Influenza A virus is an important pathogen that still causes the death of 290,000–650,000 people in seasonal outbreaks worldwide (Iuliano et al., 2018). It pertains to the Orthomyxoviridae family, which is characterized by the presence of a negative-sense, single-stranded, enveloped ribonucleic acid with a segmented genome (Cheung and Poon, 2007). There are four genera in the Orthomyxoviridae family: Influenza A, Influenza B, Influenza C, and Thogotovirus. Influenza A viruses are found in many different animals, and Influenza B viruses circulate widely only among humans. Influenza A viruses are divided into subtypes based on two proteins on the surface of the virus: hemagglutinin (HA), which is responsible for the receptor binding and membrane fusion, and neuraminidase (NA), which assists the release of the viral progeny (Cheung and Poon, 2007). Up to 18 variants of HA (from type H1 to H18) have been described till now, whereas 11 subtypes have been disclosed for NA. All known subtypes of influenza A viruses have been found among birds, except subtypes H17N10 and H18N11, which have only been found in bats. Only some subtypes (H1N1, H2N2, H3N2, H5N1, H7N7, and H9N2) have been isolated from humans, suggesting that there are restrictions to host viruses in humans.

While infections in humans directly from animals are unusual, sporadic infections and outbreaks caused by influenza A viruses have occurred in the past. The most severe pandemic was the “Spanish Flu” of 1918–1919, which caused 20–50 million deaths by influenza A H1N1 virus strain. The subsequent health challenges were the “Asian flu” in 1957 and the “Hong Kong flu” in 1968, which were caused respectively by H2N2 and H3N2 strains, and more recently the 2009 swine influenza pandemics, which pertains to the H1N1 strain (Cox and Subbarao, 2000; Hayward et al., 2014; Watson et al., 2015; Saunders-Hastings and Krewski, 2016). The latent risk of facing future pandemics underlined the Global Influenza Strategy initiative launched by the World Health Organization in 2019, which has the goal of preventing seasonal influenza, controlling the spread of influenza from animals to humans, and preparing for the next influenza pandemic (World Health Organization, 2019). This threatening scenario underscores the urgency in developing better diagnostic tools and promoting the discovery of effective influenza treatments, including vaccines and antiviral drugs.

This review aims to discuss experimental and computational efforts invested in elucidating the structural basis of the proton conducting activity of the M2 protein channel and the discovery of antiviral compounds targeting the M2 transmembrane domain. To this end, the manuscript first describes the role of the M2 proton channel in the replicative life cycle of the influenza A virus. Then, attention is focused on the structure of the proton channel, which is directly implicated in the proton transfer across the viral membrane as a preliminary event leading to the fusion of viral and host membranes and the subsequent release of the viral genetic material. Emphasis is made on the molecular mechanisms that underlie the proton conduction through the lumen of the M2 proton channel. The discussion also highlights the role played by key residues located in the transmembrane region, and especially the efforts spent in the determination of the pKa of the imidazole rings present in the His37 tetrad. Finally, the manuscript examines the mechanism of action of selected inhibitors, and the evolution followed in the design of novel compounds introduced to increase the inhibition potency and to mitigate the emergence of resistance to current treatments.

Influenza A virus replicates in the epithelial cells of the upper respiratory tract of humans, pigs, and horses. Regarding in vitro models, the virus can also affect different cells that contain targets decorated with sialic acid moieties (Julkunen et al., 2001; Sidorenko and Reichl, 2004), which act as the entrance anchor point into the cells.

Enclosed by a lipid-protein envelope, the genetic material of the influenza A virus contains 8 single-stranded RNA segments that encode the genetic information for viral proteins (Dou et al., 2018; Krammer et al., 2018; Jung and Lee, 2020): the membrane-bound hemagglutinin (HA) and neuraminidase (NA), the transmembrane M2 proton channel, the matrix protein M1, the non-structural proteins (NS1, NS2/NEP), and the heterotrimeric RNA-dependent RNA polymerase (RdRp), formed by three domains (PA, PB1, and PB2), and nucleoprotein (NP) components.

