Sensitivity-Enhanced Solid-state NMR Detection of Structural Differences and Unique Polymorphs in Pico- to Nanomolar Amounts of Brain-derived and Synthetic 42-residue Amyloid-β Fibrils

Abstract

Amyloid-β (Aβ) fibrils in neuritic plaques are a hallmark of Alzheimer’s disease (AD). Since the 42-residue Aβ (Aβ42) fibril is the most pathogenic among different Aβ species, its structural characterization is crucial to our understanding of AD. While several polymorphs have been reported for Aβ40, previous studies of Aβ42 fibrils prepared at neutral pH detected essentially only one structure, with an S-shaped β-sheet arrangement [e.g., Xiao et al., Nat. Struct. Mol. Biol. 2015, 22, 499]. Herein, we demonstrate the feasibility of characterizing the structure of trace amounts of brain-derived and synthetic amyloid fibrils by sensitivity-enhanced ¹H-detected solid-state NMR (SSNMR) under ultra-fast magic angle spinning (UFMAS). By taking advantage of the high sensitivity of this technique, we first demonstrate its applicability for the high-throughput screening of trace amounts of selectively ¹³C- and ¹⁵N-labeld Aβ42 fibril prepared with ~0.01% patient-derived amyloid (ca. 4 pmol) as a seed. The comparison of 2D ¹³C/¹H SSNMR data revealed marked structural differences between AD-derived Aβ42 (~40 nmol or ~200 μg) and synthetic fibrils in less than 10 min, confirming the feasibility of assessing the fibril structure from ~1 pmol of brain amyloid seed in ~2.5 h. We also present the first structural characterization of synthetic fully-protonated Aβ42 fibril by ¹H-detected 3D and 4D SSNMR. With procedures assisted by automated assignments, main-chain resonance assignments were completed for trace amounts (~42 nmol) of a fully-protonated amyloid fibril in the ¹H-detection approach. The results suggest that this Aβ42 fibril exhibits a novel fold or polymorph structure.

Graphical Abstract

Introduction

Alzheimer’s disease (AD) is a fatal neurodegenerative disorder characterized by the accumulation of neuritic plaques and neurofibrillary tangles outside and inside the neurons, respectively.^1-3 While neurofibrillary tangles are composed of tau aggregates,⁴ neuritic plaques mainly consist of fibrillar aggregates of 39–43 residue-long amyloid-β (Aβ) fibrils.⁵ Among these structures, 40- and 42-residue Aβ (Aβ40 and Aβ42, respectively) fibrils are the predominant isoforms.^6-7 Despite their amino acid sequences being different only in two residues located at the C-terminus, these two fibril forms show significantly different characteristics. Aβ42 has been suggested to be more neurotoxic and prone to aggregation compared to Aβ40.^8-11 Because the Aβ42 fibril is considered more pathogenic with respect to the AD,^11-13 its structural features are the key for developing an AD treatment targeting Aβ aggregation. However, conventional structural tools such as solution NMR and X-ray crystallography are not suitable for characterizing Aβ fibrils, owing to their insoluble and non-crystalline nature. Although cryo-electron microscopy (cryo-EM) has been used to study the structure of non-crystalline amyloid proteins, currently it is only applicable to determine the structure of twisted fibrils with near-atomic resolution.^14-16

SSNMR spectroscopy is a powerful technique for elucidating the atomic-level structure of Aβ and other amyloid fibrils.^17-43 Although several reports have studied the atomic-level structure of Aβ40 fibril and its mutants by SSNMR,^{18-31, 44-45} only three high-resolution structural models based on SSNMR are available for Aβ42 fibrils prepared at physiologically relevant neutral pH.^27-29 These three structures present a common structural motif, with S-shaped parallel β-sheet folds having similar ¹³C shifts. Although a recent cryo-EM and SSNMR study suggested another structural motif for an Aβ42 fibril sample prepared at a low (~2) pH,¹⁴ relatively little information is available on structural variations of Aβ42 fibrils prepared at physiologically relevant conditions, despite their pathological importance. A major obstacle to the characterization of the Aβ42 fibrils comes from their structural heterogeneity,^{19, 21, 25, 32, 46} which often prevents the preparation of samples suitable for SSNMR analysis. All atomic models of Aβ42 fibrils were obtained using ¹³C-detected SSNMR under magic angle spinning (MAS) at low to moderate frequencies (12–20 kHz). Generally, due to its low sensitivity, ¹³C-detected SSNMR requires several mg of sample to record NMR signals with signal-to-noise (S/N) ratios sufficient for signal assignment or structure determination.^{25, 27-29, 47} The requirement of large amounts of isotope-labeled amyloid samples for ¹³C-detected SSNMR analysis is extremely demanding, especially with the difficulties associated with preparing homogeneous amyloid samples. Notably, it has proven impossible to characterize the structures of patient-derived amyloid samples by SSNMR without sacrificing a relatively large section of brain tissue (1–3 g).²⁵

