The 20 flavonoid derivatives were docked into the H1N1 neuraminidase structure. Over the 10-ns MD trajectories of the H1N1 neuraminidase with tip3 water molecules and flavonoid derivatives, the overall structure of both complexes appeared to be equilibrated after 324 ps. Here, we show the RMSD profiles of 20 flavonoid derivatives (Figure 1) and the snapshot (Figure 2) of the complex system of the flavonoid derivatives 1. The RMSD values of 20 flavonoids stay within 4 Å.

The study of these 20 compounds has revealed that the amino residues can frequently interact with flavonoid inhibitors in the H1N1 neuraminidase binding site, and that these residues are responsible for the selectivity of flavonoid inhibitors. The results of our simulations are listed in Table 1 and Figure S1–S20. The inhibitors 1–3 and 14 (Table 1) belong to the 2,3-dihydrobenzofuran backbone inhibitors and the others belong to the 5,7-dihydroxychromen-4-one backbone inhibitors. In the 2,3-dihydrobenzofuran backbone inhibitors (inhibitor 1–3 and 14), Asn295, Glu119, Glu277, Thr226, Trp179 can form hydrogen bonds in the 2009 H1N1 neuraminidase/flavonoids complex structures and Asn295 most frequently forms the hydrogen bonds. Only Tyr402 has non-bonding interactions with inhibitor 1 (Figure S1). In the 5,7-dihydroxychromen-4-one backbone inhibitors (inhibitor 4–13 and 15–20), Arg152, Asn295, Asn325, Asn344, Asp151, Asp294, Glu119, Glu228, Glu277, Ser180, Ser247, Ser366, Ser367, Thr226, Trp179, Tyr402 and Val346 can form hydrogen bonds in the complex structures and Glu228 most frequently forms the hydrogen bonds. Arg368, Ile223, Pro326 and Trp179 have non-bonding interactions with the backbone inhibitors (Figure S7, 16 and 19). The overall results of our simulations suggest that Arg152, Asn295, Asn325, Asn344, Asp151, Asp295, Glu119, Glu228, Glu277, Ser180, Ser247, Ser366, Ser367, Thr226, Trp179, Tyr402 and Val346 can form hydrogen bonds between the 2009 H1N1 neuraminidase and flavonoid derivatives. Moreover, our simulations indicate that Arg368, Ile223, Pro326 and Trp179 have non-bonding interactions with these derivatives. The non-bonding interactions of the 2009 H1N1 neuraminidase/flavonoid complex structures only occurred in inhibitor 1, 7, 16 and 19 simulations. While six residues (Arg152, Asn295, Glu228, Glu277 Trp179 and Val346) more often formed the hydrogen bonds of the complex structures, Asn295 most frequently formed the hydrogen bonds.

The 2009 H1N1 neuraminidase/flavonoid (inhibitors 1–20) complex structures and the Tip3 water solvent box were constructed using the Autodock Vina docking and AmberTools 1.5 programs. The binding free energies of each inhibitor were obtained from the 10-ns MD simulation and the SIE method, with both processes using the same parameters. All the results are listed in Table 2. In the simulations, the predicted binding free energies of the 20 inhibitors was in good agreement with the experimental results (Figure 3), with the correlation coefficient being 0.75.

The hydrophobic and hydrophilic nature of 20 flavonoids was individually traced down the binding affinities with the important residue regions analyzed by the ligplot program (Arg152, Asn295, Asn325, Asn344, Asp151, Asp295, Glu119, Glu228, Glu277, Ser180, Ser247, Ser366, Ser367, Thr226, Trp179, Tyr402), and the water molecules (within a 10 Å radius of 20 flavonoids). All the results are listed in Tables S2,3 and Figures S21,22. For the hydrophobic nature analysis (aromatic groups), the two residues (Trp179 and Tyr402) have obvious binding affinities to the inhibitors. The contributions to Trp179 and Tyr402 binding free energies are −1.121~−2.961 and −1.665~−3.143 kcal/mol, respectively. For the hydrophilic nature analysis, the three elements (side chain of nitrogen atoms/Arg152, side chain of nitrogen and oxygen atoms/Asn295 and the water molecules within 10 Å of inhibitors) have obvious binding affinities to the inhibitors. Those contributions to Arg152, Asn295 and water molecules binding free energies are −0.507~−2.733, −0.185~−2.247 and −0.6~−2.6 kcal/mol, respectively. Figure S23 shows that the more hydroxyl function groups are present, the higher the binding affinities of the inhibitors.

