Structure-based drug discovery for combating influenza virus by targeting the PA–PB1 interaction

Influenza virus infections are serious public health concerns throughout the world. The development of compounds with novel mechanisms of action is urgently required due to the emergence of viruses with resistance to the currently-approved anti-influenza viral drugs. We performed in silico screening using a structure-based drug discovery algorithm called Nagasaki University Docking Engine (NUDE), which is optimised for a GPU-based supercomputer (DEstination for Gpu Intensive MAchine; DEGIMA), by targeting influenza viral PA protein. The compounds selected by NUDE were tested for anti-influenza virus activity using a cell-based assay. The most potent compound, designated as PA-49, is a medium-sized quinolinone derivative bearing a tetrazole moiety, and it inhibited the replication of influenza virus A/WSN/33 at a half maximal inhibitory concentration of 0.47 μM. PA-49 has the ability to bind PA and its anti-influenza activity was promising against various influenza strains, including a clinical isolate of A(H1N1)pdm09 and type B viruses. The docking simulation suggested that PA-49 interrupts the PA–PB1 interface where important amino acids are mostly conserved in the virus strains tested, suggesting the strain independent utility. Because our NUDE/DEGIMA system is rapid and efficient, it may help effective drug discovery against the influenza virus and other emerging viruses.

(required for the initiation of viral transcription), respectively, whereas PA has multiple roles, such as cleavage of the capped RNA 11 and serine proteinase activities 12 . Lack of PA protein expression in a reverse genetics system led to no recovery of recombinant influenza virus 13 , thereby suggesting that PA has crucial roles in the virus life cycle 14 . PA interacts with PB1 and PB1 interacts with PB2 according to crystal structure analyses [15][16][17] . The C-terminus of PA (residues 239-716) interacts with the N-terminus (residues 1-25) of PB1. According to this model, the binding between PA and PB1 involves hydrogen bonds and hydrophobic contacts, where the N-terminal residues of PB1 are inserted in the pocket of PA 15 . To explore PA-targeting inhibitors, various techniques such as NMR method 18 , phylogenetic analysis 19 and high-throughput ELISA-based screening 20,21 have been reported.
The crystal structure information accumulated for various proteins allows us to design new antivirals by structure-based drug design (SBDD). SBDD has been used widely for the development of antiviral drugs, such as the HIV-1 proteinase inhibitor, nelfinavir 22 and influenza viral NA inhibitor zanamivir 23 . Recent advances in SBDD have largely been made due to the development of computer system and docking simulation software. To facilitate the identification of novel drugs, we recently developed the Nagasaki University Docking Engine (NUDE) which efficiently runs on the DEstination for Gpu Intensive MAchine (DEGIMA) supercomputer. Using this system, we have succeeded in development of anti-prion compounds 24 and anti-influenza virus compounds that target influenza viral NP 25 . In the present study, we performed in silico screening of approximately 600,000 molecular structures with NUDE to identify compounds that can bind the pocket in PA bound by PB1. We found that a quinolinone derivative bearing a tetrazole moiety effectively suppressed the replication of influenza viruses.

In silico and in vitro screening of anti-influenza virus compounds.
