Ion Channel Formation by Amyloid-β42 Oligomers but Not Amyloid-β40 in Cellular Membranes*

A central hallmark of Alzheimer's disease is the presence of extracellular amyloid plaques chiefly consisting of amyloid-β (Aβ) peptides in the brain interstitium. Aβ largely exists in two isoforms, 40 and 42 amino acids long, but a large body of evidence points to Aβ(1–42) rather than Aβ(1–40) as the cytotoxic form. One proposed mechanism by which Aβ exerts toxicity is the formation of ion channel pores that disrupt intracellular Ca2+ homeostasis. However, previous studies using membrane mimetics have not identified any notable difference in the channel forming properties between Aβ(1–40) and Aβ(1–42). Here, we tested whether a more physiological environment, membranes excised from HEK293 cells of neuronal origin, would reveal differences in the relative channel forming ability of monomeric, oligomeric, and fibrillar forms of both Aβ(1–40) and Aβ(1–42). Aβ preparations were characterized with transmission electron microscopy and thioflavin T fluorescence. Aβ was then exposed to the extracellular face of excised membranes, and transmembrane currents were monitored using patch clamp. Our data indicated that Aβ(1–42) assemblies in oligomeric preparations form voltage-independent, non-selective ion channels. In contrast, Aβ(1–40) oligomers, fibers, and monomers did not form channels. Ion channel conductance results suggested that Aβ(1–42) oligomers, but not monomers and fibers, formed three distinct pore structures with 1.7-, 2.1-, and 2.4-nm pore diameters. Our findings demonstrate that only Aβ(1–42) contains unique structural features that facilitate membrane insertion and channel formation, now aligning ion channel formation with the differential neurotoxic effect of Aβ(1–40) and Aβ(1–42) in Alzheimer's disease.

Alzheimer's disease (AD) 3 is the most prevalent fatal neurodegenerative disease worldwide and accounts for up to 80% of all dementia cases. Global forecasts estimate the total number of people living with dementia to grow from 47 to 132 million by 2050 (1). Amyloid-␤ (A␤) peptide is a principal component of extracellular amyloid plaques found in the brain interstitium and a key hallmark of AD. A critical step in disease progression is the misfolding and self-assembly of monomeric A␤  into oligomeric and fibrillar aggregates (2).
The major A␤ alloforms found in extraneuronal plaque deposits are A␤  and the less prevalent A␤ , which accounts for ϳ10% of secreted A␤ (3). However, a range of evidence points to A␤  as the principal neurotoxic form of A␤. Familial mutations that result in early onset AD cause an increase in the ratio of A␤  to A␤(1-40) (4). Elevated A␤  plasma levels have also been correlated with the progression of late onset forms of the disease (5). A␤  is significantly more neurotoxic than A␤  both in vivo and in neuronal cell culture (6 -9), and memory impairment is believed to be driven by A␤  disruption of long term potentiation (10,11).
The following is a selection of commonly shared channel features. A␤ channels are large in size (with a conductance often greater than 250 pS), remain open for long periods of time (Ͼ500 ms), display spontaneous voltage-independent activation, and transition between multiple conductance states (28,29). There is also structural evidence to support the ion channel hypothesis. Channel-resembling structures composed of A␤ have been shown to span supported lipid bilayers by atomic force microscopy (AFM) (19,25,30), and these findings are thought to be dependent on the formation of annular pore-like A␤ assemblies with an 8 -25-nm outer diameter and a 2-6-nm pore opening (31,32). Similar annular oligomers have been reported for ␣-synuclein from Parkinson's disease that are also capable of forming ion channels (31,33,34). A shared mechanism of ion pore toxicity has therefore been proposed (31).
