Membrane Curvature Sensing by Amphipathic Helices

Background: Amphipathic helices preferentially bind highly curved lipid membranes, providing a method of protein sorting. Results: Curvature sensing requires the insertion of hydrophobic residues and is modulated by electrostatic interactions. Conclusion: The relative strength of hydrophobic and electrostatic membrane interactions determines whether helix-containing proteins sense curvature. Significance: Sensing cannot be described through simple physicochemical properties but depends on the total sum of membrane interactions. Preferential binding of proteins on curved membranes (membrane curvature sensing) is increasingly emerging as a general mechanism whereby cells may effect protein localization and trafficking. Here we use a novel single liposome fluorescence microscopy assay to examine a common sensing motif, the amphipathic helix (AH), and provide quantitative measures describing and distinguishing membrane binding and sensing behavior. By studying two AH-containing proteins, α-synuclein and annexin B12, as well as a range of AH peptide mutants, we reveal that both the hydrophobic and hydrophilic faces of the helix greatly influence binding and sensing. Although increased hydrophobic and electrostatic interactions with the membrane both lead to greater densities of bound protein, the former yields membrane curvature-sensitive binding, whereas the latter is not curvature-dependent. However, the relative contributions of both components determine the sensing of AHs. In contrast, charge density in the lipid membrane seems important primarily in attracting AHs to the membrane but does not significantly influence sensing. These observations were made possible by the ability of our assay to distinguish within our samples liposomes with and without bound protein as well as the density of bound protein. Our findings suggest that the description of membrane curvature-sensing requires consideration of several factors such as short and long range electrostatic interactions, hydrogen bonding, and the volume and structure of inserted hydrophobic residues.

Preferential binding of proteins on curved membranes (membrane curvature sensing) is increasingly emerging as a general mechanism whereby cells may effect protein localization and trafficking. Here we use a novel single liposome fluorescence microscopy assay to examine a common sensing motif, the amphipathic helix (AH), and provide quantitative measures describing and distinguishing membrane binding and sensing behavior. By studying two AH-containing proteins, ␣-synuclein and annexin B12, as well as a range of AH peptide mutants, we reveal that both the hydrophobic and hydrophilic faces of the helix greatly influence binding and sensing. Although increased hydrophobic and electrostatic interactions with the membrane both lead to greater densities of bound protein, the former yields membrane curvature-sensitive binding, whereas the latter is not curvature-dependent. However, the relative contributions of both components determine the sensing of AHs. In contrast, charge density in the lipid membrane seems important primarily in attracting AHs to the membrane but does not significantly influence sensing. These observations were made possible by the ability of our assay to distinguish within our samples liposomes with and without bound protein as well as the density of bound protein. Our findings suggest that the description of membrane curvature-sensing requires consideration of several factors such as short and long range electrostatic interactions, hydrogen bonding, and the volume and structure of inserted hydrophobic residues.
The amphipathic helix (AH) 5 is a common structural motif in cellular biology that often serves as a means for anchoring proteins to membranes. In this role AHs are important for a number of processes, including nucleation of protein coatcomplexes (1)(2)(3)(4)(5)(6)(7)(8) as well as vesicle capture and trafficking (9 -12). AHs comprise an ␣-helical secondary structure in which one face contains predominantly polar and charged residues, conferring a hydrophilic nature, whereas the opposing face consists of hydrophobic residues. When situated at the exterior of proteins, this structure is particularly well suited for binding to lipid bilayers, as the helix can align itself parallel to the bilayer axis, with the amphipathic interface level with the bilayer lipid head groups (13)(14)(15). Recently, this structural motif has been implicated in both sensing and induction of membrane curvature (16 -20), processes that are fundamental in cell division and proliferation (21,22). Importantly, AHs sense membrane curvature by responding to differences in the density of lipid packing defects in the membrane, thereby showing preferential binding to highly curved membranes (19,23). In the following, we define sensing of MC as the ratio of protein density in high membrane curvature areas over flat or low curvature areas. The ability of AHs to sense membrane curvature and thus use curvature as "molecular information" to organize processes in time and space has recently been investigated with increasing interest (7, 19, 21, 24 -26). However, little is known about how the hydrophobic and polar faces each modulate this ability and whether membrane composition also modulates binding and sensing (13).
We recently developed a new assay to investigate proteinmembrane interactions as a function of curvature in a systematic high-throughput fashion (19,(27)(28)(29). The new assay provides data for single liposome curvatures (SLiC) and thus reveals detailed information that is averaged out in bulk in vitro assays. Specifically, it allows the decoupling of two curvaturesensitive parameters that together describe membrane binding, as highlighted in recent papers (23,28). The first of these is the bound protein density ( B ), i.e. the number of proteins or peptides bound per surface area of lipid membrane. The second parameter has been termed bound fraction (B frac ) and refers to the observation that not all liposomes within a population will elicit protein binding at a given time (28). In other words, membrane binding by amphipathic helices is heterogeneous such that only a certain percentage of the observed liposomes will have protein bound to them. Importantly, the existence of heterogeneous protein binding to liposomes has severe consequences for the interpretation of bulk assay results that e.g. seek to compare membrane binding of two mutants. Because such assays measure average binding across a population of liposomes, it cannot be determined whether a mutation affects the amount of bound protein ( B ) or the fraction of liposomes with protein bound (B frac ). The latter case might affect bulk measurements of bound protein on populations of liposomes with a certain curvature without actually changing the ability of the protein to preferentially concentrate on curved membranes.
Here we used the SLiC assay to investigate two well known model systems, ␣-synuclein and annexin B12. These proteins have previously been reported to sense, and under specific conditions induce, membrane curvature through the use of AHs (30 -32); this study uses conditions where sensing, rather than induction, of curvature occurs. More specifically, we set out to investigate membrane binding in terms of B and B frac for disparate membrane curvatures, thus allowing us to systematically examine how the sensing ability of AHs is regulated by their hydrophobic residues, polar residues, and by changes in the electrostatic environment due to increased membrane charge density or the presence of calcium.
