3D-visualization of amyloid-β oligomer interactions with lipid membranes by cryo-electron tomography†

Amyloid-β (Aβ) assemblies have been shown to bind to lipid bilayers. This can disrupt membrane integrity and cause a loss of cellular homeostasis, that triggers a cascade of events leading to Alzheimer's disease. However, molecular mechanisms of Aβ cytotoxicity and how the different assembly forms interact with the membrane remain enigmatic. Here we use cryo-electron tomography (cryoET) to obtain three-dimensional nano-scale images of various Aβ assembly types and their interaction with liposomes. Aβ oligomers and curvilinear protofibrils bind extensively to the lipid vesicles, inserting and carpeting the upper-leaflet of the bilayer. Aβ oligomers concentrate at the interface of vesicles and form a network of Aβ-linked liposomes, while crucially, monomeric and fibrillar Aβ have relatively little impact on the membrane. Changes to lipid membrane composition highlight a significant role for GM1-ganglioside in promoting Aβ-membrane interactions. The different effects of Aβ assembly forms observed align with the highlighted cytotoxicity reported for Aβ oligomers. The wide-scale incorporation of Aβ oligomers and curvilinear protofibrils into the lipid bilayer suggests a mechanism by which membrane integrity is lost.


Introduction
Alzheimer's disease (AD) accounts for more than two-thirds of dementia world-wide. A large body of evidence indicates its molecular basis centres on a small hydrophobic peptide, amyloid-b (Ab). Cleaved from a large amyloid precursor protein the Ab-peptide is typically 40 or 42 amino acids in length (Ab40/ 42). 1 The self-association of monomeric Ab results in a heterogeneous mixture of small oligomeric assemblies, protobrils and amyloid brils which form extra-cellular plaques in the brains of AD patients. These assemblies have different biophysical and synapto-toxic properties. The interaction of Ab with lipid membranes is believed to impede synaptic function, causing loss of cellular homeostasis which ultimately leads to hyper-phosphorylation of tau, cell death and dementia. 1 Current understanding of Ab membrane interactions presents quite a confused picture. This may be due to different membrane systems studied in non-native conditions, and poorly dened Ab assembly states, while different imaging and biophysical techniques employed has resulted in different aspects of the Ab interaction being emphasized. There is a great deal of experimental evidence to indicate Ab assemblies, oligomers in particular, disrupt membrane integrity. [2][3][4] Ab42 oligomers have been shown to insert into cellular membranes and form large single ion-channel pores, with an internal diameter of between 1.9 and 2.5 nm. 5 Alternatively, a more wide-spread carpeting of the membrane by Ab has been proposed. This can cause a general increase in membrane conductance due to membrane thinning and the lateral spreading of lipid headgroups. [6][7][8] Furthermore, Ab has been shown to induce Ca 2+ inux 9,10 or dye release in vesicle models. 11 Ab cytotoxicity has been described extensively 12,13 and a loss of Ca 2+ cellular homeostasis is well established in AD. 14,15 However, the precise mechanisms by which Ab assembles compromise membrane integrity and molecular level imaging of this process remains poorly described.
