Determination of amyloid core structure using chemical shifts

Abstract

Amyloid fibrils are the pathological hallmark of a large variety of neurodegenerative disorders. The structural characterization of amyloid fibrils, however, is challenging due to their non-crystalline, heterogeneous, and often dynamic nature. Thus, the structure of amyloid fibrils of many proteins is still unknown. We here show that the structure calculation program CS-Rosetta can be used to obtain insight into the core structure of amyloid fibrils. Driven by experimental solid-state NMR chemical shifts and taking into account the polymeric nature of fibrils CS-Rosetta allows modeling of the core of amyloid fibrils. Application to the Y145X stop mutant of the human prion protein reveals a left-handed β-helix

Introduction

Protein aggregates are the pathological hallmark of a large variety of neurodegenerative disorders.¹ Increasing evidence suggests that a variety of aggregation intermediates are the primary toxic species, while amyloid fibrils might have protective functions.² At the same time, different fibril morphologies can result in different degrees of toxicity³ and fibril fragmentation can enhance amyloid cytotoxicity.⁴ Thus, detailed knowledge about the structure of amyloid fibrils is required to obtain insight into the toxicity of protein aggregates.

X-ray crystallography has provided unique insights into the structural properties of amyloid fibrils formed by short peptides.⁵ The structural characterization of protein aggregates, however, is challenging due to their non-crystalline, heterogeneous, and often dynamic nature. This made NMR spectroscopy a key method for the study of protein aggregation.6–8 H/D exchange coupled to solution-state NMR spectroscopy has provided residue-specific insight into the backbone hydrogen-bonding of a variety of protein aggregates (e.g. see Refs.^9–13). Solid-state NMR spectroscopy resolved the sequence specific assignment of backbone resonances of several proteins in their amyloid state (for a review see Refs.^6–8). In addition, for some amyloid fibrils intra- and intermolecular contacts were detected by solid-state NMR. So far the structurally best characterized system is the prion protein (PrP) HET-s(218–289), for which a well-defined structure of amyloid fibrils was determined on the basis of a large number of intra- and intermolecular distance restraints.¹⁴ However, in many cases the detection and interpretation of intra- and intermolecular contacts is complicated by internal dynamics and the heterogeneous nature of amyloid fibrils. In addition, the polymeric nature of amyloid fibrils complicates the interpretation of intermolecular and interfilament contacts.

Because of the difficulty to obtain a sufficiently large number of intra- and intermolecular contacts by solid-state NMR spectroscopy, the structure of amyloid fibrils of many proteins is still not known despite the availability of backbone chemical shifts (e.g. see Refs.¹⁵ and¹⁶). Here, we demonstrate that the CS-Rosetta methodology^17–20 can be used to obtain insight into the core structure of amyloid fibrils. Our study shows that CS-Rosetta—driven by experimental solid-state NMR chemical shifts and taking into account the polymeric nature of fibrils—allows modeling of the core of amyloid fibrils.

Results and Discussion

Calculations were performed using the CS-Rosetta chemical shift-based structure calculation program.¹⁷ To reflect the polymeric nature of amyloid fibrils, the CS-Rosetta protocol was modified in the following way: (i) Calculations are performed on multiple (five in the calculations reported below) identical segments that are connected by glycine-serine linkers. (ii) Distance restraints enforce interstrand distances between different layers orthogonal to the fibril axis to 4.8 Å consistent with X-ray diffraction patterns of amyloid fibrils. (iii) Hydrogen-bond distance restraints for parallel in-register alignment are used. (iv) The location of β-strands as identified by solid-state NMR chemical shifts is enforced using Hα(i)–HN(i+1) distance restraints.

The amyloid-specific CS-Rosetta protocol was tested on the PrP HET-s(218–289). Using a large number of experimental intra- and intermolecular distance restraints, the 3D structure of HET-s(218–289) amyloid fibrils was determined.¹⁴ The 3D structure revealed a left-handed β-solenoid with each molecule forming two helical windings (Fig. 1). The two helical windings are formed by residues G225-L250 and its pseudo-repeat T260-F286, respectively. Due to the polymeric nature of amyloid fibrils the core structure of the backbone of the two layers is highly similar.

