Solid State NMR Studies of Amyloid Fibril Structure

Abstract

Current interest in amyloid fibrils stems from their involvement in neurodegenerative and other diseases and from their role as an alternative structural state for many peptides and proteins. Solid state NMR methods have the unique capability of providing detailed structural constraints for amyloid fibrils, sufficient for the development of full molecular models. In this article, recent progress in the application of solid state NMR to fibrils associated with Alzheimer’s disease, prion fibrils, and related systems is reviewed, along with relevant developments in solid state NMR techniques and technology.

I. Introduction

Amyloid fibrils are self-assembled, filamentous protein or peptide aggregates with cross-β molecular structures, meaning that they contain ribbon-like β-sheets in which the β-strands run approximately perpendicular to the long fibril axis and the inter-strand hydrogen bonds run approximately parallel the long fibril axis (1). Figure 1 shows transmission electron microscope (TEM) images of typical amyloid fibrils, including fibrils discussed in detail below, as well as cartoon representations of various possible cross-β structures.

Figure 1.

Negatively-stained TEM images of amyloid fibrils. (a) “Striated ribbon” Aβ_1-40 fibrils. (b) “Twisted pair” Aβ_1-40 fibrils. (c) Brain-seeded Aβ_1-40 fibrils. (d) Amylin fibrils. (e) Sup35NM fibrils. (f) HET-s_218-289 fibrils. White scale bars are 100 nm. (g,h,i) Schematic representations of cross-β structures formed by parallel β-sheets, antiparallel β-sheets, and a β-helix, respectively.

Current interest in the structural, physical, chemical, and biological properties of amyloid fibrils has two independent sources: (i) Amyloid fibrils form in the affected tissue of a class of diseases called “amyloid diseases” (2). These include several that are major public health problems, including Alzheimer’s disease (AD) and related neurodegenerative diseases, type 2 diabetes, Parkinson’s disease, dialysis-related amyloidosis, familial systemic amyloidosis, and Huntington’s disease. In each case, a different protein or peptide forms the fibrils, and the fibrils and/or their precursors are cytotoxic. Prion proteins that cause transmissible spongiform encephalopathies (TSEs) also form amyloid fibrils or related aggregates in affected brain tissue (3). One goal of amyloid research is therefore to contribute to our understanding of the etiology of amyloid diseases and to their prevention and cure; (ii) Apart from their involvement in disease, polypeptide chains have a generic propensity to form amyloid fibrils in aqueous solution (4). In other words, the amyloid fibril is frequently a stable structural state for a polypeptide that competes with its monomeric folded state (when a folded state exists), or with other non-aggregated states, almost irrespective of the amino acid sequence or amino acid composition. Another goal of amyloid fibril research is therefore to elucidate the physical chemical factors that stabilize amyloid structures and affect fibril assembly mechanisms.

Any understanding of the properties and effects of amyloid fibrils depends on knowledge of their molecular structures. Since amyloid fibrils are inherently noncrystalline and insoluble and are assembled from high-molecular-weight units, standard experimental approaches to structure determination provide limited information. Due to the ineffectiveness of other approaches and for a variety of technical reasons, solid state NMR spectroscopy has become the “method of choice” for amyloid structure determination. Technical reasons include the feasibility of preparing isotopically labeled fibril samples in the 1-10 mg quantities typically required for solid state NMR and the fact that fibril samples have high protein or peptide concentrations (25-100%, depending on hydration level), so that small sample volumes and concomitantly high radio-frequency (rf) field strengths and high magic-angle spinning (MAS) frequencies can be employed. Amyloid fibrils also typically exhibit favorable nuclear spin relaxation properties and can have relatively homogeneous structures, resulting in high-quality solid state NMR data with sharp resonance lines. A variety of sophisticated solid state NMR techniques can therefore be applied, leading to experimental constraints on the identities of and backbone conformations in β-strand and non-β-strand segments, the supramolecular organization of β-sheets (e.g., parallel vs. antiparallel structures, and the registry of inter-strand hydrogen bonds), the conformations and inter-residue contacts of amino acid sidechains, and the overall arrangement and symmetry of cross-β structural units. When sufficient constraints are obtained, full molecular structural models can be developed from the solid state NMR data, especially when supplemented by information from electron microscopy. Most of these constraints are not available from other experimental methods.

Earlier reviews describe solid state NMR studies of amyloid fibrils up to 2005, as well as the basic principles of relevant solid state NMR methods (5, 6). Since 2005, various research groups have obtained results for fibrils formed by the β-amyloid peptide (7-10) and tau protein (11) associated with Alzheimer’s disease (AD), the amylin peptide (or islet amyloid polypeptide, IAPP) associated with type 2 diabetes (12-15), insulin (16), the α-synuclein protein associated with Parkinson’s disease (17-21), the mammalian PrP protein associated with TSEs (22-26), fungal prion proteins (27-36), the β₂-microglobulin protein associated with dialysis-related amyloidosis (37, 38), and several other peptides (39-42). A selected set of recent results from solid state NMR is discussed below. This review concludes with a brief discussion of recent advances in solid state NMR methodology that are likely to have an impact on future studies of amyloid fibrils.

