Search for efficient inhibitors of myotoxic activity induced by ophidian phospholipase A2-like proteins using functional, structural and bioinformatics approaches

Ophidian accidents are considered an important neglected tropical disease by the World Health Organization. Particularly in Latin America, Bothrops snakes are responsible for the majority of the snakebite envenomings that are not efficiently treated by conventional serum therapy. Thus, the search for simple and efficient inhibitors to complement this therapy is a promising research area, and a combination of functional and structural assays have been used to test candidate ligands against specific ophidian venom compounds. Herein, we tested a commercial drug (acetylsalicylic acid, ASA) and a plant compound with antiophidian properties (rosmarinic acid, RA) using myographic, crystallographic and bioinformatics experiments with a phospholipase A2-like toxin, MjTX-II. MjTX-II/RA and MjTX-II/ASA crystal structures were solved at high resolution and revealed the presence of ligands bound to different regions of the toxin. However, in vitro myographic assays showed that only RA is able to prevent the myotoxic effects of MjTX-II. In agreement with functional results, molecular dynamics simulations showed that the RA molecule remains tightly bound to the toxin throughout the calculations, whereas ASA molecules tend to dissociate. This approach aids the design of effective inhibitors of PLA2-like toxins and, eventually, may complement serum therapy.

Crystallographic structures. The crystal structures of both MjTX-II complexes present similar folding found in class-II PLA 2 s solved to date, including seven disulfide bridges in each protomer and the following structural features: an N-terminal α-helix; a "short" helix; a Ca 2+ binding loop (non-functional for PLA 2 -like toxins); two α-helices; two short strands of anti-parallel β-sheet (β-wing) and a C-terminal loop 10,42 .
The MjTX-II/RA crystal diffracted up to 1.60 Å and belonged to the P2 1 2 1 2 1 space group. The refinement converged to an R cryst of 15.7% (R free = 19.1%) with one RA molecule located close to the MDiS and helix 1 from monomer A and 392 water molecules (Table 1, Fig. 2a and b). Due to the lack of electron density, the side chains of the following residues were not modeled: Lys69 and Lys116 from monomer A and Lys70, Lys115, Lys128 and Lys129 from monomer B. The MjTX-II/ASA crystal diffracted up to 1.69 Å resolution and belonged to the P2 1 space group. The final refined data converged to an R cryst of 18.1% (R free = 21.8%), with two ASA molecules interacting with the hydrophobic channel from each monomer, 3 DMSO molecules, and 314 water molecules (Table 1, Fig. 2c,d). Due to the lack of electron density, the side chains of Lys7, Lys60 and Lys117 residues from monomer A and Lys53, Lys69, Lys105 and Lys117 residues from monomer B are not modeled. Both crystal structures (coordinates and structure factors) were deposited into the Protein Data Bank (PDB) under the following ids: 6MQD (MjTX-II/RA) and 6MQF (MjTX-II/ASA).
The superposition between the PLA 2 -like myotoxins (e.g., PrTX-I/RA, MjTX-II/FA14 and BthTX-I) and the structures solved here revealed that the MjTX-II/RA and MjTX-II/ASA structures present a distorted dimeric assembly (Fig. 3) as expressed by Euler angles 15 and center of mass displacement (COMdisp) values (see Methods section) ( Table 2).

Molecular dynamics.
The MD simulations of the MjTX-II/RA complex resulted in average RMS deviation of frame trajectories (ftRMSD) of the MjTX-II backbone atoms of 0.56 ± 0.11 nm (Fig. 4a) with no clear indication that the protein reached a stable conformation. The average RMS deviation of frame trajectories (ftRMSD) of the RA molecule was 0.22 ± 0.03 nm, indicating that the ligand interacts with the toxin in the same region (Fig. 4b). However, the RA effect on the MjTX-II quaternary structure results in its disturbance, forcing the toxin into a distorted conformation, evidenced by the increase of its radius of gyration (R g ) from 20.0 to 21.8 Å and COMdisp of 31.5 Å (Table 2).
