N-Benzyl 4,4-disubstituted piperidines as a potent class of influen- za H1N1 virus inhibitors showing a novel mechanism of hemagglu- tinin fusion peptide interaction

The influenza virus hemagglutinin (HA) is an attractive target for antiviral therapy due to its essential role in mediating virus entry into the host cell. We here report the identification of a class of N-benzyl-4,4,-disubstituted piperidines as influenza A virus fusion inhibitors with specific activity against the H1N1 subtype. Using the highly efficient one-step Ugi four-component reaction, diverse library of piperidine-based analogues was synthesized and evaluated to explore the structure-activity relationships (SAR). Mechanistic studies, including resistance selection with the most active compound (2) demonstrated that it acts as an inhibitor of the low pH-induced HA-mediated membrane fusion process. Computational studies identified an as yet unrecognized fusion inhibitor binding site, which is located at the bottom of the HA2 stem in close proximity to the fusion peptide. A direct π-stacking interaction between the N-benzylpiperidine moiety of 2 and F9HA2 of the fusion peptide, reinforced with an additional π-stacking interaction with Y119HA2, and a salt bridge of the protonated piperidine nitrogen with E120HA2, were identified as important interactions to mediate ligand binding. This site rationalized the observed SAR and provided a structural explanation for the H1N1-specific activity of our inhibitors. Furthermore, the HA1-S326V mutation resulting in resistance to 2 is close to the proposed new binding pocket. Our findings point to the N-benzyl-4,4,-disubstituted piperidines as an interesting class of influenza virus inhibitors, representing the first example of fusion peptide binders with great potential for anti-influenza drug development.

It was also proposed as the binding site for a class of group 1-specific aniline-based inhibitors (such as 9d in Fig. 1; [13]). On the other hand, guided by the binding mode of the anti-HA stem antibody CR6261, the conserved HA 1 -HA 2 fusion region of group 1 HAs is targeted by JNJ4796 (site B in Fig. 1; [14]), and the cyclic peptide CP141037 (not shown; [15]). Finally, MBX2329 and MBX2546 ( Fig. 1) were found to inhibit group 1 A/H1N1 and A/H5N1 strains through binding to nonoverlapping sites in the stem region of HA [16]. In this report, a novel class of piperidine-based HA fusion inhibitors is presented. The choice of a piperidine heterocycle was motivated by its pivotal role in drug design [17]. Recently reported pharmacological uses of diverse substituted piperidine derivatives include coronary heart disease, anticancer, antivirals, and antinociceptives, among others [18][19][20][21][22]. In fact, one of the most frequently used non-aromatic ring systems in small molecule drugs is the piperidine ring [23], particularly 1,4disubstituted piperidine, due to its easy synthesis and lack of stereochemical issues. A few years ago, we described [24] a structurally distinct series of 1,4,4-substituted piperidine derivatives, which were efficiently synthesized by means of the Ugi four-component reaction. These compounds of general formula I can be easily generated from amines, isocyanides, N-substituted piperidones and amino acids as ketone and carboxylic components ( Fig. 2A). Hence, starting from commercially available reagents, five points of diversity can be introduced in a one-step reaction. During broad biological evaluation, we noticed that 1 (Fig. 2B) exhibited low micromolar activity against influenza A/H1N1 virus. Starting from this hit compound, we here report the synthesis of an extended series of analogues to understand the structure-activity relationships (SAR) and explore the  antiviral mechanism of action. The most active compound, i.e. the fluorine derivative 2 (Fig. 2B), was selected for mechanistic studies including selection of resistant influenza virus mutants and influenza HA polykaryon assays. The biological findings were rationalized by in silico predictions of the binding mode within the viral HA protein, using molecular simulations. A new binding pocket that is located close to the fusion peptide at the HA 2 subunit is proposed. To the best of our knowledge, this pocket has not been so far explored by any inhibitors targeting the HA-mediated fusion process. Overall, the results show that the N-benzyl 4,4-disubstituted piperidine compounds represent a structurally promising scaffold for the design of influenza virus fusion inhibitors.

