The Cytosolic Domain of Fis1 Binds and Reversibly Clusters Lipid Vesicles

Every lipid membrane fission event involves the association of two apposing bilayers, mediated by proteins that can promote membrane curvature, fusion and fission. We tested the hypothesis that Fis1, a tail-anchored protein involved in mitochondrial and peroxisomal fission, promotes changes in membrane structure. We found that the cytosolic domain of Fis1 alone binds lipid vesicles, which is enhanced upon protonation and increasing concentrations of anionic phospholipids. Fluorescence and circular dichroism data indicate that the cytosolic domain undergoes a membrane-induced conformational change that buries two tryptophan side chains upon membrane binding. Light scattering and electron microscopy data show that membrane binding promotes lipid vesicle clustering. Remarkably, this vesicle clustering is reversible and vesicles largely retain their original shape and size. This raises the possibility that the Fis1 cytosolic domain might act in membrane fission by promoting a reversible membrane association, a necessary step in membrane fission.

The morphology of both peroxisomes and mitochondria in mammals is correlated with expression levels of Fis1, not the dynamin-like mechanoenzyme, suggesting an important role for Fis1 in fission regulation [6,15]. Fis1 was initially discovered in yeast to be essential for the maintenance of mitochondrial homeostasis and is conserved in eukaryotes [2][3][4][5]16]. This small, 17 kDa protein is uniformly localized on the cytosolic side of the mitochondrial outer membrane by a single-pass, C-terminal transmembrane domain, with a similar orientation on the peroxisomal surface [2,6]. The Fis1 cytosolic domain is known to recruit other cytosolic factors that regulate or effect the fission process including Mdv1/Caf4 [9,14,[17][18][19][20][21][22][23][24] and Dnm1 [25]. However, Fis1 also binds lipid vesicles altering membrane integrity enough to cause release of small, but not large, molecules contained within the vesicles [26]. This release of entrapped molecules occurs independently of other protein factors and may reflect an innate ability of Fis1 to alter membrane structure in a manner important for fission. Notably, release of vesicle contents did not require the transmembrane domain and can be induced by the cytosolic domain of Fis1 alone, which raises the possibility that the cytosolic domain itself plays a role in altering membrane structure, independently of its accepted role in protein recruitment.
Many soluble proteins regulate their activity through reversible interactions with membranes [27][28][29]. During membrane fusion, soluble domain interactions with membranes are essential for some SNARE-mediated fusion events, especially involving the SNARE Vam7p [30,31] and synaptotagmin [32][33][34][35]. Whether these proteins directly deform membranes to mediate fusion is unclear, but it is likely that both membrane fusion and fission require proteins to stabilize high-energy membrane intermediates. Therefore, we would expect that the soluble domain may bind to synthetic membranes and alter their structures in a reversible manner. Here, we ask whether the soluble, cytosolic domain of Fis1 shares these properties. We find this domain alone is able to bind reversibly to lipid vesicles and determine what promotes this interaction. We identify that Fis1 undergoes a membrane-induced conformational change that clusters lipid vesicles. This clustering appears completely reversible, raising the possibility that Fis1 might help drive the association of two apposing bilayers that is required for membrane fission.

Protein Purification
DNA encoding the S. cerevisiae Fis1 gene lacking the C-terminal 27 residues (Fis1DTM) was subcloned into pET29b (EMD Biosciences) including a Tobacco etch virus (TEV) protease site and a C-terminal 6xHis tag. Plasmids were transformed into chemically competent Escherichia coli Rosetta cells (Novagen) and grown at 37uC in Luria broth with kanamycin (30 mg/mL) and chloramphenicol (34 mg/mL) to A 600 of 0.7. Protein expression was induced by addition of 0.25 mM isopropyl 1-thio-b-dgalactopyranoside at 18uC and harvested 15-18 h later by centrifugation. The resulting cell pellets were resuspended in column buffer (20 mM Tris HCl, 250 mM NaCl, 1 mM dithiothreitol, 20 mM imidazole pH 7.4) containing protease inhibitors (Roche Applied Science). Cells were lysed with 4 passes through an Emulsiflex C3 (Avestin), DNase was added to 1 mg/ mL and lysates were clarified by centrifugation. Protein was isolated from the resulting supernatant by affinity chromatography using Ni-Sepharose 6 Fast Flow beads (GE Healthcare), and eluted with a 100 mL linear gradient of column buffer with 500 mM imidazole. TEV protease was added at 1/100 molar ratio, put immediately into dialysis into column buffer at 4uC for 18 hours or until the protease reaction reached completion, determined by SDS-PAGE. Fis1DTM was separated from TEV protease and further purified on a Superdex-75 16/60 prep column (Amersham-Pharmacia). Samples were pooled and concentrated to ,500 mM and stored at 4uC. Sample purity was checked by coomassie-stained SDS-PAGE and was typically greater than 95%.

