Structures of brain-derived 42-residue amyloid-β fibril polymorphs with unusual molecular conformations and intermolecular interactions

Significance

Thread-like assemblies called amyloid fibrils, formed by the 42-residue amyloid-β peptide (Aβ42), are a main component of plaques in the brain tissue of Alzheimer's disease (AD) patients. Fibrils with a variety of molecular structures have been identified in previous studies, all sharing a common S-shaped configuration for individual Aβ42 molecules. Here we report two structures that are qualitatively different in several ways, including molecular configurations that resemble Greek letters ν and υ instead of the letter S. The structures were derived from brain tissue by seeded growth, using fibrils from the tissue as seeds. This work broadens our understanding of the range of possible Aβ42 fibril structures.

Abstract

Fibrils formed by the 42-residue amyloid-β peptide (Aβ42), a main component of amyloid deposits in Alzheimer's disease (AD), are known to be polymorphic, i.e., to contain multiple possible molecular structures. Previous studies of Aβ42 fibrils, including fibrils prepared entirely in vitro or extracted from brain tissue and using solid-state NMR (ssNMR) or cryogenic electron microscopy (cryo-EM) methods, have found polymorphs with differences in amino acid sidechain orientations, lengths of structurally ordered segments, and contacts between cross-β subunit pairs within a single filament. Despite these differences, Aβ42 molecules adopt a common S-shaped conformation in all previously described high-resolution Aβ42 fibril structures. Here we report two cryo-EM-based structures of Aβ42 fibrils that are qualitatively different, in samples derived from AD brain tissue by seeded growth. In type A fibrils, residues 12 to 42 adopt a ν-shaped conformation, with both intra-subunit and intersubunit hydrophobic contacts to form a compact core. In type B fibrils, residues 2 to 42 adopt an υ-shaped conformation, with only intersubunit contacts and internal pores. Type A and type B fibrils have opposite helical handedness. Cryo-EM density maps and molecular dynamics simulations indicate intersubunit K16-A42 salt bridges in type B fibrils and partially occupied K28-A42 salt bridges in type A fibrils. The coexistence of two predominant polymorphs, with differences in N-terminal dynamics, is supported by ssNMR data, as is faithful propagation of structures from first-generation to second-generation brain-seeded Aβ42 fibril samples. These results demonstrate that Aβ42 fibrils can exhibit a greater range of structural variations than seen in previous studies.

Self-propagating, molecular-level polymorphism is a salient property of amyloid fibrils formed by many peptides and proteins, meaning that the molecular structure within an amyloid fibril is not determined uniquely by the amino acid sequence. This is particularly true of fibrils formed by amyloid-β (Aβ) peptides, as first revealed by experiments on fibrils grown in vitro (1–4). Polymorphism of 40- and 42-residue Aβ (Aβ40 and Aβ42) fibrils has been proposed to be a factor that may contribute to variations in clinical and pathological features of Alzheimer's disease (AD) or in the degree of neurodegeneration that is associated with amyloid deposition in brain tissue (5–11). Interest in both the biophysical aspects of amyloid formation and the biomedical implications of amyloid polymorphism motivate recent and ongoing studies of Aβ fibril structures.

Early in vitro studies of Aβ42 fibrils by Lührs et al. (12), based primarily on hydrogen exchange and mutagenesis experiments, led to a structural model in which Aβ42 molecules adopt a U-shaped strand-loop-strand conformation, with residues 17 to 25 and 31 to 41 forming the β-strands in two layers of in-register parallel β-sheets. This Aβ42 fibril model was similar in many respects to an early model for Aβ40 fibrils grown in vitro, developed by Petkova et al. (13) from solid-state NMR (ssNMR) data. Subsequently, Xiao et al. (14) used an extensive set of ssNMR measurements, combined with molecular dynamics (MD) simulations, to develop a structural model in which Aβ42 molecules adopt an S-shaped conformation, with three β-strand-like segments forming three layers of in-register parallel β-sheets. Higher-resolution ssNMR-based structural models for Aβ42 fibrils, also grown in vitro but under different solvent conditions and with different protocols, were then developed by Colvin et al. (15) and Wälti et al. (16). These models include an S-shaped Aβ42 conformation, similar to the conformation identified by Xiao et al., but with differences in the orientation of the certain amino acid sidechains (M35 and V36). The Aβ42 models of Colvin et al. and Wälti et al. also include two cross-β subunits (often called protofilaments), consistent with earlier mass-per-length data for Aβ42 fibrils (17). These models exhibit multiple hydrophobic contacts between the two subunits, with a twofold symmetry axis centered between their L34 sidechains.

With advances in technology for cryogenic electron microscopy (cryo-EM) and software for image analysis (18), it has become possible to develop high-resolution molecular structural models for amyloid fibrils from cryo-EM images. The first cryo-EM-based study of Aβ42 fibrils with sufficient resolution to align the amino acid sequence unambiguously was published by Gremer et al. (19), using fibrils that were grown in vitro at low pH. Their structural model shows an S-shaped conformation for residues 15 to 42 and two cross-β subunits, as in the ssNMR-based models for fibrils grown near neutral pH. Unlike the ssNMR-based models, residues 1 to 8 are structurally ordered in the model of Gremer et al., and the twofold symmetry axis is centered between V39 sidechains.