Despite the reported differential synaptotoxic effect of A␤  and A␤ , ion channel studies have largely focused on artificial lipid bilayers, and little data exist to directly compare channel forming properties of different A␤ isoforms in a cellular membrane preparation. Furthermore, the majority of ion channel studies were performed before the importance of assembly state in neurotoxicity was fully appreciated. Attention has shifted to a series of A␤(1-42) oligomers ranging from 9 to 200 kDa that have been proven to be more toxic than monomeric and fibrillar forms of A␤(1-42) (10,11,35,36). Here, we assess the relative channel forming ability and properties of a series of A␤  and A␤  preparations in membrane patches excised from a HEK293 immortalized cell line. For oligomeric preparations of A␤(1-42), we recorded large, non-selective ion channels, but channels were not observed for monomeric or purified fibrillar forms of A␤ . However, neither monomeric, oligomeric, nor fibrillar forms of A␤  formed ion channels under the same conditions. Our observations make a direct link between the unique ability of A␤  assemblies to form ion channels with a known in vivo synaptotoxicity for A␤(1-42) but not A␤(1-40).  but Not Oligomeric A␤(1-40)-Patch pipettes were backfilled with 5 M A␤ preparations. Patches of membrane were then excised from HEK293 cells, and A␤ was free to diffuse toward the extracellular face of the membrane within the pipette. Transmembrane currents were measured by clamping the patch at a series of step voltage potentials. Six different A␤ preparations were studied with between 20 and 49 patches of membrane pulled for each. Monomer, oligomer, and fiber preparations of both A␤(1-40) and A␤(1-42) were studied. The assembly state of A␤ is critical in exerting synaptic dysfunction and neurotoxicity, it was therefore important to carefully characterize the A␤ assembly types that were presented to membrane patches. A␤ preparations were characterized using transmission electron microscopy (TEM), size exclusion chromatography (SEC), and the fiber-specific thioflavin T (ThT) fluorescent dye (Fig. 1).

A␤ Ion Channels Are Observed with Oligomeric A␤
Large, voltage-independent A␤ channels with clear open/ closed step current transitions form in the presence of oligomeric and fibrillar A␤(1-42) assemblies (Fig. 1, e and f). Channels were found in 17 of 49 (35%) patches with A␤(1-42) oligomer and eight of 24 patches (33%) with A␤(1-42) fiber when monitored for 30 min. Importantly, ion channels were not observed for SEC-purified A␤(1-42) monomer (n ϭ 24) (Fig. 1d). We next wanted to determine whether channels formed by A␤(1-42) fiber preparations were in fact due to the presence of small oligomers observed to co-exist in A␤  fiber preparations by TEM (Fig. 1f). Oligomers were removed from the fiber preparation by centrifugation. The pelleted A␤(1-42) fibers were then resuspended into patch clamp buffer for delivery to the membrane at 5 M. The channel forming ability of the oligomer-depleted A␤(1-42) fibers was significantly reduced, relative to the unpurified fiber sample, with only one channel observed in 24 patches (4%). We also exposed the excised membranes to A␤(1-40) and found that A␤(1-40) monomer (n ϭ 20), oligomer (n ϭ 20), and fiber (n ϭ 20) preparations did not form any A␤-associated ion channels in a 30-min recording period. A␤ channels were also not present in A␤-free buffer-exposed control patches (n ϭ 20). Fisher's exact tests including Bonferroni correction showed channel formation by A␤(1-42) oligomers to be significant against A␤(1-40) oligomer (p ϭ 0.002) and A␤(1-42) monomer (p ϭ 0.0007). Also, A␤(1-42) fiber preparations were significantly different in their ability to form channels from A␤(1-40) fibers (p ϭ 0.005). However, removal of oligomers from A␤(1-42) fiber samples reduced the probability of ion channel formation significantly when compared with uncentrifugedA␤(1-42)fiberspreparationsthatcontainedoligomers (p ϭ 0.01) (Fig. 2).
Next, we wanted to investigate the ion channel forming properties of a more dilute A␤(1-42) oligomer preparation. A 10-fold dilution at 0.5 M monomer equivalent concentration of A␤(1-42) oligomers was delivered to membranes, and channel currents were observed in as many as six of 40 patches (15%). A further 10-fold dilution to 50 nM caused some reduction in the number of channels observed, but it was not proportional to the 100-fold dilution factor with channels observed in 10% of patches (n ϭ 20).
Endogenous ion channels were observed for many, but not all, membrane patches; they were much smaller in size, typically less than 85 pS, and exhibited voltage-dependent activation ( Fig. 3). A␤ channels were markedly larger in size with the smallest observed channel exhibiting a conductance of 270 pS.