The replicative life cycle of the virus is schematically shown in Figure 1. The first step of the cycle is the entry of the virus inside the host cell (step 1). Viral entry into the cell occurs via HA, which recognizes and binds the sialic acid of the host cell receptors. The virions penetrate inside the host cell through clathrin-dependent receptor-mediated endocytosis (step 2). The acidification of the endosome is facilitated by the M2 proton channel, which transfers protons to the interior of the viral particles. This event promotes a large-scale conformational remodeling in HA that facilitates the release of the fusion peptide, and the concomitant structural rearrangement of HA triggers the fusion of the viral and host membranes, leading to conformational changes in the M1 protein that facilitate the release of the viral ribonucleoproteins (vRNPs) to the cellular cytoplasm (step 3) (Honda et al., 2002). These vRNPs comprise complexes formed by the individual viral RNA segments and proteins, including the RNA-dependent RNA polymerase (RdRp), which is associated with the 3′- and 5′-ends of viral RNA segments, and nucleoproteins (NPs) that are coating the remainder of the viral RNA. The vRNP complexes are then transported into the nucleus to initiate transcription and replication by RdRp, which is formed by the polymerase basic protein 1 (PB1), polymerase basic protein 2 (PB2), and polymerase acidic protein (PA). The first step is known as cap-snatching, a process where RdRp (PB2 and PA subunits) interacts with the C-terminal domain of the host RNA polymerase II and removes the first 10–13 nucleotides of a nascent transcript, which is then used as a primer to initiate transcription from the viral RNA template by PB1 (De Vlugt et al., 2018). This is followed by a viral genome replication phase that involves the synthesis of full-length viral RNA and cRNA strands (steps 4 and 5) (Pielak and Chou, 2011). With a late stage of packaging, vRNPs are formed (step 6). The synthesis of M2 protein, HA and NA is carried out by ribosomes bound to the membranes of the endoplasmic reticulum, which are then glycosylated and transported to the Golgi apparatus (step 7) (Sidorenko and Reichl, 2004). Finally, a complex between the membrane-embedded proteins and vRNPs is formed and delivered to the cell membrane, leading to the assembly of new virions that will be released to the extracellular side of the host cell membrane (step 8).

The preceding discussion suffices to reveal the complexity encoded in the multi-step processes - cell entry, replication, intracellular trafficking, virion assembly, and release - involved in the generation of new virions. In this scenario, the search for effective antiviral treatments has been inspired by the identification of druggable viral targets with a relevant role in the life cycle of virus infection. Early drug discovery projects targeting the M2 proton channel primarily relied on adamantane-related compounds (Moorthy et al., 2014; Wang et al., 2015). However, the emergence of resistant strains motivated the search of novel therapeutic approaches, including the development of 1) NA inhibitors such as zanamivir and oseltamivir, although drug-resistant variants have also been reported, 2) HA inhibitors, as exemplified by the discovery of arbidol and JNJ7918, and, more recently, 3) RdRp inhibitors, such as pimodivir and baloxavir (Loregian et al., 2014; Wu et al., 2017; Han et al., 2018; Mifsud et al., 2019; Ginex and Luque, 2021). The discovery of small-molecule antiviral compounds against these targets has benefited from the progress made in disclosing the fine structural details of these proteins, the molecular mechanisms that underlie their biological function, and the characterization of the mode of action of inhibitory compounds.

The M2 proton channel is a homotetrameric integral transmembrane (TM) protein located in the viral envelope of the influenza A virus. It is a selective, pH-dependent channel that regulates the acidification of the interior of the virion, leading to dissociation of the viral RNA from its bound matrix proteins and release of the viral genetic material for replication (Pinto et al., 1992; Pielak and Chou, 2011; Manzoor et al., 2017).

From a structural point of view, the M2 channel is a homotetramer, each monomer consisting of 97 amino acids and can be divided into four main regions: 1) the highly conserved, unstructured N-terminal domain (residues 1–21) located in the viral exterior, which assists the incorporation of the M2 channel into the virion, 2) the TM domain (residues 22–46) (Figure 2), which assembles into a tetrameric domain and is involved in the proton flux, 3) an amphiphilic membrane-anchored helical domain (residues 47–67), which is located in the interface with the membrane and stabilizes the channel by inducing membrane curvature and mediates membrane scission, and 4) the C-terminal domain (residues 68–97), which is located in the viral interior and interacts with the matrix protein M1 and contributes to virus packaging and budding (McCown and Pekosz, 2006; Rossman et al., 2010; Wang and Hong, 2015a).