II. β-amyloid fibrils and prefibrillar aggregates

A. Fibrils with two-fold and three-fold symmetry

Fibrils of the β-amyloid peptide develop in the brains of AD patients, forming both localized and diffuse deposits (called plaques) in the brain tissue and brain vasculature. The full-length peptide is primarily 40 or 42 residues in length (Aβ_1-40 and Aβ_1-42). Early solid state NMR investigations focused on the β-sheet structures in fibrils formed by fragments of the β-amyloid peptide (43-48). Experiments on Aβ_16-22 fibrils showed that two-dimensional (2D) spectroscopy could be used to resolve and assign ¹³C NMR resonances in samples that contain multiple uniformly ¹³C-labeled residues, and that β-strand segments could be identified by their characteristic ¹³C chemical shift patterns (46). When 2D ¹³C-¹³C NMR spectra of full-length Aβ_1-40 fibrils with multiple uniformly-labeled residues were obtained (49), it became apparent that these spectra contained more than one set of chemical shifts, implying a coexistence of structurally inequivalent Aβ_1-40 molecules within the NMR samples. Structural equivalence could have been a consequence of layering of β-sheets within individual fibrils, as suggested by early structural models (44, 47), which would make the NMR signals from peptides in the inner layers different from those of peptides in the outer layers. However, Petkova et al. subsequently found that variations in chemical shifts were directly correlated with variations in the appearance of the fibrils in TEM images (i.e., fibril morphology), and that both the 2D ¹³C-¹³C NMR spectrum and the fibril morphology were “self-propagating” in seeded fibril growth (50). This work proved that the molecular structure in amyloid fibrils is generally not determined uniquely by the amino acid sequence of the fibril-forming polypeptide. Similar polymorphism has now been documented by solid state NMR for α-synuclein fibrils (51), amylin fibrils (13, 14), and other amyloid fibrils (40, 52).

What are the structural differences between fibrils with different morphologies? The predominant morphology in Aβ_1-40 fibril samples can be affected by subtle changes in growth conditions, specifically the presence or absence of gentle agitation of the peptide solution during fibril formation (50). Agitation leads to fibrils that appear as straight rods, approximately 6 nm in width, which associate laterally to form “striated ribbons” (see Figure 1a). The individual rods can be considered to be the “protofilaments”, i.e., the basic structural units that combine in various numbers to form ribbons of various widths. Quiescent growth leads to fibrils with an apparent modulation in width and which are less prone to lateral association. Although the modulation period in a quiescently grown sample can be heterogeneous, Paravastu et al. found that samples with a uniform appearance and a specific modulation period (120 ± 20 nm) could be prepared by repetitive seeded growth (9), meaning that many generations of fibrils were created, with generation k+1 being grown from seeds (i.e., short fragments prepared by sonication) of generation k. In negatively stained TEM images, the Aβ_1-40 fibrils prepared by Paravastu et al. appear to be “twisted pairs” of protofilaments (see Figure 1b), but in fact they are single protofilaments (i.e., they can not be separated into subunits).

Full structural models for “striated ribbon” and “twisted pair” protofilaments are shown in Figure 2. The two Aβ_1-40 protofilaments share several common features, including: (i) Residues 10-22 and 30-40 form β-strands that are separated by a bend or loop in residues 23-29. Residues 1-9 are disordered. Thus, the molecular conformation resembles a “U” in both types of fibrils; (ii) The β-strand segments form in-register, parallel β-sheets with purely intermolecular hydrogen bonds within the β-sheets. In other words, the backbone amide and carbonyl groups of residue k in one Aβ_1-40 molecule participate in hydrogen bonds with carbonyl and amide groups of residues k-1 and k+1 of a neighboring molecule. Each molecule participates in two β-sheet layers, through its two β-strand segments. This double-layered sheet is considered to be a single cross-β unit; (iii) The contacts between β-sheets within a cross-β unit include contacts between the F19 sidechains of the N-terminal β-strands and the L34 sidechains of the C-terminal β-strands. These contacts are “staggered”, meaning that F19 of one molecule interacts with L34 of neighboring molecules, not with L34 of the same molecule (53).

Figure 2.

Molecular structural models for “striated ribbon” (a) and “twisted pair” (b) Aβ_1-40 fibrils, developed primarily from solid state NMR data with additional constraints from electron microscopy. These models include residues 9-40, with hydrophobic, negatively charged, postively charged (including His), and polar (including Gly) residues colored green, red, blue, and magenta, respectively. Both models are viewed down the long axis of the fibril, with three repeats shown. The full Aβ_1-40 sequence is shown below.