Our results indicate that membrane curvature sensing by ␣-synuclein is not caused by a higher affinity for curved membranes but rather by a curvature-dependent saturation density, as reported previously for the N-terminal AHs of endophilin and amphiphysin (19). Systematically mutating residues on the hydrophobic and polar face allowed us to show that the sensing ability is strongly dependent on inserted hydrophobic residues but further modulated by changes in the polar face. Meanwhile, B frac (i.e. the ability of the protein to bind the membrane at all) is affected by both aspects and generally correlates negatively with strong sensing. This is corroborated by our examination of annexin B12, amplifying electrostatic interactions between the AHs and the membrane increased B frac as well as the B on flat membranes, thus revealing a powerful modulating element in membrane binding of AH-containing proteins. Therefore, it is the relative contributions from two sources of binding energy, hydrophobic insertion and electrostatic interactions, that determine the ability of an amphipathic helix to sense membrane curvature.

EXPERIMENTAL PROCEDURES
Proteins and Peptides-With the exception of Mut5, peptides were purchased as lyophilized powders from Copenhagenbased peptide synthesis company Schafer-N, all at Ͼ95% purity, labeled at the C terminus using a triple glycine linker.
Mut5 was prepared by automated peptide synthesis on a Syro II peptide synthesizer (MultiSynTech) by Fmoc SPPS on Tenta-Gel Rink amide resin (loading, 0.24 mmol/g, Rapp Polymere GmbH). All solvents and reagents used for peptide synthesis were of analytical reagent grade and were obtained from Sigma or Iris Biotech GmbH (Marktredwitz). N ␣ -Fmoc amino acids (5.0 equiv) were coupled using HBTU (4.8 eq), HOBt (5.0 eq), and DIPEA (9.0 eq) as coupling agents in N,N-dimethylformamide (DMF) for 2 ϫ 2 h. N ␣ -Fmoc deprotection was performed using 40% v/v piperidine/DMF for 3 min followed by 20% v/v piperidine/DMF for 15 min. The peptide amides were cleaved from the solid support by treatment with trifluoroacetic acid ( Purified Mut5 was dissolved 5 ml of acetonitrile and diluted with 6 ml of 100 mM phosphate-buffered saline (pH 7.00), and 1.1 ml of 6 M guanidinium chloride. N-(5-Fluoresceinyl)maleimide (2 equiv) was dissolved in 250 l of dimethylformamide and added to the peptide solution. After 3 h at room temperature, the reaction was quenched by mild acidification with TFA. The peptide was purified by preparative HPLC (see above) using a gradient of 35 Position 136 of human ␣-synuclein (which does not contain native cysteines) was mutated from tyrosine to cysteine by sitedirected mutagenesis and confirmed by DNA sequencing. ␣-Synuclein Y136C was expressed and purified as described previously (33). Briefly, ␣-synuclein Y136C was expressed in Escherichia coli BL21(DE3)pLysS cells. The cell pellet was resuspended in lysis buffer (100 mM Tris (pH 8), 300 mM NaCl, 1 mM EDTA (pH 8)). The lysate was boiled and then acid-precipitated. After precipitation, the supernatant was dialyzed against dialysis buffer (20 mM Tris (pH 8), 1 mM EDTA (pH 8), 1 mM DTT). Two rounds of anion exchange chromatography were performed, and proteins were eluted with a salt gradient of 0 -1 M NaCl. Labeling of ␣-synuclein Y136C with the dye Alexa Fluor 488 C 5 maleimide (Invitrogen) was performed in 20 mM Hepes (pH 7.4), 100 mM NaCl using 10ϫ molar excess of dye. The reaction was allowed to proceed at 4°C overnight, and excess dye was removed by gel filtration.
Annexin B12 contains two native cysteine residues (positions 113 and 302), both of which were replaced by alanines by sitedirected mutagenesis. Residue 162 was mutated from aspartic acid to cysteine by the same method and confirmed by DNA sequencing. Expression and purification have been previously described (34). E. coli DH5␣ cells were used for protein expression. Annexin D162C was purified by reversible Ca 2ϩ binding to phospholipid liposomes followed by gel filtration. Labeling with the dye Alexa Fluor 488 C 5 maleimide (Invitrogen) was performed in 20 mM Hepes (pH 7.4), 100 mM NaCl using 10ϫ molar excess of dye. The reaction was allowed to proceed at 4°C overnight, and excess dye was removed by gel filtration.
Single Liposome Curvature Assay-Surfaces with immobilized liposomes were prepared through biotin-streptavidin coupling as described previously (28). Proteins/peptides were kept on ice until the addition to the microscope chamber. Binding was allowed to reach equilibrium for 30 min, after which images were acquired.
Microscopy was performed on a Leica TCS SP5 confocal fluorescence Microscope, with an Acousto-Optical Beam Splitter & Acousto Optical Tuneable Filters system allowing tunable wavelength detection intervals. The objective used was an oil immersion HCX PL APO with ϫ100 magnification and numerical aperture 1.4. Labeled proteins and peptides were excited at 488 nm and detected from 495 to 555 nm. Liposomes were excited at 633 nm and detected from 640 to 790 nm. Images were acquired sequentially to avoid cross-talk. The frequency of line acquisition was 400 Hz, the resolution was 2048 ϫ 2048 pixels, and the pixel size was 25.24 nm with a bit-depth of 16 bit. All images were acquired using a line average of 3 for reducing noise. For all samples, liposomes were imaged before the addition of protein/peptide to ensure that binding would not significantly affect liposome morphology or fluorescence. The microscope was kept at a constant temperature of 22 Ϯ 1°C.
Image Analysis and Size Calibration-Images were analyzed using a specifically developed routine in Igor Pro 5.05 (Wavemetrics) for extracting single particle positions, background-corrected fluorescence intensities and identifying colocalized protein and liposome signals. Liposome diameters were determined by correlating the distribution of fluorescent intensities, as measured by microscopy, with size distributions obtained by dynamic light scattering. Both methods were described previously (19,35).