In contrast, MD simulation with the MjTX-II/ASA complex showed that the ASA molecules left their initial binding sites before 2 ns of the simulation, and subsequently, both molecules dissociated from MjTX-II or interacted non-specifically during the simulation. The unstable behavior of ASA can also be observed by the high standard deviation of ftRMSD value (0.51 ± 0.23 nm) (Fig. 4b). In addition, the MjTX-II/ASA complex did not show a clear conformation stability during the simulation (Fig. 4a) and presented a ftRMSD backbone atoms of 0.46 ± 0.08 nm. In addition, other two MD simulations of 300 ns were performed aiming to test the dynamics of the distorted quaternary conformation observed in the MjTX-II/ASA complex, using the following systems: i) unbound MjTX-II and ii) unbound MjTX-II with fatty acid (FA) molecules (supposedly an activator PLA 2 -like myotoxin) 14 . FA molecules (stearic acids) were placed in the hydrophobic channel of this structure in a similar position as observed in the MjTX-II/stearic acid crystal structure (1XXS) 43 . The MD simulation of the unbound MjTX-II presented a high ftRMSD (0.64 ± 0.11 nm) (Fig. 4c) and an increase of its radius of gyration (Fig. 4d) compared with the active MjTX-II structure (MjTX-II/FA14 18 ) ( Table 2 and Fig. 5a). In contrast, MD simulation of the MjTX-II/FA presented more stability (ftRMSD of 0.43 ± 0.04 nm) (Fig. 4c) and the model became more globular than its initial structure (see R g in Table 2 and Fig. 4d), similar to the active structure 18 .
As observed in the Table 2, different parameters (e.g. COMdisp and R g ) indicated that after MD simulation, the MjTX-II/FA complex had a conformation more similar with the active form (MjTX-II/FA14) and presented a more compact structure. The Fig. 5b shows the superposition between the MjTX-II/FA structure after the simulation and the active MjTX-II/FA14 crystal structure. In the Table 2, it is also possible to observe that the MjTX-II/ FA complex may reach, for example at 170 ns of the MD simulation, a even more compact conformation which its R g is very similar with the active MjTX-II/FA14 crystal structure (~18 Å). In addition, at this MD simulation moment the MjTX-II/FA complex presented a COMdisp decrease from 12.7 to 6.2 Å.    muscle membrane 38 . Initially, this membrane disarrangement results in the loss of permeability control to ions and macromolecules [49][50][51] . Thus, there is cellular depolarization, including the re-equilibrium of Na + , K + and Cl − ions 38,49,52 , the influx of extracellular Ca 2+ ions 53,54 and the release of intracellular Ca 2+ ions 55 . Thus, an increase in the sarcoplasmic Ca 2+ concentration promotes a series of adverse events, such as muscle contracture 49 , hypercontraction of myofilaments, mitochondrial damage and activation of proteases and Ca 2+ -dependent PLA 2 50,51 , which amplifies the process of muscle injury 52,53 . Finally, the ATP released by the injured sarcolemma diffuses into neighboring regions of muscle fibers unaffected by the toxin, activating its P2X purinergic receptors in these muscle fibers with the consequent transmembrane passage of Na + , K + and Ca 2+ ions [56][57][58][59] . This ion movement results in the depolarization of these newly affected regions and an amplification of the muscle injury process in regions away from the initially disrupted sarcolemma.
Herein, we tested the ability of RA molecules to neutralize this destabilizing membrane activity through the paralyzing effect promoted by MjTX-II. Thus, we demonstrated that RA significantly reduces the blockage of indirectly evoked muscle induced by MjTX-II (Fig. 1), similarly to observed against PrTX-I, a PLA 2 -like myotoxin from Bothrops pirajai venom 20 . Ticli and colleagues (2005), using different technical approaches, observed the antimyotoxic potential of RA against the crude venom of Bothrops jararacussu and against the main PLA 2 -like myotoxins (BthTX-I and II) from this venom, as well the enhancement effect of serum therapy by this inhibitor. Furthermore, the authors exclude the proteolytic degradation of the toxin as a potential mechanism involved in the inhibition by RA 21 . Analogously, other molecules also presented an inhibitory effect against muscle paralysis promoted by PLA 2 -like toxins, such as suramin against MjTX-II 17 and BthTX-I 41 and Zn 2+ ions 19 and chicoric acid against BthTX-I 23 .
