Molecular basis for amyloid-β polymorphism

Abstract

Amyloid-beta (Aβ) aggregates are the main constituent of senile plaques, the histological hallmark of Alzheimer’s disease. Aβ molecules form β-sheet containing structures that assemble into a variety of polymorphic oligomers, protofibers, and fibers that exhibit a range of lifetimes and cellular toxicities. This polymorphic nature of Aβ has frustrated its biophysical characterization, its structural determination, and our understanding of its pathological mechanism. To elucidate Aβ polymorphism in atomic detail, we determined eight new microcrystal structures of fiber-forming segments of Aβ. These structures, all of short, self-complementing pairs of β-sheets termed steric zippers, reveal a variety of modes of self-association of Aβ. Combining these atomic structures with previous NMR studies allows us to propose several fiber models, offering molecular models for some of the repertoire of polydisperse structures accessible to Aβ. These structures and molecular models contribute fundamental information for understanding Aβ polymorphic nature and pathogenesis.

The amyloid hypothesis (1, 2) was based on the observation that amyloid-beta (Aβ), a 39–43 amino acid peptide that forms fibrillar, β-sheet rich structures, is the main constituent of proteinaceous deposits observed in the brains of Alzheimer’s patients (3, 4). Evidence implicating Aβ in the pathogenesis of Alzheimer’s disease includes the appearance of Alzheimer’s symptoms in animal models that express the Aβ peptide (5) and the early onset of the disease coupled with massive depositions of Aβ in patients with the rare Aβ mutation, Asp23-to-Asn (Iowa mutant) (6). In vitro, the Aβ Iowa mutant forms fibers considerably faster than wild type (7). This finding that accelerated fiber formation is correlated to pathology points to fibers as the etiologic agent. In contrast, recent studies point to short fibers and soluble oligomeric forms of Aβ as the more toxic species (8–10). Thus it appears that different assemblies of Aβ are toxic, and they may share common structural features (11–13). Also supporting structural similarity of fibers and oligomers is the observation that many compounds, including analogs of the common amyloid ligands Congo red and thioflavin T, bind to both Aβ oligomers and fibers (14–17).

Despite decades of research, we lack full understanding of the molecular mechanisms of toxicity in Alzheimer’s and other aggregation diseases. Part of the problem is the lack of structural information on the proteins mainly involved in the etiology, Aβ and Tau (3, 18). Atomic structures of oligomers of Aβ have been especially elusive due to their metastability and their heterogeneity in size and shape. More structural information is available for amyloid fibers, such as those associated with Alzheimer’s disease and other aggregation diseases (19–35).

Different amyloid fibers display similar biophysical characteristics (36), most notably their common “cross-β structure” indicated by their X-ray fiber-diffraction patterns, displaying orthogonal reflections at about 4.8- and 10-Å spacings (37–39). The atomic features of the cross-β structure have been clarified by X-ray-derived atomic models of amyloid-like structures, revealing a motif consisting of a pair of tightly mated β-sheets, called a “steric zipper” (20). Steric zippers are formed from short self-complementary sequences and account for amyloid aggregation (20, 40, 41). These short peptide segments form well-ordered fibers (42) and have the biophysical characteristics of the fibers of their parent proteins (43). The structures of microcrystals of over 80 of these amyloid-like segments from different disease-associated proteins have been determined (44–49). These structures help to define cross-β structure, suggesting that stacks of identical short segments form the “cross-β spine” of the protofilament, the basic unit of the mature fiber, whereas the rest of the protein adopts either native-like or unfolded conformation peripheral to the spine (20, 50). Here, we hypothesize that steric zippers not only serve as the spine of the protofilament, but also can mediate the interactions between protofilaments that associate to form mature fibers.