Subsequently, Yang et al. (20) reported cryo-EM-based structural models for Aβ42 fibrils extracted from human brain tissue. In experiments with tissue from the cortex, hippocampus, and amygdala and from individuals with AD as well as other neurodegenerative conditions, Yang et al. succeeded in characterizing structures of two fibril polymorphs, which they call type I and type II. Both polymorphs contain S-shaped Aβ42 conformations that closely resemble the conformations in the earlier studies, with different orientations for sidechains of H14 and F19 in the two polymorphs. Both polymorphs contain two cross-β subunits, with twofold symmetry axes centered between L34 sidechains in the case of type I fibrils and K28 sidechains in the case of type II fibrils. Both brain-extracted polymorphs characterized by Yang et al. differ from the in vitro polymorphs studied by Xiao et al., Colvin et al., and Wälti et al. in the orientations of multiple sidechains (A30, I31, I32, V39, V40, I41).

These ssNMR-based and cryo-EM-based studies suggest that S-shaped molecular conformations may be a universal feature of Aβ42 fibrils, although other structural details may vary. Here we report cryo-EM studies of two Aβ42 fibril polymorphs that are derived from cortical tissue of an AD patient by seeded fibril growth (5, 21). The resulting structural models show that other molecular conformations, with qualitatively different modes of supramolecular organization, also exist in Aβ42 fibrils. Aβ42 conformations in the type A fibrils described below are approximately “ν-shaped,” with residues 28 to 42 forming a continuous surface for contacts between two subunits; Aβ42 conformations in the type B fibrils are approximately “υ-shaped,” with residues 16 to 42 forming a continuous interface between interlocked subunits. The possible relevance of these polymorphs to AD is discussed below.

Results

Polymorphic Brain-Derived Aβ42 Fibrils.

Second-generation brain-seeded Aβ42 fibrils for cryo-EM measurements were grown by incubating solutions of synthetic or recombinant Aβ42 with fibril seeds (2.3 mole%) under quiescent conditions at 24 °C (SI Appendix, Supporting Methods). Seeds were sonicated aliquots of first-generation Aβ42 fibrils that had been prepared previously from parietal lobe tissue of a patient with a typical long-duration AD [t-AD3p fibrils from Qiang et al. (5)]. Seeds were combined with synthetic or recombinant Aβ42 in a single step. Negative-stain transmission electron microscopy (TEM) and cryo-EM images (Fig. 1 A–D) showed twisting fibrils with a range of crossover distances, including populations of fibrils with crossover distances centered around 47 nm and 74 nm (SI Appendix, Fig. S1A). Fibrils without a clear twist were also observed. Two-dimensional (2D) class average images with high resolution were obtained from the cryo-EM images for a more rapidly twisting polymorph (crossover distance ≈ 50 nm), which we call type A (Fig. 1E), and a more slowly twisting polymorph (crossover distance ≈ 70 nm), which we call type B (Fig. 1F). Additional polymorphs were observed in the 2D class averages, including rod-shaped fibrils without width modulation and additional twisted polymorphs (SI Appendix, Fig. S1B). These additional polymorphs had insufficient resolution, abundance, and/or structural homogeneity for further analysis. Helical reconstruction in RELION 3.1.3 software (18) was used to reconstruct three-dimensional (3D) densities for type A and type B Aβ42 fibrils, with conditions given in Table 1. Both polymorphs were present in images of second-generation fibrils prepared with either synthetic or recombinant Aβ42. Images of fibrils prepared with synthetic Aβ42 were used for the type A density map, while images of fibrils prepared with recombinant Aβ42 were used for the type B density map. Based on the total numbers of particles extracted from the cryo-EM images and used for density reconstructions, type A and type B fibrils represent at least 20 to 30% each of the fibrils in our samples (Table 1).

Fig. 1.

Electron microscopy of brain-derived Aβ42 fibrils. Negative-stain TEM (A and B) and cryo-EM (C and D) images of brain-derived Aβ42 fibrils, prepared by seeded growth using synthetic (A and C) or recombinant Aβ42 (B and D). Red and cyan arrows indicate crossover points of different fibril polymorphs. (Scale bars are 100 nm.) Examples of 2D class average images of more rapidly twisting (type A) and more slowly twisting (type B) fibrils are shown in panels E and F, respectively. The 2D class images show the narrowest (Left) and widest (Right) regions of the fibrils.

Table 1.

