A structural and kinetic link between membrane association and amyloid fibril formation of α-Synuclein

The protein α-Synuclein (αS) is linked to Parkinson’s disease through its abnormal aggregation, which is thought to involve an interplay between cytosolic and membrane-bound forms of αS. Therefore, better insights into the molecular determinants of membrane association and their implications for protein aggregation may help deciphering the pathogenesis of Parkinson’s disease. Following previous studies using micelles and vesicles, we present a comprehensive study of αS interaction with phospholipid bilayer nanodiscs. Using a combination of NMR - spectroscopic and complementary biophysical as well as computational methods we structurally and kinetically characterize αS interaction with defined stable planar membranes in a quantitative and site-resolved way. We probe the role of αS acetylation as well as membrane charge, plasticity and available surface area in modulating αS membrane binding modes and directly link these findings to their consequences for αS amyloid fibril formation.


4
NDs, paired with their accessibility, homogeneity and stability should therefore permit quantitative insights into membrane 69 association as well as its role in aggregation.

70
Here we make use of this potential and report on a comprehensive NMR investigation of the effects of lipid charge, bilayer 71 fluidity and αS acetylation on the structural aspects of αS membrane association. We corroborate these insights with molecular 72 dynamics (MD) simulations as well as a series of complementary biophysical measurements to further characterize membrane 73 plasticity, overall affinities as well as binding and aggregation kinetics. Based on this data we correlate structural insights, such as 74 residue specific affinities and competition for accessible membrane surface area, to their potential role in modulating αS aggregation 75 properties. Our study provides insights into (i) the different lipid binding modes of αS to stable planar bilayers of defined lipid quantity 76 and composition, (ii) the effect of membrane plasticity for αS binding, (iii) the modulation of membrane plasticity through αS, and 77 (iv) the connection between binding modes and their effect on αS aggregation. Additionally, it gives an initial estimate of the number 78 of lipid-associated αS molecules that are required to induce/promote nucleation, and allows to develop a basic structural, 79 thermodynamic and kinetic model of the modulation of αS aggregation through its interaction with different membrane surfaces. This finding provides additional evidence that αS does not interact with non-charged lipid bilayers (Rhoades, et 89 al., 2006), a point that is still controversial in the literature (Davidson, et al., 1998). Additionally, it also shows that αS does not 90 interact with the membrane scaffold protein (MSP), confirming that the effects described in the following are not biased by 91 (unspecific) αS-MSP interactions.

92
In a similar way as reported previously using liposomes , Iyer, et al., 2016, we further tested the 93 influence of increasing amounts of negatively charged lipid head groups on αS membrane association, keeping a molar ratio of one 94 αS molecule per membrane leaflet (Figure 1a-d). Note that lipid ratios and proper mixing of the different lipid types inside the 95 nanodiscs was also observed by NMR spectroscopy (Figure 1figure supplement 2a). Our NMR data show a gradually increasing 96 bilayer interaction of αS with increasing lipid charge content, dividing the protein into rather distinct regions with different binding 97 behaviors (Figure 1e). The first region spans the N-terminal residues 1-38, which are already weakly interacting at 25% content of 98 5 negatively charged lipids and strongly interact at 50% (or higher) charge content. The region of residues 38-60 interacts more 99 gradually at 50% charge content. Amino-acids 60-98, corresponding approximately to the aggregation-prone non-amyloid-β 100 component (NAC region), displays some interactions with membranes containing 75% anionic lipids and strongly interacts at 100% 101 anionic lipid content. The 98-120 region is (partly) affected by 100% net charge content only. Finally, the last 20 C-terminal residues 102 never show any membrane interaction. This data is largely in line with an expected predominantly electrostatic model (the first 60 103 residues displaying a net positive charge, the last 40 residues a net negative charge, and the NAC region being mostly hydrophobic), 104 as well as the three regions dynamic model reported before using SUVs (Fusco, et al., 2014).    Using Thioflavin T (ThT) fluorescence as a reporter for fibril formation, we also measured aggregation kinetics of αS in the 119 absence and presence of the different ND compositions (Figure 1f-g). These experiments were performed under conditions where αS 120 amyloid fibrils form spontaneously, mainly by interface-driven nucleation and subsequent amplification through fragmentation 121 (non-repellent plates, glass balls, shaking, see below for complementary kinetic assays) and therefore mainly report on the 122 potential interference of nanodiscs with the lipid-independent aggregation pathway of αS (Campioni, et al., 2014, Gaspar, et al., 2017 123 Rabe, et al., 2013, Vacha, et al., 2014. Interestingly, despite the fact that the NMR data show interaction, the presence of NDs up to 124 an anionic lipid content of 50% does not appear to affect aggregation kinetics (note that the free non-acetylated αS reference ( Figure   125 1g, gray) shows a different behavior, this point will be discussed below). When increasing the negative charge content to 75% the 126 aggregation half-time slightly increases (Figure 1f,g dark blue) and a strong aggregation-inhibiting effect is detected in the presence 127 of NDs with 100% anionic lipids (Figure 1f,g purple).