Amyloid-forming proteins, and Aβ in particular, can display a bewildering variety of oligomeric and fiber forms, or polymorphs (27, 29, 31, 51–53). For example, Aβ_1–40 was suggested to form five amyloid structures with distinct β-sheet contents and fiber stabilities (28). Experiment-based models of Aβ_1–40 and Aβ_1–42 described several fiber polymorphs. Solid-state (ss) NMR (ss-NMR) provided models for Aβ_1–40 (21, 29, 31, 34, 35). A model for Aβ_1–42 was derived using hydrogen-bonding constraints from quenched hydrogen/deuterium-exchange NMR, together with information from mutagenesis and previous ss-NMR studies (19). The models suggest that in these particular polymorphs, the Aβ molecule adopts a U-shaped protofilament structure, which hydrogen-bonds with identical molecules to form a pair of in-register, parallel β-sheets. However, the models differ in the precise location of the U-turn in the sequence, as well as in the specific interactions between distal regions, demonstrating that polymorphism is present at the protofilament level. Interestingly, the protofilament structure of Aβ_1–40 fibers seeded from brain plaques was reported to differ from the earlier synthetic Aβ_1–40 structures, with the C-terminal β-sheet flipped in relation to its interface with the N-terminal β-sheet (21, 34, 35). Tycko and coworkers further expanded the range of structures available to Aβ by proposing an antiparallel protofilament structure for Asp23-to-Asn Aβ_1–40 (Iowa mutant) fibers; that is, successive stacked β-strands in each β-sheet run in opposite directions (7). Because all models reported so far were obtained from Aβ material that had been prepared by serial rounds of seeding, they presumably represent only a fraction of all conformations available to Aβ protofilaments. A higher level of polymorphism is manifested in varied modes of association of protofilaments into fibers (31, 34). For example, ss-NMR studies reported models for Aβ_1–40 fibers containing either two or three protofilaments (29). Finally, electron-density maps of Aβ fibers produced by cryoelectron microscopy displayed a variety of fiber forms (32, 51, 54, 55). It is interesting to note that in prion protein fibers, structural motifs other than tightly mated pairs of β-sheets were proposed. For example, an NMR-based model of HET-s(218–289) prion suggests a fiber formed via a β-solenoid triangular hydrophobic core (33).

The variety of polymorphs suggests that multiple interaction sites exist within each Aβ molecule, giving rise to differences in fiber morphologies and physicochemical properties on the surface of the fibers that may be correlated with varying levels of cellular toxicity (29, 31, 35, 56). This variety may provide an explanation for the poor correlation between the extent of amyloid deposition and the severity of neurological symptoms (4, 31, 57). Therefore, to better understand the nature of Aβ polymorphism in atomic detail, we report 8 previously undescribed crystal structures of Aβ segments that, together with three previously determined structures (45), span the sequence range Aβ_16–42. The 11 structures, all of which are steric zippers, reveal multiple modes of homotypic interactions (between identical segments), giving rise to a large variety of possible assemblies of Aβ molecules via different steric zipper spines. Combining our crystal structures of homotypic steric zippers with previous experiment-based models of Aβ, which suggest heterotypic interactions between distal segments in pairs of β-sheets (19, 21, 29, 34, 35), allows us to generate models of Aβ protofilament associations that exemplify the range of possible polymorphs.

Results

Identifying Fiber-Forming Segments in Aβ Using the 3D-Profile Method.

We identified fiber-forming segments of Aβ predicted to form the spines of Aβ fibers (Fig. 1A). For this we used the 3D-profile method that scores six-residue sequence segments for their propensity to form steric zippers, based on the structural profile of a canonical steric zipper with a parallel, face-to-face, β-sheet orientation (for nomenclature, see ref. 45). Generally, the strands are allowed to translate in respect to each other, but the orientation of the strands (parallel vs. antiparallel) remains fixed (41, 59). Several segments within the regions of Aβ_11–25 and Aβ_27–42 were predicted either to self-associate into homotypic steric zippers (Fig. 1A) or to form heterotypic steric zippers in which one of the two β-sheets is composed of one segment and the complementary β-sheet is composed of a second segment (Fig. 1C). These predicted heterotypic interactions correlate with ss-NMR studies of Aβ_1–40 (21, 29) and an experiment-based model of Aβ_1–42 (19). In addition, the predicted heterotypic interactions within the Aβ_27–42 region correlate with a conformation of the Aβ_28–42 segment when fused to the C-terminal region of RNase (60). In this crystal structure, Aβ_28–42 forms a small antiparallel β-sheet with a bend formed by Gly37, yielding heterotypic interaction between residues 30–36 and 38–42, similar to the predictions of the 3D-profile method (Fig. 1C).

