Aβ (amyloid-β peptide) assembles to form amyloid fibres that accumulate in senile plaques associated with AD (Alzheimer's disease). The major constituent, a 42-residue Aβ, has the propensity to assemble and form soluble and potentially cytotoxic oligomers, as well as ordered stable amyloid fibres. It is widely believed that the cytotoxicity is a result of the formation of transient soluble oligomers. This observed toxicity may be associated with the ability of oligomers to associate with and cause permeation of lipid membranes. In the present study, we have investigated the ability of oligomeric and fibrillar Aβ42 to simultaneously associate with and affect the integrity of biomimetic membranes in vitro. Surface plasmon field-enhanced fluorescence spectroscopy reveals that the binding of the freshly dissolved oligomeric 42-residue peptide binds with a two-step association with the lipid bilayer, and causes disruption of the membrane resulting in leakage from vesicles. In contrast, fibrils bind with a 2-fold reduced avidity, and their addition results in approximately 2-fold less fluorophore leakage compared with oligomeric Aβ. Binding of the oligomers may be, in part, mediated by the GM1 ganglioside receptors as there is a 1.8-fold increase in oligomeric Aβ binding and a 2-fold increase in permeation compared with when GM1 is not present. Atomic force microscopy reveals the formation of defects and holes in response to oligomeric Aβ, but not preformed fibrillar Aβ. The results of the present study indicate that significant membrane disruption arises from association of low-molecular-mass Aβ and this may be mediated by mechanical damage to the membranes by Aβ aggregation. This membrane disruption may play a key role in the mechanism of Aβ-related cell toxicity in AD.
- Alzheimer's disease
- amyloid-β peptide
- atomic force microscopy
- GM1 ganglioside
- membrane bilayer
- protein misfolding
- surface plasmon field-enhanced fluorescence spectroscopy
AD (Alzheimer's disease) is the most prevalent cause of dementia, with estimates of between 18 and 30 million sufferers worldwide . AD is characterized by the deposition of intraneuronal neurofibrillary tangles, extracellular amyloid-containing neuritic plaques composed predominantly of fibres formed from Aβ (amyloid-β peptide) and the presence of cerebrovascular amyloid fibres . Aβ is processed from the integral membrane protein APP (amyloid precursor protein) to produce predominantly Aβ40 [Aβ-(1–40)] or Aβ42 [Aβ-(1–42)]. Aβ42 is known to be the more fibrillogenic and toxic form of Aβ, and an increased ratio of Aβ42 to Aβ40 is associated with familial forms of AD . In vitro, Aβ assembles to form small prefibrillar oligomers and, over time, fibrils, and the assembly state, has been shown to modulate the cellular toxicity of the peptide . Aβ–membrane interactions may play a key role in the observed Aβ toxicity associated with AD and assist the entry of Aβ into the cytoplasm of the cell (V. Soura, T. L. Williams and L. C. Serpell, unpublished work). The influence of both Aβ40 and Aβ42 assemblies on model membranes with various compositions have been studied previously [4–8], although the majority of these studies have focused on the less disease-related form Aβ40. These studies have shown that Aβ may cause partial disruption of lipid membranes [5,7–9]. However, significant advances have been made in the last few years to ensure disaggregated solvent-free peptide [10,11]. We have previously shown that the ability of Aβ42 to disrupt the integrity of biomimetic vesicles decreases as the peptide assembles to form mature fibres, and it is therefore clear that the assembly state of the peptide plays an important role in the toxic ability .
The effect of Aβ has been linked to membrane composition, with both cholesterol  and GM1 gangliosides being implicated in mediating the effect of the peptide [14–16]. Gangliosides are glycosphingolipids composed of a hydrophilic sialic acid moiety exposed to the external environment and a hydrophobic membrane-embedded ceramide moiety. Their inclusion in membranes is typically approximately 2 mol% of cellular membranes , but this can vary between 0.5 and 13% (w/w) depending on cell type and during the development of the cell [18–20]. Interactions between GM1 and Aβ are believed to involve hydrophobic and electrostatic interactions, as greater affinity is observed between Aβ and the hydrophobic membrane-embedded ceramide portion of GM1 compared with its affinity with the hydrophilic sialic acid portion . GM1 is also implicated in seeding assembly of the Aβ associated with the membrane  and has been shown to mediate binding of Aβ40 .
We have previously used biomimetic unilamellar vesicles in solution to monitor the effect of Aβ aggregation state on the ability to cause membrane permeation using a calcein release assay . We demonstrated that as Aβ assembles from an oligomeric to fibrillar state, the ability to cause membrane permeation decreases . We have also shown that the removal of GM1 from the bilayer vesicles significantly decreases the ability of Aβ42 to permeate the membranes. In support of this, Niu et al.  and Müller et al.  showed an effect on membrane fluidity by the addition of Aβ, showing that Aβ specifically altered the acyl chain layer of cell membranes while the polar head group layer was much less affected [23,24]. Other amyloidogenic peptides, such as IAPP (islet amyloid polypeptide) have also been shown to affect the integrity of lipid membranes [25,26].
In the present study, we have examined the binding and effect of small Aβ42 assemblies on biomimetic membranes simultaneously for the first time. SPFS (surface plasmon field-enhanced fluorescence spectroscopy) is a technique that combines the ability of SPR (surface plasmon resonance) to monitor adsorption and desorption events at the air/dielectric interface with the ability to resonantly excite fluorescent molecules within the evanescent field extending 200 nm within the dielectric, and this allows us to simultaneously observe both Aβ–membrane interactions and correlate these with permeation events. SPFS has previously been used to monitor the tethering of intact LUVs (large unilamellar vesicles) to a functionalized gold substrate, and the events following the addition of the membrane-lysing protein phospholipase A2  and membrane permeation caused by cholera toxin . Permeation is the term we apply in the present study to mean the penetration of the lipid bilayer by Aβ resulting in a non-continuous membrane surface and the emergence of defects/holes or deformations within the bilayer to allow increased diffusion across the membrane bilayer. Aβ oligomers is the term we apply in the present study to mean freshly dissolved peptide in biological buffer, which is soluble and shows no defined fibrillar features.
