The mechanisms of flavonoids inhibiting conformational transition of amyloid-β₄₂ monomer: a comparative molecular dynamics simulation study

Abstract

Flavonoids can bind Aβ₄₂ to inhibit the aggregation of Aβ₄₂ monomer. However, the inhibitory mechanism remains unknown. Herein, comparable molecular dynamics simulations for a total of 710 ns were performed to study its mechanism. The in silico experiments revealed that flavonoids halt the conformational transition of Aβ₄₂ monomer by inhibiting β-sheet formation; the flavonoids push the residues D23 and K28 of Aβ₄₂ to be exposed to solvated water, destroy the salt bridge between D23 and K28, induce the conformational distribution of Aβ₄₂ into local minimization energy conformational state, and generate U-shaped Aβ₄₂ configurations, which have more stable helixes and fewer unstable random coils. Moreover, simulation results from the free energy landscape and binding free energy analyses suggest that biflavonoids are superior to monoflavonoids in inhibiting conformational transition of Aβ₄₂ monomer. These findings agree with the experimental data and may help in the design of new agents that will inhibit the conformational transition of Aβ₄₂ so as to treat Alzheimer's disease.

---

Introduction

Alzheimer's disease (AD) is a progressive, neurodegenerative disorder that is characterized by the progressive, intra-cerebral accumulation of β-amyloid (Aβ) peptides.¹ Aβ, which consists of 39–43 residues, is derived from an amyloid precursor protein (APP) through sequential cleavage by β- and γ-secretases.² Among the variants of Aβ segments, Aβ₄₂ is the predominant component in AD plaques, and displays an enhanced neurotoxicity.^3–5

Increasing evidence indicates that soluble oligomers are the main toxic agents.^6–8 Studies demonstrate that monomeric Aβ₄₂ undergoes spontaneous rearrangement of its initial secondary structure.^9,10 Aβ₄₂ is initially in a random coil or helix conformation, and subsequently transforms into β-sheet conformation. This transformation generates oligomeric and polymeric species, which have a higher content of β-sheet structures. Report suggests that the formation of β-sheets drives the amyloid self-assembly process.¹⁰ Molecular dynamics (MD) simulations demonstrate that the increased stability of the β-sheet conformation favors Aβ fibrillogenesis.^11–14 Therefore, Aβ aggregation inhibitors that prevent β-sheet formation within Aβ are promising agents for treating AD.^6,15

Flavonoids have been reported to have anti-fibrillogenic and cytoprotective properties.^16–20 Biflavonoids are more potent than monoflavonoids in inhibiting Aβ₄₂ aggregation. Circular dichroism (CD) spectroscopy reveals that biflavonoids have a higher activity than monoflavonoids in inhibiting β-sheet formation within Aβ.²¹ However, the molecular recognition mechanism remains unknown.

To data, numerous theoretical models, which were generated by molecular docking, MD simulations, and free energy analysis, have been conducted to characterize the inhibition mechanisms of current Aβ inhibitors at atomic level. For example, Liu and coworkers conducted a series of MD simulations and MMPBSA analyses to probe the molecular mechanism of the inhibition effect of (−)-epigallocatechin-3-gallate (EGCG) on the conformational transition of Aβ₄₂ monomer.²² Their computational results suggested that EGCG inhibited conformational transition of Aβ₄₂ monomer by increasing the content of helical structure and reducing the content of β-sheet structure, and the hydrophobic interactions played key role in the binding of EGCG and Aβ₄₂ monomer. Some similar works were conducted to characterize the molecular interaction mechanisms of NQTrp inhibitor and Aβ_1–28 monomer,²³ FDDNP and ibuprofen and Aβ_10–40 monomer,^24,25 EGCG and NQTrp and Aβ₄₂ dimer,^26,27 and 2002-H20, curcumin, EGCG, NQTrp, and resveratrol and Aβ_17–42 trimers.²⁸ In 2013, Zhu and coworkers adopted a strategy for identifying small-molecule binding sites in the Aβ₄₂ monomer in which MD simulations are combined with fragment-based drug design.²⁹ Based on identified binding pockets, the binding models between Aβ₄₂ and curcumin, and Aβ₄₂ and Congo red were characterized using molecular docking and MD simulations. Compared with Zhu's strategy against Aβ₄₂ monomer, Jiang and coworkers discovered fiber-binding compounds using structure-based virtual screening and bioassays.³⁰ Structure–activity relationship studies of the fiber-binding compounds and their derivatives suggested that compound binding increased fiber stability and decreased fiber toxicity. Based on the above studies, it can be concluded that knowledge of inhibition mechanism will allow us to rationally design highly potent and selective Aβ inhibitors for anti-AD drug development.

The goal of this paper is to articulate the mechanism of the molecular recognition between flavonoids and Aβ₄₂ monomer through molecular dynamics (MD) simulations, and to pave a rational way for designing highly effective and selective Aβ₄₂ inhibitors. The study started with investigating the conformational transition of Aβ₄₂ with different binding partners (apigenin, sumaflavone, 2′,8′′-biapigenin, and taiwaniaflavone, ESI Fig. S1†) to probe the driving force and the pathway of Aβ₄₂ folding. Moreover, essential dynamics and free energy landscape analyses were employed to probe the stability of Aβ₄₂ with different binding partners. Finally, the binding free energies for Aβ₄₂ with monoflavonoids, bioflavonoids, and Aβ₄₂ were measured.

Methods

Molecular docking

The structure of Aβ₄₂ used in the docking calculations was retrieved from the Protein Data Bank (PDB ID: 1IYT, model 1).³¹ Protein atom types and potentials were assigned according to the Amber force field with Kollman-united-atom charges encoded in Sybyl 7.3. The initial structures of apigenin (Ap), sumaflavone (Sum), 2′,8′′-biapigenin (Bia), and taiwaniaflavone (TF) were optimized using the MMFF force field. The Powell method was used for energy minimization via default parameters set in Discovery Studio 2.5.

Autodock 4.0 (ref. 32) was employed to identify the potential binding of mono- and biflavonoids to Aβ₄₂ using a Lamarckian genetic algorithm. A grid map with 80 × 80 × 80 equally spaced points (i.e., spaced at 0.375 Å from one another) was generated using the AutoGrid program to evaluate the binding energies between the ligands and receptors. The dimensions were large enough to cover the whole receptor. Docking parameters are set to default values, except for number of GA runs (100) and the energy evaluations (25 000 000). All docked poses of each ligand were clustered using a tolerance of 2 Å for the root mean square deviation (RMSD) and ranked on the basis of the binding docking energies. For each compound, the lowest energy conformation in the most populated cluster was chosen for further study.

