Remodeling of Lipid Vesicles into Cylindrical Micelles by α-Synuclein in an Extended α-Helical Conformation*

Background: Membrane fusion and fission events are effected by remodeling proteins. Results: Using cryoelectron microscopy, we observed the conversion of large spherical lipid vesicles into narrow protein-coated tubes. Conclusion: Tubulation is accompanied by α-synuclein switching into an extended α-helical conformation. Significance: The cylindrical micelles produced resemble a hemi-fission/fusion state of the membrane. α-Synuclein (αS) is a protein with multiple conformations and interactions. Natively unfolded in solution, αS accumulates as amyloid in neurological tissue in Parkinson disease and interacts with membranes under both physiological and pathological conditions. Here, we used cryoelectron microscopy in conjunction with electron paramagnetic resonance (EPR) and other techniques to characterize the ability of αS to remodel vesicles. At molar ratios of 1:5 to 1:40 for protein/lipid (1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoglycerol), large spherical vesicles are converted into cylindrical micelles ∼50 Å in diameter. Other lipids of the same charge (negative) exhibit generally similar behavior, although bilayer tubes of 150–500 Å in width are also produced, depending on the lipid acyl chains. At higher protein/lipid ratios, discoid particles, 70–100 Å across, are formed. EPR data show that, on cylindrical micelles, αS adopts an extended amphipathic α-helical conformation, with its long axis aligned with the tube axis. The observed geometrical relationship between αS and the micelle suggests that the wedging of its long α-helix into the outer leaflet of a membrane may cause curvature and an anisotropic partition of lipids, leading to tube formation.


␣-Synuclein (␣S) is a protein with multiple conformations and interactions. Natively unfolded in solution, ␣S accumulates as amyloid in neurological tissue in Parkinson disease and interacts with membranes under both physiological and pathological
conditions. Here, we used cryoelectron microscopy in conjunction with electron paramagnetic resonance (EPR) and other techniques to characterize the ability of ␣S to remodel vesicles. At molar ratios of 1:5 to 1:40 for protein/lipid (1-palmitoyl-2oleoyl-sn-glycero-3-phosphoglycerol), large spherical vesicles are converted into cylindrical micelles ϳ50 Å in diameter. Other lipids of the same charge (negative) exhibit generally similar behavior, although bilayer tubes of 150 -500 Å in width are also produced, depending on the lipid acyl chains. At higher protein/ lipid ratios, discoid particles, 70 -100 Å across, are formed. EPR data show that, on cylindrical micelles, ␣S adopts an extended amphipathic ␣-helical conformation, with its long axis aligned with the tube axis. The observed geometrical relationship between ␣S and the micelle suggests that the wedging of its long ␣-helix into the outer leaflet of a membrane may cause curvature and an anisotropic partition of lipids, leading to tube formation.
␣-Synuclein (␣S) 5 is a protein that localizes at neural terminals (1)(2)(3). ␣S is abundant in neural cells, accounting for up to 1% of total protein with estimated concentrations in the hundreds of micromolar range (4 -7). Its function(s) has not been precisely defined, but mutations in ␣S and duplication/triplication of the ␣S gene have been linked to familial forms of Parkinson disease (8 -11). Also, elevated levels of ␣S cause neurodegeneration in animal models (12)(13)(14)(15), reproducing the cytotoxic effects of ␣S that lead to Parkinson disease in humans. In the brains of Parkinson disease patients, amyloid deposits of ␣S accumulate in Lewy bodies (16,17), giving a hallmark of this pathology.
A relationship of ␣S to neural membrane trafficking has been indicated (18). The majority of ␣S is found at synaptic terminals (4,5), where it is seemingly involved in the control of synaptic vesicle formation and maintenance (5,19). An RNAi screen in Caenorhabditis elegans found 10 genes modifying ␣S neurotoxicity, four of which turned out to be related to the synaptic endocytosis pathway (20). Synapses of mice knocked out for ␣S showed diminished levels of synaptic vesicles in the reserve pool (2), whereas overexpression of ␣S is reported to inhibit re-clustering of synaptic vesicles, thereby reducing neurotransmitter release (15). These and other reports (21)(22)(23) suggest that ␣S interacts directly with membranes to control cycles of synaptic vesicle release.