The most striking difference between “striated ribbon” and “twisted pair” protofilaments is the overall symmetry with which cross-β units are arranged. The “striated ribbon” protofilament contains two cross-β units, related by an approximate two-fold rotational symmetry axis that coincides with the long axis of the protofilament. The “twisted pair” protofilament contains three cross-β units, related by an approximate three-fold rotational symmetry axis. The conformation of residues 23-29 is also significantly different in “striated ribbon” and “twisted pair” protofilaments. In particular, “striated ribbon” fibrils contain D23-K28 salt bridge interactions, believed to be important in fibril nucleation (54), that are absent in “twisted pair” protofilaments. The details of sidechain-sidechain contacts between cross-β units (i.e., in the central core of the protofilament) are also apparently different, creating a “steric zipper” interface (55) in “striated ribbon” fibrils, while leaving channels that may accomodate internal water molecules in “twisted pair” fibrils.

Compelling evidence for the existence of distinct Aβ_1-40 fibril structures comprised of either two or three cross-β units comes from experimental measurements of the mass-per-length (MPL) values of individual fibrils with electron microscopy (9, 50, 56). Figure 3 shows MPL data for “striated ribbon” and “twisted pair” fibrils obtained with a simple and accurate dark-field TEM technique introduced by Chen et al. (57) The symmetries discussed above and shown in Figure 2 are consistent with the observation of a single set of ¹³C NMR chemical shifts for most sites in solid state NMR spectra of morphologically homogeneous samples.

Figure 3.

MPL data for Aβ_1-40 fibrils. (a,c) Dark-field TEM images of unstained Aβ_1-40 fibrils with “twisted pair” and “striated ribbon” morphologies, respectively. Arrows indicate tobacco mosaic virus (TMV) rods that serve as intensity calibration standards. Rectangles enclose fibril segments with the indicated MPL values in kDa/nm. White scale bars are 200 nm. (b,d) MPL histograms obtained from multiple dark-field images. Heavy dashed lines indicate the predicted MPL values for the 3-fold symmetric and 2-fold symmetric structural models in Figure 2, which are comprised of three and two cross-β units, respectively. Solid lines are multiple-Gaussian fits. Minor populations comprised of a single cross-β unit appear at MPL ≈ 9 kDa/nm, possibly arising from disruption of the fibril structures during TEM sample preparation or measurement.

Structural models in Figure 2 were developed from sets of experimental constraints that are described in the relevant publications (9, 49, 50, 53). Constraints include nearly complete sets of ¹³C chemical shifts (from which backbone torsion angles can be estimated), quantitative measurements of intermolecular ¹³C-¹³C and ¹⁵N-¹³C dipole-dipole couplings and intramolecular ¹⁵N-¹⁵N dipole-dipole couplings, and semi-quantitative measurements of intermolecular ¹³C-¹³C proximities arising from sidechain-sidechain contacts. The requirements that the fibril structure be of the cross-β type and possess translational symmetry are also very powerful constraints. Together, these constraints are sufficient to define the principal structural features discussed above.

B. Fibrils from human brain tissue

Since different fibril structures may exert different biological effects (50), it would be interesting to apply solid state NMR methods to fibrils that develop in vivo. Direct structural measurements on fibrils in human tissue or on fibrils extracted from human tissue are not possible, both because the quantity of fibrils is too small and because the fibrils are not isotopically labeled. However, experiments on both amyloid and prion fibrils have shown that fibrils with a given structure can be amplified by seeded growth protocols. Typically, the seeds represent less than 10% of the total peptide or protein in the solution, producing amplification factors of 10 or more in each round of seeded growth. Milligram-scale fibril samples with isotopic labeling suitable for solid state NMR can therefore be prepared from microgram-scale fibril seeds, extracted from the relevant tissue.

The first experiments of this type were performed by Paravastu et al. (8), using highly purified amyloid from frontal lobe brain tissue of deceased AD patients. Three rounds of seeded growth produced ~5 mg samples of labeled Aβ_1-40 fibrils. 2D ¹³C-¹³C NMR spectra and other measurements showed that the predominant structure in these samples was different from both of the synthetic fibril structures discussed above and shown in Figure 2. More recently, we have modified the amyloid extraction and seeding protocols so that ~1 mg fibril samples can be generated from partially purified amyloid in a single round of seeding, starting with ~1 g of brain tissue. Figure 4 shows 2D ¹³C-¹³C and ¹⁵N-¹³C NMR spectra of such a sample, seeded with tissue from the occipital lobe of a deceased AD patient. The observation of a single set of sharp crosspeaks in these spectra indicates the presence of a single molecular structure. Again, the ¹³C and ¹⁵N chemical shifts indicate that this structure is significantly different from synthetic fibril structures.

Figure 4.

2D solid state NMR spectra of uniformly ¹⁵N,¹³C-labeled Aβ_1-40 fibrils prepared by seeding with amyloid extracted from occipital lobe tissue of a deceased Alzheimer’s disease patient. Assignments of selected crosspeaks to specific residues are indicated. (a) 2D ¹³C-¹³C NMR spectrum, obtained in a 14.1 T magnetic field with 13.6 kHz MAS, using a 2.94 ms finite-pulse radio-frequency-driven recoupling (fpRFDR) sequence (117) for spin polarization transfer in the mixing period between the two spectral dimensions. (b) 2D ¹⁵N-¹³C spectrum, obtained with frequency-selective ¹⁵N-¹³C cross-polarization followed by fpRFDR in the mixing period.

