Introduction

Alzheimer’s disease (AD) is defined neuropathologically by the presence of abundant filamentous plaques and tangles in the brain parenchyma, with the characteristics of amyloid [1]. Plaques are made predominantly of Aβ42, whereas tangles consist of hyperphosphorylated microtubule-associated protein tau. In addition, many individuals with AD have deposits of Aβ40 in leptomeningeal and parenchymal blood vessels, chiefly small arterioles and capillaries, giving rise to Aβ cerebral amyloid angiopathy (CAA) [2, 3]. CAA is the accumulation of amyloidogenic proteins, most often Aβ40, in cerebral blood vessel walls. Nearly half of end-stage AD patients have moderate-to-severe CAA [4]. Even though they share Aβ deposition, AD and CAA can occur independently [5]. Aβ CAA is mostly sporadic, but it can also be inherited [6, 7] or acquired [8, 9]. The most common clinical manifestations are lobar intracerebral haemorrhage and cognitive impairment.

In 2019, the cryo-EM structures of Aβ40 filaments from the leptomeninges of three individuals with AD and CAA were reported at a resolution of 4.4 Å [10]. For each individual, three types of filaments were observed. In the smallest filament type, two identical protofilaments were in evidence, with the C-terminus being less well resolved than the rest of the molecule. Four cross-β strands were typical of each protofilament and the filament twist was right-handed. Filament types made of four and six protofilaments were also observed. Here we report the structures of Aβ40 filaments from the leptomeninges of three individuals with AD and CAA, with resolutions up to 2.4 Å. Compared to the previously reported structures [10], our higher resolution structures reveal a difference in the sequence assignment that redefines the filament fold of Aβ40 peptides and their interactions. Using either sarkosyl or water-based methods to extract leptomeningeal Aβ40 filaments led to identical folds.

Materials and methods

Cases of Alzheimer’s disease and cerebral amyloid angiopathy

Leptomeninges from three neuropathologically confirmed cases of AD and CAA were peeled off from the region overlying the thawed cerebral cortex, before being frozen a second time. Case 1 was a male who died with sporadic late-onset AD aged 89 years; the duration of illness was 12 years and his brain weight was 1098 g. Co-pathologies were limbic-predominant age-related TDP-43 encephalopathy and cerebrovascular disease. Case 2 was a male who died with sporadic early-onset AD aged 61 years; the duration of illness was 6 years and his brain weight was 1048 g. There was a co-pathology of cerebrovascular disease. Case 3 was a female who died with sporadic early-onset AD aged 70 years; the duration of illness was 9 years and her brain weight was 1380 g. Diffuse Lewy body disease and cerebrovascular disease were co-pathologies.

Immunohistochemistry and APOE isotyping

Immunohistochemistry using anti-Aβ antibody NAB228 (1:200, Invitrogen) was carried out as described [11]. This monoclonal antibody is specific for the N-terminal region of Aβ. Leptomeningeal sections were 8 μm thick and were counterstained with haematoxylin. Isotypes of APOE were determined using HhaI digestion and gel electrophoresis of amplified DNA isolated from frontal cortex of cases 1–3, as described [12].

Filament extraction

Sarkosyl-insoluble material was extracted from the leptomeninges of cases 1–3, as described [13]. Tissues were homogenised in 20 vol buffer A (10 mM Tris–HCl, pH 7.5, 0.8 M NaCl, 10% sucrose and 1 mM EGTA), brought to 2% sarkosyl and incubated for 30 min at 37 °C. The samples were centrifuged at 10,000 g for 10 min, followed by spinning of the supernatants at 100,000 g for 60 min. The pellets were resuspended in 50 μl/g 20 mM Tris–HCl, pH 7.4, 50 mM NaCl and used for cryo-EM. For aqueous extraction, as described previously [10, 14], leptomeningeal tissues (0.2–0.5 g) were cut with a scalpel into cubes of less than 1 mm3 and washed 4 times with 500 μl Tris-Calcium buffer (20 mM Tris, pH 7.4, 138 mM NaCl, 2 mM CaCl2, 0.1% (w/v) NaN3, pH 8.0) at 4 °C. Following each wash, the samples were centrifuged at 3000 g for 1 min, with a 5 min spin at 12,000 g after the final wash. The pellets were then incubated overnight with 5 mg/ml collagenase from Clostridium histolyticum (Sigma-Aldrich) in 1 ml Tris-Calcium buffer at 37 °C. Following a 5 min centrifugation at 12,000 g at 4 °C, they were washed 5 times with 500 μl 50 mM Tris–HCl, pH 7.4, 10 mM EDTA, with a 5 min centrifugation at 12,000 g after each wash. To collect Aβ filaments, 250 μl cold water was added to the pellets, followed by a 5 min centrifugation at 12,000 g. This step was repeated nine times. The supernatants were combined and used for subsequent experiments.

