Disrupting a key hydrophobic pair in the oligomerization interface of the actinoporins impairs their pore‐forming activity

Crystallographic data of the dimeric and octameric forms of fragaceatoxin C (FraC) suggested the key role of a small hydrophobic protein–protein interaction surface for actinoporins oligomerization and pore formation in membranes. However, site‐directed mutagenesis studies supporting this hypothesis for others actinoporins are still lacking. Here, we demonstrate that disrupting the key hydrophobic interaction between V60 and F163 (FraC numbering scheme) in the oligomerization interface of FraC, equinatoxin II (EqtII), and sticholysin II (StII) impairs the pore formation activity of these proteins. Our results allow for the extension of the importance of FraC protein–protein interactions in the stabilization of the oligomeric intermediates of StII and EqtII pointing out that all of these proteins follow a similar pathway of membrane disruption. These findings support the hybrid pore proposal as the universal model of actinoporins pore formation. Moreover, we reinforce the relevance of dimer formation, which appears to be a functional intermediate in the assembly pathway of some different pore‐forming proteins.


Introduction
The organization into supramolecular complexes or oligomers that pierce the membrane is a common feature of pore forming toxins (PFTs). These proteins undergo essential conformational changes from a water-soluble state to a transmembrane oligomeric state and form cytolytic pores presumably involving the evolution of oligomeric intermediates. [1][2][3][4] However, the molecular details in the pathway of selfassembly and the role of relevant intermediates remain unclear for most PFTs. Actinoporins are among the most studied eukaryotic PFTs. [4][5][6] To date, the most studied actinoporins have been equinatoxin II (EqtII) from Actinia equina, sticholysins I and II (StI and II) from Stichodactyla helianthus, and fragaceatoxin C (FraC) from Actinia fragacea. 4,7 The 3D structure of the soluble forms of these actinoporins contain a hydrophobic b-sandwich core flanked by two a-helices, including one located in the Nterminus, which has been identified as the poreforming domain [ Fig. 1(A)].
Understanding of the assembly mechanism of actinoporins has dramatically increased after determination of the crystal structure of FraC at different stages of the lytic process, 8 complementing a body of biochemical and biophysical data previously obtained by different groups. [9][10][11][12][13][14][15][16] It is now generally accepted that actinoporins bind to membranes mainly as monomers 10,[17][18][19][20] and then undergo a conformational change that involves only the N-terminal segment. 4,8,14 Then, monomeric units oligomerize to form pores in which the N-terminal a-helix lines the walls in conjunction with lipids. 8 The average radius of actinoporin pores has been determined by different approaches and have a fixed size ($ 2 nm of diameter). 8,10,21,22 However, oligomeric structures seem to quickly switch among multiple functional stages 4 and are transient and electrically heterogeneous. 4,23 There are mainly two hypothetical models under debate which describe the architecture of the pore formed by these toxins: the classical toroidal pore [11][12][13]24 and the hybrid protein-lipid pore. 8,25 These models differ in the stoichiometry of the pore and in the relevance of the protein-protein interactions for pore stability. The toroidal pore model was initially proposed and is mainly based on functional studies with sticholysins and EqtII. 11,13 Kinetic and electrophysiological studies and more recently single molecule imaging have revealed that predominantly 3-4 monomers of these proteins form functional pores. 10,16,21,24,26 Moreover, crystalografic data of a prepore like structure of StII in lipid monolayers shows that the units that form the assembly do not establish relevant protein-protein interactions. 24 In contrast, the crystal structure of FraC shows an octameric pore where each protein molecule is associated with three molecules of lipids in a hybrid protein-lipid architecture. 8 In this architecture, both protein-lipid and protein-protein interactions contribute to the pore stability. 8 The relevance of dimers as an assembly intermediate has been highlighted by different studies. 8,16,27 In general, data supporting either the toroidal or the hybrid protein-pore model are in agreement with the relevance of dimers in the mechanism of actinoporins cell death induction. 4 It is important to note that the structure of dimeric FraC exhibits a similar protein-protein interface to that of the octameric pore, which reinforces its mechanistic relevance [ Fig. 1(B-E)]. 8 It is also significant that both the dimeric and the octameric interfaces proposed from structural data are highly conserved in other members of the actinoporin family. 8 However, there is no functional data validating the conserved role of the dimeric interface and proteinprotein interactions in the mechanism of assembly of actinoporins.
