The Double-Layered Structure of Amyloid-β Assemblage on GM1-Containing Membranes Catalytically Promotes Fibrillization

Abstract

Alzheimer’s disease (AD) is associated with progressive accumulation of amyloid-β (Aβ) cross-β fibrils in the brain. Aβ species tightly associated with GM1 ganglioside, a glycosphingolipid abundant in neuronal membranes, promote amyloid fibril formation; therefore, they could be attractive clinical targets. However, the active conformational state of Aβ in GM1-containing lipid membranes is still unknown. The present solid-state nuclear magnetic resonance study revealed a nonfibrillar Aβ assemblage characterized by a double-layered antiparallel β-structure specifically formed on GM1 ganglioside clusters. Our data show that this unique assemblage was not transformed into fibrils on GM1-containing membranes but could promote conversion of monomeric Aβ into fibrils, suggesting that a solvent-exposed hydrophobic layer provides a catalytic surface evoking Aβ fibril formation. Our findings offer structural clues for designing drugs targeting catalytically active Aβ conformational species for the development of anti-AD therapeutics.

Introduction

Neurodegenerative diseases, such as Alzheimer’s disease (AD), Parkinson’s disease, and Huntington’s disease, are ascribed to pathogenic molecular processes involving conformational transitions of amyloidogenic proteins into toxic aggregates such as oligomers and fibrils.¹⁻³ In addition to certain oligomers being deleterious aggregates,^4,5 amyloid fibrils have been demonstrated to exert cytotoxicity and play key roles in the persistence, progression, and propagation of amyloid diseases.⁶ Cross-β-structures have been found in amyloid fibrils from various amyloidogenic proteins, suggesting that in-register parallel β-sheet formation is a universal feature of amyloid fibril structures.³ Recent studies have shed light on the mechanism by which amyloid fibril surfaces can catalyze secondary nucleation.⁷ Self-catalyzed secondary nucleation is a monomer-dependent process, whereby transient monomer binding to a fibril lateral surface accelerates aggregation, giving rise to cytotoxic oligomers and short fibrils. In some instances, lipid membranes can serve as platforms for the conformational transition and subsequent fibril formation. Indeed, membranes can accelerate the rate of primary nucleation⁸ as well as elongation of amyloid fibrils, leading to bilayer disruption by increased fibril load.

AD is associated with progressive accumulation of cross-β fibrils of amyloid-β (Aβ), mainly consisting of 40 or 42 amino acids, in the brain resulting in formation of extracellular senile plaques.^9,10 It has been reported that binding of Aβ fibrils to cell membranes inhibits long-term potentiation in mouse hippocampal brain slices.¹¹ Yanagisawa et al. identified a unique Aβ species tightly associated with GM1 ganglioside, a glycosphingolipid abundant in neuronal membranes, in the cerebral cortices of human brains exhibiting early pathological AD changes.¹² A monoclonal antibody, 4396C, that specifically binds to GM1-bound Aβ species stained neurons in the cerebral cortices of AD brains.¹³ A recent mass spectrometry imaging study revealed that GM1 was present in the core region of amyloid plaques and that the amount of deposited Aβ correlated with that of GM1 in AD mouse models.¹⁴

Various in vivo and in vitro studies have demonstrated that GM1-bound Aβ species act as endogenous seeds for cerebral Aβ fibril formation, accelerating Aβ assembly.^13,15−18 Furthermore, amyloid fibrils on GM1-containing liposomes were reported to be more toxic than those formed in aqueous solution.^19,20 Interactions between Aβ and the GM1 cluster have thus far been characterized by nuclear magnetic resonance (NMR) spectroscopy and molecular dynamics (MD) simulation using model systems such as GM1-containing micelles, highlighting α-helix formation at the hydrophobic/hydrophilic interface.²¹⁻²⁸ However, the active conformational state of Aβ in GM1-containing lipid membranes remains to be elucidated, as it has only been characterized as a conformational epitope recognized by the 4396C antibody.¹³ Intriguingly, this antibody has been shown to be cross-reactive with Aβ at the ends of growing fibrils;¹³ administration of its Fab fragments significantly reduced plaque formation in AD model mice,²⁹ suggesting a common conformational epitope between Aβ molecules in GM1 clusters and those on the fibril ends.

Here, we report a structural characterization of the GM1-bound states of Aβ molecules. We established a protocol for stabilizing active Aβ catalytic species reactive with the 4396C antibody using a GM1-rich phospholipid membrane system. On this basis, we successfully unveiled the three-dimensional structures of those Aβ species by solid-state NMR spectroscopy.

