Elevated concentrations cause upright alpha-synuclein conformation at lipid interfaces

The amyloid aggregation of α-synuclein (αS), related to Parkinson’s disease, can be catalyzed by lipid membranes. Despite the importance of lipid surfaces, the 3D-structure and orientation of lipid-bound αS is still not known in detail. Here, we report interface-specific vibrational sum-frequency generation (VSFG) experiments that reveal how monomeric αS binds to an anionic lipid interface over a large range of αS-lipid ratios. To interpret the experimental data, we present a frame-selection method ("ViscaSelect”) in which out-of-equilibrium molecular dynamics simulations are used to generate structural hypotheses that are compared to experimental amide-I spectra via excitonic spectral calculations. At low and physiological αS concentrations, we derive flat-lying helical structures as previously reported. However, at elevated and potentially disease-related concentrations, a transition to interface-protruding αS structures occurs. Such an upright conformation promotes lateral interactions between αS monomers and may explain how lipid membranes catalyze the formation of αS amyloids at elevated protein concentrations.

ric amplifier (OPA) with a non-collinear difference frequency generation (NDFG) extension (TOPAS Prime, Light Conversion) to generate broadband (FWHM ∼300 cm −1 ) IR pulses at 6.1 µm.A narrowband (FWHM ∼15 cm −1 ) visible beam was generated by guiding 1 mJ of the fundamental output through a Fabry-Perot etalon.The visible and IR beams were temporally and spatially overlapped at the sample surface.The SFG signal was focused into a spectrograph (Shamrock 303i, Andor) and detected by an EMCCD camera (Newton 971, Andor), which was operated using the Andor SOLIS for Spectroscopy software.VSFG spectra are recorded using an IR pulse spectral full-width-at-half-max of ∼270 cm −1 (from 1540 to 1810 cm −1 ), for 5 minutes per sample scan (as an optimal acquisition time in terms of dark-noise reduction without obtaining so many cosmic rays per scan that removing them in an automated fashion becomes challenging) and for 15 seconds for the Au intensity calibration spectra in PPP, and in SSP (S-SF, S-visible, P-IR), PPP, SPS and PSP polarization combinations for the sample spectra.The various polarization combinations were recorded in a sequential manner, frequently going back to the SSP polarization combination in order to get an impression of the development of the overall signal intensity.The intensity of the chiral polarization combination, PSP, appears to be very weak as compared to the achiral polarization combinations, but was included in the analysis nonetheless, because (as can be seen in Main Text Figure 3) the spectral calculations of various frames do lead to chiral signals, and including this polarization combination will thus make us more sensitive in our quest of finding which structures are present in our experiment.The sample stage and IR beam path were flushed with nitrogen to avoid artifacts due to IR light adsorption by water vapor.
The spectra are processed using an in-house Python script as follows.First the spectra are frequency-corrected by calibrating the frequency axis with a calibrated water vapor FTIR spectrum by overlapping the water vapor peaks in the 1600-1800 cm −1 with the sharp dips in an unpurged VSFG sample spectrum.Then, the cosmic rays are removed by detecting sudden jumps in the spectra when going to a subsequent pixel (threshold = 15 counts, which S2 is sharp enough to identify cosmic rays, while not modifying the VSFG signal's shape, which has a maximum intensity of ∼125 counts), and after how many pixels the photon count is again within the threshold, after which the intermediate pixels are interpolated.Then, the background spectra (recorded with blocked IR) are subtracted from the sample and the gold spectra.To compensate for minor remaining amounts of visible light due to beam pointing differences or sample evaporation during the experiments, which can create a linear offset when visible light that scatters from the trough bottom hits the detector, subsequently a linear background (fitted to the average values of the first and last 50 pixels of the 1100-2200 cm −1 range of the CCD chip), is subtracted.In these experiments, this only yielded linear offsets of a few counts at the extremes of the CCD chip (while, again, the highest signals were ∼125 counts).Finally, the processed sample spectra are normalized using a reference spectrum obtained from gold, which has been processed with the same procedure.

