First, the antiviral activities of the candidate compounds against the influenza A/Puerto Rico/8/1934 (PR/8, H1N1) virus were determined in Madin–Darby canine kidney (MDCK) cells with cytopathic effect (CPE) inhibition assays. Oseltamivir carboxylate (OC) and VX-787 were used as the positive controls. Eight compounds (11C5, 11C6, 11C8, 11D2, 11D4, 12B1, 12C5, and 13D7) with novel scaffolds were found to rescue cells from virus-induced cytopathic effects at non-cytotoxic concentrations (Figure 2). The EC₅₀ values of these compounds ranged from 0.30 to 55.50 μM (Table 1). The EC₅₀ value of the most active compound, 11D-2, was 0.30 ± 0.02 μM, which was much higher than that of VX-787 (<0.005 μM) but superior to that of OC (1.78 ± 0.72 μM). To expand the range of active compounds, compounds structurally similar to 11D2, 11D4, and 12C5 were searched in the ChemBridge database and virtually screened, and 11 compounds were selected for activity evaluation (Figure 1 and Figure S2). Another two compounds, 21B1 and 21A5, were found to rescue cells from virus-induced cytopathic effects at non-cytotoxic concentrations (Figure 2 and Table 1). Further evaluation showed that four compounds, namely, 21B1, 11D4, 12C5, and 21A5, had a broad-spectrum antiviral effect on various subtypes of influenza viruses. Except for compound 21B1 showing no activity against the A/Hong Kong/8/68 (HK/68, H3N2) strain, these compounds were active against a variety of influenza A virus strains, including HK/68 (H3N2), A/WSN/33 (H1N1), A/LiaoNing-ZhenXing/1109/2010 (ZX/1109, H1N1, natural isolate oseltamivir-resistant), the PR/8-R292K mutant (H1N1, recombinant oseltamivir-resistant), the PR/8-I38T mutant (H1N1, recombinant baloxavir-resistant), and influenza B/Lee/40 virus strains (Table 1). The EC₅₀ of these compounds ranged from 0.30 to 67.65 μM. It was not surprising that the oseltamivir- and baloxavir-resistant strains were sensitive to our compounds, because our compounds have different targets other than oseltamivir and baloxavir, and thus mutations of amino acid residues in the target sites of oseltamivir and baloxavir binding have no effect on these compounds. However, our compounds had similar inhibitory effects on various wild strains of influenza A and influenza B, which was unexpected, because the sequence identity between influenza A and influenza B is only around 34% (Figure S3). In addition, according to literature reports, the activity of VX-787 against influenza A viruses and influenza B viruses differs greatly [26]. The cytotoxicity of these compounds was determined in MDCK cells, and the cell cytotoxicity (CC₅₀) observed for all 10 compounds was >100 µM (Table 2).

To confirm that these 10 active compounds are competitive ligands of PB2 CBD, biomolecular interaction analysis based on a surface plasmon resonance (SPR) competitive binding assay was conducted. Consistent with expectations based on previous molecular docking experiments, moderate binding affinities with equilibrium dissociation constants (K_d) ranging from 0.5 to 6.7 μM were found between these compounds and influenza PB2 CBD (Table 1). Compounds 11D2 and 21A5 exhibited a concentration-dependent association–dissociation pattern with a K_d of 0.21 and 0.54 μM, respectively. There was approximately an order of magnitude difference in the K_d compared with that of the control compound, VX-787 (0.054 μM).

In short, through a virtual screening and similarity search, we selected a total of 60 compounds for antiviral activity evaluation. The acceptance rate of virtual screening is about 0.0106% (Figure 1). Ten compounds were found to have obvious inhibitory effects on the H1N1 influenza virus, accounting for 16.7% of the 60 tested compounds, with 4 of these compounds exerting broad-spectrum antiviral effects on multiple influenza strains.

The X-ray crystal structure of VX-787 bound to PB2 (PDB: 4P1U) confirmed that VX-787 and PB2 are tightly bound. The azaindole ring system interacts with the protein residues Lys376 and Glu361, and the aromatic rings of VX-787 stack between the side chains of His357, Phe323, Phe363, and Phe404 to form a typical sandwich structure with His357 and Phe404. The carboxylic group of the ligand participates in two water-mediated interactions with the ε nitrogen of His357 and Gln406, as well as the main chain carbonyl of Arg355. Additional salt bridge interactions with Arg355 in the adjacent region were also observed [17]. Interaction fingerprints analysis showed that the allylpyrazole occupied the hydrophobic cavity formed by residues Phe323, Phe325, Phe404, and Met431. In addition, the methylenedioxybenzene fragment interacts with Phe323 and Phe363 through hydrophobicity (Figure S3). This is consistent with the high affinity of VX-787 toward PB2 and its significant antiviral activity.