Results and Discussion

Stabilization of Active Catalytic Aβ Species

To characterize the structure of the GM1-bound Aβ molecules, we first optimized the protocol to prepare solid-state NMR samples. An Aβ solution and GM1/1,2-dimyristoyl-sn-glycero-3-phosphocholine (DMPC) vesicle suspension were mixed on ice and immediately subjected to ultracentrifugation at 4 °C to prevent amyloid fibril formation, thereby collecting only the vesicle-bound fraction as precipitate, which was immediately lyophilized (Figure 1A). To confirm the absence of amyloid fibrils in the collected fraction, part of the lyophilizate of the Aβ/GM1/DMPC fraction was rehydrated and subjected to a thioflavin T (ThT) assay. The result indicated that rehydrated lyophilizates of the Aβ/GM1/DMPC fraction exhibited no fluorescence enhancement in comparison with GM1/DMPC alone even after prolonged incubation at 37 °C (Figure 1B). A free Aβ solution exhibited a gradual increase in ThT fluorescence intensity during incubation at 37 °C after a lag time of about 24 h. The conversion of monomeric Aβ to fibrils was markedly accelerated in the presence of the rehydrated lyophilizate of the Aβ/GM1/DMPC fraction but not without GM1 (Figure 1C). These data indicated that the GM1-membrane-bound Aβ itself was not transformed into ThT-reactive fibrils but instead catalytically promoted amyloid fibrillization of monomeric Aβ. This observation is consistent with the theory that Aβ species tightly associated with GM1 acts as an endogenous seed for amyloid fibril formation in the AD brain.^12,13 The conformational state of the GM1-bound Aβ species was probed with a series of monoclonal antibodies. The rehydrated lyophilizate of the Aβ/GM1/DMPC fraction reacted with 6E10 and 4G8, known to recognize the N-terminal segment (Asp1-Lys16) and the mid part (Leu17-Val24) of Aβ, respectively, but not with AB27, a C-terminal specific antibody (Figure 1D). These data indicate that the N-terminal and middle segments of GM1-bound Aβ are exposed to the solvent, while its C-terminal part is not accessible to the antibody. Interestingly, 4396C reacted specifically with the rehydrated lyophilizate of the Aβ/GM1/DMPC fraction but not with monomeric Aβ or that of Aβ/DMPC fraction (Figure 1D), indicating that the GM1-bound Aβ species exhibits a unique conformational epitope as suggested by previous reports.^12,13 We therefore proceeded to perform NMR analysis of this GM1-bound Aβ-structure.

Figure 1.

Preparation and characterization of active catalytic Aβ species. (A) Scheme of sample preparation of Aβ_1–40 bound to the GM1/DMPC vesicle for solid-state NMR. (B) ThT fluorescence assay of the rehydrated lyophilizate of Aβ_1–40/GM1/DMPC fraction (red), GM1/DMPC suspension (black), Aβ_1–40/DMPC fraction (blue), and DMPC suspension (gray). (C) ThT fluorescence assay of Aβ_1–40 fibrilization in the absence (black) and presence of the 1% (v/v) and 2% (v/v) rehydrated lyophilizate of the Aβ_1–40/GM1/DMPC fraction (orange and red, respectively) or in the presence of the 1% (v/v) and 2% (v/v) rehydrated lyophilizate of the Aβ_1–40/DMPC fraction (blue and green, respectively). (D) Dot blot assay of Aβ_1–40 with monoclonal anti-Aβ antibodies. Monomeric Aβ_1–40, the rehydrated lyophilizates of Aβ_1–40/GM1/DMPC and Aβ_1–40/DMPC fractions, and preprepared Aβ_1–40 fibril were blotted.