Vesicle Extrusion
All synthetic lipids were obtained from Avanti Polar Lipids (Alabaster, AL). The correct ratio of lipids were measured from chloroform stocks, mixed, and dried in a thin film under a stream of nitrogen. Dried films were lyophilized for at least 2 hours to remove excess chloroform and were resuspended in the required amount of deionized water to make a 12.6 mM solution of lipids. Lipid solutions were freeze-thawed in a dry ice/ethanol bath and 37uC water bath 11 times. Freeze-thawed solutions were extruded 11 times through a 100 nm nucleopore track etch membrane (Whatman), using an Avanti syringe extruder apparatus (Avanti Polar Lipids). Vesicle size, homogeneity and reproducibility were determined using dynamic light scattering and electron microscopy.

Vesicle Sedimentation
Vesicle sedimentation assays were performed with 5 mM protein in a 100 mL volume buffered with 20 mM potassium acetate and 20 mM potassium phosphate. Reactions were incubated with nutation for six hours at 25uC, and then subjected to centrifugation at increasing speeds (1,500, 6,000 and 18,000 rcf) for 30 minutes each [38,39]. The top 80 mL was drawn off and diluted into 46 SDS loading buffer; the remaining 20 mL, which contained a mixture of soluble protein and trace lipids, was discarded. The lipid pellet was resuspended in 100 mL of buffer, and then diluted into 46 SDS loading buffer. 15 mL of each fraction was analyzed by SDS-PAGE and stained with coomassie G-250 Blue Silver stain for 18 hours and destained until background was negligible [40]. For samples measuring reversibility, the lipids were pelleted and resuspended by pipetting up and down with a 200 mL pipette, incubated for 2 hours, spun at three speeds and then prepared for SDS-PAGE as previously noted. Gels were scanned on a Canoscan flatbed scanner (Canon USA), band volume was calculated using Image Quant TL (Amersham), fraction bound was calculated using the equation: (V pellet )/(V pellet +V supernatant ), where V pellet and V supernatant are the band volumes of the pellet and supernatant respectively. A Fis1DTM standard curve was made to assure the quantitation did not suffer from non-linear artifacts in the range of concentrations used, from the densitometry and was found to contain at most a 5% deviation from linearity. We corrected this deviation by applying a quadratic correction factor calculated from a Fis1DTM standard curve, (N = 2). Carbonate extraction buffer was 200 mM Na 2 CO 3 , pH 11.7.

Circular Dichroism
Far UV Circular dichroism was performed on a Jasco J-720 circular dichroism spectropolarimeter with 5 mM Fis1DTM in reaction buffer with the indicated lipid vesicles, 25uC using a 1 mm pathlength, 3 accumulation average, with a scan rate of 20 nm/ min, sensitivity of 20 mdeg, 2 second response from 190-260 nm. Samples were incubated for at least 2 hours before scanning at 25uC. A reference scan without protein in the presence and absence of lipids was subtracted to remove background ellipticity. Scans with vesicles alone were identical to buffer alone. Mean

Steady-State Fluorescence Spectroscopy and Acrylamide Quenching
Steady-state fluorescence measurements were done on a SLM-48000 spectrofluorometer (SLM-AMINCO, Urbana, IL). A 5 mm cuvette was used with 5 mM Fis1DTM (10 mM Trp) in reaction buffer. Excitation was at 295 nm (2 nm slit width) and emission at 310-370 nm (16 nm slit width). Polarizers were set at 90 degrees to reduce vesicle scatter. Spectra were baseline corrected using an identical sample that lacked protein. For quenching studies, acrylamide (Fisher Scientific) at the indicated concentration was added 5 minutes before the scan and the resulting data fit to the Stern-Volmer equation, F 0 /F = 1+K sv [Q], where F 0 is the fluorescence in the absence of the quencher, F is the fluorescence in the presence of the quencher, K sv is the Stern-Volmer quenching constant and [Q] is the concentration of quencher. Best fit lines were calculated using Igor Pro (Wavemetrics, Portland OR).