Figure 1.

Determination of amyloid core structure using chemical shifts. A: Backbone conformation of 9 of the 19 lowest energy CS-Rosetta structures of the fibrillar core of HET-s(218–289). The view is along the fibril axis. B: Lowest energy conformer of the fibrillar core of HET-s(218–289) obtained by CS-Rosetta. Side chains of selected residues are shown (hydrophobic: yellow; positive: blue; negative: red). C: Side view of an 8-mer based on the β-helix motif shown in (B). D, E: Fibril core and side view of the solid-state NMR structure of amyloid fibrils of the PrP HET-s(218–289) (PDB code: 2KJ3). F: Lowest energy CS-Rosetta conformer of residues 111–141 of the Y145X stop mutant of the human PrP. Selected residues are highlighted. [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.]

CS-Rosetta calculations were performed for a molecule containing three segments of G225-L250 interleaved by two segments of T260-F286. G225-L250 and T260-F286 are connected by a linker that remains—according to solid-state NMR and hydrogen/deuterium-exchange coupled to solution NMR—disordered and flexible in the amyloid structure. This HET-s specific linker was replaced by a 12-residue glycine-serine linker. The connection of T260-F286 to G225-L250 of the next molecule was also made by a 12-residue glycine-serine linker. As solid-state NMR and hydrogen/deuterium-exchange data demonstrated that residues ²⁴⁷AATL²⁵⁰ and ²⁸³GKGF²⁸⁶ are also unstructured,^{12, 14} the total length of the linkers between different layers along the fibril axis was 16 residues. CS-Rosetta fragment selection was performed using the previously reported experimental solid-state NMR chemical shifts.^{14, 21} To minimize the contribution of the flexible linkers to structure calculation, the conformation of the glycine and serine residues in the linker was driven toward the disordered state by providing random coil chemical shifts for these residues during CS-Rosetta fragment selection.

Fold assembly was driven by a total of 733 distance restraints enforcing the cross-β structure of HET-s(218–289) fibrils and the location of β-strands indicated by chemical shifts (Table I). Hydrogen bond restraints enforced intermolecular, parallel, in-register β-sheets (Table I).¹⁴ Distance restraints were supplied eight times to CS-Rosetta. Supplying the distance restraints only six times, that is, increasing the weight effectively sixfold, resulted in similar folds for HET-s(218–289) fibrils. Supplying the distance restraints only once, that is, not increasing their weight, did not result in a polymeric arrangement. Changing the distance restraints for interstrand H-bonds to O–HN 2.1 Å/O–N 3.5 Å and for interstrand polymer to 5.0 Å/10.0 Å/20.0 Å resulted in similar lowest energy CS-Rosetta structures for HET-s(218–289).

Table I.

Restraints Used in CS-Rosetta Calculations of HET-s(218–289) Fibrils

Restraints	Type	Upper distance limits	No. of restraints
Chemical shifts	Experimentala		508
	GS-linkerb		168
	Total		676
Distance restraints	Interstrand H-bonds	O–HN 1.8 Å/O–N 2.7 Å	160
	Intra β-strand Hα(i)-HN(i+1)	2.5 Å	80
	Interstrand polymer	4.8 Å/9.6 Å/19.2 Åc	493
	Total		733

^aTaken from Ref.21.

^bRandom coil values for N, CA, C of Gly and N, CA, C, CB for Ser.

^cUpper distance limits for strands i+1, i+2, and i+3 along the fibril axis enforcing the polymeric nature of the fibrils and in agreement with their cross β-structure.

Out of the 100 lowest energy CS-Rosetta structures only those structural models were retained for which the root-mean-square difference in the ϕ/Ψ backbone torsion angles between the angles observed in the model and those predicted by the program TALOS+²² on the basis of the experimental NMR chemical shifts was less than 80°. For the TALOS+ selection only those residues were considered for which the TALOS+ prediction were labeled as good. This final step ensures that the selected models are in agreement with dihedral angles predicted on the basis of the experimental chemical shifts.