Fig. 1.

Amyloidogenic propensity of Aβ homotypic and heterotypic interactions predicted by the 3D-profile method. (A) The 3D-profile method calculates the RosettaDesign energy (58) for the self-association (homotypic interactions) of six amino acid peptide segments (41, 59). The histogram of peptide segments is colored in rainbow from blue to red for segments with low-to-high predicted amyloid propensity. The Aβ amino acid sequence and residue numbering are shown (Top). (B) Aβ segments whose crystal structures have been determined are shown as arrows; blue and purple code for structures presented here or in a previous publication (45), respectively. (C) The 3D-profile method prediction for the association of hetero- and homo-Aβ segments is presented on a 2D-interaction heat map colored as in A. Each element represents the interaction energy of the hypothetical steric zipper of six residues that starts at the residues at the corresponding positions on the axes. Three main cross-peaks predicting high fiber-formation propensity are boxed.

Crystal Structures of Six to Eight Residue Segments of Aβ.

The 3D-profile method predicted several six-residue segments to be amyloidogenic (Fig. 1A). We examined these segments, as well as longer segments, for their ability to form fibers and crystals. Five segments (Aβ_16–21, Aβ_27–32, Aβ_29–34, Aβ_30–35, and Aβ_35–42) formed microcrystals, with some forming more than a single crystal form (Fig. 2). Other segments suffered from low solubility or fast fibrillation, limiting their structural characterization to fiber diffraction and electron microscopy, as described below.

Fig. 2.

Crystal structures of Aβ segments, shown in projection down the fiber axes. The Aβ segments are packed as pairs of interdigitated β-sheets, generally with a dry interface between them, termed steric zippers, forming the basic unit of the fiber (44, 45). The view here looks down the fiber axis, showing only four layers of β-strands in each β-sheet; actual fibers can contain more than 100,000 layers. Each panel is labeled with the amino acid sequence of each segment and the starting and ending residue numbers. Molecules are shown as sticks with noncarbon atoms colored by atom type. In structures with β-sheets composed of parallel strands (D, F, G, and K), the carbons are in white. Antiparallel strands forming β-sheet structures (A–C, E, H–J) are alternately colored with carbons colored white and blue. Closest partners across the dry interface share the same color. Some of the panels are split in two halves; each half represents a different dry interface within the same crystal structure.

Previously, we described eight classes of steric zipper symmetries (45). Together, the steric zippers of Aβ presented here (Table S1) and in a previous publication (45) occupy six of these eight classes, suggesting the variety of possible arrangements of Aβ associations via different spine packings. We find both parallel and antiparallel packing of β-strands within β-sheets, as detailed in the following.

Crystal structures of the Aβ segment.

The Aβ_16–21 segment crystallized in three crystal forms (Fig. 2 A–C), all displaying an antiparallel β-strand stacking of the type “face=back” (45). That is, the β-sheets are equifacial, with identical side chains at the face and back of the β-sheet, a consequence of an internal twofold screw symmetry element. Aβ_16–21 forms I and III (Fig. 2 A and C) display similar interfaces with different conformations of Lys16, and a slight registration slip of the steric zipper interface. Aβ_16–21 form II displays two steric zipper interfaces that differ in rotamer conformation of Phe20 (Fig. 2B). In all zippers, the pairs of β-sheets are packed together via hydrophobic side chains, forming a dry interface. All four steric zippers belong to symmetry class 7 (45).

Corresponding to our structures, ss-NMR characterization of a one-residue longer peptide, Aβ_16–22 (26), as well as a one-residue shorter peptide, Aβ_17–21 (30), showed an antiparallel organization of β-sheets in the fibers. Fibers of longer segments, Aβ_11–25 (30) and Aβ_34–42 (61), also display an antiparallel β-strand orientation. The antiparallel orientation might be associated with pathology seeing that it was observed for a subset of fibers of the “Iowa” Aβ mutant that is related to a familial, early onset, Alzheimer’s disease (6, 7). In addition, Aβ oligomers were also suggested to form antiparallel β-sheet structures (62–65). In contrast, the wild-type, full-length Aβ fibers display a parallel orientation (25).