Definitions of Parameters-B is defined as the ratio of fluorescent signal from the protein/peptide divided by the signal from the liposome to which it is bound. This can be translated into real densities by determining the fluorescent signal per label, but this conversion is not relevant for determining sensing ability or relative bound densities. K d is defined as the protein concentration at which half-maximal binding is observed. B frac is simply the number of observed liposomes with bound protein divided by the total number of liposomes. We have previously determined a power-law dependence for curvaturesensing, i.e. bound density ϰr Ϫ␣ , where r is the radius of a liposome, and the value of ␣ describes the curvature sensing ability of a specific protein (19).

RESULTS
Using Single Liposome Data to Investigate Membrane Curvature Sensing-In the SLiC assay individual liposomes with a wide range of diameters, and thus curvatures, were immobilized on a passivated glass surface through biotin-streptavidin coupling (Fig. 1A). Curvature-sensing proteins were allowed to bind from solution, whereupon the liposomes and protein bound on these could be imaged by confocal fluorescence microscopy through separate fluorescent labels (Fig. 1B). In the micrographs we tracked each particle/liposome ( Fig. 1C; liposome (red), protein (blue)) and identified colocalized signals in the two channels (green). From the integrated fluorescence signals (Fig. 1B, zooms) we calculated liposome size and B (19,35). This access to single liposome data allowed us to identify heterogeneity in protein binding to liposomes of the same diameter. Although the data points for liposomes of a given size showed the same B , suggesting that liposomes of identical diameter behaved the same, it was only a fraction of the liposomes that had admitted a detectable protein signal (Fig. 1C, green). From this observation we describe a new parameter, termed B frac , denoting the percentage of liposomes with bound protein (28). This parameter will be used to evaluate and compare the binding of various AHs.
The SLiC assay measures binding to several hundreds of individual liposomes of different diameters, and we were, therefore, able to accurately measure binding as a function of curvature. Plotting B as a function of liposome diameter for each colocalized signal yields a characteristic power-law dependence for curvature-sensing proteins, in this case ␣-synuclein ( Fig. 2A, black circles and fit), whereas curvature-independent binding gives a constant density for all sizes ( Fig. 2A, red circles). The power-law dependence (density ϰr Ϫ␣ , where the exponent ␣ depends on the protein in question) has been reported previously (19) and allows us to quantify the curvature sensing ability of a given protein by extracting the exponent, which is proportional to the ability of the protein to sense curvature. By plotting density versus diameter using logarithmic scales (Fig. 2B), the data are converted into a linear function where the slope, here 1.23, corresponds to the curvature-sensing ability of the protein (the value ␣ described above), thus serving as a quantitative measure of the ability of a given protein to sense curvature. Significantly, an apparent difference in sensing ability can result either from a real difference in the ability to bind preferentially highly curved membranes or from a constant increase in bind-ing to liposomes of all sizes. For curvature-sensing proteins such nonspecific binding would have a relatively larger effect on the limited binding to flat membranes and would, therefore, affect the relative increase in binding on curved membranes (see e.g. Fig. 3, D and E, in Ref. 36). Because the SLiC assay independently measures binding to small and large liposomes, it can distinguish between differences in sensing ability arising from each of these mechanisms.
Curvature Sensing by ␣-Synuclein Is Not Affinity-based-We wanted to use the SLiC assay to characterize a known AH-based membrane curvature sensor in terms of sensing ability and the separate parameters of B and B frac . ␣-Synuclein, a neuronal protein involved in synaptic plasticity regulation and implicated in Parkinson disease (37,38), is well characterized and FIGURE 1. The SLiC assay. A fluorescence-based assay for measuring curvaturesensitive binding at the single liposome level in a massively parallel manner is shown. A, fluorescently labeled liposomes with a wide range of diameters were tethered to a passivated glass surface through biotin-streptavidin coupling, and a separately labeled molecule of interest was allowed to bind from solution. Arrows indicate a preferential binding of curvature sensing proteins to small liposomes. B, liposomes (top) and peptides/proteins (bottom, in this case ␣-synuclein) are sequentially imaged using confocal fluorescence microscopy. The zooms illustrate single liposomes and their bound molecules displaying Gaussian fluorescence intensity profiles. Scale bars are 5 m. C, specialized software tracks particles in each image (red and blue circles) and identifies colocalized signals (green circles). The background-corrected gaussian fluorescence intensities shown in B were used to quantitatively calculate liposome size and the density of bound molecules for each set of colocalized signals. A, shown is a sensing graph for full-length ␣-synuclein (black circles). Curvature sensing is seen as an increase in bound density for smaller liposome diameters compared with the base-line density observed for large liposomes. This curvature-sensing can be described by a decreasing exponential function (black line) as previously described (19). Streptavidin coupling to biotinylated liposomes serves as a negative control for curvature sensing, showing equal bound densities for all liposome sizes (red circles). B, plotting data on logarithmic scales renders the exponential sensing behavior from A as a linear function. The slope of this function gives us the exponent of curvature sensing. This value, denoted sensing ability, is used as a quantitative measure of the degree to which a molecule is able to sense membrane curvature. The graphs are representative of three separate experiments.
was one of the first proteins recognized for its ability to sense membrane curvature (30,39,40). It is a 140-residue protein that is unstructured in solution but for which the N-terminal ϳ100 amino acids form an amphipathic helix upon interaction with lipid membranes (36,41,42). The C-terminal region on the other hand has considerable negative charge and has not been found to interact with membranes (36,41). We, therefore, focused our attention on the positively charged N-terminal ␣-helical region. Given the cellular role of ␣-synuclein, we chose to use liposomes reconstituted from brain lipid extract as a physiologically relevant model system.
