|Home | About | Journals | Submit | Contact Us | Français|
In dopaminergic neurons, a-synuclein (αS) partitions between a disordered cytosolic state and a lipid-bound state. Binding of αS to membrane phospholipids is implicated in its functional role of synaptic regulation, but also impacts fibril formation associated with Parkinson’s disease. We describe here a solution NMR study in which αS is added to small unilamellar vesicles of a composition mimicking synaptic vesicles; the results provide evidence for multiple distinct phospholipid-binding modes of αS. Exchange between the free and lipid-bound αS state, and between the different bound states, is slow on the NMR timescale, being in the range of 1–10 s−1. Partitioning of the binding modes is dependent on the lipid: αS stoichiometry, and tight binding with slow exchange kinetics is observed at stoichiometries as low as 2:1. In all lipid-bound states, a segment of residues starting at the N-terminus of αS adopts an α-helical conformation while succeeding residues retain characteristics of a random coil. The C-terminal 40 residues remain dynamically disordered, even at high lipid concentration, but can also bind to lipids to an extent that appears to be determined by the fraction of cis X-Pro peptide bonds in this region. While lipid-bound αS exhibits dynamical properties that preclude its direct observation by NMR, its exchange with the NMR-visible free form allows for its indirect characterization. Rapid amide-amide NOE buildup points to a large α-helical conformation, and a distinct increase in fluorescence anisotropy attributed to Tyr39 indicates an ordered environment for this “dark state.” Titration of αS with increasing amounts of lipids suggests that the binding mode under high lipid conditions remains qualitatively similar to the low-lipid case. The NMR data appear incompatible with the commonly assumed model where αS lies in an α-helical conformation on the membrane surface, and instead suggest that considerable remodeling of the vesicles is induced by αS.
α-Synuclein (αS) is a presynaptic protein which is strongly implicated in the pathogenesis of Parkinson’s disease. It is found in the form of amyloid fibrils, rich in beta-sheet structure, in the intraneuronal Lewy body lesions characteristic of the disease.1,2 Three familial mutants of αS have been discovered and correlate with aggressive inherited Parkinsonism.3–5 In presynaptic neurons, wild-type αS is abundant, estimated between 30 and 60 μM 6. It partitions between a disordered7 cytosolic and a vesicle-bound form.6,8–13 While the details of its function remain largely unknown, αS has been linked to the maintenance of reserve pools of synaptic vesicles14–18 and dopamine homeostasis.19–21 The protein binds to acidic phospholipids and anionic detergents, and then undergoes a coil to helix transition.6,11,22–27 Seven imperfect 11-mer repeats, containing the consensus sequence KTKEGV, are located in the N-terminal domain of αS and are likely to mediate its lipid binding properties. The central non-amyloid beta component (NAC, comprising residues 61-95) of αS contains a hydrophobic stretch of residues at least partially responsible for its high propensity to convert into fibrillar species.28,29 Long-range interactions between the NAC region and the acidic residues of the C-terminal tail have been identified by paramagnetic relaxation enhancement in NMR studies of samples free of lipids and detergents, and it has been proposed that this interaction stabilizes αS to keep its aggregation propensity in check.30–32 The 40 C-terminal residues are otherwise reported to lack defined structure and solution NMR studies suggest that they are inert to αS-lipid interactions.26,33,34 Based on chemical shift analysis, the same studies of lipid/detergent-free αS identified a weak α–helical propensity in the N-terminal domain. Later, RDC measurements were used to determine the SDS detergent-bound structure of αS, which features two amphipathic α-helices in the N-terminal domain bound to the micelle surface, with the 40 residue C-terminal tail remaining flexible in solution.35,36
The structure induced upon phospholipid binding and the exchange dynamics between lipid-free and lipid-bound states are likely to mediate the native state function of αS. However, fibril formation by αS is also affected by lipid interactions and, depending on conditions, membrane mimetics can either enhance or inhibit the aggregation kinetics.37–42 Furthermore, various sources of evidence suggest that the toxicity of αS is due to prefibrillar oligomeric species capable of altering membrane permeability, possibly in a manner similar to pore-forming bacterial toxins.43–47 More recently, it has been suggested that αS is capable of forming transmembrane ion channels in depolarized plasma membranes, and that disease may be triggered by the malfunction of a native role in ion conductance.47–49
The affinity of αS for phospholipid vesicles of varied chemical composition has been a topic of intense study pursued by a variety of biophysical techniques (for a review, see 50). While a consensus has not fully been reached, the requirement of an acidic headgroup for strong binding and a helical transition of αS are observed consistently. The fact that synaptic vesicles contain high phosphatidylserine (PS) content (up to 12%) has motivated the study of PS-containing synthetic vesicles under a variety of conditions. The additional presence of neutral and zwitterionic lipids, particularly phosphatidylcholine (PC) and phosphatidylethanolamine (PE), further enhances binding, perhaps because of the favorable packing properties of these mixed vesicles.13,22,23,26,51 Electrostatics alone cannot account for the binding properties of αS as lipid interactions persist, albeit weaker, even with vesicles lacking net negative charge.52,53 The degree of saturation and the length of the phospholipid chains, which determine the melting temperature and phase transitions of the bilayer, also affect αS binding.54,55 It has been proposed that αS may preferably bind to regions of lipid disorder or annealing defects and thereby stabilize these types of membrane structure.18,55,56 In contrast, other studies have identified lipid-ordered domains, such as lipid rafts and ‘raft-like’ lipids as being important for localization of αS to particular regions of the bilayer.57,58 In all cases, it is clear that αS is finely tuned to its lipid environment, and while charge interactions undoubtedly aid in recruiting αS to the bilayer, specific hydrophobic interactions appear necessary for αS to adopt and sustain a functional structure.