Supplementary protein concentration notes and calculations
We note that the 'physiological' αS concentration of ∼20 µM is based on only a single reference, 2 because this appears to be the only study that has determined this in the synapse.
The number of ">1 µM" can also be found in the literature, 3-5 but we were not able to derive from which experimental data this number was derived, and furthermore it is not specific.
The determination of the LPR values of ∼0.037, 0.091 and 37, as used in the Main Text is given in the table below.The area/lipid molecule value of 55 Å2 is taken from ref. 6.The factor 2 is incorporated to optimally compare our results to the literature studies, which are all bilayer studies.This way, the number of lipids exposed to the protein solution is the same for a given LPR, without affecting the physics behind the αS-lipid interaction, because αS is not expected to be affected by the presence of the distal lipid leaflet.that both "sides" of the α-helix horseshoe are interfaced with the monolayer.In system 6 the horseshoe was rotated 90 degrees, compared to system 1 and 2, and is only in contact with lipids via the N-terminus.
Additionally, we created two systems where αS adopts an upright conformation protruding away from the surface (system 4, 5), and one (system 7) lying flat on the membrane with several kinks, compared to PDB: 2KKW [http://doi.org/10.2210/pdb2KKW/pdb](SLASmicelle bound αS ).The αS structures used in system 4, 5 and 7 were obtained from prelim-S4 inary simulations of αS on DPPG bilayers and were found to match the low LPR ensemble (system 4,5) and the high LPR ensemble (system 7).These structures were included to diversify sampling of the αS conformational landscape.
Furthermore, two control systems were created: one without secondary structure (i.e.αS is completely disordered), in contact with the monolayer (system 3), and one where αS was placed in the vacuum above the monolayer (system 8).The starting structure of system 8 moves from the vacuum to the backside (tails) of the monolayer within ∼500 ps and displaces lipids during the 150 ns trajectory (Figure 7).All starting conformations are depicted in Molecular dynamics simulations were performed using GROMACS 10 v2021.4 on Nvidia V100 GPUs at the Centre for Scientific Computing Aarhus (CSCAA).We use a combination of the DES-amber force field 11 for protein, water, and ions, and the Slipid forcefield [12][13][14] for lipids.A non-bonded truncation cutoff of 1 nm was used and the particle mesh Ewald 15 method was used for long-range electrostatic summation.The systems were first minimized and equilibrated in two steps: (1) 250 ps (1 fs timesteps) in the NVT ensemble using the vrescaling thermostat 16 at 296.15 K (the temperature in our laser lab) with position restraints on the DPPG molecules and the protein backbone; (2) 10 ns (2 fs timesteps) using the same settings and hydrogen bonds restrained using LINCS but not restraints on DPPG.
Production runs of the eight systems were simulated for 150 ns with two repeats in each simulation using the same setting as step (2) but without position restraints on the protein.

S5
Velocities were resampled from the Maxwell-Boltzmann distribution for each repeat.Frames were saved every 50 ps.
We note that one should be careful combining force fields.However, Slipids were developed to be compatible with the AMBER99SB-ILDN force field branch which is also the foundation of the DES-amber force field, and thus this force field combination is currently the most consistent option for simulation of IDPs and lipids.However, since the LJ interactions between TIP4P-D (water model in DES-amber) and the DES-amber protein parameters are increased, compared to other protein force fields and water models, e.g.
TIP4P+AMBER99SB-ILDN, this could shift the relative interactions between protein and lipids.We stress that our MD simulations are used as a way to generate realistic hypothetical conformations of αS near a DPPG monolayer, and that the fit RSS of the VSFG spectral calculation is used to extract an ensemble consistent with the VSFG experiment.

Spectral calculations
The spectral calculations are based on a excitonic Hamiltonian approach developed first described in ref. 17.Because they are based on the 16 150 ns MD trajectories, we performed the calculations for a total of 48.016 different frames, with a varying protein structure and orientation in each frame.Because we did not find an improvement of the spectral match when we averaged over multiple frames (see Table S3), all presented spectral calculation are performed by directly obtaining the eigenmodes from each frame, by evaluating the excitonic Hamiltonian model.The residual sum-of-squares (RSS) between the experimental and calculated spectra was used to find the frames that describe the experimental response best.
In the spectral calculations (also known as frequency mapping 18 ), the exact spectral simulation parameters that are used can be very important for the outcome.The nearest-neighbor coupling was described using a dihedral-angle parameterized map of ab initio B3LYP  2 for the bulk-air phase, bulk-water phase and the refractive index of the thin interfacial lipid/protein film.The interfacial refractive index is important in the description of L zz (ω j ) in the Fresnel factor equations. 19,20It is a weak point in determining the local mode corrections, since it is particularly hard to determine with a consistent and independent approach, but the frequently-employed estimation of Zhuang et al. 19 has resulted in consistent molecular pictures in many studies. 20,21pplementary Table 2: Refractive indices used for estimation of the local field corrections.Water data from ref. 22 and the interfacial refractive index from ref. 19 Source data are provided in the Source Data file.

S8
spectra experiments, we found the slope parameter of the hydrogen-bonding effect on the local-mode frequency to be δω HB = 400 cm −1 Å−1 r CO and isolated local-mode frequency Ω 0 was 1655 cm −1 (see Figure 1).The experimental LKα14 data is recorded with a picosecond VSFG scanning setup in Seattle, as opposed to the α-synuclein data recorded in Aarhus from the femtosecond Astrella laser in the Weidner Lab.The linewidth for the two setups is different, so this parameter was therefore optimized for the α-synuclein spectrum.However, the best protein structure was found to be independent of the linewidth, within a range around the expected experimental linewidth (see Table S3).