In this study, we evaluate the applicability of ¹H-detected SSNMR to characterize homogeneous recombinant/synthetic Aβ42 fibrils and heterogeneous brain-derived Aβ42 fibrils that are available only in pico- to nano-molar amounts. ¹H-detected SSNMR is attracting growing interest as a practical tool for studying biological systems.^47-56 Especially, ultra-fast MAS (UFMAS) at a frequency of 60 kHz or above offers high ¹H resolution for SSNMR, by removing line broadening due to ¹H-¹H dipolar coupling.^{47, 50, 54} With a much higher sensitivity per given amount of sample (mass-sensitivity) compared with its traditional ¹³C-detected approach, ¹H-detected SSNMR under UFMAS is well suited for the characterization of trace amounts of biologically relevant compounds.^51-55 The sensitivity of ¹H-detected SSNMR can be enhanced by incorporating paramagnetic-assisted condensed data collection (PACC),⁵⁷ drastically reducing the repetition delays and accumulation time of the SSNMR data. By taking advantage of the high sensitivity of ¹H-detected SSNMR achieved with these methods, in this work we demonstrate the fast collection of 2D ¹³C_α/¹H_α correlation SSNMR spectra from ~200 μg of in vitro-prepared and brain-derived Aβ42 fibrils that were isotopically labeled at selected residues. Using their SSNMR spectra as fingerprints of amyloid fibril structures, we were able to characterize the structure of amyloid fibril within 33 sec to 8.7 min. As the brain-derived Aβ42 fibril was prepared by growing ¹³C- and ¹⁵N-labeled Aβ42 fibril with approximately 0.010% (~20 ng or ~4 pmol) of amyloid from AD patients as a seed, a minimal amount of patient tissues is required with the present approach. Furthermore, we successfully applied the ¹H-detected SSNMR technique under UFMAS at 90 kHz for the structural characterization of a new Aβ42 fibril polymorph prepared from a uniformly ¹³C- and ¹⁵N-labeled sample produced with bacterial protein expression (without brain seed in this case). The excellent sensitivity of ¹H-detected SSNMR and the dramatic enhancement of the ¹H resolution by UFMAS enabled us to assign most of the resonances in both the protein backbone and side chains using a combination of 3D and 4D SSNMR experiments. Importantly, we were able to record the SSNMR data using only ~42 nmol (~200 μg) of the Aβ42 fibril sample, which is 25–100 times less than the amounts used in earlier studies.^{14, 27-29, 58-60} We accomplished signal assignment of the chemical shifts based on the 2D–4D SSNMR data with the aid of automated assignment procedures. The comparison of the ¹³C chemical shifts with those of Aβ42 fibrils with known structures reported in previous studies suggests that our Aβ42 fibril may have a unique structure that was not reported to date. To the best of our knowledge, this is the first ¹H-based signal assignment reporting the chemical shifts of both the protein backbone and side chains of amyloid fibrils. Our study will promote the SSNMR-based structural analysis of patient-derived ex vivo amyloid samples that are typically available only in sub-nanomolar amounts.

Results and Discussion

Screening of Picomolar Amounts of Brain-derived Amyloid Fibrils by ¹H-detected SSNMR

The atomic details of Aβ42 fibrils provide insight into how the amyloid structure is stabilized via molecular contacts and potentially into how therapeutic agents can disrupt toxic aggregates. In addition, very little information is available on the structures of Aβ42 fibrils obtained from AD patients.^{15, 25-26, 32, 61} Therefore, investigating whether the structure of in vitro-prepared Aβ fibrils actually reflects that of amyloid species in AD brain tissue is a task of great significance. However, the difficulty in achieving mass production of brain-derived Aβ42 fibrils, due to limited sample availability, demands more sensitive and efficient SSNMR techniques to enable the structural comparison of synthetic and AD-derived Aβ fibrils. Herein, we compared the ¹H-detected ¹³C_α/¹H_α correlation 2D spectra of in vitro-prepared and brain-derived Aβ42 fibril samples (Figure 1a and b, respectively). Note that the samples were uniformly ¹³C- and ¹⁵N-labeled at selected residues (see caption).

Figure 1.

(a, b) ¹H-detected ¹³C/¹H correlation 2D SSNMR spectra of (a) in vitro-prepared and (b) brain-derived Aβ42 fibrils that were uniformly ¹³C- and ¹⁵N-labeled at Phe-20, Ala-21, Val-24, Gly-25, and Leu-34. The signal assignments are shown in the spectra. SSNMR spectra were recorded at a MAS rate of 90 kHz on an 800 MHz Bruker Avance III NMR spectrometer equipped with a JEOL 0.75-mm ¹H/¹³C/¹⁵N/²H quad-resonance probe (see Materials and Methods for details). The pulse sequence in Figure S1 was used to collect the data. The experimental time was (a) 33 sec and (b) 8.7 min. (c, d) Transmission electron micrograph (TEM) images of the (c) synthetic and (d) brain-derived Aβ42 fibril samples.

Because of the high sensitivity of our ¹H-detected SSNMR approach using UFMAS, the spectrum in Figure 1a was obtained in only 33 s for ~200 μg of the synthetic Aβ42 fibril sample (Figure 1c), which was prepared following the procedure described in ref. 62. The signal assignments of the amino acid-dependent ¹³C_α chemical shifts indicate that one relatively sharp peak correlated the ¹H_α and ¹³C_α resonances for each labeled residue, suggesting that the sample is made of a single conformer of Aβ; as reported in our previous study, the Aβ42 fibril is known to adopt an S-shaped triple β-sheet arrangement.²⁷ The brain-derived Aβ42 fibril sample (~200 μg) (Figure 1b, d), which was uniformly isotope-labeled at the same residues as those of Figure 1a, was obtained by incubating monomeric ¹³C- and ¹⁵N-labeled Aβ42 with ~0.010% of brain amyloid (~20 ng or ~4 pmol) as seed (see the Materials and Methods section for further details).