Inspection of the 19 lowest energy Rosetta structures of the HET-s amyloid core revealed three β-strands with β-strands 2 and 3 assuming two sides of a triangle. β-strand 1 (residues ²²⁶NSA²²⁸) has a more variable position including conformers in which the triangle is closed as well as ones with more open conformations. In all the 19 lowest energy Rosetta structures, the core is stabilized by hydrophobic interactions between I231, V239, and L241 [Fig. 1(B)]. The CS-Rosetta model with the lowest chemical shift score is shown in more detail in Figure 1(B). The triangular core [Fig. 1(B,C)] is similar to the conformation seen in the solid-state NMR structure (PDB code: 2KJ3) [Fig. 1(D,E)]. In both structures the hydrophobic side chains of residues I231, V239, and L241 point to the interior while polar side chains are on the surface of the fibrillar core. Importantly, no distance restraints except for those reflecting the polymeric nature of amyloid fibrils were used in the CS-Rosetta calculations, whereas the solid-state NMR structure was calculated with a large number of experimental intra- and intermolecular distance restraints.

The CS-Rosetta structure shown in Figure 1(B) only represents the fold and not a high-resolution structure. In applications to globular proteins, CS-Rosetta was demonstrated to result in backbone structures that differ by 1–4 Å from a high-resolution structure depending on the complexity and the size of the protein. In the case of HET-s(218–289) amyloid fibrils, the hydrophobic core of the CS-Rosetta model is less compact when compared to the solid-state NMR structure. In addition, the compactness of this core varies in the 19 lowest energy CS-Rosetta conformers [Fig. 1(A)]. The different degree of compactness is associated with different twists of the three strands and the β-helix [Fig. 1(A,C)]. In line with variable twists in the hydrophobic core, the turn connecting the hydrophobic core to the β-strand formed by residues N243-T246 is less bent [Fig. 1(C,E)]. Keeping the twist of the β-helix small—as suggested by electron microscopy—might further improve modeling of the amyloid core.

Next we applied the CS-Rosetta protocol to amyloid fibrils of the Y145X stop mutant of the human PrP, for which the structure is not known. Misfolding of the natively α-helical PrP into a β-sheet rich isoform is related to various human diseases such as Creutzfeldt–Jakob disease.²³ Currently, little is known about the structure of PrP^Sc.^{24, 25} Instead, a variety of studies have investigated the structure of amyloid fibrils produced in vitro from recombinant PrP.^26–35 Hereditary prion diseases include C-terminally truncated variants of the PrP, Y145X, Q160X, Y226X, and Q227X. Previously, we showed that the β-sheet content is highly similar in amyloid fibrils of the Y145X and Q160X stop mutants of human PrP.³⁶ In addition, solid-state NMR spectroscopy demonstrated that residues 112–140 assume extended β-sheet conformation in a parallel, in register alignment in amyloid fibrils of the Y145X stop mutant.29–31

We subjected the solid-state NMR chemical shifts reported by Helmus et al.²⁹ for amyloid fibrils of the Y145X prion stop mutant to the adapted CS-Rosetta protocol.¹⁷ Calculations were performed on five identical segments that were connected by 16-residue glycine-serine linkers. Distance restraints (Table II) enforced interstrand distances between different layers orthogonal to the fibril axis to 4.8 Å, in agreement with X-ray diffraction patterns of mammalian prions³⁷ and amyloid fibrils of PrP(82–146).³⁸ Hydrogen-bond distance restraints for parallel, in-register alignment—as shown by solid-state NMR spectroscopy of humPrP(23–144)³¹—were used. In addition, the location of β-strands as identified by solid-state NMR chemical shifts (Supporting Information Fig. S1) was enforced using Hα(i)–HN(i+1) distance restraints.

Table II.