We previously published that the structure and activity of FraC and EqtII are highly similar, indeed their N-terminus segment, which is responsible for pore formation is identical. 28 Thus, we would not expect that these proteins build different pore architectures as reported. 8,13,24 Because this is a controversial topic in the literature, we decided to compare the structural and functional properties of three of the most studied actinoporins (FraC, EqtII, and StII) in more detail. Here, we combined mutagenesis studies with molecular dynamic simulations to test the hypothesis that the dimeric interface of actinoporins is conserved and plays a key role in pore formation. First, a double mutant that disrupts the putative dimerization interface of actinoporins was designed by combining sequence and structure analysis with free energy calculations. We cloned, expressed, and purified the double mutants FraC V60D/F163D , EqtII V60D/F163D , and StII I58D/I161D , and characterized their structure by spectroscopic techniques and their function by biophysical tools. Although the mutants maintained their main structural features and the capacity to bind lipidic membranes of the wildtype, their pore-forming ability was abolished. Our work reinforces the relevance of dimer formation as a necessary intermediate in the mechanism of action of actinoporins. We proposed that actinoporins follow a common pathway of assembly where protein-protein interactions stabilize the oligomeric structure in the membrane.

Results
The impairment of the interaction between V60 and F163 reduces the FraC dimer stability The dimeric and the octameric structures of FraC exhibit similar protein-protein interfaces. 8 This indicates that the dimeric structure is a functionaly relevant intermediate in the actinoporins self-assembly mechanism 8 [ Fig. 1(D,E)]. Therefore, the dimer is the simplest relevant oligomeric structure for free energy calculations to identify warm/hot-spots at the protein-protein oligomerization interface. Computational Alanine Scanning (CAS) calculations indicated that the residues D58 (DDG efe 5 7.55 6 0.15 kcal/ mol), W149 (DDG efe 5 10.30 6 0.05 kcal/mol) and F163 (DDG efe 5 6.66 6 0.03 kcal/mol) were the major contributors to the binding free energy of the dimer intermediate ( Fig. 2; Supporting Information Table SI).
The pwEFED protocol was employed to gain insight into the energetic contribution of those hot spots in terms of pair-wise interactions with neighboring residues. Interactions of V60-F163, V60-W149, D58-N165, and D58-S167 between each protomer significantly contribute to the formation of the dimeric structure [ Fig. 3(A,B)]. Among these residues, the pair of interacting residues V60 and F163 showed the largest and most stable energy contribution to dimer formation (DG efeV60,F163 5 23.14 6 0.40kcal/mol). Remarkably, V60 was able to establish interaction with F163 that was three times stronger, in terms of the binding free energies, than W149 (DG efeV60,F163 5 21.03 6 0.24 kcal/mol) [ Fig. 3(A); Supporting Information Table SII]. van der Waals interactions and nonpolar solvation energies were identified as the major contributors to the formation of this pair (V60-F163) [ Fig. 3(A), Supporting Information Table SII]. Hydrophobic residues such as V60 and F163 are highly conserved among the actinoporins family. 8 We designed the double mutant FraC V60D/F163D , in order to evaluate the detrimental impact of disrupting the interaction between V60 and F163 in the dimer assembly and pore forming ability of FraC.
We calculated the RMSD values for the subset of heavy atoms belonging to the interface residues of FraC and the double mutant FraC V60D/F163D with respect to the initial structure over 100 ns of simulation [ Fig. 4(A)]. The double mutant FraC V60E/W149A , which is known to have 16 percent of the activity compared to the wildtype protein, 8 was also simulated for comparison [ Fig. 4(A)]. To further investigate the stability of the interaction of this pair of residues, the distance between them was calculated as a function of simulation time [ Fig. 4 Table I. Taking the wildtype protein as a reference, the energy of dimer formation in the double mutant FraC V60D/ F163D (DDG efe 5 6.37 kcal/mol) was 4.8-fold lower than in the double mutant FraC V60E/W149A (DDG efe 5 1.33 kcal/mol) ( Table I). The van der Waals interaction energies (DDDE vdw 5 4.78 kcal/mol) were the major energetic contributors to the stability differences between the dimeric form of both mutants (DDDG efe 5 5.04 kcal/mol) ( Table I). The less drastic conformational changes at the dimer interface observed with the FraC V60E/W149A , in which only one negatively charged amino acid residue was introduced [ Fig. 4(A,B,E)] might be a consequence of the lower electrostatic repulsion, and may also be due to an additional stabilization mediated by hydrogen bonds created through two water molecules between E60 and the E173 [ Fig. 4(F); Table I]. This finding explains why this mutant, although less active, still retains 16% of the wildtype pore-forming ability. 8 Taking into account all of these in silico results, we designed doubleresidues mutants in equivalent position of the three most studied actinoporins, FraC, EqtII, and StII, to verify its impact on the dimerization step within the mechanism of pore formation. (C-E) Distance changes of the wildtype and mutated residues during the MD simulation (initial distance-cyan) and (final distance-green). The distance and RMSD values are cumulative averages considering sliding windows of 50 frames. (F) Two water molecules between E173 (yellow) and V60 (red) indicate hydrogen bond formation between those residues in FraC V60E/W149A . Colors (red and yellow) represent different protomers of the dimeric form.