128
While the NMR data visualize the modes of αS binding to membranes of different charge contents, the ThT kinetic data 129 allow to directly link these molecular determinants to their effect on αS aggregation. In this respect, one of the most striking 130 connections is that αS interaction with NDs comprising up to 50% negatively charged lipids does not involve the NAC region and 131 that under the same conditions no detectable effect on the aggregation behavior of (acetylated) αS is found in ThT assays. When 132 further increasing the charge density above 50% negatively charged lipids, NMR data show first a partial (75% POPG, Figure 1a (Figure 1f,g, purple). Our data strongly suggest that for the tested conditions (high anionic lipid content and high lipid-136 to-αS ratios) membrane association of the NAC region seems to be the dominant factor for protecting αS from aggregation.

139
It is worth noting that the NMR results described above mainly refer to the decrease in peak intensity as a reporter for interactions, 140 which is in line with the effects seen before using SUVs (Bodner, et al., 2009). While it is clear that SUVs have particle sizes 141 (associated with slow tumbling rates) well above the detection limit of conventional solution NMR techniques, the smaller size of the 142 ND system should, in principle, allow detection of NMR signals, as has been reported before for several ND-bound or ND-integrated 143 proteins (Viegas, et al., 2016). Nonetheless, neither the usage of Transverse Relaxation Optimized Spectroscopy (TROSY) 144 (Pervushin, et al., 1997) with increased signal accumulation (i.e. 10-fold longer as for spectra shown in Figure 1a-b) nor the 145 measurement at increased temperatures (35°C) and the usage of an NMR-optimized smaller membrane scaffold protein 146 (MSP1D1ΔH5)  forming NDs of smaller size and higher tumbling rates, resulted in appearance of a new set of 147 peaks indicative for the bound sate (and the presence of slow exchange processes) or a collective shift of peaks indicative of fast on-148 off exchange processes (e.g. Figure 3 figure supplement 1). In theory, three effects may explain this observation and obstruct 7 detection of the ND-bound residues of αS: (i) the bound-to-free exchange rate is in the order of the NMR time scale (so-called 150 intermediate exchange), (ii) the presence of a non-negligible part of αS protruding out of the ND (namely at least residues 98-140), 151 slowing down molecular tumbling and increasing relaxation leading to line broadening beyond the detection limit, and/or (iii) 152 membrane-bound αS shows a significant amount of plasticity leading to inhomogeneous broadening of the NMR lines. While 153 intermediate exchange can be largely ruled out due to the observed binding kinetics (vide infra, Figure 4a

163
In order to still gain insight into the conformation of αS bound to NDs, we used magic angle spinning (MAS) solid-state 164 NMR which is not subject to size effects. Moreover, we took advantage of the very low temperatures (100 K) used in Dynamic