Our 10 experimentally validated compounds were significantly different in structure from VX-787. From the analysis of the predicted optimal binding modes of these compounds obtained from the docking results, we found that our active molecules bound deep within the PB2 CBD, similar to VX-787 and without high strain energy (Figure 3). To investigate the binding models of our active compounds in detail, the docking models of four representative compounds, 11D4, 12C5, 21A5, and 21B1, were optimized using the Refine Protein–Ligand Complex tool of Prime. The results showed that the binding modes of these compounds were somewhat similar to that of VX-787, but there were also many unique features (Figure 3, Figure 4 and Figure S6). First, similar to VX-787, compounds 12C5, 11D4, 21A5, and 21B1 all exhibited π–π stacking between their aromatic ring and the imidazole ring of His357. At the same time, these compounds were partially involved in π–π stacking interactions with the side-chain phenyl rings of Phe363, Phe323, and Phe404. Second, another similarity to VX-787 is that all four active compounds maintained hydrogen bonding with Lys376, whereas compound 12C5 simultaneously maintained the hydrogen bond with Glu361. Third, interaction fingerprint analysis showed that our compounds also retained the hydrophobic effects of the side chains of Phe323, Phe325, Met431, Phe404, and Phe363, similar to that of VX-787 (Figure S3).

Since our compounds are structurally distinct from VX-787 even though they also bind to the CBD site, they exhibited several completely different additional protein-binding modes from that of VX-787. First of all, our compounds had a different π–π stacking effect and cationic–π interactions from those of VX-787. Compound 11D4 had no π–π stacking with the side-chain benzene ring of Phe404 as VX-787 did, while a cationic–π interaction was observed between Phe404 and the amino cation of compound 11D4. For compound 21B1, no π–π stacking with the side-chain benzene ring of Phe323 and Phe363 was observed, while there was a cationic–π interaction between the guanidino group of Arg332 and the thiophene ring, and π–π stacking between the imidazole of His432 and picoline was also found. In addition, the hydrogen bond between our compound and PB2 was markedly different from that of VX-787. For compound 21A5, the pyrazole formed a hydrogen bond with the side chain of Ser321. For compound 12C5, the carboxyl group formed two H-bonds with Arg332 and Ser321. The unsubstituted nitrogen atom of allylpyrazole formed a hydrogen bond with the amide hydrogen atom of Asn429 (Figure 4). Similar to compound 12C5, compounds 11D4 and 21B1 maintained hydrogen bonding with Asn429. Besides, the carboxyl group of 12C5 also formed a salt bridge with the protonated guanidino group of the Arg332 side chain, which is absent in VX-787. One of the most significant differences between our compounds and VX-787 is that our compounds lack an acidic group to form a salt bridge with Arg355, and they do not interact with His357 and Gln406 via water molecules.

There is still a large gap between the anti-influenza activities of the compounds we obtained and that of VX-787. This is because the structures have not been optimized; thus, the compatibility of these compounds as ligands of PB2 is still far from that of VX-787. First, the azaindole ring of VX-787 forms a typical sandwich structure with His357 and Phe404, but none of our compounds (except 21B1) can form such a perfect sandwich structure. Second, the bicyclo[2.2.2]octane of VX-787 is a large hydrophobic group that has considerable contact with the hydrophobic cavity of PB2, while none of our compounds have a sufficiently large hydrophobic group. Third, our compounds have no acidic group and can neither form salt bridges and hydrogen bonds with Arg355 guanidine nor form hydrogen bond networks with His357 and Gln406 through water molecules.

Although the antiviral activities of our compound are still far behind that of VX-787, our compounds have novel chemical scaffolds and relatively small molecular weights and, thus, are suitable for optimization as lead compounds. By optimizing and modifying our compounds, such as through adding aromatic rings or adjusting the position of the aromatic rings to increase or strengthen the π–π stacking interactions, and introducing large hydrophobic groups and an acidic group to interact with the Arg355 guanidine, we believe that it will result in more active compounds.