To gain an understanding of αS-lipid interactions on an atomic scale, we have used solution NMR spectroscopy to measure the binding of αS to acidic small unilamellar vesicles (SUVs). Our analysis of NMR intensity and relaxation data lead us to propose the presence of competitive αS-phospholipid binding modes which depend on the lipid: αS stoichiometry and which have been probed over a wide range of relative concentration, from less than 1:1 to larger than 150:1. We present evidence for stable, oligomeric states of αS bound to lipids, in slow exchange with a free, monomeric and dynamically disordered aqueous state. Binding of αS to phospholipids is tight, suggesting that the equilibrium in the synapse environment is strongly shifted to the lipid-bound form. Notably, our data are inconsistent with the common “carpet” model where monomeric αS lies in an α-helical conformation on the membrane surface. On the other hand, our data do not appear to conflict with the experimental EPR observations themselves,59–62 despite the fact that they have been used to support such a surface-binding model.
In our characterization of αS-lipid interactions, we have used SUVs, consisting of a 5:3:2 DOPE:DOPS:DOPC molar ratio; vesicles that are reported to mimic the composition and curvature of synaptic vesicles.63 In agreement with previous reports.6,11,22,24 we find that αS undergoes a coil-to-helix transition in the presence of these acidic SUVs, as monitored by circular dichroism. Our study has additionally utilized solution NMR spectroscopy to quantify the equilibrium partitioning of αS, and to probe its lipid-bound state in a site-specific manner. Peaks observed in the 15N-1H correlation spectrum which coincide in position with those observed in the absence of lipids correspond to residues that retain random coil behavior; by contrast, residues that experience a different environment when αS binds to lipids will change their chemical shifts and dynamic characteristics. No new or shifted resonance positions are, however, observed upon addition of lipids to αS (Figure 1A), even though the intensity of many cross peaks is greatly attenuated upon addition of lipids (Figure 1B). This result, and the absence of any visible broadening of any of the cross peaks upon addition of lipids, indicates that residues which bind to the phospholipids are unobservable in the solution NMR spectrum. Consequently, the decrease in signal intensity in the HSQC spectrum upon addition of lipids reflects quantitatively the fraction of protein molecules for which any given residue has become NMR-invisible by lipid binding. A plot of the observed intensity in the αS spectrum as a function of position in the protein sequence demonstrates that the signal attenuation is not uniform across the protein, but rather the signals from the N-terminal residues are attenuated considerably more than residues towards the C-terminus (Figure 1B). This result importantly demonstrates that (a) a higher fraction of αS contains immobilized residues near the N-terminus than towards the C-terminus, (b) residues which are not themselves immobilized by lipid-binding remain dynamically disordered, random-coil-like, even when other parts of the same protein molecule are immobilized, and (c) multiple distinct and long-lived states of lipid-bound αS must be present. This last point is apparent when noting that contiguous residues of the protein show comparable attenuation, indicative of binding in a concerted manner, but that distinct steps in the attenuation profile are observed at certain points in the sequence. This effect is most pronounced between residues 98 and 100, but is also observed between residues 25 and 30, and at high lipid concentrations (Figure 2B) transitions at residues 65, 85, and 120 are also observed. This plateau behavior identifies the presence of distinct bound states of αS, with exchange between them being slow on the NMR time scale (<10 s−1). At low lipid to protein ratios, there exists a pool of protein where only the first ca 25 N-terminal residues are bound and NMR-invisible (referred to as SL1) and a second pool (SL2) where residues 1-97 are invisible. As will be discussed later, strong transferred NOE connectivities for residues 1-100 point to slowly tumbling helical states for SL1 and SL2.
Differences in attenuation are also observed for residues preceding and following Pro120 in the sequence, but as will be discussed later, the attenuation for the C-terminal residues is of a different origin that is likely to be influenced by the cis/trans equilibrium of Pro117 and/or Pro120 peptide bonds. A similar, less distinct break in attenuation is seen at Pro128, but no transferred NOEs are seen for any of the C-terminal 40 residues, indicating that C-terminal regions with trans Pro peptide bonds remain highly dynamic, even in the presence of lipid.
As shown in Figure 2A, the relative degree of attenuation changes when the protein concentration is lowered, indicative of competition between the different binding modes. Binding of αS to lipid can be described by equations of the type
where, in its simplest form, only two lipid-bound states (SL1 and SL2) are distinguished. The NMR data recorded at high lipid: αS ratio (Figure 2B), however, suggest that more than two distinct states exist. A highly remarkable and unexpected feature of the lipid binding is that the equilibrium of eq 1 is in the slow exchange limit on the NMR timescale, meaning the lifetime of a molecule in the free state is long compared to T2*, where the NMR line width is given by (πT2*)−1. For αS in the presence of lipids, 15N line widths ≤5 Hz are observed, indicating that the lifetime of the free state exceeds 100 ms. In addition to exchange between the lipid-free and lipid-bound states of αS, exchange between the different SLn forms may occur, but considering the absence of visible exchange line broadening this exchange must again be slow on the NMR timescale.
In order to retain a sufficient population of free, NMR-observable αS, experiments mostly are carried out at very dilute lipid concentrations (0.3 mg/mL) where the SL1 and SL2 binding events are in competition with one another due to limited lipid availability. However, when lowering the αS:lipid ratio in a stepwise manner, we asymptotically reach an attenuation pattern where competition for lipid no longer dominates (Figure 2A). The differences in attenuation observed as a function of position then provide a direct measure of the SL1 and SL2 bound-state populations. For example, for the conditions of Figure 1 (150 μM αS; 400 μM lipids), attenuation is ca 60% for residues 1-25, and ca 45% for residues 30-100. This finding corresponds to concentrations for [SL1+SL2] of 60%, and [SL2] at 45%, i.e., [SL1] at ca 15%. At lower protein:lipid ratios, the difference in attenuation between residues 1-20 and 30-100 decreases (Figure 2A), pointing to an increase of the SL2:SL1 ratio.
