A Pore-Forming Toxin Requires a Specific Residue for Its Activity in Membranes with Particular Physicochemical Properties*

Background: It is unclear how actinoporins adapt to the broad range of physicochemical landscapes of biological membranes. Results: A single mutation in fragaceatoxin C (an actinoporin) prevents the formation of lytic pores in membranes in the liquid-ordered phase. Conclusion: Phe-16 is critical for pore formation of fragaceatoxin C in cholesterol-rich membranes. Significance: Site-directed mutagenesis generates lipid phase-sensitive pore-forming proteins. The physicochemical landscape of the bilayer modulates membrane protein function. Actinoporins are a family of potent hemolytic proteins from sea anemones acting at the membrane level. This family of cytolysins preferentially binds to target membranes containing sphingomyelin, where they form lytic pores giving rise to cell death. Although the cytolytic activity of the actinoporin fragaceatoxin C (FraC) is sensitive to vesicles made of various lipid compositions, it is far from clear how this toxin adjusts its mechanism of action to a broad range of physiochemical landscapes. Herein, we show that the conserved residue Phe-16 of FraC is critical for pore formation in cholesterol-rich membranes such as those of red blood cells. The interaction of a panel of muteins of Phe-16 with model membranes composed of raft-like lipid domains is inactivated in cholesterol-rich membranes but not in cholesterol-depleted membranes. These results indicate that actinoporins recognize different membrane environments, resulting in a wider repertoire of susceptible target membranes (and preys) for sea anemones. In addition, this study has unveiled promising candidates for the development of protein-based biosensors highly sensitive to the concentration of cholesterol within the membrane.

The physicochemical landscape of the bilayer modulates membrane protein function. Actinoporins are a family of potent hemolytic proteins from sea anemones acting at the membrane level. This family of cytolysins preferentially binds to target membranes containing sphingomyelin, where they form lytic pores giving rise to cell death. Although the cytolytic activity of the actinoporin fragaceatoxin C (FraC) is sensitive to vesicles made of various lipid compositions, it is far from clear how this toxin adjusts its mechanism of action to a broad range of physiochemical landscapes. Herein, we show that the conserved residue Phe-16 of FraC is critical for pore formation in cholesterolrich membranes such as those of red blood cells. The interaction of a panel of muteins of Phe-16 with model membranes composed of raft-like lipid domains is inactivated in cholesterol-rich membranes but not in cholesterol-depleted membranes. These results indicate that actinoporins recognize different membrane environments, resulting in a wider repertoire of susceptible target membranes (and preys) for sea anemones. In addition, this study has unveiled promising candidates for the development of protein-based biosensors highly sensitive to the concentration of cholesterol within the membrane.
Pore-forming toxins (PFTs) 4 comprise a heterogeneous family of globular proteins producing pores on cellular membranes leading to cell damage and, eventually, to cell death (1). Different types of PFTs have developed, through evolution, different molecular mechanisms of insertion and diverse pore architectures in response to specific characteristics of the target cell membrane (2). For example, perfringolysin O, a cholesterol-dependent cytolysin, triggers pore formation by binding to membrane cholesterol (3). In contrast, the cytolysin of Vibrio cholerae (4) requires membrane surfaces decorated with glycans (5). Once attached to the surface of the membrane, PFTs exhibit the remarkable ability to spontaneously self-assemble into cytotoxic pores without the need of exogenous energy input. The transformation of the toxin from a water-soluble form into lytic pores is facilitated by the interplay between the protein and the membrane lipids (6 -10). In addition, membrane damage is often intensified in the presence of bidimensional lipid domains such as the so-called lipid rafts (11,12).
Actinoporins are a group of PFTs secreted by sea anemones (13)(14)(15). This unique family of eukaryotic PFTs displays a high degree of sequence identity (16). Their globular structure presents a common fold based on a ␤-sandwich core flanked by two ␣-helices (17,18). The best characterized actinoporins are equinatoxin II (19), sticholysin II (20), and FraC (16). Actinoporins have been studied as potential immunotoxins (21) and serve as models to study protein-membrane interactions, a process of great relevance in neurodegenerative diseases (22,23).