␣S is also involved in defects of other membrane trafficking pathways, with cytotoxic consequences. Overexpression of ␣S causes disruption of endoplasmic reticulum-Golgi trafficking (14), fragmentation of the Golgi apparatus (24,25), and distortion of mitochondrial membranes (26), causing them to fission (27). Taken together, these reports suggest that ␣S interacts widely with multiple organelles, but in all cases, these interactions involve membranes. Thus, it is of fundamental importance to understand the mechanism(s) of ␣S-membrane interactions.
A number of in vitro studies have found the conformation of ␣S to be highly adaptable. It has long been known that ␣S is an intrinsically disordered protein in solution (28). This view has recently been challenged by suggesting that the protein exists as a natively folded tetramer (30,31), but further experimentation, including in-cell NMR (29,72), indicate that ␣S is an unfolded monomer in vivo. Upon long incubation, it passes through toxic oligomeric states to reach an amyloid conformation (32), rich in parallel in-register cross-␤ structure (33,34), which is considered to be its state in Lewy body accumulations. However, in the presence of negatively charged membranes (35), ␣S adopts an ␣-helical conformation (36). Its amino acid sequence contains seven 11-residue repeats that are predicted to form amphipathic ␣-helices that mediate its interaction with membranes; in this respect, it is reminiscent of apolipoprotein (37). Further study of the latter structures detected two forms, depending on the experimental conditions and lipids used, viz. 1) an extended helical form (38 -40) like a curved rod; and 2) a "horseshoe"like form with broken helices that close to hairpins on small vesicles (39 -45) and SDS micelles (41,46,47). Moreover, these two conformations are reported to co-exist (48,49) in the same sample preparation, and under some circumstances, more than half of the membrane-interacting region of ␣S can remain unfolded (50).
Overall, these data suggest that the different membranebound forms of ␣S are energetically similar. Nevertheless, the state of the membrane has been found to correlate with particular structures. EPR spectroscopy revealed that the extended conformation predominates when ␣S binds to small unilamellar vesicles (SUVs). In this extended conformation, the molecule is further twisted into a right-handed super-coil akin to a coiled-coil structure (39), and the interacted vesicle remain intact. In contrast, horseshoe-like structures are formed when ␣S binds to vesicles as they are converted into smaller nonvesicular structures (39).
In the presence of large negatively charged vesicles, ␣S has been observed to remodel vesicles into tubes (51,52). Circular dichroism (CD) data have shown that this remodeling event is accompanied by a conformational change in ␣S from random coil to ␣-helix (51). However, the underlying structural change of ␣S and the tubulation mechanism has not been established. In this study, we used cryo-EM, CD, and EPR spectroscopy in combination with more closely investigated ␣S-induced tubulation of vesicles.

EXPERIMENTAL PROCEDURES
Purification of Wild-type and Mutant ␣-Synuclein-␣S was prepared as described (51). In brief, wild-type and mutant human ␣S were expressed in Escherichia coli BL21 (DE3) pLysS cells, which were lysed by boiling, followed by acid precipitation. The supernatant was passed through anion exchange columns and eluted with a 0 -1.0 M NaCl gradient.
Lipids purchased from Avanti Polar Lipids Inc. (Alabaster, AL) as a solution in chloroform were transferred to a glass tube, and the chloroform was removed by blowing N 2 gas. After the lipids were further dried under vacuum in a desiccator for at least 6 h, buffer (20 mM Hepes, pH 7.4, 100 mM NaCl) was added, and the mixture was immediately used in experiments. Preparation and incubation of lipids with ␣S were performed above 32°C to avoid any effects from phase transitions. Nevertheless, consistent results were obtained in all cases with experiments performed at room temperature.
Phospholipid Vesicle Clearance Assay-Light scattering was measured as a function of time using a Jasco V-550 UV-visible spectrophotometer to monitor the interaction of ␣S with large lipid vesicles. A monitoring wavelength of 500 nm was used with a slit width of 2 nm and medium response time. Briefly, lipid vesicles were suspended in 20 mM Hepes (pH 7.4) with 100 mM NaCl at a final volume of 500 l in a quartz cuvette. Control vesicles (no protein added) did not show any change in light scattering.
Circular Dichroism (CD)-All spectra were recorded using a Jasco J-810 spectropolarimeter with a 1-mm quartz cell. A scan rate of 50 nm/min, bandwidth of 1 nm, 0.1 nm time response, and step resolution of 0.5 nm were set for all experiments. For time course experiments, the parameters were set at a 1-s data pitch, 4-s response time, and 1 nm bandwidth. Protein concentration was determined using the extinction coefficient at 280 nm based on the number of Trp and Tyr residues in the protein. Appropriate blanks were collected under similar conditions and subtracted to obtain the final spectra. 10 mM sodium phosphate buffer (pH 7.4) was used in all CD experiments.