Chemistry
The Ugi four-component reaction is one of the most prominent isocyanide-based multicomponent reactions due to its versatility, atom economy and experimental simplicity, enabling the conversion of isocyanides, amines, aldehydes (ketones) and carboxylic acid into a great variety of bis-amide derivatives [25][26][27]. We previously applied this reaction for the efficient synthesis of a structurally diverse library of 1,4,4-substituted piperidine bis-amide derivatives ( Fig. 2A; [24]). Antiviral evaluation of these compounds allowed us to identify the N-benzyl 4,4-dipeptide piperidine analogue 1 (Fig. 2B) as a promising hit endowed with low micromolar activity against influenza A/PR/8/34 (A/H1N1) virus. The inhibitory activity was even 5-fold higher for the 4-fluorobenzyl analogue 2 ( Fig. 2B). Hence, we decided to synthesize a large series of piperidine-based analogues by modifying the R 1 -R 5 substituents and investigating the SAR for influenza virus.
The general synthetic route for these novel piperidine analogues is depicted in Fig. 2A. The synthesis was accomplished in a relatively easy way via the Ugi four-component reaction with moderate to good yields. Commercially available N-substituted 4-piperidone (A), isocyanides (B), aromatic and aliphatic primary amines (C) and a variety of polar, hydrophobic or aromatic natural L-amino acids as carboxylic acids (D) were allowed to react in methanol at room temperature over 72 h, followed by chromatographic purification. In this work, 27 novel piperidine analogues were synthesized, fully characterized as the corresponding free amine derivatives, and evaluated for antiviral activity. Using the Sirius T3 apparatus (Supplementary Material Fig. S1), a pKa value of 7.5 was measured for compound 2, indicating that its piperidine moiety should be positively charged at the acidic pH (~5) of late endosomes. We also evaluated the chemical stability at this acidic pH for the N-Boc or N-Cbz groups in the most active compounds 2 and 34. They proved very stable after 72 h incubation in acetonitrile:acetate acidic buffer solutions (pH=5.5), which mimic the acidic conditions in the endosomal lumen.

Analysis of anti-influenza virus activity
Compounds 1 and 2 served as the starting points for an extensive SAR exploration against influenza virus, performed in MDCK cells infected with strain A/PR/8/34 (A/H1N1). For analysis of the antiviral results (Table 1), the compounds were organized in three subseries with modifications at i) R 1 or R 2 , ii) R 3 , and iii) R 4 or R 5 .
In the first subseries, which comprises compounds retaining the 4-fluorobenzyl unit of 2, the effect of R 1 and R 2 substituents was assessed using a variety of commercially available N-substituted piperidones (A) and isocyanides (B). Elimination of the N-1-benzylpiperidine nucleus (3) resulted in complete loss of activity. The antiviral activity was significantly reduced when the R 1 benzyl was missing (4) or replaced by cyclohexyl (6) or methyl (5), and totally lost in analogues having a cyclohexyl (6), phenyl, 4-Cl-phenyl or phenylethyl at R 1 (7 and 8). Thus, an N-1-benzyl piperidine substituent at R 1 was absolutely required for activity. Also, the R 2 benzyl group proved to be critical since its replacement by cyclohexyl (9), t-butyl (10) or tosylmethyl (11) was detrimental.
In the second subseries, the influence of the R 3 substituent was investigated by varying the nature of the primary amine (C). Again, the benzyl moiety appeared a critical structural element since its elimination (12) or replacement by alkyl (14), cyclopropyl (15) or phenylaminoethyl group (16) was detrimental for the inhibitory activity, while a methyl substituent was partially (13) tolerated. Besides, various substituents with different electronic properties were introduced at the R 3 aromatic substituent. At position 4, a halogen was clearly preferred (cfr. 5-fold higher activity of 2 compared to 1), since the analogues carrying a methyl (17) or nitro group (19) were less active, while trifluoromethyl 8 (18) was detrimental. In line with the positive effect of the 4-F atom, even slightly higher potency was seen with the 4-chloro analogue 20.

Mechanism of inhibitory activity
The most active compound 2 was selected for mechanistic investigations, starting with one-cycle time-of-addition experiments. Compound addition time was varied relative to virus infection and the reduction in viral vRNA synthesis was monitored at 10 h p.i. The influenza virus entry process consists of virus binding to the cells, uptake in and release from acidic endosomes, and import of viral ribonucleoprotein (vRNP) complexes into the nucleus, in total taking about 1 h in MDCK cells [28].
