Based on TCMSP, 9 Chinese herbal medicines (Eclipta prostrata L, Hedyotis diffusa Willd, Lobelia chinensis Lour, Houttuynia cordata Thunb, Sedum sarmentosum Bunge, Andrographis paniculata Nees, Cynanchum paniculatum, Curcuma longa L, Pinellia ternata), which have been claimed to have a therapeutic effect on snakebite, were selected for this study. They were retrieved in TCMSP and then screened according to the criterion of drug-like properties (DL) ≥ 0.18 and oral bioavailability (OB) ≥ 30% [33] . Then those compounds which satisfied the standard, were downloaded as ligands and imported into OpenBabel for building files as “sd” or “sdf” format. Subsequently, these files were inputted into DS Minimize Ligands module for hydrogenation, adding charge and structural optimization under CHARMm force field.

The important clinical manifestations of patients who were bitten by Deinagkistrodon acutus are tissue edema and bleeding, which are closely associated with acidic PLA2 in snake venom. So the inhibition of acidic PLA2, has important significance for relieving tissue edema and bleeding. Based on it search for the acidic PLA2 in the PDB database as a receptor for molecular docking, the results showed that there is only one acidic PLA2 of known structure (PDB ID: 1IJL).The crystal structure of dPLA2 (PDB ID: 1ijl) was accessed from Protein Data Bank (PDB) of RCSB (http://www.rcsb.org/) as receptor. Subsequently, the protein was modified using DS, including removal of water, adding hydrogen, and replenishing of lost residues. By means of CHARMm force field (OPLS3e), the protein structure was optimized to assure the structural accuracy of final receptor [10] .

The compounds through pharmacokinetics screened was performed to batch molecular docking by the Dock Ligand modules. The binding sites were Y27, G29, G31, H47, D48, Y51, and D98, so the active center was X = 6.826, Y = 12.735, Z = 50.709, the grid was set as 15 × 15 × 15. Docking type was set as Libdock. Moreover, the docking score prediction of the molecules was performed, each molecule retained the highest score and arranged these scores in descending order.

The molecules with the top score conformations were further redocked using Autodock vina with the same parameters as batch molecular docking to generate binding energy, hydrogen bonding and hydrophobic interaction results. The docking results were saved in pdb format respectively. Then we imported the results into PYMOL in pdbqt format for visualization processing and into DS in pdb format to obtain 2D structure diagram.

Initial configurations of ligands were produced with ACPYPE (http://bio2byte.be/acpype/) Server to prepare the topology structures, simultaneously, the Pdb2gmx was applied to generate the protein topology and structure files.

Firstly, the protein was centered on the cube box. Then the simple point charge (SPC) water model was solved. For the purpose of energy minimization of the system, we eliminated collisions at atomic positions through implement the fastest descent method. Due to the exposed residual charge of the protein given the system -12 net charge, so 12 Na⁺ counter ions were added to conform electroneutrality ahead of the actual simulation. Next, the band constraint of LINCS algorithm was applied to the system, the long range electrostatic interaction was calculated by particle grid Ewald (PME) algorithm. Afterwards, the NVT and NPT ensembles (constant number of atoms, volume, pressure, and temperature) were performed to the system for the MD simulation studies. In addition, at 1.4 nm, the cutoff values of Van der Waals interaction and short range Coulomb were set respectively in the system, periodic boundary conditions were applied to the whole three-dimensional spaces. Finally, 100 ns intervals of MD simulations were carried out for all the complexes with the time step 2fs, as well as the following data were investigated, including RMSD, RMSF and potential energy of protein-ligand complexes, radial distribution of ligand, complexes free energy profile and hydrogen bonds.

The root mean square deviation reflects the degree of data deviation from the mean value. In the dynamic simulation of protein-ligand complexes, it represents the degree of fit or deviation between the conformation after simulated “nt” time and the initial conformation. The greater the RMSD value, the greater the difference between the conformation after “nt” time simulation and the original conformation, that is, the worse the conformation stability. On the contrary, the more stable it is. The calculation formula is below:

RMSD = 1 N ∑ i = 1 N ( δ i − δ 0 ) 2

δ i represents the position of the conformation of the I-th frame, and δ 0 represents the position of the initial conformation.

The root-mean-square fluctuation indicates the position change of an atom over a period of time. It also reflects the degree of flexibility of the atom. The calculation was derived as follows:

RMSF = 1 T ∑ t j = 1 T ( x i ( t j ) − x 0 ) 2

x i ( t j ) − x 0 represents the position of an atom at “t” time subtraction the position at the initial time.