Antiparallel β-Structure of Aβ Assemblage on GM1/DMPC Vesicles

We characterized the conformation of Aβ bound to GM1/DMPC vesicles based on solid-state NMR data. Using a uniformly ¹³C- and ¹⁵N-labeled Aβ, sequential signal assignments were achieved by the ¹³C dipolar-assisted rotational resonance/RF-assisted diffusion (DARR/RAD) experiment^30,31 in conjunction with ¹³C-detected NCO and NCA heteronuclear correlation experiments using ¹³C–¹⁵N double cross-polarization (DCP)³² (Figures 2A and S1A–C). To avoid ambiguities in spectral assignments arising from peak overlapping, Aβ analogs ¹³C- and ¹⁵N-labeled specifically at Val36 or Val40 were also used for spectral measurements (Figure S1D). The results indicate that the observed peaks originated from Val12–Val40 of Aβ bound to GM1/DMPC vesicles (Table S1). No peak multiplicity was observed, indicating the conformational equivalence of this segment. Torsion angle prediction based on Aβ backbone chemical shifts revealed that the segments Leu17–Ala21 and Lys28–Val39 form two discontinuous β strands, termed β1 and β2, respectively, upon binding to GM1/DMPC vesicles (Figures 2B and S2, Table S2). On the other hand, the segment Asp1–Glu11 was not observed, indicating structural disorder and/or high mobility of the N-terminal segment. In general, the β strands of Aβ peptides are stabilized through both intra- and intermolecular hydrogen-bonding interactions. To observe molecular contacts among Aβ peptides in the GM1/DMPC-induced structure, long distance information was acquired by DARR/RAD experiments at various mixing times up to 400 ms (Figure S3A). Intermolecular and intramolecular cross peaks were distinguished based on spectral comparison with uniformly ¹³C- and ¹⁵N-labeled Aβ diluted with unlabeled Aβ (molar ratio of [¹³C, ¹⁵N]Aβ:unlabeled Aβ, 7:3) (Figure S3B,C), thereby identifying 21 intra- and 20 intermolecular cross peaks (Tables S3 and S4). These were used as distance constraints for conformational analyses of Aβ molecules bound to GM1/DMPC vesicles. In terms of intermolecular proximity in the DARR/RAD correlation spectra, β1−β1 or β2−β2 interactions were suggested, but we could not determine whether these interactions are arranged in parallel or antiparallel orientation.

Figure 2.

Solid-state NMR characterization of Aβ bound to GM1/DMPC vesicles. (A) Aliphatic region of ¹³C–¹³C correlation MAS spectra of [¹³C, ¹⁵N]Aβ_1–40 bound to GM1/DMPC vesicles acquired by DARR/RAD with a mixing time of 10 ms. (B) Probability of secondary structures (strand, coil, and helix) estimated by TALOS+ analysis according to ¹³C and ¹⁵N chemical shifts of Aβ_1–40 bound to GM1/DMPC vesicle. The primary structure of Aβ_1–40 is presented at the top with red arrows indicating β-strand regions.

To determine the intermolecular arrangements of β-strands, we prepared a paramagnetic Aβ probe by attaching (1-oxyl-2,2,5,5-tetramethyl-3-pyrroline-3-methyl)methanethiosulfonate (MTSL) to an isotopically unlabeled Aβ molecule via a cysteine residue as its C-terminal extension. We mixed this spin-labeled Aβ (designated as Aβ-Cys-MTSL) with a 10-fold molar excess of isotope-labeled (but not spin-labeled) Aβ and added the mixture to a DMPC/GM1 suspension to prepare a GM1-bound Aβ assemblage for the DARR/RAD experiment. This enabled us to selectively observe the intermolecular paramagnetic relaxation enhancement (PRE). Our results showed that peaks originating from Ser26, Asn27, and Lys28 of [¹³C, ¹⁵N]Aβ exhibited significant line broadening due to PRE, indicating their spatial proximity with the C-terminal paramagnetic probe in neighboring molecules (Figure S4). In contrast, little or no PRE was observed for the C-terminal amino acid residues (Leu34, Met35, Val36, Val39, and Val40) of [¹³C, ¹⁵N]Aβ. These data rule out a parallel arrangement of β2 strands and instead indicate an antiparallel orientation.

As noted above, the Val12–Val40 segment exhibits conformational uniformity among Aβ molecules bound to GM1/DMPC vesicles. Based on the experimentally obtained constraints, we attempted to build a three-dimensional (3D) model of the GM1-bound Aβ assemblage. However, we could not obtain an antiparallel dimer configuration that satisfied all of the intermolecular atomic distance constraints obtained from DARR/RAD correlation spectra. Hence, we classified these constraints into two interaction modes (Figure S5A,B). More than two Aβ molecules are required to satisfy these two interaction modes. We used eight structurally equivalent Aβ molecules, which are aligned with antiparallel β1−β1 and β2−β2 interactions to make the possible terminal effect negligible (Figure S5C).

