Methionine Oxidation Changes the Mechanism of Aβ Peptide Binding to the DMPC Bilayer

Using all-atom explicit solvent replica exchange molecular dynamics simulations with solute tempering, we study the effect of methionine oxidation on Aβ10–40 peptide binding to the zwitterionic DMPC bilayer. By comparing oxidized and reduced peptides, we identified changes in the binding mechanism caused by this modification. First, Met35 oxidation unravels C-terminal helix in the bound peptides. Second, oxidation destabilizes intrapeptide interactions and expands bound peptides. We explain these outcomes by the loss of amphiphilic character of the C-terminal helix due to oxidation. Third, oxidation “polarizes” Aβ binding to the DMPC bilayer by strengthening the interactions of the C-terminus with lipids while largely releasing the rest of the peptide from bilayer. Fourth, in contrast to the wild-type peptide, oxidized Aβ induces significantly smaller bilayer thinning and drop in lipid density within the binding footprint. These observations are the consequence of mixing oxidized peptide amino acids with lipids promoted by enhanced Aβ conformational fluctuations. Fifth, methionine oxidation reduces the affinity of Aβ binding to the DMPC bilayer by disrupting favorable intrapeptide interactions upon binding, which offset the gains from better hydration. Reduced binding affinity of the oxidized Aβ may represent the molecular basis for its reduced cytotoxicity.


Fig. S1
Probability distributions P(rHr,Tr) for enthalpies Hr. Significant overlap between adjacent replica distributions is a prerequisite for efficient mixing of replicas over temperature scale.
Sampling convergence: To assess the convergence of REST simulations, we present in Fig. S2 the number of unique states Ns as a function of REST equilibrium simulation time τsim. A state (H,X) is defined by the enthalpy H, in which the bias introduced by REST is removed, and the structural probe X, where X = C, the number of intrapeptide contacts, or S2 Cl, the number of peptide-lipid contacts. Fig. S2 demonstrates that, within the timescale of our simulations, both Ns approximately level off indicating an exhaustion of new states. It is important to emphasize that saturation of new states constitutes a necessary condition for sampling convergence. The R=8 replicas in our REST simulations should exhibit a random walk over REST temperatures. To probe this expectation, we display in Fig. S3 the distribution of replicas at each REST iteration. Importantly, this figure presents a color mosaic, indicating efficient replica mixing across temperatures. Quantitatively, replica mixing can be assessed by computing the tunneling time R, the average time required for a replica to transit from the lowest to highest REST temperatures (or in reverse) [3]. Average R for all five REST trajectories is 3.2 ns, i.e., within a single REST trajectory a replica makes approximately three complete round trips through the REST temperature range.

S3
Replica mixing can be further quantified following Han and Hansmann [4]. Specifically, they defined the parameter m(T), which tracks the amount of simulation time tr a replica r spends at temperature T, as If all R=8 replicas are uniformly distributed across temperatures, m(T)=1-1/R 1/2 ≈0.65. In other words, m(T)≈0.65 is the optimum theoretical value indicating ideal replica mixing. Fig. S4 demonstrates that this is approximately true for our simulation system. Next, we examined the relaxation of energetic and structural quantities to their baseline equilibrium values. Specifically, we checked the enthalpy of the entire system H as a function of REST iteration at 330K and found that it reaches the baseline in each trajectory within the time interval not exceeding 3.8 ns. As a further test we considered the location of A peptide in the bilayer. To this end, Fig. S5 presents the positions of A and its Cterminus centers of mass ZCOM as a function of REST iteration. This figure suggests that the peptide position along the bilayer normal is well equilibrated at 330K.
