Site-directed M2 proton channel inhibitors enable synergistic combination therapy for rimantadine-resistant pandemic influenza

Pandemic influenza A virus (IAV) remains a significant threat to global health. Preparedness relies primarily upon a single class of neuraminidase (NA) targeted antivirals, against which resistance is steadily growing. The M2 proton channel is an alternative clinically proven antiviral target, yet a near-ubiquitous S31N polymorphism in M2 evokes resistance to licensed adamantane drugs. Hence, inhibitors capable of targeting N31 containing M2 (M2-N31) are highly desirable. Rational in silico design and in vitro screens delineated compounds favouring either lumenal or peripheral M2 binding, yielding effective M2-N31 inhibitors in both cases. Hits included adamantanes as well as novel compounds, with some showing low micromolar potency versus pandemic “swine” H1N1 influenza (Eng195) in culture. Interestingly, a published adamantane-based M2-N31 inhibitor rapidly selected a resistant V27A polymorphism (M2-A27/N31), whereas this was not the case for non-adamantane compounds. Nevertheless, combinations of adamantanes and novel compounds achieved synergistic antiviral effects, and the latter synergised with the neuraminidase inhibitor (NAi), Zanamivir. Thus, site-directed drug combinations show potential to rejuvenate M2 as an antiviral target whilst reducing the risk of drug resistance.


Introduction
The 2009 H1N1 "swine 'flu" outbreak dramatically illustrated the speed at which influenza pandemics can spread in the modern era due to globalisation. Whilst not as virulent as the 1918 Spanish Influenza, which claimed more than 50 million lives, swine 'flu caused increased mortality and morbidity, placing considerable burden upon even advanced health care systems. The unexpected origin (Smith et al, 2009;Solovyov et al, 2010;Zhang & Chen, 2009) of swine 'flu precluded the rapid deployment of a vaccine, making antiviral prophylaxis the only means by which to curtail the initial stages of the pandemic.
Another class of influenza antiviral, the adamantane M2 proton channel inhibitors (M2i) amantadine and rimantadine, are now clinically obsolete due to widespread resistance (Furuse et al, 2009;Zaraket et al, 2010). This is due to a near-ubiquitous S31N polymorphism within M2 (other rarer variants also occur) generating resistance at little or no associated fitness cost to the virus. Targeting rimantadine resistant M2 has been a long-standing priority yet progress targeting M2-N31 is limited compared to other minor variants (Drakopoulos et al, 2018;Li et al, 2017;Li et al, 2016a;Li et al, 2016b;Musharrafieh et al, 2019;Thomaston & DeGrado, 2016;Wang et al, 2013a;Wang et al, 2018;Wu et al, 2014). The majority of studies have focused upon adamantane derivatives with various chemical groups linked via the primary amine.
Acidic pH promotes M2 channel activity by both enhancing tetramer formation and the subsequent protonation of conserved His37 sensor residues within the channel lumen Pinto et al, 1992;Salom et al, 2000;Shimbo et al, 1996;Wang et al, 1995). This causes conformational shifts in adjacent Trp41 "gates" via a mechanism that remains debated (Andreas et al, 2015;Cross et al, 2012;Hong & Degrado, 2012;Hu et al, 2010;Leiding et al, 2010;Phongphanphanee et al, 2010;Pielak & Chou, 2010;Thomaston et al, 2019;Williams et al, 2016). More than twenty M2 atomic structures exist on the PDB, although none feature full-length protein. Instead, minimal "trans-membrane" (TM) or C-terminally extended "conductance domain" (CD) peptides have been investigated as these regions recapitulate channel function, although the CD region possesses enhanced biological activity ). Interestingly, drugbound TM and CD structures differ with respect to adamantane binding (Schnell & Chou, 2008;Stouffer et al, 2008); TM channels harbour a single lumenal amantadine molecule, whereas CD structures bind four rimantadine molecules at membrane-exposed peripheral sites, corresponding to the region largely absent from TM peptides. Ensuing controversy remains, hampered by the poor chemical probe qualities of adamantanes and a lack of confirmatory functional studies comparing TM and CD peptides (Andreas et al, 2010;Cady et al, 2011a;Cady et al, 2010;Cady et al, 2011b;Du et al, 2009;Hu et al, 2011;Koz akov et al, 2010;Ohigashi et al, 2009;Pielak et al, 2011;Pielak et al, 2009;Rosenberg & Casarotto, 2010) .