Cryo-EM data collection and density map reconstruction parameters

Data collection	Type A	Type B
Microscope	Titan Krios	Titan Krios
Camera	Gatan K2 Summit	Gatan K2 Summit
Voltage	300 kV	300 kV
Magnification	105,000	105,000
Defocus range	−0. 5 µm to −3.0 µm	−0. 5 µm to −2.5 µm
Pixel size	0.86 Å	0.86 Å
Exposure time	1.65 s	1.65 s
Total electron dose	48.69 e−/Å2	50.34 e−/Å2
No. of movie frames	22	22
Helical reconstruction
Box size	500 pixels	500 pixels
Interbox distance	44.65 Å	44.65 Å
No. of micrographs	3,383	5,557
No. of extracted particles	347,821	285,760
No. of particles in 3D reconstruction	68,481	92,019
Symmetry imposed	Quasi 21	Quasi 21
Helical rise/Helical twist	2.46 Å/−179.09°	2.48 Å/179.35°
Map resolution	2.83 Å	2.76 Å
Map sharpening B-factor	−65.20 Å2	−55.12 Å2

Both second-generation fibril samples for cryo-EM were prepared with aliquots from the same first-generation fibril sample. Images of fibrils prepared with recombinant Aβ42 showed fewer nonfibrillar assemblies but were otherwise indistinguishable from images of fibrils prepared with synthetic Aβ42.

Right-Handed, More Rapidly Twisting Aβ42 Fibrils (Type A).

The 3D density map for type A fibrils was refined with quasi-2₁ symmetry, yielding a final density with 2.83 Å resolution (Fig. 2 A and B). Helical rise and twist parameters are 2.46 Å and −179.09°, respectively, corresponding to a crossover distance of 49 nm. Calculations without symmetry or with C2 symmetry resulted in lower final resolution (SI Appendix, Fig. S2 A and C). Moreover, applying C2 symmetry yielded a density map with unphysical connections between neighboring β-strands along the fibril growth direction (SI Appendix, Fig. S2A).

Fig. 2.

Cryo-EM density map and molecular structural model for type A Aβ42 fibrils. (A) Cross-section image of the cryo-EM density map for type A fibrils. (B) Surface rendering of the 3D density, with 2.83 Å resolution. The density has a right-handed twist and quasi-2₁ symmetry, defined by a helical rise of 2.46 Å and a twist of −179.09˚. Densities for the two cross-β subunits are colored blue and orange. (C) Cross-sectional view with three repeats of the molecular structural model for type A fibrils, including residues 12 to 42. (D) Expanded view of intra-subunit contacts among hydrophobic sidechains and salt bridge interactions between sidechains of K16 and E22. (E) Inter-subunit contacts among hydrophobic sidechains and salt bridge interactions between sidechains of K28 and carboxylate groups of A42. Intra-subunit salt bridges between K16 and E22 sidechains and inter-subunit salt bridges between K28 and K42 sidechains are also seen. Sidechain carbon atoms of hydrophobic, polar, positively charged, and negatively charged residues are shown in green, magenta, blue, and red, respectively.

The final density for type A fibrils shows that these fibrils are composed of two cross-β subunits (orange and blue in Fig. 2B) in which Aβ42 molecules adopt conformations that resemble the Greek letter ν. The subunits have the in-register, parallel intermolecular alignment that is common to all mature Aβ fibrils (1, 14–17, 19–24). The molecular structural model in Fig. 2C was created in Coot (25) and refined by simulated annealing in Xplor-NIH (26), resulting in a bundle of structures from independent annealing calculations with a rmsd of 1.2 Å for backbone atom positions and 1.7 Å for all heavy atom positions in residues 12 to 42 (SI Appendix, Fig. S3 A and C). Simulated annealing conditions and structure validation statistics are given in Table 2.

Table 2.

Structural restraints in Xplor-NIH and validation statistics

Xplor-NIH potentials	Scale factor
	Type A	Type B
probDistPot	10.0, 5.0	10.0, 5.0
NCS	100	100
DistSymmpot	100	100
Repel pot	0.001 to 4.0	0.001 to 4.0
TorsionDB	0.001 to 1.0, 0.001 to 10.0	0.001 to 10.0
BOND	default	default
ANGL	0.4 to 1.0	0.4 to 1.0
IMPR	0.4 to 1.0	0.4 to 1.0
Structure validation
MolProbity score	0.94	1.34
Clash score	0.00	0.00
Ramachandran favored (%)	93.10	92.50
Ramachandran outliers (%)	3.45	0.00
Rotamer outliers (%)	0.00	3.12
No. of non-hydrogen atoms	255 per molecule	318 per molecule
No. of residues	31 (residues 12 to 42)	42 (residues 1 to 42)

In contrast to Aβ42 fibril polymorphs previously characterized by cryo-EM (19, 20), type A fibrils have a right-handed twist, indicated both by atomic force microscopy (AFM) (SI Appendix, Fig. S4A) and by the observation that corrugations of backbone density in β-strand segments align with the directions of backbone carbonyl groups when a right-hand twisted density map is used to create the structural model, but not when a left-hand twisted density map is used (SI Appendix, Fig. S4C). Assignment of the type A structure to right-hand twisted fibrils in the AFM images is supported by the observations that the right-hand twisted fibrils show an average crossover distance of approximately 52 nm in the AFM images, as well as minimum and maximum heights of 3.5 nm and 6.5 nm that are close to the cross-sectional dimensions in Fig. 2C.

