Role of the familial Dutch mutation E22Q in the folding and aggregation of the 15–28 fragment of the Alzheimer amyloid-β protein

Abstract

Amyloid fibrils, large ordered aggregates of amyloid β proteins (Aβ), are clinical hallmarks of Alzheimer's disease (AD). The aggregation properties of amyloid β proteins can be strongly affected by single-point mutations at positions 22 and 23. The Dutch mutation involves a substitution at position 22 (E22Q) and leads to increased deposition rates of the AβE22Q peptide onto preseeded fibrils. We investigate the effect of the E22Q mutation on two key regions involved in the folding and aggregation of the Aβ peptide through replica exchange molecular dynamics simulations of the 15–28 fragment of the Aβ peptide. The Aβ15–28 peptide encompasses the 22–28 region that constitutes the most structured part of the Aβ peptide (the E22–K28 bend), as well as the central hydrophobic cluster (CHC) (segment 17–21), the primary docking site for Aβ monomers depositing onto fibrils. Our simulations show that the 22–28 bend is preserved in the Aβ(15–28) peptide and that the CHC, which is mostly unstructured, interacts with this bend region. The E22Q mutation does not affect the structure of the bend but weakens the interactions between the CHC and the bend. This leads to an increased population of β-structure in the CHC. Our analysis of the fibril elongation reaction reveals that the CHC adopts a β-strand conformation in the transition state ensemble, and that the E22Q mutation increases aggregation rates by lowering the barrier for Aβ monomer deposition onto a fibril. Thermodynamic signatures of this enhanced fibrillization process from our simulations are in good agreement with experimental observations.

The presence of amyloid fibrils in the brain is a clinical hallmark of Alzheimer‘s disease (AD). The fibrils consist of large ordered aggregates of amyloid-β (Aβ) peptides, proteolytic by-products of the enzymatic cleavage of the Alzheimer amyloid precursor protein (APP). AD can be sporadic (occurring in elderly patients and characterized by a slow progression) or familial (a hereditary form characterized by early onset of the disease and aggravated severity). The majority of familial forms of AD involve single-point mutations in the 22–23 segment of the Aβ peptide. The focus of this work is on the Dutch form of AD, which involves a mutation encoded by a point substitution G to C at codon 693 of the amyloid precursor protein (APP). The result is the production of an Aβ peptide in which residue E22 is mutated to Q (E22Q mutation). Patients who have this mutation are at risk of developing cerebral amyloid angiopathy that typically leads to cerebral hemorrhage and stroke. Many in vitro experiments show that the AβE22Q peptide aggregates much more readily than its wild-type (WT) counterpart (1–3). As a free monomer in solution, the WT Aβ peptide is for the most part unstructured with residual structure observed locally in a few regions of the primary sequence (4). The Aβ peptide can populate a variety of oligomeric species, some of which may be toxic (such as transmembrane pores) (5), on route to the fibrillar state. Once bound to the fibril, the peptide adopts a β-sheet conformation (6), implying that fibrillization is accompanied by a major structural reorganization. In the case of the E22Q mutation, experiments by Maggio et al. (7) indicate that the rate of Aβ monomer deposition onto fibrils is enhanced compared with the WT. The refolding of monomeric Aβ into a conformation commensurate with the fibril presents a free energy barrier to fibril growth. The aim of the present work is twofold: (i) to investigate how the E22Q mutation affects monomer Aβ conformations and (ii) to investigate how the E22Q mutation affects the deposition reaction of monomers onto fibrils.

In aim i, we focus on the effect of the E22Q mutation on two critical regions implicated in Aβ monomer folding and aggregation, the E22–K28 region and the L17–A21 central hydrophobic cluster (CHC). NMR and molecular dynamics simulations have shown that the 22–28 region adopts a bend structure in the context of the proteolytically resistant 21–30 fragment of Aβ(8–11) and is likely the most structured region of the full-length Aβ40 and Aβ42 peptides (8). Further simulations in our group found this same bend region in fragments Aβ12–28 (12) and Aβ10–35 (13). The CHC, on the other hand, appears to serve as a recognition site for monomer binding to the fibril. Mutations in this segment (such as F19T) completely abrogate fibril growth (14). We consider two model peptides: fragments 21–30 and 15–28 of the Aβ peptide. The former will enable us to study the effect of the E22Q mutation on the structured bend motif while the latter will let us probe the effects of this mutation on the structure of the CHC region. Although we are not considering fragments that include residues distant in sequence space from residue 22, we nonetheless expect to capture the essential effect of the E22Q mutation. Indeed, experiments on Aβ12–28 (15) and Aβ10–35 (7) have shown that the E22Q mutation only induces local changes to the folding properties of the peptide that do not extend beyond neighboring residues of E22. Our simulations involve explicit solvent and the replica exchange protocol to achieve a thorough sampling of conformational space.