(B)]. A stable
V60D and F163D mutations do not affect the folding of the wildtype proteins in solution Next, we designed homologous mutants to FraC V60D/ F163D in the related proteins EqtII and StII. The solution structure of the mutants (FraC V60D/F163D , EqtII V60D/F163D , and StII I58D/I161D ) were compared with those of the wildtype variants (FraC, EqtII, and StII). UV-CD spectra of the toxins and the mutants were recorded. Typical far UV-CD spectra of wildtype FraC and FraC V60D/F163D are shown in Figure 5. The spectra of the wildtype protein and the mutant was highly similar, exhibiting a positive band centered at 195 nm and a minimum around 217 nm, typical of proteins containing mainly bsheet structure. 29 Deconvolution of these spectra according to the CONTIN 30,31 and SELCON3 32,33 algorithms located at the DICHROWEB Internet server [34][35][36] provided the contents of secondary structure given in Table II. The structure of the wild-type proteins and their mutants consist predominantly of b-sheet structures, indicating that mutation of the selected residues by Asp did not significantly modify their secondary structure in solution. Moreover, we did not observe differences in folding among the three actinoporins (Table II). The results are in agreement with data derived from far-UV-CD studies of actinoporins 27,37,38 and also with their threedimensional structures. 8,22,24,39,40 Tryptophan emission spectra of wildtype proteins and mutants were also determined. The mutations did not modify the average degree of Trp exposure to solvent since the fluorescence maximum wavelength were very similar to those of the wildtype proteins (Table II). The average location of the Trp residues was also estimated from fluorescence quenching experiments employing AA as a soluble quencher. Figure 5(B) shows the fluorescence spectra of FraC in the presence of increasing amounts of AA. Stern-Volmer plots were linear, indicating that the emitting residues can be treated as an   (Table II) and significantly different from the protein completely denaturalized by heat (data not shown).

Double mutants retain binding ability to monolayers and liposomes
To obtain insight into the interaction of wildtype proteins and their mutants with membranes, we studied their binding to lipid monolayers and bilayers of PC:SM (50:50). This has been the most used lipid composition to study the interaction of actinoporins with membrane-mimetic systems. 10,17,37,48 The increase in surface pressure (Dp) elicited by the association of a toxin to a previously formed lipid monolayer can be employed to characterize its ability to interact with organized lipids. 41 The Dp was evaluated at several initial pressures (p 0 ) of the lipid monolayer keeping the monolayer area constant. The injection of the toxin into the subphase of the lipidic monolayer triggered a Dp which stabilized after $400 s [shown in Fig. 6(A) for FraC]. The Dp at equilibrium caused by FraC and FraC V60D/F163D as a function of the p 0 are shown in Figure 6(B). As expected, Dp decreased with increasing p 0 for all proteins because tighter lipid packing prevented insertion.
Two suitable parameters for the characterization of the protein-lipid interaction are the critical pressure (p c ) and the slope of the linear fitting (m) obtained from plotting Dp versus p 0 (Table II). In particular, p c is obtained by extrapolating to Dp 5 0 [ Fig. 5(B)] and is directly related to the protein affinity for the lipids in the monolayer. 42 Each toxin and its respective mutant showed curves with similar values of slope (m). Similar values of m indicate that the effect of the mutation on binding is independent of the actual surface pressure value and consequently of the lipid concentration.