ASA is the most widely used drug in the world 60 and, since the 1970s, it has been known that mechanism of anti-inflammatory, analgesic and antipyretic activities by this molecule occurs through cyclooxygenase (COX) inhibition 35,60 . Furthermore, other studies have investigated different applications of ASA, including its action against whole venom from Daboia russelii (Viperidae) 36 and its toxic effects against catalytic PLA 2 [61][62][63][64] . Singh and colleagues (2005) proposed the structural basis of an elapid PLA 2 inhibition by ASA, but no functional assays were performed 65 . Recently, ASA was evaluated against the catalytic inhibitory activity of a pancreatic PLA 2 37 (a secreted PLA 2 with similar activity to PLA 2 s from Viperidae snake venom 66 ) by bioinformatics and affinity assays. A weak interaction between ASA and the enzyme by affinity assays was observed, and a possible interaction at the Ca 2+ binding loop was suggested by docking assays 37 . Thus, to test the ability of ASA to inhibit PLA 2 -like toxins, we performed the first tests of its interaction with a PLA 2 -like toxin and the physiological-pharmacological inhibition assays. Interestingly, despite the structural similarities between catalytic PLA 2 and myotoxic PLA 2 -like toxins (without catalytic activity), ASA did not inhibit the blockage of indirectly evoked muscle contractions promoted by MjTX-II (Fig. 1).

Structural evidence for MjTX-II inhibition by rosmarinic acid.
To date, only one crystal structure of a PLA 2 -like/RA complex, PrTX-I/RA (PrTX-I from B. pirajai venom complexed to RA), has been deposited in the Protein Data Bank. In this structure, an inhibitor molecule interacts with the N-terminal residues of a toxin monomer 20 (Fig. 6a,b), and therefore, the RA molecule physically blocks the access of fatty acid molecules to the hydrophobic channels of PrTX-I. According to the current myotoxic mechanism proposed for PLA 2 -like toxins 14,15 , the entrance of a fatty acid (FA) molecule in the hydrophobic channel of the toxin is the first step of the molecular mechanism and leads to the allosteric activation of these toxins, with the subsequent exposure and alignment of their membrane docking and disruption sites (MDoS and MDiS, respectively). Thus, the RA molecule may prevent the activation of PrTX-I by blocking its hydrophobic channel. Similar to the PrTX-I/RA complex, the crystal structure of the MjTX-II/RA complex revealed an inhibitor molecule interacting with the toxin; however, RA interacts with both monomers simultaneously and in a different region of MjTX-II (Fig. 6c). The first benzene ring region of the RA molecule (derived from the diphenyl-lactic acid portion) interacts with the N-terminus of monomer B (Asn17, Pro18 and Ala19 residues). The second benzene ring region of the RA molecule (derived from the caffeic acid portion) interacts with residues of the C-terminal region of monomer A (Leu122, Phe126 and Cys127 residues) (Fig. 6d), which includes the MDiS region.
Interestingly, RA and suramin 17 are efficient inhibitors for MjTX-II and other PLA 2 -like toxins (e.g., BthTX-I, PrTX-I, MjTX-I, BaspTX-II and Ecarpholin S), but they bind to different regions of these toxins and display different inhibition mechanisms [20][21][22]41,67 . Furthermore, it is important to highlight that several structural studies with MjTX-II complexes [16][17][18]43 demonstrated that the binding of ligands in MjTX-II is always different when compared to other PLA 2 -like toxins 20,22,39,68 . This fact was attributed to a residue insertion (Asn120) and two residue mutations (at positions 32 and 121), which cause the binding of additional ligands (e.g., fatty acids or PEG) or interactions in different regions (RA and suramin). Several functional/structural studies with PLA 2 -like toxins and ligands (e.g., aristolochic and chicoric acids, suramin and Zn 2+ ions) 17,19,23,39 have demonstrated that their binding to the MDiS region is associated with the inhibition of these myotoxins, such as that observed in the crystal structure and functional data of the MjTX-II/ RA complex presented here. Thus, an explanation of these phenomena is probably the lack of membrane disruption function of PLA 2 -like toxins.