In the present study, we show that both M2 binding sites are viable antiviral targets that enable synergistic M2-targeted combination therapy. In silico high throughput screening enriched for novel compounds with predicted preference for one or other site, validated by the first comparative TM/CD peptide screen for M2 -N31 channel activity. Several hits identified in vitro show antiviral activity versus pandemic H1N1 influenza A virus in the laboratory setting, comprising both modified adamantanes as well as unique scaffolds. Whilst a previously reported adamantane M2-N31 inhibitor rapidly selected resistance in culture, this did not occur for newly derived compounds. Excitingly, pairs of M2-N31 inhibitors achieved synergy, as did combining novel scaffolds with the NAi, Zanamivir. Together, these observations provide a firm basis for rejuvenating M2-N31 as a viable target for much needed drug combinations, which should help combat the emergence of antiviral resistance.

Robust identification of specific M2-N31 inhibitors in vitro.
We adapted an indirect liposome dye release assay for viroporin activity (Atkins et al, 2014;Carter et al, 2010;Foster et al, 2014;StGelais et al, 2007;Wetherill et al, 2012) for M2 CD region peptides derived from Influenza A/England/195/2009 (Eng195), a prototypical first wave virus from the 2009 H1N1 pandemic. In addition to the wild type Eng195 M2 harbouring N31, we included a mutated S31 peptide to allow validation with rimantadine ( Figure 1a). Both peptides induced equivalent dose-dependent release of carboxyfluorescein (CF) from liposomes, and acidic pH increased M2 activity ( Figure S1a, b). Critically, rimantadine only blocked activity of Eng195 M2-S31 peptides, confirming the ability of the assay to discriminate between susceptible and resistant M2 variants.
Modified adamantane compounds have been shown to inhibit M2-N31 activity (Li et al, 2017;Wang et al, 2011;Wang et al, 2013a;Wang et al, 2018;Wu et al, 2014). Thus, to validate the assay we tested Eng195 peptides versus a small collection of similar prototypic molecules that included inhibitors of rimantadineresistant hepatitis C virus (HCV) p7 (Foster et al, 2011). Encouragingly, from nine compounds tested, three modified adamantanes inhibited both M2-N31 and M2-S31 peptides (compounds D, H and J, Figure 1b). A further two adamantanes showed no activity (E, G), and another amiloride-related compound (L) was similarly inactive (Figure 1b). Interestingly, two amiloride-like molecules (B, K) showed activity versus M2-S31 but not N31, reminiscent of rimantadine. Lastly, the alkylated imino-sugar NNDNJ also blocked the activity of both M2 peptides, yet this compound disrupted oligomerisation, as previously shown for HCV p7(Foster et al, 2011) ( Figure 1A, S1c). D, H, J and NNDNJ displayed antiviral effects in Eng195 culture ( Figure   1c and data not shown), establishing the dye release assay as a robust means of screening for M2-N31 specific inhibitors with genuine antiviral effects.
Ambiguous predicted binding modes for prototypic M2-N31 inhibitors. To gauge how novel M2-N31 inhibitors might bind the channel complex, we generated structural homology models for Eng195 M2-N31 and S31 based upon the PDB: 2LRF CD structure from the Chou laboratory (Schnell & Chou, 2008) (Figure 2a, c). This template includes both potential rimantadine binding sites. As noted previously, N31 caused splaying of the trans-membrane domain (TMD) compared with the more lumenally oriented S31, yet the structure also differed throughout the helical bundle, consistent with a reported destabilising effect for N31 (Pielak et al, 2009) (Figure 2a, c). Surprisingly, docking of both rimantadine and novel inhibitors led to distinct binding poses at the lumenal and peripheral sites for N31 and S31 models. For the wild type Eng195 M2-N31 model, predicted binding at the peripheral site (defined by D44, R45 and F48) consistently involved H-bonding to D44, whereas the orientation of inhibitors altered in S31 models (Figure 2a, b, d and table S1). Similarly, M2-N31 lumenal interactions occurred near to the N-terminal neck of the bundle near N31 and V27 (Figure 2a, b, d and table S1), whilst binding within M2-S31 models resembled previous structures, occurring further inside the TMD just above H37. Such N/S31 dependent "flipping" within the channel lumen has been observed previously (Wu et al, 2014). Based upon these observations, we reasoned that such promiscuous pleotropic binding may result from the chemical properties of adamantane derivatives, and that molecules with improved molecular fit may exhibit less ambiguous predicted binding modes. However, it would also be necessary to validate site preferences in vitro to generate meaningful structure-activity relationships (SAR) for improved M2-N31 inhibitors.