165
Nuclear Polarization (DNP) to additionally eliminate exchange processes, as well as to increase the sensitivity of the experiment. To 166 avoid problems of signal overlap arising from severe inhomogeneous line broadening often seen in this range of temperatures (Siemer, 167 et al., 2012), we used a sparse isotope labelling scheme , leading to the simplification of 13 C-13 C spectra to 168 secondary structure sensitive Cα-Cβ cross-correlations of valines (and leucine Cβ-C). Notably according to the αS primary sequence 8 (Figure 2, top) and our solution NMR observations ( Fig. 1a-b), 95% of the valine residues (i.e. 18 out of the 19) should be membrane-170 bound at the used charge content and αS-to-ND ratio. While in the absence of NDs the DNP 13 C-13 C spectrum (Figure 2, black) shows 171 a continuous distribution of the Valine Cα-Cβ cross peaks reflecting the carbon chemical shifts of the allowed Ramachandran space 172 (expected for an intrinsically disordered protein such as αS), a very strong peak shift to a defined chemical shift range typical for α-173 helical structure is visible after addition of NDs (Figure 2b, red). The DNP data thus show that αS binds the ND lipid surface in α-174 helical conformation corroborating previous studies using CD and vesicles, solution NMR and detergents micelles and solid-state 175 NMR and SUVs (Fusco, et al., 2014, Galvagnion, et al., 2015, Ulmer, et al., 2005.     3d, red). Based on this architecture it is tempting to speculate that the positively charged residues will interact with negatively charged 244 lipid head groups, the hydrophobic residues will be oriented towards the hydrophobic lipid chains and the negatively charged residues 245 will be oriented towards the solvent (and compensate the net charge of the protein). Noteworthy these protein features are again found 246 in the next binding region (residue 39-60). In this picture, it would be likely that the lipids as well as the lysine side chains (partly) 247 rearrange, from their 'unbound' conformation, to ideally accommodate electrostatic interactions in the bound conformation. This 248 rearrangement may be favored by a more fluid lipid phase, which would explain the lower interaction found for NDs with DMPG 249 lipids below the phase transition.

11
To test this hypothesis, we performed molecular dynamics (MD) simulations of αS-membrane interactions. Our simulations 251 focus on the first 61 residues of αS and its interactions with membranes formed by a mixture of either 50% POPG -50% POPC lipids 252 in the fluid phase or a 50% DMPG -50% DMPC mix in the gel phase (see methods for more details). Indeed, the MD data confirm 253 that the lysine residues play a key role in the membrane interaction, as e.g. visible by promoting considerably more contacts to anionic

272
While our NMR data clearly reveal binding modes with different contributions of the αS primary sequence, we were also interested 273 in the overall affinity of the protein for NDs. We therefore measured interaction kinetics using bio-layer interferometry (BLI) with 274 immobilized NDs of different charge content. In line with the NMR data, no αS binding was detected when NDs containing 100% 275 DMPC were immobilized (data not shown). In the case where NDs with 100% anionic lipid content were immobilized, a clear

296
In order to obtain residue-specific insights into the membrane affinity of αS, we additionally conducted NMR titration 297 experiments using 100% negatively charged NDs (Figure 4c-f). In general, dissociation constants can be extracted from NMR 298 titrations attenuation profiles by fitting the concentration dependency of the attenuation with a single exponential decay for each 299 13 resolved peak (corresponding to one assigned residue) (Figure 4e-f). Our data reveal differential membrane affinities for different 300 regions of the αS primary sequence. The regions with differential affinities largely overlap with the regions of the different binding 301 modes identified before, i.e. four distinct regions of decreasing affinity range (1-38, 39-60, 61-98, 99-140).