A prevalent hypotheses in the AD research community is that nonfibrillar β-amyloid aggregates or oligomers, rather than mature amyloid fibrils, may be the primary neurotoxic agents (58, 59). Early support for this hypothesis came from the reported absence of a strong correlation between total amyloid deposition and cognitive impairment in AD patients, although recent data indicate otherwise (60). It is commonly reported that elderly people inevitably develop amyloid plaques (although the quantity of amyloid may vary greatly), while only about 30% of octagenerians develop AD, suggesting that amyloid fibrils alone might not destroy neurons. An alternative hypothesis is that AD results from the neurotoxic action (either direct or indirect) of specific β-amyloid fibril structures, while other structures are relatively benign (8, 50). This hypothesis can be now tested by solid state NMR measurements on fibrils prepared by seeded growth from brain tissue of patients with a variety of medical conditions.

C. Fibrils with anomalous antiparallel β-sheet structures

Evidence for in-register parallel β-sheets in Aβ_1-40 fibrils comes from measurements of intermolecular ¹³C-¹³C dipole-dipole couplings as in Figure 5a (9, 44, 45, 50). Using similar measurements, in-register parallel β-sheets have been identified in Aβ_1-42 fibrils (61), amylin fibrils (14), yeast prion fibrils (32-36), and PrP peptide fibrils (23, 25). An in-register parallel β-sheet naturally aligns hydrophobic residues with themselves, potentially maximizing favorable hydrophobic interactions for any amino acid sequence. An in-register parallel β-sheet also aligns Gln and Asn residues with themselves, maximizing favorable “polar zipper” interactions (62). Antiparallel β-sheets have also been found in amyloid fibrils, but until recently only in fibrils formed by short peptides that contain a single hydrophobic segment in a single β-strand (15, 46, 48, 63, 64).

Figure 5.

Experimental determination of β-sheet organization in wild-type and D23N mutant Aβ_1-40 fibrils with ¹³C and ¹⁵N labels at specific sites. (a) Measurements of intermolecular ¹³C-¹³C dipole-dipole couplings in wild-type fibrils using the PITHIRDS-CT technique (98). Squares, circles, and triangles are data for A30 methyl distances in “striated ribbon” fibrils, A30 methyl distances in “twisted pair” fibrils, and M35 methyl distances in “twisted pair” fibrils. Decay of ¹³C NMR signals on the 30 ms time scale results from ~5 Å intermolecular ¹³C-¹³C distances, indicating an in-register parallel β-sheet structure. (b) Measurements of intermolecular couplings among A21 methyl carbons in D23N-Aβ_1-40 fibrils containing parallel (triangles) and antiparallel (circles and squares) β-sheet structures. (c) Measurements of intermolecular ¹⁵N-¹³C dipole-dipole couplings between the backbone amide nitrogen of L17 and the methyl carbon of A21 in D23N-Aβ_1-40 fibrils containing parallel (triangles) and antiparallel (circles and squares) β-sheet structures, using the REDOR technique (106). Build-up of the normalized REDOR difference signal ΔS/S0 on the 50 ms time scale indicates ~5 Å L17-A21 distances, consistent with an antiparallel β-sheet structure with 17+k ↔ 21-k hydrogen bond registry. (d) Schematic representation of cross-β units in which the two hydrophobic segments of Aβ_1-40 (dark and light green arrows) form antiparallel (top) and parallel (bottom) β-sheets. Favorable hydrophobic interactions can occur in both structures.

Surprisingly, certain fibrils formed in vitro by a disease-associated mutant of Aβ_1-40, in which Asp at residue 23 is replaced by Asn (D23N-Aβ_1-40), have an antiparallel β-sheet structure (7). In vivo, D23N-Aβ_1-40 causes neurodegeneration by cerebral amyloid angiopathy (CAA) (65). The presence of antiparallel β-sheets is indicated by a variety of solid state NMR measurements, including the data in Figures 5b and 5c. Schematic structural models in Figure 5d illustrate how apparently similar hydrophobic interactions can be achieved with either in-register parallel or antiparallel β-sheet structures, provided that the peptide adopts the “U” conformation discussed above. The observation of antiparallel β-sheets in D23N-Aβ_1-40 fibrils suggests that CAA may be caused specifically by fibrils with these anomalous structures. Other disease-associated Aβ_1-40 mutations in residues 21-23 may have similar structural effects (66). Seeded fibril growth using tissue from CAA patients, as discussed above, may permit this possibility to be tested experimentally.