Determination of M2-N31 inhibitor binding preference in vitro. TM peptides lack the majority of the Cterminal extension present within CD peptides that contains the proposed peripheral binding site. Thus, we hypothesised that lumenally targeted compounds would inhibit both TM and CD peptides, whereas those with a peripheral site preference would show activity only against CD peptides. The dye release assay was therefore adapted to include Eng195 TM peptides, accounting for their reduced biological activity compared to CD (Figure 3a, S1a).
We first tested a published M2-N31 inhibitor, M2WJ332 (Wang et al, 2013b), a modified adamantane with activity against full-length M2-N31 shown to bind within the lumen of a TM domain NMR structure (A/Udorn/307/1972 (H3N2) M2-S31N, PDB: 2LY0, figure 3b). Surprisingly, M2WJ332 blocked the activity of Eng195 M2-N31 CD peptides and had no TM-specific activity either under standard assay conditions (up to strong functional preference for the peripheral binding site in vitro despite its location within the 2LY0 structure (Wang et al, 2013b). Accordingly, docking of M2WJ332 within the 2RLF-based Eng195 homology model resembled other adamantanes by generating poses within both the lumen and the peripheral site ( Figure 3d, S2a). Whilst structural and biophysical studies have previously compared lumenal and peripheral binding (Cady et al, 2011b;Rosenberg & Casarotto, 2010), to our knowledge this represents the first functional evidence supporting the relevance of peripherally targeted M2 ligands in vitro. Importantly, this suggests that M2-N31 possesses two potential binding sites to exploit for antiviral discovery.
Screening of novel M2-N31 inhibitors enriched for lumenal or peripheral site preferences. The next step was to enrich in silico screening libraries for compounds predicted to bind preferentially at one or other M2 site, removing as much ambiguity as possible through extensive attrition of compound characteristics. Using the Eng195 2RLF-based model as a template, grids corresponding to each site were targeted by an in silico high throughout screen (eHiTS, SimBioSys Inc.), based upon a random chemical library and a second input ligand pool derived through evolution of the compound D molecular structure (ROCS, OpenEye Scientific) ( Figure 4a). eHITS score ranked the top 1000 hits for each site and docking scores were cross-validated using a second software package (SPROUT, Keymodule Ltd.). Short-listing of compounds involved an attrition protocol directed by agreement between the two binding scores, compound molecular weight, specific binding pose and drug-like qualities. Prioritisation of compounds focused upon site selectivity, rather than merely predicted potency. Details of resultant compounds are summarised in Table 1.
Compound screens for activity versus TM and CD peptides at 40 M yielded multiple hits (defined as a ≥30% reduction in channel activity for at least one M2 peptide at 40 M) corresponding to lumenal and peripheral site preferences. Interestingly, more lumenally targeted hits presented compared to peripheral, and a third class of compound displayed specificity to TM rather than CD peptides, e.g. compound P6.4 (Table 1). A minority of compounds displayed functional site preferences contradicting docking predictions, yet rational enrichment of ligand pools in silico had significantly augmented the number of M2-N31 targeted hits, with ~50 % of compounds displaying M2-inhibitory activity in vitro compared with a hit rate of <1 % from a random prospective screen (Hansson et al, 2014).
We next titrated exemplar compounds from each class to ensure specificity corresponded to that observed in the 40 M screen and predicted interactions (Figure 4b, S2b). Interestingly, whilst lumenal compounds (e.g. L1.1) displayed equivalent activity against both TM and CD peptides, some peripherally targeted ligands (e.g. DP9) also began to exert measurable effects versus TM peptides at higher concentrations. This included DL7, which despite predictions of lumenal binding, displayed clear preference for CD peptides at lower concentrations ( Figure 4b). We hypothesise that this occurs due to inefficient interactions with the partial peripheral binding site present at the C-terminus of TM peptides. Moreover, titrations confirmed the phenotype of TM-specific ligands (e.g. P6.4) ( Figure 4b).
