Lipid oxidation controls peptide self-assembly near membranes through a surface attraction mechanism

The self-assembly of peptides into supramolecular structures has been linked to neurodegenerative diseases but has also been observed in functional roles. Peptides are physiologically exposed to crowded environments of biomacromolecules, and particularly cellular membrane lipids. Previous research has shown that membranes can both accelerate and inhibit peptide self-assembly. Here, we studied the impact of membrane models that mimic cellular oxidative stress and compared this to mammalian and bacterial membranes. Using molecular dynamics simulations and experiments, we propose a model that explains how changes in peptide-membrane binding, electrostatics, and peptide secondary structure stabilization determine the nature of peptide self-assembly. We explored the influence of zwitterionic (POPC), anionic (POPG) and oxidized (PazePC) phospholipids, as well as cholesterol, and mixtures thereof, on the self-assembly kinetics of the amyloid β (1–40) peptide (Aβ40), linked to Alzheimer's disease, and the amyloid-forming antimicrobial peptide uperin 3.5 (U3.5). We show that the presence of an oxidized lipid had similar effects on peptide self-assembly as the bacterial mimetic membrane. While Aβ40 fibril formation was accelerated, U3.5 aggregation was inhibited by the same lipids at the same peptide-to-lipid ratio. We attribute these findings and peptide-specific effects to differences in peptide-membrane adsorption with U3.5 being more strongly bound to the membrane surface and stabilized in an α-helical conformation compared to Aβ40. Different peptide-to-lipid ratios resulted in different effects. We found that electrostatic interactions are a primary driving force for peptide-membrane interaction, enabling us to propose a model for predicting how cellular changes might impact peptide self-assembly in vivo.

S3 sulfoxide (DMSO, ≥99.9%, Merck, Darmstadt, Germany) to obtain a 1 mM ThT stock solution that was stored at -20 °C and protected from light. The peptides Aβ40 (100 μM) and U3.5 Experiments were performed at least in triplicate at 37°C, and the microplate was agitated for 40 seconds before each measurement cycle (5 minutes) using double orbital shaking (300 rpm).
ThT data for Aβ40 showed higher variations between repetitions. Data were averaged, normalized to a maximum fluorescence of 1 (except in cases with inhibition of peptide aggregation) and plotted in Origin 2022 (OriginLab Corp., Northampton, MA).
Samples were mixed before each measurement by inverting the cuvettes multiple times for 30 s. Far-UV CD spectra were recorded between 260 and 195 nm at 37°C using a J-815 CD spectropolarimeter (Jasco Corp., Tokyo, Japan) at standard sensitivity as previously reported (DIT, 1 s; bandwidth, 1 nm; data pitch, 0.5 nm; continuous; scanning speed, 50 nm/min; five scans). 2 The buffer contribution was subtracted for each experiment, and experiments were repeated at least in duplicate. Representative data were plotted in Origin 2022 (OriginLab Corp., Northampton, MA). The mean residue molar ellipticities (MRE, Θmolar,λ) were determined from the measured ellipticities Θλ, the peptide concentrations c, the cuvette path lengths l, and the number of residues n, i.e., 39 peptide bonds for Aβ40 and 16 peptide bonds for U3.5). molar, = (millidegrees) 10 6 (micromolar) (millimetres) The secondary structure content for the peptides was calculated from the measured CD spectra using the BeStSel webserver (https://bestsel.elte.hu) (α-helix, β-strand, turns, and other). 3

Quartz Crystal Microbalance (QCM) Measurements
Silicon dioxide (SiO2)-coated quartz crystals with a fundamental frequency of 5 MHz (Q-Sense, Biolin Scientific, Gothenburg, Sweden) were used as sensors for the QCM experiments. The sensors were cleaned using the following protocol: After 10 min in a 2% Hellmanex II cleaning solution (Hellma, Mulheim, Germany), the sensors were rinsed with water, followed by isopropanol (>99.5%, Merck, Germany) and dried under a gentle nitrogen gas stream. The sensors were finally treated in a UV Ozone ProCleaner (BioForce Nanosciences, Virginia Beach, VA) for 20 min.
QCM measurements were performed using a Q-Sense E4 instrument (Biolin Scientific, Gothenburg, Sweden) consisting of four flow cells at 22±0.05 °C at least in triplicate. Initially, ultrapure water, followed by PBS buffer, was introduced into the measurement flow cells at S5 mM (for POPC) / 250 mM (for POPC-POPG) sodium chloride buffer) were deposited onto the silicon dioxide sensors at a flow rate of 50 μL/min until they ruptured and formed a lipid bilayer structure. This was visible from a sudden increase in frequency, followed by stable frequency values. 4,5 An increase in frequency (Δf) is correlated with a decrease in mass (Δm) through the Sauerbrey equation (Δf = -C · Δm). 6 The measurement cells were rinsed with phosphate buffer at 200 μL/min to remove any lipids that were not bound to the sensors, followed by the introduction of peptide sample solution at 25 µM at a flow rate of 50 μL/min for 15 min. Frequency changes were followed for another 45 min without flow under steady conditions, followed by a final rinse with PBS buffer for at least 10 min.