Lipid extraction by Ab oligomers from supported lipid bilayers has been imaged by atomic force microscope (AFM) and this has been likened to the effect of a detergent. 16 Others have highlighted the importance of elongating brils at the surface of membranes, which may cause extraction and incorporation of lipid into growing Ab42 brils. 17 There are also studies to indicate the lipid membrane composition, in particular levels of GM1 ganglioside, [18][19][20][21][22] and cholesterol, 23,24 can inuence Ab affinity for the bilayer. Similar effects on lipid membranes have been reported for other amyloid forming proteins, including: amylin, [25][26][27] alpha-synuclein, 28 mammalian prion protein, 29 b 2macroglobulin (b 2 M) 30 and serum amyloid A 31 which suggests a shared mechanism of membrane disruption. These behaviours draw some parallels with the toxicity mechanism described for anti-microbial peptides. 4,32 Previously, Ab interplay with lipid bilayers have been studied using predominantly AFM and negative-stain transmission electron microscope (TEM). 16,18,[33][34][35][36][37] These techniques have the capability to reveal nanoscale details of the membrane-amyloid interaction but at the same time can be artefact-prone. AFM can only be utilized for imaging on at and supported surfaces, such as mica. Heavy metal staining in TEM causes sample drying and structural artefacts such as attening of spherical objects. In contrast, cryo-electron tomography (cryoET) is an electron cryo-microscopy technique that can resolve unique structures in a native state, in three dimensions (3D) and at the macromolecular resolution range, 38 and is particularly well suited to investigate protein/membrane systems in 3D, 39 as well as amyloid brils. 40 CryoET has recently been used to study the interaction of brils from the Huntington's disease associated polyQ expanded protein, with membranes from cellular inclusion bodies in situ. 41 There is also reports of cryoET studies which focus on brils, but not oligomers, of b 2 M and their interaction with liposomes. 30 There has also been a roomtemperature tomographic study of serum amyloid A brils stained with heavy metal 31 as well as a scanning tomographic study of Ab plaques ex vivo. 42 With the impact of Ab assemblies on membrane permeability well established, here we aim to employ the latest developments in cryoET data collection strategies and hardware, including direct electron detectors, to report nanoscale 3D images of different Ab assembles impacting the surface of liposomes. In contrast to monomeric and mature brils, oligomeric and curvilinear protobrils interact extensively with the membrane surface, carpeting and inserting into the upper leaet of the bilayer. The Ab decorated membrane attracts neighbouring vesicles to form a tightly zippered network of inter-connected vesicles. CryoET imaging under near native conditions reveals the mechanism by which Ab oligomers and curvilinear protobrils can disrupt cellular homeostasis.

Results
Using an extrusion method, we have generated large unilamellar vesicles (LUVs). This lipid membrane model has the advantage that components of the bilayer can be altered, the initial lipid composition studied includes an aqueous mixture of phosphatidylcholine (PC), cholesterol and GM1-ganglioside, with a ratio 68 : 30 : 2 by weight, buffered at pH 7.4. This lipid mixture was chosen to mimic the typical composition of the extracellular face of membranes.
A few microliters of the vesicle suspension were applied to an EM grid and vitried in liquid ethane. This process offers the best possible structural preservation and is compatible with high-resolution imaging. Then, the sample was transferred to an electron cryomicroscope and a series of 2D images at discrete angles were acquired and computationally reconstructed into a 3D volume to produce the tomogram. The liposomes suspended in aqueous buffer, range in size, typically between 100 and 250 nm in diameter. The large unilamellar vesicles are intact and highly spherical, see Fig. 1 panel a, and ESI Fig. S1. † The mean lipid bilayer thickness has been measured to be 5.1 AE 0.1 nm. Multivesicular liposomes are also observed with smaller vesicles encapsulated within the larger vesicles, as shown in Fig. S1a. † Consistent with this, negatively-stained TEM images show vesicles largely spherical and undecorated, Fig. S1b. † Ab follows a nucleated polymerisation reaction in which Ab monomers form small oligomeric assemblies that then nucleate the rapid formation of large amyloid brils, Fig. S2. † We wanted to investigate the interaction of different Ab assembly forms with the lipid-bilayer. This was achieved by isolating and characterizing Ab at three stages of bril assembly. These stages were: monomeric Ab; prebrillar mixed oligomeric assemblies, taken from the end of the lag-phase; and also mature brils taken once bril assembly has reach equilibrium, as described in the experimental procedures.
The Ab assemblies taken at the end of the lag-phase contain a mixture of prebrillar structures, while appreciable monomeric Ab is still present, 43 as indicated by size exclusion chromatography. Negative-stain TEM indicates the lag-phase preparations are heterogeneous and contain a number of circular oligomeric structures typically ca. 10 nm in diameter. When imaged by cryoET smaller oligomers with typical diameters ca. 2-3 nm, and many curvilinear protobrils are observed, see Fig. S3. † Heterogeneous lag-phase assemblies are highly dynamic and rich in nucleating structures therefore no attempt was made to isolate these transient mixtures further.
