Interpeptide interactions induce helix to strand structural transition in Aβ peptides

Abstract

Replica exchange molecular dynamics and all-atom implicit solvent model are used to compute the structural propensities in Aβ monomers, dimers, and Aβ peptides bound to the edge of amyloid fibril. These systems represent, on an approximate level, different stages in Aβ aggregation. Aβ monomers are shown to form helical structure in the N-terminal (residues 13 to 21). Interpeptide interactions in Aβ dimers and, especially, in the peptides bound to the fibril induce a dramatic shift in the secondary structure, from helical states toward β-strand conformations. The sequence region 10–23 in Aβ peptide is found to form most of interpeptide interactions upon aggregation. Simulation results are tested by comparing the chemical shifts in Aβ monomers computed from simulations and obtained experimentally. Possible implications of our simulations for designing aggregation-resistant variants of Aβ are discussed.

Introduction

Alzheimer’s disease (AD) belongs a growing list of disorders caused by aberrant aggregation of polypeptide chains¹. There is a compelling evidence accumulated over years of experimental work that the onset of AD is related to extracellular aggregation of Aβ peptides², which are produced by natural cleavage of the transmembrane amyloid precursor protein. These peptides are released in a variety of lengths, among which the 40- and 42-residue species, Aβ_1–40 and Aβ_1–42, are most abundant. Aggregation pathway of Aβ peptides represents a complex cascade of structural transitions, which involves oligomerization of individual chains and formation of amyloid fibrils³. Although Aβ amyloid fibrils show cytotoxic properties⁴, recent findings suggest that Aβ oligomers are the primary cytotoxic species in AD^5–8. It has been recently reported that synaptic structure and function can be impaired even by the smallest Aβ oligomers, dimers⁹. However, even if amyloid fibrils are not the leading causative agents in AD, they still represent a reservoir of Aβ monomers and in certain cases may play a protective role¹⁰.

Under normal physiological conditions Aβ aggregation, which leads to the formation of amyloid fibrils, is irreversible. Nevertheless, on a molecular level there appears to be a dynamic equilibrium between soluble oligomeric species and amyloid fibrils^11,12. Similarly, solution-state NMR experiments indicate that there is an equilibrium between Aβ monomers and oligomers¹³. Therefore, it is conceivable that a single Aβ peptide may be recycled through different aggregated states.

Aβ monomers, which are the starting species in amyloidogenic pathway, have been a focus of numerous experimental studies. It has been shown that Aβ monomers adopt significant helical structure in the presence of chemical solvents or in a micelle-like environment^14,15. However, in aqueous solutions Aβ monomers appear to be largely unstructured lacking stable secondary or tertiary structures^16,17. Despite overall dominance of random coil structures, several regions of Aβ adopt weak β-strand and α-helix conformations¹⁸. The endproduct of amyloid assembly, Aβ fibrils, is characterized by extensive and remarkably homogeneous β-sheet structure^19–24. In contrast, relatively little is known about the structural organization of Aβ oligomers. Several experimental studies suggested the development of α-helical structure in Aβ oligomers^13,25. Whether α-helical conformers represent kinetic intermediates formed along amyloidogenic pathway or they are one of the states in the equilibrium ensemble, it is clear that Aβ amyloid assembly is associated with dramatic changes in secondary structure.

Molecular dynamics (MD) simulations have a potential to reveal microscopic details of Aβ amyloid formation²⁶. Several studies have explored the conformational properties of Aβ_1–40 monomers^27–30. Garcia and coworkers have performed explicit water replica exchange molecular dynamics (REMD) simulations and found the existence of several structured regions in generally disordered Aβ monomer²⁸. Implicit solvent REMD simulations performed by Yang and Teplow also pointed out to random coil-like structure of Aβ monomer containing independently folded subunits³⁰. Interestingly, their simulations indicated that α-helix conformations in Aβ monomers have lower free energies than β-strand structures. A useful alternative to all-atom MD in exploring amyloid assembly are the coarse grained peptide models^31–33.

In this paper, we use all-atom REMD simulations to investigate the changes in the secondary structure in N-terminal truncated Aβ_10–40 peptides upon aggregation. To this end, we study three Aβ systems - monomers, dimers, and the Aβ peptides interacting with preformed amyloid fibril (Fig. 1). The first two systems are taken to represent, on an approximate level, the initial (monomers) and intermediate (dimers) stages of Aβ aggregation. The system, in which Aβ peptides are bound to the edges of amyloid fibril, is considered as the final stage of aggregation. We term these peptides in the third system as edge peptides because of their location on the edge of amyloid fibril³⁴. Although the three Aβ systems do not exhaust all Aβ species, they do allow us to trace the changes in Aβ structure upon aggregation. We show that interpeptide interactions induce a shift in the structural equilibrium from helical toward β-strand conformations. The helix→strand conversion is moderate in Aβ dimers, but becomes very pronounced in the edge Aβ peptides. We also identify the regions in Aβ sequence, which exhibit the helix→strand conversion and are engaged in interpeptide interactions. To test our REMD simulations and support the choice of N-terminal truncated peptide Aβ_10–40 as a model for Aβ_1–40 we computed the in silico chemical shifts in Aβ monomers and compared them with the experimental values¹⁸.