Actinoporins bind to membrane phospholipids, which act as surface receptors. In particular, sphingomyelin (SM) is a lipid that facilitates binding and pore formation (24), although this lipid is not strictly necessary to permeabilize model membranes (25). Once bound to the membrane, actinoporins oligomerize and transfer their N-terminal helical region toward the hydrophobic core of the membrane, although the exact sequence of events is still under debate (19,26,27). During the insertion step, the N-terminal ␣-helix spans the entire thickness of the membrane and lines the wall of the octameric pore as recently shown by x-ray crystallography (28). Notably, all these processes are spontaneous, i.e. they do not require an exogenous input of energy.
The membrane architecture has been described as a key regulator of actinoporin toxicity (29). Intriguingly, membranes made of a single type of lipid are not susceptible to the lytic activity of the toxin (25,26,30). In contrast, mixtures of lipids containing SM and other lipids such as phosphatidylcholine (PC) or cholesterol often trigger pore formation. A well studied case is that of equinatoxin II. This actinoporin is active in model membranes composed of lipid mixtures with a coexistence of bidimensional domains of liquid-ordered (L o ), liquid-disordered (L d ), and/or solid (gel) phases (25,31,32). However, although a number of crystal structures of actinoporins are reported, including that of the lytic pore (28), the elements governing the sensitivity of actinoporins to the physicochemical properties of the membrane are poorly characterized.
Herein we show that the introduction of a single mutation at position 16 of the N-terminal region of FraC prevents pore formation in cholesterol-rich membranes. Intriguingly, this mutation has no effect in membranes with low cholesterol content. We propose that actinoporins adopt complementary strategies in response to the lipid packing density of the bilayer. These observations suggest that specific structural elements are used to widen the repertoire of biological membranes susceptible to this family of hemolytic toxins.
Expression and Purification of FraC-Expression and purification of FraC were carried out as previously described (16) with minor modifications. Briefly, Escherichia coli BL21 (DE3) was grown until the A 600 was 0.5 before inducing with 1 mM isopropyl ␤-D-thiogalactopyranoside. Cells were harvested after 5 h by centrifugation at 8,000 ϫ g for 10 min, followed by sonication in buffer A (50 mM Tris-HCl, pH 7.4). The supernatant was loaded into a Resource S cation exchange column equilibrated with buffer A. The protein was eluted in buffer A with increasing concentrations of NaCl (0 -1 M). The hemolytic fractions were pooled and subsequently subjected to size-exclusion chromatography (SEC) in a HiLoad 16/60 Superdex 75 prep grade column (GE Healthcare) equilibrated with 50 mM Tris-HCl, 200 mM NaCl, pH 7.4. Site-directed mutagenesis to generate F16P was carried out using mutagenic oligonucleotides (33). F16A and F16G muteins were generated using the KOD-Plus mutagenesis kit (Toyobo, Osaka, Japan) following the manufacturer's instructions.
Activity Measurements-Hemolytic activity was performed as described previously (16). Briefly, sheep RBC were washed in phosphate buffer and standardized at A 700 of 0.6. Kinetics of hemolysis started at the time of toxin addition (14 nM). To determine HC 50 (the protein concentration required for 50% hemolysis), washed erythrocytes were added to a 96-well plate with serially diluted toxin. The presence of hemoglobin in the supernatant was measured at A 412 after 90 min. Extraction of cholesterol was carried out as in Ref. 34 with minor modifications. Standardized cells were incubated with 2 mM methyl-␤cyclodextrin (MCD) for 30 min at 37°C with gentle agitation. Cells were then extensively washed prior to the addition of toxin as indicated above. Quantification of cholesterol was performed with BioVision (San Francisco, CA) cholesterol assay kit. Approximately 55% of the cholesterol was extracted using this methodology. The source of the RBC employed for this experiment was rabbit instead of sheep because the amount of cholesterol that can be extracted without compromising the cell integrity was higher.