Cryoelectron Microscopy-5-l drops were applied to holey carbon grids (Quantifoil) and vitrified in a Vitrobot cryo-station (FEI). The humidity was carefully controlled to avoid drying-related deformation of membranes, and the temperature was set to be above the phase transition temperature (T m ) of individual lipids. Specimens were observed with a CM200-FEG electron microscope at nominal magnifications of 38,000 and 66,000, with defocus settings in the range of Ϫ1 to Ϫ4 m. Film (SO163, Eastman Kodak Co.) was used for recording images, and digitization was done with a SCAI scanner (Carl Zeiss) at rates of 1.84 or 0.966 Å/pixel. Image Analysis-The contrast transfer function was partially corrected by phase-flipping. For averaging experiments, relatively straight tubes were selected, computationally straightened, and cut into segments using a box size of 270 Å. Image segments were binned 2-fold to increase the speed of the computation. For the analysis of ␣S-POPG at different protein-tolipid ratios, stacks of 1787 (1:40 data), 1290 (1:20 data), and 1216 (1:10 data) segments were compiled for classification and averaging. For the analysis of ␣S bound to various lipids, stacks of 4932 (DMPG), 3325 (POPG), 530 (DOPG), 944 (DLPG), and 111 (DAPG) segments were compiled for analysis of cylindrical micelles and 932 (DOPG), 552 (DLPG), and 359 (DAPG) segments for bilayer tubes.
Reference-free classification (k means) and averaging were performed using SPIDER (53) and EMAN (54). Principal component analysis was also applied for further classification of the aligned images. For the cylindrical micelles, we did not see evident morphological distinctions among averaged images, postclassification, and therefore we chose to display the average of the majority class from each experimental condition.
Fluorescence Measurements-Desiccated N,NЈ-dimethyl-N-(iodoacetyl)-NЈ-(7-nitrobenz-2-oxa-1,3-diazol-4-yl)ethylenediamine (IANBD) was obtained in amide form from Molecular Probes (Eugene, OR), kept at Ϫ20°C in the dark, and dissolved in DMSO prior to use. ␣S cysteine mutants were stored in 5 mM dithiothreitol (DTT). Immediately before labeling, each mutant was desalted in a PD-10 column (GE Healthcare) into 20 mM HEPES, pH 7.4, and 100 mM NaCl, with no DTT. IANBD was added at a 10-fold molar excess, and the protein sample was rotated at ϳ40 rpm either for 2 h at RT or overnight at 4°C. To separate labeled ␣S from free label, the sample was again passed through a PD-10 column. Samples were concentrated by centrifugation at 5000 relative centrifugal force in Millipore filter units (3000 M r cutoff). Absorbance spectra were taken from 250 to 500 nm in a Jasco V-550 spectrophotometer with a 1-cm path length quartz cuvette. Labeling efficiency was assessed with Equation 1, where A max, f is maximal absorbance of the fluorophore (478 nm for IANBD); ⑀ 280, p is the molar extinction coefficient of the protein (number of Tyr residues ϫ 1280 M Ϫ1 cm Ϫ1 ), and ⑀ max, f and ⑀ 280, f are the molar extinction coefficients of the fluorophore at maximal absorbance and at 280 nm (18,492 and 1250 M Ϫ1 cm Ϫ1 for IANBD, respectively). Emission spectra from 500 to 600 nm were taken in a Jasco FP-6500 spectrofluorometer with excitation wavelength set at 478 nm and excitation and emission slit widths of 3 and 5 nm, respectively. First, a spectrum was taken for 15 M protein alone. POPG vesicles were mixed in by pipetting up and down to a concentration of 300 M. After ϳ1 h to allow for maximal binding, the protein-with-lipid spectrum was taken.
Spectra were normalized by dividing every fluorescence value for both spectra of each ␣S mutant by the maximal fluorescence of the protein-alone spectrum for that particular mutant. This method set maximal fluorescence of protein alone to 1 and allowed for direct assessment of 5-8-fold fluorescence change upon addition of vesicles.