To our surprise, the spectrum in Figure 1b shows significantly different chemical shifts from those in Figure 1a. Furthermore, there is more than one peak corresponding to each isotope-labeled residue. Our assignment of Ala-21, for example, suggests the presence of at least one major and one minor peak, indicating two different structures. The major peak does not match the corresponding peak position for Ala-21 in Figure 1a. Similar trends were observed for the Val-24 and Gly-25 residues. These results suggest that the main conformer of the brain-derived Aβ42 fibril has a considerably different structure from the S-shaped β-sheet arrangement of the synthetic Aβ42.²⁷ The presence of the two different conformers limits the sensitivity of Figure 1b compared with that of Figure 1a. It should also be noted that the brain-derived Aβ42 sample contained considerable amounts of brain materials (5.9%) other than Aβ42 fibrils, unlike the synthetic fibril used for Figure 1a. Despite the disadvantages associated with the “native” amyloid sample, we could record a ¹³C/¹H correlation 2D spectrum of the brain-derived Aβ42 fibril within 8.7 min (compared to 33 s for the synthetic Aβ42 fibril). Thus, our analysis could be completed within 2.5 h, even from the same fibril sample of ~10 nmol prepared from ~1 pmol of brain seed. These results show that sensitivity-enhanced ¹H-detected SSNMR provides a novel high-throughput route to highlight structural differences between synthetic and patient-derived amyloid fibrils.

For a more precise analysis, we assigned the ¹³C, ¹⁵N, and ¹H resonances of the brain-derived Aβ42 fibril sample by recording ¹H-detected (H)CCH and (H)CANH 3D SSNMR spectra (Tables S1). Based on our analysis, we identified 3–4 peaks for each isotope-labeled residue, suggesting that the brain-derived Aβ42 sample may contain up to 3–4 different conformers. Then, we compared the ¹³C secondary chemical shifts of these conformers with those of the previously reported brain-derived and synthetic Aβ42 and Aβ40 fibrils (Table S2).^{14, 20, 25, 27, 30, 32, 63} The comparison clearly shows that the conformations of the fibrils in our brain-derived Aβ42 sample are distinct from those of the other amyloid fibrils at the compared sites. It should be noted that the brain material used as seeds in preparation of the patient-derived Aβ42 fibrils contained not only Aβ42 seeds (86%), but also limited Aβ40 seeds (14%) and other non-amyloid brain material. Thus, there is a possibility that cross-seeding from Aβ40 fibril might have occurred, simultaneously, although a majority should be self-seeded with brain Aβ42 fibrils considering the inefficiency of the cross-seeding.²⁷

New Polymorph for Aβ42 Fibril

Next, we performed a SSNMR analysis of a structurally homogeneous and uniformly ¹³C- and ¹⁵N-labeled Aβ42 fibril that was newly obtained from a bacteria-expressed sample following a seeding protocol described in a previous study.⁶² A protein expression system in E. coli BL21 was used to express the Aβ42 peptide using our previously published protocol,⁶⁴ with minor modifications. Figure 2 shows TEM images of the Aβ42 fibril obtained after three consecutive seeding steps. In the first round of incubation, ~40 μM unlabeled recombinant Aβ42 was incubated without any seeds for ~5 days. In the second and third rounds, 5% (w/w) of seed Aβ42 fibril from the previous round of the incubation was added to a 45–50 μM Aβ42 monomer solution (in the last one with the ¹³C- and ¹⁵N-labeled Aβ42), and we incubated the solution for 3–4 days. The TEM images display bundled fibrils with a diameter of ~10 nm, indicating the presence of a single morphology throughout the sample. Although the resolution is limited, the morphology of the fibrils appears less twisted compared with that observed in our previous studies.^{27, 62} The conformational homogeneity of the sample was further confirmed by SSNMR analysis, as discussed below. First, we examined the conformational homogeneity of the sample by recording the ¹H-detected (H)CCH 3D data, whose ¹³C/¹³C 2D projection is shown in Figure 3a, b. We observed only a single set of C_β-C_α-H_α cross-peaks for each residue type in the spectrum. For example, the amino acid sequence of Aβ42 contained only two Ser residues (Ser-8 and Ser-26), for which we could assign one weak and one strong cross-peak, respectively (green arrows in Figure 3a, b). Similarly, four well-resolved signals were obtained for four Ala residues (Ala-2, Ala-21, Ala-30 and Ala-42) (blue arrows in Figure 3a, b). This finding confirmed that our sample contained only a single structure for the Aβ peptide. The average ¹³C line width was 1.2 ppm; this relatively narrow value for a fibril sample indicates homogeneous structure. We further confirmed the higher mobility of the N-terminal part of the peptide by examining the intensities of the C_β-C_α-H_α cross-peaks for Ala and Ser residues. The Ala-2 and Ser-8 residues showed significantly weaker intensities in the 3D spectrum compared with those of the Ala-21, Ala-30, Ala-42, and Ser-26 residues, reflecting more dynamic nature of the N-terminal part of the Aβ42 peptide. Next, we compared the obtained 2D ¹³C/¹³C projection with the corresponding spectrum simulated from the known ¹³C_α/¹³C_β assignments for the Aβ42 fibril with S-shaped structure (Figure 3c).²⁷ The comparison clearly highlights a poor match between the two spectra, suggesting that the Aβ42 fibril examined in this study is likely to be a new polymorph species, as further discussed below. This finding was rather surprising, because apart from the source of the Aβ peptide (bacterial expressed vs. chemically synthesized), the incubation protocols were almost identical. Although the origin of the differences is unclear, the isolation of a new polymorph species of homogeneous Aβ42 fibril represents a substantial progress.