Simulation systems

Initial coordinates for the unbound apo form of the Aβ₄₂ in solution and the four Aβ₄₂ complexes with the active inhibitors (i.e., Ap, Sum, Bia, and TF) were taken from the docking results. Each ligand–receptor complex was first put into a suitable box, of which the minimal distance from the peptide to the box wall was 1.2 nm. Then the box was solvated with the TIP3P water model.³³ Neutralized counter ions were added to the simulation system. Energy was minimized by steepest descent followed by conjugate gradient algorithms to remove steric clashes within each system. The simulation is equilibrated in two steps lasting 200 ps; the first step involves an NVT ensemble and the second phase involves an NPT ensemble. Each system is simulated for 90 ns and trajectories were saved at 2.0 ps intervals for further analysis. Detailed of all the simulations are summarized in the ESI Table S1.† To ensure the simulations are repeatable and not a stochastic output, two independent simulations with different initial velocity distributions were performed for AP–Aβ₄₂ and TF–Aβ₄₂, and the results are documented in the ESI.†

MD simulations

MD simulations were carried out through the GROMACS 4.5.3 package³⁴ with constant number, pressure, and temperature (NPT) and periodic boundary conditions. The AMBER ff03 force field³⁵ was applied for the peptide. Parameters for the four flavonoid ligands were obtained from the ANTECHAMBER module using the Generalized Amber Force Field (GAFF).³⁶ The partial atomic charges for the ligand atoms were assigned using the RESP charge-fitting procedure with input from Hartree–Fock calculations at the 6-31G* level with the Gaussian03 program.³⁷ During the simulations, the pressure and the temperature were coupled to 1 bar with an anisotropic coupling time of 1.0 ps, the temperature was kept at 300 K, and the coupling time was 0.1 ps. Both the pressure and temperature were controlled using Berendsen coupling protocols. The short-range van der Waals interactions were cut off at 1.4 nm. Long-range electrostatic interactions were included on every step using the Particle Mesh Ewald algorithm (PME). All bond lengths were constrained using the LINCS algorithm³⁸ with time steps of 2 fs.

MD simulations were run on a 96-CPU D Dawning TC2600 blade server (Dawning, Guangzhou Computer Center, China). Analyses were performed using the tools encoded in the GROMACS package, VMD,³⁹ and in-house handwritten script. The secondary structures were characterized via the DSSP method.⁴⁰ Structural diagrams were carried out using the Pymol package (http://www.pymol.org/).

Essential dynamics analysis and dynamics similarity metrics

Essential dynamics analysis (also called Principal Component Analysis, PCA)^41,42 were employed to address the collective motions of apo–Aβ₄₂ and ligand–Aβ₄₂ complexes. Briefly, PCA is based on the diagonalization of the covariance matrix C, with the elements defined as follows:

C_ij = <(r_i − 〈r_i〉) × (r_j − 〈r_j〉) (i,j = 1,2,3,…,3N) (1)

where r_i is the Cartesian coordinate of the ith C_α atom, N is the number of C_α atoms considered, and 〈r_i〉 represents the time average over all the configurations obtained in the simulation. By projecting the MD trajectories onto the main essential direction, which corresponds to the largest eigenvector, one can visualize the extreme structures and the major fluctuations of the correlated motions.

The similarities and differences among the essential subspaces spanned by the apo–Aβ₄₂ and ligand–Aβ₄₂ (mono- and biflavonoids) complexes were compared via the rooted mean square inner product (RMSIP).^43–45 The RMSIP is defined as:

where 10 is the number of eigenvectors considered and η_i^a and ν_j^b are the ith and jth eigenvectors of two different sets from different MD simulation systems, respectively. By considering the RMSIP value as a distance parameter, it is possible to provide a quantitative measure of similarities and differences between dynamical subspaces; specifically, a neighbor list was built to describe the degree of similarity between the dynamical spaces of the apo–Aβ₄₂, monoflavonoid–Aβ₄₂, and bioflavonoid–Aβ₄₂ complexes.

Free energy landscape analysis

The energy landscape for the conformational change of a peptide or ligand–peptide complex can be obtained by an appropriate conformational sampling procedure. Conformations generated by MD simulations were used for energy analysis. To obtain a two-dimensional representation of the energy landscape, the energy landscapes were projected onto the first (PC1) and second (PC2) principal components with the highest eigen values calculated from PCA analysis for apo–peptide and ligand–peptide complexes at 300 K. The free energy landscapes are defined as follows:^46–48

G_(PC1,PC2) = −k_BT ln P_(PC1,PC2),

where k_B is the Boltzmann constant, T is the temperature, and P_(PC1,PC2) is the normalized joint probability distribution. The energy surface obtained from the raw data was further smoothed by a Gaussian function, an in-house scripting program, and the graphic views were generated with the Surfer 10 package.

Binding free energy calculation

Each ligand–receptor complex was simulated for an additional 20 ns to calculate the binding free energies for mono-flavonoid–Aβ₄₂ and biflavonoid–Aβ₄₂ complexes, and to provide insight into the interaction energies and energetic stabilities of the complexes. The simulations were conducted with the AMBER 12 program using the AMBER ff03 force field. The MM/PBSA method in the AMBER 12 suite was used to calculate the binding free energies (see the ESI†).

Results and discussion

Initial binding models of flavonoids and Aβ₄₂ from docking

Previously MD simulations indicated that the structural segment of residues 24–40 is crucial for Aβ oligomerization and that four glycine residues (Gly25, Gly29, Gly33 and Gly37) are essential for β-sheet formation.¹² Liu and coworkers suggested that the aggregation rate of Aβ correlated positively with the hydrophobic interactions of Aβ monomer and (−)-epigallocatechin-3-gallate (an Aβ₄₂ inhibitor) from MD simulations and free energy analysis.²² A synthesized Aβ₄₀ peptide analog incorporating 2-hydroxy-4-methoxybenzyl (Hmb) at residues 20, 25, 29, 33, or 37 (to protect the backbone amide) did not form fibrils.⁴⁹ Therefore, small molecules can prevent fibril formation by binding the C-terminus hydrophobic area of Aβ₄₂ monomer.

At the beginning, we docked monoflavonoids (AP) and biflavonoids (Sum, Bia, and TF) into Aβ₄₂ (see ESI Fig. S2a and d†). Both mono- and biflavonoids docked at the C-terminus hydrophobic area of Aβ₄₂ (ESI Fig. S2†). Both mono-flavonoid and bi-flavonoids were surrounded by hydrophobic residues, Leu17, Val18, Phe20, Val24, Lys28, Ile31 and Met35, in vacuum. AP can form two hydrogen bonds with Phe20 and Asn27 (ESI Fig. S2a†); Sum forms one hydrogen bond with Phe20 (ESI Fig. S2b†); Bia forms two hydrogen bonds with Val18 and Lys16 (ESI Fig. S2c†); and TF forms two hydrogen bonds with Val18 and Leu34 (ESI Fig. S2d†). These hydrogen bonds may be favorable for the binding of Aβ₄₂ and flavonoids.