In aim ii, we investigate how the E22Q mutation affects deposition rates onto preexisting amyloid fibrils by identifying the transition state (TS) ensemble for the fibril elongation reaction and calculating the activation free energy for monomer deposition onto preexisting fibrils for the WT and mutant cases.

Results

The E22Q Mutation Does Not Alter the Folded Bend in Aβ21–30.

The structure of the 21–30 region was elucidated in our earlier replica exchange molecular dynamics simulations (9). The predominant cluster found in these studies (44% population) is one in which a bend is present, spanning residues 22–28. A bend or loop structure in the 22–28 segment was reported in simulations by other researchers (11, 10), and it was also suggested to exist in fibrils of Aβ (16, 17). The bend is stabilized by a network of hydrogen bonds between the D23 O^δ atoms and the amide hydrogens of the backbone. Mutation E22Q hardly affects the backbone structure of the peptide, with an RMSD of 0.37 Å to the WT structure. The hydrogen bond network that stabilizes the bend remains intact. The population of the predominant cluster drops slightly to 33%. Fig. 1 shows the overlap of the 21–30WT and the 21–30-E22Q structures, and Fig. 2 shows the contact maps. These maps were computed assuming that a contact between two residues is formed if the minimal distance between any two atoms of these residues drops below 6 Å. For glycines, their α hydrogens were treated as side chains. It is seen that, except for a few sparsely populated contacts, the two maps are identical.

Fig. 1.

Overlay of the central structures observed for Aβ21–30E22Q (transparent) and Aβ21–30WT. Residues D23–N27 are shown in stick representation; residues E22, Q22, and K28 are shown as lines. The backbone of the peptide is highlighted in gray. Hydrogen bonds are shown as dotted blue lines.

Fig. 2.

Contact maps for WT Aβ21–30 (bottom right) and the E22Q mutant (top left) obtained in the present simulations. The values correspond to the probability of having at least one pair of atoms in a given pair of side chains within 6 Å of one another.

Folding of Aβ15–28 and Aβ15–28E22Q Is Dominated by the E22–K28 Bend Motif.

A representative structure from the most populated cluster of the WT Aβ15–28 (constituting 20% of the all sampled states) is shown in Fig. 3. A well formed bend at E22–K28, stabilized by the interactions of the side chain of D23 and amide hydrogens of the neighboring residues G25–N27, can be readily seen. A clustering analysis over all consecutive five-residue segments present in Aβ15–28 reveals that the bend region constitutes the most structured part of the peptide (data not shown). Using RMS deviations over C^α atoms of this region (E²²DVGSNK²⁸), we find that the bend conformation is ≈92% populated. In the case of the mutant Aβ15–28E22Q, the 22–28 segment remains highly structured, although its population drops to 68%. A comparative analysis reveals that the 22–28 bends in the WT and E22Q mutant of Aβ15–28 are identical in structure. In fact, they are also identical in structure to the E22–K28 segment found in Aβ21–30 and, as discussed above and shown in Fig. 1, in Aβ21–30E22Q.

Fig. 3.

All-atom representation of the most representative conformation of Aβ15–28 sampled in the present simulations. A large hydrophobic patch exposed to the solvent by residues of the central hydrophobic cluster is shown as surface.

Side-chain contacts for Aβ15–28WT and Aβ15–28E22Q are shown in Fig. 4. These contact maps clearly show that folding in Aβ15–28WT occurs at approximately the E22–K28 segment. The highest-intensity contacts (>80% populated) are observed in this segment, between side chains of D23 and G25, S26, N27, and K28. These same contacts are also dominant in the map of Aβ15–28E22Q, although their intensity is somewhat reduced. The most noticeable effect of the E22Q mutation is a weakening of the 22–28 contact, arising from the destabilization of the E22–K28 salt bridge upon mutation. In the WT sequence, the 22–28 contact is >80% populated, whereas in the mutant, its population drops to ≈50%. We note that although the salt-bridge plays a stabilizing role on the folded structure, the extent of this stabilization is small. Indeed, destabilization of the salt bridge results in a partial depopulation of the 22–28 contact, not its complete cancellation.