The p c for the wildtype proteins were similar, in the range $33-37 mN m 21 . Those values were slightly lower for the mutants ($30-32 mN m 21 ) but are still in the range where they can theoretically 30,42-44 insert into a biological membrane (Table  II). Since m was similar between wildtypes and mutants, the higher changes in surface pressure obtained with the wildtype do not seem to arise from differences in lipid affinity, but rather from differences in protein-lipid packing (i.e., oligomerization) at the interface. Our results are in agreement with previous observations with the mutants FraC 45 and StI 27,46 in which variations in the oligomeric state affected the p c values, but the slope of the curve remained almost identical. At this point, it is tempting to speculate that the monolayer system is able to detect variations in the oligomeric state of the proteins as a consequence of the variation of the overall packing of the lipid and protein molecules within the liquid-air interface. Lipid vesicles are good models to mimic the behavior of natural membranes. These aggregates allow for the evaluation of protein association to lipids and the protein function by measuring the release of entrapped solutes. 47 To obtain further insight into the insertion mechanism of the proteins into lipid membranes, binding to SUV was assessed by measuring the effect of lipid concentration on Trp intrinsic fluorescence [ Fig. 6(C)]. Addition of liposomes progressively increased the fluorescence intensity and induced a blue-shift of the spectra until a plateau was reached, indicating quantitative association of the proteins [ Fig. 6(D)]. The curves in Figure 6(D) show a similar behavior for the wildtype FraC and its respective mutant. The magnitude of the change in the fluorescence intensity due to the protein binding to lipids (F/F 0 ) can be extrapolated to high lipid concentrations to determine F/F 0 max. Moreover, the lipid concentration is related to the strength of the protein/lipid interaction and this can be characterized by the Lip 50 value (Table II). These values were similar when compared to the wildtype FraC, EqtII, and StII, as well as the mutants.
Stern-Volmer constant (Ksv) values were obtained by employing AA as fluorescence quencher (Table II). This parameter was derived from the slopes of Stern-Volmer type plots, like those shown in Fig. 6(E). Fluorescence attenuation with AA were carried out in the presence of saturating lipid concentrations in order to ensure that all the protein's molecules were bound to liposomes. For all the proteins, Ksv values in the presence of liposomes were lower than those obtained in solution. This decrease indicated protein association to the lipids since AA accessibility to the Trp is partially blocked when the protein was bound to the membrane. 37 Moreover, Ksv values from wild-type proteins were similar to the mutant variants suggesting that the mutations did not affect the step of binding to the membrane (Table II).

Double mutants do not show pore-forming ability in liposomes and cells
The pore-forming capacity of the toxins was assessed by evaluating their ability to promote the release of the fluorescent dye CF encapsulated in liposomes (Fig. 7). It is well known that actinoporins promote the release of fluorescent solutes entrapped in liposomes. 5,12,30 The fraction of vesicles bearing at least one pore (f) versus time was calculated assuming that the release of CF is simultaneous with pore formation [ Fig. 7(A)]. The dose-dependent permeabilization of liposomes elicited by the toxins exhibited a sigmoidal shape [ Fig. 7(B)]. This feature indicates that vesicles were not efficiently permeabilized until a critical number of monomers per vesicle was reached. 10 The C 50 parameter was estimated by fitting the data to a Hill sigmoid (Table II). C 50 characterizes the protein concentration necessary to promote the release of 50% of the dye entrapped into the vesicles. As we previously reported, the three actinoporins differs in activity, with StII as the most active one, followed by EqtII and FraC, which did not differ. 30 In contrast to the wildtype proteins, the mutants did not show permeabilizing activity in the concentration range assessed [ Fig.  7(B); Table II].
The hemolytic activity of the proteins toward human red blood cells was also determined to quantitatively measure the pore-forming activity of the three actinoporins in a cell model. 21,37 Figure 7(C) shows the time course of FraC-elicited hemolysis at different concentrations. The hemolytic activity of the wildtype toxins increased with a clear sigmoidal relationship to toxin concentration [ Fig. 7(D)]. The HC 50 was determined by fitting the experimental data to a Hill sigmoid model (Table II). This parameter represents the protein concentration needed to promote lysis of 50% of the cells in the assay and is inversely correlated to toxin activity. Similar to what was obtained with vesicles, wildtype proteins differ in activity and the mutants were not active against cells in the concentration range evaluated.