Approximately, 385 chemical components from 9 kinds of herbs (Eclipta prostrata L, Hedyotis diffusa Willd, Lobelia chinensis Lour, Houttuynia cordata Thunb, Sedum sarmentosum Bunge, Andrographis paniculata Nees, Cynanchum paniculatum, Curcuma longa L, Pinellia ternata), with potential antivenomous activity,were carried out to virtual screening by TCMSP. Under the above conditions, 89 compounds were selected for further assessment.

To exclude these molecules that may be harmful to human body, it is critical to screen the candidates to fit for drug-like properties. After ADME analysis,the distribution of 89 molecules are as shown in Figure 1(a). After ADME screening, 7 molecules were identified with a greater potential as the inhibitors against dPLA2 as shown in Figure 1(b). They are presented in Table 1 as, 3-[(2S,3R)-2-(4-hydroxy-3-methoxy-phenyl)-3-(hydroxymethyl)-7-methoxy-2, 3-dihydrobenzofuran-5-yl] propyl acetate; Ruvoside-qt; 14-deoxy-12-methoxyandrographolide; 14-deoxy-11-oxo-andrographolide; (3S,6S)-3-(benzyl)-6-(4-hydroxybenzyl) piperazine-2,5-quinone (MOL006957); tomentogenin (MOL005622) and Isoramanone. Their chemical structures are shown in Figure 1(c).

The toxicity of seven molecules was predicted using three models of DS: Mutagenicity, Developmental_Toxicity_Potential and Rat_Oral_LD50. A prediction score less than 0 indicates relative nontoxicity, a score greater than 0 indicates toxicity, and a larger value indicates greater toxicity. Toxicity prediction revealed that of the seven compounds, only 14-deoxy-11-oxo-andrographolide had no potential developmental toxicity. From the perspective of oral half-lethal dose in rats, the highest value was found on tomentogenin (8.12 g/kg), with the potential developmental toxicity of 0.47699, indicating that it was less toxic. From the perspective of mutagenicity, none of the 7 molecules were teratogenic, and the lowest teratogenic was Ruvoside-qt (−37.5246). Moreover the oral lethal dose of Ruvoside-qt was 6.38, which was relatively large, suggesting that it was less toxic.

Libdock batch molecular docking was performed to predict the binding possibility of the seven ligands to the active site of dPLA2. As shown in Table 1, the top docking score is Ruvoside-qt with 140.465, which indicated that Ruvoside-qt was the most likely to combine and function with dPLA2. The other compounds were sorted by score for 3-[(2S,3R)-2-(4-hydroxy-3-methoxy-phenyl)-3-(hydroxylmethyl)-7-methoxy-2,3-dihydrobenzofuran-5-yl] propyl acetate; tomentogenin; 14-deoxy-11-oxo-andrographolide; 14-deoxy-12-methoxyandrographolide et al.

Note: DTP is the potential for developing toxicity, the smaller the value indicates the smaller the potential for developing toxicity, positive value has the potential for developing toxicity (Toxic), negative value indicates no potential for developing toxicity (non-Toxic); Ames score is mutagenicity, the smaller the value indicates the lower mutagenicity, positive value indicates mutation (Mutagenicity), and negative value indicates no mutation (non-Mutagenicity).

Phosphatidylcholine (1,2-Diacyl-SN-glycero-3-phosphocholine), the natural substrate of dPLA2, is the most abundant phosphatidylcholine component in cell membranes. The hydrolase dPLA2 breakdowns membrane phospholipids, which generates inflammatory mediators, such as arachidonic acid, prostaglandin and platelet activating factor, causing inflammation in humans. Our use of phosphatidylcholine-dPLA2 complex as a positive control to screen for competitive inhibitors is highly convincing. It can also verify the accuracy of the active site.

The experiments in vitro incubation found that Gallic acid and pyrogallol shown obvious inhibitory effects on dPLA2 in a concentration-dependent manner [37] [38] . For this reason, we employed Gallic acid and pyrogallol to interact with dPLA2 respectively, to verify the feasibility of the active site, laying a foundation for the accuracy of molecular docking and the feasibility of screening.

Autodock vina was considered as an important computational tool in search of the interaction mode of the complex. It can more precisely predict the binding interactions between the compounds and the active site of the target protein. In this way, AutoDock Vina generated a docked complex for each ligand with various conformation and affinity scores. The best protein-ligand complex docked pose were analyzed and graphically visualized by DS.

