Put a cork in it: Plugging the M2 viral ion channel to sink influenza.

The ongoing threat of seasonal and pandemic influenza to human health requires antivirals that can effectively supplement existing vaccination strategies. The M2 protein of influenza A virus (IAV) is a proton-gated, proton-selective ion channel that is required for virus replication and is an established antiviral target. While licensed adamantane-based M2 antivirals have been historically used, M2 mutations that confer major adamantane resistance are now so prevalent in circulating virus strains that these drugs are no longer recommended. Here we review the current understanding of IAV M2 structure and function, mechanisms of inhibition, the rise of drug resistance mutations, and ongoing efforts to develop new antivirals that target resistant forms of M2.

Introduction

Influenza A virus (IAV) is a significant and ongoing cause of human morbidity and mortality worldwide. While prophylactic vaccines can confer immunity against circulating virus strains, their effectiveness and extent of immunization coverage vary across populations from year to year. This variability contributed in part to the recent and notable 2017–2018 seasonal outbreak which is estimated by the Centers for Disease Control and Prevention to have caused approximately 710,000 hospitalizations and up to 79,400 deaths in the United States alone (CDC, 2018; Rolfes et al., 2019). Of particular concern are less common but historically regular pandemic outbreaks in populations with limited or no immunity that can result in hundreds of thousands to millions of deaths worldwide. Moreover, there still remains a significant technical gap to expedite a rapid vaccine production response, particularly in the face of an increasingly global spread of newly-emergent virus strains (Renaud et al., 2011; Hurt, 2014). Thus, additional countermeasures against IAV in the form of prophylactic and therapeutic antivirals continue to be needed.

The M2 protein is required for IAV replication and spread and is an established antiviral target. M2 is a proton-gated proton channel that belongs to a growing viroporin family of virus proteins. M2 ion channel activity is required for IAV replication, and adamantane-based M2 inhibitors including amantadine (1) and rimantadine (2; Fig. 1 ) were historically used as influenza antivirals. However, adamantane-resistant virus strains are now prevalent to the point that these drugs are no longer recommended for therapeutic use (Bright et al., 2005; Deyde et al., 2007; Fiore et al., 2011; Hayden and De Jong, 2011). More than 95% of circulating adamantane-resistant viruses harbor a serine to asparagine mutation at amino acid position 31 in M2 (Ser31Asn; Bright et al., 2005), which distorts adamantane interactions within the pore without significantly affecting M2 ion channel activity or the fitness of viral replication (Balannik et al., 2010; Grambas et al., 1992; Pinto et al., 1992; Stouffer et al., 2008). Therefore, a major goal of current studies is to identify new inhibitors of adamantane-resistant M2.

Fig. 1

The two FDA approved adamantanes amantadine (1) and rimantadine (2).

Recent advances in molecular and structural biology, electrophysiology, and virology have produced crucial information about the basis of M2-based drug resistance. This in turn has shed light on novel drug development strategies and transitioning of those ideas to practical tools. Here we review the current state of IAV M2 viroporin research and ongoing efforts to identify and develop new antivirals against its drug-resistant forms.

The M2 viroporin of IAV

Viroporins are small (often ~60–120 amino acids), virally-encoded transmembrane proteins that regulate ion conduction across lipid membranes and are a frequent feature of both RNA and DNA viruses (Nieva et al., 2012; Nieto-Torres et al., 2015; Ouyang and Chou, 2014). Viroporins are frequently described as exhibiting a “channel-pore dualism,” where some sequences are broadly permeable to ions and small molecules, while others are tightly-regulated channels that conduct specific ions. Although the functions of most viroporins are not well understood, they are generally involved in conducting ions to facilitate viral entry, assembly, and/or release from host cells. Table 1 lists viroporins of representative viruses.

Table 1

Examples of viroporins from representative viruses.

Virus Family	Virus	Viroporin(s)	Reference
Coronaviridae	SARS-CoV	E, 3a, 8a	Castaño-Rodriguez et al. (2018)
Flaviviridae	Hepatitis C virus	p7	Madan et al. (2015)
Orthomyxoviridae	IAV	M2	Pinto et al. (1992); To and Torres (2019)
	Influenza B virus		Mould et al. (2003)
	Influenza C virus		Pekosz and Lamb (1998)
	Influenza D virus		Kesinger et al. (2018)
Papillomaviridae	Papillomavirus 16	E5	Kabsch and Alonso (2002)
Phycodnaviridae	Paramecium bursaria chlorella virus 1	KcV	Thiel et al. (2011)
Picornaviridae	Poliovirus	2 B	Aldabe et al. (1996)
Pneumoviridae	Respiratory syncytial virus	Small hydrophobic (SH) protein	Gan et al. (2012)
Polyomaviridae	JC polyomavirus	Agnoprotein	Suzuki et al. (2010)
Reoviridae	Avian reovirus	p10	Bodelón et al. (2002)
Retroviridae	HIV-1	Vpu	Strebel (2014)
Rhabdoviridae	Bovine ephemeral fever virus	α1	Joubert et al. (2014)

The role of M2 in IAV replication

The M2 protein of IAV is arguably the best understood viroporin. It consists of a 97 amino acid, type I transmembrane domain-containing protein which forms a tetrameric ion channel that is both proton-gated and proton-selective (Hong and DeGrado, 2012; Nieto-Torres et al., 2015). M2 is a multifunctional, modular protein where each segment performs one or more distinct function(s) (Wang et al., 2011c). Following viral entry, M2 expressed on the virion membrane transports protons from host cell endosomes to acidify the virion interior (Fig. 2 ). This low pH environment facilitates conformational changes in viral hemagglutinin (HA) and dissociation of viral ribonucleoprotein complex from M1 proteins to promote virion membrane fusion and release of viral RNA into the host cell. M2 proton gating function is, however, not required per se to convert HA into the membrane fusion competent state (Bui et al., 1996; Gutman et al., 1993; Martin and Heleniust, 1991). During viral egress, M2 also equilibrates the pH of trans-Golgi lumina and the cytoplasm, which may delay trafficking of virion particles and/or prevent HA from inappropriately adopting a low pH conformation (Alvarado-Facundo et al., 2015; Ciampor et al., 1992; Li et al., 2014; Pinto and Lamb, 2006; Sugrue et al., 1990). Compared to other influenza proteins, the sequence of M2 is highly conserved but has been shown to co-evolve with HA (Grambas et al., 1992; Rossman and Lamb, 2011). As such, M2's conductance rate closely correlates with HA's fusogenic propensity, where increased proton conductance by M2 correlates with more acid labile HA sequences (Grambas et al., 1992; Scholtissek et al., 1998). M2 has also been found to localize at the virus budding site on the host cell surface, where the amphipathic helix of M2 alters membrane curvature leading to membrane scission and release of new viral progeny (Chen et al., 2008; Rossman and Lamb, 2011; Rossman et al., 2010). Although protons are the preferred substrate, M2 can also act as a Na+ and K+ antiporter (Leiding et al., 2010). In particular, the ability of M2 to pass bulkier K+ ions becomes physiologically relevant in the endocytic pathway, where high Na+ concentration in early endosome clusters is replaced by K+ to allow for the subsequent release of the viral ribonucleoprotein bundle (Stauffer et al., 2014).