Mature amyloid brils were also studied, from Ab samples at equilibrium. These bril assemblies were further puried by removing any smaller oligomers using a 100 kDa molecular cutoff lter. Fibrils are typically un-branched structures, 6-20 nm in diameter and microns in length, as imaged by cryoET and negative-stain TEM, shown in Fig. S4. † Ab oligomers/protobrils but not monomers or brils, decorate the liposome surface Essentially monomeric, chromatographically puried recombinant Ab42 was added to the liposome solution directly aer elution from the size exclusion column. The nal concentration of Ab42 was 10 mM, incubated with vesicles at a concentration of 0.5 mg ml À1 . Monomeric Ab42 preparations were incubated with liposomes, for 10 minutes before freezing ready for cryoET imaging. Under these conditions there is some conversion of Ab monomer to oligomers, but this is minimal. Images for lipid membranes with and without the presence of Ab42 monomer (10 mM) shows no apparent effect, see Fig. 1b and S5a. † Indeed, the appearance of the lipid bilayer in the presence of monomeric Ab42 is indistinguishable from the bilayer in aqueous buffer, Fig. 1a and S1, † with no change in the thickness or density of the lipid bilayer.
The impact on the lipid membrane when challenged with preparations of heterogeneous oligomeric Ab42 assemblies (10 mM, monomer equivalent) are very different compared to monomeric Ab42. For these preparations many oligomers and protobrils have adhered to the surface of the membrane, aer 120 minutes' incubation with the vesicles, shown in Fig. 1c, additional images are shown in Fig. S5b. † These oligomers and curvilinear protobril assemblies are densely adhered to the bilayer, carpeting its surface, this is also highlighted in a single threshold rendered image, Fig. 2. These Ab42 oligomers and curvilinear protobrils have a higher density than the lipid bilayer and so from the continuation of these dense structures below the surface of the membrane, it is clear the oligomers are able to embed within the upper leaet of the bilayer, see   vesicle where many protruding curvilinear protobrils and oligomers are observed above the lipid surface, this is also highlighted in ESI Fig. S6. † Preparations of mature Ab42 brils incubated with the bilayer were also imaged, Fig. 1d, S5c and S6d. † The interaction of brils with the membrane is considerably less marked than those of the oligomeric samples. Indeed, the majority of images show no interaction between the brils and membrane, indeed the limited interactions that do occur in these samples are for residual oligomers present in the bril samples. Unlike the oligomers, there is not an attraction of preformed brils to the membrane surface. The lateral face of the bril does not readily adhere to the membrane in an aqueous environment, even when the bril is close to the membrane, see for example, Fig. S5c. † Also shown in Fig. S5 † are the density proles from monomer, oligomer and bril samples, these highlight the differences in their affinity for the membrane. There are some atypical examples, imaged by negative-stain TEM, of brils interacting with the membrane, these are anchored or limited to the ends of the brils, where brils do contact the membrane there are distortions on the curvature of the bilayer, ESI Fig. S7. † A similar behaviour has been reported for b 2 M brils. 30 The effects shown in Fig. 1 and 2 are consistently observed for multiple vesicles, as evidenced by inspection of typically 300 liposomes for each condition, for multiple preparations, summarized in Table 1. To quantify these effects, we surveyed liposomes incubated with Ab42 monomer, oligomers and brillar preparations. In total 299 vesicles incubated with predominately monomeric Ab42 were inspected. Only a limited number, 12%, of the vesicles (incubated for 10 min) were perturbed with evidence of some Ab42 assemblies binding to the surface of the membrane. Vesicles were also incubated with Ab42 monomer samples for 120 min, these resulted some oligomers forming but decoration of the vesicles remained relative minor, 24%. Similarly, from a total of 302 vesicles inspected aer incubation with Ab42 brils, only 13% exhibited any assemblies adhered to the lipid membrane. As with the monomeric sample, the limited number of vesicles that were perturbed by Ab42, showed only some coverage that was not marked and only partial. In contrast, of 308 vesicles inspected that were incubated with the Ab42 oligomeric preparations, 80% of these vesicles showed Ab interactions with the membrane. For these vesicles the coverage by Ab was markedly more extensive and wide spread with a carpeting effect across the membrane surface. We repeated these experiments to create independent sets of data by creating new stocks of Ab preparations to incubate with freshly prepared lipid vesicles. Very similar observations were made for each preparation, summarized in Table 1.