Because REST strengthens solvent interactions in replicas r > 0, it is important to check that REST and traditional replica exchange molecular dynamics (REMD) produce consistent results. In our previous study [2] we have performed this test for the system of WT A10-40 peptide binding to the DMPC lipid bilayer. Specifically, the averages and the distributions of quantities probing peptide and bilayer structure were in excellent agreement, including intrapeptide contact maps, distributions of amino acids along the bilayer normal, bilayer density profiles, and SCD distributions. Therefore, we believe that REST collects unbiased conformational sampling. Finally, we discuss the role of initial structures in REST simulations. To prepare unique starting structures for each trajectory, we placed peptides in random orientation near the DMPC bilayer and performed short simulations until the peptides became bound to the bilayer (Fig. S6). These bound structures were then utilized to initiate REST. The average RMSD between these initial structures was 40.1 Å (or 44.8 Å for WT A simulations) suggesting that they are conformationally dissimilar. Protocol for WT A simulations is described elsewhere [2]. Aβ secondary structure: To facilitate the analysis of Aβ secondary structure, we present in Table S1 the helical <H(k)>, turn <T(k)>, or random coil <RC(k)> fractions within the sequence regions k=R1-R4 of MetO and WT Aβ. The table reveals that in WT the only sequence region forming stable (fraction > 0.5) secondary structure is the C-terminal R4 featuring helical state. In contrast, no MetO Aβ regions adopt helical conformation, whereas three regions (R2-R4) are populated by stable turn structure. Aβ tertiary structure: To assess Aβ tertiary structure we use the difference contact map <ΔC(i,j)> = <C(i,j)> -<C(i,j)>WT, where <C(i,j)> and <C(i,j)>WT are the contact maps for MetO and WT Aβ peptides, respectively, and i and j refer to amino acids. Table S2 lists the contacts most affected by oxidation (|<C(i,j)>| ≥ 0.35). Two-third (10 out 15) of such contacts are destabilized by oxidation. In particular, all long-range (|i -j| ≥ 5) contacts and the majority (58%) of short-range (|i -j| < 5) contacts are weakened by oxidation. Aβ-bilayer interactions: Our simulations probe the equilibrium binding of the MetO Aβ peptide to the DMPC bilayer. In Table S3, we present the probabilities for Aβ sequence regions k=R1-R4 to be localized at the bilayer surface Ps(k), inserted into the bilayer hydrophobic core below lipid phosphorus atoms Pi(k), or unbound Pu(k). The definitions of these probabilities and their analysis are given in the main text. To supplement the analysis of binding interactions between A peptide and DMPC bilayer, we considered the formation of hydrogen bonds. One may conjecture that unraveling of the C-terminal helix makes hydrogen donors and acceptors in A backbone available for interactions with DMPC lipids. To assess this possibility, we calculated the numbers of hydrogen bonds <NHB(i)> between MetO A amino acids i and lipids and then determined their difference with respect to the WT peptide, <NHB(i)> (Fig. 5b). Note that DMPC lipids can only serve as hydrogen acceptors (Fig. 1c), whereas hydrogen donors must come from A. Furthermore, in the WT and MetO A C-termini, hydrogen donors are only available in the peptide backbone. In total, there are <NHB>=4.5±0.3 hydrogen bonds between MetO A and lipids compared to 3.5±0.9 formed by the WT implying that one additional hydrogen bond is established overall between MetO A and the bilayer. Importantly, Fig. 5b implicates remarkably non-uniform changes in hydrogen bonding along A sequence with the C-terminal R4 region gaining hydrogen bonds and the other regions losing them. Indeed, the MetO C-terminus forms approximately two additional hydrogen bonds with the bilayer, whereas about one less bond occurs between MetO regions R1-R3 and the bilayer. These computations support the assumption made above that helix unraveling in the MetO C-terminus makes the peptide backbone available for hydrogen bonding with the bilayer.