Fig. 1.

Three Aβ_10–40 systems studied in our simulations: (a) monomer, (b) dimer, (c) the peptides bound to the edge of Aβ amyloid fibril. N- and C-terminals are colored in purple and green. The peptides forming amyloid fibril in (c) are in grey. We use Aβ systems in (a)–(c) to represent, on a basic level, initial, intermediate, and final stages of Aβ aggregation. Aβ monomer in (a) and, to a lesser degree, dimer in (b) reveal the formation of helix structure in the N-terminal. Upon aggregation helical structure is replaced with β-strand conformations. This process is clearly manifested in Aβ peptides bound to the amyloid fibrils in (c), where β strands are shown in cartoon representation. (d) Sequence of Aβ_10–40 peptide. The residues in red are predicted to form helical structure by Predator algorithm⁶¹.

Materials and Methods

Molecular dynamics simulations

To perform simulations of Aβ peptides we used CHARMM MD program³⁵ and all-atom force field CHARMM19 coupled with the SASA implicit solvent model³⁶. Because we investigate the changes in Aβ secondary structure, it is important to choose the force field not known for structural bias. Previous simulations have indicated that CHARMM19+SASA force field does not favor particular protein secondary structure. It has been used to fold polypeptides, which contained α-helices or β-sheets^37,38. CHARMM19+SASA simulations were also employed for studying oligomerization of amyloidogenic peptides^39,40.

In this work, we use the peptide Aβ_10–40, which is the N-terminal truncated fragment of Aβ_1–40. Several reasons dictated our selection of Aβ_10–40. First, solid-state NMR resolved the fibril structure of Aβ_1–40 excluding the first nine N-terminal residues²¹. Subsequent studies have shown that the fibril structures of Aβ_1–40 and Aβ_10–40 are almost identical⁴¹. Second, it is known that the size distributions for Aβ_10–40 and Aβ_1–40 oligomers are similar⁴². These findings suggest that the N-terminal play a minor role Aβ aggregation.

Three Aβ_10–40 systems were considered: monomer, dimer, and the peptides interacting with Aβ fibril (Fig. 1). Each of these systems was subject to spherical boundary condition with the radius R_s = 90Å and the force constant k_s = 10kcal/(molŲ). The concentration of Aβ peptides is therefore of an order of mM. The system of Aβ peptides interacting with amyloid fibril was introduced in our previous studies^34,40 (Fig. 1c). Briefly, the system consists of four Aβ peptides forming a fibril fragment and two incoming (edge) peptides. The fibril structure is modeled using the coordinates of backbone atoms determined from the solid-state NMR measurements²¹. To emulate the stability of large fibril sample, the backbones of fibril peptides (Fig. 1c) were constrained to their experimental positions using soft harmonic potentials with the constant k_c = 0.6kcal/(molŲ)⁴⁰. The harmonic constraints permit backbone fluctuations with the amplitude of about 0.6Å at 350K, which are comparable with the fluctuations of atoms on the surface of folded proteins⁴³. Constraints were not applied to the side chains of fibril peptides nor to the edge peptides. The latter were free to dissociate and reassociate with the fibril. The constraints capture the rigidity of amyloid fibril and eliminate the necessity to simulate large fibril systems to achieve their stability. More generally, three Aβ systems allowed us to consider structural changes in Aβ peptides upon aggregation.

Replica exchange simulations

To achieve exhaustive conformational sampling we used replica exchange molecular dynamics (REMD)⁴⁴. This method has shown its efficiency in sampling rugged free energy landscapes and has been applied to study protein folding and aggregation^{39,40,45–50}. To simulate Aβ monomers and dimers we used 24 replicas distributed linearly with the increment of 10K in the temperature range from 300 to 530K. The exchanges were attempted every 40 ps in all neighboring replica pairs. The average acceptance rates were 67% (monomer) and 54% (dimer). In all, we produced four (monomer) and seven (dimer) REMD trajectories of the length 0.4µs each (per replica). Therefore, the cumulative simulation time for all replicas was 76µs (monomer) and 134µs (dimer). The structures were saved every 40 ps. Between replica exchanges the system was evolved using NVT underdamped Langevin dynamics with the damping coefficient γ = 0.15ps⁻¹ and the integration step of 2fs. The REMD simulations of Aβ peptides interacting with the fibril were described previously⁴⁰.

In each REMD trajectory we determined the initial equilibration interval τ_eq by monitoring the onset of equilibrium regime in the total effective energies E_eff (the sum of potential and solvation energies). As a result, the initial parts of REMD trajectories of the lengths up to 80 ns were excluded. Consequently, the cumulative equilibrium simulation times were τ_sim = 72µs (monomer) and 113µs (dimer).