Structural and dynamical analyses of apo–Aβ₄₂ and ligand–Aβ₄₂ complexes

The conformational parameters, including the radius of gyration (R_g) and the end-to-end distance, were monitored as a function of time for all the simulations. The results are depicted in Fig. 1 (parts A and E) and ESI Fig. S3.† All R_g values of apo– and ligand–Aβ₄₂ start with relatively high values and fall to lower values within 45 ns (Bia–Aβ₄₂'s R_g value fell within 5 ns, Fig. 1). After, the R_g values are steady; this suggests that the systems have all reached equilibrium status within 45 ns. The end-to-end distances of Asp1-C_α and Ala42-C_α atoms in the equilibrium systems, along with the time, are depicted in the ESI Fig. S3.† The R_g values of the backbone atoms from apo–Aβ₄₂ decrease from ∼1.7 nm (the NMR structure parameter) to ∼1.3 nm within 40 ns. After, the R_g values stay at 1.3 nm (Fig. 1a). The R_g values for monoflavonoid– and biflavonoid–Aβ₄₂ decrease significantly (i.e., from ∼1.7 nm to ∼1.1 nm for monoflavonoid–Aβ₄₂, and from ∼1.7 nm to ∼1.0 nm for biflavonoid–Aβ₄₂). Thus, both monoflavonoid and biflavonoid–Aβ₄₂ have more compact conformations in water. The compact nature of these conformations is particularly salient in biflavonoid–Aβ₄₂ complexes. The average values of the various conformational parameters obtained in the each simulation during the equilibrium period are listed in Table 1. The R_g values of backbone atoms indicate that the conformations of apo–, mono–, and biflavonoid–Aβ₄₂ complexes are changed in water. The monoflavonoid and biflavonoid–Aβ₄₂ conformations are more compact than that of apo–Aβ₄₂. Therefore, we conclude that mono- and biflavonoids can induce Aβ₄₂ to fold into a more compact structure in water; in contrast, the apo–Aβ₄₂ conformation is not as compact. Biflavonoids (Sum, Bia, and TF) are more capable of inducing compact Aβ₄₂ conformation formations than monoflavonoid (AP).

Fig. 1 The R_g (radius of gyration), as a function of time, in MD simulations. 11 hydrophilic residues: (Asp1, Glu3, Arg5, Asp7, Glu11, Gln15, Lys16, Glu22, Asp23, Asn27, Lys28) and 31 hydrophobic resides were studied. The blue line represents the radius of gyration of the backbone atoms; the red line represents hydrophobic radii of gyration of the 31 hydrophobic residue side-chains (excluding C_α atoms); and the green line represents the hydrophilic radii of gyration of 11 hydrophilic residues side-chains (excluding C_α atoms).

Table 1 Average values of the conformational parameters for Aβ₄₂^a

Complex	Rg (backbone)	Rg (hydrophobic side chain)	Rg (hydrophilic side chain)	End-to-end distance
a All results were obtained from the equilibrium period (45–90 ns) of the simulations, and are shown in nm. Standard errors of the mean are given in parentheses.
Apo–Aβ42	1.31(0.11)	1.31(0.09)	1.11(0.07)	2.09(0.39)
AP–Aβ42	1.14(0.09)	1.06(0.06)	1.34(0.09)	1.53(0.60)
Sum–Aβ42	0.93(0.03)	0.94(0.02)	1.25(0.03)	1.60(0.15)
Bia–Aβ42	1.02(0.03)	1.01(0.02)	1.24(0.03)	1.71(0.18)
TF–Aβ42	0.98(0.01)	1.01(0.02)	1.27(0.03)	1.34(0.21)

---

The R_g values of 31 hydrophobic residues side chains atoms in each simulation exhibit similar fluctuations trend with that of all backbone atoms (Fig. 1a and e) and these average values of R_g from equilibrium are also similar (Table 1). These data suggest that the compact conformation folded by Aβ₄₂ is possibly driven by hydrophobic association. It was reported that soluble Aβ is more compact conformation, which can initially interacts with an existing amyloid deposit to form amyloid fibrils, and the driving force for formation of amyloid would be based upon relieving internal stress and increasing bonding energetics via intermolecular interaction (mainly hydrophobic interaction).^50–52 Unlike the R_g values of the 31 hydrophobic residues, the R_g values of the 11 hydrophilic residues do not correlate with that of the backbone atoms (Fig. 1a and e). Moreover, the binding of monoflavonoids and biflavonoids result in greater hydrophilic R_g, values (crosscurrent for 31 hydrophobic residues) compared with that of apo–Aβ₄₂. These results suggest that the additional driven force (hydrophobic interactions between monoflavonoids and biflavonoids with Aβ₄₂) induces an extended conformation for hydrophilic residues, and induces a compact conformation for hydrophobic residues.

The end-to-end distance (i.e., the distance between Asp1-C_α and Ala42-C_α) decreases significantly from ∼4.6 nm (see the NMR structure) to ∼2.0 nm within 45 ns, and then remains steady (subject to minor fluctuations). As shown in Fig. 3, the N- and C-termini of Aβ₄₂ are closer to each other after folding. However, the end-to-end distances vary with different ligands (i.e., apo, AP, Sum, Bia, and TF) bind at equilibrium (see ESI Fig. S3† and Table 1). Compared with apo–Aβ₄₂, the Aβ₄₂ termini can be closer when Aβ₄₂ binds with flavonoids (Table 1). Initial docking studies demonstrate that the formed hydrophobic interactions between mono- and biflavonoids with Aβ₄₂ play a key role for the binding affinity. The R_g analysis suggests that the compact conformation of Aβ₄₂ is possibly driven by hydrophobic associate. To sum up, the additional driven force from the hydrophobic interaction between monoflavonoids and biflavonoids with Aβ₄₂ may promote the folding degree of Aβ₄₂ compared with apo–Aβ₄₂ in water.