Quantication indicates oligomers and protobrils are concentrated on, and embedded within the outer-leaet of lipid bilayers Next, we wanted a way of quantifying the amount of binding and incorporation of Ab on the lipid bilayer. CryoET imaging parameters included an applied defocus tuned so that the inner-and outer-leaet of the bilayer could be resolved. Prole plots with normalized integrated intensity (NII) 'grey-values' were generated across the lipid bilayer. The grey values were summed for the entire 2D projection slice, 7.6 nm thick, around the whole perimeters of each vesicle, by performing radial averaging. To quantify this effect, we measured this for multiple vesicles (n ¼ 5), which equates to a summed vesicle length of typically more than 1000 nm, for each preparation. The data was collected for monomeric, oligomeric and brillar preparations, as well as vesicles in the absence of Ab42. Comparisons with membrane controls in buffer alone, and with that of vesicles incubated with monomeric Ab42 indicates no signicant difference in the molecular density of the bilayer, Fig. 3a, b and e. In contrast, the grey-values for the membrane incubated with Ab42 oligomers shows considerable Ab42 incorporation on the surface. The outer-leaet shows a greatly increased amount of grey-value density, this indicates extensive binding on the surface and signicant incorporation of Ab42 oligomers into the membrane. Indeed, the carpeting of the bilayer has the effect of making the membrane appear thicker, as indicated in the grey-value prole plots, Fig. 3c and e. Mean thickness of the bilayer was calculated at 5.1 AE 0.1 nm for vesicles in only buffer, or in the presence of monomeric Ab42 (5.1 AE 0.2 nm) while vesicles in the presence of Ab42 oligomers/protobrils signicantly increased the thickness by an average of 1.6 nm to 6.7 AE 0.3 nm, Fig. 3e and 2 (surface rendering). This increased greyvalue density is largely restricted to the outer-leaet of the membrane, as evidenced by asymmetry between the leaets, with the inner leaet being on average 0.23 AE 0.06 less dense than the outer leaet, as shown in grey-value plot, Fig. 3c and quantied in Fig. 3f. Similar analysis of the grey-values across the lipid bilayer in the presence of mature Ab42 brils supports the assertion that there is not a widespread interaction of Ab42 brils with the membrane, Fig. 3d-f, as there is no increase in the bilayer thickness and no change in the grey-value densities within the membrane.

3D structure of curvilinear protobrils embedded in lipid membranes
The numerous curvilinear protobrils observed in the heterogeneous lag-phase preparations produce excellent high-contrast cryoET images, these three dimensional structures are represented as single-threshold surfaces, Fig. 4. Corresponding structures are also shown in movies, ESI Movies M2-M4. † These assemblies have a range of lengths, see histogram in Fig. S3b, † the majority of curvilinear protobrils are between 10 and 25 nm, and tend not to exceed 40 nm, with mean lengths of 19 AE 9 nm. While their diameters are quite consistent at 2.7 AE 0.4 nm, both values are for n ¼ 100 protobrils. The curvilinear protobrils varied from linear to branched structures, Fig. 4a. These highly irregular and branched assemblies have considerable variation in the extent of their curvature. We believe our 3D images are the rst curvilinear protobrils to be imaged using cryoET. The protobrils structures are more irregular and branched than is perhaps appreciated from 2D images and previously reported negative-stain TEM imaging. Like the shorter oligomers these protobrils interact with the lipid membrane extensively. There are a number of examples in which the ends of the protobrils have embedded into the bilayer as their heightened density continues within the upper-leaet of the bilayer, Fig. 4 and S5d. † The protobrils orientated orthogonally with the membrane surface, Fig. 4b, suggesting a displacement of the upper-leaet of lipid bilayer.
It is notable that the contrast in tomographic images for the curvilinear protobrils are more marked than that of mature bril images under the same acquisition conditions, which is highlighted in the contrast density prole between a curvilinear protobril, a bril bundle and the membrane, Fig. S4c. † This enhanced contrast is surprising as the diameter of these pro-tobrils are smaller, 2.7 AE 0.4 nm, compared to the mature brils, which are typically 10 nm, and range between 6-20 nm depending on the polymorph. 44 This is an important observation and indicates the protobrils have a higher density of biological material and more compact structure than mature brils. Interestingly, the smaller spherical oligomers have density similar to the curvilinear protobrils, Fig. S3. † The diameter of these short more spherical oligomers is between 2 and 3 nm, which is similar to the diameter of curvilinear protobrils.