Dynamic Light Scattering
Vesicle sizes were measured using dynamic light scattering on a Zetasizer nano ZS90 instrument equipped with a 633 nm laser, (Malvern). The dispersant was 425 mL of reaction buffer with 20 mM potassium acetate, 20 mM potassium phosphate using a calculated viscosity of .8910 cP and refractive index of 1.330. Before each measurement the sample was equilibrated at 25uC for 2 minutes. Each measurement was an average of 10 runs of 15 seconds and three measurements were averaged to give a final mean Z-average (Z-avg). All measurements were analyzed using the Dispersion Technology Software (Malvern) using the cumu-Fis1 Amphitropism Reversibly Clusters Membranes lants analysis on Multiple Narrow Mode. All samples collected had a polydispersity index (PDI) less than 0.2, indicating that the vesicles sizes were found to be monomodal. Protein alone under these conditions did not scatter enough light for the cumulants analysis.

Electron Microscopy
Samples were prepared by mixing Fis1DTM and LUVs at the indicated concentration in reaction buffer at the designated pH. Copper formvar coated grids (Electron Microscopy Science) were ionized and then floated on 10 mL of samples for 5 minutes. The grids were rinsed quickly in deionized water, then imbedded into a 0.2% methylcelluose, 3.2% polyvinyl alcohol, 0.4% uranyl acetate solution. Images were taken on a Philips EM 420 TEM at 100 kV equipped with a SIS Megaview III CCD digital camera (Olympus).

Light Scattering Kinetics by UV/Vis Spectrophotometry
Vesicles were added to a final concentration as noted to a 200 mL blanked buffer solution at either pH 5.0 or pH 7.0, mixed and the OD was measured on a Nanodrop 2000c UV/Vis spectrophotometer (Thermo Scientific) using a 1 cm pathlength cuvette, until the values reached equilibrium. At 450 nm, the absorbance due to the aromatic residue signal from the protein is negligible and the majority of the signal is due to light scattered at angles other than 180 degrees. For the experiment, the absorbance at 450 nm was measured every 30 seconds. An initial baseline was made with lipids and buffer alone at time 0. At approximately 1000 sec, 5 mM Fis1DTM (,500 mM stock concentration, ,1% volume change) was added, mixed by pipette and continued recording without missing a time point. At 4000 sec the solution was titrated either from pH 5.0 to pH 7.4 using a 4 M solution of fresh KOH (,1% volume change) or from pH 7.0 to pH 4.8 with HCl (,1% volume change) and recorded until 7000 sec time point. The pH of samples were confirmed using a pH meter to be within 0.5 a pH unit of the target pH.

Kinetics of Vesicle Clustering
To estimate the rate equation for Fis1-induced vesicle clustering the kinetics at high vesicle:protein ratios were measured under conditions in which vesicle collisions were not rate-limiting for clustering (up to 200 protein molecules per vesicle). At the higher protein concentrations, the vesicle concentration is effectively removed from the rate equation: where V i is the initial velocity of the reaction, k is the apparent rate constant, [Fis1DTM] is the concentration of Fis1DTM and n is the order of the pseudo-rate constant. By measuring the initial velocity as a function of Fis1DTM concentration, the data can be fitted to provide an estimate of the value of n.
The experiments were performed with 100 mM 60:40 DOPC :DOPG (the highest lipid concentration to still give a measurable scattering signal at these ratios) in 20 mM potassium phosphate, 20 mM potassium acetate at pH 5.0 with stirring. The association reaction was started by addition of Fis1DTM in 20 mL of buffer to reach a final concentration that ranged from 1 nM to 200 nM in a 2 ml final volume. Scattering of each sample was measured by monitoring the absorbance at 333 nm at 10 second intervals with a 1 cm pathlength on a Nanodrop 2000c using cuvette mode (Thermo Scientific). This sampling rate was insufficient to reasonably fit the data at protein concentrations greater than 200 nM. For each measurement, the sample was blanked with lipid and buffer only and incubated for 5 minutes to establish a baseline reading. The absorbance (OD) versus time (seconds) was fit to a single exponential function to determine the initial velocity of association for each Fis1 condition using the software Igor Pro (Wavemetrics, Portland OR). The V i reported is from 3 independent experiments. The V i was plotted as a function of [Fis1] and fit to the above rate equation for n = 1, 2, or 3.