Our observations are in qualitative agreement with an earlier NMR study which showed residues 108-140 of αS to remain fully observable in the presence of lipid vesicles while the other residues were observed with much reduced intensity.33 The low-intensity signals were interpreted as arising from a minor population of free, disordered αS. However, our results indicate the presence of distinct lipid-bound αS states, which have different sections of their N-terminal 100 residues involved in lipid binding. Distinct boundaries in the C-terminal segment of 40 residues point to distinctly different behavior for different bound forms of the protein, but as will be discussed later, we tentatively assign these boundaries to populations of protein with cis X-Pro peptide bonds, with changes in attenuation pattern coinciding with the location of proline residues. Our results differ markedly from the behavior seen when αS binds to sodium dodecyl sulfate (SDS) micelles. In that case, the first 96 residues adopt primarily an α-helical conformation and the C-terminal 40 residues retain random coil characteristics. Moreover, exchange broadening is seen when the SDS: αS molar ratio falls below ca 70:1.36 Differential attenuation of different protein segments upon lipid addition, observed in our study, cannot be attributed to exchange broadening or variations in the degree of motional restriction imparted to different residues in the molecule because the line widths are highly uniform and very narrow for all observed resonances (Figure 2C), and no change in chemical shift is observed for any of the residues upon addition of lipids. As will be shown below, rapid NOE buildup and extensive HN-HN spin diffusion in transferred NOE experiments, carried out in the presence of lipids, point to a slowly tumbling, largely helical structure for the “dark state”. The absence of significant line broadening for the resonances remaining upon addition of lipids (Figure 1) then points to a very small generalized order parameter,64 i.e. S2 < ~10−2, for these observed amide groups.
Fixing the total lipid concentration at 0.3 mg/mL, or ~0.4 mM, we lowered the αS concentration to achieve a ratio where the attenuation profile reaches a plateau (Figure 2A). This condition is reached at ca 25 μM, where 85–90% attenuation is observed for the N-terminal αS resonances. The remaining free αS concentration at 25 μM total αS concentration is ca 3 μM, which drops to ca 1.5 μM at the lowest αS concentration probed (12.5 μM). Without requiring knowledge of the number of lipids bound per αS molecule, we can define an apparent total dissociation constant, Kd,t, with respect to the monomer concentration of lipid:
where [ΣSLn ] refers to the total concentration of lipid-bound αS. As we have ensured that we are below the crowding limit, [Lipidfree] [ΣSLn], so that [Lipidfree] is estimated in the 0.25 - 0.4 mM range, yielding a Kd,t value in the range of 30–50 μM. Due to morphological changes that can occur to the SUV (discussed below), we consider all available lipid in deriving the above Kd,t, including both leaflets of the SUV.
Remarkably, even at very low lipid: αS ratios a substantial fraction of αS is converted to its “dark state” by the lipids; comparison of samples that contain 75, 150, and 300μM αS, and each 400 μM (0.03%) lipid (Supplementary Material, Figure 1), indicate that the absolute concentration of total lipid-bound αS reaches ca. 150 μM αS. These conditions represent a ≤3:1 molar stoichiometry of lipid: αS for the bound species. The surface area of a single lipid headgroup situated in a bilayer is only ~0.5 nm2, which clearly is insufficient to account for the high affinity of αS for lipids in a simple surface binding mode. A binding mode with such a small area of contact not only appears difficult to reconcile with the immobilization of a large fraction of the αS backbone but also is inconsistent with the slow exchange kinetics observed for the system. The stoichiometries therefore exclude the possibility that the bound forms simply reflect different states of αS on the surface of the SUV, and instead require a drastic rearrangement into a new species containing roughly equimolar fractions lipid and protein, i.e., dominated by an order of magnitude in mass by αS.
We also measured fluorescence anisotropy to determine the stoichiometric ratios of αS lipid binding, and the results are in close agreement with the low lipid: αS ratios determined by NMR titrations. Fluorescence anisotropy increases with the correlation time of the sampled species (up to a maximum value of r = 2/5 for samples where the rotational correlation time of the chromophore greatly exceeds its fluorescence lifetime), and is therefore an indicator of the degree of immobilization of αS in its bound form. Native αS does not contain any tryptophan residues, and we therefore monitored the fluorescence at the maximum emission wavelength of tyrosine (306 nm) as a function of lipid concentration. Upon lipid binding, fluorescence anisotropy is expected to be dominated by the Tyr39 signal, while Tyr125, Tyr133, and Tyr136 will contribute minimally due to their C-terminal position, which remains unperturbed and dynamically disordered at low lipid: αS ratios according to the above NMR data.
Figure 3 shows the rapid increase and near saturation of fluorescence anisotropy of 500 μM αS as lipid vesicles are added. By the point of addition of 0.1% SUV, where the lipid: αS molar ratio is 2.6, the fluorescence anisotropy approaches its asymptotic limit, indicating that the bulk of the change in tyrosine anisotropy signal (attributed to Tyr39) is complete. Under the same sample conditions, the NMR signal intensity data also indicate that more than 60% of Tyr39 residues do not yield a visible backbone amide NMR signal, while the amide signals of the three C-terminal tyrosine remain nearly unattenuated.
As discussed above, the absence of any significant shifts or broadening of the observed αS resonances from their random coil, lipid-free positions establishes that the exchange process between the free and bound species is slow on the NMR chemical shift timescale. Such slow exchange rates are not expected for the transient binding and dissociation of monomeric αS on a bilayer surface, but rather argue for the bound form of αS to be separated by a high energy barrier from the dynamically disordered solution state. Even while no significant broadening of the αS resonances is observed upon addition of lipid (Figure 1A), this does not exclude the possibility of slow exchange processes between the random coil and immobilized states of the protein. Provided that such processes take place on a time scale that is not much slower than the inverse of the transverse relaxation rate, R2T, of the TROSY component65 of the 15N transverse magnetization, they can be probed conveniently by measurement of these TROSY transverse relaxation rates.66,67
In the slow exchange limit, the observed R2T equals the sum of the R2T of the highly mobile random coil state, R2Trandom_coil, and the forward rate of the free-to-bound transition, kon. R2T values in the lipid-free state for αS are very low, ca 2 s−1 (Figure 4) allowing straightforward detection of even relatively small contributions resulting from kon. With the exception of several residues for which the intrinsic random coil hydrogen exchange rate is highest,68 and for which this type of exchange impacts the apparent R2T value, transverse relaxation rates in the absence of lipid are quite uniform and very low, in agreement with a highly mobile, disordered random-coil-like conformation.31,32,69,70 In the presence of lipids, a significant increase in R2T is observed for residues 1-100 (Figure 4), indicative of kon values in the 3–5 s−1 range.