Discussion
The exact sequence of events that take place during the assembly of actinoporins and functionally relevant intermediates in the membrane remains under debate. Some studies suggest the relevance of protein-protein interactions 8,22,25 while others assume that there is no need for such protein interfaces to stabilize oligomeric intermediates for the final pore assembly. 24,48 However, evidence supports the structural and functional importance of dimeric structures in the transition of the monomer structure from solution to the pore state. 16,21,27,49 The key role of dimers in actinoporins poreformation has been investigated from indirect or direct observations over the last decades. 16,21,27,49 Recently, a high resolution crystal structure of a dimeric form of FraC was obtained in a lipidic environment. 8 In this dimeric structure the putative transmembrane a-helices were found to be still attached to the protein core, but the oligomerization interface resembled the one of the octameric pore [ Fig. 1(D,E)]. This finding suggested dimer formation as an important structural step in the initial stages of the assembly of actinoporins channel. 8 In this study, we have revealed the driving forces for the spontaneous formation of FraC dimeric structure using free energy calculations with the MM-GBSA approach. 50 In particular, V60-F163 is predicted as the most relevant pair of interacting residues in the dimeric structure [ Fig. 3(A); Supporting Information Table SII]. van der Waals interactions and nonpolar desolvation are the most relevant energetic components of this interacting pair, which suggests a key role of the geometric complementarity between these residues in the stability of the dimer [ Fig. 3(A); Supporting Information Table SII]. FraC oligomers have a small protein-protein oligomerization surface and are held together by shape complementarity and hydrophobic interactions, 8 two features that are easily compromised and result in stability reduction. 45 These features are in marked contrast with other PFTs that form pores exclusively built by proteins segments (i.e., ClyA) stabilized by electrostatic interactions. [51][52][53][54] We predicted that the introduction of anionic amino acid residues (D60 and D163) in the dimeric interface of FraC allows its partial disruption as a consequence of electrostatic repulsion [ Fig. 4(A,B,D); Table I]. The reorganization observed generated an incompatible orientation of one protomer to detach its N-terminal segment once the dimer binds to the membrane. As V60 and F163 are residues of FraC conserved among the actinoporins, 8 we evaluated the effect of introducing homologous mutations in EqtII and StII in order to verify our hypothesis in others proteins of the family. These proteins differ in pore-forming activity both in cells and liposomes [ Fig. 7(B,D; Table II]. 30 However, these differences have been correlated with variations in their Nterminal segment, which accounts for mechanistic differences in the membrane insertion step and does not seem to be involved in oligomerization or binding steps. 30 Introducing two anionic charged amino acid residues at positions 60 and 163 dramatically abolished the activity of these three actinoporins [ Fig. 7(B,D; Table II]. The complete activity loss of the double mutants is not due to the impairment of protein folding in solution (Fig. 5) and seems not to be related with differences in the affinity of the mutants for the lipid bilayer [ Figs. 5(A,B) and 6(C-E)].
Structural data supporting the toroidal pore model assumes that there are no relevant proteinprotein interactions and that proteins and lipid alternate in the final pore structure. 24,48 In contrast, the recently proposed hybrid pore model argues against this property and assumes that the pore assemble is held together through relevant proteinprotein and protein-lipid interactions. 8 Here, we extend the importance of FraC protein-protein interactions in the stabilization of the oligomeric intermediates to StII and EqtII pointing out that all of these proteins follow a similar pathway of membrane disruption. We found that the interface leading to the octameric assembly seen in the FraC crystal structure 8 is critical for actinoporins pore formation. However, from our data we cannot rule out the existence of pore-forming states with different stoichoimetries. 26 Future work aimed to correlate oligomer stoichiometry with pore function should shed light about the relevance of intermediate stages, including dimeric structures, for actinoporins pore formation. Moreover, our results reinforce a model of pore assembly based on the addition of dimeric units recently proposed for the actinoporin EqtII. 16 In fact, dimer formation has been assumed as a more efficient mechanism of oligomer assembly compared with simple sequential monomer addition. 55 The dimer interface obtained by Tanaka et al. 8 is been described as the basis for the octameric pore observed in the FraC structure and, more recently, for an octameric prepore structure of this toxin. 25 However, from our results and previous published data, 16 we cannot rule out the possibility that this dimeric interface can nucleate up to 8 units by the sequential addition of single or dimeric units to complete a final ring-like assembly. EqtII was described to be organized in the cell surface as a mixture of oligomeric species mostly including those formed by even number of units. 16 However, the mutant EqtII-L26C is able to oligomerize by an alternative pathway based on the addition of monomeric units even though this alteration decreased its toxic activity. This finding, obtained in conditions closer to those existing in physiological systems, highlights the relevance of dimers for EqtII assembly and toxicity. However, the fact that the EqtII mutant was able to oligomerize through monomers addition indicates that, although more efficient, dimers nucleation could be dispensable for complex formation. 16 Moreover, it would be possible that different intermediate oligomeric states could form arc-like toroidal pores with one edge filled by protein complexes and the other with bend lipids, 55 similar to those observed in the MACPF/CDC protein superfamily 56,57 or for the Bcl2 protein Bax. 58 Further studies should confirm this hypothesis since current high quality data obtained by electron microscopy or atomic force microscopy at high protein concentration do not support the formation of arc-like structures in the mechanism of action of actinoporins. 8,22,25 One of the main differences between the toroidal and hybrid models is the stoichiometry of the final oligomeric state of the pore. While the toroidal model assumes that the channel is formed by 3-4 monomers of proteins, the hybrid pore claims for an octameric molecularity. In this regard, Gilbert 59 kindly advocated a second inspection of StII tetramers pointing out that such structures could be masking an octamer, similar to the one obtained for FraC. 8 Then, each protein unit identified in the StII toroidal pore model would correspond to dimeric structures rather than monomers. From this perspective, a consensus point between the toroidal and the hybrid pore models might be the contribution of a priming dimeric structure to trigger pore opening, 83 which is in agreement with our findings.