The TM region contains the binding site for M2 inhibitors in the interior of the pore, filling a pocket located between the tetrads formed by residues Val27, Ser31, Gly34, and His37 (Schnell and Chou, 2008; Acharya et al., 2010; Cady et al., 2010; Sharma et al., 2010). This pore is implicated in the proton conduction, and the proton current is regulated by the tetrads formed by His37 and Trp41. His37 is a pH sensor and can conduct the proton inside the channel by changing the protonation state of the imidazole side chain, whereas Trp41 acts as a gate of the channel. Accordingly, binding of compounds that fill the binding site in the interior of the M2 channel would sterically impede the passage of protons (Figure 2).

The pH-sensing gating mechanism regulates the proton flux through the channel, which is modulated by the transition from an open state at an acidic pH, where the imidazole rings of the His37 tetrad are protonated, to a closed state at alkaline pH environment, which promotes the deprotonated state of His37 residues. In the open state, the M2 channel is also permeable to some cations, such as Na⁺ and K⁺, through an antiporter-like mechanism, carrying out the metal cation efflux in conjunction to proton influx, but the permeability to protons has been estimated to be 10⁶–10⁷ higher than the conduction of alkaline cations (Chizhmakov et al., 1996; Shimbo et al., 1996; Mould et al., 2000a; Manzoor et al., 2017). There is an asymmetric conductance as the proton flux is higher when the pH outside the virus is low, consistently with an acidic endosome. This behavior has been attributed to the presence of the Trp41 gate, which reduces the conduction in the reverse direction (Ma et al., 2013). Thus, when the His37 tetrad is protonated by protons coming from the endosome, the indole moiety of Trp41 undergoes a conformational change and the channel enters an open state that would allow the passage of protons to the interior of the virus.

The functional relevance of His37 and Trp41 is revealed by the drastic effects triggered by mutations in these positions, which cause a reduction in the pH sensitivity of the channel, and a disruption in the proton conduction (Wang et al., 1995; Tang et al., 2002; Venkataraman et al., 2005). Thus, mutations in His37 have a drastic effect on the functional properties of the M2 channel, since the mutated proteins not only increase the proton conduction of the channel, but also suppress the selectivity for proton conduction. As an example, the specific activity determined for the voltage-independent component of the conductance at pH 6.2 for the wild type channel was 0.16 μA/ng, but increased to 1.36 and 30 μA/ng for the H37G and H37E mutated channels, respectively (Holsinger et al., 1994). Furthermore, the mutated ion channels H37G, H37S, and H37T were found to be deficient in ion selectivity (Venkataraman et al., 2005). On the other hand, the expression of the M2 protein in oocytes leads to an inward H⁺ current upon bathing in acidic media (pH 5.9), which is lost upon return to alkaline conditions (pH 8.5). However, whereas no currents were observed for mutated channels containing Phe41, Cys41, or Ala41 prior to acidification, outward H⁺ currents were detected by changing acidic conditions to alkaline solution.

Beyond the definition of the molecular events implicated in proton conduction, a major challenge has been the identification of small molecules that can block the proton flow through the inner pore, and hence be used for therapeutic treatments against flu. Amantadine (Amt) and its ethyl analog, rimantadine (Rmt), were patented in 1961 and 1963, respectively (Du Pont Patent, 1961; Prichard, 1967), but they are no longer recommended for the treatment of flu infection. This obeys to several reasons, such as the limited effectiveness against influenza B virus, unwanted side effects, and the emergence of adamantane-resistant influenza A viral strains, primarily single mutated variants V27A, L26F and S31N. These mutations are found in a specific area facing the interior of the pore, suggesting therefore the location of the binding site in the M2 channel (Grambas et al., 1992; Hoyden, 2006; Deyde et al., 2007). For instance, in contrast to the large and widespread chemical shifts observed upon binding of Rmt to the M2 channel, only minor changes in chemical shifts were detected upon addition of a 4-fold molar excess of Rmt to the S31N mutated channel (Cady et al., 2009; Andreas et al., 2010; Andreas et al., 2012). For our purposes here, the discussion of the molecular determinants implicated in drug binding will be limited to the subset of structures available in the Protein Data Bank that contains drug-like inhibitors (see Tables 3, 4).