Acetylsalicylic acid binds to MjTX-II but does not inhibit its myotoxicity. The crystal structure of the MjTX-II/ASA complex reveals that an ASA molecule is bound in the hydrophobic channel of each MjTX-II monomer. Thus, these two ASA ligands establish hydrophobic interactions with several residues, some of which are conserved from catalytic PLA 2 s, such as His48, Tyr52 and Gly30, and are essential for the coordination of co-factor Ca 2+ and for the catalytic activity of those enzymes 10,69 . Therefore, at first sight, the results may suggest that ASA inhibits PLA 2 -like toxins because its binding may prevent the activation of the toxin by fatty acid molecules (the first step of the proposed myotoxic mechanism). However, in the same functional experiments with PD preparations, as performed by myographic assays with MjTX-II/RA treatment, the inhibitory ability of ASA against the neuromuscular blockage induced by MjTX-II was not observed.
One explanation for these contrasting observations may be related to the low stability of the ASA molecules in the complex with the toxin, as observed in MD simulations (Fig. 4b) and due to the high content of water molecules mediating the interaction between the ligand and the toxin (Fig. 7). This eventual instability may lead to its displacement by FA molecules present in the isolated organ-bath chamber, promoting a return to the active form of MjTX-II 18 .
The presence of FA molecules in the chamber may be explained by the method employed in this assay 70 . During the removal of PD preparations from animals and their assembly in these chambers, PD preparations are sufficiently manipulated to cause initial lesions of some muscle fibers, disrupting their sarcolemma (composed of phospholipids, FA molecules and several other macromolecules) with the consequent release of their intracellular content. Although the PD preparation is subjected to washing procedures after its assembly and a stabilization period of the evoked contractions occurs before the beginning of the assays, the lesions remain in the PD preparations with continuous release of the intracellular content from injured fibers, including cytosolic phospholipases and other proteases that are activated by Ca 2+ ions 50,51 . Thus, apart from the remnants of FA molecules between the PD preparation washes, these cytosolic proteases can amplify the process of muscle injury 52,53 , releasing FA molecules prior to toxin addition into the chamber.

MD simulations with MjTX-II/RA, MjTX-II/ASA and MjTX-II/FA shed light on the inhibition of the myotoxic mechanism.
Aiming to further understand the observations achieved with myographic and crystallographic methods in the present study, the dynamic of the inhibitory process was studied by MD simulations, and four different systems were considered: (i) MjTX-II/RA, (ii) MjTX-II/ASA, (iii) unbound MjTX-II and (iv) MjTX-II/FA, as described in the results section. Thus, we observed that (i) the RA molecule interacted with the toxin throughout the simulation and the toxin reaches a even more distorted quaternary structure (Fig. 4b); (ii) ASA molecules dissociated from MjTX-II at the beginning of the simulation (Fig. 4b), but the quaternary structure presented a reasonable conformational stability throughout the simulation (Fig. 4a); (iii) the removal  Table 2); and (iv) in the presence of FA molecules, the quaternary structure of MjTX-II (starting from an initial distorted conformational MjTX-II/ASA model) became more globular and symmetric (Fig. 8). Therefore, these findings highlight the efficiency of RA as inhibitor of the toxic effects promoted by MjTX-II as observed by functional methods. The results suggest that RA inhibits this toxin not only by blocking its MDiS residues in a monomer but also by distorting the quaternary structure of MjTX-II, leading to misalignment the MDiS and MDoS regions from both monomers, which affects its myotoxic activity.