Following cytotoxicity testing in MDCK culture ( Figure S3 (DP9 excluded as precise IC50 not determined). Eng195 replication was monitored by periodic titration and sequencing of M2 RNA in supernatants. The only change in the M2 sequence was detected starting from the first analysis of M2WJ332-selected supernatants (day 5); a U>C change at position 80 (M2 cDNA sequence) led to a Val27>Ala mutation (GUC>GCC, V27A) ( Figure 6b). This polymorphism enriched over time, becoming the dominant species apparent at day 14 ( Figure 6b).
Sequencing of plaque-purified virus from day 5 supernatants confirmed the presence of V27A within 7/7 M2WJ332-selected plaques, whereas 9/9 L1.1-, and 5/5 rimantadine-selected plaques retained wild type M2-N31 sequence ( Figure 6b). Accordingly, far fewer plaques were derived from normalised L1.1 supernatants compared to M2WJ332 or rimantadine ( Figure S4), and these were eliminated by limited titration of the compound. By contrast, M2WJ332-selected supernatants still retained multiple plaques (~30 % of DMSO control) at much higher concentrations (80 M), likely reflecting the proportion of mutant virus (~30-40 %, see below) within the bulk population. Lastly, virus could only be expanded from L1.1 plaques at ≤20 M inhibitor, with cytopathic effects (CPE) taking at least 48 h to manifest. By contrast, M2WJ332 or rimantadine plaques readily expanded under 80 M inhibitor, with CPE evident by 24 h.
To investigate further whether V27A was potentially linked to M2WJ332 resistance, the fold increase in titre was determined under selection (80 M compound) for passages six and seven. Eng195 under Rimantadine, Zanamivir (known to rapidly select resistance (Correia et al, 2015;LeGoff et al, 2012)) and M2WJ332 achieved similar fold-increases compared to DMSO controls, whereas both L1.1 and DL7 significantly suppressed viral replication leading to much reduced titres compared to input ( Figure 6c). We then introduced an evolutionary bottleneck at passage eight to enrich for any minor resistant variants present within bulk populations, normalising innoculae to a multiplicity of infection (MOI) of 0.001. After a further six passages, output titres (passage 14) again revealed a significant reduction in L1.1 selected Eng195 titre, whereas DL7 selected supernatants had recovered to a similar range as controls ( Figure 6d).
Finally, deep sequencing of IAV genomes was performed comparing passage 5 and 14 supernatants to investigate minor M2 variant populations and mutations occurring elsewhere in the Eng195 genome ( Figure   6e). Only M2WJ332 selected virus showed changes in the M2-N31 sequence compared to controls, with V27A increasing from ~40 to ~80 % abundance between the two time points. Additional minor changes also occurred at position 31 (N31S/I), with another change located outside of the CD region (E70K). Low prevalence changes also occurred in the HA protein in M2WJ332 selected virus at passage 14, namely K226E, Y454H/S and N461D. L1.1 selected virus also showed a low prevalence change at HA Y454H, along with changes in PB2 D161N and M1 P55S. No variation distinct from controls was evident for DL7 selected virus at passage 5 or 14. Taken together, V27A was the only relevant polymorphism significantly enriched during chronic culture with novel M2-N31 inhibitors. This strongly suggests that Eng195 M2-A27/N31 confers resistance to M2WJ332.

Synergistic antiviral effects using M2-N31 targeted inhibitor combinations. M2-N31 specific inhibitors with
distinct binding properties provides the opportunity for antiviral combinations that could not only improve therapy, but also reduce the likelihood of resistance. Combinations of M2WJ332, DP9, DL7 and L1.1 were titrated in Eng195 MDCK plaque reduction assays and antiviral effects assessed using MacSynergy software (results corroborated using "Compusyn"). Remarkably, combinations of M2WJ332 with either L1.1 or DP9 yielded synergistic reductions of viral titre ( Figure 7a). M2WJ332 combined with L1.1 showed increased synergy proportionate to both inhibitor concentrations. However, synergy between M2WJ332 with DP9 only occurred in the lower M2WJ332 range and increased with DP9 concentration. By contrast, combinations involving the DL7 compound resulted in antagonism, whether combined with a lumenally (L1.1) or peripherally (M2WJ332) targeted partner ( Figure 7b). Lastly, L1.1 also achieved synergistic antiviral effects when combined with the NAi, Zanamivir (Figure 7c, S5), supporting that drug combinations between classes should be achievable.