Fig. 4.

Probability map of interresidue contacts computed in the present simulations for Aβ15–28WT (lower right triangle) and Aβ15–28E22Q (upper left triangle).

The E22Q Mutation Leads to a Weakening of the Interactions Between the CHC and the 22–28 Bend.

A structural analysis of the conformational space sampled by Aβ15–28WT in our simulations indicates that unlike the 22–28 region, the central hydrophobic segment 17–21 does not populate a unique conformation. Rather, three to four ensembles of structures emerge in which the CHC adopts coil and bend conformations [according to DSSP classification (18)], which have very little in common with the β-strand structure seen in this region in the fibril models of Tycko (6). The CHC is significantly less structured than the E22–K28 bend region, in both Aβ15–28WT and Aβ15–28E22Q. Multiple clusters with population levels of ≈20% are observed in the CHC region of Aβ15–28WT and with populations of ≈10% in Aβ15–28E22Q.

As can be seen in Fig. 3, the central hydrophobic segment in Aβ15–28WT is aligned with the bend structure. Interactions with the bend motif shield the CHC segment from exposure to water on one side. The other side of this segment, however, is fully exposed to the solvent. A similar pattern (partially exposed CHC interacting with the bend structure) was observed in the Aβ1–40 simulation of Garcia and coworkers (19). The conformations of CHC segment shield the bend motif from water on one side, thus increasing its population from 44% in Aβ21–30 (with no CHC present) to 92% in Aβ15–28. It appears, therefore, that the CHC and the bend segments engage in mutually beneficial interactions that increase the degree of their structuring.

In Aβ15–28WT, E22 has some interaction with the residues of the CHC, with the intensity of E22–V18 and E22–F20 contacts at ≈50%. These interactions are reduced to ≈30% in Aβ15–28E22Q, leading to the general weakening of the interactions between the CHC and E22–K28 bend region. A direct manifestation of this is the complete disappearance of the hydrophobic F19–V24 contact in Aβ15–28E22Q, which is populated ≈50% of the time in the WT sequence. It is apparent that the E22Q mutation affects the residual structure present in the CHC segment. The most notable effect is a significant increase in the β-content at residues V18 and F19, from ≈10% in WT sequence to ≈20% in the E22Q mutant.

Segments 14–20 and 30–36 Adopt a β-Strand in the Transition State for Fibril Deposition.

As outlined in supporting information (SI) Appendix, possible TS conformations were identified based on the model fibrils of Tycko (6). The TS conformations observed in our trajectories are shown in Fig. 5.

Fig. 5.

Transition state conformations of monomer deposition onto fibril identified in the present work. Two structural motifs, β-strands at positions H14–F20 and A30–V36, are shared by all observed conformations.

Although, on the whole, the TS conformations represent a strongly heterogeneous structural ensemble, they nonetheless retain several features of the original Tycko model. Clustering over all five-residue segments within Aβ9–40 reveals two regions of high structuring propensity, residues 14–20 and 30–36. These regions correspond to the location of β-strand structures in the fibril model, linked by the A21–G29 loop. The loop, on the other hand, is completely disordered in TS structures. Other disordered regions correspond to the carboxyl (residues 9–13) and amino (residues 37–40) termini. RMS deviations of the TS conformations (over C^α atoms) from their initial values are plotted in SI Fig. 10 of SI Appendix. Comparison with the pretransition clusters (observed at T = 300 K) and posttransition structures (shown in SI Fig. 10 of SI Appendix) confirms our initial inference that conformations representing the TS ensemble have increased mobility in the central loop region while keeping intact the two β-strands of the fibril.

The E22Q Mutation Leads to a Decreased Free Energy Barrier for Monomer Deposition onto the Fibril.

Our dissociation simulations of a model Aβ fibril (shown in SI Appendix) indicate that the contacts involving residues L17–F20 are the last to break under the harshest dissociative conditions considered. These simulations are consistent with the idea that the CHC residues, which are exposed to the solvent (see Fig. 3) in the monomer, serve as the docking site for Aβ molecule depositing onto a fibril. Once docked, the Aβ molecules need to undergo an additional conformational transition to become part of a fibril. The activation free energy for deposition of the monomer onto the fibril, ΔG^†, is given by two terms: refolding of L17–F20 segment into a β-strand-like structure and ordering of H14–K16 and A30–V36 segments (which are disordered in the monomer) (4). A cartoon that illustrates our model of monomer deposition onto a fibril is shown in Fig. 6.