Effects on Aβ₄₂ conformational transition

Conformational transition from α-helix to β-sheet is a crucial early step in Aβ amyloidogenesis.^{12,22,52–57} To characterize the inhibition effect of flavonoids on the conformational transition of Aβ₄₂, the secondary structures of Aβ₄₂ from each simulation were calculated through the DSSP method.⁴⁰ The profiles of Aβ₄₂ secondary structures with different binding partners along each simulation are shown in Fig. 2a and e. Obviously, the initial long α-helix of Aβ₄₂ disappeared after several nanoseconds for apo–Aβ₄₂ in water (Fig. 2a); in particular, the α-helix of residues 26–40 was converted into several short bends connected with turns or coils during the first 9 ns. Afterward, some α-helixes were occasionally converted into 3-helixes and rapidly returned to an α-helix structure within 9 to 27 ns (Fig. 2a). After 27 ns, residues 30–31 and 37–38 formed β-sheets connected with some bends or turns. However, the central hydrophobic region (residues 10–28) always kept its α-helix throughout the whole 90 ns simulation, with occasional local deviations that converted the α-helix to a 5-helix bend (or turn) structure. The residues 1–10 of Aβ₄₂ mostly adopt a random coil or α-helix, with local deviations involving 5-helix or turns. Statistically, Aβ₄₂ consists of ∼20% coil, ∼7% β-sheet, and 56% helix components in water (Table 2); this agrees with previous theoretical studies,^22,53 but is only moderately consistent with NMR observations.^3,7,52,58 The helical content, as derived from our apo–Aβ₄₂ simulation, is slightly higher than that of NMR observations. This high helix content may be due to the Amber force filed and the second structure algorithm. The mixed helix/β-sheet conformations existing in aqueous solution indicate the existence of conformational transition, which is consistent with the experimental results.⁵⁹

Fig. 2 The evolution of Aβ₄₂ secondary structures calculated by DSSP in water (a), with AP (b), with Sum (c), with Bia (d), and with TF (e). The structures were analyzed every 90 ps. The vertical coordinates represent the residue number of Aβ₄₂, and the secondary structure is color-coded (as indicated).

Table 2 The secondary structural component statistics of Aβ₄₂ with different binding partners during simulation^a

Sec-structure	NMR^31	Apo	AP	Sum	Bia	TF
a Statistical standard deviation analysis results of coil, β-sheet, β-bridge, turn, bend, and α-helix components from different simulations systems are 4.33, 3.08, 1.84, 5.50, 2.17, and 3.69, suggesting that the change of coil, β-sheet, turn, and α-helix are significant difference upon different binding partners during MD simulations.b The α-helix is the sum of α-, 5-, and 3-helixes in Fig. 4.
Coil	9.52	19.64	18.19	16.46	9.99	10.97
β-Sheet	0.00	6.86	0.00	0.00	0.00	0.00
β-Bridge	0.00	0.46	0.00	0.00	0.00	4.21
Turn	9.52	9.63	17.82	18.56	21.44	24.29
Bend	9.52	6.78	5.16	7.96	3.07	3.14
α-Helix^b	71.43	56.64	58.83	57.02	65.50	57.39

---

The transitional conformations of the monoflavonoid–Aβ₄₂ and biflavonoid–Aβ₄₂ complexes are significantly different from those of apo–Aβ₄₂ (Fig. 2b and e). The residues 10–26 (i.e., the hydrophobic region) of Aβ₄₂ remain in a helical configuration for 90 ns. This is similar to the case of apo–Aβ₄₂. There are three distinct pathways of conformation transition for apo–Aβ₄₂ and flavonoid–Aβ₄₂ complexes. First, β-sheets are not observed during the 90 ns MD simulations for flavonoid–Aβ₄₂ complexes. This suggests that flavonoids inhibit the formation of β-sheets in Aβ₄₂ in water. This result agrees with the CD experimental data.²¹ Second, residues 30–40 at the C-terminus usually fold into helixes in flavonoid–Aβ₄₂ complexes (Fig. 2). As listed in Table 2, flavonoid–Aβ₄₂ complexes have more helixes than the apo–Aβ₄₂ complex does. Statically more helixes stabilize the Aβ₄₂ conformation, and may decrease the chances of Aβ₄₂ aggregation. Biflavonoid–Aβ₄₂ complexes have fewer random coils than monoflavonoid–Aβ₄₂ and apo–Aβ₄₂ complex have. Experimental data suggest that the high content of random coil structures not only instabilize Aβ conformation, and but also drive the fibrillation of Aβ.^52,58 Our simulations indicated that flavonoids can inhibit the β-sheet formation, which may play a key role for inhibiting Aβ₄₂ aggregation. Moreover, biflavonoids induce more helixes, and reduce random coils in Aβ complexes. Biflavonoids are thus better agents for inhibiting Aβ₄₂ aggregation, which is also consistent with experimental results.²¹

Tarus and coworkers characterized the equilibrium ensemble of the Aβ_1–28 monomer with NQTrp inhibitor by means of extensive atomistic MD simulations.²³ Their results suggested that the population of α-helix was increased by a factor of 2 and the population of β-sheets was reduced by a factor of 1.5 upon NQTrp binding. Lockhart and coworkers suggested that ibuprofen binding to Aβ_10–40 was largely governed by hydrophobic effect, and its binding site in Aβ peptide was entirely composed of hydrophobic amino acids.²⁵ Moreover, conformational ensemble of Aβ monomer in ibuprofen solution revealed two structured regions, 19–25 (R1) and 29–35 (R2), composed of turn/helix and helix structure, respectively. Liu and coworkers simulation results suggested that (−)-epigallocatechin-3-gallate (EGCG) inhibited conformational transition of Aβ₄₂ monomer by increasing the content of helical structure and reducing the content of β-sheet structure.²² Similar to NQTrp, ibuprofen, and EGCG, our simulation results implicates that the helical structure of Aβ₄₂ monomer can be well preserved, and the formation of β-sheet structure is prevented by flavonoids binding.

Flavonoids destroy the D23–K28 salt bridge in Aβ₄₂

The D23 and K28 salt bridge in the Aβ monomer is important to the formation of Aβ fibrils.^7,53,60–62 The salt bridge will be formed when the distance between C_γ at D23 and N_ζ at K28 is close to 0.5 nm. The distributions of the distances between D23 and K28 in different ligand–receptor complexes are depicted in Fig. 3. The distance value peak at 0.3 nm for the apo–Aβ₄₂ complex indicates D23–K28 salt bridge formation. Statistically analysis results show that approximately 6.58% structures may form D23–K28 salt bridge. However, monoflavonoid–Aβ₄₂ and biflavonoid–Aβ₄₂ complexes do not have peaks at 0.3 nm (Fig. 3), suggesting that the binding of between monoflavonoid and biflavonoids with Aβ₄₂ may induce the disappearance of D23–K28 salt bridge.

Fig. 3 The distributions of the distances between Asp23 and Lys28 in complexes apo–Aβ₄₂, monoflavonoid–Aβ₄₂, and biflavonoid–Aβ₄₂ at 300 K. The distances between the anionic and the cationic residues are represented by the distances between the N of the NH₃ group at Lys28 and the C of the carboxylate at Asp23. The structural data were sampled every 100 ps.