In the present study, we tether biotinylated LUVs to a biotinylated thiol SAM (self-assembled monolayer) using streptavidin to covalently tether the biomimetic membranes to the functionalized surface (Supplementary Figure S1 at http://www.BiochemJ.org/bj/439/bj4390067add.htm). We ensure complete construction of the functionalized surface and tethering of the biotinylated LUVs by monitoring with SPR. SPR coupled with SPFS is then used to monitor adsorption of Aβ to the tethered membrane surface, concurrent with monitoring the change in fluorescence intensity as a result of fluorescent dye diffusion through membranes. Analysis of peptide–membrane binding kinetics is performed using a Langmuir binding isotherm to determine the equilibrium dissociation constant and fitting the data to an exponential function to analyse differences between oligomeric and fibrillar Aβ binding. The fluorescence leakage is modelled using a simple diffusion model. Using these techniques, we aim to simultaneously reveal the interactions between Aβ42, the physiologically relevant variant implicated in AD, with permeation of biomimetic membranes. SPFS with tethered LUVs is used to compare the adsorption of oligomeric and fibrillar Aβ42 with the membrane surfaces, while simultaneously monitoring the permeation of the lipid bilayers. We show that binding of the peptide to the membrane is closely followed by membrane permeation. Further to this, we report that the removal of the GM1 ganglioside from the bilayer membranes significantly reduces the binding of the oligomeric Aβ42, and results in decreased permeation of the membranes compared with when GM1 is present. This indicates that GM1 can modulate binding of the peptide to the membrane, as well as mediate an increased Aβ-induced permeation in a concentration-dependent fashion. The fibrillization of the oligomeric peptide and the permeation of membranes were also visualized by AFM (atomic force microscopy), revealing that administration of oligomeric Aβ42 resulted in visible holes within the lipid bilayer, and permeation progresses through both leaflets of the bilayer over time. The present paper reports key advances showing that: (i) oligomeric Aβ binds in a two-step process, whereby the specific adsorption of the peptide precedes permeation of the membranes, and permeation typically begins within the first phase of peptide association, whereas fibrillar Aβ binds in a one-step association; (ii) all pre-disaggregation solvents are carefully removed before addition to membranes ensuring true and reliable binding constant determination; (iii) GM1 exclusion in the membranes reduces the association of Aβ to the membrane surface, but has a much greater effect in modulating Aβ-induced permeation; and (iv) Aβ addition to bilayers can cause large visible defects and holes within the membrane surface, as a possible consequence of altered membrane viscoelasticity.
Aβ42 HFIP (1,1,1,3,3,3-hexafluoro-2-propanol) of >97% purity was purchased from rPeptide. All peptides were used without further purification.
Aβ peptides require treatment to ensure disaggregation and removal of preaggregated species , Solvents including HFIP, DMSO and TFE (trifluroethanol) have been shown to significantly affect amyloid assembly and cause membrane leakage . Aβ42 at 1 mg/ml was solubilized in HFIP >99.0% (Fluka), vortex mixed for 60 s and sonicated in a 50/60 Hz bath sonicator for 60 s. HFIP is initially used to break β-sheet structures and render the peptide α-helical. However, HFIP has also been shown to promote intramolecular hydrogen-bonding networks, so further processing is also needed . Solvent was removed using dry nitrogen and vacuum desiccation for 30 min. The amyloid was re-solubilized at 1 mg/ml with DMSO >99.9% (Sigma–Aldrich), vortex mixed vigorously for 60 s and sonicated for 60 s. DMSO is used as a proton acceptor to destroy the hydrogen-bond network to maintain the peptide in its monomeric state . Peptide (200 μl) in DMSO was added to a 2 ml Hepes (pH 7.4)-equilibrated Zeba™ desalt spin column. Once absorbed into the column resin, a 40 μl stacker of 0.22 μm-filtered water was added to the column. The column was spun in a 4 °C-controlled Mikro 22R centrifuge (Hettich) at 1000 g for 2 min. The eluted peptide was centrifuged in a 4 °C-controlled Eppendorf microcentrifuge at 16000 g for 30 min to remove contaminants and pre-formed fibrillar material. The supernatant was placed in a clean non-stick microcentrifuge tube and stored at 4 °C. The concentration was determined using a molar absorption coefficient of 1490 M−1·cm−1 and the absorbance was measured at a wavelength of 280 nm using an Eppendorf Biophotometer and resulted in peptide concentrations of 100–130 μM. The peptide stock was diluted to a 10 μM working concentration for all experiments. For the preparation of Aβ42 fibres, the oligomeric Aβ42 was statically incubated at 21 °C in a temperature-controlled Eppendorf Thermomixer Comfort and incubated for 7 days. Complete removal of HFIP and DMSO, which has been demonstrated in our previous work , was carried out to prevent the continued influence of the solvents on Aβ secondary structure and assembly kinetics .
Biomimetic membrane constituents
DMPC (1,2-dimyristoyl-sn-glycero-3-phosphocholine), DMPE (1,2-dimyristoyl-sn-glycero-3-phosphoethanolamine), DMPG [1,2-dimyristoyl-sn-glycero-3-phospho-(1′-rac-glycerol)] and DMPS (1,2-dimyristoyl-sn-glycero-3-phospho-L-serine) were purchased from Avanti Polar Lipids. Cholesterol (95%) and monosialoganglioside GM1 from bovine brain (>95% lyophilized powder) were purchased from Sigma–Aldrich. All materials were used without further purification.