To obtain potent anti-influenza compounds we targeted the influenza viral polymerase subunit PA-PB1 interaction, which is essential for the viral life cycle 15 . Approximately 600,000 compounds in the chemical compound library were ranked based on the docking score obtained by the NUDE, and the top 136 compounds designated as PA-1 to PA-136 were selected as candidates of anti-influenza compounds. Among these compounds, 99 commercially available compounds were purchased and then subjected to cell-based screening using crystal violet (CV) assays, which reflected the virus infection-induced cytopathic effects (CPE) in cells ( Supplementary Fig. S1). In this assay, when cells were infected with the virus, the cells detached from the bottom of the dish due to the appearance of CPE. We performed this assay using serially diluted compounds. The morphology of cells was observed under a microscope and the minimum inhibitory concentration (MIC) was estimated visually after CV staining. Fig. 1 summarises the structures of 14 compounds with MIC values less than 20 μM, where 11/14 compounds were quinolinone derivatives bearing a tetrazole moiety (Fig. 1a). PA-49 (ranking = 49 th , docking score = −336.77) had the best MIC value of 1.7 μM but no significant cytotoxicity ( Supplementary Fig. S2). Therefore, PA-49 was selected as representative of the quinolinone derivatives. Other different compounds (Fig. 1b) were also selected for further experiments (a 4-aza-9-fluorenone derivative (PA-37, MIC value = 1.7 μM, ranking = 37 th , docking score = −339.311), tetracyclic compound (PA-58, MIC value = 16.6 μM, ranking = 58 th , docking score = −336.051), and 8-azahypoxanthin derivative (PA-107, MIC value = 16.6 μM, ranking = 107 th , docking score = −331.554)). These results suggest that our in silico programme, NUDE 24 , is a very useful tool for screening hit compounds.
Evaluating the binding between hit compounds and PA protein using surface plasmon resonance (SPR) analysis. SPR analyses were performed to analyse the binding affinities of the hit compounds with PA protein (Fig. 2). Anti-PA antibody was immobilised on the sensor chip, and then captured approximately 2000 RU of the purified recombinant PA protein (residues 239-716). Signals obtained from the sensor chip-immobilised anti-PA antibody alone were subtracted as the background. Benzbromarone, a uricosuric agent, has been reported to bind the PB1-binding pocket of PA 26 , so we used benzbromarone as a positive control. In the presence of the compound the binding response was observed in a dose-dependent manner, suggesting that recombinant PA protein was suitable for evaluating the SPR analyses. PA-37, 49, 58 and 107 bound to PA protein in a dose-dependent manner (Fig. 2a). Using a 1:1 binding model, the dissociation constant (K D ) was calculated from the fitting curve (Fig. 2b). The hit compounds had various K D values (Table 1). PA-37 differed in terms of the MIC and K D values, which suggests that PA-37 may have a target other than PA. PA-58 had good K D values but significant toxicity ( Supplementary Fig. S1). The binding affinity of PA-107 to PA was relatively weak. By contrast, PA-49 was the most potent compound with good MIC and K D values of 1.7 μM and 7.5 μM, respectively. Therefore, we selected PA-49 for further analyses.
PA-49 suppresses nuclear accumulation of PA. We performed a nuclear transportation-inhibition assay to confirm whether subcellular localisation of PA changes in the presence of PA-49 or not ( Fig. 3 and Supplementary Fig. S3). It has been reported that the PB1 subunit is required for efficient nuclear accumulation of the PA subunit 27 . In Fig. 3, the experiment showed that in the absence of PA-49, PA dominantly localised in the nucleus where PB1 exists, and addition of PA-49 to PA-and PB1-cotransfected cells did change localisation of PA to the cytoplasm. These results suggest that PA-49 interfere with the PA-PB1 interaction.

PA-49 suppresses influenza virus propagation.
We performed screening to select anti-influenza compounds based on CV assays ( Supplementary Fig. S1) because CV staining is an alternative and rapid method for evaluating cytotoxicity and antiviral activities 28,29 . To confirm the correlations between the CV assays and antiviral activities, we compared the antiviral activity of PA-49 based on cell morphology, water-soluble tetrazolium salt (WST-1), CV staining and 50% tissue culture infective dose (TCID 50 ) assays (Fig. 4). When Madin-Darby canine kidney (MDCK) cells were infected with the influenza virus without PA-49, the cells detached from the bottom of the well (Fig. 4a, panel E). In the presence of 0.5 μM PA-49, the cells were partially attached (panel F), and the cell condition among the infected cells treated with 0.8 and 2 μM PA-49 are almost the same with uninfected control (panel A). The viral titre in the supernatant (Fig. 4b) was dose-dependent and it decreased dramatically (2.2 × 10 4 TCID 50 /mL) by nearly 1/1000 th compared to the control after treatment with 2 μM PA-49 where the cell viability was good according to WST-1 measurements. The viral titres in the presence of 20 and 100 μM oseltamivir were measured as controls that were 2.1 × 10 5 ± 1.5 × 10 5 and 2.7 × 10 5 ± 0.7 × 10 5 TCID 50 /mL, respectively ( Supplementary Fig. S4). These results suggest that PA-49 has potent antiviral activity. The viral titre appeared to be highly correlated with the relative optical density (OD) values determined in the CV assays. Therefore, we decided to use the CV assays for further evaluations of the 50% effective concentration (EC 50 ) values.