Hence, the reference compound chloroquine, which acts by increasing the endosomal pH, completely lost its activity when added at 1 h p.i. (Fig. 3). For ribavirin, an inhibitor of viral RNA synthesis [29], the time-of-addition curve was situated beyond 1 h p.i. with the steepest part between 3 and 5 h p.i. Nucleozin remained fully effective when added as late as 5 h p.i., consistent with the finding that late addition of this agent blocks cytoplasmic traffic of the vRNPs after their nuclear export [30]. For 2, the curve fully overlapped with that of chloroquine, indicating that its action takes place during the endosomal stage of the virus. The low pH inside the endosomes triggers membrane fusion induced by HA, virus uncoating that requires M2 proton channel activity, and weakening of the vRNP-M1 matrix protein interactions.
To better define the target, influenza A/PR/8/34 virus was serially passaged in the absence or presence of 2. Resistant virus emerging at passage #7 was plaque-purified and virus clones were submitted to phenotypic and genotypic analysis. As shown in Table 2, the two clones selected under 2 displayed manifest (>40-fold) resistance to 2 while being fully sensitive to ribavirin. Both clones contained no changes in the M1 and M2 proteins. As for HA, substitution HA 1 -I324T is likely irrelevant since it was previously detected upon passaging of A/PR/8/34 virus in cell culture and linked to polymorphism or cell culture adaptation [32]. The resistance to 2 was thus attributed to substitutions HA 1 -S326V and HA 2 -L99F. This concurred with its HA-subtype dependent activity and, in combination with the time-of-addition data, suggested that it affects HA-mediated membrane fusion.  c pH at which 50% hemolysis occurs, relative to the value at pH 4.6. d EC 50 : 50% effective concentration, as determined by the microscopic CPE assay or MTS cell viability assay. e Parent allantoic stock of A/PR/8/34 virus, used at the start of the passage experiment. Values shown are the mean ± SEM (N ≥ 3). ND, not determined.
The precise mechanism was revealed in the polykaryon assay, which monitors cell-cell fusion that is provoked by influenza virus HA when its conformation changes at acidic pH. At a concentration of 100 µM, 2 completely inhibited polykaryon formation in H1 (A/PR/8/34) HA-transfected HeLa cells exposed to an acidic buffer of pH 5.2 (Fig. 4). This fusion-inhibiting effect was dose-dependent with a 50% inhibitory concentration (IC 50 ) value of 12 µM. The same assay was used to determine which of the three HA mutations detected in virus resistant to compound 2 (see above) are responsible for the resistance response. In HeLa cells transfected with the HA 1 -I324T mutant protein, 2 was nicely active, and the IC 50 value was even lower compared to that of wild-type HA (Fig. 4). On the other hand, 2 was totally inactive (IC 50 >100 µM) against the HA 1 -S326V and HA 2 -L99F mutants.
As demonstrated by us in a previous study [33], resistance to influenza virus fusion inhibitors can occur at two different levels. A first type of mutations, located in the inhibitor's binding pocket within HA, directly affects the HA binding capacity of the molecule. Alternatively, HA mutations that increase the fusion pH render the HA protein less stable, thereby counteracting the HA-stabilizing effect of the fusion inhibitor. This second possibility was investigated by determining the hemolysis pH of the mutant virus selected under 2. As such, cell culture passaging without compound increased the hemolysis pH by 0.3 pH units, compared to the parent A/PR/8/34 virus grown in eggs ( Table 2).
Virus that arose under 2 had the same hemolysis pH (i.e., 5.3) as the no compound control, meaning that HA stability at low pH is not affected by the combination of substitutions HA 1 -I324T, HA 1 -S326V and HA 2 -L99F. Photographs, from left to right: cells treated with 100 µM of 2 and exposed to pH 5.2; mock-treated cells exposed to pH 5.2; mock-treated cells exposed to pH 7.0. The Table   shows the IC 50 values for inhibition of polykaryon formation induced by wild-type and mutant HA proteins (mean ± SEM; N= 3).
Finally, compound 2 was further profiled by performing some additional mechanistic experiments.
The potential effect on HA-mediated virus binding to sialylated cell surface glycans was evaluated in a virus binding experiment in MDCK cells kept at 4 °C. Compound 2 proved to have no effect (data not shown) whereas the sialylated lipid compound NMSO 3 [34], at a concentration of 200 µM, produced 96% inhibition of virus binding to MDCK cells. We also investigated whether the compounds might inhibit a cellular protease associated with influenza virus entry or HA functionality.