Fig. 6.

Cartoon illustrating how monomers of Aβ40 deposit onto amyloid fibrils. The time scale of this process is governed by the free energy difference between transition states and docked (monomeric) states, or the activation free energy ΔG^†. The activation free energy in Aβ40 with E22Q mutation is lowered compared with the WT sequence, ΔΔG^† > 0. The defining characteristics of transition state conformations are β-strands in segments H14–F20 and A30–V36. To reach these conformations, Aβ40 has to refold its central hydrophobic cluster segment L17–F20.

Because the A30–V36 and H14–K16 segments populate similar conformations in both soluble Aβ40E22Q and Aβ40WT (7), the free energy cost of refolding them into the β-strand conformations found in the fibril is the same for both of these peptides. Hence, the effect of the E22 mutations on the deposition free energy barrier in Aβ40 is primarily through the refolding of the L17–F20 segment. To estimate the free energy cost of the refolding of L17–F20, we computed the free energy profile as a function of RMSD deviation in Aβ15–28WT and Aβ15–28E22Q from the TS conformations over the positions of C^α atoms. Our results are shown in Fig. 7. Fig. 7b is for the WT sequence, and it shows that the majority of the observed conformations are ≈0.5 Å away from the β-strand structure of the TS. We estimate that the free energy cost of reaching that structure is ΔG^† = 1.3 kcal/mol. By examining free energy profile for AβE22Q (data not shown), we found that its free energy barrier is lower, at ΔG^† = 1.1 kcal/mol. Assuming that the deposition rate is given by a TS expression k ∼ e^−ΔG†/kT, we find that the ratio of deposition rate k_m in mutated Aβ to its WT sequence counterpart k is k_m/k = e^ΔΔG†/kT, where ΔΔG^† = ΔG^† − ΔG_m^† is the difference in activation free energy and T is the temperature. At temperature T = 277 K, this expression predicts that Aβ15–28E22Q should deposit onto fibrils ≈1.4 times faster than Aβ15–28. For comparison, the measured speed-up in Aβ10–35 fragment due to E22Q mutation is 2.15 times (7). We see, therefore, that the effects of E22 substitution on folding properties of the CHC fragment reasonably well capture the physics underlying the deposition rate enhancement in Aβ40 peptide. An additional validation of this result comes from the decomposition of the activation free energy difference ΔΔG^† into enthalpic and entropic contributions. We present in Fig. 7a the temperature dependence of ΔG^† observed in our simulations for Aβ15–28WT and Aβ15–28E22Q. These data are shown along with the linear approximation ΔG^† = ΔH^† − TΔS^†, where ΔH^† is the enthalpy associated with the activation process and ΔS^† is its entropy. Fitting of the simulation data to this approximation produces values ΔH^† = 1.2 ± 0.4 kcal/mol, ΔS^†/k = −3·10⁻⁴ ± 1·10⁻³ kcal/mol/K for WT and ΔH^† = 0.7 ± 0.2 kcal/mol, ΔS^†/k = −1·10⁻³ ± 7·10⁻⁴ kcal/mol/K for the E22Q mutant. It is immediately seen that the activation free energy to monomer deposition is dominated by enthalpy. This is in agreement with the recent experimental observation that enthalpy makes a significant contribution to ΔG^† (20). Activation enthalpy of both the WT sequence and the mutant are within statistical errors of each other, suggesting that the effect of E22Q mutation on deposition rate is mostly entropic. Because of large statistical noise, the fit for WT peptide in Fig. 7a is not as good as for the mutant, producing a larger error in the estimate of ΔS^†. Nevertheless, the different entropic contributions to the activation free energy in WT and E22Q peptides are clearly seen in this figure as different slopes of their ΔG^† curves.

Fig. 7.

Deposition free energy. (a) Temperature dependence of the activation free energy for monomer deposition in Aβ15–28 and Aβ15–28E22Q as determined in the present simulations. (b) Free energy profile for Aβ15–28WT as a function of the RMS deviation from the transition state conformations over C^α atoms of L17–F20 segment.