Based on recent evidence and our functional data, we can generalize the actinoporin model of pore formation by proposing that (i) dimeric units are the simplest structural intermediate with functional relevance which drive the assembly of bigger oligomeric species, (ii) the oligomerization interface is conserved across the family, and stabilized by shape complementarity and hydrophobic interactions, and consequently, (iii) the introduction of two anionic residues on this interface (V60D/F163D) dramatically abolishes the pore-forming activity. This last result clearly indicates the importance of the key hydrophobic pair V60-F163 in stabilizing the structure of the actinoporins pore. [11][12][13] Conclusions In this work, we identified two mutations in FraC (Val60D and F163D) which completely hinder dimer formation and dramatically abolish the pore-forming activity. Of mechanistic significance, we extend the importance of the FraC oligomerization interface to other members of the actinoporins family and reveal that they follow a common pathway of assembly and membrane disruption. It is highly likely that the complete activity loss of the mutants arose from their inability to form the lowest functionally relevant oligomer, the dimer, being the monomer the only molecular entity found in the membrane. Future work must be addressed to unravel this question.

Chemicals and reagents
Egg SM and 1-palmitoyl-2-oleylphosphatidylcholine (POPC) were purchased from Avanti Polar Lipids (Alabaster, AL, USA) with 99% of purity and used without further purification. Solvents and chemicals were of the highest commercial purity available.

Molecular dynamics simulations
All Energy Minimizations (EM) and Molecular Dynamics (MD) simulations were carried out using the GROMACS software package (version 4.6.5) 60 with the AMBER99sb force field, 61 and the TIP3P water model. 62 The simulation systems consist of the dimer (PDB: 4TSL) solvated in a truncated dodecahedrical box with 13640 water molecules. The protonation states of ionizable residues were assigned at pH 7.4 with the program PDB2PQR (http://nbcr-222.ucsd.edu/pdb2pqr_1.8/), which uses PROPKA for the prediction of pK a values. Cl 2 ions were added to achieve electroneutrality of the system. At each step, the electrostatic interactions were calculated with the particle-mesh Ewald method. 63 Van der Waals interactions were described by a Lennard-Jones potential with a cutoff of 1.1 nm that was switched to zero between 1.0 and 1.1 nm. Dispersion corrections for energy and pressure were applied. The SETTLE algorithm 64 was used to constrain bonds and angles of water molecules and LINCS 65 was used for all other bonds, allowing a time step of 2 fs. The equilibrium simulations were sampled for 100 ns, using the stochastic dynamic integrator 66 at 310 K and a constant pressure of 1 atm, using the weak coupling to the velocity rescaling thermostat 67 and Parrinello-Rahman barostat, 68,69 respectively. The friction coefficient (n) was set to 0.5 ps 21 in all systems as recommended. 70 Snapshots were saved at 10 ps intervals.