The HA 1 -I324T and HA 1 -S326V substitutions in resistant virus obtained under compound 2 are lying in the cleavage loop of the HA0 precursor protein. In order to become fusion-competent, HA0 requires cleavage into its HA 1 and HA 2 polypeptides, by serine proteases trypsin and human airway trypsin-like protease (HAT) [35]. We therefore tested whether the compounds could possibly act by inhibiting these proteases (Supplementary Material Table S1). We included three cathepsin enzymes, i.e. cathepsin F used as the commercially available analogue of cathepsin W [36], a cysteine protease that was linked to endosomal escape of influenza virus by an as yet unknown mechanism [37], plus cathepsin B and cathepsin L because of a reported link with influenza virus replication [38]. Enzymatic experiments were carried out with the methyl ester 2 as well as the free carboxylic acid that might be released intracellularly, i.e. 30, respectively, as well as highly active compounds 31 and 34.
Neither of these molecules produced any inhibitory effect on the five proteases tested. For comparison, 1 µM of camostat gave ~90% inhibition of trypsin and HAT and the same level of inhibition was seen with 1 µM of E64 tested against cathepsin B and cathepsin L; for cathepsin F, the inhibition was ~50% (data not shown).

Structural and molecular modeling analysis
Since 2 exerts its inhibitory activity by interfering with HA-mediated membrane fusion, its potential binding to sites A and B ( Fig. 1) was explored by docking computations using Glide [39,40]. To this end, the X-ray structure of HA (H3 subtype) bound to TBHQ (PDB entry 3EYM) was used as tem- Three and one main clusters were found for sites A and B, respectively, with scores ranging from -8.2 to -7.2 Kcal mol -1 for site A and -3.8 Kcal mol -1 for site B (Supplementary Material Fig. S2). In the predicted pose for 2 (cluster 1) bound to site A, the N-1-benzylpiperidine moiety (R 1 ) forms a hydrogen bond (HB) between the protonated nitrogen and the backbone oxygen of K51 HA2 , and the benzylamide group (R 2 ) is surrounded by Leu residues at the bottom of the cavity, whereas no stabilizing interactions were observed for R 3 , R 4 and R 5 , which are exposed to the solvent outside the binding pocket. However, this binding mode cannot explain the SAR discussed above, such as the loss of inhibitory activity upon replacement of the benzyl moiety (R 1 ) by phenyl (7), or the changes in activity found upon chemical modifications of R 3 , R 4 , and R 5 . In site B, 2 partially fills the hydrophobic groove formed by the HA 1 and HA 2 subunits, as the N-1-benzylpiperidine (R 1 ) and benzylamide groups (R 2 ) partially match some residues of the cyclic peptide CP141037. Nevertheless, no significant protein-ligand stabilizing interactions were observed, as noted in the low score of the 16 pose (-3.8 Kcal mol -1 ). Furthermore, superposition of 2 with JNJ47962 revealed weak chemical resemblance in the binding motif to site B (Supplementary Material Fig. S2).
Hence, an alternative binding pocket was searched through pocket analysis using Fpocket [42]. This led to the identification of a putative binding site located at the bottom of HA in a pocket shaped by the three HA 2 helices (Fig. 5). In this pocket, 2 was anchored through several interactions (Fig. 6A), including i) a salt bridge between the protonated nitrogen of the benzylpiperidine unit (R 1 ) with E120 HA2 , and ii) notably a direct π-stacking interaction between the benzyl ring and F9 HA2 of the fusion peptide.  (Fig. 6). The salt bridge formed between the protonated piperidine of 2 and E120 HA2 is maintained in all cases, with an average N … O distance close to 3.3 Å, which is enlarged to 6.5 Å due to the insertion of a water molecule only in the last 20ns of the trajectory obtained for replica 1 (d6 in Supplementary Material Fig. S4). Regarding the π-stacking interaction between the benzylpiperidine moiety of 2 and the benzene ring of F9 HA2 , a concerted rearrangement of the aromatic rings enabled the ligand to form an additional π-stacking interaction with Y119 HA2 (Fig. 6B)  18 The 4-F-benzene (R 3 ) unit fills a cavity formed by the side chains of K116 HA2 , E120 HA2 , Y119 HA2 , and S124 HA2 ( Fig. 6; see also Supplementary Material Fig. S5 and S6). The distance of the fluorine atom in the 4-F-benzene unit to the closest atoms in these residues varies between 3.6 and 5.4 Å, suggesting a limited tolerance for accommodating larger substituents. This may explain the slight decrease in antiviral activity observed upon replacement of 4-F in 2 by 4-Me in 17 (i.e, 3.6-fold less potent), and drastic loss of activity upon substitution by 4-CF 3 in 18, as well as the similar activity observed between 2 and its 3-F analogue (i.e., a 2.1-fold increase in EC 50 ; Table 1). Furthermore, the benzene ring of R 3 is located at around 5 Å from the protonated amino group of K123 HA2 , which in turn forms a stable H-bond with the carbonyl oxygen of NH-Boc (R 4 ), as noted in average values of 2.9±1.0, 2.5±0.6 and 2.5±0.5 Å for the three replicas (Fig. 6).