It is reported in the literature that VX-787 has a significant inhibitory effect on influenza B virus 23. In contrast, our compounds, 11D4, 12C5, 21A5, and 21B1, had similar inhibitory effects on various strains of influenza A and influenza B. To explore the reasons why our compounds exhibited different selectivity toward influenza A and influenza B compared with VX-787, we compared the sequences and structures of the PB2 CBD site of several wild strains of influenza. The protein sequence of PDB 4P1U (PB2 CBD of A/Victoria/3/1975 (H3N2) bound to VX-787) was directly extracted from the PDB file. The sequences of PB2 of the human influenza A/WSN/1933 (H1N1), ZX/1109 (H1N1), A/Hong Kong/16/1968 (H3N2), and B/Lee/40 virus strains were downloaded from the Influenza Virus Resource (http://www.ncbi.nlm.nih.gov/genomes/FLU/FLU.html) and Global Initiative on Sharing Avian Influenza Data (http://platform.gisaid.org/) databases. These protein sequences were aligned, and the similarity between the amino acid sequences in the alignments was calculated using the Align Sequence protocol in DS 3.0 (Figure S5). The identity between the PB2 CBD sequence of the A/Victoria/3/1975 (H3N2) virus and that of the A/Hong Kong/16/1968 (H3N2) virus reached 97.5%. The identity between the PB2 CBD sequence of A/Victoria/3/1975 (H3N2) and that of ZX1109 (H1N1) and A/WSN/1933 (H1N1) was also high at 92.0% and 94.5%, respectively. This suggests that the PB2 CBD is a highly conserved site in the influenza A virus. In particular, some key amino acid residues, such as Ser321, Phe323, Phe325, Arg332, His357, Glu361, Phe363, Lys376, Phe404, Gln406, Asn429, and Met432, are highly conserved (Figure S5). This fact may well explain the broad-spectrum antiviral activity of our compounds and VX-787 against various influenza A viruses.

In contrast, the PB2 CB sequence identity between the A/Victoria/3/1975 (H3N2) and the B/Lee/40 virus strains is only 35.0%, indicating that the PB2 CBD sequences of influenza A and influenza B are significantly different (Figure S4). The 3D structure overlay results of PDB 4P1U and PDB 5EFA (B/Lee/40 PB2 CBD bound to M7GTP) showed that their structural similarity is higher than their sequence similarity (Figure 5). Analysis of these two protein sequences showed that several key amino acid residues, such as Phe325/327, Arg332/334, Glu361/363, Phe363/365, Lys376/378, Phe404/406, Gln406/408, and Asn429/431 (amino acid residue numbers of PDB 4P1U/PDB 5EFA, respectively), are conserved in both type A and type B influenza viruses (Figure S5). However, some key amino acids, including Phe323/Gln325, His357/Trp359, His432/Tyr434, Asn429/Ser431, and Arg355/Gly357, are mutated. Compounds 21B1, 11D4, 12C5, 21A5, and VX-787 participate in π–π stacking with His357. Careful observation suggests that the His357/Trp359 mutation would not have an adverse effect on the activity of PB2 CBD inhibitors, because the indole ring system of Trp residues can also form a π–π stack similar to the imidazole ring of His, and as the indole ring system is larger than the imidazole ring, this π–π stack should be enhanced. On the other hand, the mutation from Phe323 to Gln325 would destroy a key π–π stack; hence, this mutation should have an adverse effect on the activities of compounds that directly interact with it, including VX-787, 21B1, 11D4, and 21A5. However, the experimental results indicate that this variation does not exert much effect on the inhibitory activities of these compounds toward influenza A and influenza B. This may be because the loss of binding energy caused by this mutation is compensated to a certain extent by the enhancement of the π–π stack effect by the His357/Trp359 mutation. The antiviral activities of 11D4, 12C5, 21A5, 21B1, and VX-787 should be little affected by the His432/Tyr434 mutation, for only compound 21B1 has a direct interaction (π–π stacking) with this residue, and the mutated Tyr434 can maintain a similar interaction. There is no direct interaction between VX-787 and Asn429, so the Asn429/Ser431 mutation should have little effect on the activity of VX-787. Among our compounds, 11D4, 12C5, and 21B1 have hydrogen bonds with Asn429. However, the experimental results show that this variation does not have much effect on the inhibitory activities of these compounds on influenza A and influenza B. The Arg355/Gly327 mutation should be key, because the carboxyl group that interacts directly with this amino acid residue plays a key role in maintaining VX-787 activity [17], and this mutation would destroy a key salt bridge and two hydrogen bonds (Figure 2, Figure 3 and Figure 4). None of our compounds contain an acid group that can interact with Arg355 as VX-787 does, and this may be the most critical reason why our compounds are not sensitive to the PB2 CBD sequence variation of influenza A and influenza B and have similar inhibitory activities toward both. At the same time, this may be a key reason why the anti-influenza virus activities of our compounds are inferior to that of VX-787.