Figure 2.

TEM images of seeded Aβ42 fibrils obtained after 144 h incubation of third-generation (G₃) fibrils. Scale bars are 20 nm (a) and 10 nm (b).

Figure 3.

(a) ¹³C/¹³C 2D projection of a ¹H-detected (H)CCH 3D spectrum of ¹³C- and ¹⁵N-labeled Aβ42 fibril. (b) Enlarged view of spectral region extracted from (a), showing the signals corresponding to C_β-C_α-H_β correlations, for comparison with the C_β-C_α assignments of Aβ42 fibrils with known S-shaped structure. (c) A simulated spectrum from the ¹³C_β-¹³C_α assignments of the Aβ42 fibril with S-shaped structure.²⁷

Sequential Assignments and Structural Profiling of the New Polymorph

Next, we demonstrate the effectiveness of the ¹H-detected SSNMR approach for signal assignments and structural profiling of the Aβ42 fibril. Previous studies of amyloid fibrils such as HET-s(218–289) by ¹H-detected SSNMR required partially deuterated samples,⁶⁵ which are generally difficult to prepare, for successful signal assignments with enhanced resolution. For the site-specific chemical shift assignment of fully-protonated Aβ42 fibrils, we recorded a combination of ¹H-detected 3D and 4D SSNMR data, in addition to the (H)CCH 3D spectrum discussed above. The ¹³C and ¹H resonances from the (H)CCH 3D experiment allowed us to identify the signals based on the residue type in the amino acid sequence of Aβ42 peptide (see Figure S3 in the Supporting Information). Figure 4a shows the representative strip plots of the (H)CA(CON)CAH (green), (H)CANH (red), and (H)CA(CO)NH (blue) 3D spectra, which illustrate the sequential connectivity of the protein backbone from Asn-27–Met-35. In particular, the (H)CA(CON)CAH 3D data (green spectrum in Figure 4a) allowed us to determine the sequential connectivity between C_α(i-1)-C_α(i) resonance pairs for most of the residues (except for the ones having similar ¹³C_α chemical shifts). Although we initially expected to see only the cross-peaks corresponding to C_α(i-1)-C_α(i)-H_α(i) correlations in the (H)CA(CON)CAH 3D spectrum (marked as black crosses), for most residues we also observed diagonal-peaks (red crosses) corresponding to C_α(i-1)-C_α(i-1)-H_α(i-1) triads via remote N_i-C_α(i-1) magnetization transfer. Thus, most of the 2D (H)CA(CON)CAH strip plots (colored green in Figure 4a) showed two peaks in each strip. Further, Gly-25, Gly-29 and Gly-37 showed cross-peaks at different geminal ¹H_α resonances in the (H)CA(CON)CAH 3D spectrum; thus, two strips are displayed in the figure for Gly-29. Moreover, the Asp-1–Gly-9 residues did not show any cross-peaks in the spectrum, presumably due to the dynamics of the N-terminus of the Aβ42 fibril. On the other hand, when two adjacent residues had close ¹³C_α resonances, the cross- and diagonal-peaks overlapped with each other, making it difficult to separate the exact ¹³C_α chemical shifts. Thus, we also obtained (H)CANH and (H)CA(CO)NH 3D data, which are traditionally used to establish the sequential connectivity via the ¹³C_α resonances with respect to ¹⁵N and ¹H_N chemical shifts.^{51, 66-68} Importantly, this analysis allowed us to determine the ¹³C_α resonances that could not be distinguished in the (H)CA(CON)CAH 3D spectrum, due to the overlap of the corresponding peaks. However, it should be noted that the traditional approach using (H)CANH and (H)CA(CO)NH 3D experiments alone was not sufficient to establish the signal assignments for the β-sheet-rich amyloid protein, which had very limited spectral dispersion in the ¹⁵N/¹H_N 2D spectrum (Figure S2a). Thus, the novel use of ¹H_α-detected (H)CA(CON)CAH 3D data was a crucial element for the successful main-chain assignment on a very small amount of Aβ42 sample. In addition, we carried out a (H)CACONH 4D experiment to assign ¹³CO resonances (Figure S2b). Since the resolution of the (H)CONH 3D spectrum was not sufficient to assign the signals, we decided to record the 4D data instead. In addition, we also acquired a (H)CX(CA)NH 3D spectrum, which correlates the aliphatic ¹³C signals with the ¹⁵N and ¹H_N resonances. These data are useful for confirming the accuracy of the sequential connectivity obtained from the analysis of the (H)CANH and (H)CA(CO)NH 3D experiments, because the pattern of the side-chain ¹³C resonances enables us to assign the peaks with similar ¹³C_α chemical shifts to their amino acid type. However, the signals of some of the side chains were not detected in the (H)CX(CA)NH 3D spectrum. This may be due to the higher degree of mobility of those side chains. Details on the side-chain assignments can be found in the Figure S3.