A␤  Channel Conductance, Pore Diameter, and Rate of Formation-Channels formed by A␤(1-42) oligomer commonly remained stable in the membrane for a period of many minutes (Ն10 min); open times were frequently in excess of 1.5 s, a value limited only by the duration of the applied voltage clamp protocol. The majority of channels exhibited flicker in which the channel transitioned briefly to lower conductance levels ( Fig. 4a). Flicker transitions were observed at all applied depolarizing and hyperpolarizing potentials.
Typical current traces recorded for A␤  oligomers are shown in Fig. 5a for three representative channels that exhibit three distinct conductances. Open/closed step current transitions at each clamped membrane potential were measured to determine operating channel conductance by generating a current-voltage relationship (37). The straight line fits used to determine the mean conductance of each channel are shown in Fig. 5b.
A plot showing the distribution of channel conductance (Fig.  5c) for a total of 34 different channels indicates that A␤(1-42) oligomers form three or more distinct channel sizes. The primary channel type (35% of channels) produced a median conductance of 337 pS (ranging between 310 and 350 pS) with at least two more subgroups of channels exhibiting 490-(477-497-pS) and 627-pS (608 -642-pS) conductance. In some cases, channel currents transition between two primary channel sizes (Fig. 4b). Current measurements were therefore made from the  most commonly observed state in which the channel remains stable.
Approximate pore size can be calculated from the channel conductance, assuming the pore is cylindrical with a bilayer spanning length assumed to be 7 nm (38). This model has been adapted to estimate channel diameter in large ion channels (see Equation 1 under "Experimental Procedures"). Using Equation 1, the three channel subtypes have estimated pore diameters of 1.7, 2.1, and 2.4 nm respectively (Fig. 5c). Groupings of conductance indicate discrete channel sizes with narrow ranges of diameters: 1.62-1.73, 2.04 -2.09, and 2.34 -2.41 nm. Standard deviations are 0.03 nm for all three channel types identified. There are also a limited number of channels with even higher conductance, ranging between 737 and 1039 pS with calculated diameters between 2.6 and 3.2 nm.
On delivery of A␤(1-42) oligomer to the membrane, channels do not form instantly; there is a delay of more than 2 min before channel current is observed (Fig. 6). Short lived current spikes associated with membrane destabilization would often precede the appearance of an A␤ ion channel. The bulk of the channels formed between 4 and 8 min following membrane exposure with 50% of channels formed within 8.5 min of exposure. The rate at which channel formation occurs appears to be unrelated to channel conductance. A Pearson correlation r value of 0.04 suggests no association between conductance and time elapsed before channel formation.
Channels formed with A␤(1-42) oligomers and oligomers from unpurified fiber samples had similar electrophysiological characteristics. In particular, channels formed by A␤(1-42) oligomer and fiber preparations were similar in size with comparable median conductances. Channels in each group formed at similar rates with the majority arising within the first 10 min.
A␤  Channels Are Not Cation-selective-The Goldman-Hodgkin-Katz equation, adapted by Spangler (39), links the concentration and permeability of Na ϩ , Ca 2ϩ , K ϩ , and Cl Ϫ ions to the reversal potential (E rev ) of a membrane (37). An E rev of 0 mV was measured in patches containing A␤(1-42) ion channels in a symmetrical pipette and bath solution. The composition of ionic solution on the internal membrane face was then altered by perfusion with two separate patch clamp buffers, which had 50 (n ϭ 2; 642 and 844 pS) and 75% (n ϭ 2; 627 and 1039 pS) of NaCl replaced with tetraethylammonium chloride. According to the Goldman-Hodgkin-Katz equation (see Equation 2 under "Experimental Procedures"), the E rev of a purely cation-selective channel would be expected to shift to  . A␤(1-42) oligomer forms discrete ion channels of varying pore diameter. a, current traces recorded from separate A␤(1-42) channels that represent three commonly observed channel sizes. b, conductance was calculated by generating a current-voltage relationship for each channel. Each data set is composed of 30 channel current measurements across six membrane potentials and fitted with a linear regression. c, conductance distribution of channels formed in the presence of A␤(1-42) oligomer and oligomercontaining fiber samples. The primary y axis represents a rank order of conductance with increasing magnitude for 34 channels. Three discrete channel subtypes are highlighted (within a 50-pS range) and labeled by a median calculated pore diameter. A secondary y axis accounts for bar chart representation of the proportion of channels formed within a 50-pS bin size. ϩ20 and ϩ28 mV for each respective solution. A modest shift of ϩ6 mV observed in each case suggests that A␤(1-42) channels are permeated by both anions and cations (Fig. 7a).