Residues 12 to 42 form the structurally ordered fibril core of type A fibrils, with residues 1 to 11 remaining disordered (SI Appendix, Fig. S5A). Residues 14 to 19, 27 to 32, 34 to 36, and 39 to 41 form β-strands (underlined segments in Fig. 2C), as defined by their extended backbone conformations and alternating sidechain directions. The ν-shaped Aβ42 conformation within each cross-β subunit is apparently stabilized by hydrophobic contacts between sidechains of L17, F19, and F20 and sidechains of I31, L34, and V36 (Fig. 2D). Stabilizing interactions between subunits include hydrophobic interactions between sidechains of A30 and I32 and sidechains of M35, V40, and I41 (Fig. 2E).

The density map and structural model for type A fibrils suggest the presence of inter-subunit salt bridges between K28 sidechain amino groups and the C-terminal carboxylate groups of A42. Since the distances between amino nitrogens (N_ζ) of K28 and carboxylate oxygens (OT) of A42 in our structural models exceed the expected salt bridge distance of approximately 3.0 Å, perhaps because electrostatic interactions were not included in Xplor-NIH calculations, we performed all-atom MD simulations on a type A fibril segment containing eight Aβ42 molecules in explicit water solvent to see whether K28-A42 salt bridges would form (SI Appendix, Fig. S6A and Movie S1). N_ζ-OT distances fluctuated in the 3 to 12 Å range over the course of a 200-ns simulation, with salt bridge lifetimes approaching 100 ns for some K28-A42 pairs (SI Appendix, Fig. S6 C and D). The transient nature of these salt bridge interactions may be a consequence of the short length of the fibril segment used in these simulations (only four molecules in each β-sheet).

The density map and structural model also suggest the existence of intra-subunit salt bridges between K16 and E22 sidechains, which would have the effect of shielding the sidechains of V18 and A21 and reducing the solvent-exposed surface area of type A fibrils (Fig. 2D). However, stable K16-E22 salt bridges were not observed in the MD simulations. Hydrogen bonds among sidechains in rows of N27 residues were observed (SI Appendix, Fig. S6H), suggesting that “polar zipper” interactions (27) contribute to the stability of the type A fibrils.

Left-Handed, More Slowly Twisting Aβ42 Fibrils (Type B).

The 3D density map for type B fibrils was also refined with quasi-2₁ symmetry, yielding a final density with 2.74 Å resolution. Helical rise and twist parameters are 2.48 Å and 179.35˚, respectively, corresponding to a 69 nm crossover distance. Calculations without symmetry or with C2 symmetry yielded similar resolutions, ranging from 2.79 to 2.93 Å, but resulted in unphysical connections between neighboring β-strands along the fibril growth directions (SI Appendix, Fig. S2 B and D).

The final density in Fig. 3 A and B shows that type B fibrils are composed of two cross-β subunits (pink and cyan in Fig. 3B) in which Aβ42 molecules adopt a conformation that resembles the Greek letter υ. Nearly the entire peptide chain is structurally ordered. Type B fibrils have a left-handed twist, as indicated by AFM (SI Appendix, Fig. S4B) and by the alignment of backbone carbonyl groups in the structural model with corrugations in the density along β-strand segments (SI Appendix, Fig. S4D). In the AFM height images, fibrils with left-handed asymmetry at crossovers exhibit an average crossover distance of approximately 68 nm. Minimum and maximum heights are 5.0 nm and 9.0 nm, respectively, consistent with dimensions in Fig. 3C and hence supporting the assignment of left-handed twist to type B fibrils.

Fig. 3.

Cryo-EM density map and molecular structural model for type B Aβ42 fibrils. (A) Cross-sectional image of the cryo-EM density for type B fibrils. (B) Surface rendering of the 3D density, with 2.77 Å resolution. The density has a left-handed twist and quasi-2₁ symmetry, defined by a helical rise of 2.48 Å and twist of 179.35°. Densities for the two cross-β subunits are colored cyan and pink. (C) Cross-sectional view with three repeats of the molecular structural model for type B fibrils, including residues 1 to 42. (D–F) Expanded views of inter-subunit contacts among hydrophobic sidechains and inter-subunit salt bridge interactions between sidechains of K16 and the carboxylate groups of A42. Sidechain carbon atoms of hydrophobic, polar, positively charged, and negatively charged residues are shown in green, magenta, blue, and red, respectively.

The structural model in Fig. 3C was refined in Xplor-NIH, resulting in a structure bundle for which the backbone and heavy-atom rmsd values are 1.6 Å and 2.2 Å for residues 1 to 42 (SI Appendix, Fig. S3 B and C). The subunits in type B fibrils interact in an interlocked manner, with residues 16 to 42 of one subunit making contacts with the same residues of the other subunit along an extended, rolling surface and with additional contacts between residues 2 to 4 of one subunit and residues 29 to 31 of the other subunit. Residues 2 to 6, 11 to 20, 30 to 36, and 39 to 41 form β-strands. While the β-sheet layers in type A fibrils have both intra- and inter-subunit hydrophobic contacts, only inter-subunit contacts between β-sheet layers occur in type B fibrils. These include interactions involving sidechains of I32, L34, and V36 from both subunits (Fig. 3D), interactions involving sidechains of A2 and F4 from one subunit and I31 from the other (Fig. 3E), and interactions involving sidechains of V18, F20, and V24 from one subunit and V39 and I42 of the other (Fig. 3F).