Leakage of Encapsulated Solutes-The leakage of encapsulated solutes was assayed as described by Ellens et al. (37). Changes in fluorescence intensity were recorded either in a F-2500 spectrofluorimeter (Hitachi, Tokyo, Japan) or a PHERAstar Plus microplate reader (BMG LABTECH, Ortenberg, Germany). The excitation and emission wavelengths were set at 355 and 515 nm, respectively. Complete release of the fluorescent probe was achieved by addition of Triton X-100 (0.1%, v/v) or n-octyl-␤-D-glucoside (1.4%, w/v). The percentage of leakage was calculated using the following equation, where F f is the fluorescence reached after toxin addition, F 0 is the fluorescence of the vesicles at the beginning of the experiment, and F 100 is the fluorescence value measured after addition of detergent. Measurements were carried out at 25°C with constant stirring in a Hitachi F-2500 spectrofluorimeter (Tokyo, Japan).
To obtain kinetic parameters, the data (expressed as the percentage of leakage) were fitted to a single-exponential time function, where p(t) is the number of pores per vesicle corrected to a uniform vesicle size and defined as follows, with being v f as the final velocity, v i as the initial leakage velocity, k as the time constant, and t as the time (38).
Purification of Oligomeric FraC-Toxin (100 M) was incubated with LUVs (lipid:protein molar ratio 200:1) for 30 min at room temperature in buffer 50 mM Tris-HCl, 200 mM NaCl, pH 7.4. The protein-liposome mixture was solubilized with 1% (v/v) Triton X-100; diluted in 50 mM Tris-HCl, pH 7.4, 0.3 mM DDM; and applied to a Resource S column. The protein was eluted with a linear gradient of 50 mM Tris-HCl, 1 M NaCl, pH 7.4, 0.3 mM DDM. The fractions containing protein were pooled and analyzed by SEC using a Superdex 200 10/300 GL column equilibrated with 50 mM Tris, 200 mM NaCl, pH 7.4, 0.3 mM DDM. The void volume of the column (V 0 ) was 8.5 ml as determined with blue dextran (not shown). The column volume V c was 21.2 ml as determined with acetone. The elution volume of monomeric FraC was V m ϭ 27.9 ml, higher than the column volume because of the unspecific interaction of the monomeric toxin with the column matrix (16).
Thermal Stability by Circular Dichroism-Thermal denaturation was monitored by CD using a J-820 spectropolarimeter (Tokyo, Japan) equipped with a Jasco PTC-423L temperature controller. The measurements were carried out using quartz cuvettes with 1-mm pathlength. Protein (0.5 mg/ml) was dissolved in 20 mM Tris-HCl, pH 7.4. Unfolding was monitored at 210 nm in the temperature range 20 -80°C with a heating rate of 60°C/h and a 2-nm bandwidth. Melting temperatures (T M ) were calculated with the software provided by the manufacturer of the instrument.
High Resolution Crystal Structure-Crystals of F16P suitable for x-ray diffraction analysis were obtained by mixing 1 l of freshly purified protein at 14 mg/ml in 10 mM Tris-HCl, pH 8, and 1 l of crystallization solution composed of 22% PEG 3350, 10 mM Tris-HCl, and 0.2 M potassium fluoride, pH 8, at 20°C. Crystals were briefly transferred to a solution of mother liquor supplemented with 20% (v/v) glycerol and subsequently frozen in liquid nitrogen. Data collection was carried out at Beamline BL5A of the Photon Factory (Tsukuba, Japan) under cryogenic conditions (100 K). The structure of F16P in the monomeric state was solved by the molecular replacement method using the coordinates of a protomer of FraC belonging to the nonameric crystal structure (PDB entry code 3LIM) with the program PHASER (39). The initial model was refined with the pro-grams REFMAC5 (40) and COOT (41). The quality of the final model was assessed with the program COOT and PROCHECK (42). The resolution achieved was 1.29 Å. Data collection and refinement statistics are given in Table 1.