MM-GBSA free energy calculations
T s (i.e., the molecular energy in the gas phase (DE gas ), the solvation free energy (DG solv ), and the entropy contribution (2TÁDS)). When the third term was neglected, the computed value is an effective free energy (DG eff ), which usually suffices for comparing the relative affinities of a series of similar ligands for a given receptor because in that case the entropy term is approximately constant. 71 The term DE gas , which includes the internal (DE int ), the van der Waals (DE vw ), and the electrostatic (DE ele ) energies, was derived from the force-field equations. The term DG solv was further decomposed into two components (i.e., the polar-solvation free energy (DG polar ) and the nonpolar solvation free energy (DG nonpolar ). 71 The former was calculated using the Generalized Born-Neck2 implicit-solvation model (GBn2, igb 5 8), 72 whereas the latter was obtained by: in which DG nonpolar (X) represents the solventaccessible surface area changeof the solute molecule X upon complex formation, while c and b are empirical constants whose values for GB models are almost always 0.0072 kcal Å 22 mol 21 and 0, respectively. 71,73 TÁDS is frequently computed by normal-mode analysis and is, therefore, the most computationallydemanding step of the MM-GBSA method. 71,73 The MMPBSA.py program of the Amber12 package was used for MM-GBSA free energy calculations 71,73 after converting the GROMACS trajectories into the Amber format using VMD v1.9.1. 74 In all cases, we followed the single trajectory (ST) approach, 71,73 in which the trajectories for both protomers were extracted from that of the dimeric form of FraC. In the ST approach only a single trajectory is required to generate all three ensembles. Furthermore, the internal potential terms (e.g., bonds, angles, and dihedrals) cancel exactly, because the conformations in the bound and unbound ensembles are the same, leading to lower fluctuations, and easier convergence of binding free calculations with the MM-GBSA approach. 71,73 The GBn2 implicit-solvation model (igb 5 8) was employed to calculate the DG GB value of each complex. 71,72 Topologies were obtained with the Amber program tleap using mbondi3 radii for GBn2 models. 71,72 Salt concentrations of 0.05 mol/L were set for each system and default solvent and solute dielectric constants (e w 5 78.3 and e in 5 1, respectively) and rgbmax cutoff (rgbmax 5 25 Å ) values were used in all GB calculations. 71 In all cases, the DSA values were determined using the linear combination of pairwise overlaps algorithm. 75 DG nonpolar was then estimated through Eq. (1). 50 Finally, DE gas was estimated from AMBER99SB parameters. 61 Mean values of the energy terms were obtained by averaging over the snapshots extracted every 10 ps from each production MD simulation. Snap shots were considered as statistically independent from each other, since previous work has determined a typical correlation time for MM-GB(PB)SA energy terms of $5 ps or less. 76 Energetically-relevant residues (i.e., warm-and hot-spots, at the interfaces of the studied complexes) were predicted by using the computational alanine scanning (CAS) protocol. 50,71 Briefly, Ala single-point mutations were generated at specific positions using the Modeller 9.14 software 77 and the topologies of the mutated complexes were obtained using tleap of Amber12. Subsequently, relative free energy values (DDG) between the native and mutated complexes were determined using MMPBSA.py. 50,71 These calculations were performed under the ST approach in which the trajectory of the mutated complex is generated from that of the native complex by simply truncating the side-chain of the residue of interest and replacing the Cg atom by a hydrogen. 78,79 Finally, the pairwise effective free energy decomposition (pwEFED) protocol of MMPBSA.py was employed to calculate interaction energies between pairs of residues (DG r1,r2 ). 50,71,78 Cloning, expression, and isolation of the mutants fraC, eqtII, and stII coding genes were cloned in the expression vector pET21a1 (Entelchon, Bad Abbach, Germany) without any tag to avoid affecting protein activity. The double mutants FraC V60D/F163D , EqtII V60D/F163D , and StII I58D/I161D were obtained by replacing the corresponding residues by the anionic charged aspartic acid aminoacid. Recombinant proteins and mutants were producedin E. coli BL21 (DE3) competent cells(F 2 ompThsdSB(rB 2 mB2) gal dcm met (DE3)). For purification we followed the protocol described by Pazos et al., 80 with a slight modification. Briefly, we used cation-exchange chromatography in CM-cellulose column (Sigma-Aldrich, Munich, Germany) employing a NaCl continuous linear gradient (from 0 to 1M) in 0.05M ammonium acetate buffer, pH 5. The identity of the recombinant proteins and mutants were verified by mass spectrometry in a MALDI TOF/TOF TM 5800 mass (Darmstadt, Germany). Protein concentration was determined employing absorption coefficients of 42,530 cm -1 M 21 (FraC), 38

Circular dichroism studies
Circular dichroism (CD) spectra were acquired from 190 to 260 nm in 1 mm path length cuvettes, at room temperature using a JobinYvon CD6 spectropolarimeter (JobinYvon, Longjumeau, France). The instrument was routinely calibrated with an aqueous solution of recrystallized D-10-camphorsulfonic acid. The reported spectra are averages of 6 scans. Samples were prepared at 10 mM in PBS buffer and spectra were baseline corrected by using control samples of solutions not containing protein. The percentage of secondary structure was calculated using the Dichroweb Internet server. [34][35][36] Surface pressure measurements on lipid monolayers Surface pressure (p) measurements were carried out with a lThrough-S system (Kibron, Helsinki, Finland) at room temperature under constant stirring. The aqueous phase consisted of 300 mL of Trisbuffered saline (TBS: 145 mM NaCl, 10 mM Tris-HCl, pH 7.4). The lipid mixture of POPC:SM (50:50) was predissolved in chloroform:methanol (2:1, v:v) and was gently spread over the surface. Then, the desired initial surface pressure (p 0 ) was attained by changing the amount of lipid applied to the airwater interface. The proteins were injected into the subphase to achieve 0.8 lM protein final concentration at which no effect on surface tension of the airwater interface was observed. 83 The increment in surface pressure (Dp) was recorded as a function of the elapsed time until a stable signal was obtained.