Fig. 2

Overview of IAV replication, with an emphasis on M2 functions. For clarity, only a subset of influenza proteins are shown.

Regulation of ion conduction by M2

The existence of M2 was initially reported by Lamb et al., (1981) (Lamb et al., 1985, 1981; Lamb and Choppin, 1981; Lamb and Lai, 1981) Pinto et al. (1992) then demonstrated its ion channel activity in Xenopus laevis oocytes injected with M2 RNA and measured by the two-electrode voltage clamp (TEVC) method of electrophysiology, which has become the most common assay by far to measure M2 conductance and its inhibition by small molecules. However, several groups have used a variety of electrophysiological techniques including whole-cell patch-clamp of mammalian cells to probe the ion conduction properties of M2 (Chizhmakov et al., 1996; Holsinger et al., 1995, 1994; Jalily et al., 2016; Shimbo et al., 1996; Tu et al., 1996; Wang et al., 1995, 1993) For example, Chizhmakov et al. (1996) expressed M2 in mouse erythroleukemia cells and also observed selective conduction of protons.

Subsequent mutagenesis studies have further defined the specific M2 amino acid residues that are required for proton conduction and regulation (Fig. 3 ). Notably, a fragment of M2 encompassing the transmembrane domain and spanning as little as amino acids 21–51 was observed by TEVC to be sufficient to produce amantadine-sensitive, proton-dependent proton currents (Ma et al., 2009). To a first approximation, transmembrane mutations that are predicted to increase the pore radius (i.e., mutation to residues with smaller side chains) result in increased proton conduction, presumably either through the enhanced formation of water wires or transfer by His37 through conformational changes, while mutations to residues with bulkier side chains that reduce the pore radius also reduce conductance. For example, the introduction of Ala at Val27, which faces the extracellular and intraluminal surfaces and is thought to form the most constricted part of the channel, increases the pore entrance size and obliterates the N-terminal gating mechanism, thereby allowing for easier pore hydration and enhanced conduction (Balannik et al., 2010; Holsinger et al., 1994; Pielak and Chou, 2010). In contrast, introduction of bulky and/or hydrophobic residues such as Phe or Trp at Val27 results in non- or low-conducting M2 proteins. Similarly, mutations that reduce pore size at other locations including Ala30Trp, Ala30Pro, and Gly34Glu also slow the rate of proton conduction and frequently result in loss-of-function (Balannik et al., 2010).

Fig. 3

X-ray crystal structures of M2-S31N (22–46) in the Inwardopen(A) and Inwardclosed(B) states (PDB: 6MJH (Thomaston et al., 2019);). In the Inwardopen state, the distance between the Trp41 indole nitrogen from opposition chains is 12.4 Å. In the Inwardclosed state, the distance between the Trp41 indole nitrogen from opposition chains is 6.7 Å.

Notably, a highly-conserved sequence of His37-XXX-Trp41 within the C-terminal end of the M2 transmembrane domain is regarded as the functional core of proton conductance (Pinto and Lamb, 2006; Tang et al., 2002; Venkatraman et al., 2005). Mutation of His37 to Gly or Gln results in enhanced conductance but also loss of proton selectivity and/or lack of pH dependence (Balannik et al., 2010; Wang et al., 1995). Moreover, mutation of Trp41 to Ala, Cys or Phe results in larger inward currents but also outward currents, indicating that Trp41 regulates unidirectional conductance (Balannik et al., 2010; Tang et al., 2002; Ma et al., 2013). Another key residue, Ser31, is likely to face the pore interior, and its mutation to hydrophobic residues such as Ala negatively affects pore hydration, resulting in diminished conduction. At the other end of the His-Trp quartet, mutating Asp44 to hydrophobic residues such as Ala affects proton exit at the C-terminal end of the channel by increasing the energy barrier (Pielak et al., 2011). These and other mutations might also impact M2 function by indirectly affecting the inherent conduction moiety defined by the His37 tetrad or the gating mechanism defined by Trp41 (Gu et al., 2013; Ma et al., 2013).

The structure of M2

Recently reported structures of M2 have been instrumental toward understanding how adamantanes inhibit this ion channel and how drug resistance overcomes them, in addition to generally informing new M2 drug discovery and ion channel biology. Experimentally-determined protein structures derived from X-ray crystallography, solid-state and solution nuclear magnetic resonance (NMR), and protein-ligand complex structures originating from computational studies have also formed the basis of structure-based drug design. As of this writing, there were more than 35 structures of wild-type (WT) and drug-resistant A/M2 proteins available in the Protein Data Bank (PDB). Most of these were solved by X-ray techniques although some were elucidated using NMR.