Isothermal Titration Calorimetry-The interaction between toxins and liposomes was measured with an ITC200 micro calorimeter (GE Healthcare) in buffer 50 mM Tris-HCl, 200 mM NaCl, pH 7.4, at 25°C. The LUV suspension was injected into a cell filled with protein under constant stirring. The heat released after each injection was determined from the raw data by integration of the individual exothermic peaks after subtraction of the baseline. The titration of WT with SM:DOPC (1:1) showed a few biphasic peaks, but only the first (major) phase was integrated. The binding isotherms were fitted to a model in which a molecule of protein binds to n molecules of lipid with the software ORIGIN 7 as described in Ref. 43. Each titration was repeated at least twice.

RESULTS
Hemolytic Activity-We predicted that residue Phe-16 would play a key role in the activity of FraC based on (i) its conservation in actinoporins, (ii) its key position within the N-terminal ␣-helical region before and after membrane insertion (28), and (iii) the structural dynamics of its side-chain ( Fig. 1). Phe-16 was mutated to proline, alanine, and glycine to produce broad alterations in the helical propensity of the N-terminal region. More specifically, proline and glycine have low propensities to form helices and are thus considered helix breakers. On the contrary, alanine has the highest propensity to form ␣-helices (44).

TABLE 1 Data collection and refinement statistics
The statistical values given in parentheses refer to the highest resolution bin.

Data collection
Space group P2 1   First, we examined the hemolytic activity of WT FraC and the three muteins prepared (Fig. 2). The destruction of RBC was monitored by the decrease of turbidity at 700 nm. As shown in Fig. 2A, FraC displayed a robust hemolytic activity, completing the destruction of RBC within 3 min. In contrast, the same concentration of muteins (F16P, F16A, or F16G) only induced a small change of turbidity that corresponded to less than 10% hemolysis. To obtain levels of hemolysis similar to that of WT protein (as estimated from the values of HC 50 ), the concentration of muteins needs to be increased by more than 2 orders of magnitude (HC 50 WT Ϸ 2.5 nM; HC 50 muteins Ϸ 800 nM) (Fig. 2B).
Notably, although Phe-16 was replaced with three different residues with varying helical propensities, the loss of activity was very similar among all the muteins. This observation suggests that the role of Phe-16 in hemolysis is independent of its propensity to form ␣-helices. Structure and Stability-To verify that Phe-16 mutation did not have an effect on the three-dimensional structure or the stability of the protein, we determined the high resolution crystal structure of F16P (presumably containing the most disrupting mutation) and examined its thermal stability by CD (Fig. 3). No significant changes with respect to WT protein were  (68); tenebrosin-C (Ten-C) from Actinia tenebrosa (69); magnificalysin III (HMg-III) from Heteractis magnifica (70); cytolysin RTX-A (RTX-A) from Radianthus macrodactylus (71); cytolysin Src-1 (Src-1) from Sagartia rosea (72); hemolytic toxin Avt-1 (Avt-1) from Actineria villosa (73); and toxin PsTX-20A (Pstx-20A) from Phyllodiscus semoni (74). The alignment was created with CLUSTALW2 and shaded with BoxShade. B and C, helical wheel diagram corresponding to the N-terminal ␣-helix of the water-soluble state (PDB entry code 3VWI) (B) and of the pore (PDB entry code 4TSY) (C). The length of the transmembrane helix increases with respect to that in the water-soluble state. Residues are colored as follows: hydrophobic (yellow), polar uncharged (purple), negatively charged (red), and positively charged (blue). Glycine and alanine are depicted in gray. Diagrams were prepared with HeliQuest (75). D, superposition of the crystal structures of water-soluble monomeric FraC (PDB entry code 3VWI, blue) and prepore oligomeric FraC (PDB entry code 3LIM, yellow). E and F, pore structure of FraC (PDB entry code 4TSY) (E) and close-up view of the transmembrane segment (F). The side chain of Phe-16 is depicted in magenta. observed. The crystal structures of the water-soluble monomeric WT (PDB entry code 3VWI) and F16P (this study) are virtually identical as determined by the overall RMSD (0.27 Å) and by the absence of conformational changes in the immediate vicinity of the mutated residue. Moreover, the thermal stability of F16P was comparable with that of WT protein as judged from their T M values (T M WT ϭ 59.4 Ϯ 0.2°C; T M F16P ϭ 58.3 Ϯ 0.2°C). Altogether, these results indicate that mutation of Phe-16 modifies neither the overall nor the local structure of the toxin in the water-soluble state, and thus we hypothesized that the loss of hemolytic activity must take place during the membrane interaction stage (binding, oligomerization, or insertion).