As noted above, the binding site for M2 inhibitors is located in the interior of the pore, filling a pocket shaped by residues Val27, Ser31, Gly34, and His37. Early neutron difraction (Duff et al., 1994) and computational (Samson and Kerr, 1993) studies already identified this pocket as the binding site that mediates the blockade of proton conduction by Amt. Detailed information about the binding mode of Amt and Rmt has subsequently been gained from structural studies based on X-ray diffraction, solution NMR and ss-NMR (Table 3; Schnell and Chou, 2008; Stouffer et al., 2008; Cady et al., 2009; Cady et al., 2010; Thomaston et al., 2018). The inspection of these structures reveals the existence of two binding sites (Figure 7). In most cases the drug is located in the interior of the pore filling the space located between Val27 and Gly34 (pore-binding model), but in two structures (PDB ID 2RLF and 6US8) the drug is also found outside the channel located in a peripheral site close to Leu43 and Asp44 in the C-terminus of the TM helical region (interface-binding model). The high-affinity site lies in the pore, as revealed from the analysis of REDOR dipolar dephasing between ¹³C-labeled M2TM and perdeuterated Amt in ss-NMR studies performed in a lipid bilayer (Cady et al., 2010). From a computational point of view, free energy calculations also pointed out that binding of Rmt to the pore-binding site is ∼7 kcal/mol more favorable relative to the interface-binding site, leading to more stable drug binding and channel inhibition (Li et al., 2008). The presence of the peripheral binding site can be attributed to the partitioning of the drug in the membrane-like lipidic environment (Duff et al., 1994; Gu et al., 2011). The studies reported by Cady et al. (2010) showed that the highest affinity binding site of amantadine is the N-terminal pore lumen, which is consistent with the known stoichiometry of binding (1:1 drug:channel) and the location of resistant mutations, such as V27A and S31N. However, increasing the drug:channel concentration up to 4:1, the drug contributes ∼7% of the amphiphiles composing the DMPC bilayer, and binding to the low-affinity, peripheral site is observed in NMR studies. Thus, when free amantadine is a major component of the membrane, Amt contacts the C-terminus of the protein, though the affinity for the peripheral site has been estimated to be ∼40-fold lower (see (Wang et al., 2015) for a detailed discussion of biophysical, computational and functional assays that addressed the binding mode of Amt and Rmt).

Inspection of these structures shows that in the pore the drug is oriented along the axis of the channel. Drug binding causes a dehydration of the channel, thus preventing the formation of a water wire from the N-terminus, although the amino group of Amt and Rmt is bound to water molecules that fill the space located between the drug and the His37 tetrad. In the specific case of Rmt, the two enantiomers bind to the pore with only slight differences in the hydration pattern, and exhibit similar values for association and dissociation kinetic rate constants and binding affinities, which was in agreement with the small differences in the relative binding free energy (around 0.3 kcal mol⁻¹) determined from free energy calculations (Thomaston et al., 2020). Furthermore, Rmt was found to be more potent than Amt, as noted in a ∼14-fold ratio of their binding affinities to the M2 channel. On the other hand, it has been shown that binding of Amt reduces the pKa of His37 and alters the propensity to form hydrogen-bond interactions in the His37 tetrad (Hu et al., 2007; Cady et al., 2011).

From a computational standpoint, fine details of the binding mode of adamantane blockers have been extensively investigated. Yi et al. paid attention to the changes in conformational states of the M2 channel upon binding of Amt through comparison of the ensembles collected for both apo and bound forms (Yi et al., 2008; Yi et al., 2009). In particular, they observed a reduction in the conformational heterogeneity upon Amt binding, as reflected in a narrower distribution of kink angles around Gly34 in the transmembrane helices compared to the apo species, which agrees with the broadened resonances observed from ssNMR studies of this latter form. In contrast, large-kink angles were not observed in simulations. The occurrence of kinks also allows the access of water molecules that may stabilize the kinked helices via hydrogen bonds with the backbone carbonyl and amide groups around the kink, which in turn may also influence the proton conduction.