Spin Labeling of Cysteine Mutants and Continuous Wave-EPR-Cysteines were reduced by adding DTT to a final concentration of 1 mM to single and double cysteine mutants and then removing it by size exclusion chromatography on PD-10 columns (GE Healthcare) in 20 mM Hepes (pH 7.4), 100 mM NaCl buffer. The protein samples were incubated overnight at 4°C with a 5-fold molar excess of the 1-oxyl-2,2,5,5-tetramethylpyrroline-3-methyl-methane-thiosulfonate spin label. Excess spin label was then removed by size exclusion on PD-10 columns.
Continuous wave-EPR spectra were recorded from tubebound ␣S at a protein-to-lipid molar ratio of 1:20 in 20 mM Hepes (pH 7.4), 100 mM NaCl buffer. Tube-bound ␣S was separated from unbound protein by pelleting it in an ultracentrifuge spun for 1 h at 120,000 ϫ g. Spectra were recorded using a Bruker EMX X-band continuous wave-EPR spectrometer with a dielectric resonator at 1.59-milliwatt incident microwave power, using a field modulation of 1.5 G. The inverse central line width values were measured from the peak-to-peak distance of the central line as described previously (38). The O 2 and nickel (II)-ethylenediamine-N,N-diacetic acid (NiEDDA) accessibilities (⌸O 2 and ⌸NiEDDA) were obtained employing a power saturation method (55). The oxygen accessibility was measured in the presence of ambient oxygen, and the sample was equilibrated with 3 mM NiEDDA for NiEDDA accessibility.
Pulsed EPR and Distance Analysis-Samples were prepared at a protein-to-lipid ratio of 1:20. 20% of spin-labeled protein, containing two spin labels per protein, was mixed with 80% of unlabeled wild-type protein and incubated for 5 min before adding it to the vesicles. This mixture was incubated for 10 min and then ultracentrifuged for 1 h at 120,000 ϫ g to separate the tubes from unbound protein. Four-pulse double electron electron resonance (DEER) experiments were performed using a Bruker Elexsys E580 X-band pulse EPR spectrometer fitted with a 3-mm split ring (MS-3) resonator, a continuous-flow helium cryostat (CF935, Oxford Instruments), and a temperature controller (ITC503S, Oxford Instruments). All samples, in 15% sucrose, were flash-frozen, and data were acquired at 78 K. The data were fitted using Tikhonov regularization (56) as implemented in DEERAnalysis2011 packages (57). The background contribution from nonspecific interaction was subtracted, using a three-dimensional model for tube-bound ␣S. Tikhonov regularization was used with parameters of 100 or less obtained from the L-curve analysis to fit distances. The distances given in Fig. 7C correspond to the maxima in the Tikhonov distance distributions.

RESULTS
␣S Converts Vesicles into Cylindrical Micelles-␣S was added to preformed large spherical POPG vesicles at various molar ratios of protein/lipid in the range of 1:40 to 1:10, and the optical absorbance was monitored at 500 nm. An immediate fall-off was observed in each case (supplemental Fig. S1), indicating that the vesicles were breaking down into smaller entities. CD measurements performed in parallel showed that the conformation of ␣S was switching from random coil to ␣-helix (Fig.  3B).
To visualize the accompanying structural changes, samples were vitrified and observed by cryo-EM ( Fig. 1). At the lowest ␣S/lipid ratio (1:40), we found relatively short segments of narrow tubes, opening out into irregular forms at their ends ( Fig.  1A, red asterisks). The diameter of the narrow segments (ϳ50 Å) shows them to be cylindrical micelles (see below). Complexes of this kind, often tangled together, covered most fields (supplemental Fig. S2). On increasing the protein content, we observed longer, continuously thin, cylindrical micelles (Fig. 1, B and C). On further increasing the ␣S/lipid ratio to 1:5, tubes became rarer and were replaced by small discoid particles 70 -100 Å across (Fig. 1D).
To visualize the cylindrical micelles in greater detail, we first cut the images into 270-Å segments (this procedure minimizes the effects of curvature on the ensuing analysis). We then aligned them and classified them computationally to obtain relatively homogeneous sets (Fig. 2). The segments were then averaged in each class to reduce the noise level. Tubes formed at each of three ␣S/lipid ratios were analyzed separately. In each case, the averaged images show two peak densities, on either side of the tube. Some class averages show slight bends (e.g. Fig.  2, 3rd panel, top row and 4th panel, middle row). However, the main source of variability is in width. The thinnest tubes (48 -54 Å) were formed at all three protein/lipid ratios, but there is a greater prevalence of slightly wider tubes at ratios of 1:20 and 1:40 (widths summarized in Fig. 2 legend were mea-sured using the full width at half-maximum criterion for edge detection). Also, with decreased ␣S (1:40), there is a greater tendency of tubes to meander (supplemental Fig. S2).