Starting from this arrangement, we performed structural modeling by an MD calculation with the constraints of dihedral angles and intra- and intermolecular distances estimated based on solid-state NMR data. We calculated the probability of Aβ forming β-structures during the MD simulation, showing that the interaction mode reproduced the β-structure probability calculated from solid-state NMR data using TALOS+ (Figure S5D). The most characteristic feature of the obtained 3D structural model of the Aβ octamer is alternative formation of β1−β1 and β2−β2 hydrogen-bonding interactions, giving rise to two layers of discontinuous antiparallel β-structures (designated as β1- and β2-layers, respectively) (Figure 3A). The final β1 and β2 strands were composed of Lys16-Ala20 and Ile31-Val36, respectively. In the β1-layer, intermolecular hydrogen bonds were identified between the NH of Val18 and the CO of Val18 and the CO of Lys16 and the NH of Phe20 (Figure 3B). In the β2-layer, intermolecular hydrogen bonds were found between the NH of Gly33 and the CO of Gly33 and between the CO of Ile31 and the NH of Met35 (Figure 3C). We suppose that the octamer is just a section out of the GM1-bound assemblage. This is supported by the absence of peak multiplicity, indicating the conformational equivalence of Val12-Val40 in GM1/DMPC-bound Aβ and the minimal influence of the terminal molecules on the spectrum.

Figure 3.

Antiparallel β-structure of Aβ assemblage on GM1/DMPC vesicles. (A) 3D structure (upper) and diagrammatic representation (lower) of Aβ_12–40 segments in the assemblage. Close view of intermolecular β1−β1 (B) and β2−β2 (C) interactions.

In contrast to the cross-β-structure composed of in-register parallel β-sheets commonly shared among Aβ fibrils, the assemblage visualized in this study is characterized by a double-layered antiparallel β-structure formed through alternative β1−β1 and β2−β2 hydrogen-bonding interactions along the long axis (Figure 3). It has been reported that a D23N-Aβ_1–40 mutant can form fibrils characterized by a double-layered in-register highly stacked antiparallel structure,¹⁸ structurally distinctive from the loosely packed GM1-bound Aβ assemblage identified in this study. Aβ peptides corresponding to the β1 strand, such as Aβ_16–22 and Aβ_11–25, have been reported to form single-layered in-register antiparallel β-sheets.³³⁻³⁵ To determine the relative orientation of β1- and β2-layers with respect to the membrane surface, we used TEMPO (2,2,6,6-tetramethylpiperidine 1-oxyl), which is embedded in the hydrophobic interior of the lipid bilayer and thereby acts as a source of distance information. The attenuation of the peaks originating from the β2-layer and C-terminal segment was observed in the Aβ/GM1/DMPC sample with TEMPO, indicating that the β2-layer is located closer to the vesicle interface than the β1-layer (Figures 4 and S6). This is consistent with the antibody-binding data indicating solvent exposure of the N-terminal and middle segments of GM1-bound Aβ and the inaccessibility of its C-terminal part (Figure 1D). The N-terminal segment (Asp1-Glu11) was not included in the structural model because it is disordered and flexible. We thus examined the possible effect of the deletion of N-terminal segments on antibody binding (Figure S7). We confirmed that the truncated mutant, Aβ_10–40, lacked reactivity with 6E10, but still retained binding capacity to 4396C when bound to GM1/DMPC vesicles. This indicates that the disordered N-terminal segments are not integral parts of the epitope recognized by 4396C, suggesting that the β1-layer formed on GM1-containing membranes is solvent exposed, constituting the conformational epitope to this antibody. Our NMR and antibody-binding data showed that the β1-layer is antibody accessible, reaching out from the glycan cluster, whereas the β2-layer is inwardly set on the membrane (Figure 5). The previously reported antiparallel fibrils, including the D23N-Aβ_1–40 fibril, are considered transient intermediate species which eventually evolved into parallel fibrils.³⁶⁻³⁸ In contrast, the nonfibrillar assemblage on the GM1 membrane appeared stable: our experimental data show that the antiparallel β-assemblage itself was not transformed into ThT-reactive fibrils on GM1 membranes but could promote conversion of monomeric Aβ into fibrils (Figure 1B,C). The antiparallel β-sheet of Aβ_16–22 has been reported to catalyze Aβ_1–40 aggregation through a surface-catalyzed secondary nucleation mechanism.³⁵ In mature Aβ fibrils, the hydrophobic surface of cross-β-structure appears to be pivotal for secondary nucleation.³⁹⁻⁴¹ In the double-layered Aβ assemblage identified in the present study, the β1-layer provides a solvent-exposed hydrophobic surface on the GM1–glycan cluster, thereby providing a catalytic surface for Aβ fibril formation.