The above analysis is oversimplified and strictly speaking applies only to cases where the equilibrium involves a single observable state. As discussed above, the case of αS is more complicated because the observed resonance intensity corresponds to the sum of completely free αS and the motionally unrestricted segments of the various different bound forms of αS. For example, for αS engaged in the SL1 type binding mode, residues 25-140 contribute to the random coil intensities observed in Figure 1B. One may expect that αS molecules whose N-terminal 20–24 residues have been immobilized by binding have shorter T2T values for observable residues that immediately follow this bound segment, thereby giving rise to an increase in the apparent R2T, even if kon ≈ 0. The absence of significant multi-exponentiality for the transverse magnetization decay curves for these residues suggests the presence of rather sharp transitions between ordered and disordered residues, as observed previously for amyloid fibrils.71
A second complication that arises in the analysis of the R2T data relates to the possibility of exchange between different bound forms. For example, the NMR observable resonances for the 25–100 region of SL1 will become NMR invisible upon an SL1 → SL2 transition, which may occur at its own specific rate. Therefore, each TROSY transverse relaxation process could occur with different R2T values for the different components that contribute to a given random coil resonance position. In practice, decay curves are reasonably close to mono-exponential (Supplementary Material, Figure 2), indicating that transitions between different bound forms, if present, do not occur on a timescale that is much faster than the random-coil to lipid-bound transitions.
A second procedure to evaluate the timescale of the exchange process is based on saturation transfer NMR experiments. By selectively saturating the magnetization of the phospholipid methylene resonances at 1.16 ppm, magnetization transfer to the αS amide protons manifests itself in an attenuation of a subsequently recorded HSQC spectrum (Figure 5). Experiments were carried out in the same manner as used previously to monitor magnetization exchange between aliphatic resonances of the detergent SDS and αS molecules 100% partitioned on the SDS micelle surface.35 The experiments were carried out in a difference mode, permitting quantitative measurements to be made of the fractional attenuation of the various resonances in the αS spectrum. The experiments show relatively uniform, ca 2%, attenuation for the NAC region of the αS backbone (residues 65-96). Although an increase in the NOE effect is evident towards the N-terminus, its magnitude is roughly proportional to the fractional attenuation in the reference spectrum, i.e., it reflects the population of each residue in the lipid-bound “dark state”. The fact that magnetization can be transferred from the “dark” lipid-bound state to the isotropic free state is consistent with a transition from bound to free αS on a timescale that is comparable to the longitudinal relaxation rate of the αS amide protons (~1 s). Although the scatter in the attenuation profile increases towards the N-terminus, this effect is largely a consequence of the lower signal-to-noise ratio observed for the N-terminal residues as a result of their lower intensities in the presence of lipid (Figure 1,,2);2); indeed the attenuation due to NOE transfer from the lipid resonances appears quite uniform and gradual. However, as will be discussed below, this observation does not exclude differential contacts between protein residues and lipid molecules: 1HN-1HN NOE equilibration in the dark state is shown to be rapid, and will therefore largely smoothen out any such differential lipid-amide NOE transfer.
Lipid-bound residues of αS are likely to tumble at a rate that is determined simply by the Brownian diffusion of the “dark state” particle; this diffusion has been shown to be slow and therefore will cause very rapid transverse relaxation of the nuclear spin magnetization, precluding its contribution to signals in HSQC or TROSY-HSQC NMR spectra. However, slow exchange of molecules between the immobilized and the random coil states of the protein allows the characteristics of the NMR-invisible “dark state” to be probed indirectly via 1H-1H NOE correlations, effectively similar to the well-known transferred NOE (trNOE) experiment.72,73 We therefore recorded two sets of 3D NOESY spectra in the presence of 0.06% lipid. One experiment, HMQC-NOE-HMQC,74,75 labels in the F1 dimension the 15N frequency of the amide from which proton magnetization originates prior to NOE mixing, and then reads out the 15N (F2) and 1H (F3) frequencies of the observed amide group. In another, complementary 3D NOESY-HMQC experiment,76 the frequency of the originating amide proton is mapped on the F1 dimension of the NMR spectrum, with the F2 and F3 dimensions again revealing the frequencies of the spins involved. This complementary set of 3D NOESY spectra allows identification of both the 1H and 15N frequencies for each of the two amides associated with any given NOE cross peak, thereby resolving any ambiguities arising from extensive spectral degeneracy in any individual 1H or 15N dimension. The experiment utilized only amide-selective pulses, of the EBURP and REBURP variety,77 allowing the experiment to be repeated relatively rapidly,78 thereby affording high spectral resolution in each of the indirect dimensions.
For spins which are located on a protein initially in the random coil state, but which exchanges to the dark state and back again during the NOE mixing time, the transfer of magnetization from each amide proton to its proximate spins, taking place during the time the backbone is immobilized, will be observed as NOE cross peaks. Indeed, the 3D NOE spectra (Figure 6) show large numbers of i to i ± n NOE cross peaks (up to n = 6) for this perdeuterated protein, despite a relatively short (100 ms) NOE mixing time. A control experiment, recorded in the absence of lipids, shows virtually no NOE intensity other than very weak sequential (n = 1) connectivities (data not shown), confirming that the NOE buildup has taken place in the dark state. The majority of the observed signal (ca. 75%) in fact corresponds to spins that never have exchanged to the dark state during the NOE mixing time, and these spins give rise to intense diagonal resonances. Sequential cross peaks between adjacent amides, however, correspond to spins that have switched to the immobilized state and back during the NOE mixing period, as a result of physical exchange of protein molecules from the free state to the “dark state” and back. While immobilized, NOE transfer of magnetization is fast, giving rise to extensive spin diffusion; therefore, for molecules that contribute to (i, i ± 1) cross peaks, diffusion of magnetization from i to i ± 2, i ± 3, and even further is also observed. Indeed, for the vast majority of amides in the SL1 and SL2 regions, HN-HN connectivities to eight or more amide protons adjacent to each other can be detected at contour levels somewhat lower than shown in Figure 6. The rapid spin diffusion along the peptide chain, together with the universal decrease in i to i ± n HN-HN cross peak intensity with increasing n, indicate that sequential distances (|n| = 1) are shorter than medium or long range HN-HN distances (|n| ≥ 2), strongly pointing to a largely helical conformation for the dark state, in agreement with the α-helical circular dichroism results. The absence of any long-range NOEs (n > 8) is consistent with such an α-helical conformation.