Preparation of vesicles
The lipid mixture of POPC:SM (50:50) was dissolved in chloroform:methanol (2:1, v:v), mixed and evaporated thoroughly at 508C, and submitted to vacuum for not less than 2 h. For permeabilizing assays, multilamellar vesicles (MLV) were obtained by subsequent hydration in the presence of 80 mM carboxyfluorescein (CF), pH 7.4 in water and subjected to six cycles of freezing and thawing. For binding assays, MLV were hydrated in TBS. Small unilamellar vesicles (SUV) were prepared through sonication of MLV suspensions using an ultrasonicator (Branson 450, Danbury, USA) equipped with a titanium tip and subjected to 15 cycles of 1 min sonication between resting intervals of 1 min. Titanium particles released from the probe were removed by further centrifugation at 10 000g for 10 min. To remove untrapped CF in the permeabilization assay, the vesicles were filtered through mini-columns (Pierce, Rockford, USA) loaded with Sephadex G-50 (medium) pre-equilibrated with TBS. Phospholipid concentration was measured by inorganic phosphate quantification. 84

Intrinsic fluorescence measurements
Fluorescence spectra were recorded in a Hitachi F-4500 spectrofluorimeter, using 1 cm path length quartz cuvettes at room temperature. Slit widths of 5 nm nominal band pass were used both in the excitation and emission beams. Intrinsic fluorescence emission spectra of the proteins (2 mM) were recorded from 300 to 450 nm in solution. Excitation was performed at 295 nm to obtain fluorescence spectra derived only from Trp residues. Changes in the intrinsic fluorescence of the proteins upon addition of increasing quantities of SUV composed of POPC: SM (50:50) were also measured. Background intensities measured in samples without proteins were subtracted.
Quenching of proteins was achieved by adding the water-soluble quencher acrylamide (AA) in the presence of buffer or saturating concentration of SUV (40 mM). The maximum fluorescence intensities without and with the quencher (F 0 and F AA , respectively) were determined and F 0 /F AA values were plotted as a function of acrylamide concentration ([AA]). The experimental data were analyzed according to the Stern-Volmer equation: The slopes of the best-fit linear plots were used to determine the Stern-Volmer quenching constants (K SV ). 85 Leakage studies from CF-containing SUV SUV permeabilization was assessed at room temperature, using a FLUOstar OPTIMA microplate reader (BMG Labtech, Offenburg, Germany) by measuring the fluorescence (exc at 490 nm and em at 520 nm) of CF released. The toxins were two-fold serially diluted in the well microplates in a final volume of 100 lL of TBS and the reaction was started by adding the same volume of SUV (10 lM of final lipid concentration). After mixing vesicles and toxins, the release of CF produced an increase in fluorescence (f) due to the dequenching of the dye into the external medium, which was resolved in time. Spontaneous (toxin-free) leakage of dye was negligible under these conditions. Maximum release was always obtained by adding 1 mM Triton X-100 (final concentration) and provided the fluorescence value f max . The fraction of fluorophore release (f) was calculated as follows: where f 0 and f t represent the value of fluorescence before and at time t after toxin addition, respectively. C 50 was calculated by fitting dose-dependence curves of permeabilization induced by the toxins to a Hill sigmoid using Origin 8.0, Microcal (USA).

Hemolytic activity
Hemolytic activity (HA) was evaluated turbidimetrically at 600 nm at room temperature in a microplate reader (Labsystems, Helsinki, Finland). Erythrocyte suspension was prepared using pooled fresh human red blood cells collected intravenously from at least four healthy volunteers. Cells were washed by repeated centrifugation (600g, 15 min), the cell pellet resuspended in TBS and finally diluted to an apparent absorbance of 0.1 at 600 nm. Toxins were twofold serially diluted in a flat-bottom 96-well microplate in a final volume of 100 mL of TBS. The hemolysis was started by adding the same volume of red blood cell suspension. The decrease in optical density (O.D) was recorded as a function of time with intermittent shaking. The hemolytic activity (HA) was calculated as follows: where OD t , OD 0 , and OD min represent the O.D. at time t, at time zero, and in the presence of an excess of toxin, respectively.