Computation of structural probes

To characterize intra- and interpeptide interactions we computed the number of side chain contacts as described in our previous studies⁵¹. Backbone hydrogen bonds between NH and CO groups were assigned according to Kabsch and Sander⁵². Secondary structure in Aβ peptides was assigned by evaluating their dihedral angles (ϕ, ψ)^53,54. To this end, the grid with the spacing of 18° was superimposed on the Ramachandran plot. The β-strand conformations are enclosed by the vertices of the polygon (−180°,180°), (−180°,126°), (−162°,126°), (−162°,108°), (−144°,108°), (−144°,90°), (−50°,90°), (−50°,180°); helix structure is confined to the polygon (−90°,0°), (−90°,−54°), (−72°,−54°), (−72°,−72°), (−36°,−72°), (−36°,−18°), (−54°,−18°), (−54°,0°). (Note that these definitions do not distinguish α-helix, 3₁₀-helix, or π-helix.) Using these definitions the fractions of residues in helix and β-strand conformations, H and S, can be computed in any structure.

We have also used the program STRIDE for secondary structure assignment⁵⁵. Although the assignments of helices based on (ϕ, ψ) angles and STRIDE are very similar, the correspondence between β-structure is not straightforward. Because STRIDE defines β-sheet structure using HBs, its fraction is low in disordered systems such as Aβ dimers and monomers. As a result STRIDE underestimates extended β-strand-like structures by assigning them to turns and coils. Consequently, in STRIDE about one-third of turn structure and half of coil structure correspond to β-strand conformations defined on the basis of (ϕ, ψ) angles. Because our aim is to monitor structural changes in Aβ peptides upon aggregation, we chose to assign secondary structure using (ϕ, ψ) angles. However, for completeness we list respectively the values of extended (β-sheet), α-helical, and β-turn structure contents obtained from STRIDE: monomer − 0.01, 0.38, 0.41; dimer − 0.06, 0.19, 0.49; edge peptides − 0.30, 0.06, 0.35. Because the amount of bridge structure is negligible, the remaining fractions correspond to coil.

Throughout the paper angular brackets < ‥ > indicate thermodynamic averages. Because dimer and fibril systems include two indistinguishable peptides, we report averages over two peptides. The distributions of states produced by REMD were analyzed using multiple histogram method⁵⁶.

Convergence of REMD simulations

To evaluate the quality of REMD sampling we consider the number N_s of unique states (E_eff, N_mhb), which were sampled in the course of simulations at least once. Each state (E_eff, N_mhb) is defined by the effective energy of the Aβ system E_eff and the number of intrapeptide HBs, N_mhb. Fig. 2 shows N_s for Aβ dimer as a function of the cumulative equilibrium simulation time τ_sim. At τ_sim ≳ 10µs N_s approximately levels off suggesting approximate convergence of REMD. To further test the reliability of REMD we divided the dimer simulations into two equal subsets and analyzed them independently. The resulting errors in the intrapeptide and interpeptide HBs, < N_mhb > and < N_dhb >, were 2% and 4%, respectively. In Fig. 2 (inset) we show the REMD error analysis for the secondary structure content. The average errors for the helix and strand fractions at individual sequence positions were 8% and 7%. The errors in monomer system are smaller than those in Aβ dimer. The error analysis for the fibril system was reported earlier⁴⁰.

Fig. 2.

The number N_s of the new states (E_eff, N_mhb) in Aβ dimer not previously sampled in REMD as a function of the cumulative equilibrium simulation time τ_sim. Continuous and dotted lines indicate N_s for each of the two peptides in a dimer. The inset shows the fractions of helix < H(i) > and β-strand < S(i) > structure formed by individual residues i in the dimer. The data in grey refer to the first and second half-sets of REMD trajectories, the data in black are for the full trajectory set. The errors in < H(i) > and < S(i) > are increased in the N-terminal due to interpeptide interactions.

Results

Using REMD we investigated the distribution of structures sampled by Aβ_10–40 monomers, dimers, and the Aβ_10–40 peptides bound to amyloid fibrils (Fig. 1). We have previously showed that Aβ peptides docked to amyloid fibrils assume ordered β-sheet conformations at the locking temperature T_l ≈ 360K⁴⁰. Consequently, to compare different Aβ species on equal basis, we report their structural properties at the temperature T_s = 350K < T_l. Following the allocation of β-structure in Aβ_1–40 amyloid fibril²¹, we distinguish three sequence regions in Aβ_10–40 peptide - N-terminal (residues 10 to 23), which corresponds to the first fibril β-strand, C-terminal (residues 29 to 39), which corresponds to the second fibril β-strand, and the turn region (residues 24 to 28) (Fig. 1d).

Conformational propensities of Aβ monomers

With the decrease in temperature Aβ_10–40 monomer collapses to the structures with the average radius of gyration < R_g >≈ 14.0Å. The collapse of Aβ_10–40 is very gradual and spans the entire REMD temperature range. This observation is likely to be the consequence of the lack of Aβ stable tertiary fold. However, if the monomer melting temperature is attributed to the maximum in d < R_g(T) > /dT, then it is about 470 K.

To probe the interactions in Aβ monomers we computed the number of side chain hydrophobic contacts < C_mHH > and the number of backbone hydrogen bonds (HB) < N_mhb >. The collapsed conformations are stabilized by hydrophobic interactions (< C_mHH >≈ 7.6) and intrachain HBs (< N_mhb >≈ 15.2). Fig. 3a shows the thermal averages of the number of HBs < N_mhb(i) > formed by NH and CO groups of the residues i in Aβ monomer. Visual inspection of Fig. 3a suggests that N-terminal should be more structured than the C-terminal. Indeed, the average numbers of HBs per residue in the N- and C-terminals are 1.2 and 0.9, respectively.