First, in line with previous reports (14,30) we confirmed that ␣-synuclein adopts an ␣-helical structure upon binding to liposomes using circular dichroism (supplemental Fig. 1). Using the SLiC assay, we then observed that ␣-synuclein senses membrane curvature by binding at a higher density to liposomes of decreasing diameter ( Fig. 2A), in overall agreement with previous literature (30,39). However, the single liposome nature of the assay further allowed us to deconvolve the influence of B frac on the apparent curvature sensing. Thus, the curvature sensing of ␣-synuclein was quantified more rigorously as a sensing ability (␣) of 1.23 Ϯ 0.08, which makes the protein as efficient a curvature sensor as the N-terminal AHs of endophilin and amphiphysin (␣ ϭ 1.3 (19)). This corresponds to a 17-fold greater concentration on highly curved liposomes compared with the base-line bound density, i.e. six times higher than previously estimated using bulk assays (30). We further investigated whether ␣-synuclein exhibited fractional binding on the single liposome level and found that, as observed for other AHs, only a subpopulation of liposomes had admitted protein. Average B frac was 13 Ϯ 1.5%, similar to the results reported for the AH of endophilin (28).
For the AHs of endophilin and amphiphysin, preferential binding on more highly curved membranes has been reported to result from an increased density of apparent binding sites (B max ) rather than an increased affinity for the curved membrane (28). In other words, the K D of membrane interaction varied only by a factor of three for changes in membrane curvature, which cannot account for the corresponding 80-fold changes in B . This further holds true for lipid anchor motifs and NBAR domains and, thus, appears to be a characteristic trait of curvature sensing mediated by hydrophobic insertion. Because ␣-synuclein exhibited membrane curvature-sensing properties similar to the N-terminal AH of endophilin and amphiphysin, we proceeded to assess whether binding of ␣-synuclein was similarly limited by apparent binding sites on the membrane. In Fig. 3A, we show B as a function of liposome diameter for three representative concentrations of ␣-synuclein (10 nM, 200 nM, and 1 M). In total, 8 concentrations between 5 nM and 1 M were tested. For each concentration, B was recorded at seven different liposome diameters ranging from 60 to 600 nm. Three of these diameters are highlighted as gray columns in Fig. 3A. In Fig. 3B we plot the B for these representative diameters as a function of concentration (for all eight concentrations). We thus observe that although B increased for lower diameters, there was no obvious shift in the binding curve, indicating that K d is independent of liposome curvature. By plotting the extracted K d from the binding curve FIGURE 3. Curvature sensing of ␣-synuclein is not affinity-based. Measuring the dissociation constant of ␣-synuclein as a function of liposome size reveals that sensing arises from a greater density of binding sites on curved membranes (B max ) rather than from a higher affinity for these. A, sensing fits for ␣-synuclein at three representative concentrations are shown; in total, binding was measured for 8 concentrations between 5 nM and 1 M. The bound density increased as a function of concentration for all liposome sizes. By extracting the concentrationdependent density at specific diameters (60, 100, and 400 nm, gray columns), binding curves can be plotted for specific curvatures. The three highlighted diameters are representative of seven values ranging from 60 to 650 nm. B, the concentration-dependent density at the three diameters highlighted in A are plotted and fitted with Langmuir isotherms. Higher bound densities are recorded for smaller liposomes, but the isotherms are not shifted on the concentration axis. C, the size dependence of the apparent K d was extracted from the fits in B. The K d value was more or less constant at 190 nM, illustrating that the increased bound densities arise from a higher number of binding sites. Error bars are the uncertainty in fitting the K d from the binding curves in B. All graphs are based on two separate experiments. DECEMBER 9, 2011 • VOLUME 286 • NUMBER 49 for all seven diameters (Fig. 3C), it becomes evident that the affinity of ␣-synuclein for membranes of different curvature remains constant at ϳ200 nM. Several previous studies have examined the affinity of ␣-synuclein for lipid vesicles either calorimetrically or by fluorescence, yielding dissociation constants around 1 M for liposome compositions, comparable with ours (39,(43)(44)(45). These studies have also revealed marked differences from even minor differences in membrane charge, and thus our findings for brain lipid liposomes can be considered in good correspondence with the previous data. Regarding the influence of curvature, one earlier report showed no significant difference in affinity for liposomes extruded at 50 versus 100 nm (45). A later study found an ϳ2-fold difference in affinity for synthetic liposomes extruded at diameters ranging from 30 nm to 200 nM (46). Given that compositional differences compared with our brain extract liposomes, their observation does not seem incongruent with our results. The authors also reported an ϳ15-fold difference between sonicated and extruded samples, which was not examined in this study. Our finding that the curvature sensing of ␣-synuclein is not primarily affinity-based is, therefore, in good agreement with previous results.

Membrane Curvature Sensing by Amphipathic Helices
Model Peptides for Curvature Sensing by Amphipathic Helices-The helical region of ␣-synuclein consists of a conserved 11-mer helical motif repeated seven times (14,47), suggesting that the curvature-sensing ability of this motif would be representative of ␣-synuclein as a whole. To test this theory, we created two helical peptides from the sequence of ␣-synuclein, each comprising two repeat motifs (residues 2-23 and 21-42). We found that with these two peptides, no liposomes had detectable amounts of protein bound at 200 nM peptide and only showed a detectable signal at very high concentrations of 5 M (data not shown). Although it is clear from structural studies that the amphipathic helical repeat is important for membrane curvature sensing, these results suggest that the cooperative effect of several repeats is necessary for efficient binding, in agreement with the notion that ␣-synuclein takes up a very long, almost 100-amino acid-containing amphipathic helix (14, 36, 48 -50).
Nonetheless, we chose peptide 2-23 (p2-23), which has a strongly amphiphilic character (hydrophobic moment ϭ 0.354), as a model system for investigating the helical traits that influence the curvature sensing of ␣-synuclein. For such a systematic investigation, it is preferable to work with a short peptide derived from the helix rather than the entire helix because this results in a stronger effect from specific single or double mutations, allowing the role of each element to be determined without the potentially obscuring influence of the rest of the helix. Thus we created several mutants from p2-23, as described in the following sections, focusing on amino acid substitutions predicted to improve binding. Although we had done a concentration-dependent study to elucidate the binding mechanism of ␣-synuclein, we chose a single concentration (the K d 200 nM) to compare the mutants in this analysis.