D. Nonfibrillar aggregates

Aggregated species with various morphologies are typically observed during the incubation of fibril-forming polypeptide solutions before the appearance of mature amyloid fibrils. Some of these species (referred to as “amyloid-derived diffusible ligands” (67), “globulomers” (68), or simply oligomers) appear as roughly spherical objects in TEM or atomic force microscope images, with diameters in the 5-50 nm range, while others (called “protofibrils” (69)) appear as worm-like objects, with diameters similar to those of fibrils, but shorter and with greater curvature. In cell cultures, nonfibrillar aggregates can exhibit higher levels of toxicity on a per-molecule basis than mature fibrils (70). Such species have also been shown to occur in human tissue (71, 72). Thus, the hypothesis that nonfibrillar aggregates are the primary causative agents in AD and other amyloid diseases has attracted considerable attention.

The first solid state NMR studies of nonfibrillar aggregates were performed by Ishii and coworkers, who focused on spherical Aβ_1-40 oligomers with demonstrated toxicity and a likely on-pathway role in fibril formation (70, 73). 2D ¹³C-¹³C NMR spectra of these oligomers, which have diameters in the 15-35 nm range and were prepared specifically at 4° C and pH 7.4, are nearly identical to spectra of Aβ_1-40 fibrils, indicating very similar molecular conformations. Measurements of intermolecular ¹³C-¹³C dipole-dipole couplings indicate that the β-strands form parallel β-sheets. Sharpe and coworkers have reported related studies of spherical oligomers formed at pH 4.6 by residues 106-126 of the human prion protein (PrP_106-126) (74). Again, the solid state NMR data indicate a structure that is very similar to the structure of PrP_106-126 fibrils, which form at pH 8.0 (23). In particular, both the oligomers and the fibrils contain in-register parallel β-sheets formed by residues 115-124 (sequence AAAAGAVVGG), with antiparallel stacking of the parallel β-sheets.

The similarity of molecular conformation and supramolecular structure in spherical oligomers and amyloid fibrils is quite surprising, given their qualitatively different morphologies. Future measurements are likely to shed more light on this issue and to provide structural constraints on other aggregated species such as protofibrils.

III. Prion fibrils

A. Mammalian PrP

The mammalian prion protein PrP causes infectious neurodegenerative diseases (TSEs) by structural conversion and aggregation (75). PrP is a 231-residue protein with a ~100-residue N-terminal segment that is disordered in the monomeric state and a ~100-residue C-terminal globular domain that has primarily helical secondary structure (76). The TSE-causing state has not been shown definitively to be a cross-β, amyloid-like state (77), but PrP does form amyloid fibrils (78-80). Moreover, the properties of molecular-level polymorphism and structural self-replication in seeded growth discussed above for amyloid fibrils would account for the existence of self-propagating “strains” or “variants” of TSEs (3, 75). Hydrogen/deuterium exchange and electron paramagnetic resonance data indicate that residues 164-220 (approximately) of PrP form the core of recombinant PrP fibrils, and that the core contains in-register parallel β-sheets (78, 79).

Jaroniec and coworkers (22, 24) have reported studies of fibrils formed by a truncation mutant of PrP (residues 23-144, or PrP_23-144) that produces familial CAA. Signals in 2D solid state NMR spectra of uniformly ¹⁵N,¹³C-labeled samples are exceptionally well resolved at 0° C and above, both because the NMR lines are intrinsically narrow (about 1.0 ppm linewidth for ¹⁵N lines at 11.7 T; about 0.6 ppm linewidth for ¹³C lines) and because only 28 residues are sufficiently immobilized in PrP_23-144 fibrils to contribute to the solid state NMR spectra. From multidimensional spectra, the solid state NMR signals were assigned to residues 112-141 (24). Quantitative measurements of motionally-averaged heteronuclear dipole-dipole couplings indicate that these residues have rapid fluctuations of backbone C_α-H_α bond directions on the order of ±10°, while the N-terminal segment gives rise to NMR signals that are detectable with solution NMR techniques and have chemical shifts similar to random-coil values, indicating that the N-terminal segment is dynamically disordered (22). A structural model for the rigid core of PrP_23-144 fibrils has not yet been reported.

B. Yeast prions

As originally demonstrated by Wickner and coworkers (81, 82), several heritable phenotypes of yeast arise from intracellular aggregation of specific proteins into amyloid states. Such proteins constitute yeast prions. Yeast prions identified to date have Gln- and Asn-rich segments, which are the amyloid-forming segments (or “prion domains”), and which have been shown to be sufficient for prion infectivity and propagation in several cases (83, 84). In a series of solid state NMR studies, amyloid fibrils formed by the prion domains of the Sup35, Ure2p, and Rnq1 prion proteins have been shown to possess in-register parallel β-sheet structures (32-36). These studies depend on measurements of intermolecular ¹³C-¹³C dipole-dipole couplings, as in Figure 5a, but performed on samples that are biosynthetically ¹³C-labeled at backbone carbonyl or sidechain methyl sites of all residues of a specific type (e.g., all Leu or all Tyr residues). The data reflect intermolecular ¹³C-¹³C distances as long as the labeled residues do not occur consecutively in the amino acid sequence. The in-register parallel β-sheet structure of yeast prions is consistent with the primacy of polar zippers among Gln and Asn sidechain amide groups as stabilizing interactions for the amyloid fibrils (62). Interestingly, scrambling the sequence of a yeast prion domain does not prevent prion propagation or amyloid formation (85), and preserves the in-register parallel β-sheet structure (33). This is understandable, as scrambling the sequence does not change its capacity to form polar zippers.

