Effects of Familial Mutations on the Monomer Structure of Aβ₄₂

Abstract

Amyloid beta (Aβ) peptide plays an important role in Alzheimer’s disease. A number of mutations in the Aβ sequence lead to familial Alzheimer’s disease, congophilic amyloid angiopathy, or hereditary cerebral hemorrhage with amyloid. Using molecular dynamics simulations of ∼200 μs for each system, we characterize and contrast the consequences of four pathogenic mutations (Italian, Dutch, Arctic, and Iowa) for the structural ensemble of the Aβ monomer. The four familial mutations are found to have distinct consequences for the monomer structure.

Amyloid beta (Aβ) peptides have long been thought to play a central role in Alzheimer’s disease (AD). Usually 40 or 42 residues in length, Aβ peptides are proteolytic products of the Aβ precursor protein and they aggregate to form the fibrillar plaques in AD patients’ brains. Besides fibrillar plaques, Aβ oligomers are also neurotoxic. The significance and nature of Aβ oligomerization has recently become a focus of intensive research studies and debates (1,2). Notably, numerous pathogenic mutations have been identified in the Aβ precursor protein sequence and in the enzymes involved in Aβ processing (3). These mutations generally lead to early onset of AD or cerebral amyloid angiopathy. Understanding how the pathogenic mutations alter Aβ oligomerization/aggregation is essential to our understanding of the disease mechanism.

Four of these pathogenic mutations (Italian E22K, Dutch E22Q, Arctic E22G, and Iowa D23N) cluster in the region of E22 and D23 in the Aβ sequence (distal from proteolytic cleavage sites) and they have higher neurotoxicity compared to wild-type (WT) Aβ (4). These mutations are thought to modify the physicochemistry of the peptide. For example, kinetic studies (4) show that the E22K and E22Q mutations lead to faster peptide aggregation, whereas the E22G and D23N mutations result in slightly slower aggregation than WT Aβ₄₂ (although the E22G mutation shows increased protofibril formation (5)). Recent solid-state NMR studies also suggest that rather than the in-register β-sheet conformation adopted by WT Aβ, the Iowa D23N mutant forms amyloid fibrils with antiparallel β-sheet structure (6).

To understand how the mutations modify the peptide oligomerization/aggregation it is critical to characterize the starting point of the process, the monomers. Unfortunately, investigating the early phase of the oligomerization process experimentally is a challenging task due to the high aggregation propensity of Aβ and its intrinsic disorder. Therefore, a number of computational approaches have been adopted to investigate the consequences of mutations for the monomer structure (7–16). However, due to the high computational demands of explicit-solvent molecular dynamics (MD) simulations to simulate full-length Aβ peptides, most of these computational studies are either on Aβ fragments (to decrease the system size) using explicit-solvent simulations (8–12) or on full-length Aβ using implicit-solvent simulations (which are less computationally demanding and enable longer simulation times, but lack explicit water molecules in the simulations to fully describe water-peptide interactions) (13–15). In a very recent report, explicit-solvent simulations were used to study the effects of the E22Q mutation on full-length Aβ; however, rather limited data (<10 μs) were collected (16). Thus, characterizing full-length Aβ monomers remains quite a daunting task even with simulations.

To characterize the effects of mutations on full-length Aβ monomer using explicit-solvent MD simulations, we employed distributed computing (17) to simulate the WT Aβ₄₂, Aβ₄₂-E22K, Aβ₄₂-E22Q, Aβ₄₂-E22G, and Aβ₄₂-D23N monomers. MD simulations of >200 μs were performed for each system and AMBER ff99sb (18) and the tip3p water model (19) were used for force field parameters. Peptide configurations in the MD trajectories were clustered with the root mean-square deviation metric to identify representative conformations (i.e., states) and transitions between these states were counted. Markov state model analysis was then performed where the master equations were solved and the equilibrium population of each state deduced (20). Details of the MD simulation procedures and Markov state model analysis can be found in the Supporting Material.

Each of the five Aβ monomer systems exhibits great structural diversity and can only be characterized in an ensemble fashion (rather than described by a handful of representative configurations). This is in accord with the notion that full-length Aβ peptides are intrinsically disordered (21,22). Using the Dictionary of Secondary Structure of Proteins program (23) to assign secondary structure, it is clear that the five Aβ monomer systems are found overall not well structured, although small β-hairpins and α-helices are observed. In Fig. 1 we plot the residue-dependent extended β propensity and α-helix propensity, in the top and bottom panels, respectively, for each Aβ monomer system. Although we are reasonably confident of the convergence behavior of the α-helix propensity, we note that the convergence of the extended β-propensity might be more challenging and demand a much longer sampling time than the current aggregate simulation time of ∼200 μs (24).