Molecular Dynamics (MD) Simulations
MD simulations were performed at 303.15 K using the GROMACS 4.5.7 software package. [7][8][9][10] Model membranes were simulated with lipid compositions consistent with experimental conditions to understand peptide-membrane interactions at the molecular detail. The GROMOS 54A7 united-atom force field was used to describe the peptides, water and ions. 11 All lipid force fields were used as previously reported in the literature. 12,13 POPC molecules were parametrized based on the Berger force field 14 with double bond correction introduced by Bachar et al. 15,16 The cholesterol force field is GROMOS based. 17 The molecule types CH2/CH3 in the cholesterol force field were changed to avoid overcondensation of the bilayer as previously implemented. 12,18 The force field for deprotonated PazePC was obtained from Khandelia and Mouritsen 19 who derived it from the POPC force field by shortening the oleoyl chain at the double bond and replacing it with a carboxyl group, correcting geometry and using partial charges from amino acids. The force field for protonated PazePC was adapted from Khandelia and Mouritsen 19 by adding hydrogen to the deprotonated PazePC together with ad hoc partial charges to obtain a total charge of zero, as reported by Ferreira et al. 13 POPG was described using a Berger based force field. 20 The force field parameters for POPC,

S6
cholesterol and PazePC were received from Ollila et al. [21][22][23][24] and for POPG from the MemBuilder II webserver. 25 Each lipid bilayer consisted of 128 lipid molecules, i.e., two layers of each 8x8 (64) lipid molecules. The structure files of the pure POPC bilayer and the POPC-PazePC bilayer (7:3; 90 POPC and 38 PazePC molecules) were obtained from Ollila et al. 21,23,24 The POPC-cholesterol bilayer structure (4:1; 102 POPC and 26 cholesterol molecules) was obtained from Ollila et al. 22 and manually adapted for the correct amount of cholesterol (8 POPC molecules were removed and manually replaced by cholesterol molecules). The POPC-POPG bilayer structure (4:1; 102 POPC and 26 POPG molecules) was generated using MemBuilder II. 25 The Aβ40 peptide was used with both a random coil (PDB Utilities Server) 26 and an α-helical (adapted from PDB: 1IYT) 27 starting structure from the Protein Data Bank (PDB), while the U3.5 peptide was studied with a random coil and an α-helical starting structure, as obtained in nuclear magnetic resonance (NMR) experiments in sodium dodecyl sulfate (SDS) micelles. 28 The partially folded 310 helix structure of Aβ40 (PDB: 2LFM) could have been used alternatively. 29 N-and C-termini of the Aβ40 peptide were charged, whereas the C-terminus of U3.5 was uncharged (amide modified).
The following simulation parameters were used for all membrane simulations: Periodic boundary conditions were applied. The Particle Mesh Ewald (PME) method with a grid of 0.12 nm, a fourth order spline interpolation and a Coulomb cut-off at 1.0 nm was used to describe electrostatic interactions 30,31 and a Lennard-Jones cut-off distance of 1.0 nm was used to describe van der Waals interactions. The neighborlist was updated every tenth step with a time step of 2 fs. Centre of mass motion was removed for the system at every step. All bonds were constrained to their equilibrium values using the LINCS algorithm. 32 Explicit water (Simple Point Charge, SPC) 33 was constrained using the SETTLE algorithm. 34 The temperature was coupled separately for lipids and peptide/water/ions to 303.13 K using the velocity-rescale method with a coupling constant of 0.1 ps -1 . 35 The pressure was semiisotropically coupled to 1 bar with the Berendsen barostat. 36 Prior to adding peptide, all lipid bilayers were equilibrated for at least 50 ns. One or five peptide molecules were randomly positioned outside the lipid bilayer and solvated with about 5700-7700 water molecules. 150 mM NaCl was added as salt and to electroneutralize the systems (see Figure S1 and Table S1 for an overview). All simulations were run for 100 ns in triplicate. Further, each peptide was simulated in water without any lipid present for 100 ns S7 in triplicate, serving as a reference. The parameters were used as for the membrane simulations, but the temperature was coupled separately for peptide and water/ions and the pressure was isotropically coupled to the system.
MD simulation snapshots were visualized in VMD 1.93. 37 Representative structures of the simulation trajectories were determined using clustering analysis of the last 10 ns of all repetitions each (3 x 10 ns, gromos method, RMSD cut-off 0.2 nm, g_cluster). 38 The central

Molecular Dynamics (MD) Simulations
MD simulations were performed for the Aβ40 and U3.5 peptide with α-helical and unstructured (random) starting structure each. While representative data are shown in the main manuscript, a comprehensive overview of the results of all simulations is included as part of the Supporting Information (Figures S5 -S17).    S23 Figure S16. Average distances between the phosphate head groups of POPC in the outer membrane leaflet and the peptide Cα atoms of (a, b, e, f) Aβ40 and (c, d, g, h) U3.5 (perpendicular to the membrane along z-axis) during the last 10 ns simulation time of all replicates. U3.5 peptide shows stronger binding to the membrane surface compared to Aβ40. Note that the symbols are used to distinguish the data sets and each residue has a data point. The vertical lines at residues 5 (arginine), 16 (lysine) and 28 (lysine) for Aβ40 and at residues 7 (arginine), 8 (arginine) and 14 (lysine) for U3.5 indicate the positively charged residues in both peptides to guide identifying the closest peptide-membrane interactions. Note that the symbols are used to distinguish the data sets and each residue has a data point.