While the mechanisms of proton shuttling that enable conduction remain incompletely understood, it is likely that experimental conditions such as pH, peptide length, lipid/detergent composition and thickness, and binding of small molecules affect the fundamental properties of M2 (Acharya et al., 2010; S. Cady et al., 2011a, Cady et al., 2011b; Du et al., 2012; Kolokouris et al., 2019; Kovacs et al., 2000; Thomaston et al., 2013; Zhou and Cross, 2013). Among the available M2 structures in PDB, 2RLF and 3C9J were among the first structures solved using solution NMR and X-ray crystallography in micelles, respectively (Schnell and Chou, 2008; Stouffer et al., 2008). Hu et al. (2010) and Sharma et al. (2010) subsequently applied solid state NMR and molecular dynamics to explore the activation and conduction mechanisms of M2 at low pH (using PDB entries 2KQT, and 2L0J; Cady et al., 2010; Sharma et al., 2010). Although both studies support a “proton-relay” model of conduction, they differ with regard to the side-chain conformation of His37. The study by Hu et al. (2010) proposes a tightly-packed His37 tetrad which undergoes numerous reorientations and side chain rotations to relay protons. In contrast, the model by Sharma et al. (2010) suggests three distinct states for the His-Trp quartet: the ‘‘histidine-locked state’’, the ‘‘activated state’’, and the ‘‘conducting state’’, with transition between these states allowing for proton conductance. In this model, the “histidine locked state” exists when a proton is shared between two adjacent histidine residues (HisH+-His) and the channel is locked in the non-conducting state. They further described the activation cascade at low extracellular pH, when a single hydronium ion approaches the non-protonated histidine in the HisH+-His dimer, breaks this ionic bond, and transitions the channel to the “activated state”, resulting in the formation of stable cation-p interactions between His37 and Trp41. The Trp41 tetrad forms a gate at the C-terminal end, and the perturbation of this cation-p interaction leads to conformational changes which in turn transition the quartet to the “conduction state”. Here, protons are then donated to water molecules and relayed to the intracellular side (Fig. 4 and Sharma et al., 2010 ). The transition between these three steps requires small changes of the χ2 angles of the His37 and Trp41 side chains. This contrasts with the model by Hu et al. (2010), where considerable reorientation of His37 and Trp41 residues are required to allow proton passage. Further experimental and theoretical data supporting both models are described in detail elsewhere (Hong and DeGrado, 2012; Thomaston et al., 2019; Watkins et al., 2019; Zhou and Cross, 2013).

Fig. 4

Water molecules in the wild-type M2 proton channel at different pH conditions. The structures were determined by X-ray free-electron laser (XFEL) at room temperature. (A) Low pH (5.5) structure (PDB: 5JOO). (B) Intermediate pH (6.5) structure (PDB: 5UM1). (C) High pH (8.0) structure (PDB: 5TTC). Waters are shown as spheres (Thomaston et al., 2017).

Mechanisms of M2 inhibition by adamantanes

Advent of adamantanes as IAV M2 inhibitors

Ion channel activity of WT M2 is effectively antagonized by two FDA-approved adamantane-class drugs, amantadine (1) and rimantadine (2) (Fig. 1), in addition to numerous other adamantane and non-adamantane derivatives (Table 2, Table 3 ; Tataridis et al., 2007; Wang et al., 2013a; Wang et al., 2013b; Wu et al., 2014). During virus entry, amantadine's inhibition of M2 results in incomplete dissociation of M1 from ribonucleoprotein complexes, which in turn fail to enter the nucleus to initiate further replication. As M2 can also equilibrate pH across the trans-Golgi network and the cytoplasm, amantadine-based M2 inhibition may also disrupt viral egress (Lamb, R.A. and Krug, 2001; Takeda et al., 2003).

Table 2

Examples of adamantane-based inhibitors of drug-resistant M2. NR, not reported.

Compound	Name	Structure	Activity				Reference
			Assay	IC50/EC50	Virus Strain	MDCK cell toxicity (CC50)
1	Amantadine	Image 1	TEVC	WT: 16 μMV27A: > 500 μMS31N: 200 μM	A/Udorn/72	>100 μM (multiple reports)	Wang et al. (2013a)
			PRA	WT: 0.3 μMS31N: 22 μM	A/WSN/33
2	Rimantadine	Image 2	TEVC	WT: 11 μMS31N: > 2 mM	A/Udorn/72	>100 μM (multiple reports)
3	M2WJ352	Image 3	TEVC	S31N: 14 μM	A/Udorn/72
			PRA	S31N: 14 μM	A/WSN/33	NR
4	M2WJ332	Image 4	TEVC	S31N: 16 μM	A/Udorn/72	100 μM (Li et al., 2016a)
			PRA	S31N: 0.1 μM	A/WSN/33
5	M2WJ379	Image 5	TEVC	S31N: 16 μM	A/Udorn/72	NR
			PRA	S31N: 1 μM	A/WSN/33
6		Image 6	TEVC	S31N: 74.4% inhibition at 100 μMWT: 12.0% inhibition at 100 μM	A/Udorn/72	33.9 μM	Li et al., 2016a, Li et al., 2016b
			PRA	S31N: 1.7 μM	A/WSN/33
7	Benzdiol	Image 7	TEVC	WT: 60 μMS31N: 35 μM	A/Udorn/72	NR	Wang et al. (2013b)
			PRA	WT: complete inhibition at 1 μMS31N: 3.2 μM	A/WSN/33
8	BC035	Image 8	TEVC	WT: 77% inhibition at 100 μMS31N: 76% inhibition at 100 μM	A/Udorn/72	123 μM	Wu et al. (2014)
			PRA	WT: 4.6 μM
				S31N: 1.8 μM	A/WSN/33
9	Spirane-adamantane amine	Image 9	TEVC	WT: 18 μML26F: 6 μMV27A: 0.3 μM	A/Udorn/72	27.6 μM	Hu et al. (2017a); Wang et al. (2011a)
			PRA	WT: 0.3 μM	A/Udorn/72
				N31S+V27A: 1.8 μM	A/WSN/33
10	Spirane-adamantane dithiane	Image 10	TEVC	WT: 64% inhibition at 100 μMV27A: 0.4 μM	A/Udorn/72	74.8 μM	Hu et al. (2017a)
			PRA	WT: 0.07 μM
				N31S+V27A: 1.0 μM	A/WSN/33
11	Organosilane	Image 11	TEVC	WT: 9% inhibition at 100 μMS31N: 86% inhibition at 100 μM	A/Udorn/72	40.7 μM	Hu et al. (2017b)
			PRA	S31N: 0.4 μM	A/WSN/33
12		Image 12	TEVC	WT: 4.1 μMV27A: 3.6 μM	A/Udorn/72	>100 μM	Barniol-Xicota et al. (2017)
			PRA	WT: 0.14 μM	A/HK/7/87
13		Image 13	TEVC	WT: 1.9 μMV27A: 16.2 μM	A/Udorn/72	10 μM
			PRA	WT: > 50 μM	A/HK/7/87
14		Image 14	TEVC	S31N: 75.5% inhibition at 100 μM	A/CA/07/2009	59.3 μM	Li et al. (2017); Musharrafieh et al. (2018)
			PRA	S31N: 1.2 μM
				S31N+V27A: 23.7 μM
15		Image 15	TEVC	S31N: 84.3% inhibition at 100 μM	A/CA/07/2009	146.6 μM	Musharrafieh et al. (2018); Wang et al. (2018)
			PRA	S31N: 0.3 μM
16		Image 16	TEVC	S31N: 47.9% inhibition at 100 μM	A/WSN/33	>300 μM	Wang et al. (2018)
			PRA	S31N: 0.5 μM

Table 3

Examples of non-adamantane-based inhibitors of drug-resistant M2. NR, not reported.