Evidence of non-selectivity is further supported by depletion of both anions and cations. Three more channel-incorporated patches were perfused with patch clamp solution that had an 80% volume replacement with equimolar sucrose. Under these conditions, the E rev of a purely cation-selective channel would be expected to shift ϩ40 mV (Equation 2). However, a small E rev of Ϫ8 mV was measured (336, 622, and 642 pS) (Fig. 7b).

Discussion
The ability of A␤ to form membrane-spanning ion channels is a mechanism by which A␤ might exert synaptic neurotoxicity (17)(18)(19). However, the relationship between A␤ pore formation and AD pathology has been questioned as previous studies have made little distinction between the abilities A␤(1-40) and A␤  to form ion channels (20,25,26,40). Hence, a mismatch has existed between observed ion channel formation by A␤(1-40) (17)(18)(19)(20)(21) and what is known about the relative pathogenicity of A␤(1-40) and A␤  in vivo.
Here, we show that only A␤(1-42) oligomer preparations form channels; A␤(1-40) oligomer and fibers do not form channels in HEK293 cell-excised membrane. Therefore, our study aligns the ion channel hypothesis of A␤ toxicity with studies that have established oligomeric A␤(1-42) as the toxic entity of AD pathology (10,11,36,41). Furthermore, we provide evidence to suggest that monomeric A␤(1-42) is incapable of forming ion channels from solution, whereas A␤(1-42) oligomer will readily insert into the lipid bilayer to form membrane-spanning pores.
The ion channel properties observed in this investigation are consistent with those observed in other voltage clamp studies. In particular, the channels were of large conductance (270 -1039 pS), voltage-independent, and transiently flickered between multiple conductance states (17,20,29). A question remains as to why others have reported A␤(1-40) channels (17-21, 40) but we have only observed channels for A␤ (1-42).
Most of the disparities in the literature may be explained by the method in which A␤ is delivered to the membrane. A␤(1-40) channels were typically generated using peptide already reconstituted into vesicular lipid bilayers (17)(18)(19)21), and as such channel formation would not be dependent on the insertion of A␤ into a preformed lipid bilayer. Our data suggest the additional Ile 41 -Ala 42 residues found in A␤  are important in the formation of cell membrane-spanning ion channels. Although there are some reports of channel formation from solution for A␤   (20,26,40,42), the majority of these studies report A␤ insertion into artificial model bilayers rather than cellular membrane. Importantly, this is the first study to report A␤(1-42) ion channels in excised cellular membrane, and in this case membrane composition may favor A␤(1-42) insertion given that single molecule imaging has previously demonstrated A␤(1-42) oligomers to have a greater affinity for hippocampal membrane than A␤(1-40) (43). Molecular dynamics simulations also suggest A␤(1-42) to be twice as efficient at inserting into lipid bilayers when compared with A␤(1-40) (44). It is therefore plausible that a differential membrane interaction and insertion of A␤ isoforms may explain a lack of ion channels formed by the A␤(1-40) preparations studied here.
Size exclusion chromatography-purified A␤(1-42) monomer preparations did not form ion channels. Similarly, removal of low molecular weight, diffusible oligomer from A␤(1-42) fibers almost completely removed the ability of fiber preparations to produce ion channels. This indicates channel formation to be dependent on the self-assembly of A␤(1-42) into oligomeric structures that are capable of inserting into the membrane to form membrane-spanning pores. Previous studies have assumed that monomeric conformations of A␤ form ion channels (17,18,21,24,29,40). However, monomeric A␤ preparations were not SEC-purified or structurally characterized, and the solubilization protocols used are likely to have generated oligomeric assemblies. A more recent study has implicated a select population of 4-mer to 13-mer assemblies in triggering A␤-induced ion flux for both A␤(1-40) and A␤  in planar lipid bilayers (42); however, perforations were sporadic in nature. Current transitions were not well defined, and open pore lifetimes as low as several milliseconds were recorded, suggesting greater pore instability compared with ion channels observed here.