302
Importantly, it appears that one ND with 100% negatively charged lipids can simultaneously interact with up to 16 αS 303 molecules (8 per bilayer side) in the course of the NMR time scale, as seen from the almost complete disappearance of the signals of 304 the very N-terminal residues (Figure 4e-f, light blue). Moreover, while these first residues are interacting with the membrane, 305 independently of the number of αS molecules bound to one ND, the binding of the NAC region is strongly modulated by the number 306 of bound αS molecules. This suggests that, due to the higher lipid affinity of the N-terminal region as compared to the NAC region, 307 the free energy of the system is minimized by favoring N-terminal interactions in cases where the accessible membrane surface is 308 limited. Note that due to the geometry of the used nanodiscs up to five αS molecules can simultaneously bind with a 38-residue long

311
To characterize the effect of the accessible membrane surface area (as given by the αS-to-ND ratio) and the resulting stoichiometry 312 on αS aggregation behavior, we measured ThT aggregation kinetics on samples with different αS-to-ND ratios ranging from 2:1 to

323
indicates that a perturbation of the aggregation pathway by lipid incorporation is very unlikely in our case.

324
We additionally carried out the same BLI measurements, NMR titrations and ThT assays for ND containing only 50% 325 POPG lipids. For these NDs no clear signature of binding could be obtained in the BLI measurements (data not shown), suggesting 326 a weak affinity and/or too fast off rates to allow detection via BLI. This is in line with SEC profiles that also point to a more transient 327 interaction (Figure 1figure supplement 2).

328
NMR titrations, however, show clear concentration dependent attenuation profiles that allow the calculation of residue 329 specific affinities (Figure 5a-d). Noteworthy, the affinities for the αS residues in the first binding region (residues 1-38) are 330 14 comparable to the values obtained for 100% charged NDs (Figure 4e-f). In contrast, for the following binding regions much lower 331 affinities are found (at the edge of detection for residues 39-60, and no interaction for residues > 60), including the absence of 332 interactions of the NAC region. In line with an exposed NAC region the ThT data for NDs with 50% negatively charged lipids at low 333 αS-to-ND ratios are consistently showing no effect on aggregation half-times. The data at higher ratios are less reproducible and show 334 a slight tendency to prolonged half-times (Figure 5e-f). At this point, it is not clear whether this feature has mechanistic relevance or 335 is just an artefact caused by the limited reproducibility of this condition.

337
The molecular, thermodynamic and kinetic determinants of membrane-modulated αS aggregation.

338
It has recently been shown in a systematic study that under appropriate conditions, which minimize the intrinsic nucleation rate

348
As expected, the ThT assay under the quiescent conditions does not show any aggregation in the absence of NDs ( Figure   349 4j,k grey). However, the samples with 100% POPG NDs (in a αS-to-ND ratio of 16:1) do display aggregation, indicating that this 350 type of membrane is indeed enhancing primary nucleation to a degree sufficient that the subsequent secondary nucleation leads to 351 detectable quantities of amyloid fibrils (Figure 4j,k blue). Interestingly, since our data also allow an estimation of the total number of 352 αS monomers that are brought in close proximity due to their interaction with the same ND, this result may also provide a first 353 approximation of the number of αS monomers needed for the formation of a nucleus. Our data suggest that this 'minimal critical 354 nucleation number' has an upper limit of around 8 αS molecules.

355
In order to further disentangle the effect of NDs on the various individual steps in the αS aggregation pathway, we next 356 designed strongly seeded aggregation assays under quiescent conditions at neutral pH, where fibril elongation is the only process that    1c-i). This data is difficult to explain given that elongation is, in all cases of amyloid formation, responsible for the generation of the 378 bulk of fibril mass. Hence, its inhibition should slow down the overall aggregation kinetic, also under non-seeded conditions.

379
Noteworthy, we observe that ThT signal intensity can be strongly affected by the presence of NDs and that the absolute ThT intensity 380 does not correlate with absolute fibril mass when comparing data recorded in the absence or presence of NDs (as seen from SDS-381 PAGE, Figure 5figure supplement 1e). Although our NMR data show that these NDs do not interact with the αS, the corresponding 382 ThT signal (of the identical samples) show very different intensities ( Figure 5figure supplement 1f). It is therefore likely that the 383 ThT (unspecifically) interacts with NDs, leading to an overall decrease in ThT intensities. We therefore normalized most ThT assays.