Compound	Name	Structure	Activity				Reference
			Assay	IC50/EC50	Virus Strain	MDCK cell toxicity (CC50)
17	Polycyclic pyrrolidine	Image 17	TEVC	WT: 3 μMV27A: 0.3 μM	A/Udorn/72	10 μM	Rey-Carrizo et al. (2013)
			CPE	WT: 0.37 μM	A/HK/7/87
18	Polycyclic amine	Image 18	TEVC	WT: 18 μML26F: 8.6 μMV27A: 0.7 μM	A/Udorn/72	49 μM	Rey-Carrizo et al. (2014)
			PRA	V27T+ S31N: 1.8 μM	A/PR/8/34
19	Pinanamine derivatives	Image 19	TEVC	WT: 95% inhibition at 100 μMS31N: 27% inhibition at 100 μM	A/Udorn/72	251.5 μM	Dong et al. (2015); Zhao et al. (2012)
			CPE	WT: 3.2 μM	A/HK/68
				S31N: 95.5 μM	A/WSN/33
20		Image 20	TEVC	WT: 98% inhibition at 100 μMS31N: 3.5% inhibition at 100 μM	A/Udorn/72	200.2 μM	Dong et al. (2016)
			CPE	WT: 2.5 μM	A/HK/68
				S31N: 3.4 μM	A/WSN/33
21	Spiranamine	Image 21	TEVC	WT: 12.6 μMV27A: 85 μM	A/Udorn/72	NR	Balannik et al. (2009)
22	Silaspirane derivative	Image 22	TEVC	WT: 13.7 μMV27A:31.3 μM	A/Udorn/72	NR	Wang et al. (2011b)
23	Hexamethylene amiloride	Image 23	SEVC	WT: 1.3 μMS31N: 10% inhibition at 100 μM	A/HK/1073/99	4.7 μM	Balgi et al. (2013); Jalily et al. (2016)
24	Acylguanidine derivatives	Image 24	SEVC	WT: 0.2 μM	A/HK/1073/99	>100 μM	Jalily et al. (2016)
			PRA	WT: 2.3 μM	A/PR/8/34
25		Image 25	SEVC	WT: 0.6 μMS31N: 4.4 μM	A/HK/1073/99	55 μM
			PRA	WT: 40 μMS31N: 18 μM	A/PR/8/34
26		Image 26	SEVC	WT: 20% inhibition at 100 μMS31N: 42 μM	A/HK/1073/99	25 μM
			PRA	WT: 6.9 μMS31N: 1.5 μM	A/PR/8/34
27	Divalent copper complex	Image 27	TEVC	S31N: 90% inhibition at 100 μM	A/Udorn/72	147 μM	Gordon et al. (2017)
			PRA	WT: 3.7 μM	A/Victoria/03/75
				S31N: 0.7 μM	A/CA/07/09
				S31N: 2.1 μM	A/WSN/33
28	Salinomycin	Image 28	Virus-like particle conductance assay	WT 54% inhibition at 100 μMS31N: 72% inhibition at 100 μM	A/PR/8/34	35.6 μM	Jang et al. (2018)
				M2 from influenza B: ~60% at 100 μM	B/Lee/40.
			CPE	S31N: 0.7 μM	A/PR/8/34
				S31N: 1.9 μM	A/CA/07/09

Amantadine first obtained FDA approval in 1966 under the brand name “Symmetrel” for systemic use in humans for prophylaxis of Asian influenza (H2N2) (Davies et al., 1964; Schwab et al., 1969; Wingfield et al., 1969). However, during the first decade following its licensure, Symmetrel was not widely prescribed due in part to side effects such as agitation, confusion, and hallucinations (Keyser et al., 2000). Additionally, the molecular mechanism describing a direct link between amantadine and virus inhibition was not described until more than 25 years later (Duff and Ashley, 1992; Pinto et al., 1992). Rimantadine, sold under the brand name Flumadine, is a methylated derivative of amantadine which was licensed by FDA for the same indications in 1994.

Interactions of adamantanes with M2

Two pharmacologically relevant binding mechanisms were initially proposed for adamantanes (Fig. 5 ). In the first reported crystallographic structure of M2-bound amantadine at pH 5.3 (PDB: 3C9J), Stouffer et al. (2008) suggested that amantadine binds in a pocket located inside the intraluminal cavity and surrounded by the nonpolar side chains of Val27, Ala30, the Cβ group of Ser31, and the Cα group of Gly34 (Fig. 5A). In this pore binding model, the authors proposed that amantadine, by entering the lumen of the channel, “plugs” the M2 pore and thus prevents the transport of protons. The model further suggests that Ser31Asn confers adamantane resistance by decreasing the pore size and occluding stable access of amantadine to the pore (Stouffer et al., 2008).

Fig. 5

Drug binding site controversy of the IAV M2 proton channel. (A) X-ray crystal structure of M2 (22–46) (PDB: 3C9J). (B) Solution NMR structure of M2 (18–60) (PDB: 2RLF).

In a separate series of studies, an alternative but initially controversial binding mechanism for adamantanes, located external to the M2 pore, was also proposed (Pielak et al., 2009; Schnell and Chou, 2008, Fig. 5B). In these studies, rimantadine was found to bind to Asp44 on the C-terminal, lipid-facing side of the helices. The solution NMR structure (PDB: 2RLF), solved at pH 7.5, showed an allosteric mechanism of inhibition where four drug molecules were proposed to stabilize the inactivated state of M2, thereby preventing proton-gated M2 opening and proton transfer. In this binding mechanism, perturbation of the extra-luminal drug binding pocket as a result of Ser31Asn mutation rendered rimantadine incapable of effectively binding to the allosteric site (Pielak et al., 2009).