Supplementary discussion of spectral match
The resulting spectra with the lowest RSS (see Figure 3) closely match the relative SSP and PPP intensities, as well as their lineshape.The SPS match is slightly less good, which can be understood from the fact that selection based on the total RSS of all polarization combinations results in a higher sensitivity to stronger polarization combinations.The lack of any (chiral) PSP intensity is reproduced well, and the transparent band that indicates the bandwidth of the spectra calculated for the frames within the ensembles shows that including this polarization combination filters out structures that have a chiral response, contrary to the experimental observations.It is interesting to note that most deviations within the transparent band are in the 1660-1680 cm-1 region, where turns are thought to absorb IR light. 26Also based on other recent results obtained with this Hamiltonian approach, we think that this is due to force-field imperfections that lead to too-steep turns.Happily, when taking the average of the structures in the ensemble the effect averages out, or at least has a negligible effect on the spectral lineshapes.S9

Supplementary discussion of spectral-calculation results for the MD simulations
A more detailed comparison of the molecular dynamics(MD) simulations and the spectralcalculation derived RSS values reveals that, for all the trajectories, the kinked α-helical starting structure partly unfolds, resulting in shorter helical segments linked by disordered regions (see Figures 2(C) and 12).The α-helical contents (quantified by the DSSP algorithm 27 ) generally decreases throughout the 150-ns trajectories, in some cases ∼20-30%.
Interestingly, there is no clear correlation between the obtained RSS values and the helicity, indicating -as expected -that the VSFG response is sensitive to other structural aspects besides the secondary structure alone.Furthermore, although the difference is not large, from Figure 10, it appears thatwhile most segments have a very similar secondary structure distribution, there is a distinct subset of structures with a helicity of ∼30% at low concentrations, which is absent in the high concentration ensemble.The capability to resolve such detailed structural differences is a valuable feature that the frame-selection method allows.

Supplementary discussion of spectral calculations for literature models
For 2KKW [http://doi.org/10.2210/pdb2KKW/pdb](SLAS-micelle bound αS), for both LPRs, all available models within the PDB entry were tried in order to find the model with the best match.This resulted in model 26 for the low-LPR dataset, and in model 18 for the high-LPR dataset.For the structures (of αS fragment (residues 9-89) bound to 7:3 POPC/POPS small unilamellar vesicles 28 ) provided by the Langen group, only the first structure out the identified ensemble was analyzed, as all of the structures were structurally very similar.Although a small part of the N-terminus and the full C-terminus is missing, we expect that this will only minutely influence the result, as random-coiled structures hardly contribute to VSFG signals. 20For the low-LPR dataset, the best 2KKW orientation results in a reasonably extended protein orientation, and also the degree of helicity is not very S10 different in the best-matching MD-simulation frames, which explains why the RSS of this structure and orientation is relatively low.Lorentzian widths of both 8 and 12 cm −1 have been tried, finding that a width of 12 cm −1 gave the best match.Most importantly, for the low-LPR dataset, the orientations reported in the associated publications do not appear to be consistent with our experimental results at all, given the very large RSS values at y rotation = ∼ 90 • for the horseshoe structure 9 and at both x and y rotation = 90 • for the Langen structure, 28 while for the high-LPR dataset these orientations actually lead to a close match, which corroborates the similar orientations derived with the frame-selection method described in the main text, thus benchmarking this novel approach.

Atomic-force microscopy
The AFM samples have been prepared by performing a Langmuir-Schaefer deposition of the sample surface on freshly-cleaved mica, followed by drying for 10 min at the ambient environment.Subsequently, the sample on mica was imaged.
The AFM imaging was conducted on the Multimode ® VIII AFM (Bruker, CA) in a Peakforce tapping mode.An silicon nitride probe Scanasyst-Air with the spring constant of 0.4 N/m was used under the ambient environment.The setpoint is less than 200 pN in order to protect sample.The raw images were processed by the software SPIP™(Scanning Probe Image Processor) software package (Image Metrology, Denmark).S11 2 Supplementary calculation results

RSS values for various Lorentzian widths and frame-averaging sizes
Supplementary Table 3: Table of key parameters from the investigation of the dependency of the best matching α-synuclein structure on the broadening parameter Γ. "Best structure" refers to the structure with the lowest RSS value.The half-widthat-half-max (HWHM) Γ = 8 cm −1 is used as it is the expected experimental linewidth of the visible pulse coming out of the etalon, but it is also observed that the conclusion from the calculation is stable within ∼ 2 cm −1 .The structure in frame 2599 is shown in main text figure 2(C).The best matching structures from Γ = 6 −12 cm −1 are all very similar in structure and orientation.Source data are provided in the Source Data file.