A second, simple experiment to probe NOE magnetization transfer in the dark state selectively inverts 1H magnetization of individual types of methyl groups. The data shown in Figure 7 illustrate this experiment for Leu Cδ1H3. Transfer of magnetization to nearby amides after an NOE mixing period can then be measured by recording the difference between spectra with and without methyl group inversion. For this purpose, a perdeuterated αS sample with specific methyl protonation of Ile, Val, and Leu was prepared79 and again examined in a sample containing 0.06% lipids. In the 13C spectrum, the Leu Cδ1 (24.5 ppm) nuclei resonate sufficiently downfield (> 4 ppm) from Val Cγ1, Val Cγ2, and Ile Cδ, such that their attached protons can be inverted selectively by a bilinear pulse80,81, while also partly inverting the Leu Cδ2H3 proton spins at 23.1 ppm. αS contains four Leu residues, at positions 8, 38, 100, and 113 in the sequence, and the Leu Cδ1H3 → HN NOE spectrum (Figure 7A) shows cross peaks only to the intraresidue amides and sequentially adjacent amide groups. This NOE transfer must arise from the corresponding Leu Cδ1H3 group while in the dark state, and the NOE-difference effect is large for Leu 8 and Leu 38 and extends over a significant number of residues (Figure 7B). In contrast, only weak intraresidue and (i, i+1) cross peaks are observed for the dynamically disordered residues Leu 100 and Leu 113 in the 2D NOE difference spectrum (Figure 7).
Dynamic light scattering (DLS), transmission electron microscopy (TEM), and cryo-electron microscopy (cryo-EM) confirm the lipid SUVs in the absence of αS to be nearly spherical unilamellar vesicles of ca 25 nm diameter (Figure 8A). However, addition of αS at the relatively high protein-to-lipid ratios used in our NMR experiments seriously disrupts these SUVs and results in a rearrangement forming much larger structures, including multilamellar vesicles, branched vesicles, and tubular structures (Figure 8b,c,d). DLS is strongly weighted towards the presence of large particles and, under all conditions studied, reveals the presence of a small fraction of very large aggregates (>5 micron), prohibiting characterization of the size distribution of smaller particles in the solution. Both TEM and cryo-EM also indicate the formation of large structures. The TEM measurements were, however, carried out following dehydration with the substrate adhering to a lipophilic thin carbon coating; this procedure may therefore affect the structures formed and observed by TEM. Similarly, sample blotting prior to cryo-EM flash freezing has an unknown effect on the protein:lipid ratio, and adhesion to the lacey carbon mesh of the grid may impact morphology of the protein-lipid aggregates, making it difficult to evaluate to what extent the images (Figure 8) reflect the distribution of particles in solution.
In order to characterize the size of the “dark state” particles under the conditions of our NMR sample solutions, we therefore have also investigated the particles by translational diffusion rates as measured by NMR pulsed field gradient (PFG) techniques.82 The diffusion rates of both the protein and the lipid SUVs were measured as isolated species, as well as in mixed samples. Because the 1H signals of the fatty acid lipid signals overlap with those of the aliphatic αS signals, measurements were carried out on perdeuterated αS, using its amide signals as markers for the protein diffusion rate, and the terminal alkane methyl signals for the diffusion of the lipids. When observing amide 1H signals in a PFG diffusion experiment, care needs to be taken to avoid interference between the rates of diffusion of water and protein molecules, as amide protons can exchange with water during the relatively long mixing period of the diffusion experiment. In order to minimize this effect, measurements were carried out at 10 °C, pH 6, where the intrinsic hydrogen exchange rates for most amide protons can be calculated and fall well below 1 s−1.68 The impact of hydrogen exchange on the measured diffusion decay curve can also be reduced by using the shortest possible diffusion delay, during which hydrogen exchange between solvent and protein is taking place, but this limits the measurement for slowly diffusing large particles where both longer diffusion delays and strong gradients are required. Instead, we therefore mitigate the impact of hydrogen exchange on the PFG diffusion measurements by starting the measurements at gradient strengths, Go, where the rapid initial intensity decay caused by diffusion of free water is mostly complete (compare Figure 9 and Supplementary Figure S5).
Experimental measurements of the hydrogen exchange rates for free αS are generally in good agreement with the values predicted for unstructured peptides68 (Supplementary Material), as expected for this intrinsically unstructured protein. In the presence of lipids, however, the apparent hydrogen exchange rate measured from the effect of presaturating the water on the intensity attenuation of the amide signals35 becomes much greater (Supplementary Material). This can be attributed to rapid exchange of protons of the serine and threonine hydroxyl groups with solvent, which then causes rapid equilibration of their nuclear spin magnetization with backbone amide protons through homonuclear NOE effects in the slowly tumbling “dark state”. Consequently, amide signal decay in the PFG diffusion experiment of lipid-containing αS samples is governed by a combination of the diffusion rates of water, free αS, and lipid-bound αS. With a PFG mixing time (500ms) that is considerably longer than the inverse of the free to lipid-bound αS conversion rate (ca 200 ms), the tail of the PFG decay curve at high gradient strengths is governed by the average of lipid-free and lipid-bound αS, weighted by the relative amide intensity observed for these two states. Under the low lipid conditions (See Figure 9), analysis of the intensity attenuation pattern indicates that 68% of the total integrated amide 1H intensity corresponds to lipid-free αS. With the diffusion rate of free αS known from a separate measurement, this results in a diffusion rate for lipid-bound αS of 4.1 × 10−11 m2/s (Table 1). While this diffusion rate is only 30% slower than for free αS, it corresponds to a radius of hydration, Rh of 37 ± 2 Å, or to a total mass of ca 150 kDa if the lipid-bound state is assumed to be relatively compact. Under the high-lipid conditions of Figure 9, where effectively all of the αS is lipid bound, the signal decay as a function of gradient strength is observed to be highly non-exponential, even at gradient strengths where all the 1H2O signal has decayed, and decay becomes much slower than even that observed for isolated SUVs (Figure 9), pointing to particle sizes much larger than that of the SUVs. The highly non-exponential decay of the lipid-bound αS indicates the presence of a wide distribution of particle sizes, ranging from smaller than the 25 nm SUVs to much larger aggregates, ca. 150 nm, consistent with the cryo-EM observations (Figure 8).