Fig. 3.

Thermal averages of the number of intrapeptide HBs < N_mhb(i) > formed by the backbone NH and CO groups of the residues i in Aβ monomer (a) and dimer (b). The data in grey and black are for NH and CO groups, respectively. The allocation of N- and C-terminals is shown by boxes in this figure and in Figs. 4,5,7,8. Large number of HBs formed in the N-terminal of Aβ peptide is indicative of helix formation.

To further investigate structural propensities we computed the distribution of secondary structure. The average fractions of residues in β-strand and helix conformations are < S >≈ 0.22 and < H >≈ 0.34, respectively. The random coil fraction is about 0.44. Fig. 4a displays the distributions < S(i) > and < H(i) >, which report the fractions of β-strand and helix structure formed by individual residues i. The average fractions of residues in β-strand and helix states in the N-terminal are < S_N >≈ 0.15 and < H_N >≈ 0.53, respectively. Within the C-terminal a weak preference for β-structure is observed (< S_C >≈ 0.27 for β-strand and < H_C >≈ 0.16 for helix). Thus, helix structure in Aβ monomers is observed more frequently than β-strand states.

Fig. 4.

Thermal distributions of the fractions of helix < H(i) > and β-strand < S(i) > structure formed by individual residues i in Aβ monomer (a), dimer (b), and the edge Aβ peptides bound to amyloid fibril (c). Compared to Aβ monomers there is a uniform decrease in helix content and simultaneous increase in β-population in dimers, and, especially, in edge Aβ peptides.

Within the N-terminal the distributions of < H(i) > and < S(i) > show gradual variations (Fig. 4a). In contrast, due to structural flexibility of four Gly residues (Gly29, Gly33, Gly37, Gly38) pronounced variations in β-strand and helix contents are observed in the C-terminal. At these four Gly positions the fraction of random coil structure reaches 0.9. In contrast, the N-terminal (Fig. 1d) lacks Gly residues. It is also worth noting that the concentration of Gly residues in the C-terminal is consistent with the dominance of β-turns in this region. According to STRIDE⁵⁵ all residues within the C-terminal have the probabilities to form β-turn < T(i) > in excess of 0.5. For comparison, the average probability to form β-turns in the N-terminal is only 0.29.

Conformational propensities of Aβ dimers

What is the impact of interpeptide interactions on the structural propensities of Aβ peptides? To answer this question we first consider the distribution of intrapeptide HBs < N_mhb(i) > in Aβ dimers (Fig. 3b). Similar to the distribution computed for Aβ monomer in Fig. 3a, N-terminal appears to have elevated number of HBs. The average numbers of HBs per residue in the N- and C-terminals are 0.8 and 0.6, which are lower than in Aβ monomer. The thermal average of the total number of intrapeptide HBs < N_mhb > is ≈ 10.2, whereas the number of intrapeptide hydrophobic contacts is < C_mHH >≈ 6.3. Thus, compared to Aβ monomers the number of intrapeptide HBs in dimers is decreased by one-third. Despite the decrease in intrapeptide interactions, Aβ peptides in dimers have approximately the same dimensions (< R_g >≈ 14.2Å) as monomeric species.

Similar to the analysis of Aβ monomers our next step is to probe the distribution of secondary structure in Aβ dimers. From REMD data we obtained the average β-strand and helix contents in Aβ peptides to be < S >≈ 0.35 and < H >≈ 0.23, respectively. The random coil fraction is about 0.42. To get better insight in the secondary structure propensities we plot in Fig. 4b the fractions of helix < H(i) > and β-strand < S(i) > structure formed by individual residues i. Although the profiles of < H(i) > and < S(i) > remain qualitatively similar to those computed for Aβ monomers (Fig. 4a), a considerable redistribution of secondary structure is evident. For example, the average fractions of residues in helix and β-strand states in the N-terminal become equal in dimers (< H_N >≈< S_N >≈ 0.34), whereas a strong preference for β-conformations is observed in the C-terminal (< S_C >≈ 0.35 versus < H_C >≈ 0.12). As in Aβ monomer a considerable drop in β-strand content is observed at four Gly residues (Gly29, Gly33, Gly37, Gly38) in the C-terminal. Hence, in contrast to Aβ monomers β-structure in dimers is observed more frequently than helical states.

Interpeptide interactions are quantified by the numbers of interpeptide HBs < N_dhb > and hydrophobic side chain contacts < C_dHH >. Using REMD we found that < N_dhb >≈ 3.8 and < C_dHH >≈ 5.5. The probability of associated state, i.e., the probability of forming any interpeptide HB, is 0.86. To obtain residue-specific information about interpeptide interactions we computed the thermal contact map < C(i, j) >, which displays the probabilities of forming interpeptide contacts between amino acids i and j (Fig. 5a). Two observations follow from this plot. First, N-terminal of Aβ tends to form more interpeptide interactions than the C-terminal. From the contact map < C(i, j) > we calculated that the numbers of interpeptide side chain contacts formed by the N- and C-terminals are 16.6 and 8.8, respectively. Therefore, in Aβ dimers the interpeptide interface primarily involves the N-terminal.