In addition to these hydrophobic interactions, inter-subunit salt bridges between the sidechain amino group of K16 and the carboxylate group of A42 may contribute to the association between two cross-β subunit (SI Appendix, Fig. S3E). In all-atom MD simulations on a type B fibril segment (SI Appendix, Fig. S6B and Movie S2), distances between N_ζ of K16 and OT of A42 fluctuated over the 200-ns trajectory but with average values in the 3.5 to 5.0 Å range (SI Appendix, Fig. S6 E and F). This result suggests that K16-A42 salt bridges contribute to the stability of the type B fibril structure. Although the structural model suggests that sidechains of D1 and K28 may also engage in inter-subunit salt bridge interactions (Fig. 3E), D1-K28 salt bridges did not develop in the MD simulations. Instead, stable intra-subunit D23-K28 salt bridges were observed (SI Appendix, Fig. S6G), along with transient interactions between R5 sidechains and sidechains of E3 and D7. Rows of N27 sidechains engaged in polar zipper interactions, and S26 sidechains engaged in hydrogen bonds with D23 sidechains, concomitantly with the D23–K28 salt bridges (SI Appendix, Fig. S6 I and J).

The interlocked association of υ-shaped subunits in type B fibrils creates two internal pores that run along the length of the fibril. In MD simulations, these cavities are filled with water. The density map shows high-density features within these pores, near backbone atoms of residues 32 to 34, that may indicate partially ordered water molecules, i.e., positions where water molecules have high occupancy (asterisks in Fig. 3D). In the MD simulations, water molecules were found to reside at these positions with higher-than-average probability (SI Appendix, Fig. S7). We have no experimental evidence regarding the identity of material in the internal pores of type B fibrils. It should be noted that our fibrils were grown in solutions that contained approximately 1.7% v/v dimethyl sulfoxide (SI Appendix, Supporting Methods).

Additional density maxima are apparent outside the type B fibril core, apparently making contact with sidechains of Q15 and L17 (Fig. 3C). These features are weaker than the main density into which we fit our structural model, as can be seen by varying the threshold value used to visualize the density (SI Appendix, Fig. S5B). These features may indicate fractional populations of alternative conformations for the N-terminal segment in which residues 2 to 4 make intra-subunit contacts to residues 15 to 17, rather than inter-subunit contacts to residues 29 to 31.

As originally observed in ssNMR studies of Aβ40 fibrils (28), peptide segments in different layers of a multilayered cross-β structure do not necessarily lie directly on top of one another. Instead, interacting segments of a given molecule are frequently “staggered” or shifted relative to one another along the fibril growth direction. In type A fibrils, residues 17 to 20 of Aβ42 molecule k make contacts to residues 31 to 36 of molecules k-1 and k in the same subunit (SI Appendix, Fig. S8A). In type B fibrils, inter-subunit interactions of a given Aβ42 molecule span five β-sheet repeats in the opposite subunit (SI Appendix, Fig. S8B). Residues 2 to 4 of molecule k in one subunit make contacts to residues 29 to 31 of molecules k′+3 and k′+2 in the other subunit, residues 18 to 24 of molecule k make contacts to residues 39 to 41 of molecules k′+2 and k′+1, residues 32 to 36 of molecule k make contacts to residues 32 to 36 of molecules k′+1 and k′, residues 29 to 31 of molecule k make contacts to residues 2 to 4 of molecules k′-1 and k′-2, and residues 39 to 41 of molecule k make contacts to residues 18 to 24 of molecules k′ and k′-1.

Supporting Evidence from ssNMR.

Second-generation brain-seeded fibrils were grown with uniformly ¹⁵N,¹³C-labeled recombinant Aβ42, using the same first-generation seeds as in sample preparations for cryo-EM and AFM. Conditions for growing isotopically labeled fibrils for ssNMR measurements were nearly identical to conditions for growing the fibrils used for cryo-EM measurements, aside from differences in the quantities of fibrils that were prepared and minor variations in the molar ratio of initially solubilized Aβ42 to Aβ42 in seeds (SI Appendix, Supporting Methods). 2D ¹³C-¹³C and ¹⁵N-¹³C ssNMR spectra in Fig. 4 A and B were obtained at 17.5 T with standard ¹H-¹³C, ¹H-¹⁵N, and ¹⁵N-¹³C cross-polarization methods (29, 30), high-power ¹H decoupling, magic-angle spinning (MAS) at 18.00 kHz, and dipolar-assisted rotational resonance (31) in ¹³C-¹³C mixing periods. Under these conditions, nuclear spin polarization transfers that give rise to strong ssNMR signals are driven by magnetic dipole-dipole couplings, requiring that rapid, large-amplitude molecular motions be absent. Thus, these spectra report on immobilized segments of Aβ42 in the fibrils. The relatively broad crosspeak signals in these spectra are consistent with the extensive polymorphism observed in TEM, cryo-EM, and AFM, which precludes site-specific assignment of most signals. However, the unique NMR chemical shifts of the C-terminal carboxylate carbon of A42 (181.7 ppm), the ε-carbon of R5 (159.6 ppm), and the N-terminal amino nitrogen of D1 (19.6 ppm) allow crosspeak signals from these residues to be assigned, implying that a substantial fraction of the Aβ42 fibrils have immobilized N-terminal and/or C-terminal segments.