Results
The Fis1 cytosolic domain binds vesicles with lipid compositions that mimic the mitochondrial outer membrane Previously the cytosolic domain of Fis1 (Fis1DTM) was shown to induce permeabilization of the small molecule ANTS from DOPC:DOPG (6:4) lipid vesicles at pH 5.0, but not at pH 7.0 [26]. To better understand this permeabilization, we first tested whether Fis1DTM could bind to lipid vesicles that mimic the mitochondrial outer membrane (Mitomix) [37]. In a vesicle sedimentation assay, we measured the amount of Fis1DTM bound to Mitomix lipid vesicles as a function of increasing lipid concentrations (Figure 1). At pH 5.0, Fis1DTM binding increased in a sigmoidal manner and saturated near 80% protein bound (1:100 protein:lipid) ( Figure 1A). At pH 7.0, with the highest lipid concentration tested (1:400, P:L), ,5% Fis1DTM bound to vesicles ( Figure 1B). By contrast using a similar composition of lipid vesicles to that used in our earlier work (Br 4 DSPC:DOPG,6:4), we found a similar trend. At pH 5.0, Fis1DTM bound to a greater extent than the Mitomix vesicles and reached saturation at approximately 90% bound, (Figure 1C). At pH 7.0, with the highest lipid concentration tested (1:400, P:L), the interaction was ,15% bound ( Figure 1D). We consistently observed this increase in protein binding to Br 4 DSPC:DOPG compared to Mitomix vesicles, which may arise from an increase in the amount of negatively charged lipid head groups compared to Mitomix vesicles (20% anionic headgroup for Mitomix vs. 40% for Br 4 DSPC:DOPG mixture).

Electrostatic forces stabilize membrane partitioning of Fis1DTM
To test the effect of electrostatic forces on this interaction, we titrated the negatively charged lipid, DOPG, from 40% to 0% while maintaining the total lipid concentration using DOPC, and measured fraction protein bound. At both pH 5.0 and 7.0, decreasing the amount of negatively charged lipid (DOPG) decreased the amount of Fis1DTM bound (Figure 2). At pH 5.0, the largest decrease occurred between 20 to 10% DOPG (Figure 2A), whereas the largest decrease at pH 7.0 occurred from 40% to 30% DOPG ( Figure 2B). These data indicate an electrostatic contribution to vesicle binding that arises from both the negative charge of the vesicles and from protonation of ionizable groups on Fis1DTM at pH 5.0. Altering the DOPE concentration did not affect Fis1DTM membrane binding (data not shown) suggesting that neutral lipids had little affect on binding. Consistent with this interpretation, the fraction of Fis1DTM bound to Mitomix vesicles (,80% bound at P:L of 1:400 with 20% anionic content in Figure 1A) is similar to the fraction of Fis1DTM bound to PC/PG vesicles with the same ratio of negatively charged lipids (20% DOPG). The pH and anionic lipid dependence to Fis1 vesicle binding indicated a strong electrostatic contribution to the interaction. To further test this idea, we measured Fis1DTM binding to PC/PG membranes as a function of increasing concentration of KCl and observed a strong attenuation of binding at high salt (Figure 3). At 300 mM KCl, Fis1 binding was barely detectable in the vesicle fraction. This result was not a function of K + because Na + had a similar effect (data not shown). Since neither KCl nor NaCl significantly affects Fis1DTM structure or stability, we interpret these data to indicate that membrane binding is favored by electrostatic interactions between the anionic headgroups of the lipid bilayer and the protein. Because the lipid headgroups are not thought to be titrated from pH 7.0 to 5.0 [42,43], we attribute this affect to protonation of ionizable groups on Fis1.