Figure 4.

(a) Representative strip plots of (H)CA(CON)CAH (green), (H)CANH (red), and (H)CA(CO)NH (blue) 3D spectra, showing the sequential connectivity from Asn-27 to Met-35. Diagonal peaks corresponding to intra-residue magnetization transfer in the (H)CA(CON)CAH 3D spectrum are marked with red crosses. (b) Graphical representation of successful chemical shift assignments for ¹³C (green), ¹⁵N (blue), and ¹H (yellow) resonances in this study. Side-chain ¹H and ¹⁵N are omitted. Gray squares denote unassigned resonances.

Signal assignments based on a set of the 3D data were performed with the semi-automated assignment program MAGRO NMRView.⁷⁰ The ¹³CO assignments were made manually by analyzing the 4D data using the NMRFAM-SPARKY software.⁷¹ Chemical shifts obtained by semi-automated methods were manually validated with the 4D data. The assignments were further validated by HIGHLIGHT REDOR experiments on Val-reverse ¹³C- and ¹⁵N-labeled Aβ42 fibril sample (see Supporting Information).⁵¹ The completeness of the signal assignment for the Aβ42 fibril is summarized in Figure 4b. Without any deuterated samples, we were able to assign most of the backbone resonances (86% of ¹³C_α, 76% of ¹⁵N, 71% of ¹³CO, 85% of ¹H_α, and 76% of ¹H_N) and also the majority of the side-chain ¹³C and ¹H resonances (69% of ¹³C and 54% of ¹H). In the N-terminal residues 1–9, the ¹H_N and ¹⁵N signals were largely missing or weak, presumably reflecting the highly dynamic nature of this region. The average ¹H_N line width was ~0.8 ppm, which is considerably broader than the corresponding line width of 0.2 ppm observed for GB1 microcrystals in the same conditions.⁵¹ Nevertheless, our approach enabled the first successful assignment of both protein backbone and side-chain signals from a trace amount (~200 μg, instead of several milligrams) of Aβ42 fibril.

Based on the assigned chemical shifts of the Aβ42 peptide, torsion angles were calculated using the TALOS-N software (Table S6).⁶⁹ The results indicate that the present Aβ42 fibril is composed of three β-strands spanning the Tyr-10–Asp-23, Asn-27–Vla-36 and Val-39–Ile-41 residues (Figure 5a). Thus, we compared the locations of the β-strands observed in this work with those previously reported for Aβ42 fibril structures (Figure 5b). The present locations of the β-strands are markedly different from those reported in previous studies for S-shaped fibril structures.^27-29 This indicates that the Aβ42 fibril examined here possesses a different conformation, compared with Aβ42 fibrils with S-shaped structure. However, the locations of the β-strands obtained in this work show some similarity to the pattern of the β-strands in the fibrils prepared at low pH by Gremer et al.¹⁴ Therefore, we further compared chemical shifts of our Aβ42 fibril with those in the previous studies, to investigate any conformational similarities.

Figure 5.

(a) Torsion angles of Aβ42 fibril obtained from TALOS-N analysis.⁶⁹ Predicted β-strands are shown as arrows on the top of the figure. (b) Comparison of the locations of the β-strands obtained in this study with those of previously reported Aβ42 fibrils. Red arrows represent β-strands predicted in this study. Green, blue, orange, and purple arrows correspond to the β-strands reported by Xiao et al.,²⁷ Colvin et al.,^{28, 58} Ravotti et al.,^{29, 59} and Gremer et al.,¹⁴ respectively. Curved lines represent non-β-strand regions. Dotted lines represent the regions where there are no predicted secondary structures due to the lack of SSNMR data.

More significant conformational differences emerged when we compared the ¹³C secondary chemical shifts of the present Aβ42 fibril with those reported in the previous studies mentioned above. Figure 6a-d show the difference between the ¹³C_α and ¹³C_β secondary chemical shifts of the Aβ42 fibril in this study and those of the fibrils reported in other studies for the residues Leu-17–Ala-42. The ¹³C secondary chemical shifts obtained here are largely inconsistent with those reported in previous studies, with R² values of (a) 0.29, (b) 0.30, (c) 0.32, and (d) 0.64 denoting poor linear fitting, as shown in Figure 6a-d. For all comparisons, 55%–70% of the residues showed a deviation in the secondary chemical shifts exceeding 1.0 ppm. Similar comparisons of the differences between the ¹³C_α and ¹³C_β secondary chemical shifts of Aβ42 fibrils reported by Xiao et al.,²⁷ Colvin et al.,^{28, 58} and Ravotti et al.^{29, 59} show nearly perfect linear fittings, with R² values of 0.96 and 0.98 for the same residues, as shown in Figure 6e and f, respectively. These three groups reported essentially identical S-shaped structures for Aβ42 fibrils prepared at a neutral pH (pH 7.4–8),^{27-29, 58-59} as shown in the insets of Figure 6a-c. The ¹³C chemical shift of the side-chain carboxyl group at Asp-23 (179.3 ppm in the DSS reference) indicated the non-protonated form,⁷² which is consistent with the neutral pH. These results suggest that the present study involves a conformationally distinct novel polymorph of the Aβ42 fibril, prepared at a physiologically relevant neutral pH. We also confirmed that the ¹³C chemical shifts of the novel polymorph are notably different from the corresponding chemical shifts for any of the four forms of our brain-derived Aβ42 fibrils (Table S2). At this point, the origin of the different structure of the fibril examined in this work remains unclear.