To reconcile these findings, a series of both experimental and computational studies were rapidly reported (Jing et al., 2008; Balannik et al., 2009; Cady et al., 2010; Carpenter et al., 2008; Chuang et al., 2009; Stouffer et al., 2008). Jing et al. (2008) demonstrated that the intraluminal binding site is more physiologically relevant through conducting whole-cell electrophysiology and antiviral assays using mutant and chimeric M2 forms. By mutating the critical Asp44 residue to alanine, Jing and coworkers continued to observe inhibition by amantadine in both electrophysiological and antiviral plaque assays, suggesting that the allosteric-binding site is not pharmacologically essential (Jing et al., 2008). Additionally, in a chimeric ion channel based on the M2 of influenza B, which is normally amantadine-insensitive due to a larger and more hydrophilic transmembrane pore (Mould et al., 2003), substitution with the M2 transmembrane domain from influenza A restored amantadine sensitivity (Jing et al., 2008). Using solid-state NMR, Cady et al. (2010) also reported that the channel pore was the preferred binding site (PDB: 2KQT), where amantadine preferentially localized to a hydrophobic cage formed by Ala30, Ser31, and Gly34. Hydrophobic interactions between Val27 side chains and the adamantane cage were also observed, and the ammonium group of amantadine showed hydrogen bonding to water molecules and pore-facing residues that stabilized its interactions in the occluded pore. The authors also reported that interactions with the allosteric binding site were observed only at high amantadine concentrations and were thus less probable as the primary mode of M2 inhibition. In further support of the pore-binding model, mutation of Asp44, proposed as the key residue of the allosteric binding mechanism, to six different amino acids (Asp44Ala, Asp44Lys, Asp44Asn, Asp44Phe, Asp44Gly, and Asp44Thr) did not affect inhibition by amantadine (Balannik et al., 2010). Notably, in further testing their initial hypothesis of an allosteric binding site of amantadine to M2, Pielak et al. (2011) next generated an M2 chimera where the C-terminus was derived from the amantadine-insensitive M2 of influenza B virus (PDB: 2LJB). Contrary to their previously proposed allosteric binding mechanism, using solution NMR, they described the rimantadine binding pocket to be preserved within the pore of the chimeric M2 protein. In a series of surface plasmon resonance experiments, the pore-binding site was also found to be more energetically favorable than the allosteric binding site, although the latter did remain sensitive to amantadine (Rosenberg and Casarotto, 2010).

Another computational study using small molecular probes and solvent mapping techniques found the pore-binding site to be more energetically favored, although the allosteric site was also observed under some conditions (Chuang et al., 2009). Gu et al. (2011) also applied molecular dynamics approaches to study the antagonistic effects of known M2 inhibitors and observed that, while interactions with the pore-binding site were more thermodynamically stable, ligands bound slowly and poorly dissociated due to a high energy barrier of binding. In contrast, the allosteric binding was readily accessible, the energy barrier for binding was minimal, and drug binding was less stable and more readily dissociated. Moreover, the initial NMR structure (PDB: 2RLF) did not show evidence of an intraluminal binding site but rather binding at the allosteric peripheral site in the lipid interface. However, this can also be interpreted by noting that the NMR structure was narrower as compared to the counterpart X-ray structure (PDB: 3C9J) due to a smaller tilt angle which resulted in transmembrane helices being overly parallel. This finding further alerted the field to the importance of the protein crystallization environment and emphasised the risk of crystal packing distortions across all published crystal structures (Cross et al., 2012).

Later, through synthesizing organosilane probes and measuring intermolecular NOESY spectra in DPC micelles, Wang et al. (2011a) also mapped the drug binding site at the N-terminal lumen of A/M2 near Val27. Recent high-resolution X-ray crystal structures of M2 in complex with adamantanes and other M2 channel blockers published by DeGrado lab also unambiguously recognized the intra-luminal binding site (Thomaston et al., 2018, 2019). Taken together, these follow-up studies support that amantadine inhibits IAV M2 by plugging the pore (Fig. 5A).

More recently, the binding affinities of the R and S enantiomers of rimantadine to M2 have also been questioned. Surprisingly, by comparing the isotropic chemical shift changes of rimantadine's two enantiomers through ssNMR experiments and molecular dynamics simulations, Wright et al. (2016) proposed that the 2-R enantiomer may bind to full-length M2 differentially and with a higher affinity relative to the 2-S enantiomer. However, a subsequent series of in vitro and cellular assays, electrophysiology experiments, and molecular dynamics simulations indicated that both enantiomers exhibited similar channel blockade in TEVC experiments, comparable antiviral activity in plaque assays, and similar free energies of binding in isothermal titration calorimetry and computational simulations (Drakopoulos et al., 2017). The latter observations are also consistent with experimental data showing similar efficacy of these two enantiomers in mice (Aldrich et al., 1971).

The rise of drug-resistant M2

In addition to Ser31Asn (and further consistent with the pore-binding model described above), the mutations Leu26Phe, Val27Ala, Ala30Thr and Gly34Glu also confer adamantane resistance (Bright et al., 2006; Cady et al., 2009a, 2009b). Large-scale sequence analyses of circulating IAV strains indicate that Leu26Phe, Val27Ala and Ser31Asn are the three major adamantane-resistance mutations found in transmissible viruses (Furuse et al., 2009a, 2009b). Among these, Ser31Asn is the prevalent mutation in more than 95% of adamantane-resistant strains (Dong et al., 2015; Nelson et al., 2009), although a recent study reported a potential rise in the frequency of virus strains containing the Val27Ala and Ser31Asn double mutations (Durrant et al., 2014). A recent crystallography study reported that elimination of hydrophobic contacts that are essential for adamantane drug binding is the underlying cause behind adamantane resistance to Val27Ala mutant (Thomaston et al., 2020). In contrast to the Val27Ala single mutation that results in complete resistance to amantadine, M2 sequences containing Ser31Asn can still be inhibited by amantadine in vitro, albeit with approximately 200-fold higher concentrations in TEVC experiments and plaque reduction assays (PRAs) and substantially higher than what is therapeutically-achievable (Wang et al., 2013a). Notably, the Ser31Asn mutation has been isolated in curated virus strains that were in circulation before use of adamantanes, and so the prevalence of the Ser31Asn mutation in the wild may not be exclusively the result of drug selection pressure. This contrasts with mutations such as Val27Ala which appear more likely than Ser31Asn to have evolved as a result of drug selection pressure (Furuse et al., 2009a, 2009b).