Figure 4.

Relative orientation of β1- and β2-layers of Aβ bound to GM1/DMPC vesicles. Plots of the peak intensity ratios between two ¹³C–¹³C correlation MAS spectra of Aβ_1–40 bound to the vesicles acquired in the presence and absence of TEMPO. The relative intensity ratio was calculated by normalizing the intensity ratio of the Phe20β-γ cross peak as 1.

Figure 5.

Schematic drawing of Aβ assemblage on GM1-containing membrane which catalytically promotes amyloid fibrillization. The distance between β1- and β2-layers of the assemblage is almost the same as the GM1 glycan dimension. The β1-layer provides a catalytic hydrophobic surface evoking fibril formation in GM1-sugar clusters.

Conclusion

The present solid-state NMR study identified a unique antiparallel β-structural assemblage of Aβ specifically formed on GM1 ganglioside clusters. Despite previous structural analyses of amyloid fibrils formed in the presence of lipid membranes,^42,43 this study is the first to uncover the structure of nonfibrillar assemblage trapped on GM1-containing membranes.

The present data indicate that the hydrophobic surface of the β1-layer is exposed and provides a catalytic surface for Aβ fibril formation. The β1-layer is also accessible for 4396C binding, suggesting that the conformational epitope recognized by 4396C is located in the antiparallel β1 strands. Noteworthy, aside from GM1-bound-Aβ, 4396C binds an Aβ fibril specifically at its end, thereby preventing fibril elongation in vitro.¹³ This indicates that the Aβ molecule at the fibril end is conformationally distinct from the remaining fibril parts but shares structural similarities with GM1-bound Aβ species in terms of the β1 conformational state. Recently, various therapeutic antibodies have been developed for specifically targeting oligomeric or fibrillar Aβ aggregates.⁴⁴ These antibodies are supposed to multivalently bind the N-terminal segments of the assembled Aβ molecules. Hence, the 4396C epitope, i.e., the antiparallel β1-structure found in the GM1-bound Aβ assemblage, can be an alternative therapeutic target for suppressing fibril formation in the membrane environment and at the fibril end. Our findings offer structural clues for designing drugs targeting catalytically active conformational species of Aβ for the development of anti-AD therapeutics.

Methods

Preparation of the Aβ Solution

Expression and purification of uniformly ¹³C- and ¹⁵N-labeled Aβ_1–40 was performed as previously described.²² The unlabeled Aβ_1–40 peptide with an extra cysteine residue at its C terminus (Aβ-Cys) was expressed and purified according to the protocol used for wild-type Aβ_1–40 with slight modifications. The reaction of Aβ-Cys with the nitroxide spin label MTSL (Toronto Research Chemicals) was carried out as previously described.²³ Synthetic Aβ_1–40 labeled with ¹³C/¹⁵N selectively at Val36 or Val40 was purchased from AnyGen Co. N-Terminally truncated Aβ mutant, Aβ_10–40, was purchased from ABclonal Inc. Both recombinant and synthetic Aβ proteins were dissolved at an approximate concentration of 5 mM in a 0.1% (v/v) ammonia solution. This Aβ solution was used for preparing samples in subsequent experiments.

Preparation of Small Unilamellar Vesicle Suspension

Powdered DMPC and ganglioside GM1 were purchased from Avanti Polar Lipids Inc. and Carbosynth Ltd., respectively. Small unilamellar vesicles (SUVs) composed of GM1 and DMPC were prepared by dissolving 2.8 mg of DMPC and 26.0 mg of GM1 in 1 mL of a methanol/chloroform (1:1) solution (molar ratio of DMPC:GM1, 2:8). The solvents were removed from the DMPC/GM1 suspension with nitrogen gas, followed by complete vacuum drying. The dried DMPC/GM1 sample was resuspended in a total of 2 mL of 5 mM potassium phosphate buffer (pH 7.4) and homogenized by six cycles of successive freezing in liquid nitrogen, thawing at 50 °C, vortexing at room temperature, and subsequently sonicated for 10 min (2 min, 5 times) using a probe-type sonicator. Metal debris from the titanium tip of the probe was removed by centrifugation. DMPC SUV was prepared in the same manner. SUV composed of GM1 and DMPC with TEMPO (2,2,6,6-tetramethylpiperidine 1-oxyl) was prepared in the same way, except that 6.25 mg of TEMPO dissolved in chloroform was added. The supernatants, i.e., GM1/DMPC suspension (containing 2 mM DMPC and 8 mM GM1), DMPC suspension (containing 10 mM DMPC), and GM1/DMPC/TEMPO suspension (containing 2 mM DMPC, 8 mM GM1, and 20 mM TEMPO), were used for subsequent experiments.