Figure 6.

Comparison of the ¹³C secondary chemical shifts obtained in the present work with previous studies of Aβ42 fibrils. The difference between the ¹³C_α and ¹³C_β secondary chemical shifts of the Aβ42 fibril obtained in this study was plotted against that of the fibrils reported by (a) Xiao et al.,²⁷ (b) Colvin et al.,^{28, 58} (c) Ravotti et al.,^{29, 59} and (d) Gremer et al.¹⁴ The reported structures of the fibrils are shown in the insets; their PDB identifiers are (a) 2MXU , (b) 5KK3, (c) 2NAO and (d) 5OQV. For comparison, the difference between the ¹³C_α and ¹³C_β secondary chemical shifts of the Aβ42 fibril of Xiao et al.²⁷ was plotted against those reported by (e) Colvin et al.^{28, 58} and (f) Ravotti et al.^{29, 59} The corresponding plots show excellent linearity, suggesting that the Aβ42 structures of these studies are very similar to each other.

Conclusion

In this study, we demonstrated for the first time that ¹H-detected SSNMR is an effective tool for characterizing the structural differences between synthetic and brain-derived Aβ42 fibrils for the samples in a range of nano- to pico-moles. Despite the limited sample amounts available, especially in the case of brain-derived Aβ fibrils, we were able to acquire 2D SSNMR spectra with sufficient S/N ratios within a reasonable experimental time (33 sec to 8.7 min), allowing us to identify the main structural differences between these samples. Our SSNMR data clearly indicate that Aβ42 fibrils in AD brain tissues show different structural features from their synthetic counterparts. Four sets of the cross peaks were identified by ¹H-detected 2D–3D SSNMR, suggesting the presence of polymorphs for the brain-derived Aβ42 fibrils. Moreover, the comparison of our SSNMR data for the brain-derived Aβ42 fibrils with those of previously reported Aβ42 and Aβ40 fibrils highlights substantial structural differences.

Furthermore, we report the first (to our knowledge) spectral assignments and structural characterization of fully-protonated Aβ42 fibril using sensitivity-enhanced ¹H-detected SSNMR approach under UFMAS. Owing to the excellent sensitivity achieved by the ¹H-detection, the analysis could be performed using trace amounts of sample, compared with the larger amounts required by conventional ¹³C-detected SSNMR. The morphological and conformational homogeneity of the Aβ42 fibril prepared using a seeding protocol allowed us to analyze the sample using ¹H-detected SSNMR. Since the resolution of the SSNMR spectra acquired under UFMAS is comparable to that of previous studies, a similar approach could be used for the chemical shift assignment. A combination of ¹H-detected 3D and 4D SSNMR spectra was used to perform the assignment of the ¹³C, ¹⁵N, and ¹H chemical shifts of both the protein backbone and side chains. In addition, we used a ¹³C/¹³C correlation 2D spectrum to assign the ¹³CO resonance of Ala-42, and some of the aromatic and carbonyl side-chain ¹³C chemical shifts. In the structured region of the Aβ42 peptide (Tyr-10–Ala-42), our analysis successfully assigned 100% of the ¹³C_α and ¹H_α, 97% of the ¹⁵N and ¹H_N, and 91% of the ¹³CO resonances of the protein backbone, along with 81% of the ¹³C and 59% of the ¹H resonances of the side chains, within an experimental time of 12.7 days and using only ~200 μg (~42 nmol) of the fibril sample.

We also identified the secondary structural elements in the Aβ42 fibril. The torsion angles calculated by the TALOS-N software⁶⁹ predicted three β-strands in Tyr-10–Asp-23, Asn-27–Val-36 and Val-39–Ile-41 residues. The comparison of the locations of the β-strands and ¹³C secondary chemical shifts clearly indicated that the Aβ42 fibril examined in this study is a new polymorph with a different structure from those of previously reported Aβ42 fibril species. The way in which structural differences affect the biological function of the Aβ42 fibril is currently unclear, and further investigations are needed to clarify this aspect. It should be also noted that seeding efficiency of brain-derived amyloid is likely to vary from patients to patients and that the presented data could be a favorable case. Nevertheless, this study demonstrated high propensity of Aβ42 to form multiple forms of fibrils in both in vitro and in vivo (i.e. brain), which was not obvious from the previous studies. We believe that our study can open a new avenue for the analysis of trace amounts of biological systems such as amyloid fibrils, oligomers, and membrane proteins, for which ¹³C-detected SSNMR might be ineffective.