Fis1DTM is amphitropic and binds membranes reversibly
An important consideration of the Fis1-membrane interaction is whether Fis1DTM binds reversibly to membranes as a peripheral membrane protein, or is more characteristic of a permanently associated integral membrane protein. Therefore, we tested whether the Fis1DTM-membrane interaction is reversible by pH change, ionic strength change, and carbonate extraction. In these experiments, we bound protein to membrane vesicles (pH 5.0, 1:20, P:L), collected the lipid vesicles by sedimentation, resuspended in the indicated wash buffer, incubated for 2 hours, and determined the fraction bound. As expected, resuspension of pelleted vesicles with Fis1DTM in pH 5.0 buffer left all the protein in the lipid vesicle fraction (Figure 4, pH 5.0). In contrast, upon resuspension in pH 7.0 buffer, little protein was left in the lipid fraction (Figure 4, pH 7.0). This finding is similar to our initial binding results without first pre-binding Fis1DTM to the membranes ( Figure 1) and shows that the pH-dependence of Fis1DTM binding to membrane vesicles is reversible. Increasing concentrations of KCl also resulted in partitioning of protein into the soluble fraction from the lipid fraction, (Figure 4, K150 and K300). A typical test to determine whether a protein interacts peripherally with the membrane is a carbonate extraction experiment, which resulted in ,5% of protein in the vesicle fraction ( Figure 4, HCO 3 ). Together these results are consistent with the cytosolic domain of Fis1 binding to membranes in a manner that resembles a peripherally bound protein.

Fis1DTM undergoes a conformational change upon membrane binding
Reversible protein-membrane interactions are often associated with conformational changes of the protein from the soluble conformation to a membrane-bound conformation [29]. To assess this, we measured Far-UV CD spectropolarimetry of Fis1DTM in the presence and absence of lipid vesicles. In the absence of vesicles, the spectra are typical of well-folded a-helical proteins, with minima at 208 and 222 nm at both pH 7.0 and pH 5.0    Figure 5). Although slow-timescale dynamics differ for regions of Fis1 between these two pH values [44], pH 7.0 and 5.0, CD spectra showed little differences ( Figure 5A and 5B). Upon addition of 500 mM lipid vesicles at pH 5.0, the a-helical signal increased indicating an increase in secondary structure upon vesicle binding ( Figure 5A). Using the method of Luo and Baldwin, vesicle binding increased the Fis1DTM helicity by 20%, from 58 to 78% [41]. No change was found after addition of vesicles at pH 7.0 ( Figure 5B).

Membrane binding buries Fis1DTM tryptophan residues
Changes in tertiary structure upon membrane binding were assessed by intrinsic tryptophan fluorescence in the presence and absence of lipid vesicles. Fis1DTM contains two tryptophans that are in close proximity to each other and are partially protected from bulk solvent in the NMR structural ensemble (1y8m.pdb) [20]. Tryptophan 7 lies in the N-terminal arm that is thought to block access to a concave binding surface that includes tryptophan 47 and other evolutionarily conserved residues [25]. At pH 5.0 in the presence of 500 mM lipids (1:100, P:L), the fluorescence signal increased significantly and was accompanied by a blue shift in l max by 6 nm (Figure 6A). At pH 7.0 in the presence of 500 mM lipids, the fluorescence signal slightly increased with little change in l max consistent with the conditions where we found ,10% of Fis1DTM bound ( Figure 6B). While changes in fluorescence intensity are difficult to interpret, the blue shift in l max upon membrane binding indicates a change in the Trp environment that is more non-polar.
In order to compare the solvent accessibility between the solution and membrane-bound conformations of Fis1DTM, we measured the Stern-Volmer quenching coefficient (K SV ) for acrylamide, a water-soluble quencher of intrinsic tryptophan fluorescence. The acrylamide quenching data were linear at both pH 5.0 and pH 7.0 ( Figure 6C and 6D) indicating that the two tryptophans in Fis1DTM have similar solvent accessibilities. This linearity also allows the data to be fit to determine K SV . At pH 5.0 and 7.0 in the absence of vesicles, we found that Fis1DTM tryptophans are equally protected with K SV values of 4.62 M 21 at pH 5.0 and 4.77 M 21 at pH 7.0. The K SV value provides a measure of solvent accessibility and these values are typical for solvent exposed to moderately exposed tryptophans in soluble proteins [45][46][47]. By contrast, under conditions in which Fis1DTM robustly binds vesicles at pH 5.0, the K SV value decreased significantly to 0.91 M 21 (from 4.62 M 21 ) indicating that Fis1DTM tryptophans are significantly more protected from solvent upon membrane binding ( Figure 6C). In the presence of lipid vesicles at pH 7.0, we found only a slight increase in protection consistent with the intrinsic tryptophan fluorescence data ( Figure 6D). We interpret the fluorescence and circular dichroism data collectively to indicate that Fis1DTM undergoes a dramatic structural rearrangement upon membrane binding that results in burial of its tryptophan residues while retaining its secondary structure. We also note that this conformational change is completely reversible (data not shown).