Based on the above analysis and visual observation of Figure 5, the three residue mutations that have a greater impact on the binding abilities of influenza PB2 CBD inhibitors should be Phe323/Gln325, Asn429/Ser431, and Arg355/Gly357. When attempting to obtain compounds with high activities against both influenza A and influenza B, it is critical that the impact of these mutations be considered fully, particularly the impact of Arg355/Gly357.

The Madin–Darby canine kidney (MDCK) cell line was purchased from the American Type Culture Collection (ATCC, Manassas, VA, USA) and maintained at 37 °C in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 10% fetal bovine serum, 100 μg/mL streptomycin, and 100 U/mL penicillin. Influenza A PR/8 (H1N1), ZX/1109 (H1N1), the PR/8-R292K mutant (H1N1, oseltamivir-resistant, recombinant strain), the PR/8-I38T mutant (H1N1, baloxavir-resistant, recombinant strain), A/WSN/33 (H1N1), HK/68 (H3N2), and influenza B/Lee/40 were propagated in 8- to 10-day-old embryonated chicken eggs or MDCK cells for 3 days at 37 °C. The virus strains were stored in our laboratory.

MDCK cells were seeded and grown for 18–24 h to a confluent monolayer in a 96-well plate. The medium was exchanged with DF-12 medium containing 2 μg/mL TPCK-trypsin after washing twice with phosphate-buffered saline (PBS). Cells were infected with each influenza virus strain at a multiplicity of infection (MOI) of 0.005 in DF-12 medium containing 2 μg/mL TPCK-trypsin in the presence of various concentrations (ranging from 0.005 to 100 μM by three-fold dilution) of each test compound. After a 72 h incubation at 37 °C in a CO₂ incubator, the antiviral activity of the test compounds was measured using the CellTiter-Glo viability assay (Promega, Madison, WI, USA). The concentration required for 50% maximal effect (EC₅₀) was calculated using Origin 8 software.

The cytotoxicity of the compounds was evaluated in MDCK cells using the CellTiter-Glo viability assay according to the manufacturer’s instructions. Briefly, cells were seeded at a density of 1.5 × 10⁴ per well into 96-well plates and grown for 18–24 h to a confluent monolayer. The medium was exchanged with DF-12 medium containing 2 μg/mL TPCK-trypsin, and the test compounds were added to the cells in a three-fold dilution series. DMSO was added for the control. After a 72 h incubation at 37 °C in a CO₂ incubator, the luminescence of each well was read with a SpectraMax M5 microplate reader (Molecular Devices, San Jose, CA, USA). The 50% cytotoxicity concentration (CC₅₀) was calculated using Origin 8.

For H1N1 PB2 CBD protein expression and purification, a 507 bp region of the target gene, PB2 from influenza A/WSN/1933 (H1N1, a.a. 318–486), GenBank ID: BBB04699.1, was amplified from the plasmid, spET24-6H-JS30-2, constructed by Sino Biological, Inc. (Beijing, China), using primers 190822-F2 and 190822-R1. This 507 bp fragment was used as the template to amplify a 531 bp region using primers 190822-F3 and 190822-R1 to add a 3C protease cleavage site. Finally, the 531 bp fragment was cloned into the vector, spET24-SUMOstar, constructed by Sino Biological, Inc., via homologous recombination using the StuI and XhoI restriction sites, in frame with an N-terminal histidine and SUMO tag. The protein was expressed in Escherichia coli, and the fusion protein was captured with a Ni-IDA column. The N-terminal tags were removed by cleavage with HRV 3C protease. The digested protein mixtures were further purified with a Ni-IDA column, yielding the purified PB2 CBD protein of influenza A/WSN/1933 (H1N1, a.a. 318–486). All primer sequences are listed in the supporting information (Table S2).

H1N1 PB2 was immobilized to a CM5 sensor chip (GE Healthcare, Chicago, IL, USA) to a level of approximately 4900 response units (RUs) using a Biacore T200 (GE Healthcare) and a running buffer composed of 1× PBS-P buffer (PBS, pH 7.4, containing 0.02 M phosphate buffer, 2.7 mM KCl, 0.137 M NaCl, and 0.05% Tween 20). Serial dilutions of the compounds (VX-787, 11C5, 11C6, 11D2, 11D4, 11C8, 12B1, 12C5, 13D7, 21A5, and 21B1) were injected at concentrations ranging from 20 to 0.156 µM. The resulting data were fit to the affinity binding model using the Biacore Insight Evaluation Software 2.0.2 (GE Healthcare).