Electrophysiological characterization of A␤ channels recorded in this study allows comparisons to be made with proposed structural features of A␤ pores based on AFM micrographs. Pore size diameter can readily be estimated using a measured channel conductance. Here, the median conductances of the three main channel subtypes were measured to be 337, 490, and 627 pS, which imply approximate pore diameters of 1.7, 2.1, and 2.4 nm, assuming a lipid bilayer of 7 nm. These values are closely in line with pore diameters taken from molecular dynamics simulations and AFM studies (19,25,30,45). Spacefilling simulations predict a preferred A␤ 16-mer to 24-mer channel configuration with estimated pore diameters of ϳ1.6 and ϳ2.5 nm, respectively (45). Recently, a combination of biophysical studies (CD, protease K digestion, and solution NMR) on pore-forming A␤(1-42) oligomers has revealed data consis-  rev was recorded by measurement of channel current over seven voltage steps. a, the E rev of channels preformed in symmetrical buffer was measured to be 0 mV (black Ⅺ, n ϭ 4). Replacement of NaCl with tetraethylammonium chloride by 50% (blue E, n ϭ 2) and ϳ75% (redछ, n ϭ 2) minimally shifts E rev to 6 mV. b, a second solution in which 80% of anions and cations were also replaced with non-polar sucrose was also exposed to three other channels. E rev shifted from 0 mV in symmetrical solution (black Ⅺ, n ϭ 3) to Ϫ8 mV (yellow E, n ϭ 3). Error bars represent S.D. tent with a ␤-barrel conformation. In the future, these preparations in a membrane-mimicking environment may be amenable to 3D structure determination (27). Channels were also of comparable size with those observed by AFM; channel-resembling structures that spanned artificial lipid bilayers had 1-2-nm measured pore diameters (19,25,30). Porelike structures of comparable scale with the larger channels formed in this study have also been observed (31,32). Annular assemblies with a 2.5-4-nm pore diameter have been found localized to the membrane of AD model mice and human AD frontal cortex (46). Transitions between conductance states (Fig. 4b) suggest that A␤ ion channels are dynamic structures that can change in pore dimensions. Similar transitions have also been reported for pores formed by the misfolding amyloid protein ␣-synuclein from Parkinson's disease (33,34).
Ion channel formation follows a biphasic process; channels were not observed within the first few minutes after exposure to A␤. This delay may represent the time taken for A␤ oligomers to insert into the membrane and perhaps undergo structural rearrangement of A␤ into membrane-spanning pores. Most channels form within 10 min of exposure to A␤. The mechanism by which A␤ coalesces into a membrane-spanning channel assembly is yet to be defined (22,47). A␤ is capable of forming channel-resembling structures in the form of annular pore assemblies (31,32). However, preformed annular assemblies are surprisingly ineffective at permeabilizing lipid bilayers. Instead, spherical oligomers that precede annular pores are required (32). The lack of ion channels for monomeric A␤(1-42) may be explained by a relative inability of monomer to insert into the membrane. Indeed, previous data suggest A␤  oligomers to preferentially interact with hippocampal cell membranes relative to A␤(1-42) monomer (43). Neutron diffraction studies support this as A␤(1-42) monomer was seen not to insert into cholesterol -containing bilayers; A␤  insertion was seen only after monomeric self-assembly into oligomers (48).