Fig. 5.

(a) The thermal contact map < C(i, j) > (lower right corner), which displays the probabilities of forming interpeptide contacts between amino acids i and j in Aβ dimer. Difference contact map < ΔC(i, j) >, which reflects the change in intrapeptide contact probabilities in Aβ dimers relative to monomers, is displayed in the upper left corner. < C(i, j) > and < ΔC(i, j) > are color coded according to the scales shown. Areas in white correspond to the intrapeptide contacts, which occur with the probability < 0.1 in monomers. The contact map < C(i, j) > indicates that most interpeptide interactions in Aβ dimer are formed by the N-terminals. The difference contact map < ΔC(i, j) > shows that the intrapeptide contacts destabilized most in the dimer are those involved in helix formation in monomers. (b) Probability distribution P(cosϕ) for the cosine of the angle ϕ, which describes the orientation of Aβ N-terminals in dimer. Antiparallel docking of Aβ peptides in the dimer is weakly preferred.

The orientation of Aβ peptides in the dimer was probed using the vectors R⃗(k) = r⃗_Cα(23, k) − r⃗_Cα(10, k), where r⃗_Cα(23, k) and r⃗_Cα(10, k) are the radius-vectors of C_α atoms of the residues Asp23 and Tyr10 in the peptide k. Because the fraction of ordered structure (helix or strand) in the N-terminal is high (~ 0.7) and, consequently, this region is relatively rigid, R⃗(k) approximately describes the orientation of the N-terminal. Fig. 5b presents the distribution P(cosϕ), where ϕ is an angle between R⃗ (1) and R⃗(2). The distribution P(cosϕ) is weakly bimodal, reflecting a tendency of Aβ peptides to form antiparallel docking interface as shown in Fig. 1b. Consequently, the probability of antiparallel orientation (cosϕ < 0) is about 0.6.

Second, although none of the probabilities < C(i, j) > are larger than 0.1, there are several side chain contacts, which occur with elevated probabilities, namely Tyr10-Phe20, Tyr10-Asp23, Glu11-Glu11, Glu11-Asp23, His13-Phe20, His13-Asp23. Most of these contacts are involved in the formation of antiparallel dimer interface in the N-terminal. The probability of forming intermolecular salt bridge Asp23-Lys28, which is important for the stability of Aβ fibrils²¹, is low (0.05).

Finally, we evaluate the change in Aβ peptide structure by computing the difference in intrapeptide side chain contact probabilities in Aβ dimers relative to monomers. The difference contact map < ΔC(i, j) > in Fig. 5a shows that the most notable change is related to the destabilization of the helix-like contacts (i, i + 4), where i = 15, ‥, 21 and i = 27, ‥, 33. With the exception of very few side chain pairs the intrapeptide contacts tend to become less stable in Aβ dimer. Indeed, the average value of < ΔC(i, j) > is about −0.1. This observation is consistent with partial unraveling of helix structure in Aβ dimers. The probability of intrapeptide salt bridge Asp23-Lys28 is reduced from 0.31 (monomer) to 0.22 (dimer).

Conformational propensities of the Aβ peptides interacting with amyloid fibril

We now analyze the secondary structure and interactions in the Aβ peptides bound to the edge of amyloid fibril⁴⁰. The numbers of intrapeptide HBs < N_mhb > and hydrophobic side chain contacts < C_mHH > are reduced to 4.5 and 5.5, respectively. Therefore, compared to Aβ monomer there is a three-fold decrease in < N_mhb >. There is also a significant extension of the edge Aβ peptides as their radius of gyration < R_g > increases from 14.0 (Aβ monomers) to 18.7Å. The loss in intrapeptide interactions is compensated by the dramatic rise in peptide-fibril interactions. The numbers of HBs < N_ehb > and hydrophobic side chain contacts < C_eHH > linking edge Aβ peptide with the fibril are increased to 10.8 and 10.0. Similar to Aβ dimers most of such interpeptide side chain contacts are formed by the N-terminal. From the thermal contact map < C(i, j) > (data not shown) we obtained that the numbers of peptide-fibril side chain contacts formed by the N- and C-terminals are 21.8 and 12.0, respectively.

In the edge Aβ peptides the average β-strand < S > and helix < H > contents are 0.52 and 0.11, whereas the random coil fraction is 0.37⁴⁰. Fig. 4c shows the distributions of helix < H(i) > and β-strand < S(i) > structure formed by individual residues i in Aβ peptides bound to the fibril. This plot demonstrates that the probability of β-strand states exceeds that of helix for all residues. The average fractions < S > in the N- and C-terminals are < S_N >≈ 0.55 and < S_C >≈ 0.50, respectively. For comparison, the helical fractions in these regions are 0.16 and 0.05. Fig. 4c also suggests that as in Aβ monomers and dimers four Gly residues (Gly29, Gly33, Gly37, Gly38) may impart considerable flexibility to the C-terminal. Thus, in the edge Aβ peptides there is a strong dominance of β-strand structure and most of intrapeptide interactions are replaced with peptide-fibril HBs and side chain contacts.