Although the D23–K28 salt bridge is unstable, it is crucial for the Aβ aggregation process.⁶⁰ Our MD simulations demonstrate that approximately 6.58% structures may form D23–K28 salt bridge in apo–Aβ₄₂. Tarus and co-workers⁶⁰ suggested that the major reason for the D23–K28 salt bridge's instability is that the hydrophilic side chain of D23 and K28 from Aβ_10–35 can form many hydrogen bonds with solvate water. These hydrogen bonds disrupt the formation probability of D23–K28 salt bridge. As shown in the ESI Fig. S4,† many hydrogen bonds are formed between the side chain of D23 and K28 from Aβ₄₂ and water molecules. The R_g analyses confirm that hydrophobic interactions induced by monoflavonoids and biflavonoids push the D23 and K28 residues to be exposed to solvated water. This keeps D23 and K28 separate from each other, such that they form more hydrogen bonds with water (Fig. 3 and ESI Fig. S4†).

Flavonoids induce conformational distribution of Aβ₄₂ into local minimization energy

We clustered the conformations sampled from MD simulations at 300 K to investigate the differences among apo–Aβ₄₂ and flavonoid–Aβ₄₂ complexes. The relative population distributions of the complex clusters are depicted in Fig. S5.† 2.5 Å is the RMSD threshold for determining whether two conformations belong to the same cluster. Detailed clustering results can be found in the ESI Tables S2 and S3.† The conformations of the flavonoid–Aβ₄₂ complexes are more uniform than that of the apo–Aβ₄₂. For example, the relative populations of the five most popular clusters of the TF–Aβ₄₂ complex are 56.9%, 22.1%, 7.5%, 3.7%, and 2.1%, respectively. In comparison, the relative populations of the five most popular clusters of the apo–Aβ₄₂ are 28.3%, 10.9%, 5.2%, 4.5%, and 2.6%, respectively. The conformations of the apo–Aβ₄₂ are more heterogeneous than that of the monoflavonoid– and biflavonoid–Aβ₄₂ complexes. Therefore, monoflavonoid and biflavonoids can induce Aβ₄₂ to assume more stable conformations.

Fig. 4a and e shows the free energy landscapes projected onto the first two principal components of the apo–Aβ₄₂ and flavonoid–Aβ₄₂ complexes. The free energy landscapes vary for different binding complexes. The size and shape of the minimal energy area (in blue) indicate the stability of a complex. Smaller and more centralized blue areas suggest that the corresponding complex is more stable. As indicated in Fig. 4, the stability order of the complexes is: TF–Aβ₄₂ > Bia–Aβ₄₂ > Sum–Aβ₄₂ > AP–Aβ₄₂ > apo–Aβ₄₂. Thus, flavonoids induce Aβ₄₂ to enter the local energy minimal state. Biflavonoids are better than monoflavonoids. This is consistent with experimental observations.²¹

Fig. 4 Free energy landscapes generated by projecting the principal components, PC1 and PC2, of flavonoid–Aβ₄₂ complexes in MD simulations at 300 K. The free energies are represented by −k_BT ln P_(PC1,PC2) with P_(PC1,PC2) being the distribution probability calculated using the structures sampled at 300 K. The unit of free energy is k_BT, where T is the temperature, and k_B is the Boltzmann constant.

Binding free energies analysis

To further confirm the interactions between flavonoids and Aβ₄₂, we performed additional 20 ns MD simulations on the central representative conformation of the flavonoid–Aβ₄₂ complexes, which was derived from the clustered conformations generated by previous MD simulations. MMPBSA and MMGBSA were applied for the binding free energy calculation. As listed in Table 3, the binding free energies of biflavonoid–Aβ₄₂ complexes (Sum, Bia, and TF) are superior to that of mono–Aβ₄₂ (AP); this is consistent with bioassay results.²¹ The three biflavonoids (Sum, Bia, and TF) had similar binding free energy values (−38.81, −36.67, and −37.73 kcal mol⁻¹, respectively). These values are consistent with the bioassay results (which revealed IC₅₀ values of 4.48, 2.90, and 2.41 μM for Sum, Bia, and TF, respectively, IC₅₀ is defined as the concentration of the compound required to reduce the rate of polymerization).²¹

Table 3 Averaged binding free energies for flavonoid–Aβ complexes^a

Energy terms	PB method				GB method
	AP–Aβ42	Sum–Aβ42	Bia–Aβ42	TF–Aβ42	AP–Aβ42	Sum–Aβ42	Bia–Aβ42	TF–Aβ42
a The energy unit is kcal mol^−1.b Non-bonded van der Waals.c Non-bonded electrostatics.d Polar component to solvation.e Non-polar component to solvation.f Total gas phase energy.g Sum of nonpolar and polar contributions to solvation.h Final estimated binding free energy calculated from the terms above. Standard errors of the mean are given in parentheses. IC50 values of AP, Sum, Bia, and TF^21 are 23.3, 4.48, 2.90, and 2.41 μM, respectively.
ΔEvdw^b	−19.33 (2.78)	−53.99 (3.12)	−50.53 (5.12)	−50.18 (2.52)	−19.33 (2.78)	−53.99 (3.12)	−50.53 (5.12)	−50.18 (2.52)
ΔEele^c	−2.74 (2.15)	−3.98 (4.84)	−13.31 (5.34)	−8.21 (3.97)	−2.74 (2.15)	−3.98 (4.84)	−13.31 (5.34)	−8.21 (3.97)
ΔEele,solv^d	9.94 (6.06)	24.01 (4.48)	36.08 (10.47)	26.26 (3.13)	−10.57 (5.67)	25.60 (4.31)	33.77 (9.29)	26.65 (3.33)
ΔEnonpol,solv^e	−14.76 (1.60)	−32.20 (1.55)	−33.60 (3.56)	−31.36 (1.08)	−2.78 (0.30)	−6.44 (0.23)	−6.59 (0.62)	−5.99 (0.15)
ΔGgas^f	−22.07 (6.72)	−57.97 (5.56)	−63.84 (12.68)	−58.39 (4.50)	−22.07 (2.28)	−57.97 (5.56)	−63.84 (12.68)	−58.39 (4.51)
ΔGsolv^g	−4.82 (3.05)	−8.19 (4.49)	2.48 (1.25)	−5.10 (3.07)	7.80 (5.61)	19.16 (4.28)	27.18 (8.90)	20.66 (3.31)
ΔGbinding^h	−26.90 (3.74)	−66.15 (4.32)	−61.36 (6.56)	−63.48 (3.50)	−14.27 (2.35)	−38.81 (2.83)	−36.67 (4.84)	−37.73 (2.45)

---

Table 3 indicates that both non-bonded van der Waals (ΔE_vdw) and non-bonded electrostatics (ΔE_ele) are favorable for the formation of the binding complex, but ΔE_vdw values are at least 4-fold stronger than ΔE_ele values (∼13 fold for Sum, ∼4 fold for Bia and ∼6 fold for TF). Non-polar component to solvation (ΔE_nonpol,solv) values are also favorable for mono- and biflavonoid bindings. ΔE_ele values are favorable for the formation of the binding complex (e.g., −8.21 kcal mol⁻¹ for TF), while ΔE_ele,solv values do not favor TF–Aβ₄₂ complex formation. Similar observations apply to AP–Aβ₄₂, Sum–Aβ₄₂, and Bia–Aβ₄₂ complexes. Therefore, the intermolecular van der Waals and the nonpolar solvation interactions are the dominant forces involved in stabilizing flavonoid–Aβ complexes.