Behavior of FraC on Model Membranes Made of SM:DOPC (1:1)-To gain insight into the mechanistic differences between WT FraC and muteins, their pore forming ability was evaluated in model membranes composed of the binary mixture SM:DOPC (1:1), a lipid composition often used for the quantification of actinoporin activity (27,45) (Fig. 4). In agreement with previous reports, the WT protein displays a robust lytic activity (77% release of solutes) (Fig. 4A) (16,46). Remarkably, the activity of the muteins was essentially identical to that of WT FraC (percentages of release for F16P, F16A, and F16G were 75%, 79%, and 80%, respectively). This observation was unexpected, because the hemolytic activity of the muteins was considerably weaker than that of the native toxin. We selected F16P for an in-depth comparison of the lytic mechanisms of mutant and native FraC in model membranes.
First, we examined the ability of WT FraC and F16P to oligomerize, a necessary step for pore formation (46). WT and F16P were preincubated with SM:DOPC (1:1) vesicles, followed by solubilization of the proteoliposome mixture with detergents, repurification, and subsequently analysis by SEC. The elution profiles of WT and F16P were essentially identical (Fig.  4B). In both cases, the protein eluted as a large peak centered at 10.8 Ϯ 0.1 ml. We note that this volume is larger than V 0 (8.5 ml). The presence of protein in the elution peaks was verified by SDS-PAGE (Fig. 4B, inset). The elution profile observed during the purification of the octameric transmembrane pore used for crystallization is nearly identical to that described just above (28).
To examine the thermodynamics of pore formation, we titrated the toxins with vesicles composed of SM:DOPC (1:1) using the high resolution technique ITC (Fig. 4, C and D). The shape of the binding isotherm and the thermodynamic parameters of WT toxin and F16P were similar to each other ( Table 2). The interaction of WT and muteins with liposomes is exother- Effect of Phospholipid Acyl Chain Length and Bond Saturation on Pore Formation-To explain the differences in hemolytic activity between WT FraC and the F16P mutein, we screened lipid compositions in which the length and saturation of the acyl-chain moiety were varied. These two properties of phospholipids affect the activity of both, integral membrane proteins and proteins interacting with membranes (47,48). Because the commercial availability of multiple SM species is limited, only the PC component was systematically modified.
First, we evaluated the initial leakage velocity (v i ) (Fig. 5A) and the range of concentrations where the toxins are active (

Pore Formation of Actinoporins in Cholesterol-rich Membranes
among all the variants tested (Fig. 5A). However, the values of v i obtained for the muteins were at least 2-fold greater than that determined for WT FraC. This is surprising because it indicates that the muteins act faster than the native toxin (Table 3). This observation was corroborated in the concentration dependence experiment, where it is shown that all the muteins, and especially F16G, displayed greater lytic activity than WT FraC (Fig.  5B). We selected F16P as a representative mutein to evaluate the leakage of encapsulated solutes from vesicles made of other lipid compositions. The PC species examined differed from DOPC in their acyl chain length (20:1(⌬11-cis), 22:1(⌬13-cis), and 24:1(⌬15-cis), corresponding to lipids DEiPC, DErPC, and DNPC, respectively), the cis/trans configuration of the double bond (18:1(⌬9-trans) as in DElPC), and the relative position of the double bond (18:1(⌬6-cis), as in DPSPC). eggPC, which contains a mixture of PC species, but mainly DPPC (34%) and DOPC (32%), was also investigated. The extent of lysis induced by WT FraC or F16P was similar in all the compositions tested. Interestingly, the lysis induced by the mutein occurred faster than that by WT FraC in liposomes composed of SM:DElPC (1:1), SM:eggPC (1:1), and SM:DNPC (1:1) as judged from the values of v i (Fig. 5C and Table 3). We also observed that v i decreased as the length of the acyl-chain of PC increased (Table  3). Altogether, the results obtained so far with model membranes made of PC and SM have failed to explain the different hemolytic activities shown by WT FraC and the muteins at position Phe-16.