Discussion

Interpeptide interactions drive a shift in secondary structure

We performed REMD simulations to sample conformational states of Aβ monomers, dimers, and the edge peptides bound to amyloid fibril. The main result of this investigation is that a significant shift in secondary structure distribution is observed in dimeric and edge Aβ species compared to Aβ monomers. Fig. 6a summarizes the changes in helix and β-strand populations at different stages of Aβ aggregation. Aβ monomer tends to adopt generally random coil structure augmented by significant helix (< H >= 0.34) and, to a lesser extent, β-strand (< S >= 0.22) populations. Large helix fraction is primarily due to a strong propensity to form this structure in the N-terminal (Fig. 3a). Formation of dimeric structures triggers partial conversion of helix into β-strand structure as the β-strand content < S > increases by ≈ 50% from 0.22 to 0.35. The decrease in helix population from 0.34 to 0.23 matches in reverse manner the increase in < S >. The association of Aβ peptides with large aggregates (amyloid fibrils) causes further helix → strand conversion. Compared to Aβ dimers β strand content increases by a factor 1.5 from 0.35 to 0.52, while the helix population decreases in half (from 0.23 to 0.11). The difference contact map in Fig. 5a is consistent with the destabilization of helical structure in Aβ dimers. Note also that the helix → strand conversion is in line with a dramatic extension of edge peptides compared to monomers or dimers (from < R_g >≈ 14.0 to 18.7Å).

Fig. 6.

(a) Probabilities of helix, β-strand, and random coil states in Aβ monomers, dimers, and edge peptides. The plot summarizes the conversion of helix into β-strand structure upon aggregation. (b) Two-dimensional projections of the free energy ΔF(H, S) as a function of the helix and strand contents, H and S. The minima in free energy are set to zero. The values of ΔF(H, S) are color coded according to the scale. (c) One-dimensional profiles of the free energies ΔF(H) and ΔF(S). The data for monomer, dimer, and edge peptides are shown by squares, open and filled circles, respectively. The states with zero helix or strand contents are set to have zero free energy. The plots in (b) and (c) demonstrate that upon aggregation free energy minimum shifts from helix to β-strand region. ΔΔF_S and ΔΔF_H are the total changes in free energy of strand and helix states occurring between Aβ monomers and the peptides bound to amyloid fibril. The free energies in (b,c) are computed at T_s = 350K.

It is natural to suggest that the helix → strand conversion is driven by interpeptide interactions. With respect to Aβ monomer the number of intrapeptide HBs < N_mhb > is reduced by 30% in dimers and three-fold in edge peptides (from 15.2 to 10.2 and 4.5, respectively). Simultaneously, in edge peptides compared to dimers the number of interpeptide HBs is increased almost by a factor of three (from 3.8 to 10.8). Thermal contact maps < C(i, j) > for dimers (Fig. 5a) and edge peptides suggest that about two-third of interpeptide interactions are formed by Aβ N-terminals. It is therefore indicative that the most significant change in secondary structure occurs within the N-terminal, in which the fraction of β-strand < S > almost quadruples (from 0.15 to 0.55). For comparison, the change in < S > in the C-terminal is less than two-fold.

The helix → strand conversion occurring upon Aβ aggregation can be further illustrated by free energy computations. Fig. 6b shows free energies ΔF(H, S) as a function of the helix H and β-strand S fractions. In monomers, a compact free energy minimum is located at H ~ 0.35 and S ~ 0.2. In dimers, the minimum becomes stretched to include a broad range of H (0.05 to 0.4) and S (0.1 to 0.6) states. For the edge peptides, ΔF(H, S) reveals a well-defined minimum centered at H ≲ 0.1 and S ~ 0.6. It is worth noting that the plot of ΔF(H, S) for dimers suggests that large (compared to other Aβ species) fluctuations in secondary structure should be observed. It follows from one-dimensional free energy profiles ΔF(H) shown in Fig. 6c that with respect to monomers the helix state is destabilized in dimers by 2.8RT and in edge peptides by ΔΔF_H = 4.0RT. In contrast, compared to monomers the free energy of β-strand structure decreases by −1.5RT in dimers and by ΔΔF_S = −5.3RT in edge peptides. Taken together, our data indicate that upon aggregation a dramatic helix → strand conversion occurs in Aβ peptides.