Crystal structure of the Aβ segment.

The Aβ_27–32 segment forms a parallel β-sheet stacking with two different steric zipper interfaces. Both interfaces show β-sheets packed together via interdigitating hydrophobic side chains, typical of symmetry class 1 (45). One interface shows a “face-to-face” orientation and the other “back-to-back” (Fig. 2D).

Crystal structure of the Aβ segment.

The Aβ_29–34 segment forms an antiparallel β-sheet with a dry steric zipper interface displaying a “face-to-back” orientation (Fig. 2E). The two nonequifacial antiparallel β-sheets are related to each other by a simple translation vector, corresponding to symmetry class 6 (45). The registration between neighboring antiparallel strands is such that the last two residues in each strand fall outside the hydrogen-bonding pattern of the β-sheet. Specifically, Gly33 deviates from β-sheet geometry, placing Leu34 outside the β-sheet. The conformation of Gly33 and Leu34 are different in the two antiparallel strands in the asymmetric unit, which correspond to neighboring strands within the β-sheet.

Crystal structure of the Aβ segment.

The Aβ_30–35 segment forms a parallel β-sheet with a dry steric zipper interface of the type face-to-back, symmetry class 2 (Fig. 2F). This steric zipper interface resembles a “knobs-into-holes” type of packing (66); i.e., Ile32 and Leu34 from one β-sheet form the “knob” that enters the “hole” between Ile31 and Met35 of the mating β-sheet, created by the presence of Gly33 (lacking a side chain).

Crystal structures of the Aβ segment.

The Aβ_35–42 segment crystallized in two crystal forms displaying both parallel (face-to-back orientation, symmetry class 2) (Fig. 2G) and antiparallel (face = back orientation, symmetry class 7) (Fig. 2H) β-sheet stacking. Interestingly, the two steric zippers display a similar interface with minor conformation differences of side chains, and a knobs-into-holes type of packing similar to that described for Aβ_30–35. The knob is formed by Val39 and Ile41 that accommodate the hole formed by the presence of glycine residues.

Previously described structures of Aβ_35–40 and Aβ_37–42 segments (45) are shown in Figure 2 I–K. Aβ_35–40 crystallized in two forms, both displaying antiparallel β-sheets with a face = back orientation, symmetry class 8 (Fig. 2 I–J). Aβ_37–42 forms parallel β-sheets with a face-to-back orientation, symmetry class 4 (Fig. 2K).

Quasicrystalline Fibers of Long (11–20 Residue) Aβ Segments.

Our attempts to crystallize longer segments of Aβ (Aβ_11–25, Aβ_16–35, Aβ_22–35, and Aβ_30–42) resulted in highly disordered microcrystals. The X-ray diffraction patterns (Fig. S1) show a mix of crystalline and fiber diffraction, termed quasicrystalline fiber diffraction (67). These diffraction patterns display a distinguishable feature at reciprocal spacing of 4.8 Å, which is consistent with parallel, in-register, β-sheet structures. A feature at a spacing of 9.6 Å, expected for either antiparallel β-sheets or out-of-register parallel β-sheets, is not present.

Fiber Formation of Aβ Segments Analyzed by Electron Microscopy.

We examined the fiber-forming propensities of Aβ segments, including 6–8 residue segments that form microcrystals (Aβ_16–21, Aβ_30–35, Aβ_35–40, Aβ_35–42, and Aβ_37–42) (Fig. S2A), as well as 11–20 residue segments (Aβ_11–25, Aβ_16–35, Aβ_30–40, and Aβ_30–42) (Fig. S3A). All of the Aβ segments formed fibers. It is noteworthy that Aβ_30–35 forms small microcrystals even under fibrillation conditions, and fibers can grow from the tip of microcrystals (Fig. S2A), suggesting common structural features for fibers and microcrystals (48).

Can Distal Aβ Segments Associate to Form the Spine Structures of Amyloid Fibers?