Behavior of FraC on Model Membranes Containing Cholesterol-An obvious difference between the lipid composition of RBC membranes and the model membranes employed so far is the presence of cholesterol in RBC. This lipid favors the appearance of domains consisting of L o phases at the expense of other lipid phases. In particular, the relative amount of L d and solid phases is greatly reduced in comparison with those in vesicles of SM:PC (1:1) (49,50). We first examined the ability of the toxins to permeabilize model membranes containing a high percentage of cholesterol (SM:eggPC:Chol (50:10:40)). The concentration of cholesterol in these vesicles is comparable with that of sheep RBC membranes as determined by us from the purified lipids (44% cholesterol; data not shown) and consistent with previous data (51).
When cholesterol is incorporated into the vesicles, large differences between the lytic activity of WT FraC and that of the muteins were observed (Fig. 6A), mirroring the results obtained in the hemolytic assay shown in Fig. 2. Whereas the percentage of release of encapsulated solutes in cholesterol-containing vesicles treated with WT FraC was similar to that observed in vesicles without cholesterol (ϳ80%), the muteins F16A, F16G, and F16P were virtually inactive, giving rise to 16%, 15%, and 2% release, respectively. The F16P mutein displayed the lowest activity and therefore was selected for further analysis.

TABLE 3 Initial leakage velocities (v i ) of WT FraC and muteins
Velocities are expressed as percentages of leakage per second (30). First, we examined the lipid-mediated oligomerization of the toxins by SEC as described above. Interestingly, the elution profiles of WT FraC and F16P treated with SM:eggPC:Chol (50:10: 40) vesicles were essentially identical (Fig. 6B). Both proteins displayed a large peak of protein at 10.8 Ϯ 0.1 ml. This peak is preceded by a smaller peak (V E ϭ 8.4 Ϯ 0.1 ml), which corresponds to the void volume and, most likely, represents insoluble proteoliposomes (52). These experiments indicate that mutation of Phe-16 does not alter the oligomerization state of the membrane-bound protein in cholesterol-rich vesicles, even considering that the toxin is inactive.

Lipid composition Acyl chain of PC
Next, we determined the thermodynamic parameters of the toxin-lipid interaction by titrating the toxins with vesicles made of SM:eggPC:Chol (50:10:40) (Fig. 6). Significant differences were observed between WT and muteins. The enthalpy change was approximately twice as large in WT FraC than in F16P (⌬H°W T ϭ Ϫ17.9 Ϯ 0.1 kcal mol Ϫ1 ; ⌬H°F 16P ϭ Ϫ11.1 Ϯ 1 kcal mol Ϫ1 ), thus giving rise to different entropy compensation patterns (ϪT⌬S°W T ϭ 7.9 Ϯ 0.4 kcal mol Ϫ1 ; ϪT⌬S°F 16P ϭ Ϫ0.5 Ϯ 1.3 kcal mol Ϫ1 ). Interestingly, the affinity constant of F16P for the cholesterol-containing vesicles was ϳ20-fold greater than that of WT FraC (K D WT ϭ 49 Ϯ 33 nM; K D F16P ϭ 2.8 Ϯ 2 nM). These results suggest uncoupling of the binding and oligomerization steps with respect to the insertion of the N-terminal helix in the hydrophobic core of the membrane.