The effects of the Ser31Asn mutation on M2 pore size and proton conduction have been elucidated in exquisite detail. For example, the solution NMR structure (PDB: 2RLF; Schnell and Chou, 2008) presents Ser31 between two alpha-helices, and its mutation to the larger asparagine expands the pore by creating a kink in each monomer which would be expected to result in higher conductance. In contrast, the crystal structure (PDB: 3C9J; Stouffer et al., 2008) shows Ser31 facing the pore, where mutation to asparagine decreases the pore size and would be expected to reduce conductance. However, Holsinger et al. (1994) did not observe a difference in the rate of proton conduction due to the Ser31Asn mutation as measured by single electrode electrophysiology recordings. This result was further confirmed by Balannik et al. (2010) in two-electrode voltage clamp assay in which S31N mutant had indistinguishable specific conductance as M2 WT. Regardless, being located at a pore lining residue, the Ser31Asn mutation is likely to alter the diameter as well as the polarity and dynamics of the channel pore, resulting in changes between the observed interactions of the Ser31Asn channel and amantadine when compared to WT M2 (Gleed et al., 2015). Interestingly, in the solved X-ray crystal structure for M2 containing the Ser31Asn mutation (PDB: 5C02; (Thomaston and DeGrado, 2016; Thomaston et al., 2018, 2019), in the absence of a drug molecule in the pore and in the Inwardopen state, Asn31 was found to face the pore and was stabilized by H-bonds formed with the carbonyl groups of neighboring Asn31 (Fig. 3A). This arrangement sterically constricts the adamantane binding site in the pore. In contrast, in the Inwardclosed state, Asn31 was tucked away from the centre of the pore and stabilized by forming H-bonds with carbonyls at the monomer-monomer interface, thereby twisting the helices and narrowing the pore near the binding site (Fig. 3B). Alternatively, using a solution NMR structure Wang et al., 2013a, Wang et al., 2013b demonstrated that the Asn31 carboxamide could be stabilized in a pore-facing conformation through a bidentate interaction with the M2-S31N inhibitor M2WJ332 (compound 4; Table 2; Fig. 6 B). These observations further emphasize that the orientation of Asn31 in the pore is dependent on the conformational state of the protein and/or the presence of interacting drug molecule in the lumen of the pore.

Fig. 6

Structures of M2 in complex with channel blockers. (A) X-ray crystal structure of M2-WT (22–46) in complex with amantadine (1) (PDB: 6BKK; Thomaston et al., 2018). (B) Solution NMR structure of M2-S31N (19–49) in complex with (4) (PDB: 2LY0 (Wang et al., 2013a, Wang et al., 2013b);). (C) X-ray crystal structure of M2-V27A (22–46) in complex with spiro-adamantyl amine (9) (PDB: 6NV1; Thomaston et al., 2020).

Development of inhibitors to drug-resistant M2

An extensive combination of molecular dynamics simulations (Khurana et al., 2011; Wang et al., 2011a), X-ray crystallography (Acharya et al., 2010; Stouffer et al., 2008), and NMR spectroscopy (Cady et al., 2011a, Cady et al., 2011b; Wang et al., 2009, 2011a, 2013a) along with iterative cycles of medicinal chemistry, electrophysiological testing, and antiviral assaying have led to the discovery of several classes of compounds that inhibit at least one of the three major drug-resistant mutations (Leu26Phe, Val27Ala, or Ser31Asn) with efficacies comparable or better than amantadine against WT M2. While several other compounds are reported throughout the literature to inhibit M2 Ser31Asn-containing viruses, the mechanism of action of many of them is not confirmed to be mediated through M2 as many of these compounds were not tested in M2-specific assays. The most promising compounds have therefore been validated by both antiviral and M2 conductance assays, the vast majority of which have been assessed by TEVC. These “second-generation” inhibitors of drug-resistant M2 can be broadly categorized in two main groups: adamantanes (Table 2) and a series of chemically diverse non-adamantane molecules (Table 3). While numerous compounds are reported by many groups, in addition to those cited in Table 2, Table 3, we have highlighted particular compounds with electrophysiology-based IC50s and PRA-based EC50 s at low or sub-micromolar concentrations, which approximate the activities of amantadine and rimantadine in these assays with wild-type M2. Clearly, the efficacy of a given inhibitor as measured by TEVC and PRA does not necessarily translate to efficacy in vivo. However, a subset of these inhibitors has recently advanced to preclinical and animal efficacy studies, where promising results are reported. Additional compounds that inhibit WT M2, in addition to early but weaker inhibitors of M2 Val27Ala and Ser31Asn, have been reviewed elsewhere (Duque et al., 2011; Wang et al., 2015).

Adamantane-based inhibitors

Among the many molecules synthesized and tested by the DeGrado group, compounds 3 and 4 emerged as the first promising candidates of M2 Ser31Asn inhibition. TEVC experiments revealed 50% inhibitory concentration (IC50) values of 14 and 16 μM for 3 and 4, respectively, against the Ser31Asn form of M2, which was comparable to amantadine's inhibition of WT M2 in the authors' hands (16 μM; Wang et al., 2013a). Moreover, compound 4 bound to Ser31Asn M2 with such high affinity that structure determination by solution NMR was possible. In their structure model (PDB: 2LY0), 4 was found to be clamped between the side chains of three Asn31 residues while having its amine group pointing towards the M2 N-terminus (Fig. 6B). The adamantane cage fit within the hydrophobic pocket between Asn31 and Gly34, and the positively charged ammonium group of the adamantane formed hydrogen bonds via water molecules with two asparagines, while the third asparagine and the nitrogen of isoxazole of 4 formed a bidentate interaction. The side-chain carbonyl of Asn received a hydrogen bond from the ammonium group, and the carboxamide of the same Asn donated a hydrogen bond to the endocyclic nitrogen of isoxazole. On the other end of compound 5, the thiophene moiety sit against four methyl groups of Val27 and was stabilized by hydrophobic interactions. Notably, this distal hydrophobic group could be substituted by other similar moieties such as benzene (3) and cyclopropane (5) without substantially compromising efficacy. Molecular dynamics experiments also supported the potential drug-protein interactions observed in NMR experiments. Numerous additional inhibitors of M2 Ser31Asn have since been reported by the DeGrado and Wang laboratories: for example, compound 6 inhibited 74.4% of M2 Ser31Asn activity by TEVC and virus replication with a 50% effective concentration (EC50) of 1.7 μM (Li et al., 2016a, 2016b; 2017) in PRA. Notably, compound 6 also inhibited 12% of WT M2 channel activity by TEVC at 100 μM (Li et al., 2016a).