384
Noteworthy, this effect is less pronounced for ND with 100% POPG (as e.g. visible in the data in Figure 4m

390
As visible in Figures 1 and 3-5, in addition to using the N-terminally acetylated form of αS, which represents the native post-391 translational modification of αS and is known to be relevant for membrane association (Nemani, et al., 2010, Theillet, et al., 2016,

392
we also recorded most experiments with the non-acetylated form of αS. In line with previous findings (Maltsev, et al., 2012), N-393 terminal acetylation leads to clear chemical shift perturbations in the NMR spectra for the first 10 residues of αS (Figure 1a,b).

394
Overall, most of the above discussed features of αS membrane interaction are rather similar in acetylated and non-acetylated 395 αS, however there are a number of distinct differences. For instance, the peaks which are already shifted in free αS due to the 396 acetylation are also the ones that are affected most by the presence of nanodiscs with low amount of charges (close to physiological 397 concentration). Our data (Figure 1a-d) show a rather small but significant increase in the membrane association of the first 15 residues 398 due to the N-terminal acetylation, which is in line with previous observation using SUVs (Bartels, et al., 2014, Dikiy and Eliezer, 399 2014). αS acetylation is known to increase N-terminal helix propensity (Kang, et al., 2013, Maltsev, et al., 2012, which may facilitate 400 formation of the initial binding mode and be of significance for naturally occurring processes.

401
A similar effect is also seen for the global binding as determined by BLI, which shows a (slightly) higher membrane affinity 402 of the acetylated (Figure 4a, KD of 60 nM) as for the non-acetylated αS construct (Figure 4b, KD of 100 nM). When looking deeper 403 into the NMR titration data, it appears that another effect takes place, namely a slightly increased membrane affinity of the NAC 404 region for the non-acetylated αS NAC region (Figure 4e vs. f and Figure 5c vs. d). At current stage, it is not easy to explain why a 405 modification at the N-terminus will affect the lipid interaction of a protein region that is sequentially separated by roughly 60 residues. Such a behavior could however either be related to intermolecular interactions and/or long range intramolecular interactions (in a 407 'horseshoe'-conformation) that may or may not be artificially introduced by the limited surface area of the NDs.

408
Strikingly, the reference kinetic curve of non-acetylated αS reproducibly shows under the applied conditions a strongly 409 delayed aggregation as compared to the acetylated reference. In the setup used, primary nucleation processes are likely to happen at 410 the air-water or plate-water interface (Campioni, et al., 2014, Vacha, et al., 2014, thus a lower hydrophobic propensity of non-411 acetylated αS could explain this effect. While this may be the dominant process in the absence of lipids, it may not be the case 412 anymore in the presence of NDs (Galvagnion, et al., 2015), either because NDs shield these interfaces or because nucleation happens 413 primarily at the membrane surface. The much lower differences due to acetylation state in the presence of NDs fit this explanation, 414 as well as additional tests we ran using different types of plates (data not shown). Higher order processes, namely different 415 fragmentation behaviors, can however not be excluded.

416
It appears that the biggest effect of acetylation is related to assay parameters that are normally not the matter of interest, 417 which nevertheless may be important for future studies (Iyer, et al., 2016). Still the results from systematic measurement of the effects 418 of N-terminal acetylation via different methods point to subtle changes in membrane interaction in respect to NAC region specific 419 affinities at high lipid charge densities as well as to N-terminal binding at a lipid charge density comparable with the overall 420 composition found for membranes in e.g. synaptic vesicles (Fusco, et al., 2014). Hence, both effects may be of particular importance Overall our data demonstrate that the nanodisc system allows to study the interaction of αS with stable, planar membranes in a 427 quantitative and site-resolved way. It also provides insights into the correlation between the identified membrane binding modes as 428 well as binding kinetics and their consequences for αS amyloid fibril formation, both in respect to the nucleation and elongation 429 process. In summary, our data show that (i) residue specific αS-membrane affinities are rather similar for the N-terminal αS region 430 for 100% and 50% negatively charged NDs, (ii) for 100% anionic lipids the αS can adopt a substantially expanded binding mode as 431 compared to 50% anionic lipids content, leading to considerably higher global affinities, (iii) the exchange rate between free αS in 432 solution and membrane-bound αS is rather slow in the 100% charged case and likely to be rapid in the 50% case, (iv) region-specific where membranes/NDs with only low charge densities are present (scenario 1 and 2) αS interacts with its N-terminal residues and most likely 444 forms a fast exchanging equilibrium between soluble and membrane associated αS monomers. This equilibrium does not seem to strongly 445 interfere with the slow process of αS nucleation, it may however (slightly) decrease the pool of free monomers available for fibril formation.