Cation selectivity has been reported for A␤  pores (17,20,26,40). A␤(1-42) channels formed in this study exhibited a lack of selectivity for cations; this is in agreement with a previous model lipid bilayer study (26). A difference in ion selectivity between A␤(1-40) and A␤(1-42) could be due to changes in localized membrane composition. In particular, increased lipid bilayer cholesterol has been suggested to shift A␤(1-42) channel selectivity in favor of anion permeation (49). Given that molecular dynamics simulations predict A␤  to insert more readily into cholesterol-containing phosphatidylcholine membranes (44), the more hydrophobic A␤(1-42) could conceivably target non-polar cholesterol raft environments and exhibit non-selectivity of ions as a result. Alternatively, cation selectivity could be altered by changes to the structural configuration of negatively charged residues postulated to reside in the pore lining (50,51).
A similar incidence of channel formation was observed with channels forming in 35 and 33% of patches for A␤  oligomer and A␤(1-42) fiber preparations, respectively, that had not had the oligomeric component depleted by centrifugation. A␤(1-42) oligomer and fiber preparations appear to be equally effective in loading a membrane with A␤. This is somewhat surprising given that the concentration of free A␤(1-42) oligomer is significantly lower in fiber preparations. This suggests the membrane is still saturated with channel-forming assemblies, even at lower concentrations. This hypothesis is supported by subsequent experiments in which 10-fold diluted 500 nM A␤(1-42) oligomer preparation also yielded channels in 15% of patches. Even when A␤(1-42) oligomers are diluted 100-fold to 50 nM, an appreciable proportion of channels (10%) is observed. Therefore, at 500 nM-5 M, the proportions of patches in which A␤ channels formed are, to a large extent, independent of A␤ oligomer concentration.
It is interesting to speculate why, if the membrane is saturated, we only observed channels in 35% of patches. One possibility is that channel formation is dependent on bilayer composition found only within a third of patches pulled. Perhaps membrane proteins can modulate channel formation (52). Others have highlighted the importance of membrane composition for effective A␤ insertion and perforation of model lipid membranes, specifically pointing to GM1 and cholesterol as key mediators of membrane interaction and perforation (16,53,54). Such sites may be critical in the formation of ion channels in cells, although we note that ion channels formed in synthetic bilayers do not require GM1 or cholesterol (17,26).
An apparent membrane saturation would be consistent with a study that suggests that A␤ saturates a model lipid bilayer at very low peptide to lipid ratios; as little as 2 nM was enough to saturate all membrane binding sites (55). Membrane patches excised in this study have an approximate surface area of 7 m 2 . A typical neuron with a membrane surface area of 250,000 m 2 might theoretically incorporate more than 10,000 A␤(1-42) channels with the cytotoxic potential to increase intracellular calcium levels.
The channel forming capacity of oligomeric A␤(1-42) assemblies, but not monomeric A␤  or any form of A␤(1-40), makes a direct correlation between pore formation and the known pathology of various A␤ assemblies in AD. This adds weight to the amyloid pore hypothesis, and as such molecules that block these ion channels may be therapeutic in protecting neurons against the synaptotoxic effects of A␤.

A␤ Peptide Preparations and Characterization
Monomer-Lyophilized A␤(1-40) and A␤(1-42) were purchased from Cambridge Research Biochemicals and EZBiolab Inc. A␤ peptides were synthesized using solid phase F-moc (N-(9-fluorenyl)methoxycarbonyl) chemistry, producing a single elution band in HPLC with correct mass verified by mass spectrometry. Peptides were solubilized in ultrahigh quality (UHQ) water (0.7 mg/ml; pH 10.5) and left at 4°C for 12 h. It was necessary, particularly in the case of A␤ , to remove any remaining nucleating, oligomeric aggregates by SEC (Superdex 75 10/300 GL column, GE Healthcare) to generate a seed-free preparation.
Seed-free A␤, termed here as monomeric, had no ThT fluorescence signal and exhibited a clear lag phase to the nucleation polymerization reaction (Fig. 1a). Furthermore, SEC-purified A␤ gave a single elution band, and assemblies were not observed by negative stain TEM (Fig. 1). Concentrations were determined by absorbance of stock solutions at 280 nm (⑀ ϭ 1280 M Ϫ1 ⅐ cm Ϫ1 ) using a Hitachi U-3010 spectrophotometer. Monomeric samples were stored immediately after SEC elution at Ϫ80°C.