The inhibition activity of a flavonoid against Aβ₄₂ (IC₅₀) is positively correlated with its A log P value (R² = 0.97, ESI Fig. S6†). This means that flavonoids interact with the hydrophobic residues of Aβ₄₂. This is confirmed by residue-based free energy decomposition analysis (Fig. 5a and d). Generally, if the interaction energy between the residue and the substrate is lower than −1 kcal mol⁻¹, those residues are considered to be important in substrate binding (i.e., they are hot residues).^63,64 As shown in Fig. 6, mono- and biflavonoids have different interaction patterns, which indicate that they have different binding modes (Fig. 6a and d). The hot residues are Leu34, Met35, and Val39 for monoflavonoid binding (AP, Fig. 6a). AP binds the similar U-shaped structure of Aβ₄₂ and are surrounded by Val39, Met35, Leu34, Val39, Leu17, and Gly38 (Fig. 6a, left). Met35 and Val39 form an arene–H stack interaction with the aromatic group of the AP molecule (Fig. 6a, right), and Leu34 forms a hydrophobic association with AP. These interactions occur mainly in the side chains (ESI Fig. S7a†). Biflavonoids form more interactions with Aβ₄₂ residues than monoflavonoids do (Fig. 6b–d). Taking TF as an example, the hot residues are His14, Leu17, Val18, Phe20, Val24, Ala30, Gly33, Leu34, Gly38, and Val39 (Fig. 5d). Previously MD simulations and MMPBSA results also demonstrated that interactions between EGCG and Aβ₄₂ monomer were major from the origin of hot residues (Phe19, Phe20, Leu34, Glu22, Lys28, Gly29, Arg5, and Gly38).²² Five arene–H stack interactions are formed, as well as one hydrogen bond between Gly38 and one hydroxyl of TF (Fig. 6d, left). Similar results can be recognized in Sum– and Bia–Aβ₄₂ complexes (Fig. 5b and c, 6b and c). For biflavonoids, the major ligand–protein interactions also involve the Aβ₄₂ side chain residues (see ESI Fig. S7b–d†). The binding free energy calculation and decomposition results (Table 3 and Fig. 5) are consistent with the binding modes analyses, and agree with the bioassay results.

Fig. 5 Decomposition of binding free energy on a per-residue basis for each protein–inhibitor complex. The unit for energy contribution per residue is kcal mol⁻¹.

Fig. 6 Binding modes for flavonoid–Aβ₄₂ complexes. (a) AP–Aβ₄₂, (b) Sum–Aβ₄₂, (c) Bia–Aβ₄₂, and (d) TF–Aβ₄₂. The coordinates of each complex are derived from the central representation structure in the conformation clustering analysis. The figures on the right side are 2-D ligand–interaction diagrams. Hydrogen bonds are highlighted in red dotted lines.

The hydrophobic interactions of flavonoids with Aβ come from the hydrophobic side-chains of Aβ₄₂ (see ESI Fig. S7a and d†). These hydrophobic interactions stabilize the Aβ₄₂ conformation in water, and induce Aβ folding to generate a U-shaped structure, which results in the hydrophilic residues being pushed away to the solvate phase, thus disrupting the D23–K28 salt bridge and preventing β-sheets being formed (Fig. 7a and d).

Fig. 7 Collective motions obtained by principal component analysis (PCA) on the conformational trajectories of the flavonoid-A complexes derived from the MD simulations. (a) AP–Aβ₄₂, (b) Sum–Aβ₄₂, (c) Bia–Aβ₄₂, and (d) TF–Aβ₄₂. PC1and PC2 accounts for ∼70% of the total movements. The cones and their lengths represent the direction of the protein residue C_α atom and its vibration amplitude, respectively. (e) Quenched hydrogen deuterium-exchange NMR structure of Aβ_17–42 fibrils⁶⁶ (PDB ID: 2BEG). (f) Solid-state NMR structure of Aβ_9–40 fibrils⁶⁷ (PDB ID: 2LMP). A similar U-shaped repeating monomer structure was extracted from the Aβ_17–42 fibrils and Aβ_9–40 fibrils.

As shown in Fig. 7, U-shaped conformations exist in all flavonoid-A complexes. It seems that flavonoid-induced U-shaped conformation blocks Aβ₄₂ aggregation. It should be noted that the U-shaped conformations are not identical across different complexes. However, the value of the rooted mean square inner product (RMSIP) is greater than 0.5 for all the complexes (see ESI Table S3†). This means that the U-shaped conformation is popular in the flavonoid-induced Aβ₄₂ complexes and show similarity dynamics folding space.

The conformations of the transition state and U-shaped state were representative conformations during the MD simulations (see ESI Fig. S8 and S9†). The transition state conformation is the first snapshot of the β-sheet formed in the apo–Aβ₄₂ complex in water. Liu and coworkers discovered a compound (DC-AB1) inhibiting Aβ₄₀ monomer aggregation based on virtual screening of the Aβ₄₀ transition state.⁵⁷ The Aβ₄₂ transition state was also used for structure-based virtual screening in our lab.⁶⁵ Thorsten and coworkers reported a 3D-structural model of the Aβ_17–42 fibrils using quenched hydrogen/deuterium-exchange NMR.⁶⁶ The Aβ_17–42 fibrils consist of a similar U-shaped structure derived from the Aβ_17–42 monomer (Fig. 7e). Another structural model for Aβ_9–40 fibrils was obtained through solid-state NMR and also consists of a similar U-shaped repeating structure derived from the Aβ_9–40 monomer (Fig. 7f).⁶⁷ Therefore, the U-shaped Aβ monomer may be the initial framework for formation of Aβ fibrils (Aβ_1–40 or Aβ_1–42). These U-shaped Aβ monomers consist of a β-sheet–turn–β-sheet motif that lacks a helical secondary structure (Fig. 7e and f).