446
Notably these conditions are more likely to better resemble the average charge densities found in physiological membranes. However, 447 specific, abnormal and/or stochastic processes may also lead to highly charged membrane surfaces with limited (scenario 4) or not limited 448 surface access (scenario 3). αS will strongly interact with the latter in a binding mode that will largely inhibit both, αS nucleation and fibril 449 elongation. In cases where several αS monomers compete for a limited highly charged membrane surface area (scenario 4), the amyloid fibril 450 nucleation process can be accelerated, most likely due to an αS binding mode that brings exposed NAC regions of several αS monomers in 451 close proximity. Under these conditions the fibril elongation rate is largely unperturbed, probably due to sufficient monomers with limited 452 membrane association and/or due to induction of higher order processes.

459
Our data suggest that clusters of around 60-80 negatively charged lipids suffice to form a strong interaction (this may 460 however not be the lower limit). Sporadically formed lipid charged clusters could also induce a competition of several αS monomers 461 for the accessible surface area. Our data show that due to the different residue-specific membrane affinities this will generate a binding 462 mode that, once the rather low αS critical oligomerization number is reached, can act as an aggregation seed. Such a scenario could 463 promote the initial step of primary nucleation in the pathogenesis of Parkinson's disease and is in line with recent in vivo findings 464 suggesting that shielding αS from membrane interactions can inhibit initial steps of amyloid fibril formation including the formation 465 of cell toxic species (Perni, et al., 2017).

476
Sparsely labelled αS for DNP experiments was non-acetylated, expression was done in a similar way, in M9 medium using 0.4% [2-13 C]-glucose 477 and 0.2% 15 NH4Cl. Isotope labelling of Phe, Gln, Glu, Pro, Asn, Asp, Met, Thr, Lys, and Ile was suppressed by supplementing sufficient quantities 478 (150 µg/ml of each) of these unlabeled amino acids in the expression media as reported previously .

479
Purification of αS or acetylated αS was carried out as previously described (Hoyer, et al., 2002), some changes to the original protocol have been  As reported in before (Bayburt, et al., 1998)      3.2 (Bruker) and analyzed with CCPN (Vranken, et al., 2005). Peaks were automatically integrated and the ratio of volumes in the presence and 526 absence of NDs plotted against the primary sequence. Outliers as results of peak overlap and/or ambiguities were removed.

530
Assays were conducted in 20 mM sodium phosphate buffer pH 7.4 or 20 mM acetate buffer pH 5.3 with 50 mM NaCl, 0.02% NaN3 and 10 µM 531 Thioflavin T. Unless otherwise stated, triplicates of 120 µl were pipetted into 96-well half area well plates with non-binding surface (Corning

535
For the seeded experiments, fibril seeds of αS or acetylated αS were prepared as follows: 300 µl of 100 µM αS or acetylated αS was fibrillated at 536 37 °C and 800 rpm for 3 days in a 2 ml tube containing a glass ball in a Thermomixer (Eppendorf). The fibril solution was diluted to 50 µM and 537 sonicated with a tip sonicator (Bandelin Sonopuls HD3200, BANDELIN electronic) at 10% power (20 W) for 60 s, with 1 s pulses on and 4 s off in 538 between. Seed solution was diluted 20-fold for the aggregation assays (2.5 µM, 5%).