Identification of fiber-forming segments in Aβ using the 3D-profile method predicted the association of distal segments to form heterotypic steric zippers (Fig. 1C). Based on these predictions, we carried out cocrystallization screens of 1∶1 mixtures of distal peptide segments (Aβ_16–21 + Aβ_30–35, Aβ_16–21 + Aβ_35–40, Aβ_16–21 + Aβ_35–42, Aβ_16–21 + Aβ_37–42, Aβ_11–25 + Aβ_30–40, Aβ_15–25 + Aβ_30–40, Aβ_11–25 + Aβ_30–42, and Aβ_15–25 + Aβ_30–42), but failed to produce diffracting crystals containing two differing peptide segments. Nonetheless, electron micrographs of fibers grown from certain mixtures display a morphology that is distinct from the morphologies of the individual segments (Figs. S2B and S3B).

Discussion

Aβ, as well as several of its peptide segments, readily forms fibers (Figs. S2 and S3) (4). Eleven short segments (6–8 residues) also form microcrystals permitting us to determine their structures in atomic detail (Fig. 2). These structures represent 13 diverse steric zipper interfaces, each of which can serve as the spine for fiber formation (20). In previous work, we termed this phenomenon segmental polymorphism (48).

Types of Amyloid Polymoprhism.

Four steric zipper structures (Aβ_16–21, Aβ_27–32, Aβ_35–40, and Aβ_35–42) show a second type of amyloid polymorphism, termed packing polymorphism, in which the same sequence can form distinct steric zipper structures by virtue of different packing in the spine (48) (Fig. 2). Aβ_35–42 shows a previously undescribed mode of packing polymorphism, with β-sheets stacking via both parallel and antiparallel β-strands (Fig. 2 G and H). Of particular importance, this previously undescribed type of polymorphism may be related to Aβ toxicity. The Aβ Iowa mutation (Asp23-to-Asn) (6) is the determinant for familial, early onset, Alzheimer’s disease. The majority of fibers formed from the mutant Aβ_1–40 suggest an antiparallel orientation (7) and deposit massively compared to wild-type Aβ fibers, which exclusively exhibit a parallel orientation (25). Of interest, the two types of polymorphs (parallel vs. antiparallel) were observed within the same sample of the Iowa mutant Aβ_1–40 (7). This observation can be explained structurally by our crystal structures. Two different polymorphs of Aβ_35–42 showing parallel and antiparallel β-sheet orientation nevertheless show similar interfaces of the two steric zippers with only slightly dissimilar side-chain conformations (Fig. 2 G and H). Also, antiparallel β-sheet structures have been reported for Aβ oligomers (62–65), which recent studies point as more toxic than fibers (8–10).

The steric zipper structures of Aβ segments (Fig. 2), as well as of segments from other disease-related amyloid proteins (44–49), all show homotypic interactions, with the pair of β-sheets formed from the same segment of the protein. Heterotypic interactions, between β-sheets formed from different Aβ segments, were proposed based on NMR studies (19, 21) and the interpretation of cryoelectron microscopy maps (32, 51, 54, 55). Our predictions, based on the 3D-profile method, suggest the association of distal segments to form heterotypic steric zippers (Fig. 1C). Our observations of fiber formation of the different Aβ segments are compatible with this notion, as fibers formed from mixed pairs of Aβ segments display different morphologies compared to fibers formed from individual segments (Figs. S2 and S3). The heterotypic interactions suggest a fourth mode of amyloid polymorphism, heterotypic polymorphism, which is an example of combinatorial polymorphism suggested in previous work (48). With the numerous modes of segmental, packing, and heterotypic polymorphism available for fiber formation, a given Aβ fiber may contain more than a single type of protofilament, each displaying a different kind of polymorphism, as discussed in the following.

Pseudoatomic Aβ Fiber Models.