Discussion

Several types of familial forms of AD have been identified in addition to the more common sporadic form of this disease. Familial forms of AD result from a single point mutation in Aβ peptides and are characterized by early onset and aggravated severity. In this article, we focus on the Dutch AD type, which is associated with the E22Q substitution (21). Numerous in vitro experiments have shown that Aβ peptides with the Dutch mutation aggregate more readily than their WT Aβ counterparts, in addition to exhibiting increased cytotoxic activity (1, 3, 22). These observations have led to the hypothesis that increased propensity for aggregation in AβE22Q underlies the main mechanism of pathology of the Dutch form of AD (3, 23).

De novo formation of amyloid fibrils is a multistage process that includes two obligatory steps: nucleation and growth (24). During the first step (the slower of the two), a fibril of minimal dimensions (the “nucleus”) gets created; fibril nucleation can be either spontaneous or assisted (through, for instance, interaction with other intracellular proteins). In the second step, the newly created fibrils extend, primarily through addition of Aβ monomers at fibril ends, although other mechanisms have been reported (25). Because nucleation is such a slow process, the fibril growth stage has been identified as the relevant element for explaining the increased amyloidogenicity of the Dutch form of AD (7, 26). The deposition of a monomer onto a fibril consists of two consecutive steps: an initial “docking” in which monomers bind weakly and reversibly to the open end of the fibril, and a “locking” step involving the structural transition of the monomer into a conformation compatible with the structure of the fibril (27–29). The locking process was seen as the rate-limiting step of the entire reaction (27, 29). Time-resolved studies of Maggio et al. (7) have shown that monomer deposition onto amyloid plaques extracted from AD brains proceeds two to four times faster in the Dutch-type mutant Aβ40 than in the case of the WT sequence (27).

Refolding of a soluble conformation of Aβ into a conformation found in amyloid fibrils is the first and critical step on the pathway that leads to fibril growth. In the past few years, a number of hypotheses have been put forth to explain how the E22Q mutation affects the refolding process of Aβ in an attempt to rationalize the effect of this mutation on the rate of fibril formation (30–32). These hypotheses involve (a) a proposed reduction in the α-helical structure in the 10–24 N-terminal domain of Aβ accompanying the E22Q mutation (30), (b) an overall increase in β-sheet structure in the E22Q Aβ mutant (31), and (c) unfolding of the CHC upon E22Q mutation (33). Hypotheses a and b are not particularly compelling because previous FTIR (34) and more recent NMR (7) studies show little difference in overall secondary structure between the WT and mutated peptide and no dramatic conformational change upon mutation. The third hypothesis is based on very short (1 ns) simulations (32) initiated from the Aβ10–35 NMR structure of Lee et al. (33). These simulations show that the CHC, the most structured region in the proposed NMR model, partially unfolds as a result of the E22Q mutation. More recent experiments (8) and simulations in our group (9, 12, 13) and others (19, 35) on fragments and on the full-length Aβ peptide, however, indicate that the CHC is not the most structured element of the Aβ peptide. Rather, a bend involving residues E22–K28 emerges as the most structured part of the peptide.

The simulations presented here lead to a model for Aβ monomer deposition onto fibrils (shown in Fig. 6) that suggests a mechanism for the enhanced aggregation rates seen in the E22Q mutant. Our replica exchange molecular dynamics simulations show that the E22Q mutation does not alter the structure of the 22–28 bend, but rather, that this mutation leads to a weakening of the interactions between the CHC and this bend. In the WT peptide, our simulations show that the CHC does not fold into a unique structure, but to an ensemble of three to four structures, all aligned with the bend motif and all populated no more than 20% of the time. The mutation enables the CHC to sample conformational space more readily and adopt a structure commensurate with the fibril. Simply reducing the amount of α-structure in the Aβ peptide (hypothesis a) or a random disordering of the CHC (hypothesis c) will not lead to increased aggregation rates. For enhanced aggregation rates, it is necessary for the peptide to adopt a conformation that bears close similarity to the TS structures for fibril deposition. In our simulations, we see that the E22Q mutation does precisely that: the E22Q mutation induces changes in the CHC away from random coils and to β-strand-like structures seen in the TS ensemble. This structural shift upon mutation translates into a decreased free energy barrier to monomer deposition and, as a consequence, leads to an increased aggregation rate. In agreement with experiments (7), we find that the entropic contribution dominates the difference in the activation free energy barrier between Aβ15–28WT and Aβ15–28E22Q. The origin of this entropic effect can be due to the conformational entropy of the Aβ15–28, indicating that there are more TS structures in the E22Q mutant than in the WT sequence. Alternately, the additional entropy might arise from solvation differences in E22 and Q22 residues due to their different hydrophobicity. The exact mechanism of entropy change and its relation to other theories of enhanced aggregation such as those proposed for polyglutamine (36) will be determined in a separate study.