The cytosolic domain of Fis1 affects membrane structure
Fis1 is proposed to mediate membrane fission by recruiting the dynamin-like mechanoenzyme, Dnm1, to sites of fission. During endocytosis, dynamin mediates membrane scission along with other molecules that also alter membrane structure and are thought to be important for this process [48,49]. Given the reversible, membrane-induced conformational change of Fis1, we asked whether Fis1 itself might alter membrane structure. To test this idea, we used dynamic light scattering to determine the size distribution of lipid vesicles in the presence and absence of Fis1DTM and as a function of increasing lipid concentration. Incubation in the absence of protein had no effect on vesicle size distribution over a 24 hr period, with a mean Z-average of 10461 nm, PDI = 0.17 ( Figure 7A, control). At pH 5.0, Fis1DTM incubation increased the Z-average size of the vesicles in all conditions tested, with the highest protein:lipid ratios giving the species with the greatest apparent size ( Figure 7A  We next used negative staining electron microscopy to assess the origin of increased light scattering. Vesicles in the absence of protein at pH 5.0 or pH 7.0 were indistinguishable except for a slightly darker staining at pH 5.0. At a 1:40 protein:lipid ratio, we found large aggregates that appeared composed of clustered vesicles that retained their original spherical shape ( Figure 7B). Unlike protein-mediated vesicle fusion [50], we found no evidence for large spherical vesicles. Our results are inconsistent with Fis1DTM mediating a simple fusion of membranes, and are more consistent with closely apposed membranes that maintain the original curvature of the 100 nm vesicle. At lower protein:lipid ratios large clusters still formed, but at a much lower frequency (,1% of total vesicles). Vesicles incubated with protein at pH 7.0 were indistinguishable from vesicle-alone controls ( Figure 7B), consistent with the light scattering data, which indicates that significant binding of Fis1 is necessary for vesicle clustering.
The conclusions from electron microscopy were confirmed using fluorescence light microscopy of 100 nm vesicles that have been doped with the fluorescent lipid Rh-DOPE. Under similar conditions as before (pH 5.0, 1:20, P:L), larger particles (.1 mm) were visible after 5-10 seconds and grew with time (data not shown).
Vesicles appeared to cluster in a random manner, often forming inhomogeneous strings of vesicles or large vesicular clumps. Incubation overnight or longer, resulted in clusters visible to the naked eye. At all times, the most abundant and largest complexes were seen at the highest protein:lipid ratios. Vesicle clustering also occured at lower protein:lipid ratios but was less frequent and resulted in smaller aggregates.
Given that Fis1DTM binding to lipid vesicles is reversible, we asked if the Fis1-induced membrane clustering was also reversible. We first induced clustering at pH 5.0 by adding Fis1DTM at a 1:40 protein:lipid ratio. This sample after 30 minutes was opaque to the naked eye indicating vesicle clustering. After 1 hour, we titrated this solution from pH 5.0 to pH 7.0, incubated for 1 hour and then analyzed the vesicle population by negative-stain EM. Surprisingly, we found that the vesicles returned to their original size and shape (Figure 7b). To determine the time-dependence of vesicle clustering, we monitored light scattering as measured by a change in optical density at 450 nm (OD 450 ), where the OD 450 of the solution is proportional to vesicle clustering. At pH 5.0, protein addition to vesicles (P:L, 1:20) dramatically increased the OD 450 signal ( Figure 8A). Upon addition of concentrated base to bring the pH to 7.0, the OD 450 signal precipitously fell to values similar to those in the absence of protein ( Figure 8A, KOH arrow). Since little Fis1DTM is bound under these conditions at pH 7.0, we interpret these data to indicate a reversal of Fis1-induced vesicle clustering within 30 sec. Analysis of these vesicles by dynamic light scattering gave a Z-average of 164 nm (data not shown). However, this value is skewed by a small number of larger vesicles that remain clustered (PDI = 0.254 and .2000 nm as determined by DLS and EM). These data indicated that clustered vesicles predominantly returned to their original size distribution. We next pre-incubated Fis1DTM with vesicles at pH 7.0 and found no change to the OD 450 signal, consistent with our earlier observations of little Fis1DTM bound under these conditions ( Figure 8B). By contrast, acidification of this solution to pH 5.0 increased the OD 450 in a near linear fashion ( Figure 8B, HCl arrow). Similar titrations of protein or vesicle samples alone did not change the OD 450 signal (data not shown). The rate of vesicle clustering in these two experiments differed, which likely arises from differences in local protein concentration that may indicate a requirement for protein oligomerization.
To assess whether Fis1 oligomerization is involved in vesicle clustering, we conducted further kinetic measurements at a constant lipid concentration (100 mM) that was not rate-limiting for clustering ([Lipid]&[Protein]). We assumed under these conditions, that the initial rate of clustering is primarily dependent on the concentration of protein (see materials and methods). We PLoS ONE | www.plosone.org measured the initial velocity of clustering at a constant vesicle concentration as a function of increasing concentrations of Fis1DTM to determine the overall order of the reaction, which is related to the number of Fis1 molecules required for each clustering event. We found that the initial velocity increased in a non-linear manner when plotted against [Fis1DTM] ( Figure 9A) and was best fit to a reaction rate model that was pseudo-second order in Fis1 concentration with an overall apparent rate constant of k = 1.82610 26 nM 21 s 21 and a coefficient of determination of R 2 = 0.9866. By comparison, reaction rate models that were either pseudo-first order or pseudo-third order gave poor fits to the data (R 2 = 0.9525 and R 2 = 0.9285, respectively, Figure 9B-D). These data are consistent with at least two Fis1 molecules being necessary for vesicle clustering.