Using our atomic structures of steric zippers of Aβ segments, combined with models of the protofilament structure (19, 34), we constructed several atomic models of Aβ fibers that exemplify the numerous possibilities for fiber morphology (Fig. 3 and Figs. S4–S6). The protofilament models, namely pairs of tightly mating β-sheets, one of Aβ_1–40 derived from ss-NMR studies (34), and another of an experiment-based model of Aβ_1–42 (19), show a U-shaped structure. In Aβ_1–40, residues 23–29 form a bend in the backbone to bring two β-sheets, composed of residues 10–22 and 30–40, to form a heterotypic interface (34). In Aβ_1–42, the heterotypic interactions are between β-sheets formed by residues 18–26 and 31–42 (19). Few examples of Aβ fiber models are constructed from the steric zipper structures of Aβ_35–42, Aβ_16–21, or Aβ_27–32 mediating the interactions between the two different types of protofilaments (Fig. 3 and Fig. S4). The interprotofilament interface suggested by the model of Fig. S4A differs from those of Fig. 3 and Fig. S4B and involves the pairing of the N-terminal β-sheets for the former and the pairing of the C-terminal β-sheets for the latter. A quaternary model that includes the association of the C-terminal β-sheets was previously suggested by ss-NMR studies (34). The interprotofilament interface in this NMR model covers residues 30–40, which is longer than the interfaces of Fig. 3 and Fig. S4B, which cover residues 35–42 or 27–32, respectively, and thus might represent a more stable polymorph.

Fig. 3.

Models of protofilament associations. The crystal structure of Aβ35-42 Form II was used to model interactions between two protofilaments. The protofilament structure is derived from experiment-based models of Aβ1-40 (residues 1–9 are disordered in the fiber) (34) (Upper) or Aβ1-42 (residues 1–17 are disordered in the fiber) (19) (Lower). See also Figs. S4–S6 for other models of Aβ polymorphs based on the structures of Fig. 2.

Overall, our models (Fig. 3 and Fig. S4) incorporate different segments as the spines, exemplifying segmental polymorphism. They display diverse interfaces within the fiber, incorporating variation within the protofilament structure, as suggested by experiments (19, 34), as well as variation in the interactions between protofilaments composing the mature fiber. Additional fiber models displaying the association of multiple protofilaments via several different core regions illustrate a higher level of segmental polymorphism (Fig. S5).

Our predictions of fiber-forming segments show a cluster of predicted interactions within residues 30–42 of Aβ (Fig. 1C). Structures of segments within this region, of Aβ_30–35 and Aβ_35–42, show steric zippers forming a knobs-into-holes type of packing (66) (Fig. 2 F–H). Correspondingly, we modeled a steric zipper that is longer than those determined by the crystal structures and spans residues 31–42, displaying a similar kind of knobs-into-holes packing between two protofilaments (Fig. S6A). In this longer model, residues Val39 and Ile41 protrude into the void created by Gly33, and Met35 protrudes into the void created by Gly37-Gly38, similar to the structure of Aβ_35–40 (Fig. 2J). The NMR-based quaternary model of Aβ_1–40 (34) displays a similar, but slightly shifted, knobs-into-holes interface, with Ile31 forming the knob that protrudes into the void created by Gly37-Gly38, and Met35 forming the knob that protrudes into the void created by Gly33. Finally, in order to demonstrate polymorphism that is associated with the disease-related Iowa Aβ mutant (6, 7), we constructed a fiber model based on the crystal structure of Aβ_16–21 displaying an antiparallel orientation (Fig. S6B).

Our results offer a molecular basis for amyloid polymorphism. Thirteen different steric zipper interfaces display a variety of polymorphic arrangements (Fig. 2). By combining our crystal structures with previous NMR studies, we offer fiber models that illustrate the structural variety of Aβ assemblies. Polymorphism produces a variety of structures with a variety of cellular toxicities, and a molecular view into the different structures may advance our understanding of the mechanisms of amyloid toxicity.

Materials and Methods

Materials.

Peptide segments (custom synthesis, minimal purity of 98%) were purchased from CS Bio. Chemicals were purchased from Thermo-Fisher and Sigma-Aldrich.

Crystallization Conditions.

All crystals were grown at 18 °C via hanging-drop vapor diffusion. Details of crystallization, structure determination, and refinement are provided in SI Text.

Modeling of Full-Length Aβ Fibers.

Models of Aβ fibers were contracted based on experiment-based models of Aβ_1–40 and Aβ_1–42 protofilaments (19, 34) aligned with the steric zipper interfaces and refined as described in SI Text.

Fiber Formation Assessed by Electron Microscopy.

Samples were prepared as described in SI Text.