On the other hand, efforts have been conducted to predict the differences in binding affinity for adamantane inhibitors using free energy calculations, including implicit solvent/implicit membrane molecular mechanics Poisson-Boltzmann surface area (MM-PBSA) approach (Homeyer et al., 2016) and free energy perturbation techniques (Gkeka et al., 2013; Ioannidis et al., 2016). Gratifyingly, alchemical free energy calculations performed for a set of 10 adamantane ligands reproduced satisfactorily the binding potency determined experimentally using isothermal titration calorimetry (ITC) against the M2 channel at high pH (pH = 8), especially keeping in mind that they cover a binding affinity range of only ∼2 kcal mol⁻¹. Nevertheless, it is worth noting that the agreement between computational and experimental data was affected by the nature of the lipids used to model the membrane. In particular, a higher correlation between experimental and computed relative binding free energies was obtained when calculations were performed for the tetramer in a 1,2-dimyristoyl-sn-glycero-3-phosphocholine bilayer, possibly due to a larger mimetic resemblance with the dodecylphosphocholine micelles used for ITC measurements.

Furthermore, the entire recognition process of Amt to the M2 channel has been explicitly explored using multiple-walkers well-tempered metadynamics calculations (Llabrés et al., 2016). The results suggested that Amt retains some degree of conformational motion in the pore allowing the adoption of two main orientations where the amino group is oriented to the N-terminus and C-terminus (denoted up and down, respectively). Binding of Amt would follow a sequential mechanism that would involve trapping of the protonated Amt by the negative electrostatic field created by the tetrad of Asp24, followed by the adoption of a transiently populated intermediate with Amt in the up state, and finally the reorientation in the interior of the pore to the down state, which is the thermodynamically favored orientation (Figure 8). Release of Amt from the down state to the channel mouth is destabilized by ∼12 kcal mol⁻¹. The agreement with the experimental binding affinity (∼8–9 kcal mol⁻¹) is reasonable, since the theoretical estimate omits the free energy contribution due to the release of Amt to the aqueous solvent, and the fact that the apparent K _i might not properly correspond to an equilibrium measurement of the inhibitory constant (Wang et al., 1993; Ma et al., 2009; Rosenberg and Casarotto, 2010). As a particular remark, this study also pointed out that the down → up transition depends on the presence of chloride anions in the C-terminus of the channel, since the stabilization of the down state can be enhanced by ∼6 kcal mol⁻¹ due to the modulation of the electrostatic field in the interior of the pore. Finally, the barrier from Amt dissociation was estimated to be ∼19 kcal mol⁻¹, which compares a value of 22 kcal mol⁻¹ determined from electrophysiological assays (Wang et al., 1993).

Mutation of Val27 to Ala causes a drastic alteration in the free energy surface for Amt binding, as noted in a stabilization of the up state less than 1 kcal mol⁻¹, which agrees with the experimentally observed lack of inhibitory potency for the Amt-resistant V27A channel. This effect can be attributed to the increased accesible volume in the inner pore due to the V27A mutation, which encouraged the search of compounds with an expanded hydrophobic cage (Duque et al., 2011; Wang et al., 2011; Rey-Carrizo et al., 2014; Rey-Carrizo et al., 2015; Thomaston et al., 2018). In the case of the spiroadamantane 1 ((1r,1′S,3′S,5′S,7′S)-spiro [cyclohexane-1,2′-tricyclo [3.3.1.1∼3,7∼]decan]-4-amine) shown in Table 4 (Thomaston et al., 2018; Thomaston et al., 2020), comparison of the X-ray structures solved for the wild type and the V27A mutant shows that the compound exhibits a similar arrangement in the ion channel, which is found in the closed conformation (Figure 9) (Thomaston et al., 2020). The ligand exhibits only a slight shift depending on the nature of the residue at position 27. In the V27A complex, the amino group of 1 occupies approximately the same position as the ammonium group of Amt in complex with the wild type channel (PDB entry 6BKK) (Thomaston et al., 2018). This is facilitated by the larger free volume enabled by the presence of Ala27. Nevertheless, in the complex with the wild type channel, the adamantyl group of 1 and Amt overlap, but the ammonium group binds deeper in the pore at the expense of displacing few water molecules. MD simulations performed for the V27A complex with 1 reflected the shift of the ligand toward the N-terminus.