Determination of IC 50 by plaque inhibition assay.
Determining EC 50 values using the CV assay is a feasible and rapid method for evaluating many samples. We also evaluated the antiviral activity of PA-49 using a plaque inhibition assay, which is the gold standard method for evaluating antiviral activity 30 , in order to determine whether plaque formation in MDCK cells was suppressed by PA-49 treatment. MDCK cells in six-well plates were infected with approximately 200 plaque-forming units (pfu) of virus and the cells were then overlaid with agarose solution containing various amounts of PA-49. The number of plaques formed decreased in a dose-dependent manner ( Fig. 5a and b), where the 50% inhibitory concentration (IC 50 ) determined by the plaque inhibition assay was 0.47 ± 0.13 μM (Table 1). These results suggest that PA-49 effectively suppressed influenza virus propagation in the cell culture.    analysed by western blotting (Fig. 6b and c). At 9 hpi, in the presence of PA-49, the band intensity was reduced in a dose-dependent manner. These results suggest that PA-49 suppresses influenza virus replication by inhibiting viral RNA synthesis.

PA-49 affects virus replication but not virus particles, cells and adsorption. PA-49 may inhibit
the formation of viral polymerase PA-PB1 subunit, but PA-49 might also affect other parts of the virus life cycle, such as virus attachment or entry into cells. Thus, time-of-addition assays were performed to confirm the mode of action for PA-49 (Fig. 7a). The pre-treatment of cells and pre-treatment of virus with 1 and 2 μM PA-49 had no effects on the relative plaque numbers, thereby suggesting that PA-49 does not bind to a cellular receptor and that it lacks a virucidal activity. However, a significant reduction (over 99%) in the relative plaque number was observed when 2 μM of PA-49 was added to the agarose solution ('in agarose'). Thus, PA-49 had significant antiviral activity against influenza virus when replication cycle occurs. These results were also supported by RT-PCR analysis ( Supplementary Fig. S7). To exclude the possibility that PA-49 inhibits viral binding to the cellular receptor, a hemagglutination inhibition assay was performed using chicken red blood cells (RBCs) (Fig. 7b).
As expected, hemagglutination occurred when RBCs were mixed with influenza virus (Fig. 7b, no drug), where the HA titres for A/WSN/33(H1N1) and A/Aichi/2/68(H3N2) were 320 and 160 HA units/mL, respectively. The presence of PA-49 (1-20 μM) did not affect the hemagglutination reaction. These results suggest that PA-49 has no effect during the binding/entry step of the infection process.

Antiviral effect of PA-49 on various influenza viruses. Antiviral effect of PA-49 on various influenza
viruses were tested using the CV assay ( Table 2). The EC 50 value of oseltamivir against A/WSN/33 was found to be similar to that reported previously 29 . When PA-49 was used in the analysis, the EC 50 values against A/Virginia/  Tables 1 and 2). We also tested the antiviral activity of PA-49 against several negative-stranded RNA viruses that have RdRp including severe fever with thrombocytopenia syndrome   31,32 . RdRp of these viruses is a single polypeptide called L protein, whereas that of influenza virus is composed of subunits of PA, PB1 and PB2. As expected, PA-49 did not show significant antiviral activity against these RNA viruses ( Supplementary Fig. S5b). These results suggest that PA-49 can specifically inhibit the replication of influenza viruses types A and B.