Fig. 4.

Characterization of N- and C-terminal mobility in brain-derived Aβ42 fibrils. (A and B) 2D ¹³C-¹³C and ¹⁵N-¹³C ssNMR spectra of polymorphic second-generation brain-seeded Aβ42 fibrils with uniform ¹⁵N and ¹³C labeling, recorded under conditions that show strong signals from immobilized peptide segments (i.e., nuclear spin polarization transfers driven by magnetic dipole-dipole couplings). Observation of crosspeak signals from D1, R5, and A42 indicates that the N- and C-termini are immobilized in at least a substantial fraction of the fibrils. (C) 2D ¹H-¹³C spectrum of the same sample, recorded under conditions that show strong signals from highly mobile segments (i.e., polarization transfers driven by scalar couplings). Observation of signals from amino acids contained in residues 1 to 9 of the Aβ42 sequence indicates that the N-terminal segment is mobile in a substantial fraction of the fibrils. Contour levels increase by successive factors of 1.15 in panel A and 1.20 in panels B and C.

The 2D ¹H-¹³C spectrum in Fig. 4C was obtained from the same sample, but using low-power ¹H decoupling and polarization transfers driven by one-bond ¹H-¹³C scalar couplings. Under these conditions, strong signals arise only from residues that are highly mobile, executing nearly isotropic motions on sub-microsecond timescales. Although these data do not permit site-specific assignments, residue-type assignments shown in Fig. 4C can be made by assuming that the ¹H and ¹³C chemical shifts of highly mobile residues are close to standard “random coil” values (32). These residue-type assignments match residues 1 to 9 of Aβ42, indicating that a substantial fraction of the Aβ42 fibrils have highly mobile N-terminal segments. Taken together, the 2D spectra in Fig. 4 are consistent with the coexistence of polymorphs with mobile and immobile N-terminal segments (type A and type B fibrils, respectively) and with immobilized C-termini (type A and type B).

Discussion

Comparisons with Previously Determined Aβ42 Fibril Structures.

Aβ42 fibrils are known to be polymorphic, with structural variations that may correlate with variations in AD characteristics (5–11). Previously reported high-resolution Aβ42 fibril structures share similar S-shaped backbone conformations (Fig. 5A). In ssNMR-based structures for fibrils grown in vitro (14–16) (Protein Data Bank (PDB) IDs 2MXU, 5KK3, and 2NAO), the S-shaped conformation in the fibril core comprises residues 15 to 42, with intramolecular K28-A42 salt bridges. In a cryo-EM-based structure for fibrils grown in vitro at low pH (19) (PDB ID 5OQV), the S-shaped conformation comprises residues 10 to 42, and K28-A42 salt bridges are replaced by inter-subunit K28-D1 salt bridges. In a cryo-EM study of brain-extracted Aβ42 fibrils, structures of two polymorphs were reported (20). Similar S-shaped Aβ42 conformations were found in type I and type II polymorphs (PDB IDs 7Q4B and 7Q4M, respectively), differing in the length of the structurally ordered segment (residues 9 to 42 for type I, residues 12 to 42 for type II) and sidechain orientations of H14, Q15, V18, and F19. Inter-subunit interaction surfaces in type I brain-extracted fibrils resemble those in the ssNMR-based structures (14–16), with the axis of quasi-twofold symmetry being centered between L34 sidechains of the two subunits in both cases (Fig. 5B). Inter-subunit contacts in type II brain-extracted fibrils are qualitatively different, with the quasi-symmetry axis being centered between K28 sidechains, which form inter-subunit salt bridges to the A42 carboxylate groups (Fig. 5C).

Fig. 5.