The ability of solution-state NMR to measure processes taking place over a wide range of timescales has made it a powerful tool for probing the dynamics of αS in the presence of lipid vesicles. Although the long correlation time of lipid-bound αS precludes direct NMR observation of the lipid-bound state of αS, the phenomenon of slow exchange between bound αS and its disordered free state provides a convenient condition for spying on the ‘invisible’ lipid-bound state which is of particular interest in the present study. The slow exchange kinetics result in the observation of narrow lines for the disordered state, and allow us to capitalize on conventional NMR techniques to probe structural and kinetic details of the αS/lipid complex. Thus, transverse relaxation measurements of the backbone 15N nuclei have been utilized to determine the rates at which αS exchanges from a disordered random coil state to the lipid-containing “dark state”. In addition, NOE transfer experiments have provided information about the contacts made between the protein backbone and lipid moieties, and also reveal dynamical characteristics of the bound state.
In stark contrast to detergent micelles studied previously, a marked loss of NMR signal intensity in the 1H-15N HSQC spectrum of αS is observed upon addition of lipid vesicles; residual intensity varies in a stepwise manner along the αS sequence, without significant concomitant line broadening of the attenuated resonances. This observation unequivocally establishes the presence of long-lived lipid-associated “dark” states that are characterized by long rotational correlation times. Although, at least in principle, formation of the “dark state” could be lipid-free, i.e., lipids could be acting as a catalyst for formation of annular or tubular pore-like oligomeric structures, previously observed in the absence of lipids,45 the magnetization transfer between lipid and protein resonances (Figure 5) argues strongly against the presence of lipid-free aggregates in our samples.
The attenuation profile of the amide proton NMR signals, which shows a series of more or less distinct plateaus when plotted as a function of position in the protein sequence (Figures 1,,2),2), provides evidence for a number of distinct lipid-bound species. As has been shown for flexible regions flanking structured regions of molten globule83 or amyloid fibril cores,71 the transition between immobilized and flexible domains can be quite sharp, spanning just a few residues. Moreover, αS increasingly partitions into the NMR-unobservable, membrane-bound state with increasing lipid concentration; indeed, even at the lowest concentrations (lipid αS molar ratio ≈ 1) nearly 50% of the protein population has its N-terminal region (residues 3-25) sequestered into the lipoprotein complex. Assuming all lipid is sequestered, this near-50% attenuation of signals of the N-terminal amide protons of αS in the presence of a near-equimolar concentration of phospholipids points to a lipid: αS stoichiometry as low as 2:1 for the lipid-bound state.
At lipid: αS molar ratios above ~15, the resonance intensities of the 40 C-terminal amide groups are also substantially attenuated, indicating that this region, which previously was thought not to interact with lipids, also undergoes a transition into an NMR-unobservable state. It is important to note, however, that neither the 3D transferred NOE spectrum nor the methyl-to-amide NOE difference spectrum (Figure 7) show any evidence for NOEs beyond what is expected in a dynamically disordered random coil for this group of C-terminal residues. Similarly, no NOEs were observed between the amides of these residues and lipids (Figure 5), even in the presence of much higher lipid concentrations than those used for Figures 5–7, where attenuation of these amide signals is substantial (Figure 2B). Interestingly, the “break points” in the attenuation profile of the C-terminal residues (Figure 2) coincide closely with the presence of proline residues (at positions 108, 117, 120, 128 and 138). Knowing that cis-trans isomerization of X-Pro peptide bonds, even in unstructured polypeptide chains is slow at the temperatures used in our study (10–20 °C), and that formation of cis X-Pro bonds is favored in non-polar environments,84,85 we speculate that attenuation of the NMR signals for the C-terminal 40 residues of αS is caused by a shift in equilibrium to the cis conformation for a fraction of the X-Pro peptide bonds in this C-terminal region. Increased lipid affinity of these cis-containing protein fractions causes their signals to be invisible in the presence of lipids, yielding the observed attenuation of the remaining all-trans-peptide-bond protein signals. The slow kinetics of the cis-trans peptide bond equilibrium therefore could be a major factor in preventing us from studying the conformation of this region of the protein in the lipid-bound state.
If αS were to be bound on the surface of SUVs, with its N-terminal 100 residues in a contiguous α-helical conformation, as concluded from recent EPR measurements59,60 it would occupy a minimum of 1400 Å2, an interface that corresponds approximately to the surface area of 28 phospholipid headgroups. With two leaflets per bilayer, the minimal stoichiometry for such a binding mode therefore requires at least 56 lipids per αS molecule, assuming the surface of an SUV to be 100% covered by αS. Our present study, which shows evidence for stable lipid-αS interactions down to a 2:1 stoichiometry, clearly requires an alternative lipid-binding mode. On a more qualitative note, the slow exchange kinetics between free and bound αS also appear inconsistent with such a “helix on a surface” binding mode as one would expect fraying at the ends of the helices, where there would be rapid exchange between bound and free states, and concomitant chemical shift changes and line broadening relative to the free state.