Augmenting the Hydrophobic Face of AHs Improves Curvature Sensing-AHs sense curvature by inserting hydrophobic residues into lipid packing defects in the membrane, and thus the first step of our mutational analysis was to investigate how curvature sensing is changed by altering the hydrophobic face of the helical peptide. Under the hypothesis that a more prominent hydrophobic face would lead to stronger interactions with lipid tails, we created a mutant peptide based on p2-23. The mutant peptide (Mut1) had an alanine and a glycine in the hydrophobic face replaced with phenylalanines (G13F, A17F), which have aromatic side chains that are both bulkier and more strongly hydrophobic (see Fig. 4A). Thus, both the volume and hydrophobicity of the hydrophobic face were increased, as evidenced by the increase in mean helical hydrophobicity (0.354 -0.408, see Table 1 and supplemental Fig. 2). It should be noted that the replaced glycine is known to perturb the folding of helices, so one might imagine that the mutation would also facilitate helix formation. The Mut1 peptide was able to bind liposomes at 200 nM, implying that binding was indeed improved compared with the native p2-23 (Table 1). Whether this was the result of a higher B frac or a higher affinity cannot be said at present. Mut1 showed slightly improved curvature sensing compared with full-length ␣-synuclein, with a sensing ability of 1.40 Ϯ 0.05. This is comparable with previous reports in which alanine substitution of three bulky hydrophobic residues in ArfGAP1 significantly reduced the sensing ability (51) and that the introduction of tryptophan into the endophilin A1 helix improves tubulation (52). The fraction of liposomes with bound protein (B frac ) was similar to the value for the full-length protein, at 9.5 Ϯ 1.5% compared with 13 Ϯ 1.5% (although of course higher than p2-23, for which no binding was observed at this concentration).
To dissociate the effects of hydrophobicity and hydrophobic volume, we created another mutant peptide where the same alanine and glycine residues were replaced by valines rather than phenylalanines (Mut2; G13V, A17V, see Fig. 4A and supplemental Fig. 2). Valine is similarly strongly hydrophobic, but with a far smaller, aliphatic side chain (53). As with Mut1, the fact that this peptide measurably bound liposomes at 200 nM, whereas p2-23 did not, shows that binding was again improved by the enhancement of the hydrophobic face. Furthermore, as for Mut1, curvature sensing was considerably increased compared with full-length ␣-synuclein (1.57 Ϯ 0.21 versus 1.23 Ϯ 0.08), whereas the observed increase over Mut1 fell within the experimental uncertainty (Table 1). B frac was again lower than for the full-length protein and, at 3.8 Ϯ 1.5%, even lower than for Mut1. Interestingly, this results from a large decrease in B frac on large liposomes, whereas it actually increased on small liposomes compared with Mut1 (Fig. 5). Thus, B frac as well as the B is sensitive to membrane curvature. In a bulk assay this redistribution of bound protein would have manifested as a drastic increase in apparent curvature sensing because the increase in B frac on small liposomes could not be distinguished from a genuine increase in B on these liposomes.
For both mutants lower B was observed for all liposome sizes compared with ␣-synuclein, with the most significant decrease on large liposomes; the improved sensing of these mutants compared with full-length synuclein thus originates FIGURE 5. Size dependence of bound fraction is strongly affected by changes to hydrophobic and polar helix faces. Bound fraction (B frac ) is depicted as a function of liposome diameter (binned to four diameter ranges) for full-length ␣-synuclein (A) and the mutant peptides Mut1, Mut2, and Mut3 (B-D). Although Mut1 and Mut2 show generally impaired binding compared with the full-length protein, the largest decrease is seen for large liposomes (of low curvature); this effect is particularly pronounced for Mut2, which surprisingly also showed improved binding to small liposomes compared with Mut1. Conversely, Mut3 shows improved binding to liposomes of all sizes, but the effect is most noticeable for small liposomes. The graphs represent the average of three separate experiments.

TABLE 1 Augmenting the hydrophobic face of helices improves curvature sensing
The binding behavior and helix properties of the model peptide p2-23 and the mutants Mut1 and 2 were compared to full-length ␣-synuclein. Both mutant peptides showed improved binding compared to p2-23, which did not bind at 200 nM. Compared to ␣-synuclein, both have increased sensing ability but lower bound fractions. The effect is particularly pronounced for Mut2. Neither hydrophobicity (H) nor hydrophobic moment (H) correlates well with the measured parameters. Bound percentage and sensing ability are the averages of three separate experiments. Sensing ability ␣ is the slope of a linear fit to the log-log graph of protein density over vesicle diameter.

Sensor
Bound primarily from lower B on flat membranes (Fig. 4B). Although increasing the hydrophobicity of the inserting region increased curvature sensing for both mutants, the difference in the volume and structure of the inserted residues leads to significantly different binding behavior in terms of B frac even without changing the hydrophobicity. This might theoretically stem from the difference in insertion energy for the two side chains (54) and certainly indicates that hydrophobicity is not sufficient for describing curvature-sensitive binding. Charge Distribution on Amphipathic Helices Modulates Binding and Curvature Sensing-Another way that binding energy can change is through altered electrostatic interactions, and we therefore proceeded to examine whether modifying the net charge on the polar face or the distribution of charges would lead to altered binding behavior, as has been reported previously for other AH-containing proteins (28,55,56). To investigate the effect of positive and negative charges in the different regions of the AH on curvature sensing, we chose to base further mutants on Mut1 described above rather than on p2-23. In this manner, both mutations that improve and impair binding could be examined; the examined mutants and their binding behavior are summarized in Fig. 6 and Table 2.
The third mutant (Mut3) had two further amino acid substitutions, E12A and E19A, both in the polar face directly opposing the hydrophobic wedge. This raised the net charge of the peptide from ϩ2 to ϩ4, yet the affected side chains were not expected to interact strongly with the lipid membranes because of their location (14). For Mut3, we observed a significantly higher B frac than for Mut1 (35 versus 9.5%) as well as a lower sensing ability (0.92 versus 1.40). Although both B frac and B increased for all liposome sizes compared with Mut1, it is noteworthy that for Mut3 B frac increased most markedly for smaller liposomes (Fig. 5D). Meanwhile,. the largest increase in B was observed for the large liposomes (ϳ3-fold, compared with ϳ2-fold for small liposomes), resulting in a smaller difference between small and large liposomes and consequently a decreased sensing ability (Fig. 6B). Both these results are the opposite of those observed when the hydrophobic face was augmented in Mut1, suggesting that the improved binding observed for Mut3 compared with the source peptide is not mediated by insertion. From these results it is clear that charges on the peptide affect the probability of membrane binding and may also alter the sensing ability of a given peptide by increasing B on larger liposomes. Furthermore, it is worth noting that although B frac and overall B represent distinct factors in membrane binding, between the three mutant peptides and the fulllength protein there seems to be some degree of correlation between these parameters.