Ab protobrils remain on the outer leaet and can cluster and link liposomes together at their interface
The observation that Ab42 incorporation into the lipid bilayer is restricted to the outer leaet, poses the question as to whether Ab42 oligomers are able to pass through the bilayer and be observed inside vesicles. Some of the examined liposomes are multivesicular and contain a smaller vesicle encapsulated within the larger vesicle. Observation of the encapsulated vesicles, and the accompanying quantication using grey-scale analysis, indicates Ab42 assemblies do not tend to migrate across the bilayer, as detectable Ab42 oligomers are not observed bound to internal vesicles, Fig. 5, S8 and Movie M5. † The vesicles imaged indicate that oligomers do not only have a strong attraction to the membrane surface, but they also have the effect of binding two adjacent vesicles. The presence of curvilinear protobrils cause the contacting interface of the vesical to extend ('zipping up' the interface) for more than 50 nm (a typical interface of 2000 nm 2 ) causing a distortion and attening of the vesicles at the interface. This results in the creation of a network of liposome linked together by the binding of Ab42, as shown in Fig. 6a and S9. † In the absence of Ab, additional biological density is not observed between vesicles, even when the vesicles are observed contacting each other, Fig. 6b. This is highlighted by the inserts showing the density between vesicles. Furthermore, in the absence of Ab42 oligomers liposome only briey contact each other (ca. 5 nm; a contact area of just 20 nm 2 ) and tend not have a distortion in the spherical nature of the liposome.  Interestingly this effect is more apparent when low levels of oligomers are present, typically for monomer samples that have been allowed to form a limited number of oligomers and pro-tobrils over a 2 h incubation with vesicles. The uid nature of the bilayer facilitates the moving and clustering of the proto-brils at the interface between two vesicles. We note for the lower abundancy oligomer samples, the oligomers and proto-brils are exclusively observed at the interface between vesicles and not distributed elsewhere around the liposome, Fig. 6a. In the situation of preparations with more abundant levels of oligomers (taken at the end of the lag-phase) the membrane becomes so saturated with Ab42 oligomers, the linking between of vesicles is less widespread.

Liposomes imaged by TEM with negatively stained samples
In addition to cryoET the same sample preparations have been imaged by negative-stain TEM, so as to compare the appearance of vesicles using this related technique. The same liposomes, but at 0.05 mg ml À1 have been incubated with recombinant and synthetic Ab42 and Ab40 monomer and oligomers (10 mM monomer equivalent). Two different heavy-metal stains were used; uranyl-acetate and phosphotungstic acid (PTA). Similar to the tomographic images, monomeric Ab42 does not impact the appearance of the membrane, while Ab42 oligomers causes considerable disruptions of the lipid bilayer. The proportions of liposomes perturbed by Ab42 monomers (8%); oligomers (85%); brils (15%), is very similar to that observed in our cryo-ET data. This behaviour is also echoed in the Ab40 negatively stained images, also summarized in ESI Table S1. † Particularly, for images in the presence of uranyl-acetate most vesicles, incubated with Ab oligomers, exhibit a very distorted membrane surface, marked curvatures of the membrane, and the appearance of budding-off of the membrane, as shown in Fig. S10. † We also investigated the effect of recombinant Ab40, there are no signicant differences between the effects caused by Ab42 and Ab40 oligomer preparations, see Fig. S10 and Table S1. † Similar studies with synthetic Ab preparations show the same effects on the lipid bilayer.