EPR studies have provided clear evidence for the helical conformation of αS in its lipid-bound state.59,60,86 The accessibility of this helical state to paramagnetic O2 and chelated Ni has unambiguously shown that the hydrophobic side of the αS helix is shielded from the solvent, with the hydrophilic side being solvent-accessible.60 We note, however, that binding modes other than a helix embedded on an SUV surface are also compatible with these EPR data. For example, a bundle of αS helices with a modest number of phospholipids at its core could equally agree with these observations. Indeed, our TEM and cryo-EM images of lipid vesicles in the absence and presence of αS demonstrate that the SUVs are grossly rearranged byαS, with the presence of numerous rod-like structures, many of which appear to emerge from the vesicles (Figure 8). Such rearrangements are consistent with earlier reports that even very low concentrations of αS can result in membrane disruption and permeability, as well as vesicle leakage.23,38,39,87
The composition of the small unilamellar vesicles used in our studies has been selected to mimic that of synaptic vesicles. In purified fractions of synaptic vesicles, phosphatidylserine accounts for 12% of the total phospholipid content,63 but an asymmetric distribution of PS, due to the activity of aminophospholipid translocase, nearly doubles the local concentration of PS on the cytosolic surface of the vesicle. In our studies, we observe no discontinuity in the data recorded for samples with lipid-to-αS ratios ranging from <1 to 150, but rather see a smooth progression in the NMR spectra and other biophysical characteristics from free and disordered αS to a completely sequestered form of the protein. We do not, for example, observe the onset of fast exchange dynamics, as we would expect if αS were to be surface-bound when the protein-lipid ratio is raised. This strongly suggests the presence of relatively stable, oligomeric lipid-bound species of αS, separated by a high energy barrier from free and highly unstructured αS. While the majority of experiments carried out in this study were conducted at dilute lipid conditions in order to optimize the quality of the NMR spectra, titration to increasingly higher lipid concentrations shows no discontinuities and qualitatively similar behavior is observed over the entire range of lipid: αS ratios (Figure 2); it is therefore likely that these conditions mimic well the physiological interactions of αS with synaptic vesicles. Our data also suggest that, in a non-diseased presynaptic environment, the population of free αS is very low, with virtually all αS molecules bound to lipids, the majority through their N-terminal 100 residues, but a substantial fraction additionally engaging lipids through their C-terminal 40 residues. The latter mode of binding is likely to be modulated by the cis-trans equilibria of the peptide bonds preceding the five proline residues located in this region, and provides a possible explanation for the recently observed presence of the cis-trans isomerase protein Pin1 in Lewy bodies (Ryo et al 2006).
Behavior very similar to that seen here for the SUVs designed to mimic synaptic vesicles has been observed for a variety of compositions of phospholipid vesicles, with the αS-membrane affinity being enhanced by negatively charged phospholipids. Fluorescence correlation spectroscopy results point to a αS affinity for 1-palmitoyl-2-oleoyl-sn-glycero-3-phospho-L-serine (POPS)53 that is comparable to the Kd of 50 μM, estimated from our study.
The presence of a set of stable but distinctly different binding modes, which engage different lengths of the αS sequence, differentiates our results from earlier studies. It is interesting to speculate how the presence of these stable binding modes may relate to αS fibril formation, often considered the hallmark of Parkinson’s disease. Biophysical characterization has implicated the so-called NAC region of the protein (residues 65-95) to engage in the extended β-sheets that make up the fibrils.28,29 Clearly, such fibril formation is not directly possible for the SL2 binding mode, where all 100 N-terminal residues are in a helical state, but it would be readily possible for SL1, where only the 25 N-terminal residues are found to be sequestered in a helical state. In fact, one would expect the SL1 binding mode to promote fibril formation as the anchoring of the N-terminal region of many αS proteins in the “dark state” particle leaves a very high local concentration of dynamically disordered fibril-formation-prone NAC regions on its surface.27 These results also explain the previous paradox, where the presence of low concentrations of phospholipids was found to promote fibril formation,37–42 as they would correspond to elevated SL1 binding modes, whereas high phospholipid concentrations were found to protect against fibril formation.37–42 The latter result is also expected on the basis of our results as high phospholipid concentrations promote not only the SL2 binding mode, but also binding modes that are likely to protect the C-terminus, protecting the NAC-region from forming β-sheet. Prior explanations for the increased fibril formation at low lipid or detergent concentrations simply focused on the increased local concentration of αS;27,40 our results suggest that it is the anchoring of the α-helical N-terminal domains of αS molecules which not only results in an increased local concentration of unstructured NAC domains, but also causes them to be lined up in a roughly parallel arrangement, thereby promoting the formation of parallel β-sheet fibril structures.
The disease-associated αS gene-triplicating mutant88 increases the αS concentration in vivo and thereby promotes the SL1 binding mode, elevating the concentration of fibril-formation-prone NAC regions by much more than just the effect of the increased cellular concentration. Similarly, it is likely that the disease-causing A30P mutant,4 which is helix-disrupting, will shift the equilibrium of binding modes towards SL1 because the helical conformation of residue 30 in the native binding modes (i.e. SL2) is energetically unfavorable. In this respect it is interesting to note that the A30P mutation increases, relative to wild-type, the population of free αS in an NMR study employing POPA/POPC vesicles.13 NMR studies of the lipid-binding properties of this mutant, as well as the other disease-causing mutants E46K and A53T, are currently in progress.
Human wild-type αS cloned in the kanamycin-restricted pET41a vector was overexpressed in E. coli. For NMR samples, 15N/2H isotopic enrichment of the protein was achieved by culturing cells in M9 minimal media: 6 g/L Na2HPO4·7H2O, 3 g/L KH2PO4, 5 g/L NaCl, 1mM MgSO4, 100 μM CaCl2, 3.0 g/L perdeuterated D-glucose, 1 g/L 15NH4Cl, and 1 g/L 15N,2H rich growth medium supplement (Celtone, Cambridge Isotope Laboratories, Andover, MA) in D2O. For protein with specific 13C1H3 labeling of valine, leucine, and isoleucine sidechains, E. coli was grown without rich growth medium supplement and the precursors [3 2H] α-ketoisovalerate (100 mg/L) and [3,3 2H2] α-ketobutyrate (50 mg/L) were added an hour prior to induction according to the protocol of 79. Protein expression was induced with 1mM IPTG at A600 = 0.6–0.8 for 4.5 h. Lysis of the cells and purification of the protein was achieved by freeze-thaw cycles followed by heat precipitation (15 min, 80 °C) in 50mM Tris-HCl, pH 7.4, 500mM NaCl. Anion-exchange chromatography on a Q-Sepharose column (Amersham Biosciences, Piscataway, NJ) was employed as a final purification step and yielded αS in > 98% purity as analyzed by SDS-PAGE. The purified protein was dialyzed into water and lyophilized for storage.