Interestingly, experiments show that the deposition of the longer and more cytotoxic variant, Aβ42-E22Q, is not significantly faster than that of the Aβ42 WT counterpart, suggesting that these two variants may play different roles in familial AD. A compelling feature of our model is its ability to explain why monomer deposition proceeds much faster in Aβ42 than in Aβ40 (27). Unlike Aβ40, monomers of Aβ42 contain ordered β-strand structures in their A30–V36 segment (4). Hence, for Aβ40, there will be an additional free energy cost for ordering this segment. [We assume here that Aβ40 and Aβ42 have the same deposition TS ensemble, which is likely given the similarity of their fibril structures (37).] It follows that ΔG^† is lower in Aβ42 than in Aβ40, leading to a higher deposition rate.

In addition to the Dutch mutation E22Q discussed in the present work, a number of other mutations at position 22 (including the Italian E22K and Arctic E22G mutants) and at position 23 (the Iowa mutant D23N) are known to lead to familial types of AD. Several in vitro experiments on Aβ40 with these mutations have reported enhanced aggregation rates (3, 22). Whether these mutations affect aggregation in the same manner as the E22Q mutation remains at the present time an open question.

Model and Methods

The all-atom OPLS/AA force field (38) in conjunction with the explicit TIP3P model of water (39) were used to model the solvated amyloid β peptide system. All simulations were performed using the GROMACS software set (40, 41). Covalent bonds of the water molecules were held constant by SETTLE algorithm (42). The bonds involving hydrogens of the peptide were constrained according to LINCS protocol (43). The integration time step was 2 fs. Nonbonded Lennard–Jones interactions were tapered starting at 8 Å and extending to 10 Å cut-off. Neighbor lists for the nonbonded interactions were updated every 10 simulation steps. Electrostatic interactions were included using the particle-mesh Ewald approach (44).

Three sets of simulations were performed. First, structural characterization of the Aβ15–28 (sequence: QKLVFFAEDVGSNK) and its E22Q mutant were carried out. To improve conformational sampling, the constant-pressure replica-exchange (REMD) algorithm was used (45). The pressure was controlled using Parrinello–Rahman algorithm (46) A total of 60 replicas of the original system were considered, at temperatures exponentially spaced between 277 and 600 K. Normal pressure of 1 atm was applied up to T = 384 K, after which the pressure was exponentially increased to 2,500 atm to avoid blowing -up of the simulation box. These high temperatures are not relevant for the present analysis. Exchanges of replicas at adjacent temperatures were attempted every 250 simulation steps. This lead to replica exchange probabilities of 19% and higher. The simulations were started from a random extended conformation, same for all replicas, and equilibrated for 5 ns. Subsequent equilibrium sampling runs were performed over 28 ns. To check the convergence of our data, an independent simulation of the E22Q mutant of the same length was performed. It produced the same results as the original simulation. Both Aβ15–28WT and E22Q mutant were acetylated and amidated at the amino and carboxyl terminus, respectively, and solvated in a cubic box with length of 4.5 nm containing 3,000 water molecules. To neutralize the total change in Aβ15–28E22Q, one chlorine ion was added to the system. Similar procedures were used in the second set of simulations, to characterize folding of Aβ21–30E22Q. Forty replicas, all at 1 atm, were used for temperatures ranging from 277 to 600 K in a simulation box with a length of 4 nm. Two trajectories were generated, 25 ns each.

In the third set, constant-temperature dissociation simulations of a fibril model (6) of Aβ9–40 were carried out. The Amino acid sequence of this peptide was GYEVHHQKLVFFAEDVGSNKGAIIGLMVGGVV. To neutralize charges at carboxyl and amino termini, acetyl and amide groups were used, respectively. The total charge of the peptide was −1 at pH 7. Two chains solvated in a rectangular box with dimensions 7.5 × 5.0 × 5.0 nm and containing 5,866 water molecules were considered. The charge of this system, −2, was neutralized by the addition of two sodium ions. Ten independent trajectories were run for 5 ns, using the Nosé–Hoover algorithm to keep the temperature constant. To model different dissociation conditions, simulations were performed at 300 and 500 K.