Solid state NMR data for yeast prions discussed above are incompatible with alternative structural proposals, including a β-helical model for Sup35 prions (86) and a model for Ure2p prions in which the globular C-terminal domain (rather than the Gln- and Asn-rich N-terminal domain) is the major structural element in the fibril core (31). The observation of strong, sharp solid state NMR signals from the globular C-terminal domain in full-length Ure2p fibrils by Loquet et al. has been presented as support for the latter model (31), but this observation is also consistent with a fibril structure in which the N-terminal domains form the cross-β core while the C-terminal domains (which are largely immobilized both by tethering to the core and by dimerization) form a shell around the core. Baxa et al. have shown biochemically that the C-terminal domains of a variety of chimeric Ure2p protein fibrils retain their globular structures (87).

C. HET-s and functional amyloid

Several cases have been identified in which amyloid fibrils appear to have specific biological functions (88-90). One such case is that of the HET-s protein, whose aggregation into an amyloid state has been shown to be essential for heterokaryon incompatibility in the filamentous fungus Podospora anserina. Briefly, fungal cells that carry the het-s gene, which encodes HET-s, are incapable of fusing viably with cells that carry the het-S gene, which encodes the homologous protein HET-S, but only if HET-s is in its aggregated state. HET-s is therefore considered to be a fungal prion protein, since its structural state encodes a specific phenotype. Unlike yeast and mammalian prions, whose aggregated states are apparently accidental and deleterious to their hosts, HET-s appears to have evolved to control a biological process that protects het-s cells from viral infections (88).

Meier and coworkers have performed solid state NMR studies on the HET-s prion domain (residues 218-289, or HET-s_218-289), which is not Gln- or Asn-rich (27-30). HET-s_218-289 fibrils prepared in vitro are not polymorphic, consistent with an evolved function and the absence of distinct HET-s prion strains or variants. Initial measurements revealed that the ¹³C and ¹⁵N NMR lines from hydrated HET-s_218-289 fibrils are remarkably sharp (linewidths of 0.5 ppm or less), allowing nearly complete resonance assignments to be obtained with standard 2D and 3D solid state NMR techniques from uniformly ¹⁵N,¹³C-labeled samples (30, 91). Only about 60% of the HET-s_218-289 sequence contributes strong signals to solid state NMR spectra. These signals arise from two segments (residues 226-248 and 262-282) with homologous sequences, suggesting a cross-β structure in which the two segments alternate within a parallel β-sheet, which then contains both intramolecular and intermolecular hydrogen bonds (unlike the in-register parallel β-sheets discussed above). MPL data that indicate a monomer repeat distance of 9.6 Å along the fibril length support this arrangement (57, 92), and conclusive evidence was obtained from measurements of nonsequential, interstrand contacts between H_N and H_α sites using the 2D NHHC solid state NMR technique (28). Additional measurements of nonsequential contacts involving specific sidechains and estimates of backbone torsion angles from ¹³C chemical shifts provided sufficient constraints to develop the model in Figures 6a and 6b. This model closely resembles the β-helix motifs that have been observed in crystal structures of certain monomeric proteins (93), such as the pertactin structure in Figures 6c and 6d. The HET-s_218-289 structure therefore supports earlier suggestions that β-helices might be important structural motifs within amyloid fibrils (94, 95).

Figure 6.

Comparison of the HET-s_218-289 prion fibril structure (a,b) with that of the β-helical protein Bordetella pertussis pertactin (c,d), based on Protein Data Bank files 2RNM and 1DAB, respectively. Structures are viewed perpendicular (a,c) and parallel (b,d) to the long axis of the fibril or β-helix. For HET-s_218-289, five repeats of residues 226-278 are shown, with each repeat forming two “rungs” of the β-helix. For pertactin, residues 1-285 is shown. Hydrophobic, negatively charged, postively charged (including His), and polar (including Gly) residues are colored green, red, blue, and magenta, respectively.