Figure 1.

Ensemble-averaged %population of β-strand (top) and α-helix (bottom) propensity for all five monomer systems. The sequence of the WT Aβ₄₂ is given on the x axis.

We observe in Fig. 1 that all five Aβ monomer systems share a rather similar residue-dependent tendency to form an extended β-structure, although minor differences are present. On the other hand, these pathogenic mutations alter the α-helix propensity quite significantly. The E22K and E22Q mutations increase the α-helix propensity in the region of residues 20–23. All four mutations (E22K, E22Q, E22G, and D23N) decrease the α-helix propensity in the region of residues 33–36.

Notably, we find that in all five systems only short stretches of α-helices are formed. That is, when a residue is involved in α-helix formation, it participates in forming mostly short helical segments (consisting of only four helical residues). To provide more insight into the changes of α-helix propensity due to the mutations, in Fig. S1 we plot the tendency of forming short α-helices along the sequence for all five systems. Each data point in Fig. S1 represents the propensity to form an α-helix of four residues in length, ending at the specific residue. For example, in the structural ensemble adopted by the WT peptide, ∼5.5% of the conformations have a short α-helix of size four, involving residues 15–18. We see from Fig. S1 that the E22K and E22Q mutations induce the formation of two short helices in residues 19–22 and 20–23. The higher α-helix propensity in this region for the E22K mutant compared to the WT was previously attributed to the elimination of the electrostatic repulsion between E22 and D23 in the WT by the mutation and the longer aliphatic chain of K22 in the mutant compared to E22 in the WT (9,22). This is consistent with the observation that the E22Q mutation also induces helix formation in this region (by eliminating the electrostatic repulsion between E22 and D23 in the WT) but to a lesser extent, possibly due to the shorter aliphatic chain of Q22 compared to K22.

In the E22G mutant, although the mutation eliminates the electrostatic repulsion between E22 and D23 in the WT peptide, glycine is known to be a helix breaker (25), leading to diminished α-helix propensity in the region around residue G22 seen in Fig. S1.

In the D23N mutant, although the mutation eliminates the electrostatic repulsion between E22 and D23 in the WT peptide, it does not induce (or rather even slightly decreases) helix formation around residue 23. This may be due to the short aliphatic chain of N23 but it is possible that the mutation induces some nonlocal effects on the peptide structure, disfavoring helix formation in this region.

It is worth noting that all four mutations (E22K, E22Q, E22G, and D23N) virtually eliminate the α-helix propensity in the region of residues 33–36. This region is rather far away from the mutation sites in sequence but its α-helix propensity is nonetheless affected. The origin of such a nonlocal effect is less straightforward to explain and further analysis will aid untangling this behavior. Nonetheless, the diminished α-helix propensity in the region of residues 33–36 appears to be a consistent feature across all four mutants.

The four mutations studied here (E22K, E22Q, E22G, and D23N) have been thought to modify the physicochemistry of the peptide and alter the oligomerization/aggregation process, leading to higher neurotoxicity. In predicting intrinsic aggregation propensities using peptide sequences, all four mutants are suggested to be more aggregation prone (26). On the other hand, kinetic studies show that only the E22K and E22Q mutants aggregate more quickly, whereas the E22G and D23N mutations result in slightly slower aggregation than WT Aβ₄₂ (4). Our simulation results suggest these pathogenic mutations have complicated effects on the monomer structure—all four mutations decrease helix propensity in residues 33–36, whereas only the E22K and E22Q mutations increase helix propensity in residues 20–23. It is interesting to note that α-helix propensity is generally thought to anticorrelate with aggregation propensity; however, recent studies have suggested an important role of α-helical intermediates in amyloid oligomerization (27–29). Our studies suggest that it would be of great value to investigate how the distinct patterns of α-helix propensity in these five systems may propagate to give rise to different oligomerization kinetics or even mechanisms. The pathogenic mutations studied here have complex effects on the oligomerization of the peptide. The characterization of the monomer structural ensembles reported here should aid understanding of such an important and complicated process.