Fibers-Amyloid fiber preparations of A␤  and A␤  were generated in a 96-well plate incubation of 10 M peptides, 160 mM NaCl, and 30 mM HEPES (pH 7.4). Assembly kinetics were monitored with ThT fluorescence, and adjacent sample wells with no ThT added were used in all patch clamp experiments. At equilibrium, A␤ assemblies have the typical amyloid fibrous appearance according to TEM. A␤  fibers were many micrometers in length and 10 -20 nm in diameter (Fig. 1c). A␤  is typically more heterogeneous in appearance with oligomeric and protofibrillar assemblies present in addition to amyloid fibers (Fig. 1f).
Oligomers-Similarly, A␤(1-40) and A␤(1-42) assemblies with predominantly circular oligomeric structure were obtained from the well plate toward the end of the lag phase as monitored by ThT fluorescent dye in separate wells (Fig. 1a). Samples were immediately stored at Ϫ80°C to prevent further assembly. TEM images show a large number of circular oligomeric assemblies between 5 and 20 nm in diameter (Fig. 1, b and  e). A preparation of A␤(1-42) oligomers and protofibers was also obtained for A␤  by solubilizing at pH 10.5, but these samples were not purified by SEC. These oligomer samples did not have a clear lag phase before elongation of fibers (Fig. 1a). Numerous protofibrillar assemblies, typically 200 nm long, were also present in addition to a limited number of short fibers.
Oligomer-depleted Fibers-Oligomer-depleted fiber preparations were generated by centrifugation of A␤(1-42) fiber samples at 16,100 ϫ g for 5 min as described previously (56). The supernatant, which contained small diffusible oligomers, was removed, and the fiber pellet was resuspended for use in additional patch clamp experiments at a final concentration of 5 M.

A␤ Assembly
ThT fluorescence upon fiber formation was measured using BMG-Galaxy and BMG-Omega FLUOstar fluorescence 96-well plate readers. Well plates were subjected to mild double orbital shaking for 30 s every 30 min followed by a fluorescence reading, 20 flashes per well per cycle with 4-mm orbital averaging. Fluorescence excitation and emission detection were at 440 and 490 nm, respectively. ThT fluoresces when bound to amyloid fibers to give a fluorescent signal proportional to the amount of amyloid fiber present. It has been shown that ThT does not markedly affect the formation or kinetics of fiber growth (57).

Size Exclusion Chromatography
A Superdex 75 10/300 GL column was used to purify and elute a monomeric fraction of solubilized A␤ preparations using Ä KTA FPLC. A stock 90 M A␤ solution in UHQ water was loaded onto the column, and the monomer peak was eluted in a solution of 160 mM NaCl, 30 mM HEPES using a 0.5 ml/min flow rate at 4°C. Eluted monomer had a typical concentration of 30 M.

Transmission Electron Microscopy
A␤-containing preparations were aliquoted (5 l) onto glowdischarged carbon-coated 300 mesh grids using the droplet method and washed with UHQ water, and then a negative phosphotungstic acid (2% (w/v), pH 7.4) stain was applied before a final wash step. Images were taken with a JEOL JEM-1230 electron microscope operated at 80 keV paired with the Olympus iTEM software package. Representative images were selected from multiple images taken across multiple grids and fields.

Cell Culture
HEK293 immortal cells of neuronal origin were cultured in 30-ml cell culture flasks. Cells were grown in a 37°C in a 5% CO 2 incubator in Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% fetal bovine serum, 200 units penicillin, and 0.2 mg/ml streptomycin. On reaching confluence, Ca 2ϩ -and Mg 2ϩ -free phosphate-buffered saline (pH 7.2) was used to dissociate cells from culture flasks for splitting. Cultured cells were split every 5-7 days and used between passages 22 and 65. With each round of splitting, a fraction of cells was plated into 35-mm diameter Petri dishes and maintained in supplemented DMEM until the day of a patch clamp experiment. Plated cells were cultured for 2-3 days before use in patch clamp experiments. Reagents and media were purchased from Sigma-Aldrich and ThermoFisher Scientific (Invitrogen).