The tight-packed arrangement of the ligand should shield the network of salt bridges formed between the protonated piperidine nitrogen of 2 and the carboxylate group of E120 HA2 , which in turn is stabilized by contacts with K116 HA2 , from the bulk solvent, thus reinforcing these interactions. This The key role of the N-1-benzylpiperidine moiety is also noted in the drastic decrease in inhibitory activity upon replacement of the benzyl group by cyclohexyl (6), phenyl (7) or phenylethyl (8), since these would imply a loss of the stacking interaction, as well as by the reduced activity found upon replacement by H (4) or methyl (5).
To confirm the suitability of this binding mode, compounds 38-40 were rationally designed and synthesized to determine their antiviral activity ( Table 1). The presence of a quaternary nitrogen in 40, though preserving the positive charge of the ligand should affect the salt bridge with E120 HA2 , leading to a drastic reduction in inhibitory activity, as confirmed experimentally (EC 50 >100; Table 1).
On the other hand, 38 and 39 were chosen to explore the ability of the benzyl ring attached to the piperidine nitrogen to fill the pocket between F9 HA2 and Y119 HA2 , whose size is hindered by residues A5 HA2 and V115 HA2 . Accordingly, whereas the insertion of a single F atom in meta position (38) may fill the void space between the benzene and G134 HA2 , the insertion of two F atoms (39) would be penalized by unfavorable repulsion with the lone pairs of carbonyl oxygens in V115 HA2 and K116 HA2 (Supplementary Material Fig. S4). Indeed, the biological evaluation revealed that the antiviral activity of 2 was retained in compound 38, but abolished in 39 (Table 1). this appears to be a conserved mutation, this substitution in the three monomers of trimeric HA triggers a substantial reduction in the volume of the cavity, which would affect the binding of compound 2 to site C (Fig. 7). Interestingly, the analysis of HA sequences retrieved from 3DFlu [44] for different subtypes of group 1 (H1, H2, H5, H6, H8 and H9) and group 2 (H3, H7 and H10) HA proteins ( Supplementary Material Fig. S9) revealed that the K123 HA2 R mutation is not only present in both HA groups, but also frequently accompanied by other changes in site C, which could explain why these compounds are specific against the H1 HA subtype (Supplementary Material Fig. S10). On the other hand, let us note that the mutation S326 HA1 V, which renders the A/PR/8/34 virus and its HA 20 protein resistant to compound 2, is mapped closely to the proposed new binding pocket (Supporting Information Figure S11).

Conclusion
We here report the synthesis and antiviral evaluation of 40 N-benzyl-4,4-disubstituted piperidines, which were easily synthesized by an Ugi four-component reaction. Several displayed lowmicromolar activity against A/H1N1 influenza virus (i.e., the A/PR/8/34 strain) but not the A/H3N2 subtype. Mechanistic studies including virus resistance selection and polykaryon assays with compound 2 demonstrated that it represents a new class of H1 HA-specific membrane fusion inhibitors.
The inhibitory activity is proposed to be mediated through binding to a new site in the HA 2 subunit close to the fusion peptide, which is so far unexplored to design influenza virus fusion inhibitors.