Discussion
This work lays the foundation for future combination therapies targeting rimantadine-resistant influenza A viruses, which could form a vital addition to the current pandemic antiviral repertoire. We have moved beyond the controversy surrounding two potential binding sites within M2 channel complexes, instead showing that synergistic antiviral therapy is achievable using compounds targeting both regions, which ultimately should reduce the incidence of new resistance mutations. Finally, whilst adamantanes still contribute to the M2-N31 chemical toolbox, we describe multiple distinct scaffolds that should provide a start-point for the next steps in antiviral drug discovery.
The question of how amantadine and/or rimantadine block the activity of M2 from sensitive influenza strains has been debated since two contrasting atomic structures were published in 2008 (Schnell & Chou, 2008;Stouffer et al, 2008). However, these and other ensuing studies generally compared TM with CD peptides, which is likely to bias where prototypic adamantanes bind. This is due to the C-terminal extension in CD peptides inducing a more compact helical bundle that is less favourable for lumenal interactions compared with the much broader structure seen for TM peptides, which also lack the majority of the peripheral binding in biophysical or other studies, likely due to the technical difficulties associated with the use of membrane bilayers compared with membrane-mimetic detergents. Notably, the lipidic 2L0J membrane bundle is less compact compared with other CD structures, and the orientation of the C-terminal extension, comprising basic helices that form the core of the peripheral binding site, differs significantly in 2LOJ compared to the detergent-solubilised template used for the present study, 2RLF (Schnell & Chou, 2008;Sharma et al, 2010). This may explain why fewer peripherally targeted compounds were selected compared to the lumen; accordingly, re-docking peripheral compounds into 2LOJ using E-Hits results in altered binding poses and affinity scores ( Figure S6).
To our knowledge, the present study is the first to compare TM and CD peptides using an in vitro functional assay. Assuming that the presence/absence of the C-terminal extension discriminates peripheral binding, the identification of CD peptide-specific hits is the first direct evidence that the peripheral binding site represents a druggable target for M2. Whilst we cannot rule out that CD and TM peptides adopt altered conformations within liposomes, the equivalence seen for lumenally targeted compounds suggests that this is likely not the case, at least for the trans-membrane region. Many other M2-N31 targeted studies, such as that describing M2WJ332 (Wang et al, 2013b), combine TM-derived structures with compound efficacy data from full-length protein, thereby presuming that the two are directly related. Nevertheless, discovering that M2WJ332 showed a functional preference for the peripheral binding site was unexpected. It is conceivable that M2WJ332 interacts differently with Eng195 M2 compared to the protein in the reported TM structure (A/Udorn/307/1972 (H3N2) M2-S31N, PDB: 2LY0). Moreover, 2LY0 was solved using relatively harsh detergent conditions (Chipot et al, 2018;Wang et al, 2013b), rather than lipid, and in the presence of millimolar rather than micromolar inhibitor concentrations. Nonetheless, whether or not Eng195 and Udorn M2 are directly comparable as proteins, the present study serves as precedent for effective influenza A virus inhibitors targeting the M2-N31 periphery for at least one influenza A virus strain. Interestingly, a third class of compounds showed in vitro preference for TM peptides. However, with the exception of a mild antiviral effect for P6.4 (Table 1, Figure 5a), none of these displayed activity in Eng195 culture making the relevance of these compounds unclear.