Materials and Methods

Sample Preparation

The protocol for the preparation of a synthetic Aβ42 fibril sample uniformly ¹³C- and ¹⁵N-labeled at Phe-20, Ala-21, Val-24, Gly-25, and Leu-34 residues can be found in ref. 62. The preparation of the brain-derived Aβ42 fibril sample, which was uniformly ¹³C- and ¹⁵N-labeled at the same residues, is described here. Tissue containing Aβ from the temporal lobe of the AD brain was extracted from a patient (87 year old male) diagnosed with AD and cerebral amyloid angiopathy at The University of Chicago, following the protocol described in ref. 73. The brain tissue material (0.029 mg) was suspended in 70 μL of 10 mM phosphate buffer (pH 7.4) at a concentration of 0.42 mg/mL. This tissue-material sample contained ~3.1 ng/μg and ~0.48 ng/μg of Aβ42 and Aβ40, respectively; the results were suggested by quantitative mass spectroscopy of the C-terminal peptides (VGGVV and VGGVVIA) after cleavage of the Aβ sequence beyond Met-35 with CNBr. To truncate the fibrils and obtain more amyloid seeds, the suspension was sonicated on ice at 65% amplitude of 200 W with 40% duty factor in a 20 sec cycle; the cycle was repeated for 10 min in total with a 5 min rest in the middle. The seeds were introduced into a freshly prepared monomeric Aβ42 solution containing 1 mg of Aβ42 at 40.7 μM in a 10 mM phosphate buffer (pH 7.4). The brain-tissue templated Aβ42 was fibrillized under gentle circular agitation for 2 days at room temperature. Despite the small amount of Aβ seed from the tissues (~0.01%) in the sample, the effectiveness of the seeding was confirmed by thioflavin-T fluorescence (see Figure S5 in SI). The brain-tissue templated fibril sample was pelleted and lyophilized following the same protocol in ref. 27, 62. The use of the human-tissue derived sample for this research was approved by the Tokyo Institute of Technology Human Subjects Research Ethics Review Committee (#2019045, #2020061) and RIKEN Research Ethics Committee (H30-16(2)).

The uniformly ¹³C- and ¹⁵N-labeled Aβ42 fibril sample was prepared as follows. The Aβ42-expression construct was prepared as described in ref. 64. Briefly, the genes for Aβ42 were cloned into the pGEX-2T vector (Sigma) at the BamHI and EcoRI sites, using the N-terminal GST tag and factor Xa cleavage site (IEGR) encoded in the vector. The fusion protein GST-IEGR-Aβ42 was expressed in the BL21-CodonPlus (DE3) bacterial strain. The isotopically labeled protein was expressed in the M9 minimal media containing 5 g/L of ¹³C₆-glucose and 1 g/L of ¹⁵N-NH₄Cl, with overnight IPTG induction at 27 °C. Then, the cell pellet was collected and the fusion protein was purified by first lysing the cells in a 25 mM Tris-HCl buffer (pH 8.4) containing 0.1% (v/v) of Triton X-100, 0.05% (w/v) of deoxycholic acid, and sodium salt, using a sonic dismembrator (84 W amplitude, 12 cycles of a 10 sec pulse in each minute). The cell debris containing the inclusion bodies were pelleted at 27,000 rpm for 10 min at 4 °C, and then washed in 25 mM Tris-HCl buffer (pH 8.4) containing 0.2% (w/v) of N-lauroylsarcosine (NLS). Eight additional cycles of sonication were incorporated during the wash step followed by centrifugation as described above. The GST-IEGR-Aβ42 fusion protein was dissolved by vortexing in the 25 mM Tris-HCl buffers (pH 8.4) containing 2% of NLS (NLS amount was in a range between 1% to 5% (w/v)). The vortex and centrifugation steps were repeated for multiple rounds until most of the fusion protein was collected in the supernatant. The supernatants containing the GST-IEGR-Aβ42 fusion protein were pooled together and filtered through a 0.8-μm syringe filter.

The fusion protein was cleaved by bovine factor Xa (30–80 units of enzyme per milligram of fusion protein) in 25 mM Tris-HCl (pH 8.4), 0.2 mM CaCl₂, and 0.35% NLS, for 16 h at 12 °C quiescently. The cleavage mixture was filtered through a 0.22-μm syringe, and then subjected to gel-filtration by a G-25 column in 10 mM Tris base (pH 10), to desalt the NLS molecules. In the eluate, most of the uncleaved fusion protein and the GST-IEGR tag in dimeric forms were trapped in a spin concentrator (50 kDa molecular weight cutoff, Amicon Ultra-15 UFC905096, EMD Millipore) after centrifuging at 4.7×10³ g for 12 min at 4 °C. The flow-through from the centrifugation contained the crude Aβ42 peptide. Subsequent purification and seeded fibrillization were conducted in the same manner as described in ref. 27, 62, except for the injection part for HPLC. The flow-through solution (10 mL) was injected for HPLC purification, and the column was washed at 4 mL/min with mobile phase containing 5% acetonitrile to remove salts for 7 min. Then, we started HPLC purification with acetonitrile gradient, as described in the references. The purity of Aβ42 (> 95%) was verified by MALDI TOF/TOF mass spectroscopy. Briefly, the lyophilized peptide after HPLC purification was weighed and then completely dissolved at 2 mg/mL in an aqueous solution containing 30% acetonitrile (Fisher Scientific) and 0.1% trifluoroacetic acid (TFA; American Bioanalytical) at 4 °C; the solution was subsequently lyophilized again. The lyophilized peptides were stored with drying reagents in a freezer at −20 °C. Before each incubation, the peptide was warmed to room temperature and dissolved in hexafluoroisopropanol (HFIP) (Sigma-Aldrich) at a concentration of ~2 mg/mL; after 1 h, the solution was subsequently lyophilized. This dissolution-lyophilization cycle was repeated twice, according to the previously published protocol. The HFIP-treated peptide was first dissolved in a 10 mM NaOH solution (Fisher Scientific) to 0.6 mM, and then the Aβ solution was diluted to 60 μM at pH 7.4 with a 10 mM phosphate buffer. The fresh Aβ42 peptide solution was filtered by centrifugation with a 50-kDa molecular-mass-cutoff filter (EMD Millipore Amicon Ultra-15 filter, UFC905096) at 4.8 × 10³ g for 3 min in order to remove any undissolved peptide or preformed aggregates. The final Aβ monomer concentration was typically ~40 μM. The uniformly ¹³C- and ¹⁵N-labeled Aβ42 fibril was obtained after three consecutive seeding steps. The first two seeding steps were performed with the unlabeled recombinant Aβ42, and the last one with the ¹³C- and ¹⁵N-labeled recombinant Aβ42. The final fibril sample was pelleted by centrifugation; after the buffer was removed, the pelleted sample was lyophilized following the same protocol in ref. 27, 62.