Lipid bilayer composition; GM1-ganglioside is important for Ab interaction on lipid bilayers
Next we were interested in how membrane composition might inuence the extent by which Ab42 oligomers interact with the lipid-bilayer. GM1-ganglioside has been shown to have a heightened affinity for Ab. [18][19][20][21] Phosphatidylcholine (PC) with cholesterol (70 : 30 by weight) vesicles were therefore produced in the absence of GM1 and incubated (120 min) with Ab42 oligomers. CryoET images show a large reduction in the extent to which Ab binds to the membrane for GM1 free vesicles. This was consistently observed across many liposomes (n > 50) and multiple preparations. Many vesicles remain largely free of oligomers, Fig. 7a, which instead remain at the air/water interface or on the carbon grid support. Occasionally, but only for images in which vesicles are contacting each other do we observe protobrils on the membrane. Here the protobrils are concentrated only at the interface between vesicles, Fig. 7b. The very different behaviour of the same Ab42 oligomer preparation interacting with lipid bilayer that contain GM1 (2% by weight) is also shown, Fig. 7c.
A similar behaviour can be observed from negative-stained samples of GM1 free vesicles, imaged by TEM. In these images there is a large reduction in the extent of membrane disruption, Fig. S11 and Table S2. † As observed for many vesicles and multiple preparations of Ab42 and Ab40 oligomers. Indeed, most (85%) of the liposomes are unaffected by Ab42 or Ab40 oligomers when GM1 is absent from the PC/cholesterol liposomes. This data suggests that Ab oligomers and proto-brils display a lower affinity for GM1 decient liposomes.
Cholesterol levels are a known risk factor in AD 45 and cholesterol has been suggested to impact interactions of Ab with membranes. 23,24 We therefore also investigated the effect of varying the levels of cholesterol in our vesicle preparations. Phosphatidylcholine vesicles were produced with 2% GM1 and either 9% or 39% (by weight) of cholesterol. These lipid mixtures also produce stable vesicles, as imaged by negativestain TEM. Neither a reduction nor increase in cholesterol had a noticeable effect on the extent of Ab-induced membrane disruption, ESI Fig. S12. † Some images exhibit an extensive amount of budding-off of membrane from the main vesicle, imaged only when using uranyl-acetate negative-stain. In these micrographs there was also the appearance of spherical structures, the majority of which are believed to be micelles typically ca. 12 nm in diameter, the larger micelles/vesicles formed are ca. 20 nm in diameter, Fig. S13. † These, very smooth appearing, circular structures were not detected in control vesicles in the absence of Ab, or images of Ab oligomers alone which are less smooth and regular in appearance. Lipid extraction and small micelle formation is not apparent in our cryoET images, aer incubated with monomer, oligomers or brils; even aer 48 h incubation at room temperature with Ab42 oligomers. In addition, we looked for the appearance bril elongation nucleated at the membrane surface aer 48 h incubation with oligomers, but this was not observed.

Discussion
We report the rst 3D macromolecular description of the impact of Ab assemblies on lipid membranes imaged by cryoET. Preparations of lipids vesicles with various Ab assemblies in aqueous buffer produce excellent quality images in a nearnative environment. In particular, the resolution achieved by cryoET makes it possible to directly distinguish impacts on the outer-and inner-leaet of the bilayer.
We observe very different behaviours of the various Ab assembly forms with wide-spread insertion and carpeting of the membrane by Ab oligomers and curvilinear protobrils, but minimal interaction by monomers or brils (Fig. 1-3 and Table  1). The carpeting of the surface of the membrane and incorporation of oligomers and protobrils, is likely to have a major impact on the bilayer properties, reducing the integrity of membrane and causing leakage and the inux of ions such as Ca 2+ in to the cell. This type of membrane permeability in the presence of Ab oligomers has been widely reported, as indicated by an increase in membrane conductance, 6-8 Ca 2+ inux 9,10 or dye release 11 and loss of cellular homeostasis. [13][14][15] The increase membrane conductance observed has been described as a thinning of the membrane, although we have shown that the insertion and carpeting of Ab on the membrane surface actually has the effect of making the membrane appear thicker although with a reduced lipid content. The insertion of Ab into the bilayer is localized to the outer-leaet of the membrane. This is an important observation as it suggests that extracellular Ab oligomer assemblies do not readily migrate to the cytosol (Fig. 5). If trafficking of Ab oligomers into the cytosol does occur in vivo, our data suggests it happens slowly, or by endocytosis, or an additional membrane protein would need to assist this process. We note that our cryo-ET observations do not rule out monomeric and dimeric Ab diffusion, across the bilayer, which has been reported by a number of studies. [46][47][48] When membranes are imaged by negative-stain with uranylacetate, the Ab oligomers, destabilizes the membrane sufficiently to cause budding-off of vesicles and the formation of micelles ( Fig. S10 and S13 †). This effect has been likened to the action of a detergent. 