Eng195 chronic culture in the presence of M2WJ332 led to the rapid evolution of a V27A change within the M2 sequence. Both plaque purification and monitoring the titre of selected bulk populations supported that this change confers specific resistance. V27A is a known amantadine resistance mutation (Barniol-Xicota et al, 2017;Hu et al, 2017), albeit less prevalent compared to S31N. Given the ambiguity surrounding amantadine/rimantadine binding and the nuances of TM and CD peptide detergent structures, it is unclear how such resistance mutations relate to amantadine, or other inhibitor binding. Amantadine binds proximal to H37 in the central portion of the trans-membrane helical bundle (Cady et al, 2010), meaning that S31N and V27A are too distant for these mutations to affect direct contacts with the drug. However, lumenally docked compound D and M2WJ332, as well as the 2LY0 structure, predict direct contact with V27 and N31 ( Figure   2d, 3a). V27 is also proposed to form a secondary gate/constriction at the neck of the channel lumen (Yi et al, 2008), meaning that V27A might promote a more open-form channel complex. For Eng195, the channel is already less compact due to S31N pushing apart the trans-membrane helices (Figure 2a). Directly related to this alteration in structure, both polymorphisms also mediate resistance to peripherally bound rimantadine via the destabilisation of channel complexes (Pielak et al, 2009). Hence, it is neither possible to reinforce nor contest in vitro data on M2WJ332 peripheral binding based upon the location of V27A.
Notably, naturally occurring M2-N31/A27 double variant isolates exist, implying a low genetic barrier in nature as well as in cell culture (Durrant et al, 2015). Interestingly, other minor M2 variants selected by M3WJ332 included the revertant N31S, which mediates resistance to another published M2-N31 specific adamantane derivative; a more dramatic N31D mutation also mediated resistance to a dual M2 -N31/S31 inhibitor (Ma et al, 2016). However, changes in M2, including at N31, did not occur in L1.1-selected Eng195, which forms predicted interactions with N31 but not V27 due to sitting lower in the lumen interacting with the H37 tetrad ( Figure S2). Accordingly, this compound maintained suppression of Eng195 bulk titre throughout the course of the experiment. DL7 appeared to behave similarly to L1.1 at early times, but titres recovered following the introduction of the evolutionary bottleneck at passage eight. No sequence changes relative to controls occurred at passage five or fourteen, making it unclear whether resistant variants were initially selected.
Irrespective of whether resistance may or may not arise to new M2-N31 specific compounds, the most clinically important observation from this study is that both lumenal and peripheral binding sites are viable drug targets that allow combinations of inhibitors to be used for therapy. Moreover, synergistic, rather than additive, antiviral effects were achieved for two of the four combinations tested, and L1.1 behaved similarly when combined with Zanamivir. Hence, whilst the compounds herein represent the initial stages of hit identification for both binding sites, the indications are that further development will eventually enable double, triple, or even expanded therapeutic regimens upon inclusion of other agents. Such strategies are applied to antiviral treatment for other highly variable RNA viruses and there is growing consensus that such approaches represent the best way forward for influenza A virus. Furthermore, combinations should combat potential shortcomings for individual agents in terms of lower genetic barriers to potential resistance.
Overall, future exploitation of both druggable sites within M2-N31 using specific inhibitors has considerable potential to rejuvenate this essential ion channel protein as a drug target, providing an important additional resource to combat emergence of future pandemics.

Materials and Methods
Peptide synthesis and reconstitution. Peptides (Eng195 M2-N31 CD and TM,  Open access small molecules libraries were used for an unbiased molecular binding study, with eHiTS (SimBioSys Inc.) used to dock compounds onto the two pre-defined binding regions of the Eng195 homology model. Ranked by eHiTS score the top 1000 hits, at each site, were manually assessed for their binding pose and drug-like qualities, resulting in seven predicted lumenal binding compounds L1-L7 and six peripheral binding compounds P1-P6 being selected for testing.
In addition, a biased screen was carried out utilising a rapid overlay of chemical structure (ROCS) approach and centred on compound D. ROCS (OpenEye Scientific) software was used and the top 1000 hits were docked against the homology model using eHiTS. Compound docking at both sites was validated using SPROUT (Keymodule Ltd.) software. A protocol of attrition was carried out to select DL and DP compounds, focussing on docking scores, molecular weight and specific interactions with the M2 tetramer. Full details are listed in (Section 3.5.1). Briefly, compounds were selected based on agreement between the two binding scores, molecular weight and specific interactions with the protein. Analogues of selected compounds were found via the online tool eMolecules (www.emolecules.com). Selected compounds were subsequently docked against the 2RLF-based E195 M2 homology model using eHiTS.   Figure 7 are from MacSynergy, Compusyn data was comparable and is available upon request.