539
Kinetic curves were corrected by subtracting the curve of buffer (containing NDs) in the presence of ThT and normalized to highest fluorescence 540 intensity (in line with comparable fibril mass seen in SDS-PAGE after the aggregation assay). The corresponding triplicates are shown as transparent 541 circles in order to visualize the reproducibility of each experiment. In the case of quiescent nucleation and seeded assays no normalization was applied 542 and data were recorded without the presence of glass balls and without plate shaking.  As starting conformation for the MD simulations, the NMR structure of micelle-bound αS (PDB 1XQ8) was used, considering only the first 61 563 residues in order to concentrate on the membrane binding region of αS. The Amber99sb-ILDN force field (Lindorff-Larsen, et al., 2010) was used 564 for αS, which was simulated in its non-acetylated form (i.e., with NH3+ at the N-terminus) and with a C-terminal N-methyl amide capping group to 565 account for the fact that αS would continue beyond residue 61. All lysine side chains were modeled as positively charged, glutamate and aspartate 566 as negatively charged, while glutamine and histidine residues were considered to be neutral corresponding to a pH of 7.4. The protein was placed 567 either 0.5 nm or 1.5 nm above the membrane surface. A starting orientation with the negatively charged side chains pointing away from the membrane 568 and the lysine side chains being oriented towards the membrane surface were chosen (Fig. 3d). For modeling the lipid bilayer, membrane patches 569 consisting of POPC/POPG (1:1) or DMPC/DMPG (1:1) involving 512 lipids (256 lipids per leaflet) were built using CHARMM-GUI (Lee, et al., 570 2016) and modeled with Slipids force field parameters Lyubartsev, 2012, Jambeck andLyubartsev, 2013). Before αS was added, both 571 lipid bilayers were solvated and simulated for 500 ns (POPC/POPG) or 1000 ns (DMPC/DMPG) to obtain relaxed membranes. Here, the same 572 simulation procedure was employed as described below. αS was placed above the membrane, the protein-membrane complex solvated using the 573 TIP3 water model, Na+ and Cl-added to neutralize the system and to mimic the Na+ concentration used in the experiments. The ion parameters of 574 Smith and Dang (Smith and Dang, 1994) were used. The system was then subjected to steepest descent energy minimization, followed by MD 575 equilibration in the NVT ensemble for 1 ns at 10 °C using the V-rescale thermostat (Bussi, et al., 2007) with a time constant of 0.5 ps and separate 576 temperature coupling for the protein, membrane and water/ions. Afterwards, 1 ns of NPT equilibration was performed using the Nose-Hoover 577 thermostat (Hoover, 1985, Nosé, 1984 and Parrinello-Rahman barostat (Parrinello and Rahman, 1981) with semiisotropic pressure scaling, a 578 reference pressure of 1 bar, a time constant of 10.0 ps and an isothermal compressibility of 4.5 × 10-5 bar-1. During both equilibration steps, restraints 579 were applied to the positions of the P-atoms of the lipids and terminal C-atoms of their tails with a force constant of 1000 kJ mol-1 nm-2. All bond 580 lengths were constrained using the Lincs algorithm (Hess, et al., 1997). The Coulombic interactions were calculated using the Particle mesh Ewald 581 (PME) method (Darden, et al., 1993, Essmann, et al., 1995 with a cut-off of 1.0 nm for the short-range interactions and a Fourier spacing of 0.12 582 nm. The cut-off for the van der Waals interactions was set at 1.4 nm. Periodic boundary conditions were employed in all directions. For the MD 583 production runs the same settings as for the NPT equilibration were used, except that all position restraints were removed. All MD simulations were 584 performed at 10 °C with a time step of 2 fs for integration using the GROMACS 4.6 molecular dynamics package (Hess, et al., 2008). For the 585 analysis, which was performed using Gromacs and Membrainy tools (Carr and MacPhee, 2015), only the last 250 ns of each production run was 586 used. An overview of the production runs, can be found in Figure