The average width of all the remodeled tubes measured in the 1:40 experiment was 62 Å (S.D. 9.1 Å). The corresponding values were 56 Ϯ 2.3 Å for the 1:20 experiment and 50 Ϯ 1.6 Å for the 1:10 experiment. We take the two peaks to mark the posi-tions of the lipid headgroups, in which case the observed dimensions are consistent only with the tubes being micellar (a single lipid leaflet is ϳ20 Å thick (58)). The density peaks should also reflect, to some extent, protein bound to the outer surface. However, no definite protein-associated pattern is seen, and it appears that signal from the protein is smeared out and affects the observed density profiles primarily by damping the white interference fringes along the sides of the tubes. (These fringes result from phase-contrast imaging.) Moreover, the inferred ␣-helices are only about 10 Å thick and are likely to be partially submerged into the lipid layer (see EPR data below and "Discussion").
␣S-induced Membrane Remodeling Occurs with Lipids with Various Acyl Chains-The central core of a cylindrical micelle, seen in the density profiles as the part between the two peaks, is occupied by the acyl chains. Based on simple geometric considerations, short and saturated acyl chains should be more amenable to micellar tube formation, although longer and bulkier acyl chains might favor the formation of bilayer tubes. To test this hypothesis, we investigated the interaction of phosphoglycerol lipids with various acyl chains, ranging from the short chain 1,2-dimyristoyl-sn-glycero-3-phospho-(1Ј-rac-glycerol) to (DMPG) to the long chain and polyunsaturated 1,2-diarachidonoyl-sn-glycero-3-[phospho-rac-(1-glycerol) (DAPG) (supplemental Fig. S3).
As with POPG vesicles, ␣S reduced the light scattering from vesicles of various other phosphoglycerol lipids within minutes of addition and did so in a concentration-dependent manner (Fig. 3), again reporting the conversion of large vesicles into smaller entities. The most rapid remodeling was observed for DMPG vesicles, where the light scattering fell precipitously within a few seconds (Fig. 3A). These remodeling events were   accompanied by ␣S switching to an ␣-helical conformation (Fig. 3, B and C).
Acyl Chain Bulkiness Affects Tube Morphology-To compare the morphological changes in membranes with different acyl chains, we observed tubes formed at a fixed protein/lipid ratio (1:20) by cryo-EM (Fig. 4, B-F). Two types of tubes were observed as follows: cylindrical micelles and wider tubes, with varying diameters. It is apparent, both from the images directly and from averaged density profiles, that the wider tubes are tubular bilayer membranes (Fig. 5, bottom row). The ratio of micellar versus bilayer tubes correlated with acyl chain size. In the case of DMPG, which contains the shortest acyl chains (14 carbons), we observed exclusively cylindrical micelles. As the bulkiness of the acyl chains (i.e. chain length and degree of unsaturation) increased, more bilayer tubes were observed (72% with DAPG). The averaged images for the major classes of cylindrical micelles obtained with the respective lipids yielded widths of 60 Ϯ 1.3 Å for DMPG, 56 Ϯ 2.3 Å for POPG, 44 Ϯ 1.9 Å for DLPG, 52 Ϯ 2.9 Å for DOPG, and 40 Ϯ 0.8 Å for DAPG (Fig. 5). The thicknesses of the membranes of the bilayer tubes, measured in similar fashion, were 48 Ϯ 2.0 Å for DLPG; 49 Ϯ 1.0 Å for DOPG; and 46 Ϯ 1.6 Å for DAPG. With DMPG and POPG, no bilayer tubes were observed.
It is not clear why there are differences in tube widths for different kind of lipids; nevertheless, in general, the width of bilayer tubes increased for lipids containing more unsaturated acyl chains. This trend led even to ameba-like shapes on adding DAPG (Fig. 4F), which has eight double bonds in its acyl chains (supplemental Fig. S3). We also performed some preliminary experiments with other anionic lipid compositions and observed similar effects (supplemental Fig.  S4).