Discussion
To explore novel anti-influenza compounds using SBDD, focusing on the PA-PB1 interaction which co-crystal structure of PA and PB1 has also been reported 15 , we screened for compounds that targeted the major pocket in PA binding PB1 and inhibits the replication of influenza virus. We expanded our in-house library to approximately 600,000 compounds, and the top 99 compounds were then subjected to in vitro assays. Among that, the activity of PA-49 was promising against various influenza viruses, including types A and B (Table 2). Our plaque inhibition and CV assays suggested that PA-49 inhibits viral replication (Fig. 5 and Table 2). At 9 hpi, the expression levels of late viral proteins M1 and HA, which are synthesised from progeny vRNP, were effectively suppressed compared with those of early viral proteins PA, PB1 and NP, which are synthesised from incoming vRNP. PA-49 could suppress the PA-PB1 interaction (Fig. 3) when newly synthesised vRNP assembly occurred, whereas the incoming vRNP had already formed the PA-PB1 complex. The effect of PA-49 on the expression levels of viral proteins (Fig. 6b) was relatively weak compared with that on the viral titre in the supernatant (Fig. 4b), probably because the appropriate ratio of viral proteins must be expressed for efficient virion formation 13 (Table 2), which suggests that PA-49 has a broad spectrum of anti-influenza virus activities. In general, the amino acid sequence of PA is highly conserved. It should be noted that the amino acid residues required to form the hydrogen bonds and hydrophobic contacts across the PA-PB1 interface 15 are also conserved among various influenza A virus strains, including the H1N1, H3N2 and H5N1 subtypes (Fig. 8c). The docking simulation results suggested that PA-49 located in the centre of PB1-binding pocket of PA ( Fig. 8a and b). The important amino acid residues for PA-PB1 interface are conserved in the virus strains tested (Fig. 8c), thereby suggesting the possible utility of PA-49 as an anti-influenza viral drug. To clarify the binding mode of PA-49, further analysis is needed based on structural analysis of the protein-compound complex by NMR and more accurate simulations including molecular dynamics simulation or fragment molecular orbital calculation.
Influenza A (H1N1)pdm09, H3N2 strain and two type B viruses are currently used for vaccine production, but it is likely that these viruses rapidly acquired resistance to oseltamivir and other NA inhibitors. The amino acid sequence of PA is also conserved in the oseltamivir-resistant clinical isolate of A(H1N1)pdm09 virus A/ Quebec/147365/2009(H1N1) strain (Fig. 8c), which has an H275Y mutation in NA 33 , thereby suggesting that PA-49 may be effective against oseltamivir-resistant clinical isolates. Recently, we reported that a novel influenza virus inhibitor NUD-1 can suppress influenza virus replication by targeting the influenza virus NP 25 , but it does not inhibit the replication of type B viruses. By contrast, the inhibitory activity of PA-49 was observed against type B virus with EC 50 values less than 1 μM (Table 2).
Interestingly, many quinolinone compounds like PA-49 were selected by our screening analysis. Quinolinone is a major scaffold and useful pharmacophore in many drugs 34 , such as the bronchodilator procaterol 35 , antithrombotic agent cilostazol 36 and antipsychotic agents aripiprazole 37 . In addition, a similar monomethyl quinolinone with a tetrazole group has been reported as a novel tyrosine phosphatase inhibitor 38 . Eleven hit compounds with anti-influenza activity (MIC values less than 20 μM) had quinolinone moiety (Fig. 1). Other hit compounds that do not have this moiety (PA-37, 58 and 107) showed weak binding properties to PA protein ( Table 1). The important role of this moiety on the binding to PA should be elucidated.