Moreover, the dissociation of ASA in the beginning of the MD simulation suggests low stability and/or affinity of the ligand in its binding site. This observation is compatible with functional assays that did not detect myotoxicity reduction or neutralization in its presence, as discussed in the previous sections.
FA molecules have an important role for oligomeric changes that lead to the activation of the myotoxic mechanism proposed for PLA 2 -like toxins 14,15 , but in the case of MjTX-II, their binding is not necessary for the alignment of the functional sites of this particular toxin 18 . However, the MD simulations performed here showed that the distorted starting model with no ligands (unbound MjTX-II) was not able to assume the active state (symmetric conformation), which may only obtain after the addition of FA molecules. These findings reinforce previous structural data with MjTX-II that showed that while FA molecules do not lead to important quaternary conformation changes, their binding has a direct influence on some side-chain orientations in the dimeric interface 18 , leading to stabilization of the dimeric conformation of MjTX-II.

Conclusions
In the present study, functional, crystallographic and bioinformatics assays involving MjTX-II, a PLA 2 -like toxin with structural particularities, and two potential inhibitors were performed. RA, previously tested as an efficient inhibitor for a PLA 2 -like toxin (PrTX-I), was demonstrated to be highly efficient for the inhibition of MjTX-II in mouse PD preparations. Interestingly, despite similar inhibitory effects against MjTX-II and PrTX-I, the RA binding sites in these toxins are different. In the case of MjTX-II, the inhibitory effect is attributed to binding in the MDiS region, which prevents the disruptive activity of the toxin. In addition, the distorted quaternary conformation of MjTX-II after RA binding may also contribute to its inactivity. In contrast, ASA, a ligand previously tested in catalytic PLA 2 , was not able to prevent the paralyzing effect of MjTX-II, despite its binding in the MjTX-II/ASA crystal structure. The MD simulation with MjTX-II/ASA showed the low stability of this ligand in the hydrophobic channel of MjTX-II, suggesting that this molecule may be replaced with another molecule with higher affinity (such as fatty acids) in the functional assays, which can then activate the toxin. Finally, MD simulations of a MjTX-II/FA model led to a symmetric and stable structure, which reinforces the importance of fatty acids for the stabilization of the toxin. This combination of functional, structural and bioinformatics assays used here can methodologically contribute to the design of effective antiophidic molecules.

Experimental Procedures
Toxin isolation and ligands source. Freeze-dried crude venom (150 mg) was solubilized in 0.05 M ammonium bicarbonate pH 8.0 and subjected to ion exchange chromatography. The fraction corresponding to MjTX-II was obtained by a gradient of 0.05 to 0.5 M ammonium bicarbonate pH 8.0, as described by Soares and colleagues 71 . For contaminant removal, this fraction was subjected to reversed-phase chromatography, with a gradient of 0-66.5% acetonitrile (in 0.1% trifluoroacetic acid) in a C18 column (Shimadzu). Acetylsalicylic acid (ASA) and rosmarinic acid (RA) were purchased from Sigma-Aldrich, St. Louis, Missouri, USA. Functional studies. Adult male mice (25-30 g) were euthanized by exsanguination after cervical dislocation to remove the phrenic nerve-diaphragm muscle and mounted vertically (under a resting tension of 5 g) in a conventional isolated organ-bath chamber containing 15 mL of physiological solution (135 mM NaCl, 5 mM KCl, 1 mM MgCl 2 , 2 mM CaCl 2 , 15 mM NaHCO 3 , 1 mM Na 2 HPO 4 and 11 mM glucose). This solution was constantly bubbled with carbogen (95% O 2 and 5% CO 2 ) and maintained at 35 ± 2 °C. The phrenic-diaphragm (PD) preparation was attached to an isometric force transducer (Grass-Telefactor, FT03) for recording twitch tension. The