S17
Supplementary Figure 5 (preceding page): AFM images taken of regions with defects of Langmuir-deposited interfaces (on freshly-cleaved mica), which show that there are no fibrillar species present at the water-lipid interface.(A) Height map of a DPPG monolayer that has been incubated for 5-hours with 50 µM of αS (LPR = 0.037), indicating that at these incubation times and concentrations the proteins form a ∼1.5 nm thick monolayer on top of an ∼1.8 nm DPPG layer (consistent with the lipid lengths found in the MD simulations).On top of these layers, oligomeric species are observed, which might either be an artifact from the drying step in the sample preparation, or from protein aggregation during the experiment.(B1-3) Three associated height profiles.(C1,C2) Height and adhesion maps of a pure DPPG monolayer.(D1,D2) Similar, but for a DPPG monolayer incubated for 24 hours with an αS solution at LPR = 37.As can be seen in the height profiles depicted in (C3) and (D3-5), the lipid monolayer is ∼1.6 nm high, the monomeric protein layer is ∼0.8 nm high (consistent with the diameter of an α-helix), while the oligomers (mostly to be found on the lipid layer, but there are also some found on the protein layer) are ∼10 nm high.It is interesting to note that the largest oligomers are found on the edge of the areas where the lipids have been displaced by the proteins.

Figure 6 . 15 M
Figure 6.All αS residues were built in standard protonation states (pH 7.4) with H50 being neutral and in the δ-tautomer.Protein termini were uncapped and charged.The monolayer systems consist of a water region (100,000 water molecules) bordered by two DPPG monolayers (250 lipids in each, head groups facing the water).The monolayer was built using an area per lipid of 55 Å2 , mimicking the liquid-condensed (LC) monolayer phase.A large vacuum region separates the DPPG monolayers.The water region contains 768 Na + , and 259 Cl − ions (0.15 M NaCl).The box size of the systems are ∼11.7 nm x 11.7 nm x 81.5 nm.
/6-31G+(d) coupling from N-Methyldacetamide (NMA) deuterated at the amide group's N S6 atom.Transition dipole momenta were based on a map of Mulliken charges from an ab initio B3LYP/6-31G+(d) calculation.The frequency shift applied for the five proline residues in αS was δω proline = 19 cm −1 .The refractive indices used to describe the local-field corrections are listed in Table

18 1. 4 . 1 Supplementary Figure 1 :
Calibration of the spectral calculations with LKα14To calibrate the local-mode frequency parameters (the frequency of the uncoupled local modes and influence of hydrogenbonding on the local-mode frequency, estimated by a linear correlation with the C=O bond length23,24 ), we compare experimental and calculated spectra of the LKα14 peptide.25 LKα14 is a short peptide of leucine and lysine residues (Ac-LKKLLKLLKKLLKL).It is specifically engineered to fold into a stable α-helical structure at a hydrophilic/hydrophobic interface.LKα14 is simulated for 300 ns in 3 replications on a water-hydrofobic (PG lipid-tails) interface.The experimental data is taken from ref. 25 of LKα14 at a hydrophobic polystyrene interface in water.The reflections of the sum-frequency-, visible-and infrared-beams, as well as the different VIS-wavelength (λ = 532 nm) beams were compensated for with the appropriate local-field corrections.By optimizing the spectral parameters of the calculations to optimally match the SSP and PPP S7 Intensity (arb.units) Lk Peptide from f3000to4000_ Exp.data: SSP Exp.data: PPP Simulated SSP, RSS = 0.370915 Simulated PPP, RSS = 0.564351 Experimental (rugged) and calculated (smooth) spectra of LKα14, used to calibrate the local-mode frequency parameters of the spectral calculations.Source data are provided in the Source Data file.

9 :
(E,F) Height and adhesion histograms of the AFM image for DPPG (top) and αS -incubated DPPG (bottom) corroborate the layer assignment through the different adhesion values for the lipid and the protein layers.All scale bars are 1 µm.Source data are provided in the Source Data file.Distances to the nearest DPPG molecules for each residue of the conformations within the high-LPR and low-LPR ensembles.

Table 1 :
Calculation of lipid-protein ratios.Same method was applied for the determinations of LPR = 0.091 and 37 for [αS] = 20 µM and 50 nM, respectively.Source data are provided in the Source Data file.

Table 4 :
−1] Lowest RSS MD frame of best structure RSS of frame 2599 RSS as a function of the number of successive frames over which the local-mode frequencies (that are a function of the hydrogenbonds) are averaged, for each of the production runs.