Next we designed two new mutants to investigate the importance of charge at the interface between the polar and hydrophobic regions of the helix. In the first mutant we replaced the four interfacial lysine residues with arginines (Mut4; K9R, K11R, K20R, K22R; see Fig. 6A), which conserve the positive charge but enable stronger interactions with lipid head groups through additional hydrogen bonds formed by the guanidinium group. The other mutant with altered interfacial charges had the interfacial lysines replaced by aspartic acids, whereas an aspartic acid and two glutamic acids in the polar face were replaced with arginines (Mut5; D1R, K9D, K11D, E12R, E19R, K20D, K22D; see Fig. 6A). This redistribution of charge, yielding negative residues in the interfacial region and positive residues in the polar region, would be expected to impair binding due to unfavorable interactions with negatively charged lipid head groups. An additional motivation for these mutations was that the protein DivIVA has recently been reported as a nega- FIGURE 6. Charges on the amphipathic helix strongly affect binding. A, three additional mutant peptides were created based on the Mut1 peptide described in Fig. 4. The first of these (Mut3, top right) had two glutamic acids in the polar face replaced by alanines (A), reducing net charge likely without affecting membrane interactions in the bound state. As shown in B, this peptide bound at significantly higher densities than Mut1; this effect was most marked on large liposomes (3-fold versus 2-fold on small liposomes), resulting in a lower sensing ability. The second peptide (Mut4, middle right) had four interfacial lysines replaced by arginines (R), which are able to interact more strongly with lipid head groups. This peptide caused liposome disruption at a concentration of 200 nM, suggesting that its membrane interaction was significantly stronger than for the Mut1 peptide. The last peptide (Mut5, bottom right) had the same four interfacial lysines replaced by aspartic acids (D), whereas three negative charges in the polar face were changed to arginines (R). This arrangement resembles a previously reported sensor of negative curvature (DivIVA), but the Mut5 peptide was unable to bind liposomes in our assay. The graphs are representative of three separate experiments.
tive curvature sensor through an amphipathic helix with negatively charged interfacial residues (6,57). For the Mut5 peptide, a complete lack of binding was observed even when we increased the concentration 12-fold. It should be noted that we cannot say from this mutant alone whether the lack of binding was due to the decreased net charge caused by these mutations or the unfavorable interfacial interactions nor can we rule out that this (DivIVA-like) peptide could bind to membranes with negative curvature as previously reported (7,57). Indeed, from these data we cannot make predictions about the potential sensing ability of this peptide; the absence of signal precludes any conclusions regarding binding behavior.
In contrast, Mut4 was designed to have increased interfacial interactions with lipid headgroups without a change in net charge through additional hydrogen bonding by the interfacial arginines. Here, the result was dramatically different; at 200 nM, Mut4 caused disruption of the liposomes (data not shown). This may be linked to induction of membrane curvature, as previously reported for ␣-synuclein (32). Thus, binding parameters could not be quantified, but the ability to deform liposomes at this concentration clearly represents stronger interaction with the lipid membrane than exhibited by the other peptides or indeed by ␣-synuclein. These membrane interactions are comparable with a similar amphipathic helix in the influenza virus M2 protein, which was reported to disrupt membranes (58).
Mut4 and -5 along with Mut3 described above strongly suggest an important role for interfacial charges in modulating the binding and consequently the curvature sensing of amphipathic helices. This is in agreement with a previous report where the introduction of interfacial positive charges in the AH of ArfGAP1 diminished sensing by increasing binding preferentially on large liposomes (55). Apart from the effect of interfacial charges, we also found that polar residues distal to the membrane-interacting surface do affect curvature sensing and that hydrogen bonding can have drastic effects on binding.
Membrane Charges Modulate AH Binding through B frac -Based on these findings, we decided to further investigate the role of electrostatic interactions between helix and membrane by adjusting the membrane composition. It seems logical that altering membrane charge should affect electrostatic interactions between helix and membrane just as does altering the charge distribution on the helix. To test this we produced liposomes consisting of neutral porcine brain L-␣-phosphatidylcholine and negatively charged L-␣-phosphatidylserine phos-pholipids at either 20, 33, or 67 mol % L-␣-phosphatidylserine. These liposomes represented simpler systems than the brain lipid mixtures used thus far and allowed controlled variation of membrane charge density.
The binding of ␣-synuclein at 100 nM to liposomes of increasing L-␣-phosphatidylserine content is summarized in Table 3 and is consistent with the results presented in the previous section. Changing the amount of negative charges in the membrane from 20 to 33% and then 67% led to an increase in B frac , from 5.5 Ϯ 3.0 to 27 Ϯ 1.0 and then 34 Ϯ 10%. Thus, when membrane charge density was increased, B frac went up just as we observed when increasing peptide net charge. Interestingly, there was no significant difference in sensing ability when increasing membrane charge from 20 to 33 and 67%. The sensing ability was slightly reduced compared with the brain liposomes (1.06 Ϯ 0.04 versus 1.26 Ϯ 0.15), but this was not unexpected based on the difference in compositional complexity of the lipid mixtures. However, it is noteworthy that increasing membrane charge did not increase B on liposomes of any size (Fig. 7). Therefore, the presence of charges on either peptide or membrane affect protein binding in different ways; a change of the peptide charge (e.g. phosphorylation) may alter the peptide ability to sense membrane curvature, whereas a change in FIGURE 7. The effect of the membrane charge on curvature sensing by ␣-synuclein. The complex brain lipid mixture making up our reference liposomes was replaced by three compositions containing neutral PC and negatively charged PS at 4:1, 2:1, and 1:2 ratios respectively. Binding of ␣-synuclein at 100 nM on these liposomes is shown in the graph, illustrating that neither bound density nor curvature sensing was significantly affected by membrane charge. The graphs are representative of three separate experiments.