Discussion
In this study, we show that the cytosolic domain of fission protein, Fis1, undergoes a membrane-induced conformational change that results in lipid vesicle clustering. This clustering does not result from membrane fusion, but is reversible and vesicles largely return to identical size and shape. This reversibility is rapid and the Fis1-membrane association is favored by a high protein:lipid ratio. Whether vesicle clustering is required for organelle fission is not known. However, many proteins involved in membrane dynamics mediate a close membrane association for their cellular function [51]. For membrane fission, the mechanics of each scission event requires the close apposition of two bilayers. In order to complete scission, each of the apposing bilayers must fuse adjacent to the site of scission. Thus each membrane scission event requires two membrane fusion events and would likely be aided by proteins that could promote reversible membrane association [52]. Our data leads us to speculate that vesicle clustering induced by the Fis1 cytosolic domain might play a role in organelle fission.
Several issues need to be addressed for this speculation to be correct. First, we expect vesicle clustering to be specific to Fis1. Although visible aggregation between positively charged proteins and negatively charged membranes has been reported [53,54], the structure-based PROPKA method [55] estimates the mean charge of Fis1 at pH 5.0 to be 21. Therefore, it is unlikely that Fis1 associates with negatively charged membranes using a simple mechanism of net positive charge, especially given the random distribution of charges in its structure [20]. The conformational change upon membrane binding is also greater than one might expect from a non-specific interaction. Notably, the soluble domains of other tail-anchored proteins, Bcl-xL and CED-9, have not induced vesicle clustering in similar experiments to those presented here [38,56,57], supporting the idea that reversible vesicle clustering may be specific to Fis1.
The apparent requirement for protonation to induce vesicle clustering may seem unlikely in vivo, but two considerations may address this concern. First, Fis1 does indeed bind to membranes at pH 7.0, though weakly, and this interaction could be enhanced by other factors that are not present in our experiments. For instance, the cytosolic domain is normally tethered to the mitochondrial outer membrane by its C-terminal transmembrane domain and this tethering likely enhances the protein-membrane interaction significantly [58]. This tethering may also reduce the apparent need for protonation in clustering vesicles, as has been shown for the tail-anchored Bcl-2 protein, Bcl-xL [38]. However, for Fis1 we have been unable to test this idea because the full-length molecule has proven difficult to isolate and to reconstitute into membranes in a form similar to the solution structure. Second, it may be that the effective pH at the surface of intracellular membranes is lower than the cytosol. This idea is based on recent evidence that a negatively charged membrane surface, such as that found on peroxisomes and mitochondria, has an effective pH that is lower than bulk solution [59][60][61][62][63]. This effective decrease in local pH at the surface of these membranes would be expected to promote the Fis1-membrane interaction. In addition, other factors, such as change in the molecular environment around the membrane or membrane curvature may also be important for inducing a Fis1membrane interaction.