Conclusion

The molecular recognitions of flavonoids by Aβ₄₂ monomer have been investigated via MD simulations. The study suggests that the activities of flavonoids halting the conformational transition of Aβ₄₂ are attributed to (i) disrupting the formation of the β-sheet structures in Aβ₄₂, (ii) expanding the more stable secondary structure (i.e., the helix) and reducing random coils, (iii) disrupting the D23–K28 salt bridge in Aβ₄₂, and (iv) inducing Aβ₄₂ conformational change into the local energy minimized state.

Binding free energy analyses suggest that hydrophobic interactions are critical for ligand–Aβ₄₂ binding. The hydrophobic interactions lead to (1) the stabilization of an Aβ₄₂ conformation in water, and (2) the inducement of Aβ₄₂ folding, so as to make the D23–K28 salt bridge disappear, and prevent the formation of the β-sheet.

Our study also suggests that compounds with inhibiting activity against Aβ₄₂ monomer also introduce a similar U-shaped Aβ structure with more of helical secondary structures, and which lacks the β-sheet structure. We conclude that if a compound can bind the similar U-shaped structure of the Aβ monomer, it will disrupt the formation of the initial framework (i.e., the β-sheet–turn–β-sheet motif), and inhibit Aβ aggregation. Thus, it would be promising to design Aβ₄₂ inhibitors based on the U-shaped structure of Aβ₄₂.

Acknowledgements

This work was funded in part by China Postdoctoral Science Foundation (No. 2015M572325), the Fundamental Research Funds for the Central Universities (No. 2015ZM049), the Open Project Program of Guangdong Key Laboratory of Fermentation and Enzyme Engineering, SCUT (FJ2015006).