Restraints Used in CS-Rosetta Calculations for Residues 111–141 of Amyloid Fibrils of the Y145X Stop Mutant of the Human Prion Protein

Restraints	Type	Upper distance limits	No. of restraints
Chemical shifts	Experimentala		510
	GS-linkerb		224
	Total		734
Distance restraints	Interstrand H-bondsc	O–HN 1.8 Å/O–N 2.7 Å	176
	Intra β-strand Hα(i)–HN(i+1)d	2.5 Å	85
	Interstrand polymer	4.8 Å/9.6 Å/19.2 Åe	714
	Total		975

^aTaken from Ref.29.

^bRandom coil values for N, CA, C of Gly and N, CA, C, CB for Ser.

^cH-bond restraints enforce a parallel, in register alignment in agreement with solid-state NMR data for fibrils of the Y145X stop mutant of the human prion protein.³¹

^dResidues ¹¹²MAGA¹¹⁵, ¹²⁰AVVG¹²³, ¹²⁸YMLGSAMSR¹³⁶, ¹³⁸IIHF¹⁴¹ are in a β-strand conformation according to fragments selected by CS-Rosetta (Supporting Information Fig. S1).

^eUpper distance limits for strands i+1, i+2, and i+3 along the fibril axis enforcing the cross-β structure of the fibrils.

The lowest energy CS-Rosetta amyloid structure of the amyloid core of the Y145X stop mutant is shown in Figure 1(F). It revealed a left-handed β-helix comprising three β-strands that are formed by residues ¹¹¹HMAGA¹¹⁵, ¹²⁰AVVG¹²³, and ¹²⁸YMLGSAMSR¹³⁶. At P137 the third β-strand twists such that residues ¹³⁸IIHF¹⁴¹ extend the β-helix. The palindromic sequence ¹¹³AGAAAAGA¹²⁰ encompasses a loop that connects two β-strands. The fibrillar core is formed by the most hydrophobic part of the human prion sequence from H111 to S135. The β-strand conformation of residues ¹²⁸YMLGSAMSRPIIHF¹⁴¹ shown in Figure 1(F) is highly similar to the structure of amyloid fibrils of humPrP(127–147) as determined by solid-state NMR spectroscopy.³⁹ In addition, electron microscopy of 2D crystals of the 27–30 kDa infectious fragment, PrP27–30, suggested that PrP amyloid fibrils might consist of left-handed β-helices with the core formed by residues 89–140.^{24, 40}

In summary, we show that the CS-Rosetta methodology allows modeling of the core of amyloid fibrils. In line with the application of CS-Rosetta to globular proteins, the calculated conformations are not high-resolution structures. At the same time, the backbone folds can be useful for designing additional experiments and to obtain insights into the mechanism of protein aggregation. Additional experimental information such as spatial proximity between side chains based on mutations and experimental distance restraints obtained from solid-state NMR spectroscopy can be readily included into the calculations. The combination of CS-Rosetta modeling with intra- and intermolecular medium- and long-range distance restraints is expected to improve the accuracy of the structure of amyloid fibrils.

Materials and Methods

CS-Rosetta calculations were performed according to the authors' manual.¹⁷ Note, that chemical shifts and distance restraints (hydrogen bonds, distances) are used in distinct stages of the CS-Rosetta protocol. Chemical shifts are only used during fragment selection and rescoring of CS-Rosetta models, while distance restraints (hydrogen bonds, distances) are only used during the fragment assembly stage of CS-Rosetta (see http://spin.niddk.nih.gov/bax/software/CSROSETTA/). Hydrogen bonds are enforced as distance restraints (Tables I and II) and have the same weight as all other distance restraints during CS-Rosetta fragment assembly. To ensure that the distance restraints are the major driving force during CS-Rosetta fold assembly, they were supplied eight times to CS-Rosetta (see Supporting Information). Thus, the weight of the distance restraints was effectively increased eightfold compared to the intrinsic CS-Rosetta energy terms. Alternatively, the CS-Rosetta distance restraint force constant might be increased by a factor of eight. About 2500 structures were calculated. For further details please see the main text.

Supplementary material

Additional Supporting Information may be found in the online version of this article.