Comparison of experiments and simulations: Aβ monomers

To test our simulations of Aβ monomers we computed the chemical shifts δ_sim(i) for C_α and C_β carbons in residues i using the program SHIFTS⁵⁷. To compute δ_sim(i) we used the lowest REMD temperature (T = 300K). We choose to analyze C_α and C_β chemical shifts because of their sensitivity to α-helix and β-strand structures¹⁸. In Fig. 7 the distributions of δ_sim(i) are compared with the corresponding experimental chemical shifts δ_exp(i) measured by Zagorski and coworkers¹⁸. The experimental chemical shifts were obtained for Aβ_1–40-Met^red monomers at 278K. Several important observations follow from Fig. 7. First, despite a 20K difference in experimental and simulation temperatures, an excellent agreement between C_β δ_exp(i) and δ_sim(i) is observed (Fig. 7a). The corresponding correlation factor is r = 0.9995. The agreement between C_α δ_exp(i) and δ_sim(i) is also reasonably good, although some discrepancy can be seen in the N-terminal (Fig. 7b). However, the overall correlation between C_α chemical shifts is still very strong (r = 0.984, inset to Fig. 7b). Second, the experiments studied the full-length Aβ_1–40 and in our simulations we used the truncated peptide Aβ_10–40. A good agreement between the distributions of δ_exp(i) and δ_sim(i) suggests that the truncation of the residues 1–9 does not lead to qualitative changes in the structural properties of Aβ monomers. This conclusion is in line with the arguments for selecting Aβ_10–40 presented in Methods^21,41,42. It is also interesting to note that the recent study has also demonstrated that C_β chemical shifts are in better agreement with simulation data than those obtained for Cα⁵⁸.

Fig. 7.

Experimental δ_exp(i) (open circles) and simulation δ_sim(i) (full triangles) chemical shifts in Aβ monomer as a function of residue number i: (a) C_β atoms, (b) C_α atoms. The inset to (b) shows δ_sim(i) vs δ_exp(i) for C_α atoms. Chemical shifts of Gly C_β atoms are set to zero.

Because our simulations suggest the formation of the helix in the N-terminal of Aβ monomers (Fig. 4a), it is important to evaluate the reliability of this prediction. Three lines of evidence support helix structure. First, it is instructive to compare C_α δ_exp(i) in the N-terminal with the reference values of chemical shifts for α-helix δ_α, β-strand δ_β, and random coil structures δ_RC reported in the literature⁵⁹. For all residues in the N-terminal the reference chemical shifts are ordered as δ_β < δ_RC < δ_α. For the positions Val12, His14, Gln15, Leu17, Val18, Phe19, and Phe20 experimental δ_exp fall between random coil and α-helix values, i.e., δ_RC < δ_exp < δ_α (Table 1). At other positions, δ_exp ≲ δ_RC and the presence of helix is less likely. Therefore, the data in Table 1 are consistent with the formation of helix at seven out of 13 positions in the N-terminal. Recent NMR study of Aβ_1–40 monomer determined eight J-coupling constants in the sequence region 10 to 23²⁸. Because the coverage of measurements is not continuous, it is difficult to make reliable identification of helical structure, although three constants are within the helical range (³J_HNHα < 6Hz)⁶⁰. Furthermore, out of eight measured J-coupling constants only two exceeded the value of ³J_HNHα = 7Hz normally attributed to β-structure (with the same accuracy as for single helical J-couplings)⁶⁰.

Table I.

Experimental and Reference C_α Chemical Shifts for N-Terminal Residues in Aβ Monomer

Residue	δβa (ppm)	δRCa (ppm)	δαa (ppm)	δexpb (ppm)
Glu11	55.55	56.39	59.30	56.38
Val12	60.72	61.80	65.96	62.80
His13	54.80	55.78	59.62	55.79
His14	54.80	55.78	59.62	56.03
Gln15	54.33	55.94	58.61	56.20
Lys16	55.01	56.40	59.11	56.35
Leu17	53.94	54.85	57.54	55.04
Val18	60.72	61.80	65.96	61.87
Phe19	56.33	56.94	60.74	57.31
Phe20	56.33	56.94	60.74	57.23
Ala21	50.86	52.67	54.86	52.51
Glu22	55.55	56.39	59.30	56.29
Asp23	53.41	54.09	57.04	53.97

^aData from Ref.60.

^bChemical shifts from Ref.18.

Second, the support for helix structure comes from simulations. Recent MD study, which used different force field and implicit solvent model (AMBER and Generalized Borne model), demonstrated that the propensity to form α-helix at the positions 14 to 22 is larger than that for β-strand³⁰. For the residues Leu17, Val18, and Phe20 the combined probability of all types of helices reached about 0.3. The simulations have also shown that the free energy of α-helix conformers is lower than of any other conformational state. In another study, which employed explicit solvent and GROMACS force field, the probability of helix conformers in Aβ_1–40 monomer exceeded that of β-structure about three-fold²⁷. Interestingly, helical conformations persisted within the sequence positions 14 to 21 after being melted in other sequence regions. (It is important to note that the memory of initial helix-favoring conditions taken from micellular environment could have biased the sampling, which was unlikely to reach equilibrium on ~100 ns time scale.)

Third, secondary structure analysis programs do not give consensus prediction concerning the N-terminal of Aβ_10–40. Predator⁶¹ predicts α-helix conformations at the Aβ_10–40 positions 13 to 17 (Fig. 1d), whereas PHD⁶² assigns extended structure in that region. It is worth noting that the truncation of Aβ residues 1 to 9 could elevate the helical content in the N-terminal of Aβ^10–40. Our preliminary simulations of Aβ_1–40 monomer⁶³ indicated that the fraction of helix structure < H > is reduced from 0.34 (Aβ_10–40) to 0.29 (Aβ_1–40), whereas the strand content rises from 0.22 to 0.28, respectively. Therefore, the dominant structure in Aβ_1–40 is random coil (0.43), whereas the rest is about equally divided between helix and beta conformers. Consequently, the amount of helix in Aβ_1–40 may be somewhat smaller than reported in our study for Aβ_10–40.