For SPFS, 40 mg/ml stock solutions of the bilayer constituents were solubilized in 2:1 (v/v) chloroform/methanol in a glass vial from the stock solutions (see Table 1 for compositions) and the solvent was removed using dry nitrogen and vacuum desiccation overnight. The lipid films were rehydrated with 5 μM Alexa Fluor® 647 succinimidyl ester (Molecular Probes), hydrolysed to form the carboxylic acid, dissolved in 10 mM Hepes (pH 7.4), 100 mM NaCl, 1 mM EDTA and 0.05 mM sodium azide (all purchased from Sigma–Aldrich; referred to as Hepes pH 7.4), and vortex mixed vigorously for 30 min. The resulting suspension was passed 19 times through an Avestin extruder fitted with two stacked 100 nm polycarbonate membranes (GC Technology). For the GM1-deficient LUV composition see Table 1.
For the AFM bilayer composition see Table 1. The lipid films were rehydrated with Hepes pH 7.4 and vortex mixed vigorously for 30 min. The resulting suspension was passed 19 times through an Avestin extruder fitted with two stacked 100 nm polycarbonate membranes (GC Technology).
For SPFS, clean high refractive index glass (SFL6, n=1.7988, UQG Optics) was coated with a thin ∼50 nm gold layer (99.99%, Advent Research Materials) using an Emitech K975 thermal evaporator. The gold was thermally annealed at 500 °C for 90 s. The gold-coated wafers were ozone cleaned in a UV TipCleaner for 10 min and then placed in absolute ethanol for 10 min and dried with nitrogen to ensure a clean surface. 11-Mercaptoundecanoic-(8-biotinoylamido-3,6-dioxactly)amide , subsequently referred to as biotin-thiol, was self-assembled in an ethanolic solution of 0.05 mM biotin-thiol and 0.95 mM 11-mercapto-1-undecanol (99%, Sigma) to form a SAM on to the gold substrate for 16 h. The biotin surface was coupled with streptavidin (500 nM in Hepes pH 7.4), to create the ‘capture’ surface for the biomimetic lipid vesicles. After LUV attachment the system was rinsed with buffer to remove non-encapsulated fluorophore and non-bound vesicles from the system; the experimental setup is shown in Supplementary Figure S1.
For AFM, 200 μl of 1 mg/ml LUVs were adsorbed to mica and incubated at room temperature (approximately 25 °C) for 3 h. The bilayer formed was rinsed with 10 mM MgCl2 for 10 min to facilitate complete vesicle rupture, and then rinsed in Hepes pH 7.4.
SPR and SPFS measurements and data analysis
SPR was used to follow the construction of the modified surface, and the experimental setup has been described previously [33,34]. The surface construction was monitored as a change in reflectivity angle scan, from which the mass adsorption can be calculated from a simple empirical relationship of adsorption: where Δθ is the resonance minimum and Δθ/σ is the surface concentration correlation (0.1868°). The surface construction and adsorption/desorption events following Aβ addition to the membrane surface was also followed kinetically as the change in reflection intensity at an angle of approximately 1.5° lower than the resonance angle. Therefore changes in the reflected light intensity at a fixed angle allows real-time measurement of peptide–membrane binding. The equilibrium dissociation constant (Kd) is fitted to the binding isotherm: where Req is the equilibrium response, Rmax is the maximum signal response, [A] is the analyte concentration, koff is the dissociation rate constant and kon is the association rate constant, using Origin 7 data analysis software (OriginLab). The equilibrium dissociation constant (Kd) was determined, and all changes in reflectivity were normalized.
The Aβ–membrane response was also analysed using an exponential functional expression to determine mono-, bi- and/or tri-phasic association of the oligomeric and fibrillar Aβ with the membrane: derived from 1−exp(−t/t1). t1, t2 and t3 are the time scales associated with the mono-, bi- and tri-phasic processes respectively, which is equal to 0 at t=0 and goes to 1 when t=∞, a1, a2 and a3 are constant factors for the amplitude of observed change for the mono-, bi- and tri-phasic exponential function respectively, and a0 is the time correction factor for the y value as y does not equal 0, using Grace-5.1.22.
The evanescent wave is used to resonantly excite fluorescence dye molecules within 200 nm of the surface and is detected by a photomultiplier tube, allowing us to monitor any emitted fluorescence encapsulated within the membrane vesicles, but not in the bulk solution. A simple theoretical model based on Fick's first law of diffusion was used to fit the decrease in fluorescence–time curves obtained following Aβ-induced membrane permeation as previously used to model cholera–membrane interactions . The fluorophore flux through permeated membranes, calculated as the apparent dye diffusion coefficient (D*), is effectively deter-mined from the gradient of the linearized change in fluorescence over time, and takes into account vesicle membrane internal volume and membrane area and thickness. Therefore D* is not a function of total encapsulated fluorescent molecules lost from the vesicle aqueous space, but the extrapolated linearized slope of the dye diffusion rate. To be clear, the fluorescence scale shown in Figure 2 is arbitrary and the changes in fluorescence between experiments cannot be directly compared with each other. These fluorescence traces are extrapolated and processed using Origin analysis software, therefore the apparent dye diffusion coefficient (D*) provides a direct means of comparing permeation rates.
The experiments were performed in duplicate (Supplementary Figure S2 at http://www.BiochemJ.org/bj/439/bj4390067add.htm) and the kinetic traces shown in Figure 2 are representative of each experiment. Initially, we monitored the construction of the functionalized surface by following the angle-resolved SPR reflectivity curves (Supplementary Figure S3 at http://www.BiochemJ.org/bj/439/bj4390067add.htm) and the change in resonance minima allowed us to calculate the change in mass to the surface (Supplementary Table S1 at http://www.BiochemJ.org/bj/439/bj4390067add.htm). Scanning electron microscopy was used to confirm the covalent attachment of whole intact lipid vesicles (Figure 1a).