Comparisons of brain-derived type A and type B fibril structures with Aβ42 fibril structures reported in previous cryo-EM and ssNMR studies. (A) Superposition of peptide backbones from monomers in structures with PDB IDs 2MXU, 2NAO, 5KK3, 5OQV, 7Q4B, and 7Q4M (light green, yellow, purple, dark green, light pink, and gray, respectively). The S-shaped conformation for residues 15 to 42 in these previously reported structures for in vitro (2MXU, 2NAO, 5KK3, and 5OQV) and brain-extracted (7Q4B and 7Q4M) fibrils differs qualitatively from the ν-shaped and interlocking υ-shaped conformations in the type A and type B structures in Figs. 2 and 3. (B–D) Comparison of inter-subunit interfaces in type I (7Q4B), type II (7Q4M), and type A fibrils, respectively. Purple dots indicate the locations of twofold symmetry axes. (E–G) Comparison of contacts between sidechains of A30 and I32 and sidechains of M35 and V40, which occur between cross-β subunits in type A fibrils (blue and orange) and within each subunit in type I and type II fibrils (light pink and gray). (H–J) Comparison of inter-subunit contacts involving sidechains of L34 and V36, which are part of a wider interface between β-sheets in type B fibrils (magenta and cyan) than in type II fibrils (gray). Purple dots indicate the locations of twofold symmetry axes.

Aβ42 conformations in brain-derived type A and type B fibrils are ν-shaped and υ-shaped, rather than S-shaped. Interactions that apparently stabilize the type A and type B fibril structures also differ in a variety of ways from the stabilizing interactions in previously reported Aβ42 fibril structures. In type A fibrils, residues 28 to 42 form the inter-subunit interface, with contacts among hydrophobic sidechains of A30, I32, M35, V40, and I41, as well as possible K28-A42 salt bridges (Fig. 5D). Such an extended interface, formed by a single segment of the peptide sequence, does not exist in previously reported high-resolution Aβ42 fibril structures. Interestingly, type A fibrils and brain-extracted fibrils (both type I and type II) show similar contacts between A30 and V40 and between I32 and M35. However, these contacts are between cross-β subunits in type A fibrils, rather than within each subunit as in the brain-extracted fibrils (Fig. 5 E–G).

Inter-subunit contacts involving sidechains of L34 and V36 occur in both type B and type I fibrils. In type B fibrils, L34 and V36 are contained within a seven-residue β-strand, creating wider β-sheet surfaces for the inter-subunit interface (Fig. 5 H–J). In fact, the inter-subunit interface in type B fibrils is formed by a continuous 27-residue segment, from K16 to A42 (Fig. 3C), even longer than the 15-residue interface-forming segment in type A fibrils (Fig. 2C). The interlocking arrangement of subunits and the water-filled internal pores described above are unique structural features of type B fibrils.

Residues 2 to 8 form β-sheets both in type B fibrils and in Aβ42 fibrils grown in vitro at low pH (19). In the low-pH fibrils, ordering of this N-terminal segment is apparently driven by inter-subunit D1-K28 salt bridges and intra-subunit F4-V36 and F4-L34 contacts. In type B fibrils, inter-subunit A2-I31 and F4-I13 contacts apparently contribute to the stability of the N-terminal β-sheets.

The type B fibril structure resembles a structure for in vitro Aβ42 fibrils that was proposed by Schmidt et al. (33) from lower-resolution cryo-EM data (approximately 7 Å resolution), antibody binding results, and molecular modeling. Residues 17 to 42 in the model of Schmidt et al. adopt a conformation similar to that in Fig. 3C, but with different sidechain orientations and different inter-subunit interactions for residues 30 to 36. The N-terminal β-sheet and inter-subunit contacts between A2, F4, and I31 in Fig. 3E are not present in the model of Schmidt et al. Instead, their fibrils may contain intra-subunit contacts between the N-terminal segment and residues 15 to 17 (33), as discussed above.

A structure of Aβ40 fibrils extracted from meninges of AD patients has been reported by Kollmer et al., based on cryo-EM images (34). Although this structure differs in many respects from the type A and type B Aβ42 fibrils described above, the Aβ40 fibrils from meninges were found to have a right-handed twist as in type A fibrils.

Possible Relevance to Alzheimer's Disease.

Structures for type A and type B fibrils were determined from cryo-EM measurements on second-generation Aβ42 fibrils, prepared with synthetic and recombinant Aβ42 by seeded growth using first-generation fibrils as seeds. The first-generation fibrils were prepared with synthetic Aβ42 by seeded growth, using an amyloid-enriched extract from the parietal lobe tissue of an AD patient as the source of seeds as described by Qiang et al. (5). With the seeded growth protocols used in these preparations, we expect second-generation samples to contain the same fibril polymorphs as first-generation samples. This expectation is supported by previously reported ssNMR measurements on first- and second-generation brain-derived Aβ40 fibrils, which show accurate self-propagation of fibril structures in seeded growth (21, 35), and by the ssNMR data for first- and second-generation Aβ42 fibrils shown in the SI Appendix, Fig. S9. Self-propagation of Aβ42 fibril structures from the original brain tissue to first-generation samples can not be verified directly by ssNMR, because fibrils in the original brain tissue are not isotopically labeled and have insufficient quantity.