According to our results, among the 99 commercially available compounds that we screened, 88 compounds had molecular weights more than 500 (ranging from 303.2-596.7, median = 532.7). Because the PA cavity for PB1 binding is relatively wide (Fig. 8a and b), it is possible that a fairly large molecule is necessary to specifically inhibit (with high affinity) PB1 binding to the PA cavity. It has been suggested that lead compounds with molecular weights of less than 500 conform to Lipinski's rule of five for drug-likeness 39 . Indeed, previously reported inhibitors targeting PA-PB1 interaction 26,40,41 conform to this rule. However, recent advances in the development of protein-protein interaction inhibitors have demonstrated the suitability of medium-sized compounds with molecular weights of more than 500 42 . Recently, medium-sized compounds such as the antitumor agent eribulin (molecular weight = 729.9) 43 and the BCL-2 inhibitor venetoclax (molecular weight = 868.4) for chronic lymphocytic leukaemia with 17p deletion 44 have been approved.
In summary, we efficiently identified potent anti-influenza virus compounds that bind to PA by SBDD. PA-49 could be a lead compound in the development of novel anti-influenza drugs. Optimisation of the structure of PA-49 to increase its anti-influenza activity and in vivo experiments are currently in progress. Focusing on the inhibition of viral protein-protein interactions will facilitate the development of antiviral drugs and our strategy using NUDE/DEGIMA system may be applicable in emergent drug screening for many other fatal RNA viruses.

Methods
In silico screening. In silico screening used an original docking simulation programme NUDE, which was designed to run on a GPU system 24,25 . The algorithm employed by NUDE is based on evolutionary Monte Carlo (EMC) techniques 45 and an empirical free energy model is used to evaluate the binding energy of a ligand. We used an original chemical compound library comprising approximately 600,000 compounds. For each compound, a maximum of 50 conformations were generated using Open Babel software 46 , which were followed by energy minimisation with the GAFF force field 47 . In our docking simulation, the X-ray structure of influenza A virus PA was downloaded from the Protein Data Bank (PDB code: 2ZNL) 15 as the receptor and a cubic space that included PB1-binding pocket of PA was employed as the search region. The docking simulation was performed using the NUDE/DEGIMA system.

Chemicals, cells, viruses and antibodies.
Hit compounds were purchased from Namiki Syoji Co. Ltd (Tokyo, Japan) and dissolved in 100% dimethyl sulfoxide (DMSO). Benzbromarone was purchased from Tokyo Chemical Industry Co. Ltd (Tokyo, Japan) and dissolved in 100% DMSO. Oseltamivir phosphate purchased from F. Hoffmann-La Roche Ltd (Basel, Switzerland) was dissolved in phosphate-buffered saline (PBS) at a concentration of 10 mM. All of the compounds were maintained at −30 °C until use. Before performing the experiments, the compounds were diluted with Eagle's minimum essential medium (MEM) supplemented with 1% 100× vitamin solution (MEM vitamin). HeLa cells were obtained from Dr Takujiro Homma (Nagasaki University, Japan) and maintained in Dulbecco's modified Eagle's medium (Sigma Aldrich, St Louis, MO) containing 10% fetal bovine serum (FBS). MDCK cells kindly donated by Dr Kyosuke Nagata (Tsukuba University, Japan) were grown in MEM supplemented with 5% FBS. These cells were maintained at 37 °C in an atmosphere of 5% CO 2 . Influenza viruses were prepared as described previously 48   Hydrogen atoms are omitted for clarity. In order to obtain binding structure of PA-PA-49 complex, hydrogen atoms were added to the complex structure obtained from the docking simulation, followed by 1,000 steps of energy minimisation without any restraint using AMBER 10 software package 51 . For minimisation, AMBER ff99SB 52 and GAFF 47 force fields were used for the protein and ligand, respectively. Colour code for atom: N (blue), O (red), S (yellow) and F (light green). (c) Amino acid sequence alignment of influenza virus PA protein (residues 601-716) including H1N1, H3N2 and H5N1 subtypes. Residues that are required for PA-PB1 interface are labelled with yellow. Molecular structures in (a) and (b) were drawn using UCSF Chimera package 53 .