All bilayer images were obtained under filtered Hepes pH 7.4 buffer using a Nanoscope IV Multimode atomic force microscope (Veeco). Bilayer formation was verified by imaging defects and classic bilayer morphology (4.8–5.0 nm step heights for this bilayer composition), or a series of force–distance curves and a complete force–volume map. Aβ42 solution was injected into the liquid cell via flexible tubing so as not to disturb the imaging process too greatly (although the change in temperature often necessitated a resetting of scan parameters). A flexible gasket was used to seal the liquid cell to prevent evaporation and maintain the buffer concentration, and allow for overnight incubations if necessary. Temperature was not directly controlled, although the laboratory temperature was stable, and the cell temperature equilibrated to 26 °C. Both contact mode and tapping mode were used throughout, using a selection of Veeco NP cantilevers depending upon the mode. Images were processed using Nanoscope version 5.12r30 software. Area measurements were performed using the Bearing Analysis function.
Transmission electron microscopy
Electron micrograph images of Aβ were prepared as described previously . Briefly, a 4 μl droplet of the peptide solution was adsorbed on to formvar/carbon-coated 400 mesh copper grids (Agar Scientific) for 60 s, and blotted dry. A 4 μl aliquot of 0.22 μm-filtered water was added to the grid and immediately blotted, and then negatively stained with 4 μl of 2% (w/v) uranyl acetate for 60 s and blotted dry. The grid was allowed to air dry before examination on a Hitachi 7100 microscope fitted with a Gatan Ultrascan 1000 CCD (charge-coupled device) camera (Gatan). Aliquots of samples at the stock concentration were taken at time points for each experiment to monitor the fibrillization state and morphology. Measurements were made using ImageJ .
Scanning electron microscopy
LUVs (100 nm) were tethered to the biotinylated-thiol SAM as described above for the SPFS measurements, and then the glass substrate was removed from the flow-cell. The tethered LUVs were fixed by immersion in 2% glutaraldehyde solution (25% electron microscopy grade, Agar Scientific) for 30 min. The fixed LUVs were then immersed in a 1% osmium tetraoxide solution (2% solution, Agar Scientific) for 20 min to biologically fix and stain the sample. The substrate was rinsed carefully with Hepes pH 7.4 and air dried before examination on a JEOL JSM6310 microscope.
RESULTS AND DISCUSSION
Transmission electron microscopic characterization of Aβ42
The assembly characteristics of Aβ42 were monitored using transmission electron microscopy, Aβ42 used immediately following disassembly treatment is from now on referred to as ‘0-h-incubated Aβ’. 0-h-incubated Aβ showed a variety of morphological species, including small circular peptides of between 10 and 20 nm in diameter, 30–50 nm annular structures, larger 60–70 nm curvilinear oligomers and 70–140 nm amorphous peptide assemblies (Figure 1b). At 20 h the peptide formed long, twisted, unbranched and straight fibres with defined edges (Figure 1c). We have previously shown that Aβ42 was capable of forming morphologically indistinguishable fibres when grown in the presence of lipid membranes, and LUVs changed from regular spherical 100 nm-diameter lipid vesicles with apparently smooth bilayers to irregularly shaped rough vesicles where the bilayer began to bleb and swell in the presence of Aβ42 .
Oligomeric Aβ42–membrane interaction characterization
Monitoring the oligomeric Aβ42–membrane interactions by SPFS
The association of different bound Aβ assemblies with membranes and their subsequent effect on biomimetic membrane stability was followed by SPR and SPFS as changes in resonance angle shifts, reflected light intensity and fluorescence intensity. The experiments were performed in duplicate (Supplementary Figure S2) and the kinetic traces shown in Figure 2 are representative of each experiment. Initially, we monitored the construction of the functionalized surface by following the angle-resolved SPR reflectivity curves (Supplementary Figure S3) and the change in resonance minima allowed us to calculate the change in mass to the surface (Supplementary Table S1). Scanning electron microscopy was used to confirm the covalent attachment of whole intact lipid vesicles (Figure 1a). As a control, we re-circulated Hepes pH 7.4 buffer containing no peptide into the flow-cell for 18 h to monitor the stability of the tethered LUVs. The passive diffusion of Alexa Fluor® 647 through the membranes was calculated as the apparent dye diffusion coefficient, D*=7.29×10−22 m2·s−1 (Supplementary Figure S4 at http://www.BiochemJ.org/bj/439/bj4390067add.htm) and this served as a baseline.
For the purpose of the following experiments, we have selected a concentration of 10 μM Aβ to ensure a visible effect between Aβ–membrane interactions in the time course of our studies. This is comparable with concentrations that show a significant toxic effect on neuroblastoma cells (V. Soura, T. L. Williams and L. C. Serpell, unpublished work). An Aβ concentration titration was initially performed and maximal peptide–membrane binding was observed at this peptide concentration (results not shown). 0-h-incubated Aβ42 (10 μM) was injected into the flow-cell using a peristaltic pump at a flow rate of 1.8 ml/min to ensure continuous delivery of the sample to the membrane surface and minimize mass-transfer effects. An increase in reflectivity was observed within 10 min as a result of Aβ42 binding to the membrane surface (Figure 2, left-hand panel, primary y-axis). The peptide proceeded to reach an initial equilibrium within the first 60 min as there is a short plateau in normalized reflectivity [R]% during Aβ binding at this time, which results from the specific association of the oligomeric Aβ42 with the LUVs. Over the subsequent 400 min there is a second phase of peptide binding. This biphasic interaction suggests an initial period of specific adsorption for the first 60 min, which is then followed by a slower phase of association. After equilibrium was reached, the system was washed with fresh buffer (after approximately 500 min), whereby the reflectivity [R]% drops, which is indicative of non-specifically or weakly bound peptide being washed into solution. From rinse-off data, the dissociation constant can be calculated (see the Experimental section). An angle-resolved reflectivity scan was performed to measure the shift in resonance minima, where it is calculated that there is a 0.59 ng/mm2 increase in mass at the supported surface, and the Kd was calculated to be 2.68×10−8 M. Previous studies have reported SPR-derived Kd constants for oligomeric Aβ42 to vesicles containing DMPC/GM1 of 5.2×10−7 M . The discrepancy between Kd values is likely to be due to differences in bilayer composition and the method of Aβ preparation. In the present study, we have prepared Aβ42 to ensure removal of preformed aggregates. The increased complexity of our membranes, which include physiologically relevant GM1, sterols and negatively charged phospholipids, tend to favour the association of Aβ with membranes, therefore the determined Kd value in the present study suggests stronger binding compared with simpler membrane systems, and careful preparation to ensure all peptide disaggregation solvents are removed ensures that no external influence in membrane binding occurs. At the end of the measurement 5 g/l SDS was added to the system to lyse all attached membrane vesicles, and a change in resonance angle scan minimum was determined, showing that 11.78 ng/mm2 of material was lost from the surface, indicating the almost complete removal of streptavidin-tethered vesicles from the functionalized surface (Supplementary Table S1).