Additional adamantane-based compounds have also been reported to inhibit multiple forms of M2. For example, Wang et al. (2013b) proposed that installation of aromatic groups to the amine group of amantadine might also improve inhibitory activity against Ser31Asn M2 while maintaining activity against WT M2. This design hypothesis lead to the synthesis and discovery of benzyl-substituted amantadine derivatives, among which a benzdiol (7) was reported as the first potent WT and Ser31Asn M2 dual-inhibitor with IC50 values of 60 and 35 μM, respectively, as measured by TEVC. By plaque reduction assay, 7 further showed a EC50 of 3.2 μM against the A/WSN/33 M2 Ser31Asn strain of IAV and complete inhibition of WT (Ser31 M2-containing) virus at 1 μM. Another series of dual inhibitors, exemplified by compound 8, inhibited 77 and 76% of currents from both M2 WT and Ser31Asn, respectively, as measured by TEVC. Compound 8 additionally inhibited amantadine sensitive and resistant viruses with EC50 values of 4.6 and 1.8 μM, respectively, as measured by PRA.

Wang et al. (2011a) also reported design of a spirane-adamantane derivative (9; Fig. 6C) that inhibited WT (IC50 = 18 μM), Leu26Phe (IC50 = 6 μM), and Val27Ala (IC50 = 0.3 μM) forms of M2, as measured by TEVC, which in turn were comparable to amantadine's activity against WT channel in this study (IC50 of 15.7 μM). Additionally, based on their previous findings in identifying a dual inhibitor of WT and Val27Ala M2 (10), Hu et al. (2017b) designed and reported the first class of organosilanes that exhibited potent antiviral activity against amantadine-resistant and oseltamivir-resistant viruses. Their most potent organosilane (11) was able to inhibit M2 Ser31Asn currents in TEVC experiments by ~86% at 100 μM and inhibited an A/WSN/33 (H1N1) virus encoding M2 Ser31Asn with an IC50 of 0.4 μM (Hu et al., 2017b). Barniol-Xicota et al. (2017) also reported a series of dual M2 WT and Val27Ala inhibitors. However, while both compounds 12 and 13 were potent inhibitors of both the WT M2 and M2 Val27Ala, with IC50s ranging from 1.9 to 16.2 μM by TEVC, only 12 exhibited any antiviral activity (EC50 = 0.14 μM; Barniol-Xicota et al., 2017), while 13 also exhibited toxicity in MDCK cells (50% cytotoxic concentration (CC50) = 10 μM). The authors proposed that the antiviral activity of these dual inhibitors is highly dependent on “slow and steady” binding of inhibitors, where a kd of 10−6 or less is needed to be observed by TEVC.

An important question is whether these second-generation inhibitors are capable of giving rise to additional drug-resistant forms of M2 and influenza viruses with comparable replicative fitness. In beginning to address this, the Wang group first reported that in vitro passaging of influenza virus in the presence of increasing concentrations of the M2 Ser31Asn inhibitor 5 resulted in a novel M2 mutation, Leu26Ile, which conferred drug resistance. This mutation was also co-observed with an occasional Asn31Ser mutation which reverted back to Asn31 upon drug withdrawal. In contrast, passaging with the dual inhibitor compound 8 resulted in Asn31Asp and Ile32Thr mutations which also reverted after drug withdrawal (Ma et al., 2016). Both sets of mutations were confirmed to be resistant to their respective selecting drugs by TEVC.

The Wang group have also reported results from in vitro passaging of two M2-S31N inhibitors 14 and 15, which potently inhibited both M2 current activity by TEVC (75.5 and 84.3% at 100 μM) and adamantane-resistant viruses (EC50s = 1.2 and 0.3 μM, respectively; Li et al., 2017; Wang et al., 2018; Musharrafieh et al., 2018). Remarkably, unlike amantadine, which readily gave rise to resistance after a single passage in vitro, compounds 14 and 15 required 4–5 passages before two resistance mutations (Val27Ile and Leu26Ile respectively) were detected in vitro, suggesting a higher genetic barrier to resistance. Interestingly, reversion of M2 back to Ser31 was not observed. Further passaging of Ser31Asn + Leu26Ile virus with 15 at higher concentrations selected for a third Ala30Thr. Notably, while both M2 and viruses containing Ser31Asn + Val27Ile or Leu26Ile exhibited similar proton conductance properties and viral replication fitness relative to the single Ser31Asn mutant, the addition of Ala30Thr to Ser31Asn and Leu26Ile resulted in substantially reduced M2 proton conductance and viral replication. These observations may suggest that the viable evolutionary space of M2 in response to selection with these compounds may be limited.

In further M2 resistance studies, Compound 16 was also reported as a potent inhibitor of both M2 Ser31Asn by TEVC (47.9% inhibition at 100 μM, although its proposed slow-binding kinetics may underestimate its efficacy) and in vitro virus replication (EC50 as low as 0.2 μM; Wang et al., 2018) Serial passaging of virus in the presence of increasing concentrations of 16 resulted in a novel and unexpected drug-resistance mutation, Leu46Pro, positioned at the distal end of the M2 transmembrane helix. Interestingly, Leu46Pro did not significantly change proton conductance or amantadine sensitivity on its own in WT M2. However, in the presence of Ser31Asn, the Leu46Pro mutation rendered compounds 14–16 ineffective in TEVC experiments. Through molecular modelling and molecular dynamics simulations, the authors proposed that Leu46Pro, despite being located outside of the canonical M2 binding sites of adamantanes, could affect the size of the M2 pore where compounds 14–16 were proposed to interact (Musharrafieh et al., 2019). Taken together, these results suggest that when compared to amantadine and WT M2, the genetic barrier for drug resistance is likely to be higher for adamantane-derived compounds that target M2 Ser31Asn, at least in vitro.

The preclinical and in vivo potential of M2 Ser31Asn-inhibiting adamantane derivatives are now beginning to be elucidated. For example, Compounds 5, 8, 10, and 16 have been reported to inhibited virus strains with resistance to the licensed neuramindase inhibitor oseltamivir (Hu et al., 2017a; Ma et al., 2016; Wang et al., 2018). In oseltamivir-sensitive viruses, 5 and 16 also synergized with oseltamivir, raising the possibility of combination therapies where lower doses of drugs can be used to minimize the risk of toxicities without sacrificing antiviral efficacy (Ma et al., 2016; Wang et al., 2018). Compound 16 also exhibited favorable in vitro pharmacokinetic properties such as half-life of at least 145 min in mouse and human liver microsome stability assays. It also exhibited good cellular permeability in Caco-2 cells (~28 × 10−6 cm/s) and did not inhibit a panel of five CYP enzymes. (Wang et al., 2018). Additional M2 Ser31Asn adamantane-derived inhibitors have shown similar supportive preclinical results, indicating that these properties are fairly common for this chemical class (Hu et al., 2018). Furthermore, injection of mice with the dual-inhibitor compound 9 at up to 100 mg/kg/day resulted in no changes in body weight. Importantly, at up to 100 mg/kg/day, compound 9 also rescued mice from lethal infection of viruses containing either WT M2 or M2 Val27Ala, thereby demonstrating in vivo efficacy (Hu et al., 2017a). While initial results are encouraging, in vivo results for most adamantane derivatives remain limited and await further study.