Preparation of Membrane-Bound Aβ Samples

An aliquot (0.1 mL) of the Aβ/ammonium solution was mixed with 1 mL of the DMPC suspension, the DMPC/GM1 suspension, or the DMPC/GM1/TEMPO suspension on ice. After adjusting the pH to 7.4, each mixture was immediately subjected to ultracentrifugation at 604 000g for 6 h at 4 °C. The supernatant was removed, and the precipitate was immediately lyophilized. A lyophilizate of the Aβ_1–40/DMPC/GM1 fraction was subjected to solid-state NMR spectroscopy, ThT assay, and dot blot assay. The lyophilizate of the Aβ_1–40/DMPC fraction was subjected to the ThT assay and dot blot assay; that of Aβ_10–40/DMPC was subjected to a dot blot assay only.

Solid-State NMR Experiments

All solid-state NMR measurements of the lyophilizate of the Aβ_1–40/DMPC/GM1 fraction were carried out under a 14.1-T magnetic field (¹H resonant frequency of 600 MHz) using a Bruker Avance-600 spectrometer equipped with a ¹H–¹³C-¹⁵N triple resonance 2.5 mm magic angle spinning (MAS) probe. The MAS rate was actively controlled at 13.5 kHz ± 5 Hz using a Bruker MAS II automatic controller. Net sample temperature was controlled at 15 °C using a variable temperature controller by considering sample heating due to MAS. Representative radio frequency (RF) fields for ¹H and ¹³C were 100 and 88 kHz, respectively. Samples were packed into a 4 mm space at the sample tube center using an original Diflon spacer of 1 mm thickness to maintain RF homogeneity. ¹³C chemical shifts were referenced externally to the ¹³C CH signal of adamantane at 29.5 ppm on the TMS scale, and ¹⁵N chemical shifts were referenced to the ¹⁵N signal of NH₄Cl at 39.25 ppm on the liquid ammonia scale. Recycle delay was 2 s. ¹H heteronuclear decoupling during the detection period was achieved using small phase incremental alteration-64 (SPINAL-64)⁴⁵ using a ¹H RF field of 100 kHz. Initial magnetizations of rare nuclei were enhanced by using ¹H-X cross-polarization (CP) with 20% amplitude sweep of the ¹H spin locking field⁴⁶ (X = ¹³C and ¹⁵N). For ¹H–¹³C CP, an average ¹H spin locking field of 86.5 kHz and a ¹³C spin locking field of 73 kHz were used, while for ¹H–¹⁵N CP, an average ¹H spin locking field of 66.5 kHz and a ¹⁵N spin locking field of 53 kHz were used. Contact times (CTs) were 1200 and 700 μs for ¹³C and ¹⁵N, respectively.

The two-dimensional (2D) ¹³C–¹³C correlation was achieved by DARR³⁰/RAD³¹ experiments. A constant ¹H RF field of 13.5 kHz corresponding to the spinning rate was applied during the mixing period. A total of 300, 250, 185, and 185 complex t₁ points were acquired with an increment of 20 μs for mixing times of 10, 100, 200, and 400 ms. For each t₁ point, 1200–2480 scans were accumulated. ¹H decoupling during the t₁ period was achieved by SPINAL64. In order to observe paramagnetic effects, the samples containing paramagnetic probes were also subjected to the measurement with a mixing time of 10 ms.

2D NCO and NCA ¹³C–¹⁵N correlation experiments were carried out by DCP³² with 78 complex points for t₁ time domain with an increment of 148 μs. For each t₁ point, 2400 scans were accumulated in both experiments. CTs of 700 and 1900 μs were used for ¹H–¹⁵N and ¹⁵N–¹³C CPs, respectively. ¹⁵N–¹³C CP was achieved by a ¹³C spin locking field of 20 kHz and an average ¹⁵N spin locking field of 33.5 kHz with a 5% amplitude sweep. CW decoupling at a ¹H RF field of 100 kHz was applied during the ¹⁵N–¹³C CP. ¹³C π pulse was applied at the middle of the t₁ period to achieve ¹³C–¹⁵N heteronuclear J-decoupling. During the t₁ period, ¹H decoupling was achieved by two pulse phase modulation (TPPM)⁴⁷ with a ¹H RF field of 100 kHz.