16,17 Thus, we have a picture of rapid and wide-spread carpeting of membranes by Ab oligomers and protobrils that will destabilize the bilayer and may then lead to extraction of lipids in the long term. Indeed, Ab amyloid plaques have a high lipid content in the AD brain. 49,50 The high contrast density of the curvilinear protobrils facilitates the rst report of 3D images of these assemblies. These images indicate highly irregular, branched structures, that insert into the upper-leaet of the membrane, many of which extend out from the membrane orthogonally (Fig. 4). These structures are 2.7 AE 0.4 nm in cross-section and so could accommodate a single Ab molecule in the one plane of the protolament. Very similar heights, for curvilinear protobrils have been reported using AFM, while a width of 6 nm is reported, which may reect the lower resolution of AFM in this dimension. 51 A recent AFM study of Ab42 assemblies reports long and straight protobrils with a regular twist, that represent the smallest of bril structures, with a typical 5.5 nm height. 52 The higher contrast for the curvilinear assemblies and the smaller oligomers suggests a more compact structure than mature brils (Fig. S3 and S4 †), although brils are known to form tightly packed b-sheets. 53,54 The compact nature of these possible oligomer structures has been reviewed. 55 Based on, in particular, the contrast density, it appears the oligomer assemblies are closely related to curvilinear protobrils and these oligomers may simply be thought of as short curvilinear protobrils, both of which have a diameter of ca. 3 nm and very similar contrast density. While the grey-scale values suggest mature brils may be more structurally distinct (Fig. S4c †). Oligomers with a diameter of 3 nm suggest a protein volume ca. 12-21 kDa in size. These structures can extend to form the short protobrils which may be building blocks to longer curvilinear protobrils. [56][57][58][59] Tetrameric and octameric Ab42 structures have been described which forms an anti-parallel b-sandwich structure in lipid bilayers. 60 The adhesive properties of Ab curvilinear protobrils on the surface of the lipid membrane (Fig. 6) can link vesicles together. Previously unreported, the protobrils become concentrated at the interface between two vesicles. It is interesting to speculate that the 'gluing' together and concentrating of curvilinear pro-tobrils at the interface between membranes might profoundly impact synaptic cles; which are typically separated by a gap of 20-40 nm, a space similar in length to the Ab curvilinear pro-tobrils. We also note this process of linking membrane surfaces could be a mechanism by which Ab oligomers and curvilinear protobrils may spread from one cell surface to another, by attaching to cells or exosomes. This suggest a mechanism by which the prion-like spreading of misfolded Ab might occur in the Alzheimer's disease brain. [61][62][63] Our cryoET data provide a more complete picture of Abmembrane interactions and builds on previous studies, see reviews. [2][3][4] The lack of interaction and impact on the membrane by monomeric Ab is in agreement with AFM studies of supported lipid bilayers 16 and aligns with what is known about the relative cyto-toxicity of Ab monomers and oligomers. 12,64,65 AFM studies have reported widespread extraction of lipid from a mica supported lipid-bilayer which results in the formation of large $50 nm holes. 16,18 This type of extraction and hole formation is not apparent in our cryoET images. The lipid within the bilayer of vesicles can behave as a uid and ll any holes generated, if lipid extraction occurs. While in AFM studies the lipids are more immobile supported on a mica surface, which may explain the different appearance observed.
Our cryoET studies in amorphous ice are well placed to study brils, and show minimal affinity and interaction with the membrane surface (Fig. 1d, 3d, S5c and S6d †). Lateral association of brils on the surface of lipid membranes has been reported. However, images acquired using negative-stain and AFM are obtained by drying of samples by blotting, this causes brils to lay-down on the membrane surface as water is lost. Thus, for these images the interaction of brils may appear more widespread. 16,18,33,35,36 Amyloid bril membrane interactions have been reported, by cryoET for b 2 M brils. 30 In this study the ends of brils tend to interact with the surface of the membrane. We have also observed this effect for Ab, although the interactions with oligomers are considerably more marked. The ends of brils may have more exposed hydrophobic sidechains, while the lateral surface of the bril has less of an affinity for the membrane.