␣S Switches to an ␣-Helical Conformation on the Membrane Surface-Next, we addressed the conformational change in ␣S that accompanies vesicle tubulation. The CD spectra indicate that ␣S increases in ␣-helicity (Fig. 3, B and C). To further characterize this transition, we performed fluorescence spectroscopy. ␣S was labeled with the polarity-sensitive IANBD at various positions along its amino acid sequence. In all cases, the fluorescence was generally low in solution (for example, residue 31, Fig. 6A, black dotted line). Upon exposure to membranes, the fluorescence of all N-terminal sites (residues 31, 48, 52, 70, and 76) increased markedly (Fig. 6A), suggesting that this region interacts directly with the membrane. In contrast, two C-terminally labeled derivatives (at positions 124 and 136) did not show any detectable change in fluorescence, implying that the C-terminal region does not interact with the membrane. A minor change was detected when residue 100 was labeled, suggesting that the end of the membrane-interacting region is close to this position. These data are consistent with the structure of ␣S in its vesicle-bound form, where the N-terminal but not the C-terminal regions interact with the membrane (39).
␣S Adopts an Extended ␣-Helical Conformation-The N-terminal region of ␣S harbors seven 11-residue repeats and previ-ous studies of vesicle-bound ␣S demonstrated that each repeat forms three ␣-helical turns (39). Within a given repeat, positions I, V, and IX were found to face away from the membrane surface, whereas positions III, VII and XI face into its interior (Fig. 6B). To further investigate the conformational change of ␣S and possible similarities between tubule-bound and vesiclebound ␣S, we employed site-directed spin labeling together with continuous wave-EPR spectroscopy and R1-spin labeling of the residues shown in Fig. 6B. The respective X-band EPR spectra indicate a transition from a highly dynamic structure in solution to an ordered structure on the tubulated membrane, as with vesicle-bound ␣S (supplemental Fig. S5). The spectra are missing strongly immobilized components, indicating the lack of tertiary or quaternary contacts. These spectra are completely different from those produced by the amyloid fibrillar form of ␣S in which a cross-␤ conformation is adopted (32,34).
Next, we performed accessibility measurements using the paramagnetic colliders NiEDDA (⌸NiEDDA) and O 2 (⌸O 2 ). Accessibility to the more hydrophobic O 2 molecule is strongly enhanced in lipid phases, although NiEDDA preferentially partitions into the aqueous phase. Using these reagents, we calculated the depth parameter (ln(⌸O 2 /⌸NiEDDA)), an established measure of membrane immersion depth (59). -values were plotted for each residue in both tube-bound and vesiclebound situations (Fig. 6C). First, we observed a clear positive correlation in -values between tube-bound and vesicle-bound ␣S. This means that the two conformations are nearly identical. Second, the -values can be grouped into two sets (green and red circles in Fig. 6C). Residues 31 and 76 in the green circles in Fig. 6C have reduced -values, located in positions IX, V, and I on the helical wheel (Fig. 6B, colored in green) on the solventexposed surface. Residues 37,41,44,48,52, and 70 have larger -values (Fig. 6C, red circle) and are membrane-inserted. Of particular note are residues 41 and 44 that are ␣-helical in the extended conformation but are part of an inter-helical loop in the broken conformation (47). The former conformation is assumed when ␣S binds to vesicles and the helix breaks into a horseshoe shape upon exposure to small amounts of SDS. These results indicate that on tubular membranes ␣S assumes an extended, amphipathic, and ␣-helical conformation, as in its vesicle-bound mode.