Mass spectrometry

Sarkosyl-insoluble pellets were resuspended in 200 μl hexafluoroisopropanol. Following a 3 min sonication at 50% amplitude (QSonica), they were incubated at 37 °C for 2 h and centrifuged at 100,000 g for 15 min, before being dried by vacuum centrifugation. Matrix-assisted laser desorption/ionization time of flight (MALDI-TOF) mass spectrometry was carried out.

Electron cryo-microscopy

Three μl of the sarkosyl-insoluble fractions were applied to glow-discharged (Edwards S150B) holey carbon grids (Quantifoil Au R1.2/1.3, 300 mesh) that were plunge-frozen in liquid ethane using a Vitrobot Mark IV (Thermo Fisher Scientific) at 100% humidity and 4 °C. Cryo-EM images were acquired using EPU software on Titan Krios G2 and G4 microscopes (Thermo Fisher Scientific) operated at 300 kV. Images for case 1 were acquired using a Falcon-4i detector (Thermo Fisher Scientific) in electron-event representation mode with a flux of 11 electrons/pixel/s and a Selectris-X energy filter (Thermo Fisher Scientific) with a slit width of 10 eV to remove inelastically scattered electrons. Images for case 2 were acquired using a Falcon-4i detector without energy filter. Images for case 3 were acquired using a Gatan K3 detector in super-resolution counting mode, using a Bio-quantum energy filter (Gatan) with a slit width of 20 eV. Images were recorded with a total dose of 40 electrons per A2. See Additional file 1: Tables S1 and S2 for further details.

Helical reconstruction

Datasets were processed in RELION using standard helical reconstruction [15]. Movie frames were gain-corrected, aligned and dose-weighted using RELION’s own motion correction programme [16]. Contrast transfer function (CTF) was estimated using CTFFIND4-1 [17]. Filaments were picked manually. Following 2D classification (Additional file 2: Fig. S1), polymorphs were identified using the clustering approach FilamentTools (https://github.com/dbli2000/FilamentTools) [18]. For case 1, most filaments corresponded to 2D class averages that showed twisting Aβ40 filaments suitable for 3D reconstruction, and multiple different types were identified (Additional file 2: Figs. S2 and S3). Because, for case 2 (Additional file 2: Fig. S4) and case 3 (Additional file 2: Fig. S5), filaments clumped together in the micrographs, the majority of picked filaments gave rise to relatively featureless 2D class averages that were not suitable for 3D reconstruction, and only a single type of twisting Aβ40 filaments was identified. Numbers of picked particles and numbers of selected particles for the different filament types of each dataset are given in Additional file 1: Table S2. Initial models were generated de novo from 2D class average images using relion_helix_inimodel2d [19]. Three-dimensional auto-refinements were performed with optimisation of the helical twist and rise parameters once the resolutions extended beyond 4.7 Å. To improve the resolution, Bayesian polishing and CTF refinement were used [20]. Final maps were sharpened using standard post-processing procedures in RELION and resolution estimates calculated based on the Fourier shell correlation (FSC) between two independently refined half-maps at 0.143 [21] (Additional file 2: Fig. S1).