TABLE 3
The effect of the membrane charge on curvature sensing by ␣-synuclein The extracted parameters bound fraction and sensing ability are summarized for ␣-synuclein at 100 nM on liposomes containing 20, 33, or 67% membrane charge as well as on the brain lipid reference mixture. Bound fraction increased for rising membrane charge density, whereas sensing ability was effectively unaltered. Both values were changed slightly on these membranes compared with the more complex reference mixture. Values are the averages of three separate experiments.

Vesicles
Bound fraction Sensing ability  membrane charge density might only affect B frac binding in a curvature-independent manner. The lack of effect on sensing might suggest that only a certain number of negatively charged head groups are able to interact with positive residues at the helical interface; although the membrane charge density always affects long-range attraction to the membrane (and thus B frac ), additional charges do not interact with the amphipathic helix upon insertion and, therefore, have no effect on B.
Overall, these results support the idea that electrostatic interactions are important both for long-range attraction of the AH to the membrane and for modulating membrane interactions, as has been reported previously (31,59). Attractive electrostatic interactions increase the background binding to membranes, most significantly to areas of low curvature, and consequently affect curvature sensing.
Membrane Binding of Annexin B12 Confirms the Role of Electrostatics-With this in mind, we wished to determine whether this role of electrostatics is conserved for other membrane-inserting proteins. Annexin B12 is another interesting AH-containing protein that has been reported to sense curvature (31). It binds membranes in different conformations by either a calcium-dependent mechanism or in a calcium-independent mechanism, where the latter conformation has been reported to be sensitive to curvature (31,60,61). In solution, the annexin monomer consists of four repeats, each containing two helix-loop-helix hairpins. In the absence of calcium ions, monomers can undergo a conformational change causing the hairpins to form continuous amphipathic helices, which are thought to insert into curved membrane in the same manner as ␣-synuclein (31,62). In the presence of calcium, annexin forms trimers that coordinate multiple calcium ions at the loop regions; this results in minimal insertion into the membrane, with the positively charged calcium ions mediating membrane binding (62). This is illustrated in Fig. 8. We used our assay to investigate the membrane curvature-sensing properties of annexin B12, which are summarized in Table 4. .
As expected, the Ca 2ϩ -independent binding of annexin (3 M) to brain liposomes showed a clear curvature-sensing ability in the SLiC assay, manifesting as a sensing ability of 1.64 Ϯ 0.10, corresponding to a 44-fold increase on a 40-nm liposome compared with the base line at 400 nm. This is a remarkable sensing ability, constituting one of the best curvature sensors examined in the SLiC assay to date. B frac was 19 Ϯ 3.5%, comparable with other AH-inserting proteins like ␣-synuclein. Surprisingly, we found that contrary to a previous report (31), the Ca 2ϩ -dependent binding of annexin is in fact moderately selective toward high membrane curvatures, with a sensing ability of 0.83 Ϯ 0.05, although much less so than its Ca 2ϩ -independent binding. B frac was very high for all liposome sizes (78 Ϯ 6.0%, size dependence data not shown), as expected from the strong electrostatic interaction mediated by the chelated calcium ions. Thus, electrostatics mediate effective binding to liposomes of all sizes, which diminishes the curvature-sensitive binding caused by membrane insertion. Although it must be noted that calcium-dependent binding also involves a conformational change in the protein, these findings nonetheless fit well with the results obtained using peptide mutants in the previous sections and with existing reports in the literature (31,55).
The fact that the Ca 2ϩ -dependent binding mode of annexin B12 did show weak sensing in our studies could be attributed to the more complex membrane lipid composition used here compared with previous reports (31) or to the greater sensitivity of our assay. It does, however, harmonize with data showing that the Ca 2ϩ -dependent binding of synaptotagmin is higher on curved membranes (63). To test whether the observed curvature sensing could be abolished by increasing electrostatic interactions, we measured calcium-dependent binding to brain PC/PS liposomes with controlled membrane charge densities, as previously described for ␣-synuclein. These results are summarized in Table 5. For membranes with 20, 33, or 67% negative charge, the sensing ability of annexin B12 was drastically decreased (to around 0.5) and was not significantly higher than negative controls. We also observed that at higher charge ratios (33 and 67% negative charge), B frac was effectively complete (ϳ100%). The abolished sensing could thus be attributed to effective binding to liposomes of all sizes and as such corresponds well with previous studies on such two-component membrane systems (31,55). This reinforces the notion that sufficiently strong electrostatic interactions may completely overshadow curvature-sensitive binding, even when some insertion of hydrophobic residues does occur, by enabling potent binding to membranes of any curvature and thereby obliterating any differential B mediated by curvature sensing.

DISCUSSION
The SLiC assay used in this study is a uniquely suited method for accurately describing various aspects of curvature sensitive binding. By measuring binding at the single liposome level, we FIGURE 8. Membrane binding of annexin B12 with or without calcium. Two binding modes of annexin B12 are illustrated. In the absence of calcium, the annexin monomer (left) undergoes a conformational change to allow the helical hairpins to form continuous amphipathic helices that insert into the membrane (right, top). This insertion results in highly effective curvature sensing. In the presence of calcium, annexin forms trimers that coordinate calcium ions jointly with the lipid membrane (right, bottom; calcium ions are illustrated as blue pentagons). This binding mode involves very minimal insertion into the membrane and is far less sensitive to curvature. The graphs are representative of three separate experiments.

TABLE 4 The calcium-dependent and -independent binding modes of annexin B12 differ greatly
The binding behavior of annexin B12 (ANX12) in the absence and presence of calcium is summarized in terms of bound fraction and sensing ability ␣. Binding with coordination of calcium ions exhibits vastly lower sensing, with a corresponding increase in the bound fraction compared to the calcium-independent binding by insertion. Values are the averages of three separate experiments.