To further test the way in which ␣S takes up an extended structure on tubes, we measured the distance between two spin-labeled sites by 4-pulse DEER experiments. Three key distances were measured between residues 11R1/26R1, 22R1/ 52R1, and 63R1/81R1 (Fig. 7A). Fits of the time evolution data (Fig. 7B), performed using Tikhonov regularization (56, 57), yielded distance distributions consistent with those expected for an extended helical structure. To confirm the robustness of these distance distributions, we also used Gaussian fits. The peaks moved by only small amounts (Fig. 7, legend), attesting to their robustness. Of particular note, the 22R1-52R1 spacing show the solvent-exposed residues, and red circles show lipid-exposed residues. C, on small unilamellar vesicles (SUVs) plotted against ⌽ on tubes. The ratios of the accessibilities to O 2 and NiEDDA for residues are expressed by the depth parameter ϭ ln(⌸O 2 /⌸NiEDDA), with increasing values indicating deeper membrane immersion. FIGURE 7. Spin-label intramolecular distances from 4-pulse DEER experiments. A, positions of double mutants in a bent helix (SDS-bound structure of ␣S) versus extended helix (SUV-bound structure of ␣S). B, 1st column depicts the dipolar evolution for each of the indicated spin label pairs. 1st column, black traces denote background-corrected experimental data and the red curves depict fits made using Tikhonov regularizations. 2nd column shows the resulting inter-molecular distance distributions. To optimize signal-to-noise ratios, short acquisition times were used for shorter distances, and long acquisition times were used for longer distances for which dipolar evolutions are of lower frequency. For shorter distance scans, the base-line correction was verified by scans of longer time base. Because of significantly better signal-tonoise ratios, however, shorter scans were used to carry out Tikhonov regularization. The increased uncertainty arising from shortened time scans was compensated by enhanced signal-to-noise ratios. Because the spectra were somewhat noisy, we also calculated the distance distributions by Gaussian regularization (data not shown) and found that the peaks moved only by small amounts, from 26 to 27 Å, 50 to 45Å, and 28 to 27Å, respectively. C, comparison of experimental spin label inter-electron distances obtained for ␣S bound to tubes with an ideal helix, ␣S bound to an SUV, SDS micelle, or SLAS micelle. was determined to be 45-50 Å. The corresponding distance was measured to be 23 Å for the broken helix (Fig. 7C, SDS/ SLAS) and 49 Å for the extended helix (Fig. 7C, ideal helix). These distances also showed consistency between the tubebound ␣S and the vesicle-bound ␣S (Fig. 7C) (39). In sum, these data strongly indicate that ␣S forms an extended helical structure on tubes, like that of vesicle-bound ␣S. Similar results were obtained for all phosphoglycerol tubes that we tested (other data not shown).

DISCUSSION
Previously, we showed that when large POPG vesicles are exposed to ␣S, the protein adopts an ␣-helical formation as it remodels the vesicles into tubes (51). This study characterizes the tubulation reaction further. We have found with protein/ lipid ratios between 1:10 and 1:40 that most tubes produced are cylindrical micelles, a membrane topology previously described for lipid/detergent mixtures (60 -62) and for some lipoprotein tubules induced by the endocytic protein endophilin (63). The cylindrical micelles cannot be mistaken for amyloid ␣S fibrils because of the following. 1) Cryo-EM shows them to have a low density core, unlike ␣S amyloid fibrils (see Fig. 1A in Ref. 33). 2) They are observed to diverge continuously into wider tubes (Fig. 4E), again unlike amyloid fibrils. 3) Tubes form within a few seconds of incubation whereas amyloid fibrillation takes many hours. 4) Our CD and EPR experiments reveal the presence of ␣-helix as opposed to the ␤-sheets of amyloid, consistent with the previously reported lack of thioflavin T fluorescence (51). 5) EPR shows the ␣S to be interacting with a membrane, whereas fibrils are made of ␣S alone.
With some dependence on the protein-to-lipid ratio used and on the acyl chains of the lipids, we found that the majority species formed in each case was cylindrical micelles. In these assemblies, the lipid is organized in what is often referred to as the "hemifusion" or hemifission" states. These states have been much discussed but seldom observed directly. Our study shows that, under appropriate conditions, they can be an abundant and stable assembly form.
Although the inferred transition from initial to final state was not captured in detail, we observed hints as to how it may proceed. When specimens were vitrified immediately after mixing ␣S and POPG vesicles, an initial elongation of vesicles seemed to precede tube formation (supplemental Fig. S6). The observation of bilayer tubes at lower protein/lipid ratios suggests that they could be intermediates in a pathway that leads at higher protein concentration to micellar tubes.
Anisotropic Interaction of ␣S along the Tube Surface-The surface of ␣S-induced tubes appears rather smooth, even though the protein is present on the tube surface, according to gold labeling EM (51) and the present EPR and fluorescence data. The extended amphipathic structure indicated by the latter data accounts for the difficulty to visualize the protein. The assigned ␣-helix length is ϳ140 Å. Given that the cylindrical micelles are only about 50 Å in diameter, the helix should run nearly parallel to the tube axis if it is to remain in continuous contact with the membrane (Fig. 8). A single ␣-helix generates little contrast in cryo-EM and that of ␣S may be partly sub-merged in the lipid layer, providing an explanation for why we do not explicitly visualize coating molecules.