Val-reverse ¹³C- and ¹⁵N-labeled Aβ42 fibril sample was prepared following a similar protocol used to prepare uniformly ¹³C- and ¹⁵N-labeled Aβ42 fibril with the following additional step; before the IPTG induction, 1.25 mM of unlabeled L-Val was added to the M9 medium.⁵¹

For SSNMR experiments, unless otherwise mentioned, the prepared lyophilized Aβ42 fibril samples (~200 μg) were first packed into 0.75-mm JEOL SSNMR MAS rotors, and then ~0.5 μL of 200 mM Cu-EDTA solution was added for paramagnetic doping and incubated for 3 days. For SSNMR experiments of Val-reverse ¹³C- and ¹⁵N-labeled Aβ42 fibril, for which all the residues except for Val and some scrambled Leu were unformly ¹³C- and ¹⁵N-labeled, a sample was packed into a 1-mm (~500 μg) JEOL SSNMR MAS rotor and doped with 200 mM Cu-EDTA solution (~0.5 μL).

SSNMR Experiments

Unless otherwise mentioned, all the SSNMR experiments in this study were recorded at a MAS rate of 90 kHz on a Bruker Avance III 800 MHz NMR spectrometer at RIKEN NMR Facility using a JEOL 0.75-mm ¹H/¹³C/¹⁵N/²H quad-resonance probe. During the SSNMR experiments, the temperature of the sample was maintained at ~30 °C by the application of the cooling N₂ gas at a flow rate of 400 L/h from a Bruker BCU II unit set to the strong cooling power. SSNMR data for Val-reverse ¹³C- and ¹⁵N-labeled Aβ42 fibril were collected at a MAS rate of 80 kHz on a Bruker Avance III 750 MHz spectrometer at the UIC Center for Structural Biology using a JEOL 1-mm ¹H/¹³C/¹⁵N/²H quad-resonance MAS probe while maintaining the sample temperature at ~30 °C by the application of cooling N₂ gas at −20 °C. Unless otherwise mentioned, all the SSNMR data were recorded under ¹H-decoupling at 10 kHz using WALTZ-16 decoupling during the evolution periods.⁷⁴ To eliminate ¹³C-¹⁵N J-coupling, a ¹³C or ¹⁵N π-pulse was applied at the middle of the ¹⁵N or ¹³C evolution periods, respectively. During the ¹H acquisition, 10 kHz and 5 kHz WALTZ-16 decoupling was used on ¹³C and ¹⁵N channels, respectively. For Val-reverse ¹³C- and ¹⁵N-labeled Aβ42 sample, ¹H-decoupling during t₁ and t₂ evolution was set to 10 kHz using SPINAL-64 scheme. Other experimental parameters used in the SSNMR measurements are summarized in Tables S7-9. All multidimensional SSNMR spectra were processed using the NMRPipe software.⁷⁵ The ¹H and ¹³C/¹⁵N dimensions of the time-domain data were apodized with 45°- and 60°-shifted sine-bell window functions, respectively. All the chemical shifts presented in this work were calibrated based on the DSS standard for ¹H and ¹³C and liquid ammonia for ¹⁵N. Secondary chemical shifts were determined based on the random coil chemical shifts given in ref. 76.

Assignment Protocol

Site-specific chemical shift assignments of the 3D SSNMR data collected on uniformly ¹³C- and ¹⁵N-labeled Aβ42 fibril were mainly performed via the MAGRO-NMRView automated assignment software, which is an upgraded version of Kujira, a package of integrated tools for NMR analysis.⁷⁰ The NMRFAM-SPARKY software⁷¹ was used for assignments of 4D SSNMR (H)CACONH data, which were performed manually to assign ¹³CO resonances. Since the 4D data had a better resolution than (H)CA(CO)NH 3D spectrum, a few resonance assignments from the 3D data that were inconsistent with the 4D data were corrected manually. The chemical shift data for the uniformly ¹³C- and ¹⁵N-labeled Aβ42 fibril were deposited in Biological Magnetic Resonance Bank (BMRB; Entry ID: 26307). Additionally, a comparison of the observed ¹³C_α, ¹⁵N and ¹H_N resonances with a Val-reverse ¹³C- and ¹⁵N-labeled Aβ42 fibril sample was performed to confirm the accuracy of the assignment (see Supporting Information).