Patch Clamp Recording
Patch clamp recordings in voltage clamp mode were made from excised membrane of HEK293 cells. Patch pipettes were backfilled with patch clamp buffer containing a series of A␤ preparations containing 5 M, 500 nM, or 50 nM monomer equivalent concentration of A␤ at pH 7.4. A␤ was free to diffuse toward the extracellular face of the membrane within the patch pipette, and transmembrane currents were measured by clamping the patch at a series of voltage potentials between ϩ60 and Ϫ60 mV. Patches of membrane were excised, in an inside-out configuration, into a 35-mm dish containing buffer of identical ionic composition. A stepwise voltage ramp protocol using an Axopatch 200B amplifier (Axon Instruments, Union City, CA) was applied via personal computer using PCLAMP-10 software (Axon Instruments). Data were further processed using Clampfit software package, and a low pass boxcar filter was typically applied. Patches were pulled from thick walled filament borosilicate glass (Harvard Apparatus, Edenbridge, Kent, UK) with a needle diameter of ϳ1-3 m and a resistance of 4 -6 megaohms when filled with recording solution. Junction potentials generated at boundaries of ionic asymmetry were accounted for using an applied pipette offset potential. Recordings were sampled at a rate of 2 kHz with 500-s intervals with a low pass four-pole Bessel filter frequency of 1 kHz.
In the excised patch, the holding potential was set to 0 mV. Recordings were made in symmetrical solution adjusted to pH 7.4 containing 121.4 mM NaCl, 10 mM CsCl, 9 mM NaHEPEs, 1.85 mM CaCl 2 , 1.87 mM MgCl 2 , 2.16 mM KCl, and 0.1 mM EGTA. Transmembrane patch currents were recorded for 30 min, and if current spikes indicative of membrane destabilization were observed in this time period, the recordings were extended to 45 min. The majority (90% of channels) were formed within the first 30 min.
Channel current transitions between open and closed states were measured using Clampfit software at each membrane potential. An average from typically five measurements for any given membrane potential was used to generate a current-voltage relationship. The slope of the current-voltage relationship was calculated to estimate channel conductance (pS). Each conductance value for an individual channel is obtained from typically five or more current measurement values at the six different potentials; a straight line is then fitted to the Ͼ30 data points to obtain the conductance value. Fisher's exact tests (SPSS software) were performed with a Bonferroni correction to assess statistical significance comparing each A␤ preparation (58).

Calculation of Channel Pore Diameter Using Channel Conductance
Theoretical pore diameter has been calculated using a model developed by Hille (38) and adapted by Cruickshank et al. (59). Pore diameter (d) is calculated in a solution of defined resistivity () for a pore of measured length (l) and ionic conductance (g) (Equation. 1). A membrane-spanning conformation with a pore length (l) of 7 nm (45, 51) and a solution resistivity of 80 ohms cm has been assumed (60).

Calculation of Channel Selectivity
Patches were pulled in solution of symmetrical ionic composition, and then ionic concentration gradients were generated by gravity perfusion of solutions of altered composition onto the internal face of the membrane. Ion channel selectivity was probed by measurement of a reversal potential.
Channel selectivity has been calculated using a modified Goldman-Hodgkin-Katz equation (39) (Equation 2). The Goldman-Hodgkin Katz equation (Equation 2) has been used to predict membrane E rev given the concentration and permeability (P) of ions on the intracellular ([ion] int ) and extracellular ([ion] ext ) face of a membrane. The reversal potential is the membrane potential in which ions are at equilibrium and zero current flows. E rev measured in the current investigation deviated from an expected E rev according to previous studies. The cation-selective ion permeability ratio of 1 Na ϩ :1.3 K ϩ :1.3 Ca 2ϩ :2 Cs ϩ calculated by Arispe et al. (17) was used to predict an expected E rev under our own experimental conditions. Junction potentials appearing on ion replacement were nulled by zeroing the holding current at 0 mV and assumed no ionic selectivity for the patch leakage current. Measured channel currents are normalized against their maximum recorded channel current.
The universal gas constant, R, is 8.314 J ⅐ K Ϫ1 ⅐ mol Ϫ1 ; temperature, T, is 296 K; and the Faraday constant, F, is 96,485 coulombs ⅐ mol Ϫ1 . y is solved as follows.