There are a number of studies that describe the elevated affinity of Ab for GM1-ganglioside compared to other lipids. [18][19][20][21][22] We image, for the rst time, the very different impact Ab oligomers have on lipid membranes, which do not contain GM1 ( Fig. 7 and S11 †). Ab is negatively charged, at neutral pH, and electro-static attraction to the polar carbohydrate groups may promote the Ab-bilayer interaction. This effect may be important as GM1-ganglioside is particularly abundant in the outer-leaet of neuronal plasma-membranes. 66 In conclusion, 3D nanoscale imaging of liposomes suspended in a near-native aqueous environment have revealed new details and insights in to the interaction of Ab assemblies with lipid membranes. Wide-scale impacts on the membrane are restricted to oligomeric and curvilinear protobrillar structures that saturate the outer-leaet of the membrane, while in the case of isolated monomers, or even brillar Ab42, the lipid bilayer remains relatively unperturbed. This carpeting and insertion has previously been shown to impact membrane integrity and cellular homeostasis, [6][7][8]13,14 and is in line with the relative cytotoxicity of Ab oligomers compared to brillar assembly states. 12,64,65 The conclusions drawn here may have many parallels for anti-microbial peptides, 67,68 and other amyloid forming proteins such as: amylin, 25-27 alpha-synuclein, 28 mammalian prion protein, 29 b 2 M 30 and serum amyloid A. 31 Therapeutic molecules, that block insertion of Ab oligomers into membranes may help maintain neuronal homeostasis and slow the cascade of events that culminates in dementia.

Ab sample preparation
The puried lyophilized Ab40 and Ab42 both recombinant and synthetic peptides were solubilized at 0.7 mg ml À1 in water at pH 10. See ESI methods for details. † Monomeric Ab was isolated using size exclusion chromatography with a Superdex 75 10/300 GL column (GE Healthcare). Ab40 and Ab42 monomer (10 mM) were placed in a 96-well plate in NaCl (160 mM) and HEPES (30 mM) buffer at pH 7.4. Ab40 and Ab42 prebrillar assemblies with predominantly oligomeric and curvilinear protobril structures, were obtained from the well plate towards the end of the lag-phase, as monitored by ThT uorescent dye in separate wells. Oligomeric samples were used immediately or stored at À80 C to halt further assembly. Lag-phase mixed prebrillar assembles were characterize by cryoET and TEM, ESI Fig. S3. † At equilibrium (as indicated by ThT uorescence in separate wells) Ab assemblies had the typical amyloid brous appearance according to TEM, see ESI Fig. S4. † Fibril preparations were centrifuged using 100 kDa molecular cut-off lter (Amicon Ultra) to remove any low molecular weight oligomers. Fibril preparations have a low level of Ab monomer and oligomer content.

Cryo electron tomography (cryoET)
Large unilamellar vesicles (LUVs) were produced using an extrusion method described previously. 16,36 See ESI methods for details. † The nal lipid vesical concentration was 0.5 mg ml À1 for cryoET imaging with recombinant Ab42 (5 mM monomer equivalent). Ab42 was incubated with the vesicles for 10 and 120 min for Ab42 monomer preparations; 120 min and 48 h for lag-phase oligomers preparations; and 120 min for Ab42 brils before plunge-freezing. Vesicle solutions were plunge-frozen onto Quantifoil R2/2 holey carbon grids using a Thermo Fisher Vitrobot.
Electron cryo-tomography was performed using a Thermo Fisher Glacios TEM operating at 200 kV, equipped with a 4k Â 4k Falcon 3EC direct electron detection camera at a magnication of 73k, corresponding to a pixel size of 1.9Å at the specimen level. Specimens were tilted from approximately À60 to +60 with a 3 increment using the dose symmetric scheme. The defocus was set between 3 and 4 mm, and the total dose for each tilt series was approximately 100 eÅ À2 . Final tomograms were binned 4Â, with a pixel size of 7.6Å. Tomographic slices were typically shown as an average of 10 slices, 7.6 nm thick. Experimental detail of image processing and TEM can be found in ESI methods.