A combination of computational techniques, biophysical studies, functional assays and classical medicinal chemistry approaches has been used also to design potent inhibitors of V27A. This mutation imposes a reduction in the steric constraint found in the wild type channel, suggesting the compounds with a size-expanded hydrophobic cage would be better suited to fit the larger volume of the pocket in the V27A channel (Wang et al., 2011). An example is the spiroadamantane 4, which has an IC₅₀ value of 18.7 μM against the wild type channel and of 0.3 μM against the V27A species. MD simulations showed that the ammonium group occupies the lower aqueous site forming solvent-mediated hydrogen bonds with the His37 residues in both wild type and V27A, but in this latter case the compound was slight shifted toward the N-terminus due to the extra space afforded by Ala27.

Other efforts to develop size-expanded analogs have led to the pyrrolidine derivative 5 reported by Vázquez and coworkers (Figure 10) (Duque et al., 2011; Rey-Carrizo et al., 2014). This compound inhibits the wild type M2 channel with an IC₅₀ of 18 μM, while being ca. 26-fold more potent against the V27A mutant. MD simulations revealed an orientation consistent with the down state of Amt, with the amine nitrogen pointing toward the His37 tetrad with an average tilt angle (i.e., the deviation of the amine nitrogen from the pore axis) of 16°. Nevertheless, in the V27A complex 5 adopted the down orientation, but was also found in a up-like arrangement, presumably facilitated not only by the larger volume in the pore, but also by the widening of the helices at the location of Ala27, as the cross-diagonal distance between Cα atoms at position 27 was enlarged by 1.1 Å compared to the wild type channel. The effect of including other polycyclic scaffolds as the hydrophobic cage have led to analogs 6 and 7 (Figure 10), which exhibit a micromolar potency against the wild type channel (IC₅₀ ∼ 2 μM), but a lower inhibition in the V27A mutant (IC₅₀ ∼ 17.2 and 184.6 μM for 7 and 6, respectively) (Rey-Carrizo et al., 2015), suggesting a limiting effect in the size of the hydrophobic cage even for the V27A channel. At this point, MD simulations suggested that even the center-of-mass of 6 and 7 was located between the planes formed by S31 and A27 tetrads, the protonated amine was pointing to the N-terminus in five out of six MD simulations.

The resistance of the S31N channel, which is the most prevalent in currently circulating influenza viruses, to inhibition by Amt is reflected in the more variable motion and orientation of the ligand in the mutated channel. Thus, whereas Amt occupies a stable position in the wild type channel, it showed larger positional fluctuations in the pore of the mutated channel. This trait is reflected in a decreased energetic stability of the ligand in the proximity to the Asn31 side chains, as reflected in potential of mean force calculations. Finally, this destabilization was accompanied by an increase in the density of water molecules around and above Amt in the S31N channel compared to the wild type species, enabling water molecules to fill laterally the available space in the channel pore (Gleed and Busath, 2015). Furthermore, studies conducted for the binding of Rmt to the wild type M2 channel and the S31N variant have related the resistance with the higher dissociation rate constant determined for the mutated channel (Drakopoulos et al., 2018). In turn, this effect was attributed to the reduction in van der Waals interactions due to the shift of the ligand toward the C-terminus due to due to enhanced repulsive forces of the Asn side chains with the adamantyl ring in the mutated channel, and the concomitant weakening of the stabilizing contacts with Val27.

While these features reveal the challenges for developing inhibitors against the M2 S31N variant, significant progresses have been made in the last years. An example is compound 2 ((3S,5S,7S)-N-{[5-(thiophen-2-yl)-1,2-oxazol-3-yl]methyl}tricyclo [3.3.1.1∼3,7∼]decan-1-aminium), also denoted M2WJ332, Table 4). This compound inhibits the proton conduction in electrophysiological assays of the S31N channel with an IC₅₀ of 16 μM, being however much less effective against the wild type ion channel (Rey-Carrizo et al., 2014). Remarkably, the NMR structure reveals that the compound binds the S31N channel in a different orientation compared to Amt and Rmt (Figure 10), since the adamantly ring is located facing the His37 tetrad, whereas the thienyl group is located at the level of the Val27 tetrad. This arrangement is facilitated by the slight expansion (∼1.0 Å) observed at the N-terminus of the TM helices relative to the wild type M2 channel. As noted in the analysis of MD simulations performed for the complex with 2, this binding mode is reinforced by hydrogen-bonds formed by the protonated amine and the isoxazole ring with three Asn31 residues (Wang et al., 2013).