Model building and refinement

Atomic models were built manually in the reconstructed cryo-EM densities using Coot [22]. Model refinements were performed using Servalcat [23] and REFMAC5 [24, 25]. Models were validated with MolProbity [26]. Figures were prepared with ChimeraX [27] and PyMOL [28].

Results

Structures of Aβ40 filaments from leptomeninges extracted using sarkosyl

We determined the cryo-EM structures of Aβ filaments from the leptomeninges of three cases of AD and CAA (Fig. 1), with APOE genotypes: ε3/ε3 (case 1); ε3/ε3 (case 2); ε3/ε4 (case 3). There were two main types of filaments, the minority type (type 1) comprising a pair of identical protofilaments and the majority type (type 2) comprising two protofilament pairs running side by side. All filaments had a right-handed twist, as established from the densities for backbone oxygen atoms in the cryo-EM maps. Structures from case 1 reached the highest resolution. The structure of filaments made of two pairs of protofilaments were determined at 2.4 Å; the structure of filaments made of one pair of protofilaments was determined at 2.7 Å (Fig. 2a; Additional file 2: Fig. S1).

Fig. 1
figure 1

Aβ immunoreactivity of leptomeningeal blood vessels. Anti-Aβ antibody NAB228 was used to stain the leptomeninges from Alzheimer’s disease cases 1 (a), 2 (b) and 3 (c). Scale bar, 200 μm

Full size image
Fig. 2
figure 2

Structures of Aβ40 filaments from leptomeninges extracted using sarkosyl. a Cross-sections of Aβ40 filaments from cases 1–3 perpendicular to the helical axis, with a projected thickness of approximately one rung. Percentages of filaments are shown on the top right. The resolutions of the cryo-EM maps are indicated on the bottom left. Filaments were made of one (type 1) or two (type 2) pairs of protofilaments. Scale bar, 1 nm. b Cryo-EM density maps (transparent grey) and atomic model (cyan) of type 2 Aβ40 filaments with two pairs of protofilaments. c Schematic of the structure shown in panel b. Negatively charged residues are shown in red, positively charged residues in blue, polar residues in green, non-polar residues in white, sulphur-containing residues in yellow and glycines in pink. Thick connecting lines with arrowheads indicate β-strands. d Amino acid sequence of the core of Aβ40 protofilaments, which extends from D1 to G38. Beta-strands (β1–β4) are indicated as thick arrows

Full size image

The ordered core of each protofilament consists of residues D1–G38 and comprises four β-strands (β1–β4) that extend from residues 2–8, 10–13, 15–22 and 34–36, respectively (Fig. 2b–d). In type 1 filaments, the individual protofilaments pack against each other with 2-start (pseudo-21) helical symmetry through a large interface, made of residues H14–G37 in an extended conformation. This interface is stabilised by hydrophobic interactions of L17, F19, A21, V24, I32 and M35 from both protofilaments and hydrogen bonds from the polar side chains of H14, Q15 and N27 to the main chain of the opposite molecules. The N-terminal strands β1 and β2 fold back on the opposite side of β3, enclosing a network of electrostatic interactions and hydrogen bonds formed by the polar side chains of H6, S8, E11, H13 and K16, and capped by the interlocking aromatic side chains of F4 and F20. The N-terminal substructure formed by the β1–β3 region is similar to that derived from the low-resolution map [10].

Type 2 filaments comprise two pairs of protofilaments that pack against each other with C2 symmetry (Fig. 2b). Pairs of leptomeningeal Aβ40 protofilaments interact through salt bridges between E3 from one dimer and R5 from the other dimer, and vice versa. A similar association of dimeric filaments into tetramers via electrostatic interactions of the surface residues K16 and E22 was observed in Aβ42 Type 1b filaments [13].

Besides the density for the ordered core of Aβ filaments, there were additional densities in the cryo-EM reconstruction that could not be explained by Aβ. There was a strong density in front of Y10, which was situated at a right angle turn between β1 and β2, which may correspond to a post-translational modification of this residue and/or an external cofactor. There was another strong density opposite a hydrophobic patch made of I31, L34 and V36 in the C-terminal region. In the previously reported low-resolution maps [10], there was a disconnected density in the equivalent location that was assigned to the C-terminal residues V39 and V40. The unambiguous sequence assignment of our higher-resolution structures renders such an assignment impossible. In principle, the C-terminal residues of Aβ42 or Aβ43 peptides could reach this disconnected density, but these peptides were uncommon in the cases studied here.