NMR data were processed by TOPSPIN2.1 (Bruker Biospin, Japan). FIDs for 2D ¹³C–¹³C DARR correlation experiments were zero-filled to 4 K points both for t₁ and t₂ time domains, respectively, and apodized with a trapezoid window function for both t₁ and t₂ time domains, prior to Fourier transformation (FT). FIDs for 2D ¹³C–¹⁵N correlation experiments were zero-filled to 4 and 8 K points for t₁ and t₂ time domains, respectively. FIDs were apodized with trapezoid window functions and Gaussian broadenings for the t₁ and t₂ time domains, respectively, prior to FT.

The cross-peaks first observed in DARR spectra at a mixing time of 100 ms were assigned to a short-range distance of <5 Å, while those first observed in the DARR spectra at mixing times of 200 and 400 ms were assigned to a long-range distance of 5.0 ± 2.5 Å.

Structure Determination

To determine the 3D structure of the Aβ assemblage, we performed structural modeling by a molecular dynamics (MD) calculation of the Val12–Val40 segments of Aβ_1–40, i.e., Aβ_12–40, with the constraints obtained from solid-state NMR experiments. Regarding the β2 strands, the intermolecular distance constraints (listed in Table S4) could be classified into two sets: In one set (termed set A), the average of the two residue numbers (residue numbers of Atom 1 and Atom 2 in Table S4) is about 33 (as shown in Figure S5A), while in the other set (set B), it is about 36 (as shown in Figure S5B). There is an additional intermolecular distance constraint between Phe20 and Ala21 in the β1 strand. Phe20 is also close to Leu34 and Val36 judging from the intramolecular constraints between β1 and β2 (Table S3), suggesting that the region around Gly25–Gly29 forms a loop so that Phe20 and Ala21 are close to the Leu34–Val36 segment in one Aβ molecule. Based on the above considerations, an initial structure of the Aβ assemblage had eight Aβ_12–40 peptides arranged in antiparallel fashion, assuming that Aβ molecular pairs forming intermolecular hydrogen bonds between odd-numbered residues and those between even-numbered residues satisfy sets A and B, respectively (Figure S5C).

One Aβ assemblage, consisting of eight Aβ_12–40 peptides, was placed with eight sodium ions as counterions and 52 537 water molecules in a cubic simulation box with a side length of 118.075 Å. The total number of atoms was 161 123. The N-terminus of each Aβ_12–40 peptide was capped with an acetyl group to avoid electrostatic repulsion between the N-termini. The AMBER parm14SB force field⁴⁸ was used for peptides and counterions. The TIP3P rigid body model⁴⁹ was used for water molecules adopting the symplectic⁵⁰ quaternion scheme for rigid body molecules.^51,52 In addition, to include the constraints from the solid-state NMR experiment, the DIANA force field⁵³ was applied: the total potential energy E is given by

where E_AMBER is the AMBER potential energy and T_DIANA is the DIANA target function. The weight factor w for the target function was set at 10 kcal/mol. The electrostatic potential was calculated using the particle mesh Ewald method. The cutoff distance was set as 12 Å for the Lennard–Jones potential.

We used the generalized-assembled Molecular Biophysics program to perform the MD calculation. This program was developed by one of the authors (H.O.) and has been used for several protein systems.⁵⁴⁻⁵⁸ Reversible multiple time-step MD techniques were also applied.⁵⁹ The time step was Δt = 0.5 fs for bonding interactions of peptide atoms, Δt = 2.0 fs for nonbonding interactions of peptide atoms and those between peptide atoms and solvent molecules, and Δt = 4.0 fs for the interaction between solvent molecules. Since the symplectic rigid body algorithm was used for water molecules, Δt was as long as 4.0 fs.⁵² After energy minimization, an MD calculation was performed for 100 ps at 300 K. The temperature was controlled using the Nosé–Hoover thermostat.⁶⁰⁻⁶² The molecular graphics were prepared using PyMOL (Schrödinger, New York, USA).