Sensor
Bound fraction Sensing ability ANX12-Ca 2ϩ 19 Ϯ 3.5% 1.64 Ϯ 0.10 ANX12 ϩ Ca 2ϩ 78 Ϯ 6.0% 0.83 Ϯ 0.05 evaluated curvature sensing for a continuum of curvatures and thus in much greater detail than earlier ensemble-based methods that are limited by low numbers of data points and overlap in liposome populations (23). The ability to decouple membrane binding into two separate, curvature-dependent components, B (and from this, sensing ability) and B frac , allowed us to quantitatively compare and delineate how the hydrophobic and polar face as well as membrane charge density influence membrane curvature sensing by amphipathic helices. Although curvature sensing originates from insertion of hydrophobic residues into lipid packing defects and is governed by the nature and number of inserted residues and the amount of defects in the membrane (19,23,28), this insertion is modulated by electrostatic interactions that can facilitate binding to charged membranes of any curvature and, therefore, reduce the apparent curvature sensing ability of the protein (31,55). Thus, it is the relative strength of each type of interaction that determines a given protein's ability to bind membranes in a curvature-sensitive manner.
It has previously been suggested (13,59) that amphipathic helix insertion occurs in three steps; attraction to the membrane, insertion of hydrophobic residues, and folding into the helical structure. Although this study did not address the folding step, our results seem in accord with this three-step model. Specifically, we propose that the initial step is to a great extent reflected in the observable B frac , whereas B is highly dependent upon membrane insertion and/or helix folding. This is consistent with a previous report that B frac represents the probability of nucleation for cooperative binding (28). When we increased charge density in either the AH or lipid membrane, we observed a large increase in B frac (Tables 2 and 3), illustrating that this parameter is heavily modulated by overall charge as one would expect for long-range electrostatic attraction between two objects. Remarkably, we observed that membrane charge density did not affect B or sensing ability, which were, however, dependent on both hydrophobic and charged residues and their position on the AH (Figs. 4 and 6). This may reflect a state where negatively charged lipid head groups are abundant such that the exact configuration of short-range interactions with the AH depend largely on the nature and availability of amino acid side chains. This model is in good agreement with a recent examination of the HIV-1 Nef protein, for which membrane association was described by a fast kinetic phase with a strong electrostatic component followed by a slower kinetic phase that depended heavily on conserved hydrophobic residues and was influenced by membrane curvature (64).
Another analysis of these data addresses the relative role of hydrophobic and electrostatic interactions. Our findings consistently indicated that augmenting the hydrophobic face of the AH improved binding (both B frac and B ) as well as sensing ability (Table 1 and Fig. 5). Meanwhile, mutations that improved its ability to interact with the bilayer through electrostatics consistently improved binding in terms of B frac and B while inhibiting sensing ability (Table 2 and Fig. 5). These observations on peptide mutants of ␣-synuclein were corroborated by our observations for annexin B12 in the presence and absence of calcium (Table 4). Thus, curvature sensing depends critically on two factors of the total binding energy of a protein; 1) a positive contribution from a curvature-dependent mechanism, conferring the ability to sense curvature (a prime example of this is the insertion of hydrophobic residues or lipid anchors into curvature-induced membrane defects (23)), and 2) a net gain in energy from membrane association without which membrane binding will obviously not occur. A recent review (13) concluded that hydrophobic and electrostatic membrane interactions must be in opposition for membrane curvature sensing to occur. Our study, in line with previous findings (28,30,65), shows that this is not a necessary condition. Several of the examined peptides showed strong curvature sensing despite positive interactions between interfacial lysines and the negatively charged membrane. The "opposing forces" hypothesis correctly predicts that if the contribution of (1) is minor compared with the overall binding energy, curvature sensing will be weak or undetectable. However, it is the combined interplay of all influences on binding that ultimately determine protein binding and sensing behavior; it is possible to show curvature sensing in combination with positive electrostatic interactions with the membrane provided that the hydrophobic interactions are sufficiently strong. Similarly, a protein that has weak interactions with the membrane overall but a strong hydrophobic contribution can be a very strong sensor, as seen for the Mut2 peptide with the hydrophobic face augmented by two Ala Val substitutions (Table 1 and Fig. 4). Conversely, too strong electrostatic interactions led to ubiquitous membrane binding, thus precluding curvature sensing and potentially inducing membrane deformation. Therefore, curvature sensing is more likely to be observed in peptides and proteins showing weak, reversible membrane binding, which could explain the combination of recurring hydrophobic and hydrophilic residues in the hydrophobic face of ␣-synuclein (14).
Our results indicate that the inserting AH itself is not the only feature of membrane-binding proteins that influences curvature sensing. Rather, factors such as net protein charge and charge distribution can also have a significant effect in modulating sensing, and the case of Mut3 illustrates that charges distal to the interface are also important in this regard. Furthermore, in terms of the amphipathic helix itself, commonly used physicochemical properties such as mean hydrophobicity and hydrophobic moment are not sufficient to describe the traits that determine curvature sensing. From this study there is no direct correlation between these parameters and binding either in terms of sensing ability or B frac . Thus, to better understand TABLE 5 The effect of the membrane charge on calcium-dependent binding of annexin B12 To determine the importance of membrane charge density on binding not mediated by hydrophobic insertion, calcium-dependent binding of annexin B12 was tested on three membrane compositions with different ratios of PC and PS (4:1, 2:1, and 1:2). Increasing membrane charge leads to an increase in bound fraction to practically complete binding at 33% PS and above. The sensing ability ␣ (see main text for definition) was completely diminished on these membranes. Values are the averages of three separate experiments.

Vesicles
Bound and compare AH recognition of membrane curvature, it is necessary to look at specific helical motifs, at the nature and environment of the amino acids and their position, and use accurate and quantitative techniques to measure curvature-sensitive binding. The conclusions of this study were made possible by the ability of the SLiC assay to deconvolve three parameters describing membrane binding ( B , B frac , and sensing ability) and reveal new aspects of membrane curvature sensing by amphipathic helices.