By mass spectrometry, Aβ1-40 showed the highest peak in all three cases (Fig. 3), indicating that most filaments were made of Aβ1-40. In case 3, a comparable peak was observed for Aβ1-38, consistent with the ordered filament core extending from D1 to G38. Our findings on sarkosyl-extracted filaments are similar to those reported previously following aqueous extraction of Aβ filaments from leptomeninges [10].

Fig. 3
figure 3

Mass spectrometry. MALDI mass spectra of Aβ from the sarkosyl-insoluble fractions of the leptomeninges of cases 1–3. The peptides with their molecular mass are labelled on top of the spectra

Full size image

Structures of Aβ40 filaments from leptomeninges extracted using aqueous extraction

We also used aqueous extraction of Aβ filaments from the leptomeninges of case 1. Cryo-EM structures of type 1 and type 2 filaments, at 3.1 Å and 2.7 Å resolution, respectively, revealed the same Aβ40 fold as observed using sarkosyl extraction (Fig. 4a–c). Aqueous extraction also yielded a third type of filaments with three such pairs of protofilaments (type 3), which was solved at 4.0 Å resolution (Fig. 4a). The all-atom root mean square deviation (r.m.s.d.) between protofilaments from type 1 filaments of sarkosyl and aqueous extractions was 1.1 Å. Minor differences were due to a distinct side chain orientation of Y10. The additional density next to this residue was less prominent and more diffuse than in the maps of sarkosyl-extracted filaments, suggesting that a putative cofactor associated with this residue was partially lost during aqueous extraction.

Fig. 4
figure 4

Structures of Aβ40 filaments from leptomeninges extracted using a water-based method. a Cross-sections of Aβ40 filaments from case 1 perpendicular to the helical axis, with a projected thickness of approximately one rung. Percentages of filaments are shown on the top right. The resolutions of the cryo-EM maps are indicated on the bottom left. Filaments were made of one, two or three pairs of protofilaments. Scale bar, 1 nm. b Cryo-EM density maps (transparent grey) and atomic model (pink) of type 2 Aβ40 filaments with two pairs of protofilaments. c Superposition of the structures of type 2 Aβ40 filaments that were extracted using either sarkosyl (cyan) or an aqueous solution (pink)

Full size image

Comparison with previous structures

A comparison of our type 1 filament structure at 2.7 Å with the previously reported structure at 4.4 Å [10] revealed differences in the atomic models (Fig. 5a, b). The same differences were also present when comparing the previously reported structure of type 1 filaments with the pairs of protofilaments in our 2.4 Å structure of type 2 filaments (not shown). Compared to our model, the previously reported model has the same sequence assignment for the N-terminal 23 residues of Aβ40, but the rest of the sequence is misaligned by up to two residues (Fig. 5c, d). Whereas the main-chain assignment is unambiguous in our high-resolution maps, the density in the 4.4 Å map was less clear. It is possible that the misalignment arose from an erroneous assignment of the C-terminal residues of Aβ40 into the disconnected density near the end of the contiguous segment. Fitting our model into the 4.4 Å map (EMD-10204) [10] and comparing it with the deposited model (PDB:6SHS), shows that our model is as good a description of the 4.4 Å map as the deposited model (Fig. 5e).