The concomitant change in fluorescence upon Aβ42-induced membrane permeation results in a brief lag of 8 min until an effect in the fluorescence signal is observed (Figure 2, left-hand panel, secondary y-axis), which may result from the time required by the peptide to specifically adsorb to the membranes and begin to permeate the bilayers. Subsequently, there is another brief period of approximately 15 min where there is a phase of fluorescence enhancement which is not observed with the buffer control measurement. We have previously attributed this increased fluorescence intensity to the partial dequenching of the hydrated fluorophore as it is stripped of some of the surrounding water layer as it traverses the permeated membrane channels . The fluorescence intensity begins to slowly decrease after approximately 15 min as the fluorophore leaks out through the permeated membranes and continues progressively to decrease as the dye diffuses through the permeated membranes into solution. The change in fluorescence intensity was modelled and calculated as D*=3.9×10−20 m2·s−1. The diffusion of the dye through 0-h-incubated Aβ42-permeated membranes is an approximately 50-fold faster diffusion rate compared with the passive leakage of the dye through non-permeated control membranes (Table 2). Previously it has been shown that Aβ40 was able to form cation-selective channels across planar lipid bilayers when the peptide was reconstituted into the lipid mixture prior to bilayer formation . It was shown that preformed oligomeric Aβ reconstituted in membranes had channel activity. Complementary to this work, we suggest that Aβ42 can also cause permeation and defects within the lipid bilayer larger than discrete ion channels, capable of allowing a fluorescent dye molecule ∼130 kDa in size to easily pass through the permeated defects. It has also been suggested that Aβ is able to induce membrane conductance in the absence of discrete ion channel or pore formation, but was instead a result of Aβ-soluble oligomers enhancing the ability of ions to move through the lipid membrane on their own . Therefore the ability of Aβ to cause membrane damage may in fact be a collection of mechanisms including channel formation, membrane thinning and/or alteration of the viscoelasticity of the bilayer membranes.
The binding data were analysed in more detail by fitting the oligomeric Aβ42–membrane response to one, two or three exponential functions, revealing that the oligomers binding to a membrane is a multivalent bi- or tri-phasic process (Supplementary Figure S5A at http://www.BiochemJ.org/bj/439/bj4390067add.htm). Using the biphasic exponential we obtain a0=0.9081, a1=0.0603, t1=220 min, t2=7min, χ2=0.00183 and correlation coefficient=0.9963. Using the triphasic exponential, a slight improvement in the fit was observed, where a0=0.9064, a1=0.0319, a2=0.0269, a3=0.0622, t1=103 min, t2=5 min, t3=877 min, χ2=0.00143 and correlation coefficient=0.9971 (Supplementary Table S2 at http://www.BiochemJ.org/bj/439/bj4390067add.htm). However, the triphasic exponential resulted in extended time points (t3=877 min), which is beyond the time scale of the experiment, and would result in Aβ–membrane processes that would be complicated by fibril formation at this time point. Therefore the triphasic exponential was rejected for the purposes of our analysis. A single monophasic 1:1 exponential resulted in a poor fit. Therefore oligomeric Aβ–membrane interactions do not fit well and can therefore be excluded. However, the biphasic exponential modelling reveals that there are two phases of Aβ oligomer binding to phospholipid membranes, suggesting an initial phase of Aβ binding to the membranes, followed by a second phase that may be either non-specific association of the peptide to the membrane surface or self-association of the Aβ from the bulk solution to existing membrane-bound Aβ resulting in amyloid assembly and the beginnings of fibrillization. From the biphasic exponential function, the t2=7 min corresponds to the beginning of fluorescence leakage through the permeated membrane. At this time point, we have attributed this to the phase of fluorescence enhancement, where the fluorophore begins to diffuse through the permeated membranes. The t1=220 min for the biphasic exponential function corresponds with the beginning of the linear range of fluorescence diffusion, and may result from the formation of the largest defects in the bilayer membrane. Both t1 and t2 time phases are important points in the leakage of the fluorophore through the permeated membranes.
Monitoring the fibrillar Aβ42–membrane interactions by SPFS
Aβ42 fibres were prepared from stock 10 μM 0-h-incub-ated Aβ42 and statically incubated for 7 days and confirmed by transmission electron microscopy (Figure 1d); fibres were vortex mixed vigorously to break-up fibrillar clusters prior to injection into the flow-cell (Figure 2, middle panel). This adsorption slowly increases until equilibrium and showed markedly different kinetics of association compared with the adsorption of 0-h-incubated Aβ42 (Figure 2, left-hand panel). There is no immediate sharp increase in reflectivity, which is indicative of the specific adsorption of the peptide to the membrane (as observed with 0-h-incubated Aβ42). Instead, there is a continual and gradual increase in reflectivity until equilibrium is reached at approximately 600 min. This fibrillar association suggests non-specific or weak binding between the fibres and the tethered LUVs, and if binding is mediated by the ends of the molecules, this could reflect the relatively fewer free ends compared with oligomeric species. Upon rinsing of the flow-cell with fresh Hepes pH 7.4 at 600 min, the weakly adsorbed Aβ42 fibres disassociate from the tethered membranes and wash off into solution. A 0.06 ng/mm2 an increase in mass is observed, which is approximately 10-fold lower in mass adsorption compared with 0-h-incubated Aβ42. The equilibrium dissociation constant is calculated to be 8.80×10−6 M; this is a significantly lower affinity (2-fold decrease in affinity) compared with 0-h-incubated Aβ42 (Table 2). Only a Kd determination between Aβ40 and membranes has previously been reported, and SPR binding of Aβ42 fibrils to lipids has not been previously studied. Using the exponential functional fit a0+a1(1−e−t/t1), fibril association with the tethered membranes is shown to fit to a single-exponential function (Supplementary Figure S5B and Supplementary Table S2), a0=0.9473, a1=0.0567, t1=245 min, χ2=0.00149, correlation coefficient=0.9967. Addition of a second fit did not improve the fit, suggesting that fibril binding to the membrane occurs in a monophasic 1:1 binding model, which is significantly different compared with the biphasic process associated with oligomeric Aβ42 interactions with the membrane.