Remarkably, a direct π-stacking interaction of the N 1 -benzylpiperidine moiety with a F9 HA2 residue located on the fusion peptide is reinforced by the formation of a second π-stacking with Y119 HA2 . The proposed binding model successfully rationalized the SAR results and the observed selectivity The antiviral efficacy of the compounds is limited to the A/PR/8/34 H1N1 virus so far, thus reducing the practical therapeutic application of this type of inhibitors. However, the unique binding mode proposed for compound 2, which may directly stabilize the fusion peptide leading to an alternative inhibition mechanism in comparison with reported small-molecule fusion inhibitors, deserves further study. These compounds therefore appear to be promising hits, although two major challenges need to be dealt with. The first improvement would be search for chemical modifications able to form interaction with all three fusion peptides in the trimeric structure of HA, which would likely enhance the inhibitory potency. The second would be to obtain inhibition of other HA subtypes to confer broad anti-influenza A virus activity. Finally, given their direct interaction with the fusion peptide, these compounds may also serve as a scaffold to design chemical tools for exploration of the fusion process, more specifically the molecular events that occur during low pH-induced HA refolding leading to release of the fusion peptide.

Chemistry
The chemical synthesis and characterization of novel 1,4,4-trisubstituted piperidine analogues 3, 5- were described in Supplementary Material. For 1, 2 and the remaining piperidine compounds, the synthesis was reported in detail elsewhere [24]. The reference compounds ribavirin, chloroquine and nucleozin were from commercial sources. The sulphated sialyl lipid NMSO 3 was a generous gift from G. Wright (Microbiotix, Worcester, MA).

General information
Microanalytical results obtained with a Heraeus CHN-O-RAPID were within 0.4% of the theoretical values. Electrospray mass spectra were measured on a quadrupole mass spectrometer equipped with an electrospray source (Hewlett Packard, LC/MS HP 1100). Analytical thin-layer chromatography (TLC) was performed on silica gel 60 F254 (Merck). Compounds were purified by flash column chromatography with silica gel 60 (230-400 mesh) (Merck), by preparative centrifugal circular thinlayer chromatography (CCTLC) on a Chromatotron (Kiesegel 60 PF254 gipshaltig (Merck), layer thickness of 1 mm, flow rate of 5 mL/min) or by MPLC using SNAP 12 g KP-C18-HS cartridges in an Isolera One system (Biotage). The purity of the compounds was analyzed using an analytical Agilent Technologies (model 1120 Compact LC) ACE 5 C18-300 column (150 mm x 4.6 mm). Gradient conditions were: mobile phase CH 3 CN/H 2 O (0.05% TFA); flow rate, 1 mL/min; detection, UV (254 and 217 nm). All retention times are quoted in minutes. HPLC-MS was performed on an HPLC Waters 2695 instrument connected to a Waters Micromass ZQ 2000 spectrometer, and a photodiode array detector. The column used was a Sunfire C18 (4.6 mmx50 mm, 3.5 mm), and the flow rate was 1 mLmin -1 . NMR spectra were recorded with Varian Inova-300, Varian Inova-400 or Varian System-500 spectrometers operating at 300, 400, or 500 MHz for 1 H NMR, and at 75, 100, or at 125 MHz for 13 C NMR with Me 4 Si as an internal standard. The purity of novel compounds was also determined to be >95% by elemental analysis. Chemicals and reagents were obtained from commercial sources and used without further purification.

4.3.General synthetic procedure for the Ugi reaction
To a solution of the ketone (1.32 mmol) in methanol (2 mL), 2 equivalents of the corresponding amine, 2 equivalents of the amino acid and 2 equivalents of the isocyanide were successively added.
The resulting mixture was stirred at room temperature for 4 days. Then, a 1.2 M solution of HCl in MeOH was added and the mixture was stirred at room temperature for 30 min. The solvent was removed under reduced pressure. The residue was redissolved in ethyl acetate and was successively washed with saturated NaHCO 3 (3 x 10 mL) and brine (3 x 10 mL). The organic phase was dried

Methyl (S)-4-((1-benzyl-4-(benzylcarbamoyl)piperidin-4-yl)-4-fluorobenzyl)amino)-3-
To perform the polykaryon assay [33], the coding sequence for A/PR/8/34 HA was cloned into a pCAGEN plasmid [46]. Specific HA mutations were introduced by site-directed mutagenesis and verified by cycle sequencing. Plasmid transfection into HeLa cells was carried out in 12-well plates as described [13,33]. Two days later, surface-exposed HA0 was first activated for 15 min with TPCK-treated trypsin. Next, the cells were preincubated for 15 min with test compound; exposed for 5 min to pH 5. To measure the hemolysis pH of wild-type or mutant virus [33], virus was added to microcentrifuge tubes together with an equal volume of 2% chicken red blood cell (RBC) suspension. After 10 min incubation at 37 °C, unbound virus was removed by centrifugation. Next, the cell pellets were resuspended in acidic buffer, i.e. PBS-CM that was acidified with acetic acid to a pH ranging from 4.6 to 6 with 0.1 increments. After 25 min incubation, the suspensions were neutralized with NaOH and intact RBC were removed by centrifugation. The extent of hemolysis in the supernatants was quantified by measuring the absorbance at 540 nm using a plate reader. The hemolysis pH was defined as the pH at which 50% hemolysis occurred relative to the value at pH 4.6.