Influence of Cholesterol and Other Sterols-To examine in detail the contribution of cholesterol to the activity of FraC, we used liposomes made of SM:eggPC:Chol containing a fixed amount of SM (50%) and increasing concentrations of choles-terol. Vesicles with cholesterol contents ranging from 0 to 40% were equally susceptible to the action of WT toxin (Fig. 7A). However, the lytic activity of the muteins was comparable with that of FraC only when the cholesterol concentration was 25% or less. When the cholesterol concentration exceeded this threshold value, their lytic activity of the muteins decreased abruptly. At the highest concentration tested (40% cholesterol), the extent of leakage was practically negligible for all of them (0.6, 17, and 16% for F16P, F16A, and F16G, respectively). Interestingly, the 25% threshold coincides with the disappearance of L d phases in vesicles consisting of ternary mixtures of SM, PC, and cholesterol (49,50). These results indicate that, in contrast to the native toxin, the muteins are unable to operate in conditions where cholesterol-rich L o domains are predominant.
We also examined the release of encapsulated solutes from liposomes made of SM and cholesterol, where L d domains are not present (49). In these vesicles, the activity of WT FraC increased from 3 to 76% as the percentage of cholesterol within the membrane increased from 20 to 50% (Fig. 7B and inset  therein). In contrast, the lytic activity of the muteins was negligible (under 20% leakage in all cases). The activity of the toxins was also investigated in vesicles made of binary membranes of eggPC and cholesterol. As shown in Fig. 7C, these vesicles were virtually insensitive to all variants of FraC tested, including the WT protein. This behavior is also observed for the actinoporin equinatoxin II and can be explained by the absence of SM from the vesicles (25). Overall, these results indicate that above a definite threshold of cholesterol concentration, the presence of  (31). Increasing concentrations of ergosterol inhibited the activity of the muteins in a fashion similar to that of cholesterol but not that of WT (Fig.  7D). In contrast, the lytic activity of the muteins was similar to that of WT when the sterol employed was cholestenone (Fig.  7E). Overall, these observations strengthen the idea that Phe-16 is a critical element for the lytic activity of the protein in L o phases induced by sterols such as cholesterol.
To verify whether the removal of cholesterol from RBCs specifically sensitizes the cells, we employed MCD (see "Materials and Methods"). This molecule specifically extracts cholesterol from cells (34). The hemolytic activity of WT and F16P were examined with RBC treated with or without MCD (Fig. 7F). Whereas the hemolytic profile of RBC in the presence of WT is not affected by the treatment with MCD (HC 50 values were 0.8 nM in both cases), the hemolytic potency of F16P increases 4-fold from ϳ2 M in untreated RBC to 0.5 M in cells treated with MCD. The hemolytic activity of F16P is still significantly lower than that of WT toxin, but it is necessary to take into account that a significant amount of cholesterol is still present in the membranes after the treatment with MCD (ϳ50% with respect to the untreated samples), and the complexity of RBC membranes is higher than that of model membranes (53). In summary, reducing the amount of cholesterol in the membrane of RBC sensitizes the cells for the hemolytic action of F16P, consistent with the results obtained in model membranes.

DISCUSSION
The physicochemical properties of bilayers, such as lipid packing and membrane lipid order, are governed by the chemical identities of the constituent lipids (i.e. the nature of their head groups and their acyl chains), and the interactions among them (54). For instance, some lipids facilitate their segregation into two-dimensional platforms known as lipid domains, which can exist in the L d , L o , or gel phases (55). These phases and lipid domains play a key role in the regulation of the structure and function of membrane proteins (55)(56)(57). Notably, the so-called lipid rafts (L o phases rich in cholesterol and sphingolipids) have been shown to modulate the activity of membrane-active protein toxins (12,58).