The change in fluorescence upon fibrillar permeation of the LUVs results in a brief 5 min period of fluorescence enhancement, similar to that observed with the addition of 0-h-incubated oligomeric Aβ42. The fluorescence intensity begins to slowly decrease after approximately 5 min as the fluorophore leaks out through the permeated membranes. The fluorophore leakage progressively decreases as the dye diffuses through the permeated membranes into solution. The calculated apparent dye diffusion coefficient of D*=3.65×10−22 m2·s−1 is not significantly different compared with the passive leakage of the dye through non-permeated control membranes, and results in significantly less leakage (1.96-fold decrease) compared with oligomeric Aβ42-induced membrane permeation (Table 2). Although the overall decrease in normalized fluorescence may appear to be similar between fibres and oligomers, it is not relevant to compare these decreases in fluorescence values and encapsulation efficiencies, and LUV capturing may and does vary from experiment to experiment, so they cannot be directly compared. The only true means of comparing the permeation is by comparing the apparent dye diffusion coefficients. We have calculated a D* value of 3.9×10−20 m2·s−1 from Aβ oligomers compared with D*=3.65×10−22 m2·s−1 for fibrils. Our results, showing that 0-h Aβ binds and causes permeation of LUVs more strongly than fibrillar Aβ, suggest that binding to the membrane and subsequent elongation are linked to the resulting membrane damage leading to permeation.
GM1 influences both oligomeric Aβ42-membrane binding and permeation
Monitoring the oligomeric Aβ42–GM1-deficient membrane interactions by SPFS
The GM1 ganglioside expression in biological membranes has been reported to affect the interactions between the Aβ peptide and the membranes, and in our previous work we have shown that the removal of GM1 from the composition of the membranes results in a greater than 50% decrease in permeation of membranes . In the present study we examined the effect of GM1 within the tethered membrane vesicles, in simultaneously modulating the binding and permeation of 0-h-incubated Aβ42. The addition of 10 μM 0-h-incubated Aβ42 to tethered LUVs without GM1 (see Table 1 for the LUV composition) causes an immediate increase in reflectivity as the Aβ42 binds to the membranes within the initial 20 min (Figure 2, right-hand panel). The preliminary binding kinetics appear to be slower compared with when GM1 is present, and the initial binding kinetics are approximately 3-fold slower to reach initial equilibrium compared with when GM1 is present in the bilayers (Figure 2, left-hand panel). The adsorption of the Aβ42 to the membranes not containing GM1 then follows a slower phase of binding until equilibrium is reached approximately 900 min after Aβ42 injection. Upon reaching equilibrium, there is a small decrease in reflectivity, which may be the result of partial loss of non-specifically adsorbed peptide from the tethered LUVs. There is a 0.19 ng·mm−2 increase in mass at the supported surface, which is approximately 3-fold less than when GM1 is present. The equilibrium dissociation constant is calculated to be 8.90×10−7 M; this is an approximately 1.8-fold lower binding affinity compared with when GM1 is present (Table 2). From the binding kinetics, this suggests that GM1 is involved in modulating Aβ42 oligomer binding, and its exclusion from the membranes decreases the binding affinity. Binding of the oligomeric peptide to the tethered membranes is still observed, suggesting that Aβ can also bind to other components found within our LUVs, such as cholesterol and/or the phospholipids. The binding of Aβ peptide to GM1 appears to be concentration-dependent, and with increasing GM1 concentration there is an increase of Aβ42 binding. The addition of 10 μM 0-h-incubated Aβ42 to tethered LUVs containing various physiologically relevant proportions of GM1 (0, 2, 5, 10 and 15 mol%) causes various increases in reflectivity and minima shifts, which were plotted against GM1 concentration (Figure 3). From this plot the Req is determined as 1 mol%. Above 5 mol% GM1 the Aβ42 binding becomes saturated and equilibrium is reached, whereby no further increase in Aβ42 binding is observed. Ensuring that the GM1 molar fraction is between the Req and Rmax ensures maximal binding efficiency without saturating the LUV membranes.
SPFS for tethered vesicles without GM1 shows very little change in fluorescence on addition of Aβ oligomers (Figure 2, right-hand panel), and the fluorophore leakage shows similar kinetics to the buffer control experiment. The change in fluorescence intensity is calculated as D*=1.41×10−22 m2·s−1, and is significantly different (1.99-fold decrease) compared with the oligomeric permeation of membranes (Table 2). The decrease in fluorescence leakage observed when no GM1 is present suggests that GM1 may play a role in modulating Aβ permeation.