(5′-GGATCCCTATCTCAATGCATGTGTGAGGAA-3′). The PCR products were subcloned into T-vector pMD19. The resultant plasmid was digested with NdeI and BamHI, and the fragment was subcloned into the NdeI-BamHI site of pET15b. To express the PA 259-716 protein, the pET15b-PA 259-716 plasmid was transformed into BL21(DE3) pLysS. The protein was induced at 23 °C by adding 0.5 mM IPTG at an OD 600 value of 0.7 for 5.5 h. The bacteria were disrupted by sonication in buffer containing 20 mM Tris-HCl (pH 7.9), 150 mM NaCl, 25 mM imidazole, 500 mM urea, 10 mM 2-mercaptoethanol and proteinase inhibitor cocktail (No. 03969-21, Nacalai Tesque), before purification using His60 Ni Superflow ™ Resin (Takara Bio Inc, Shiga, Japan). His-PA 259-716 was eluted with the same buffer containing 100-400 mM imidazole. The purified protein was dialysed in dialysis buffer containing 20 mM Tris-HCl (pH 7.9), 150 mM NaCl, 20% glycerol and 5 mM DTT, and then stored at −80 °C until use. To construct pCAGGS-PA and pCAGGS-PB1 plasmids, the WSN PA and PB1 genes were inserted into the multiple cloning site of pCAGGS vector, respectively. SPR analysis. Interactions between recombinant PA and compounds were evaluated by SPR using the Biacore T200 system (GE Healthcare UK Ltd, Buckinghamshire, UK), as described previously 49 with some modifications. Briefly, we immobilised the anti-PA antibody on a CM5 sensor chip (GE Healthcare, BR-100530) using an amine coupling kit (GE Healthcare, BR-1000-50) and running buffer (10 mM HEPES pH 7.4, 150 mM NaCl and 0.1% Tween 20 [Sigma Aldrich]). The amount of immobilised antibodies was ~8,000 RU. Next, the recombinant PA protein were loaded at a flow rate of 30 μL/min for capture by the antibodies on the sensor chip. The amount of PA protein captured reached ~2,000 RU. Compounds were analysed using the running buffer containing 5% DMSO. Data were corrected by using the anti-PA antibody immobilised sensor chip as a control. EC 50 and CC 50 evaluations in CV assays. The anti-influenza virus activities of compounds were evaluated as described previously 48 with some modifications. To evaluate the anti-influenza virus activities, MDCK cells were seeded into 96-well plates at a density of 3.0 × 10 4 cells/well in 100 μL of MEM containing 5% FBS, and then incubated overnight. Cells were washed with MEM vitamin and 100 μL of the serially diluted compound was then added. Cells were subsequently infected without (determination of CC 50 ) or with (determination of EC 50 ) 100 μL of virus solution in MEM vitamin equivalent to 100 TCID 50 for type A viruses or 30 TCID 50 for B/Lee/40. The culture plates were incubated at 37 °C for 48 h. After incubation, cells were fixed with 70% EtOH and stained with 0.5% CV. After washing with water and air drying, the absorbance was measured at 560 nm using an Infinite M200 pro plate reader (Tecan Japan Co. Ltd, Kanagawa, Japan). To determine the yield of influenza virus, the supernatant from each well was collected and the virus yield was determined using the TCID 50 assay. EC 50  WST-1 assay. The WST-1 assay was performed as previously described 29 with some modifications. Briefly, MDCK cells were seeded in 24-well plates at a density of 1.8 × 10 5 cells/well in MEM containing 5% FBS and incubated overnight. Cells were infected with the influenza virus in the presence of PA-49 and the culture plates were then incubated at 37 °C for 46 h. The medium was replaced with 0.6 mL of MEM vitamin containing 15 μL of Cell Proliferation Reagent WST-1 (Roche) and then incubated at 37 °C for 1 h. The absorbance was measured at 450-650 nm using the plate reader. The viability of cells was determined as described above. The plates were subsequently fixed and stained with CV as described above.