Fig. 5
figure 5

Comparison of the maps and models of Aβ40 filaments described here with those of Kollmer et al. [10]. a Map of Aβ40 filaments from leptomeninges with two protofilaments extracted using sarkosyl (present work; in orange). N- and C-termini are indicated. b Map of Aβ40 filaments from leptomeninges with two protofilaments extracted using a water-based method (previous work; in grey). N- and C-termini are indicated. c Atomic models of Aβ40 filaments for the maps shown in panels a and b (present work in orange; previous work in grey). The blue arrow points to V24, from which the two models differ in the register of their main chains. d Zoomed-in view of the structures in panel c. e Atomic model from present work (orange) and from previous work (grey) fitted into the 4.4 Å map from the previous work

Full size image

Discussion

We report the cryo-EM structures of Aβ filaments from the leptomeninges of three cases of sporadic AD and CAA with resolutions up to 2.4 Å. In agreement with a previous report [10], we observed three types of right-handed Aβ40 filaments that consisted of one, two or three pairs of protofilaments. Each protofilament extended from D1 to G38 and consisted of four β-strands. It follows that Aβ monomers in these filaments have an intact N-terminus and are truncated or disordered for the two C-terminal residues. This contrasts with the structures of Aβ42 filaments from the brains of individuals with AD [13], in which the N-terminal eight or ten residues are disordered and partially truncated, and the C-terminus is ordered. Aβ40 filaments also differ from Aβ42 filaments by the presence of a strong density associated with Y10.

Sarkosyl extraction and aqueous extraction gave nearly identical filament structures, demonstrating that sarkosyl did not influence the structure of Aβ40 filaments. The same was true of Aβ42 filaments and paired helical tau filaments [29]. When using sarkosyl extraction of AD brain tissues, we observed Aβ42 filaments [13]. Only Aβ40 filaments were found when leptomeninges from cases of AD and CAA were extracted with sarkosyl. Both Aβ40 and Aβ42 filaments were found in the brain of a patient with mutation E22G in Aβ (Arctic mutation) following sarkosyl extraction [30].

Our atomic models differ from those reported previously for Aβ40 filaments from the leptomeninges of three individuals with sporadic AD and CAA [10]. Our unambiguous sequence assignment differs from that proposed previously for a low-resolution map [10], thereby redefining the fold of Aβ40 peptides and their interactions. It has been recommended that de novo atomic modelling of amyloids should not be performed in RELION reconstructions, when the resolutions are worse than 4.0 Å [19].

Seeded aggregation of amyloid filaments is often used, with the implicit assumption that these filaments have the same structures as the initial seeds. However, this is not necessarily the case, as was shown for seeded aggregation of recombinant α-synuclein with seeds from the putamen of individuals with multiple system atrophy [31]. The cryo-EM structures of synthetic Aβ40 filaments seeded from vascular deposits of brain parenchyma of individuals with CAA have recently been reported [32]. One of the two structures described in this preprint appears to be identical to the structure of the type 1 Aβ40 filaments that we report, including the new sequence assignment of the 15 C-terminal residues. Without the results described here, this difference in sequence assignment of the seeded structure could have been attributed to a difference between Aβ40 filaments from CAA and CAA + AD, a difference between Aβ40 filaments from leptomeningeal and parenchymal blood vessels, or an artefact of seeded aggregation. However, the present results suggest that the seeds used in [32] contained the same type 1 Aβ40 filaments that we extracted and that their structures were replicated correctly during seeded aggregation.

The cryo-EM structures of Aβ42 filaments from plaques and Aβ40 filaments from blood vessels are different, consistent with the view that AD and CAA are the result of distinct pathogenic processes. For assembled tau and α-synuclein, it has also been shown that specific amyloid structures define different diseases [33]. Medin has been reported to co-aggregate with Aβ40 in blood vessels [34], but we did not observe mixed filaments.

The present findings may have consequences for the development of disease-modifying treatments of AD. Current immunotherapies suffer from common and sometimes serious side effects, also known as amyloid-related imaging abnormalities (ARIAs) [35, 36], which appear to be linked to the abundance of CAA [37]. Existing antibodies seem to bind to both parenchymal and blood vessel deposits of Aβ. Based on the structural differences, it may be possible to produce conformational Aβ antibodies that bind specifically to either parenchymal or blood vessel deposits. It may also become possible to develop imaging agents that distinguish between parenchymal and blood vessel deposits of Aβ. Accurate knowledge of the atomic structures of the different amyloids will facilitate these developments.