Phospholipids were purchased as a lyophilized mixture (5:3:2 DOPE:DOPS:DOPC) from Avanti Polar Lipids (Alabaster, AL) and used without further purification. The lipid mixture was weighed out and 20mM Na2HPO4 pH 7.0 added to achieve a15% (w/v) opaque, viscous slurry. The mixture was alternately subjected to vortex mixing and probe sonication to achieve vesicles of small diameter (20–40nm as determined by dynamic light scattering, data not shown. After sonication, the vesicle solution was hazy blue in color and much less viscous than the starting suspension. More dilute solutions of SUVs were made as necessary for experimental purposes, but were always diluted from a stock solution prepared as described above at 15% (w/v).
NMR spectra were recorded at 293K on Bruker spectrometers operating at 1H frequencies of 600, 750, and 800 MHz equipped with cryogenic (600 and 800 MHz) or room temperature (750 MHz) probes. Data were processed using NMRPipe software89 and analyzed by NMRPipe and Sparky software (Goddard TD, Kneller DG, unpublished, UCSF). Samples were prepared from lyophilized protein dissolved in 20mM Na2HPO4, pH 6.0, 94%/6% H2O/D2O, 0.02% NaN3, and concentrations were determined by UV-Vis spectra using an extinction coefficient of ε280 = 5120 M−1cm−1.
For experiments in which lipids were titrated into the sample, HSQC spectra were recorded with a data matrix consisting of 1280 (t2, 1H) × 448 (t1, 15N) complex points, t2,max = 152 ms, t1,max = 334 ms. Lipid to amide resonance NOE transfer experiments were recorded in difference mode as two interleaved HSQC experiments, applying a 1.5 s presaturation pulse with an RF field strength of 21 Hz centered at either 1.16 or −10 ppm. The 15N TROSY experiments for measurement of R2T 66,90 were recorded as nterleaved experiments with twelve T2 delays (20, 30, 50, 70, 90, 120, 150, 180, 220, 270, 400, and 500 ms) and acquisition times matching those of the regular HSQC experiments. The 3D HMQC-NOE-HMQC spectrum was recorded with a mixing time of 100ms and at a 1H frequency of 600 MHz, as a data matrix consisting of 2048 (t3, 1H) × 84 (t2, 15N) × 84 (t1, 15N) complex points, t3,max = 106 ms, t2,max = 60.8 ms, t1,max = 60.8 ms. The complementary 3D NOESY-HMQC spectrum was recorded with the same mixing time and 2048 (t3, 1H) × 84 (t2, 15N) × 72 (t1, 1H) complex points, t3,max = 106 ms, t2,max = 60.8 ms, t1,max = 48 ms. Amide-selective EBURP and REBURP pulses,77 of 1.8 ms duration each and centered at 8.3 ppm were used for both experiments.
Leu-Cδ1H3 inversion experiments were measured by interleaving 1H-15N HSQC experiments where, in alternate scans, the Cδ1H3 proton resonance is inverted by means of a bilinear pulse, taking advantage of the fact that Leu 13Cδ1 resonates more than 5 ppm downfield from the other methyl groups of valine and isoleucine. A bilinear pulse,80,81 of the type 90x(1H)-τ-[180x(1H)/180x(13Cδ1)on/off]-τ-90x(1H), where the pulses in square brackets are centered relative to one another, and the 180x(13Cδ1) is a 7-ms REBURP-shaped pulse, and τ = 1.2 ms, was used for the selective inversion. The NOE mixing time between the bilinear pulse and the HSQC experiment was 250 ms.
Pulsed field gradient NMR experiments for determining self-diffusion rates of protein and lipid were carried out at a 1H frequency of 600 MHz, using a room temperature three-axis gradient triple resonance probehead. Experiments were recorded as one-dimensional spectra read out after variable strength, two-axis (yz) gradient encoding and decoding91 separated by a 500 ms diffusion delay. The reference intensity spectrum was taken at 6% of the maximum gradient strength (87 G/cm). All spectra were recorded at 10 °C.
For cryo-EM imaging, vitrified samples were prepared in a Vitrobot apparatus (FEI Company, Hillsboro, OR) at a relative humidity of 85%. A 5 μL sample volume of 0.03% DOPE:DOPS:DOPC SUV in either the absence or presence of 600 μM αS was applied to a plasma holey-carbon coated grid (Quantifoil, Hatfield, PA), and blotted with filter paper for 1 s to leave a thin film of solution. Blotted samples were immediately plunged into liquid ethane at its freezing point (−196 °C) and stored under liquid nitrogen prior to imaging in the microscope. Samples were examined using a Philips CM120 electron microscope, operating at 100kV, using a Gatan 626 cryo holder cooled with liquid nitrogen to temperatures below −180 °C. Digital images were acquired on a Gatan 791 MultiScan CCD camera with the DigitalMicrograph software package (Gatan, Pleasanton, CA).
Fluorescence anisotropy measurements were taken on a Jobin Yvon Fluoromax3 fluorimeter with automatic polarizers. Data were recorded at 25 °C with an excitation wavelength of 280 nm, 2 nm bandwidth, and 3 s integration. Emission values were recorded at 306 nm, the emission maximum for tyrosine. Each data point is the average of ten measurements.
This work was funded by the Intramural Research Program of the NIDDK, NIH (to AB) and by the Leverhulme Trust and the Wellcome Trust (to CMD). We thank James Masse, Nick Fitzkee, Frank Delaglio and Alex Grishaev for help with data analysis, Jenny Hinshaw (NIDDK), Dan Shi and Sriram Subramanian (NCI) for help with cryo-EM data collection and interpretation, and John Christodolou for valuable help in the initial phase of this study.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.