ThT Assay

Each lyophilizate of the Aβ_1–40/GM1/DMPC and Aβ_1–40/DMPC fractions was rehydrated with 1 mL of ultrapure water. The Aβ_1–40 solution was diluted to 20 μM in 5 mM potassium phosphate buffer (pH 7.4) as a monomeric Aβ solution. The rehydrated lyophilizates of the Aβ_1–40/GM1/DMPC and Aβ_1–40/DMPC fractions, the GM1/DMPC suspension, the DMPC suspension, the monomeric Aβ solution, and the monomeric Aβ solution with 1–2% (v/v) rehydrated lyophilizate of the Aβ_1–40/GM1/DMPC or Aβ_1–40/DMPC fraction were supplemented with 40 μM ThT from a 2 mM stock solution. These samples were then pipetted into multiple wells (80 μL per well) of a 96-well half area, low-binding polyethylene glycol coating plate (Corning 3881) with a clear bottom, followed by incubation at 37 °C under quiescent conditions in a plate reader (Infinite 200Pro; TECAN). ThT fluorescence was measured through the bottom of the plate with a 430 nm excitation filter and a 485 nm emission filter.

Dot Blot Assay

Along with mouse monoclonal antibody 4396C (Immuno-Biological Laboratories Co., Ltd.), mouse monoclonal antibodies 6E10 (COVANCE), 4G8 (COVANCE), and BA27 (FUJIFILM Wako Pure Chemical Corporation), directed against amino acid residues 1–16, 17–24, and the C-terminal region of human Aβ_1–40, respectively, were used for the dot blot assay.

Aβ_1–40 and Aβ_10–40 solutions were diluted to 0.1 mM in 5 mM potassium phosphate buffer (pH 7.4). Aβ fibrils were prepared using 0.1 mM Aβ_1–40 or Aβ_10–40 by incubation at 37 °C under quiescent conditions. Each lyophilizate of the Aβ_1–40/GM1/DMPC, Aβ_1–40/DMPC, and Aβ_10–40/GM1/DMPC fractions was rehydrated with 1 mL of ultrapure water and then diluted at 0.1 mM in 5 mM potassium phosphate buffer (pH 7.4).

Aβ solutions, Aβ fibrils, and the rehydrated lyophilizates were blotted onto nitrocellulose membranes (BioRad) as described previously.²¹ The blots of these Aβ samples reacted with 4396C (1:2500), 6E10 (1:10 000), 4G8 (1:10 000), or BA27 (1:10 000) and subsequently with horseradish peroxidase-conjugated antimouse IgG (Cell Signaling Tec). Bound-enzyme activities were visualized with an enhanced chemiluminescence system (GE Healthcare).

Data Availability Statement

All data needed to evaluate the conclusions in the paper are present in the paper and/or Supporting Information. Assigned chemical shift data for Aβ_1–40 bound to the GM1/DMPC vesicle were deposited in the BMRB under the accession number 36495. The atomic coordinates and structure factors are deposited in the Protein Data Bank with accession code 7Y8Q. Additional data related to this paper may be requested from the authors.

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acschemneuro.3c00192.

• NMR spectra, dot blot assay, summary of NMR chemical shifts and intra- and intermolecular restraints used for structure calculations, and backbone torsion angles (PDF)

Author Contributions

M.Y.-U., K.Y., K.K., and K.N. designed the research. M.Y.-U. and K.N. prepared the samples. M.Y.-U. and K.N. performed NMR experiments and data analysis. M.Y.-U. performed ThT and dot blot analyses. S.G.I. and H.O. performed computational analyses. M.Y.-U., H.O., K.K., and K.N. contributed to manuscript writing. All authors contributed to discussions throughout the research and participated in editing or commenting on the manuscript.

This work was supported in part by JSPS KAKENHI (Grant Numbers JP19K07041 to M.Y.-U., JP21K06040 to S.G.I., JP21K06118 to H.O., and JP16K05858 and JP19K05552 to K.N.), by JST PRESTO (Grant Number JPMJPR22AC to M.Y.-U.), by Grant-in-Aid for Research in Nagoya City University (Grant Numbers 2212008 and 2222004 to M.Y.-U.), and by Joint Research of the Exploratory Research Center on Life and Living Systems (ExCELLS program Nos. 22EXC338 and 23EXC305 to K.Y.). The computation was performed using Research Center for Computational Science, Okazaki Research Facilities, Japan (Projects: 20-IMS-C155, 21-IMS-C172, and 22-IMS-C186). Solid-state NMR measurements were conducted at the Institute for Molecular Science, supported by Nanotechnology Platform Program “Molecule and Material Synthesis” (JPMXP09S17MS1095, JPMXP09S18MS1055, JPMXP09S18MS1087, and JPMXP09S19MS1049) of the Ministry of Education, Culture, Sports, Science and Technology (MEXT), Japan.

The authors declare no competing financial interest.