As shown in Table 2, molecular docking revealed that phosphatidylcholine

Note: Table 2 shows the docking results of the seven small molecules with dPLA 2 molecules (top 7), and the docking results of dPLA 2 with phosphatidylcholine, gallic acid and catechol (last 3). In Table, MOL006957 to MOL008213 are the potential active components obtained in virtual screening according to the size of docking binding energy; MOL008290 to MOL000106 are the small molecules that inhibit PLA2 and used as a positive control. The abbreviated VDW applied in the table represents the van der Waals.

combine well with the protein in the position of TRP30, GLY31, HIS47 and LYS60 through hydrogen bonds as shown in Figure 2(b), hydrophobic interactions with CYS44, HIS47 and PRO120, anion interactions with TRP30, charge attraction interactions with LYS60, and C-H bonds with GLY32. These binding

sites are similar to the dPLA2 active site. Meanwhile comparing with phosphatidylcholine, gallic acid and pyrogallol have similar dPLA2 binding sites, both gallic acid and pyrogallol form hydrogen bonds with GLY29 and HIS47 as shown in Figure 2(a). Besides, all three form hydrogen bonds and hydrophobic interactions with HIS47. It then suggested that HIS47 may perform an important part in the catalytic activity of dPLA2.

In terms of binding energy, pyrogallol and phosphatidylcholine have the same binding energy of −5.7 kJ/mol, gallic acid −6.3 kJ/mol and combine with Ca²⁺ by metallic bond. Thus, according to the interaction of three controls with dPLA2, the catalytic active center of the receptor is a calcium-ion binding ring composed of TYR27, GLY29, GLY31, HIS47 and ASP48. Moreover HIS47 and GLY29 may be key targets in the catalytic activity.

After docking, 7 molecules are able to combine with the target protein of dPLA2. The binding energies are all less than −5.7 kJ/mol. The lowest binding energy is MOL006957 (−9.0 kJ/mol). The second least binding energy for Isoramanone being −8.1 kJ/mol. As it is known that the lower the energy, the stronger the bond, they were the ligands with the most binding affinity to the receptor. In terms of interaction, MOL006957 forms hydrogen bonds with GLY29 and GLY31 as shown in Figure 2(d), hydrophobic bonds with TRP30, metallic bonds with Ca²⁺, cations interaction with LYS60, and CYS 44 forms the Pi-sulfur bond and the C-H bond with HIS47, which is most similar to phosphatidylcholine in general. Isoramanone forms hydrogen bond interaction with HIS47 and ASP48 as shown in Figure 2(c), hydrophobic interaction with LEU2, TRP30 and TYR51, and metallic bond with Ca²⁺. Consequently, MOL006957 is most likely to be the ligand of dPLA2, which was further verified by molecular dynamics simulations.

Figure 3(a) shows RMSD of the backbone of the seven protein-ligand complexes which changed with the time over 100 ns. A conclusion can be drawn from the picture that after simulation, the conformations of these seven complexes tended to converge. Comparing with the initial conformation, the deviation of all the complexes were less than 0.4 nm. In particular, the RMSD of MOL005622-dPLA2 and MOL006957-dPLA2 complexes were lower than other compounds. The result indicated that the two compounds were relatively stable over 100 ns. Then, in comparison to dPLA2, RMSD of MOL005622-dPLA2 and MOL006957-dPLA2 complexes was lower overall and it suggested that the structural flexibility of dPLA2 were reduced after binding ligands, as shown in Figure 3(b).

RMSF reflected the vibration of each complex compared with the initial conformation, as shown in Figure 3(c). In comparison to other compounds, there was the smallest molecular vibration in MOL006957-dPLA2 complex, which reflects that it is more stable than other compounds at the molecular level. Simultaneously comparing with dPLA2, MOL006957-dPLA2 molecular vibration was

significantly weakened, as shown in Figure 3(d).

FEL is able to evaluate comprehensively the free energy of different conformations in simulation process through the following two parameters: root mean square error (RMSD) and gyrate. As shown in Figure 4(a), the lower the energy (the darker the blue), the more stable the conformation. Through the parameter files, it was found that the lowest energy conformation was frame 3670 and frame 3680, and compared the lowest energy conformation with the initial conformation, as shown in the Figure 4(b). Furthermore, hydrogen bonds play an important role in ligand-receptor interactions, and the number of hydrogen bonds is also an important index to judge the binding stability of protein-ligand complexes. As shown in the Figure 4(c), the overall number of hydrogen bonds increased with the extension of simulation time, and finally remained at 3 to 4.