Our methods for preparing the amyloid-enriched extract and for growing first-generation samples are designed to avoid the selection of specific fibril polymorphs and to facilitate structural self-propagation. In particular, our extraction protocol results in amyloid-enriched material but is not a rigorous purification of Aβ fibrils that might select a subset of polymorphs (21). We use a single primary amplification step in which solubilized Aβ42 is incubated quiescently with sonicated extract for 4 h, after which abundant fibril growth from seeds is verified by TEM (5, 21). In control experiments, fibrils are not found after a 4-h incubation period when extract from amyloid-free brain tissue is used (5, 21). Additionally, fibrils grown from synthetic Aβ40 or synthetic Aβ42 with amyloid-free extracts have different ssNMR spectra than fibrils grown with amyloid-containing extracts (5, 11). Our seeded growth methods differ substantially from methods in a recent study of brain-derived α-synuclein fibrils, in which seeded fibrils did not appear before 20 h of incubation with intermittent agitation, and in which self-propagation of fibril structures was not observed (36). Nonetheless, the existence of the type A and type B structures in the original brain tissue is uncertain for two reasons: i) accurate propagation of structures from the original tissue may depend on cofactors that are lost in the extraction protocol, and ii) cross-seeding between Aβ40 and Aβ42 could create new polymorphs.

Yang et al. developed structural models for type I and type II fibrils from cryo-EM images of material that was extracted from brain tissue (20), without fibril amplification by seeded growth. In order to obtain fibrils that were sufficiently separated from one another and free of extraneous material from the tissue, they treated homogenized tissue with sarkosyl detergent and used centrifugation with alternating periods of moderate and high force to remove extraneous material and (presumably) masses of self-associated fibrils while retaining fibrils that were suitable for cryo-EM measurements. As is usually the case in cryo-EM studies, the majority of particles extracted from the images were not used in their final density maps. Therefore, it is conceivable that type A or type B fibrils were present in the brain tissue used by Yang et al. but were lost in centrifugation steps or could not be analyzed due to self-association or association with extraneous material. We note that fibrils resembling the type II fibrils of Yang et al. were also present in our cryo-EM images, but with insufficient quantities for high-resolution 3D reconstruction (SI Appendix, Fig. S10). Type I and type II fibrils have recently been observed in aqueous extracts from AD brain tissue, without sarkosyl treatment and after ultracentrifugation (37).

In conclusion, we have characterized the molecular structures of two Aβ42 fibril polymorphs with a variety of conformational features and sets of stabilizing interactions that have not been observed in previous studies. These fibrils were derived from brain tissue of an AD patient by seeded growth. We have not established that the same structures exist in AD brain tissue, but their existence in AD brain tissue is a possibility. Structures of type A and type B fibrils show that the compact S-shaped conformation seen in previous studies is not required for fibril formation by Aβ42 and that Aβ42 fibrils with wide, continuous interfaces between cross-β subunits can form. By expanding our understanding of amyloid polymorphism, these results may have implications for structure-based design of amyloid imaging agents and aggregation inhibitors.

Methods Summary

Full details of sample preparations, measurements, data analyses, and simulations are given in SI Appendix. Briefly, second-generation Aβ42 fibrils were grown from synthetic or recombinant peptide, using sonicated first-generation brain-seeded fibrils described by Qiang et al. (5) as seeds. TEM and AFM images were obtained as previously described (5, 24). Cryo-EM grids were screened with a Thermo Fisher Glacios microscope. Final images were obtained with a Thermo Fisher Titan Krios microscope. Cryo-EM images were processed and density maps were reconstructed with RELION software (18, 24). ssNMR data were obtained at 14.1 T and 17.5 T, using Tecmag Redstone spectrometers, magic-angle spinning probes obtained from the research group of Drs. Ago Samoson and from Black Fox, LLC, and standard pulse sequence methods. MD simulations were performed with NAMD software and analyzed with VMD software (38, 39).

Supplementary Material

Appendix 01 (PDF)

Movie S1.

Molecular dynamics simulation for type A Aβ42 fibrils. All-atom simulations for eight Aβ42 molecules (residues 12–42) were performed in explicit water solvent at 303 K with 100 mM NaCl, starting with the type A fibril structure in Fig. 2 and running for 200 ns. Sidechain carbon atoms of hydrophobic, polar, negatively charged, and positively charged residues are colored green, magenta, red, and blue, respectively. Water oxygen atoms are also colored red. To keep the Aβ42 octamer structure centered in each movie frame, overall rotations and translations were applied to minimize the peptide backbone RMSD relative to the initial structure.

Movie S2.

Molecular dynamics simulation for type B Aβ42 fibrils. Same as Movie S1, but starting with the type B fibril structure in Fig. 3 and including residues 1–42 of Aβ42.

Data, Materials, and Software Availability

Density maps for type A and type B fibrils are available from the Electron Microscopy Data Bank, codes 28740 and 28741, respectively.  Atomic coordinates for structural models are available from the Protein Data Bank, ID codes 8EZD and 8EZE.  MD trajectories and 2D ssNMR spectra are available at https://doi.org/10.17632/jznmb7mxpm.1. Alternative density maps calculated without symmetry and with C2 symmetry are available at https://doi.org/10.17632/wr9yvx8wjf.1. Previously published data were used for this work [published in Qiang et al. (5)].