic nature of amyloid oligomers has been suggested to contribute to their capacity to penetrate and insert into membranes, coat or lie on the surface of the membranes, or potentially act as cell-penetrating peptides [77].

Three structurally divergent modes of membrane-mediated toxicity have been proposed for Aβ, which include carpeting of the peptide on one leaflet of the membrane surface, resulting in an asymmetric pressure between the two leaflets and the leakage of small molecules [78]. This model is believed to be of minor physiological relevance for amyloid diseases because it has been demonstrated to cause similar membrane damage for both hIAPP and non-amyloidogenic mouse IAPP [46]. The carpet model was also proposed to explain the exponential leakage kinetics and absence of a lag phase in hIAPP and mouse IAPP-induced LUV permeation [46].

The formation of stable pores and ion channels is the second model proposed for amyloid-induced toxicity. The disruption of Ca2+ homeostasis has been recognized as a potential mechanism associated with AD, and was shown to be involved in the amyloid cascade hypothesis, where elevated Ca2+ was suggested to be a consequence of both tau-phosphorylation and cell death [79]. The formation of Ca2+ channels in lipid bilayers was proposed in AD cytotoxicity because the incorporation of Aβ40 into planar phosphatidylserine bilayers formed channels that generated linear current–voltage relationships in symmetrical solutions [80]. The channels were directly observed in planar membranes by AFM, which consisted of an 8–12 nm doughnut-shaped structure with a 1–2 nm internal pore cavity that protrudes ∼ 1 nm above the embedded bilayer surface [43,81]. Computational modelling of Aβ42 insertion into a lipid bilayer that approximately matches the thickness of DOPC bilayers showed that Aβ42 octamers separate into distinct tetrameric units. By achieving this structure, the tetrameric units stabilize the membrane-inserted octamer, and lead to the conclusion that the pore formed by the octamer might consist of tetrameric and hexameric β-sheet subunits, which were considered to be consistent with reported AFM studies [82]. The transmembrane pore has been proposed to form between six hexamers, which span the bilayer and merge to form a more stable 36-stranded β-barrel arrangement. This model was favoured because it was proposed that the parallel β-barrels formed by the N-terminal segments of the peptide form the lining of the pores and could account for the cationic selectivity observed for metal ions such as Zn2+ [83].

The third proposed model is based on the detergent-like effects of amyloid-forming peptides on lipid membranes. This mechanism of permeation is proposed to occur through the membrane association of the amyloid-forming peptides in the form of micelle-like structures. Membrane permeation occurs at high local concentrations of the peptide on the membrane surface, either after the surface is covered with peptide monomers or oligomers, or through the association between membrane-bound amyloid [84]. The initial interaction is electrostatically driven, where the peptide preferentially binds to either the phospholipid headgroup or receptors on the membrane surface, which is followed by alignment of the peptide so that the hydrophilic surface faces the phospholipid head groups. The peptide orientates so that the hydrophobic residues reside towards the hydrophobic core of the membrane, and this is followed by disintegration of the membrane by disruption of the bilayer curvature [84]. The detergent effect results from the surfactant-like properties associated with the amphiphilic peptide causing a reduction in membrane surface tension, where it forms a hole by the removal of lipid from the bilayer, either in the outer leaflet, which results in membrane thinning, or both leaflets, which results in holes [78]. hIAPP has been reported to extract lipids from cellular membranes and be incorporated into the forming amyloid deposits; confocal microscopy of rhodamine-labelled giant unilamellar vesicles showed a loss of barrier function of the membranes and the colocalization of IAPP and DOPC/phosphatidylethanolamine lipids using amyloid-specific and lipid-specific dyes [85]. Structural analogies and sequence homologies between Aβ and IAPP have been reported and it was suggested that Aβ may therefore possess an ability to cause lipid extraction of membranes in a similar manner to IAPP [86,87]. The mechanism of toxicity associated with amphiphilic amyloid-forming peptides may not be exclusively related to a single mechanism such as pore formation or detergent-like effects but more likely a collection of these mechanisms (Fig. 3). Each mechanism may be involved in the disruption of biological membranes, either at a particular stage during the assembly of the amyloid-forming peptide or during a particular pathway taken during the peptide fibrillization. As shown with other amphiphilic peptides, the carpet and detergent models may well only occur when the peptide is in its monomeric or small oligomeric state and result in nonspecific permeation of the membranes; the formation of amyloid-induced pores or ion channels may occur as a result of specific receptor-amyloid-induced permeation.

Figure 3.

 Schematic diagram that depicts three possible mechanisms of Aβ-induced membrane damage: carpeting, pore formation and the detergent effect. Adapted with permission from Butterfield and Lashuel [86].

Concluding remarks

The mechanism that leads to cell dysfunction and death caused by amyloidogenic peptides related to disease remains controversial. Effects on organelles and cellular homeostasis mechanisms have been suggested [3]. One way of explaining all of these mechanisms from plasma membrane permeation to mitochondrial dysfunction and lysosomal leakage is the potential effect of oligomers on membranes. Therefore, membrane damage could represent a unifying explanation for all different avenues of the toxic effect.

We have reported on studies that aimed to examine the mechanisms of Aβ-related membrane damage giving weight to the idea that Aβ can disrupt membrane integrity. However, any conclusions made regarding the mechanism of membrane disruption by amyloidogenic peptides, such as Aβ, are hampered by differences in peptide preparation and experimental conditions. It may be that several mechanisms are responsible (Fig. 3) and it remains to be seen which of the proposed mechanisms is most important.