S6
We further analyze Aβ-lipid interactions by computing the difference <ΔCl(i,k)> in the number of contacts formed between amino acids i and lipid structural groups k=G1-G5 caused by oxidation. Specifically, <ΔCl(i,k)>=<Cl(i,k)> -<Cl(i,k)>WT, where <Cl(i,k)> and <Cl(i,k)>WT are the peptide-lipid contact maps reporting the number of contacts between i and k for MetO and WT Aβ, respectively. The difference contact map <ΔCl(i,k)> is presented in Fig. S7, whereas Table S4 lists the contacts most affected by oxidation (|<ΔCl(i,k)>| ≥ 0.35). Consistent with Fig. 5b, this figure demonstrates that the interactions between the MetO A C-terminus R4 and all lipid structural groups k=G1-G5 are generally enhanced. This outcome is contrasted with elsewhere in the sequence, which features mostly a loss in A-lipid contacts due to methionine oxidation. Interestingly, the interaction most strengthened by oxidation as reflected in the largest positive <ΔCl(i,k)> in Table S4 occurs between oxidized Met35 and the lipid phosphate group G2. In WT Aβ, the number of contacts between Met35 and G2 is 0.02, but it increases to 0.83 upon oxidation becoming the most stable peptide-lipid group interaction. The next two interactions enhanced by oxidation in descending order are those between G1 and Gly33 or Met35 (Table S4). Aβ-lipid interactions most destabilized by oxidation are the contacts S7 between Phe20 and fatty acid tails G4 and G5. These changes reflect the expulsion of MetO Aβ R2 region from the bilayer. Importantly, out of 12 contacts with positive <ΔCl(i,k)> in Table S4 11 are associated with the C-terminal R4 region. These results are consistent with the enhanced binding interactions established between MetO Aβ C-terminus and the bilayer. Interestingly, stronger binding interactions are particularly evident for choline (G1), phosphorous (G2), and glycerol (G3) groups, which contribute 82% of the most affected contacts in Table S4.   Fig. S8b, we found that <> averaged over distant lipids is 149±0°. In the proximal region around the MetO peptide, the tilt angle is marginally smaller than near WT A (144±1° vs 145±0°, respectively). Similar result follows if we consider the centers of proximal regions near MetO and WT peptides (140±1° against 141±3°). Thus, the oxidized A peptide induces marginally stronger structural perturbation in lipid fatty acid tails compared to the WT. This finding appears at variance with the much weaker impact of MetO A binding on the DMPC bilayer structure compared to its WT counterpart displayed in Fig. 6. In Discussion we rationalize these conflicting observations. (a) (b) Fig. S8 (a) The lipid carbon-deuterium order parameter -<SCD(i)> computed for each carbon i in sn-2 fatty acid tails. Vertical bars represent the standard error about the mean calculated from n=5 trajectories. (b) Probability distributions P(γ) for lipid sn-2 fatty acid tails to be tilted at an angle γ from the bilayer normal. In both panels data for MetO and WT Aβ are in black and red, respectively. The solid and dashed lines refer to the proximal and distant lipids. The panel reveals that binding of MetO A results in marginally stronger structural disordering of the DMPC fatty acid tails than of the WT.
Computation of hydrophobic moment: To compute hydrophobic moment, we created an idealized -helix formed by all 12 C-terminal A residues. To this end, the positions of C atoms of amino acids were placed on the circle with the 1 Å radius. Each successive C atom is rotated =100° relative to the previous one. Following [5] a hydrophobic , Hk is a hydrophobicity of amino acid k [6], and ⃗ and ⃗ are unit vectors. Their initial orientation is arbitrary and was chosen to direct vector ⃗ as in Fig. S9.  Fluctuations in A structure: To assess rigidity of Aβ structures, we computed the standard deviations in peptide backbone dihedral angles φ(i) and ψ(i) for each amino acid i. Fig.  S11 plots the differences in the dihedral angle standard deviations, Δδφ(i) and Δδψ(i), between MetO and WT Aβ. Visual inspection suggests that MetO Aβ experiences higher fluctuations than the WT. Detailed analysis is given in the main text. Free energy of A binding to the DMPC bilayer: To compute the free energy of binding of A peptide to the DMPC bilayer, we used equilibrium replica exchange conformational sampling and MM-GBSA approach. The free energy of a system G is S11 given by Eq. (1). The description of individual terms in G is given in the main text, whereas their values are listed in Table S5. It is important to check the consistency of computation of solvation free energies Gsolv,p and Gsolv,ap. The polar solvation free energy Gsolv,p is calculated using Generalized Born Implicit Solvent model, which offers two sets of parameters for estimating effective Born radius, denoted as OBC I and OBC II [7]. In addition, we tested the effect of changing the nonpolar surface tension coefficient  from 1=0.005 to 2=0.02 kcal/mol/AA 2 [8]. We found that depending on the specific combination of parameters, the difference in the free energy of binding between MetO and WT peptides Gb (Eq. (3)) is 17 kcal/mol (OBC II and 1, used in Table S5), 22 kcal/mol (OBC II and 2), 17 kcal/mol (OBC I and 1), and 23 kcal/mol (OBC I and 2). These computations indicate that irrespective of specific choice of solvation parameters, oxidation decreases the affinity of A binding to the DMPC bilayer.
Our computations of Gb revealed a vanishing entropic contribution -TS≈0 kcal/mol. This result suggests that oxidation induces similar changes in entropy in the MetO peptide bound to the bilayer and solvated in lipid-free water.