When ␣S attaches to the membrane surface at the lowest ␣S concentration used (1:40), we observed some slightly wider tubes with less uniform diameters (Fig. 2, 1:40, 2nd panel). This feature could indicate that the more uniform width of the narrower tubes reflects a denser (saturating) packing of ␣S molecules on the surface of the tube.
We can estimate the protein/lipid ratio at saturating binding of ␣S to tubular micelles with partially embedded 140-Å-long ␣-helices aligned with the tube axis and separated by single rows of lipid molecules. The observed widths of the cylindrical micelles and that of an ␣-helix (ϳ10 Å) suggest that there would be a maximum of 8 -10 ␣-helices in cross-section. If we allow ϳ15 lipids per molecular length of ␣S, we obtain molar ratios in the range of 1:15. Although this estimate awaits experimental testing, it is consistent with our observation that further increasing the content of the lipophilic protein correlates with switching the system toward a different kind of complex (the discoid particles) that is more protein-rich.
Mechanism of Curvature Induction-An extended helical structure running almost parallel to the tube axis would provide a plausible mechanism for the induction of membrane curvature. As indicated in Fig. 8, we expect the lipids to be arranged with their acyl chains facing inward into a somewhat crowded interior, so that increasing curvature would push the head groups apart, circumferentially. By running parallel to the axis of the cylindrical micelle, ␣S would fill this gap between head- groups and effectively stabilize the curved lipid arrangements. In fact, the extended helical structure would provide a highly anisotropic curvature constraint that would promote the induction of curvature perpendicular to the tubule axis. It remains to be tested whether the great length of the ␣S helix contributes to the remarkable longevity of the cylindrical micelles, which are stable for days (data not shown), or whether shorter helices might also be able to cause the formation of stable cylindrical micelles.
A number of membrane curvature-inducing proteins, including BAR domain-containing proteins (64 -66), insert amphipathic helices into the membrane. However, they often contain additional scaffolding domains, making it more difficult to assign individual contributions to curvature induction. Inasmuch as ␣S does not present a pre-existing scaffold but folds up as a consequence of its interaction with the membrane, the present data demonstrate that amphipathic helices can be sufficient to induce stable bilayer tubes as well as cylindrical micelles.
Effect of Lipid Bulkiness on Tube Formation-We observed that the propensity for forming micellar or bilayer tubes is modulated by the length and bulkiness of the lipid acyl chains. Although smaller acyl chains might help to reduce crowding at the center of the cylindrical micelles, long and bulky acyl chains might destabilize the cylindrical micelle in favor of other curved structures. This expectation is largely borne out by the data. The overall trend is that short acyl chains, like those of dimeristoyl (DM) or palmitoyl oleoyl (PO), favor tubulation into cylindrical micelles whereas bulkier acyl chains result in more bilayer tubes. With diarachnidonoyl (DA), a very bulky lipid with eight double bonds, the remodeling products were larger and had a more bulbous morphology.
Implications of ␣S-coated Cylindrical Micelles and Lipoprotein Particles in a Cellular Context-The membrane remodeling phenomena that we observe in vitro take place at relatively high concentrations of the protein principle (␣S). For them to be operative in a cell would require some mechanism to achieve the needed local concentrations. However, as ␣S is an abundant protein accounting for 1% of total protein mass in neural cells (4,5), there appear to be realistic prospects that such is the case. Molecular crowding in the cellular milieu (67) could also promote the reaction. However, further in situ data are needed to clarify the status of this hypothesis.
As yet, there has been no demonstration that ␣S-coated cylindrical micelles are formed in situ, although they form readily and persist in vitro. However, there have been reports of thin filamentous structures in neural tissues of humans and a mouse model that are ␣S-positive according to immuno-gold EM with anti-␣S antibodies (68,69). The present findings raise the possibility that they may represent cylindrical micelles instead of or as well as amyloid fibrils.
The main focus of this study was to gain structural and mechanistic insight into the ability of ␣S to induce curvature in tubular membranes. In the process, we also discovered that at higher protein/lipid ratios, ␣S induces the formation of discoid particles, 70 -100 Å across (Fig. 1D). Whether these structures derive from tubular precursors and what structure ␣S may assume in them remain to be established. If it is important to maximize contact between the ␣S molecule and the lipids, a bent or segmented helical structure would be preferred to an extended one (Fig. 8). Regardless of the details, the ability to induce discoid membranous particles could be of physiological relevance. It may also be another property that ␣S shares with apolipoproteins (70,71), a family of proteins involved in the transport of lipids whose ability to form similarly sized membrane discs is well established.