transducer signal output was amplified and recorded on a computer, via a transducer signal conditioner (Gould Systems, 13-6615-50), with the AcquireLab Data Acquisition System (Gould). Indirect contractions were evoked by rectangular pulses (0.2 Hz, 0.5 ms) and a supramaximal intensity delivered from an electronic stimulator (Grass-Telefactor, S88K) and applied to the phrenic nerve by means of a suction electrode. Crystallization and X-ray data collection. The purified fraction of MjTX-II used for co-crystallization was concentrated up to 10 mg.mL −1 and solubilized in 20 mM ammonium bicarbonate, pH 8.0. ASA was dissolved in 100% dimethyl sulfoxide (DMSO) and RA was dissolved in 95% ammonium bicarbonate (20 mM, pH 8.0) and 5% ethanol to obtain a molar ratio of 1:8 (protein:ligand, for both compounds) in crystallization drops. Crystals from MjTX-II/RA and MjTX-II/ASA complexes were obtained by a conventional hanging drop vapor-diffusion method 72 at constant temperature of 291 K for approximately 20 days, from a mixture of 1 µL of protein/inhibitor solution (previously incubated for 30 minutes) and 1 µL of reservoir solution and equilibrated against a reservoir (500 µL). The reservoir solution was similar to that previously found in the literature for this toxin and was composed of PEG 4000, Tris HCl pH 8.5 and lithium sulfate (MjTX-II/RA) and PEG 4000, isopropanol and sodium citrate (MjTX-II/ASA) [16][17][18] .
The crystals were mounted in a nylon loop and flash-cooled in a stream of liquid nitrogen using no cryoprotectant for X-ray diffraction data collection. The datasets were obtained using a synchrotron radiation source (MX2 station, Laboratório Nacional de Luz Sincrotron (LNLS), Campinas, Brazil) and a PILATUS 2 M detector (Dectris) at a wavelength of 1.458 Å (at 100 K).
Both crystal structures were solved by the Molecular Replacement Method using the software PHASER 75 from PHENIX software package v.1.12 76 and the monomer A atom coordinates of MjTX-II/FA14 (PDB access code 6B80) as a search model. The modeling, ligand insertion, solvent molecules and manual refinement process were performed using Coot v.0.8.9 software 77 . Structural automated refinement of models was performed by PHENIX software package v.1.12 76 , and the structural quality was checked using PHENIX software package v.1.12 and MolProbity software 78 16 and PrTX-I/RA (Piratoxin-I from Bothrops pirajai -PDB id: 3QNL) 20 were used. Molecular comparison of the structures was performed using Coot v.0.8.9 software 77 . All structural figures was generated using PyMOL v.1.3 software 79 and LigPlot + v.1.4.5 80 .
In order to analyze the conformations, the center of mass (COM) displacement measure was calculated. This measurement considers the active MjTX-II structure (MjTX-II /FA14, PDB id: 6B80) as reference. For each subset the structures (Table 2), both monomers A were superposed (C α atoms) and COM was calculated for each monomers B; subsequently, the COM distance between both monomers B was calculated by Euclidean distance, resulting in a displacement value called COMdisp. Molecular dynamics (MD) simulation. All MD simulations were performed using GROMACS (Groningen Machine for Chemical Simulation) v.5.0.5 81 under GROMOS96 54a7 force field 82 . MjTX-II protonation states were set to pH 8.0 using PROPKA3 server 83 , and each complex was placed in a triclinic box. The systems were solvated using simple point charge (SPC) water models to maintain the crystallographic water molecule positions and were equilibrated with 100 mM NaCl. A minimization step was applied using the Steepest Descent algorithm to reach a system energy below 100 kJ/mol/nm, and then, MjTX-II, RA, ASA and FA were restrained to perform an 1 ns NVT step using a V-rescale thermostat 84 at 310 K followed by an 1 ns NPT step adding Berendsen barostat 85 at 1 bar to accommodate the systems. Furthermore, unrestrained MD simulations were performed for each system using a Nose-Hoover thermostat 86,87 and Parrinello-Rahman barostat 88 .