The removal of the GM1 gangliosides from the membrane composition has been reported to potentially destabilize membranes , therefore in our experiments we would theoretically observe a potentially greater Aβ-induced membrane permeation. However, the decrease in D* and lower apparent dye diffusion observed between Aβ and membranes not containing GM1 is probably the result of GM1 modulating Aβ–membrane binding and not a result of altered membrane stability. The increased binding observed between Aβ and GM1-containing membranes may be the result of the ganglioside ability to form clusters, which are able to interact with various membrane components and membrane proteins , thereby increasing the local concentration of bound Aβ on the membrane and increasing the potential for Aβ–Aβ interactions. This is supported by previous studies that have suggested a role for GM1 in seeding aggregation of Aβ . The influence of electrostatic interactions is believed to play a key role in modulating Aβ–GM1 interactions, as at pH 5.5 Aβ40 is slightly positively charged and the headgroup of GM1 is negatively charged, whereby the peptide shows greater insertion pressure compared with when Aβ40 inserts into GM1-containing membranes at pH 7.2 where Aβ is negatively charged .
Observation of Aβ42–membrane interactions by AFM
In order to directly visualize the association and effect of oligomeric or fibrillar Aβ on membranes, we have used AFM allowing the monitoring of elongation and permeation. AFM images reveal heights of the sample above the surface whereby associations with the bilayer will appear light and holes or defects appear dark. A freshly prepared mica-supported bilayer was scanned in tapping mode AFM (Figure 4a) prior to the addition of Aβ42 to ensure complete bilayer formation and for comparison with Aβ-adsorbed bilayers. Injection of 10 μM 0-h-incubated Aβ42 shows the association of the peptide to the membranes (Figure 4b). A variety of morphological oligomeric Aβ42 species is observed on the membrane surface. An aliquot of the Aβ42 injected into the AFM flow-cell was used to prepare a transmission electron micrograph for comparison (Figure 4f). The transmission electron microscopy shows similar morphological peptide species to that observed on the membrane via AFM, which include circular oligomers ranging between 20 and 50 nm in diameter, and curvilinear oligomers that are 70–85 nm in length. AFM images were scanned in tapping mode upon incubation of the 0-h-incubated Aβ42 with the bilayers over a time course of 1 h to visualize the changes in peptide and bilayer (Figures 4b–4e). Deposition of Aβ42 on to the membrane surface occurs immediately following injection, with a visible roughening of the smooth membrane surface. Increasing amounts of Aβ42 assemblies are observed during the incubation of the peptide with the bilayer, which include a great abundance of circular oligomers, and the emergence of a few short fibrillar species (Figures 4g–k). After further incubation, an increasing number of small circular oligomers, protofibrils and a small number of short fibrils are observed (Figure 4e). After incubating the peptide and bilayers for 1 h a greater abundance of 200 nm protofibrils appear adsorbed to the membrane and much longer 600–1000 nm fibres. The assembly of Aβ42 oligomers into fibres can clearly be visualized by AFM (Figures 4g–4i and 4j and 4k), and structures looking like ‘strings of beads’ can be seen to begin to form fibres.
The adsorption of the peptide to the bilayer causes the appearance of defects in the bilayer structure, as the membrane changes from an apparently smooth structure with a few discrete double bilayer patches and defects revealing the mica substrate 5 nm below (Figure 4a), to a rough bilayer as the Aβ adsorbs (Figures 4b–4e). Small irregular imperfections and indentations ranging from 10 to 100 nm begin to emerge within 30 s of the Aβ42 adsorbing to the membrane surface (Figure 4b), and these ill-defined defects appear to only permeate through the first lipid monolayer of the bilayer membranes. A cross-section plot shows a depth of approximately 3–4 nm (Figure 5a). Upon further incubation of the Aβ42 with the bilayers, these defects become obvious holes with well-defined circular edges peppered across the entire bilayer surface; these holes range between 10 and 100 nm in diameter (Figure 4d), and appear to permeate through both leaflets of the bilayer membrane. From profile scans the holes were calculated to be 6.2±0.2 nm deep (Figure 5b). The depth of the holes is slightly deeper than the unpermeated membrane (5 nm) before Aβ addition, which could either result from the adsorption of the peptide to the bilayer or as a result of the Aβ causing the acyl chains of the phospholipids to swell and change the fluidity of the membranes . During the 1 h time course of Aβ42–bilayer incubation, the regions of permeation tend to vary in diameter and the defects within the membrane are seen to range from a few nanometres up to large areas of permeation up to 280 nm in diameter where several defects merge to form larger holes. A 5.4±0.8% (S.D. 1.8%) loss of bilayer by Aβ42 permeation resulted, suggesting that Aβ may act by a combination of both pore formation and detergent-like fashion causing the removal of lipids from the bilayer surface and therefore altering the viscoelasticity of the membrane. Therefore AFM gave us a direct means to visualize the adsorption of Aβ42 in the presence of bilayer membrane, but also showed the permeation of these membranes, initially forming defects which did not permeate the membrane fully (Figure 5a), but during the incubation these defects became holes that permeated all the way through the bilayer (Figure 5b). Imaging of the bilayer with the lightest force showed quite large quantities of loosely bound peptide, together with patches of ejected bilayer floating on top. More than a single image of the same area resulted in the holes refilling with bilayer, a process driven by the scanning AFM probe. A zoom out following an initial scan revealed holes on the unscanned perimeter, with a fully smooth square with closed-up holes from the previous scan (results not shown). The refilled holes were obvious as fresh bilayer with little adsorbed peptide (Figures 4g–4i and 4j and 4k). To avoid this AFM-driven healing effect, the scan area was changed each time with a manual translation. The holes, once formed, appeared to be stable over time. Bilayer is ejected from the surface as the holes appear, and as the bilayer is solid supported there is no excess lateral pressure to drive the holes to close up. Conversely, in a vesicle or real cell membrane, the flexible membrane could change shape and size to some degree, minimizing or even closing up the hole. Therefore the hole is less transient and possibly much larger. Similar effects have been previously observed for the amylin association with lipid bilayers at lower magnification by confocal microscopy, showing immediate defects in the membrane following administration of oligomeric amylin . The authors suggest a direct and strong interaction between the fibrillizing peptide and the lipid molecules.