Starting from compound 2, a successful rational design led to compound 3 ((3s,5s,7s)-N-[(5-bromothiophen-2-yl)methyl]tricyclo [3.3.1.1∼3,7∼]decan-1-aminium; Table 4) (Drakopoulos et al., 2018), which exhibits a similar potency in inhibiting both the wild type and the S31N channel (around 77% inhibition in two-electrode voltage clamp assays, and EC₅₀ values of 4.6 and 1.8 μM, respectively). The chemical scaffold of compound 3 combines an adamantane moiety with a bromothiophene unit. However, the most remarkable structural feature is that 3 adopts completely flipped arrangements in the wild type and mutated channels (Figure 11A). Thus, the bromothiophene unit of the ligand faces the His37 tetrad in the wild type channel, but it is pointing toward the N-terminus in the S31N variant. MD simulations confirmed the structural stability of the flipped orientations in the wild type and mutated channels, enabling an interpretation of the structure-activity relationships (Wu et al., 2014). Inspection of the NMR structures (2MUW and 2MUV) does not support the involvement of halogen bonding, as this noncovalent interaction requires specific geometric features between the halogen atom and an electron-rich atom that are not fulfilled in these structures (see Kolár and Hobza, 2016 for details of halogen bonding). Furthermore, MD simulations revealed the formation of weak interactions between the bromine atom and water molecules, supplemented with transient, nonspecific contacts with the imidazole ring of His37 and the carbonyl group of Gly34 in the wild type channel, and with the Val27 side chains in the S31N channel (Wu et al., 2014).

It is also interesting to highlight the synthesis of aminoadamantane-CH₂-aryl derivatives (8 in Figure 10) designed as sensitive probes for blockage of the wild type and S31N ion channels (Tzitzoglaki et al., 2020). These compounds retain the aminoadamantane unit present in 2, but the isoxazole ring has been replaced by a substituted phenyl ring. MD simulations have shown that compound 8 exhibits a stable binding in the S31N mutated channel. The binding mode locates the adamantyl group between Val27 and Gly34, and the phenyl unit fills the space between the side chains of Val27 residues. The results obtained from electrophysiological assays in oocytes evidence that blockage of the proton conduction in the wild type channel is relatively insensitive to chemical changes in the scaffold of 8, such as the insertion of a methylene between the adamantane and amino units or the replacement of adamantane by diamantane or triamantane. However, these changes have a drastic effect on the current blockage of the S31N channel, leading to lower association and higher dissociation rate constants relative to the parent compound 8. These results point out the larger sensitivity of the S31N ion channel to the chemical features of inhibitors upon binding to the pore.

Superposition of the compounds shown in Tables 3, 4 is useful to exemplify the ability of the inhibitors to exploit different sites and binding modes along the interior of the pore in order to prevent proton transport in the wild type M2 channel and the V27A and S31N mutated species. (Figures 11B,C). This feature has been highlighted in previous studies that combined the structural analysis of bound channels with metadynamics calculations (Wang et al., 2013; Gianti et al., 2015), which pointed out the existence of distint locations that can be occupied by the protonated amine along the axis of the channel. These studies also remarked the role played by waters molecules inside the pore, as they contribute to stabilize the inhibitor by facilitating hydrogen-bond bridges with carbonyl groups of the transmembrane helices as well as with the His37 tetrad. From a practical perspective, the presence of multiple interaction sites for the protonated amine and the structure of water clusters that stabilize the inhibitor can be useful hallmarks for guiding the design of new drugs. In this regard, recent studies have indicated that the ammonium group of these inhibitors can act as a hydronium mimic upon binding to the channel pore, suggesting that the adamantane-like compounds function as mechanism-based inhibitors as long as they exploit structural features of the channel implicated in proton transport (Watkins et al., 2020).