Notes and references

1. G. G. Glenner and C. W. Wong, Biochem. Biophys. Res. Commun., 2012, 425, 534–539.
2. D. J. Selkoe, Nat. Cell Biol., 2004, 6, 1054–1061.
3. M. Ahmed, J. Davis, D. Aucoin, T. Sato, S. Ahuja, S. Aimoto, J. I. Elliott, W. E. Van Nostrand and S. O. Smith, Nat. Struct. Mol. Biol., 2010, 17, 561–U556.
4. T. E. Golde, C. B. Eckman and S. G. Younkin, Biochim. Biophys. Acta, Mol. Basis Dis., 2000, 1502, 172–187.
5. G. W. Small, S. C. Huang, G. Cole, N. Satyamurthy, E. D. Agdeppa, Z. Kiziloglu, V. Kepe, A. Petric, H. V. Vinters, M. E. Phelps and J. R. Barrio, Neurobiol. Aging, 2001, 22, 335.
6. J. Hardy and D. J. Selkoe, Science, 2002, 297, 353–356.
7. T. Luhrs, C. Ritter, M. Adrian, D. Riek-Loher, B. Bohrmann, H. Doeli, D. Schubert and R. Riek, Proc. Natl. Acad. Sci. U. S. A., 2005, 102, 17342–17347.
8. R. Kayed, E. Head, J. L. Thompson, T. M. McIntire, S. C. Milton, C. W. Cotman and C. G. Glabe, Science, 2003, 300, 486–489.
9. P. Nguyen and P. Derreumaux, Acc. Chem. Res., 2014, 47, 603–611.
10. W. Zhuang, N. G. Sgourakis, Z. Y. Li, A. E. Garcia and S. Mukamel, Proc. Natl. Acad. Sci. U. S. A., 2010, 107, 15687–15692.
11. G. H. Wei, A. I. Jewett and J. E. Shea, Phys. Chem. Chem. Phys., 2010, 12, 3622–3629.
12. Y. C. Xu, J. J. Shen, X. M. Luo, W. L. Zhu, K. X. Chen, J. P. Ma and H. L. Jiang, Proc. Natl. Acad. Sci. U. S. A., 2005, 102, 5403–5407.
13. A. Morriss-Andrews, G. Bellesia and J. E. Shea, J. Chem. Phys., 2012, 137, 145104–145109.
14. B. Urbanc, L. Cruz, S. Yun, S. V. Buldyrev, G. Bitan, D. B. Teplow and H. E. Stanley, Proc. Natl. Acad. Sci. U. S. A., 2004, 101, 17345–17350.
15. C. L. Moore and M. S. Wolfe, Expert Opin. Ther. Pat., 1999, 9, 135–146.
16. F. S. Yang, G. P. Lim, A. N. Begum, O. J. Ubeda, M. R. Simmons, S. S. Ambegaokar, P. P. Chen, R. Kayed, C. G. Glabe, S. A. Frautschy and G. M. Cole, J. Biol. Chem., 2005, 280, 5892–5901.
17. K. Ono, H. Naiki and M. Yamada, Curr. Pharm. Des., 2006, 12, 4357–4375.
18. K. Ono, Y. Yoshiike, A. Takashima, K. Hasegawa, H. Naiki and M. Yamada, J. Neurochem., 2003, 87, 172–181.
19. C. W. Lee, H. J. Choi, H. S. Kim, D. H. Kim, I. S. Chang, H. T. Moon, S. Y. Lee, W. K. Oh and E. R. Woo, Bioorg. Med. Chem. Lett., 2008, 16, 732–738.
20. S. S. Kang, J. Y. Lee, Y. K. Choi, S. S. Song, J. S. Kim, S. J. Jeon, Y. N. Han, K. H. Son and B. H. Han, Bioorg. Med. Chem. Lett., 2005, 15, 3588–3591.
21. A. Thapa, E. R. Woo, E. Y. Chi, M. G. Sharoar, H. G. Jin, S. Y. Shin and I. S. Park, Biochemistry, 2011, 50, 2445–2455.
22. F. F. Liu, X. Y. Dong, L. Z. He, A. P. J. Middelberg and Y. Sun, J. Phys. Chem. B, 2011, 115, 11879–11887.
23. B. Tarus, P. H. Nguyen, O. Berthoumieu, P. Faller, A. J. Doig and P. Derreumaux, Eur. J. Med. Chem., 2014, 16, 43–50.
24. C. Lockhart and D. K. Klimov, Biophys. J., 2012, 103, 2341–2351.
25. C. Lockhart, S. Kim and D. K. Klimov, J. Phys. Chem. B, 2012, 116, 12922–12932.
26. T. Zhang, J. Zhang, P. Derreumaux and Y. Mu, J. Phys. Chem. B, 2013, 117, 3993–4002.
27. T. Zhang, W. X. Xu, Y. G. Mu and P. Derreumaux, ACS Chem. Neurosci., 2014, 5, 148–159.
28. Y. Chebaro, P. Jiang, T. Zang, Y. Mu, P. H. Nguyen, N. Mousseau and P. Derreumaux, J. Phys. Chem. B, 2012, 116, 8412–8422.
29. M. Zhu, A. De Simone, D. Schenk, G. Toth, C. M. Dobson and M. Vendruscolo, J. Chem. Phys., 2013, 139, 035101.
30. L. Jiang, C. Liu, D. Leibly, M. Landau, M. L. Zhao, M. P. Hughes and D. S. Eisenberg, eLife, 2013, 2, e00857.
31. O. Crescenzi, S. Tomaselli, R. Guerrini, S. Salvadori, A. M. D'Ursi, P. A. Temussi and D. Picone, Eur. J. Biochem., 2002, 269, 5642–5648.
32. G. M. Morris, R. Huey, W. Lindstrom, M. F. Sanner, R. K. Belew, D. S. Goodsell and A. J. Olson, J. Comput. Chem., 2009, 30, 2785–2791.
33. D. J. Price and C. L. Brooks, J. Chem. Phys., 2004, 121, 10096–10103.
34. M. J. Abraham and J. E. Gready, J. Comput. Chem., 2011, 32, 2031–2040.
35. Y. Duan, C. Wu, S. Chowdhury, M. C. Lee, G. Xiong, W. Zhang, R. Yang, P. Cieplak, R. Luo, T. Lee, J. Caldwell, J. Wang and P. Kollman, J. Comput. Chem., 2003, 24, 1999–2012.
36. W. Zhang, T. J. Hou, X. B. Qiao and X. J. Xu, J. Phys. Chem. B, 2003, 107, 9071–9078.
37. G. W. T. M. J. Frisch, H. B. Schlegel, G. E. Scuseria and M. A. Robb, Gaussian 03, Revision E.01, Gaussian, Inc., Pittsburgh PA, 2004.
38. B. Hess, J. Chem. Theory Comput., 2008, 4, 116–122.
39. W. Humphrey, A. Dalke and K. Schulten, J. Mol. Graphics, 1996, 14, 33–38.
40. W. Kabsch and C. Sander, Biopolymers, 1983, 22, 2577–2637.
41. A. Amadei, A. B. Linssen and H. J. Berendsen, Proteins, 1993, 17, 412–425.
42. M. Stepanova, Phys. Rev. E: Stat., Nonlinear, Soft Matter Phys., 2007, 76, 051918.
43. A. Amadei, M. A. Ceruso and A. Di Nola, Proteins, 1999, 36, 419–424.
44. A. Merlino, L. Vitagliano, M. A. Ceruso, A. Di Nola and L. Mazzarella, Biopolymers, 2002, 65, 274–283.
45. A. Merlino, L. Vitagliano, M. A. Ceruso and L. Mazzarella, Biophys. J., 2004, 86, 2383–2391.
46. E. Papaleo, P. Mereghetti, P. Fantucci, R. Grandori and L. De Gioia, J. Mol. Graphics Modell., 2009, 27, 889–899.
47. G. Colombo, G. Morra, M. Meli and G. Verkhivker, Proc. Natl. Acad. Sci. U. S. A., 2008, 105, 7976–7981.
48. W. Li, J. Zhang, Y. Su, J. Wang, M. Qin and W. Wang, J. Phys. Chem. B, 2007, 111, 13814–13821.
49. A. B. Clippingdale, M. Macris, J. D. Wade and C. J. Barrow, J. Pept. Res., 1999, 53, 665–672.
50. J. Talafous, K. J. Marcinowski, G. Klopman and M. G. Zagorski, Biochemistr, 1994, 33, 7788–7796.
51. S. Zhang, N. Casey and J. P. Lee, Folding Des., 1998, 3, 413–422.
52. K. I. S. Zhang, M. J. Lachenmann, J. W. Peng, S. Li, E. R Stimson, Y. Lu, A. M. Felix, J. E. Maggio and J. P. Lee, J. Struct. Biol., 2000, 130, 130–141.
53. C. Yang, X. Zhu, J. Li and R. Shi, J. Mol. Model., 2010, 16, 813–821.
54. L. O. Tjernberg, A. Tjernberg, N. Bark, Y. Shi, B. P. Ruzsicska, Z. Bu, J. Thyberg and D. J. Callaway, Biochem. J., 2002, 366, 343–351.
55. M. Coles, W. Bicknell, A. A. Watson, D. P. Fairlie and D. J. Craik, Biochemistry, 1998, 37, 11064–11077.
56. W. B. Stine Jr, K. N. Dahlgren, G. A. Krafft and M. J. LaDu, J. Biol. Chem., 2003, 278, 11612–11622.
57. D. Liu, Y. Xu, Y. Feng, H. Liu, X. Shen, K. Chen, J. Ma and H. Jiang, Biochemistry, 2006, 45, 10963–10972.
58. G. Bitan, M. D. Kirkitadze, A. Lomakin, S. S. Vollers, G. B. Benedek and D. B. Teplow, Proc. Natl. Acad. Sci. U. S. A., 2003, 100, 330–335.
59. M. D. Kirkitadze, M. M. Condron and D. B. Teplow, J. Mol. Biol., 2001, 312, 1103–1119.
60. B. Tarus, J. E. Straub and D. Thirumalai, J. Am. Chem. Soc., 2006, 128, 16159–16168.
61. K. L. Sciarretta, D. J. Gordon, A. T. Petkova, R. Tycko and S. C. Meredith, Biochemistry, 2005, 44, 6003–6014.
62. W. M. Berhanu and U. H. E. Hansmann, PLoS One, 2012, 7, e41479.
63. H. Chen, Y. Zhang, L. Li and J. G. Han, J. Phys. Chem. B, 2012, 116, 10219–10233.
64. W. Li, Y. Tang, H. Liu, J. Cheng, W. Zhu and H. Jiang, Proteins, 2008, 71, 938–949.
65. Y. Y. Cao, L. Wang, H. Ge, X. L. Lu, Z. Pei, Q. Gu and J. Xu, Mol. Diversity, 2013, 17, 515–524.
66. T. Luhrs, C. Ritter, M. Adrian, D. Riek-Loher, B. Bohrmann, H. Dobeli, D. Schubert and R. Riek, Proc. Natl. Acad. Sci. U. S. A., 2005, 102, 17342–17347.
67. A. K. Paravastu, R. D. Leapman, W. M. Yau and R. Tycko, Proc. Natl. Acad. Sci. U. S. A., 2008, 105, 18349–18354.

---

Footnote

† Electronic supplementary information (ESI) available. See DOI: 10.1039/c5ra12328c

---