Enzymatic assays with diverse proteases. Molecular docking. Docking was performed using Glide [39,40] with the standard precision mode to explore the potential binding of compound 2 to sites A, B, and C in A/PR/8 H1N1 HA. Docking in sites B and C was performed using the PDB structure 1RU7 of A/PR/8. For site A, the homology model of A/PR/8 in an "open" conformation (based on PDB entry 3EYM) produced in a previous work [13] was used. A cubic grid of 25 Å was applied to define each of the three binding sites. A RMSD value of 1.0 Å and an atom displacement of 2.6 were set to filter poses during clustering. A total of 50 poses were generated per ligand and four clusters were identified for each of the three analysed sites.
Molecular dynamics simulations. Amber16 [49] was used to perform MD simulations on the selected ligand-protein complex previously generated by docking analysis. The general Amber force field (GAFF) was used to parameterize the ligand [50], and the partial charges were derived at the B3LYP/6-31G(d) level, after preliminary optimization of the molecular structure, by using the restrained electrostatic potential (RESP) fitting method [51] implemented in Gaussian09 [52] and Antechamber. The A/PR/8/34 complex with compound 2 (cluster 1) obtained in site C was solvated with a truncated octahedral (TIP3P) [53] water box with a layer of 18 Å and neutralized by adding Na + ions. For the protein, disulphide bonds were built by using the "bond" command in tleap.
Energy minimization was accomplished in three-stages that involved firstly all hydrogen atoms, then water molecules, and finally all the system with a maximum number of minimization cycles of 10,000 for the latter stage. The minimized system was then heated from 0 to 300 K in five steps, the first being performed at constant volume and the rest at constant pressure. The system was then equilibrated for 5 ns at constant pressure. Langevin dynamics with a collision frequency of 1.0 ps-1 was applied for temperature regulation during the heating. A force constant of 10 kcal mol -1 Å -2 was applied to restrain some protein-ligand contacts and thus avoid conformational distortions during heating and equilibration. These harmonic restraints were gradually eliminated during the first 50 ns of MD production. A total of 150 ns (50 for each MD replica) of un-restrained MD production was generated at constant volume and temperature in periodic boundary conditions.
The SHAKE algorithm [54] was applied to constrain bonds involving hydrogen atoms. Cut-off for non-bonded interactions was set to 10 Å. Electrostatic interactions beyond the cut-off within the periodic box were computed by applying the Particle Mesh Ewald (PME) method [55]. The weakcoupling algorithm with a time constant of 10.0 ps was used to stabilize the temperature during the simulation.
Binding free energy. MMPBSA.py was used to compute the the binding free energy of compound 2.
The method estimates the free energy ( G ) of the protein-ligand complex, and the separate protein and ligand as the sum of enthalpic ( E gas ), solvation ( G solv ) and entropic ( S ) terms ( Eqs. 1-3).
where E int , E elec and E vdW are the internal, Coulomb and van der Waals energy components in the gas phase, G GB is the polar contribution, which was evaluated by using the generalized Born solvation model, and G SURF stands for the nonpolar term, which was determined using a linear dependence on the solvent-accessible surface area ( SASA ; Eq. 4).
where the surface tension (γ ) is set to 0.0072 Kcal mol -1 Å -2 , and b is a correction term that was assumed to be zero in present calculations.
The binding free energy ( ΔG bind ) was evaluated as noted in Eq. 5.
ΔG bind = G complex − G protein − G ligand (5) where G x is the average value determined for each species (x: complex, protein, ligand) using an ensemble of 100 snapshots taken from the last 50 ns of the MD trajectory of the complex within the framework of the single-trajectory approach. The vibrational entropy term was not determined since it was assumed to cancel out in the comparison between the three complexes.

Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.