All flavonoid derivatives inhibitors were constructed and minimized using VEGA ZZ [41] and ISIS/DRAW [42] programs. We aligned the H1N1 neuraminidase structure (PBD ID: 3NSS) and the template (PDB ID: 3B7E) [43], and then used the template structure drug (zanamivir) as a template to generate the active site of the H1N1 neuraminidase structure. Next, the Autodock vina docking program [44] was used to dock the 20 flavonoid derivatives inhibitors into the active site of the H1N1 neuraminidase structure, which was defined as all residues within 0.15 nm from alignment with the template structure drug. Autodock vina is a fast and accurate way to dock small compounds into fixed protein binding sites, utilizing NNscore [45] and several types of genetic algorithms. Four thousand conformations were obtained from docking for the 20 inhibitors, and these then scored by NNscore. The conformations of the best NNscores were then selected for subsequent MD simulations.

Calculations were performed with the NAMD molecular dynamics software [46] using the AMBER FF99 all-hydrogen amino acid and general amber force field (GAFF) parameters. The GAFF partial atomic charges are often based on the RESP fitting procedure of the electrostatic potential obtained at the HF/6–31G(d,p) level of theory [47]. Those levels of theory overestimate the gas-phase partial atomic charges giving rise to an effectively polarized force field. The geometries of the 20 flavonoid derivatives inhibitors were fully optimized and their electrostatic potentials were obtained using a single-point calculation; both operations were carried out at the HF level with the 6–31 G(d,p) basis set using the US GAMESS [48] program. Subsequently, their partial charges were obtained by the restrained electrostatic potential (RESP) using R.E.D tools [49]. From the docking simulations, the complex structures were inserted into the tip3p water box. All MD simulations were performed in the NVT ensemble (temperature equal to 310 K), unless noted, using a Verlet integrator, an integration time step of 0.002 ps, and SHAKE [50] data for all covalent bonds involving hydrogen atoms. In electrostatic interactions, atom-based truncation was undertaken using the PME method. In addition, the switch van der Waals functions were also used with a 1.8 nm cutoff for atom-pair lists. The complex structures were minimized for 20,000 conjugate gradient steps. The minimized complex structures were then subjected to a 10 ns isothermal, constant volume MD simulation. The MD simulation trajectories were converted to the Amber type coordinates by the AmberTools 1.5 program [51]. All MD simulation results were used to initiate the functionally important residues and the binding free energies calculations.

In a diseased target proteins, not every residue is equally important. Some residues are essential for important enzymatic functions and protein structure stability. Thus identification of functionally important residues can provide a clear insight into the structural aspects of H1N1 2009 influenza neuraminidase. In this work, the structure-based approach was applied to identify functionally important residues, while MD simulations were used to identify the residues involved in the binding pocket. In the MD simulations, the important residues regions were analyzed by the ligplot program [52]. The results of functional important residues were applied to binding free energies calculations.

The binding free energies calculations were performed by the solvated interaction energies method (SIE). The binding free energies between receptor and inhibitors were calculated for snapshot structures taken from the MD trajectory of the system. From the 10 ns protein-ligand MD trajectories, 100 snapshots were taken at regular intervals for the binding energies analyses. The SIE function [53] to estimate the binding free energies is written as:

where E_C and E_vdw are the intermolecular Coulomb and van der Waals interaction energies in the bound state, respectively. These values were calculated using the AMBER molecular mechanics force field (FF99) with an optimized dielectric constant. ΔG_bind^R is the change in the reaction field energies between the bound and free states and is calculated by solving the Poisson equation with the boundary element method program, BRI BEM, and using a molecular surface generated with a variable-radius solvent probe. The ΔMSA term is the change in the molecular surface area upon binding. The following parameters are calibrated by fitting to the absolute binding free energies for a set of protein–ligand complexes: AMBER van der Waals radii linear scaling coefficient (ρ), the solute interior dielectric constant (Din), the molecular surface area coefficient (γ), the global proportionality coefficient related to the loss of configurational entropy upon binding (α), and a constant (C). The optimized values of these parameters are α = 0.1048, Din = 2.25, ρ = 1.1, γ = 0.0129 kcal/(mol Ų), and C = −2.89 kcal/mol. The SIE calculations were carried out with the program sietraj [53]. The hydrophobic (non-hydroxyl group) and hydrophilic (hydroxyl group) nature of flavonoids can affect the binding abilities [29–31]. Thus these natures were individually traced down to the binding affinities with the whole H1N1 neuraminidase, the important residue regions analyzed by the ligplot program (hydrophilic and hydrophobic parts listed in Table S4), and the water molecules (within a 10 Å radius of 20 flavonoids).