Solid state NMR data from HET-s_218-289 fibrils resemble data from microcrystalline proteins, in that the NMR lines are extraordinarily sharp and the nuclear spin T₂ and T_1ρ relaxation times are long (leading to strong and well-resolved signals in multidimensional spectra). Data from PrP_23-114 fibrils are similar (24). In our experience, the quality of solid state NMR data from amyloid fibrils formed by different polypeptides can be significantly different. Variations in the extent of structural heterogeneity, including heterogeneity in sidechain conformations, sidechain packing, and lateral association (i.e., “bundling” of fibrils), may contribute to variations in linewidths. The time scale and completeness of dynamical averaging of structural heterogeneities probably plays an important role, with sharp solid state NMR lines and long relaxation times occurring only when hetereogeneities are averaged out on sub-microsecond time scales and when the time-averaged structure is homogeneous. For both HET-s_218-289 and PrP_23-114, solid state NMR linewidths have been shown to increase significantly at low temperatures (24, 96), presumably because molecular motions that average out structural inhomogeneities become too slow.

Curli fibrils of Eschirichia coli, which are formed by the CsgA and CsgB proteins (89), have an adhesive function. The CsgA sequence contain five pseudo-repeat segments, each about 22 residues in length. Solid state NMR measurements on uniformly ¹⁵N,¹³C-labeled CsgA fibrils indicate that nearly the entire sequence is immobilized. Measurements of ¹³C-¹³C dipole-dipole couplings in selectively ¹³C-labeled samples, as discussed above, show only weak couplings, arguing against a simple in-register parallel β-sheet structure. MPL data indicate a monomer repeat distance greater than 9 Å (97). Together, these data suggest a β-helical structure for CsgA fibrils, perhaps with the five pseudo-repeats forming successive rungs of the β-helix.

IV. Advances in solid state NMR methodology

A. New approaches to structural constraints

Under MAS, internuclear distances are measured with dipolar recoupling techniques, i.e., rf pulse sequences synchronized with sample rotation in such a way that the desired dipoledipole couplings are no longer averaged to zero (5). In the case of ¹³C-¹³C distances, quantitative distance measurements are possible with established methods when ¹³C labels are introduced at specific sites (98), as in Figures 5a and 5b. When samples are uniformly ¹³C-labeled, strong couplings between directly-bonded ¹³C pairs effectively attenuate the weaker couplings between more distant pairs because of the quantum mechanical non-commutivity of different pairwise couplings, preventing quantitative distance measurements by standard recoupling techniques. This phenomenon is commonly known as “dipolar truncation”. Several approaches to overcoming dipolar truncation have been developed. One approach is to detect long-range couplings by the presence of crosspeaks in 2D ¹³C-¹³C spectra obtained with simple “spin diffusion” or “rf-assisted diffusion” techniques (99, 100), i.e., techniques in which ¹H-¹³C couplings are not removed during the ¹³C-¹³C polarization exchange period, making the ¹³C-¹³C exchange process essentially incoherent. While these approaches allow proximities of ¹³C pairs to be detected at distances of at least 6 Å (53), they do not provide quantitative distance information. Variants in which ¹³C-¹³C exchange proceeds through second-order, three-spin interactions have the advantage of suppressing directly-bonded crosspeaks and enhancing longer-range crosspeaks (101), although dipolar truncation is not eliminated.

A second approach is to devise “frequency selective” recoupling techniques, which take advantage of chemical shift differences to selectively recouple only ¹³C pairs with particular chemical shifts, thereby avoiding quantum mechanical interferences. Although early versions of frequency selective recoupling required relatively low MAS frequencies and large chemical shift differences (102), the more recent SEASHORE (shift evolution assisted homonuclear recoupling) methods are in principle applicable with high-speed MAS and arbitrary chemical shift differences (10, 103). As shown by Hu and Tycko, a zero-quantum SEASHORE technique can be used to measure couplings among backbone carbonyl ¹³C sites (103). When combined with measurements of couplings among backbone amide ¹⁵N sites by non-selective recoupling techniques (9, 98), these data can provide quantitative constraints on backbone conformations in uniformly ¹⁵N,¹³C-labeled peptides and proteins that may improve the precision and accuracy of structural models such as those in Figures 2 and 6.

We have recently proposed a third approach, called stochastic dipolar recoupling, in which the rf pulse sequence intentionally varies from scan to scan in such a way that the recoupled ¹³C-¹³C interactions fluctuate randomly and in a nearly uncorrelated manner. Fluctuations are induced either by random excursions in the rf carrier frequency, in the case of double-quantum stochastic recoupling (104), or random variations in delays between recoupling blocks, in the case of zero-quantum stochastic recoupling (91). In effect, stochastic recoupling mimics the random fluctuations of dipole-dipole interactions that occur in solution NMR due to molecular rotational diffusion. Nuclear spin polarization transfers then become incoherent and can be described by simple kinetic equations, as in the nuclear Overhauser effects observed in solution NMR, rather than by coherent quantum mechanical dynamics (104, 105). In principle, this allows internuclear distances to be extracted from polarization transfer rates in a quantitative manner and without adjustable parameters.

In the case of ¹⁵N-¹³C dipolar recoupling methods, the issue of dipolar truncation does not arise in the same way, since the heteronuclear dipole-dipole couplings created by dipolar recoupling techniques such as REDOR (106) commute with one another. Nonetheless, similar effects occur when couplings are measured from non-selective ¹⁵N-¹³C polarization transfer data. Therefore, frequency-selective ¹⁵N-¹³C recoupling techniques have been developed, and have been shown to furnish quantitative distance constraints in uniformly ¹⁵N,¹³C-labeled proteins (107).