Selection in culture using M2-specific compounds and plaque purification of single variants. For serial passage using increasing compound concentrations, MDCK cells were seeded into 6 well plates 4 h before the initial infection with Influenza virus was carried out as described above at an MOI of 0.001, with 2.5 µM compound. At 24 hpi virus containing media was removed, 1/10 th volume was used to infect freshly seeded MDCK cells, as a blind passage and the remainder was snap frozen. This process was repeated, each time increasing the concentration of compound present in the media two-fold, until 80 µM was reached. At selected time points the titre of viral supernatants was determined via plaque assay. These supernatants could then be used at MOI 0.001, with fresh 80 µM compound, in subsequent infections.
For plaque purification, MDCK cells were seeded in 12 well plates 4 h before infection. Virus was diluted 1:250 in SF media containing between 5 and 80 µM compound and used to infect cells for 1 h, before cells were set under overlay media, containing 2 µg/ml TPCK, 0 -80 µM compound and agar. At 72 hpi, agar plugs were picked and placed in 300 µl SF media for 2 h prior to it being used to infect fresh MCDK cells, in the presence or absence of compound (5 -80 μM) for 1 h at 37 0 C, 5 % CO2. Once infectious supernatant was removed, it was replaced with SF media + 1 µg/ml TPCK and 0 -80 µM compound and plates returned to 37 0 C, 5 % CO2. Once > 40 % CPE was observed, infectious supernatant was clarified, prior to vRNA extraction.
Extraction, purification and sequencing of virion RNA (vRNA). vRNA was extracted from clarified supernatants using a QIAamp Viral RNA Mini Kit (QIAGEN) according to the manufacturer's instructions.
Resultant eluted vRNA was kept at -20 0 C for short term storage, or transferred to -80 0 C for long term storage. vRNA was synthesised into first strand cDNA using SuperScript® III (SSCIII) (Invitrogen™) and a Eng195 segment 7 specific forward primer (sequences available upon request). A negative control of vRNA but no SSCIII was included in each experiment.
cDNA was amplified via polymerase chain reaction (PCR) using the proof reading Phusion® high fidelity (HF) polymerase (Phusion) (New England Biolabs). Reactions were heated to 98 0 C for 30 s, followed by 35 cycles of the following steps; denaturation at 98 0 C for 10 s, annealing at 48 0 C for 30 s and extension at 72 0 C for 40 s and a final incubation at 72 0 C for 7 min. Amplified cDNA was purified using a QIAquick PCR Purification Kit (QIAGEN) according to the manufacturer's instructions, with eluted DNA concentrations determined using a nanodrop spectrophotometer and DNA visualised by Tris-acetate-EDTA buffered agarose gel electrophoresis. Samples were stored at -20 0 C. Direct sequencing of virus-derived cDNA. Standard dsDNA sequencing was conducted using the Mix2Seq kit (Eurofins Genomics), with forward internal primer Eng195_s7_Fint (5'-GGCTAGCACTACGGC-3') or reverse primer Flu_s7_R2 (5'-AGTAGAAACAAGGTAGTTTTTTACTCTAGC-3').
Next Generation sequencing of total genomic viral RNA. vRNA was reverse transcribed by Superscript III (Invitrogen) and amplified by Platinum Taq HiFi Polymerase (Thermo Fisher) and influenza specific primers (Zhou et al, 2009) in a 1-step reaction. Library preparation was performed using a Nextera kit (Illumina). Libraries were sequenced on an Illumina MiSeq using a v2 kit (300-cycles; Illumina) giving 150-bp paired end reads. Reads were mapped with BWA v0.7.5 and converted to BAM files using SAMTools (1.1.2).
Variants were called using QuasiBAM, an in-house script at Public Health England.            with "?" indicating potential binding to lumen or partial peripheral binding site based upon compound titrations; bold text indicates differences from predicted binding. Several compounds were tested versus Eng195 in culture (80 M) and the order of magnitude titre reduction across at least three assays is shown.
Finally, IC50 was determined for four compounds selected for synergy experiments.