In contrast with the dramatic effect of oligomeric Aβ42 on lipid bilayers, the addition of 10 μmol·dm−3 fibrillar Aβ42 to the bilayer surface resulted in no apparent fibrillar association with the membranes and despite fibrils being observed for the samples by electron microscopy, no fibrils were observed by AFM and no images could be obtained. This may have resulted from the weak association of the fibrillar peptide being swept away by the AFM probe during imaging, even when using the lightest force. This supports the SPFS results showing that fibrillar binding to membranes is non-specific, and weakly bound peptide does not have high affinity for the biomimetic membranes. In contrast, fibres that had been grown from 0-h-incubated Aβ42 in the presence of the bilayer remained strongly adhered and could be imaged by AFM (Figure 4).
Three structurally divergent models for Aβ–membrane permeation have been proposed: (i) carpeting of the peptide on one leaflet of the membrane surface, which results in an asymmetric pressure between the two leaflets ; (ii) the formation of stable pores and ion channels has been proposed for amyloid-induced toxicity, whereby the disruption of Ca2+ homoeostasis has been observed as a potential mechanism associated with AD; (iii) detergent-like effects of amyloid-induced membrane damage are believed to occur through the association of the amyloid peptides in the form of micelle-like structures with the membrane surface. At high local peptide concentrations on the membrane surface, either after the surface is covered with peptide monomers or oligomers or through the association between membrane-bound amyloid, a detergent effect is observed .
AFM has previously shown apparent channel formation by Aβ42 that had been reconstituted into liposomes . In contrast, in the present study we show that Aβ42 administered to the lipid bilayer causes immediate defects upon binding. This result is also in contrast with a study showing that TFA-solubilized Aβ42 association with total brain extract bilayers assembled to form fibrils with no apparent damage to the membrane . However, advances in Aβ42 peptide preparation and formation of oligomers may be responsible for these differences. In the present study, Aβ42 has been solubilized using newly established methods to ensure disaggregation and also removal of all solvents . The AFM studies reveal almost immediate and dramatic changes to the lipid bilayer in response to binding of Aβ42 and its assembly on the membrane, showing for the first time the changes associated with membrane permeation of oligomeric Aβ42.
The assembly of Aβ42 has been implicated in the cytotoxicity effects, and has been reported to be associated with early oligomeric amyloid species. Oligomeric Aβ affects the integrity of the membrane structure, and may also provide a route for the uptake of Aβ into the cytosolic environment of the cells or affect internal organelles (V. Soura, T. L. Williams and L. C. Serpell, unpublished work). In the present study we have shown that 0-h-incubated Aβ42 oligomers bind with greater affinity and cause greater permeation to synthetic membranes compared with fibrillar Aβ42. Oligomeric Aβ42 binding appears to follow a biphasic interaction with the membrane, whereby there is an initial phase of specific adsorption, followed by a secondary phase of either non-specific association of the peptide with the membrane or the self-association of the peptide as it assembles and fibrillizes. Moreover, the association of fibrillar Aβ42 with membranes follows a monophasic exponential fitting. We postulate that this assembly from oligomer to fibrils may cause some mechanical defects in the membranes. The SPFS methods have allowed us to directly correlate the binding with the permeation of the lipid vesicles. The GM1 ganglioside appears to affect the binding of the oligomeric peptide to the membrane surface and also modulates the permeation, whereby the exclusion of GM1 from the membrane composition results in decreased Aβ42 binding and decreased permeation. It has been previously suggested that GM1 may form microdomains that seed association and aggregation of Aβ42 . Direct visualization of the permeation of the membranes by AFM reveals initial defects where the peptide begins to penetrate the upper monolayer of the bilayers that develop with time to become discrete holes up to 200 nm in diameter. This membrane damage appears to be directly correlated with growth of the peptide to form fibrillar deposits on the membrane. From these unique findings in the present study of Aβ-directed membrane permeation, we suggest that soluble oligomeric Aβ may possess several mechanisms of inducing cell death, and the formation of holes in the membrane and altered viscoelasticity of the membrane may be crucial in modulating cell toxicity. Taken together, our results reveal a direct correlation between binding and assembly of Aβ42 on the membrane with the observed leakage from vesicles and holes forming in lipid bilayers. These observations have important implications for the cytotoxic activity of the Aβ42 oligomers relevant to AD pathology.
Thomas Williams and Louise Serpell wrote the paper. Thomas Williams designed and carried out the experiments and analysed the data. Benjamin Johnson helped with the experimental design and collected data. Brigita Urbanc produced mathematical models to analyse SPFS data. Toby Jenkins contributed to experimental design and analysis of SPFS data. Simon Connell and Thomas Williams performed and designed the AFM experiments. Louise Serpell managed the project and critically assessed the data.
This work was supported by Alzheimer's Research UK (previously Alzheimer's Research Trust) (to T.L.W. and L.C.S.); and the National Institutes of Health [grant number A6027818 (to B.U.)].
We thank Dr Julian Thorpe for all his help and guidance with electron microscopy. We thank Professor Stephen Evans (Leeds) for his guidance and support. We are grateful for the generous researcher exchange prize given by the Synthetic Components Network, and the Leeds EPSRC Nanoscience and Nanotechnology Research Equipment Facility (LENNF). We would also like to thank Dr Petra Cameron and Eleanor Johnson from the University of Bath for help and advice with the SPFS.
Abbreviations: Aβ, amyloid-β peptide; AD, Alzheimer's disease; AFM, atomic force microscopy; DMPC, 1,2-dimyristoyl-sn-glycero-3-phosphocholine; DMPE, 1,2-dimyristoyl-sn-glycero-3-phosphoethanolamine; DMPG, 1,2-dimyristoyl-sn-glycero-3-phospho-(1′-rac-glycerol); DMPS, 1,2-dimyristoyl-sn-glycero-3-phospho-L-serine; HFIP, 1,1,1,3,3,3-hexafluoro-2-propanol; LUV, large unilamellar vesicle; SAM, self-assembled monolayer; SPFS, surface plasmon field-enhanced fluorescence spectroscopy; SPR, surface plasmon resonance
- © The Authors Journal compilation © 2011 Biochemical Society