As a final remark, it is worth noting a very recent study by Kolokouris and coworkers where a systematic analysis of setup conditions and methodological details, covering from docking calculations to MD simulations, is presented for the M2 channel (Kolocouris et al., 2021). This study examines distinctive traits, such as the usage of M2 (22–46) and M2 (22–62) constructs, lipid content in the bilayer (1,2-dimyristoyl-sn-glycero-3-phosphocholine,1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine, and 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoethanol-amine), and the effect of cholesterol, paying particular attention to the location of channel blockers and water structure, the presence of chloride anions, structural properties of the helical bundle, and the binding site of cholesterol. While this study deserves practical interest for the choice of parameters in simulations of the M2 proton channel (and related viroporins), we limit ourselves to highlight three findings. First, the results obtained for M2 complexes with Amt at pH 8 revealed that chloride anions approached Arg45 and occupied a position located between Trp41 and Arg45 (at a distance close to 15.5–17.5 Å from the Amt nitrogen atom) presence of chloride anions stabilized in the interior of the pore. The anions approached in the C-terminus. The anions approached Arg45. Second, the simulations with CHARMM36 preserved the structure of the amphipatic helices, providing a correct description for the wedge shape geometry that might be required to modulate the saddle-shape curvatuve of the membrane for the release of virions (Wang and Hong, 2015b; Martyna et al., 2017). Finally, the ability of cholesterol to interact with the tetrameric channel, forming van der Waals interactions with Ile39, Leu40, Leu43, Leu46, Phe47, Ile51 and Phe54, and a hydrogen bond with Ser50, which is in agreement with previous experimental evidence (Ekanayake et al., 2016; Elkins et al., 2017; Elkins et al., 2018), paving the way to future studies about the role of cholesterol in mediating the M2 protein function.

Resistance to current pharmacological treatments is a severe health challenge, as illustrated by the emergence of viral strains resistant to drugs interfering with the M2 proton channel, such as Amt and Rmt, but also to antiviral agents targeting neuraminidase, such as oseltamivir, as noticed in a resistant strain associated to the His275Tyr mutation in this protein (Meijer et al., 2009; Ginex and Luque, 2021). Currently circulating seasonal H1N1 and H3N2 strains are now resistant to adamantanes, and oseltamivir is no longer effective against the pandemic H1N1 subtypes (Hussain et al., 2017).

Finding new active antiviral compounds is an urgent need, which may be explored resorting to the application of a combinatorial therapy that should potentiate the effect of two or three drugs endowed with distinct mechanisms of action, hopefully being less susceptible to inactivating mutations (Dunning et al., 2014). A long-term strategy is the identification and characterization of the molecular mechanisms associated to the activity of proteins relevant for the life cycle of the virus, as this knowledge may disclose novel strategies for the design of antiviral compounds.

This approach is illustrated by the research efforts invested in the M2 proton channel. In spite of the apparent simplicity posed by the membrane-embedded four-helix bundle, the flow of protons reflects a complex synergy between different molecular events, involving the acidity of the His37 residues, the structural coupling between His37 and the gating Trp41 residue, the structure of the hydrating water molecules in the pore, the pH-dependent conformational arrangement of the TM helices, the electrostatic funnel created by acidic residues in the N-terminus of the channel, and the influence of the amphiphilic helix and cytoplasmic tail on the flexibility of the TM domain, in addition to the potential influence exerted by other factors such as the nature of the lipid environment in the viral capside.

Although a full understanding of the interplay between these factors, which encompass a diverse range of spatial and time processes, has not yet been achieved, the progress consolidated in the last decades through the combined use of structural, biophysical, physiological and computational techniques is impressive, and the successful development of compound 3 is encouraging. This scientific background defines the framework to address novel questions about the mechanisms of drug resistance and the guidance to novel antiviral and treatment approaches, such as the potential dependence between drug resistance and the nature of the viral strain or the infected cell type, which have been the subject of recent studies (Musharrafieh et al., 2019; Musharrafieh et al., 2020). Hopefully, the outcome will be valuable to enrich our current anti-influenza therapeutic arsenal in the form of effective antivirals less susceptible to drug resistance.