B. Paramagnetic effects

Since the magnetic moments associated with electron spins are ~10³ larger than those associated with nuclear spins, dipolar hyperfine couplings between paramagnetic centers and nuclear spins can provide information about significantly longer distances than can be probed by couplings between nuclear spins. Recent work has demonstrated the feasibility of estimating electron-nucleus distances in proteins in the 10-20 Å range from effects on nuclear spin T₁ or T₂ relaxation times from hyperfine couplings to nitroxide spin labels or paramagnetic metals (108, 109) or from pseudo-contact shifts from hyperfine couplings to paramagnetic metal ions (110).

The general reduction in ¹H T₁ values that results from paramagnetic doping permits more rapid signal averaging, increasing the effective sensitivity of solid state NMR measurements. As shown by Ishii and coworkers (111), when high-speed MAS is employed, ¹H decoupling powers can also be greatly reduced, permitting the use of recycle delays less than 100 ms without excessive sample heating. Data acquisition times can be reduced by factors of 10-100, or sample quantities can be reduced by factors of 3-10.

C. Low-temperature measurements and dynamic nuclear polarization

The sensitivity of solid state NMR measurements can be enhanced by reducing sample temperatures to increase thermal equilibrium nuclear spin polarizations according to the Curie law. We have recently developed an ultra-low-temperature MAS system that allows standard biomolecular solid state NMR techniques to be applied at 25 K, providing sensitivity enhancement factors greater than ten relative to room-temperature measurements (112). Still greater sensitivity enhancements can be achieved by dynamic nuclear polarization (DNP), a phenomenon in which partial saturation of electron spin resonance transitions leads to anomalously large nuclear spin polarizations. Griffin and coworkers have pioneered the development of high-field DNP for biomolecular solid state NMR, using nitroxide dopants and high-power gyrotron microwave sources to generate sensitivity enhancements in the 100-200 range at sample temperatures near 100 K (113, 114). Similar NMR signal amplitudes can be obtained with low microwave powers at lower temperatures (115, 116).

Debelouchina et al. have demonstrated the applicability of DNP to amyloid fibrils, in studies of partially-labeled GNNQQNY peptide fibrils (41). Ultra-low-temperature MAS and DNP technology is likely to have a major impact on future studies, especially of amyloid-forming systems that can be selectively labeled (where the increased MAS NMR linewidths that generally accompany low temperatures are not a serious limitation). A particularly promising application is to studies of oligomers and other intermediate states in amyloid formation, where DNP enhancements should allow solid state NMR measurements at all stages of structural evolution to be performed in frozen solutions of amyloid-forming polypeptides at the 0.01-1.0 mM concentrations that are typically employed in vitro.

Summary Points.

1. Modern solid state NMR data provide unique insights into the molecular structures of amyloid fibrils that are of real biological and biomedical significance. Full structural models can be developed with assistance from electron microscopy.

2. Solid state NMR data show that amyloid fibrils are frequently polymorphic, with structural variations that can include variations in symmetry and in the nature of β-sheet structures.

3. Through seeded growth, structural studies of fibrils that develop in human tissue are now possible. Current data indicate that β-amyloid fibrils in Alzheimer’s disease brain are structurally distinct from fibrils prepared in vitro.

4. In-register parallel β-sheet structures are most common in amyloid and prion fibrils, but antiparallel and β-helical structures also occur.

5. Ongoing developments in solid state NMR techniques and technology promise to improve both sensitivity and structural information content substantially.

Future Issues.

1. In Alzheimer’s disease and other amyloid diseases, does disease severity or progression correlate with molecular structural variations in the associated fibrils?

2. How do molecular structures of oligomeric and protofibrillar intermediates relate to the structures of mature fibrils?

3. Are functional amyloids in general structurally distinct from disease-associated amyloids?

4. How will new solid state NMR methods expand the range of amyloid-forming systems that can be structurally characterized?

Glossary

amyloid fibril

A filamentous protein or peptide aggregate with a cross-β structure

cross-β

A ribbon-like β-sheet structure in which the β-strands run perpendicular to and the backbone hydrogen bonds run parallel to the long axis

prion

A protein with a self-propagating transmissible structural state (usually an amyloid state) that produces a specific phenotype (usually a disease)

dipolar recoupling

Solid state NMR pulse sequence applied in synchrony with MAS to restore nuclear magnetic dipole-dipole couplings

Acronyms

NMR

nuclear magnetic resonance

TEM

transmission electron microscopy

MAS

magic-angle spinning

DNP

dynamic nuclear polarization

Aβ_1-40

40-residue β-amyloid peptide associated with Alzheimer’s disease

AD

Alzheimer’s disease

TSE

transmissible spongiform encephalopathy

CAA

cerebral amyloid angiopathy