The high concentrations of Fis1 necessary for vesicle clustering might also seem unlikely, in vivo. Fis1 is thought to be distributed uniformly on the surface of mitochondria [2], but recruits proteins specifically to sites of scission. Therefore a physiological, yet unknown, mechanism exists to determine a site of scission, which may involve assembling Fis1 molecules. Our data supports at least PLoS ONE | www.plosone.org two Fis1 molecules being necessary for vesicle clustering. Previously, mammalian Fis1 has been shown to oligomerize in cross-linking and blue-native PAGE experiments [64,65] and the cytosolic domain of human Fis1 crystallized as a dimer [66]. Whether Fis1 from budding yeast oligomerizes in vivo is not known, however, we have found that it dimerizes in solution upon recombinant expression (unpublished data). This raises the possibility that Fis1 could act in a SNARE-like manner to tether two vesicles together, although our data cannot exclude the possibility of dimerization on the same vesicle or two molecules that remain independent. This work lays the foundation for identifying mutations in Fis1 that disrupt vesicle clustering, possibly by disrupting dimerization. Such mutagenesis studies will be useful for determining the mechanism of clustering and its physiological role.
Reversible protein-membrane interactions are well-appreciated in the regulation of many cellular processes [27][28][29]. These processes often involve amphitropic domains that exist in both distinct soluble and membrane bound conformations, similar to the cytosolic domain of Fis1. Amphitropic proteins often utilize a conformational change as a result of a cellular signal to regulate their biological activities [67,68]. In one case similar to that reported here, the amphitropic nature of creatine kinase may be oligomerization-dependent in a manner important for mitochondrial inner membrane morphology [69]. More generally, our work also highlights a different kind of protein amphitropism that involves transmembrane-anchored proteins, which play important roles in a variety of biological processes [37,[70][71][72]. The human genome is estimated to contain approximately 400 tail-anchored proteins [37,72]. Given the ability of the transmembrane domain to increase the effective concentration of the cytosolic domain at the membrane, it is reasonable to ask whether the cytosolic domains of these proteins undergo reversible membrane interactions. Indeed, such interactions are evolutionarily conserved between the pro-and anti-apoptotic Bcl-2 proteins where a critical event appears to be the permeabilization of the mitochondrial outer membrane by the cytosolic domain of proapoptotic Bax [73][74][75][76][77][78]. Tail-anchored proteins, such as these, must have evolved mechanisms distinct from other proteins that either prevent (non-amphitropic) or promote (amphitropic) interactions with the membrane. Such mechanisms are currently unknown, and difficult to measure in vivo, but our work here suggests that controlling the electrostatic forces of these interactions may be critical.
In conclusion, we have shown that a single protein domain has the ability to mediate dramatic changes in membrane structure through reversible clustering. We speculate that the Fis1membrane interaction is important in mediating membrane association that occurs during membrane fission.