For instance, in the family of cholesterol-dependent cytolysins, this lipid promotes the insertion of segments of the protein in the membrane, followed by the formation of transmembrane ␤-barrel pores (34). Cholesterol is also involved in the mechanism of pore formation of ␣-hemolysin from Staphylococcus aureus (59), ClyA hemolysin from E. coli (60), and cytolysin from V. cholerae (61).
Vesicles containing raft-like domains rich in cholesterol are also susceptible to actinoporin activity (29,31,32,62). These domains are particularly abundant in lipid mixtures containing SM and cholesterol. They are characterized by a high packing density of lipids and a relatively high lateral motion (63). Notably, cholesterol-depleted vesicles containing two types of lipid domains are also susceptible to the lytic action of actinoporins (16,29,31).
We have discovered a single mutation in the actinoporin FraC sensitive to the cholesterol content of the membrane. Specifically, cholesterol concentrations above 25% inactivate muteins of FraC lacking a phenylalanine residue at the N-terminal region (position 16), regardless of the identity of the substituting residue (Ala, Gly, or Pro) or its propensity to form ␣-helices (Fig. 7, A and B). Interestingly, 25% corresponds to the concentration of cholesterol at which L d domains are replaced by L o domains (49,50).
The mutation of Phe-16 does not cause structural defects in the protein (Fig. 3, A and B), nor does it reduce its thermal stability (Fig. 3C). In addition, our data indicate that Phe-16 is not crucial for the initial binding of FraC to the membrane or the oligomerization of the toxin (Fig. 6B). Instead, Phe-16 is essential for the formation of active pores specifically in cholesterol-rich membranes. In the mutated toxins, the absence of Phe-16 reduces the activity in vesicles composed predominantly by L o domains. However, the mutation does not affect pore formation of FraC in other lipid mixtures with low cholesterol content. Indeed the mutated versions of the toxin act faster than WT protein in some of the lipid compositions examined (Table 3).
Why is Phe-16 such a critical residue in membranes with high cholesterol content? After oligomerization on the surface of the membrane, the N-terminal ␣-helix of FraC must penetrate the membrane before producing the lytic pores. From this point on, two complementary mechanisms are contemplated based on the cholesterol content of the membrane (Fig. 8). On the one hand, in susceptible membranes displaying L d and solid domains (cholesterol-depleted membranes), the insertion of the N-terminal region is achieved at domain-domain interfaces (16,25,29). Presumably, the N-terminal ␣-helix finds no major obstacles during its insertion across the bilayer. On the other hand, in susceptible membranes with high lateral packing (L o sterol-rich membranes), the penetration of the ␣-helix is less feasible. Under these conditions, Phe-16 is necessary to increase the hydrophobicity of the helix and to optimize its insertion in the membrane. In addition, helix-helix interactions and possibly lipid-protein interactions mediated by Phe-16 are also important factors as evidenced by the larger enthalpy change observed in WT with respect to the mutein F16P. The thermodynamic parameters shown in Table 2 support this hypothesis, because the enthalpy for the interaction with membranes with high cholesterol content is more favorable for WT FraC than for the F16P mutant. The larger entropy change observed in WT would point to a more structured transmembrane helix, a step needed for pore formation. A correct helical folding across the L o bilayer could be assisted by the favorable interaction of Phe-16 with the ordered lipids, as well as with the adjacent helices to stabilize the membrane-inserted complex. Leucine, found in the same position in a few actinoporins (16), might play the same role as Phe-16. Overall, our study is one of the first examples in which a single residue is shown to be necessary for a PFT to adapt to different lipid compositions and physicochemical environments.
In summary, we have shown that residue Phe-16 of FraC facilitates its pore forming activity in membranes with high cholesterol content. Hence, Phe-16 is a critical residue in cholesterol-rich membranes, and its conservation in actinoporins expands the spectrum of potential target membranes (and preys) of sea anemones. Moreover, we revealed single-residue mutants of FraC potentially useful as biosensors in diseases affecting the concentration and/or distribution of cholesterol in cell membranes (64).