Taken together, the arguments cited above and overall good agreement between simulation and experimental chemical shifts support the formation of helical conformations in Aβ monomers.

Solution NMR studies are in general agreement that Aβ monomers in aqueous solutions lack stable tertiary structure^16–18. In particular, NMR data do not reveal the presence of stable long-range interactions in Aβ monomer. This observation is consistent with the low probability of long-range side chain contacts in Aβ monomers in Fig. 5a. Experimental studies have also demonstrated that several regions in Aβ monomer are relatively structured. For example, NMR dynamics in Aβ_1–40 monomer revealed higher mobility of the C-terminal compared to the N-terminal⁶⁴. Similar conclusion has been reached in all-atom explicit water simulations²⁸. Garcia and coworkers established that the N-terminal (residues 1 to 18) samples restricted conformational space compared to the C-terminal (25 to 40). These findings are consistent with the formation of helix structure in the N-terminal in our simulations.

To provide direct evaluation of Aβ monomer rigidity, we computed using REMD the standard deviations δϕ(i) and δψ(i) of the backbone angles ϕ and ψ for the residues i. Fig. 8 shows that the amplitude of fluctuations in the N-terminal is generally lower than in the C-terminal. Indeed, the average of the combined set of δϕ(i) and δψ(i) values in the N-terminal is 42, but it grows to 67 in the C-terminal. The increase in backbone flexibility is mainly caused by four Gly residues, which show dramatically higher values of δϕ(i) than other residues. It is conceivable that large structural fluctuations caused by Gly residues destabilize interpeptide interactions originating in the Aβ C-terminal. Consequently, as shown in our previous studies^34,40 and in this work Aβ peptides tend to aggregate through their N-terminals. A similar suggestion that Aβ C-terminal may not represent the primary aggregation interface was put forward by Hou et al.¹⁸ and by Melquiond et al.⁶⁵.

Fig. 8.

Standard deviations δϕ(i) = (< ϕ²(i) > − < ϕ(i) >²)^½ and δψ(i) = (< ψ²(i) > − < ψ(i) >²)^½ of the backbone angles ϕ(i) and ψ(i) for the residues i in Aβ monomer. The data for δϕ and δψ are shown in grey and black, respectively. Due to backbone steric constraints δϕ(i) are smaller than δψ(i) for all residues except Gly. The distributions of δϕ(i) and δψ(i) suggest that the C-terminal is more exible than the N-terminal.

Comparison of experiments and simulations: Aβ oligomers

Experimental studies of Aβ_1–40 oligomers have revealed significant population of α-helix states^13,25. α-Helix states may represent kinetic intermediates in amyloid assembly or these states may occur at equilibrium with other conformations¹³. These results appear to support our observation of significant helix content in Aβ dimers. Furthermore, solution-state NMR has been used to identify amino acids involved in interpeptide interactions in Aβ_1–40 dimers¹³. Although unambiguous assignment of interpeptide interactions to specific amino acids is difficult, it has been concluded that the sequence positions 15 to 24 contribute most to dimer stability. This region of Aβ chain approximately coincides with the N-terminal of Aβ_10–40, which form most of interpeptide interactions. Interestingly, our simulations suggest that the largest number of interpeptide contacts is formed by Tyr10 (1.7 contacts, on an average). In solution-state NMR this residue also provides large contribution to interpeptide interactions¹³.

In summary, by taking into account the in silico distributions of chemical shifts and interpeptide interactions, overall dominance of random coil conformations, and flexibility of the C-terminal it appears that our simulations are in reasonable agreement with the experiments and previous computational studies.

Conclusions

In this paper, we used REMD simulations to compute the structural propensities of Aβ monomers, dimers, and Aβ peptides bound to the amyloid fibril edge. In a crude way, these three systems represent initial, intermediate, and final stages in Aβ aggregation. We showed that Aβ_10–40 monomers form helical structure in the N-terminal. Upon aggregation interpeptide interactions induce a dramatic shift in the structural equilibrium, away from helical states toward β-strand conformations. This helix → strand conversion is relatively moderate in Aβ dimers, but is very pronounced in the Aβ peptides interacting with amyloid fibrils. We also determined that the N-terminal of Aβ_10–40 peptide forms most of interpeptide interactions. To test our REMD simulations we computed in silico chemical shifts for Aβ monomers and compared them with the experimental data¹⁸. Their good agreement suggest that our study fairly well represents the distribution of states sampled by Aβ peptides.

Our result that the N-terminal of Aβ peptides forms helical structures and is predominantly involved in interpeptide interactions may have implications for controlling Aβ amyloidogenesis. By site-directed mutations one may enhance α-helical propensity in Aβ peptides and therefore block helix → strand conversion occurring upon aggregation. Hence, it is of interest that the consensus secondary structure prediction⁶⁶ demonstrates that substituting Val18 and Phe19 with Glu triggers uniform assignment of α-helical structure in the Aβ_10–40 N-terminal (positions 14 to 23) by all prediction algorithms. This mutation may be considered in the design of Aβ variants resistant to aggregation.