Non-adamantane-based inhibitors

Several non-adamantane chemical scaffolds have also been successfully explored for their ability to block adamantane-resistant M2. Examples of non-adamantanes that inhibit drug-resistant M2 are shown in Table 3.

A polycyclic pyrrolidine (17) reported by (Rey-Carrizo et al., 2013) was reported to inhibit both WT and Val27Ala M2 by TEVC, with IC50 values of 3 and 0.3 μM, respectively. Another related compound (18) was reported by the same authors to exhibit triple inhibitory efficacy by TEVC against WT, Leu26Phe, and Val27Ala forms of M2 with IC50 values of 18, 8.6, and 0.7 μM, respectively (Rey-Carrizo et al., 2014). Unfortunately, the cytotoxicity of these derivatives may impede their further development as therapeutic agents. In an independent attempt to change the shape and bulk of the adamantane cage, Zhao et al. (2012) reported the discovery of pinanamine derivatives as inhibitors of M2 Ser31Asn. Specifically, an imidazole derivative of pinanamine (19) showed a moderate inhibitory effect against A/WSN/33 (H1N1) in a cytopathic effect inhibition assay (CPE; EC50 = 95.5 μM). Dong et al. (2016) subsequently reported a related compound (20) with improved activity against both WT A/HK/68 (H3N2) as well as A/WSN/33 (H1N1) encoding M2 Ser31Asn with EC50s of 2.5 and 3.4 μM, respectively. In a separate endeavor, derivatives of spiranamine (21) were also explored (Balannik et al., 2009; Wang et al., 2011a). For example, a silaspirane derivative (22) exhibits activity against both WT and Val27Ala with 95% and 68% inhibition at 100 μM in TEVC experiments. (Wang et al., 2011b).

Another notably explored chemical class is exemplified by hexamethylene amiloride (HMA, 23), an amiloride-based compound which is reported to block viroporins and/or replication of IAV, hepatitis C virus, HIV-1, Dengue virus, and severe acute respiratory syndrome coronavirus (SARS-CoV; Gazina and Petrou, 2012; Pervushin et al., 2009; Premkumar et al., 2004). However, HMA has several off-target effects and is cytotoxic at low micromolar concentrations (Jalily et al., 2016; Premkumar et al., 2004). Recently, a series of HMA derivatives were designed through iterative medicinal chemistry and electrophysiological testing approaches (Jalily et al., 2016). Among these compounds, 24 exhibited activity against WT M2 in single electrode voltage clamp (SEVC) experiments, which was comparable to amantadine in these studies (IC50 0.2 μM vs 0.6 μM), but was not obviously toxic to cells. While 24 was found to be active against WT M2 only, an elongated derivative (25) exhibited dual-inhibitory effects against WT and Ser31Asn M2 in SEVC (respective IC50s = 0.6 μM and 4.4 μM) and PRA (respective EC50s = 40 μM and 18 μM). Additionally compound 26, although inhibiting WT M2 with lower efficacy (20% inhibition at 100 μM) than 25 in SEVC, was more effective at inhibiting viruses encoding WT and Ser31Asn M2 in PRA (EC50s = 6.9 and 1.5 μM, respectively). Preliminary data indicate that 23 exhibits stability in human microsome assays (t1/2 = 70 min) and mouse plasma following injection in vivo (t1/2 = 60 min). Moreover, injection of compound 23 at 2 mg/kg did not affect the monitored ECG or hemodynamic parameters in mice, and 200 μM of cumulative dosing of 23 over a course of 3 h did not cause obvious cardiac toxicity (P. Jalily, I. Tietjen, M. Pourrier, and D. Fedida, unpublished data).

The inhibitory effect of Cu2+ on M2 conductance was initially studied by Gandhi et al. (1999), where they showed that the inhibition is bi-directional across the membrane and dependent on the presence of His37. The WT protein also exhibited high specificity for Cu2+ and was only partially inhibited by high concentrations (1 mM) of bivalent nickel, zinc, or platinum (Gandhi et al., 1999). More recently, Gordon et al. (2017) reported the discovery of divalent copper complexes such as compound 27 as non-selective M2 inhibitors. The authors proposed that the mechanism of action of these copper complexes is independent of mutation at Leu26, Val27, or Ser31 but is instead dependent on the conduction moiety, His37, as no inhibition was observed with 27 against His37Ala mutant. Compound 27 was able to inhibit 80% of M2 proton currents at 100 μM as measured by TEVC and had an EC50 of 0.7 μM against A/CA/07/2009 encoding M2 Ser31Asn in PRA.

Finally, Jang et al. (2018) reported that the monovalent ionophore salinomycin (28) was able to counteract the proton conductance function of both influenza A and B in vitro. However, it indirectly inhibits M2 channel function by preventing cellular endosome acidification, similar to the mechanism of chloroquine. Salinomycin inhibited proton conduction by both M2 WT and Ser31Asn incorporated into virus-like particles (54% and 72% vs. WT and N31 at 100 μM), in addition to proton conduction by the highly divergent M2 channel of influenza B (~60%). Salinomycin also synergized with oseltamivir in vitro. However, while salinomycin on its own exhibited limited antiviral activity in vivo; it did rescue mice treated with a sub-optimal concentration of oseltamivir and lethally-infected with oseltamivir sensitive or resistant viruses. (Jang et al., 2018).

Observations from drug-resistant M2 inhibitor studies

The two-electrode voltage clamp assay remains the most effective assay for M2 ion channel studies, as it is less technically demanding and more likely to be replicated across independent laboratory groups when compared to SEVC. In general, there is a liner correlation between the percentage channel blockage from TEVC and the antiviral activity (Li et al., 2017). However, we also note that the efficacies of certain M2 inhibitors as measured by TEVC and PRA can be discordant. For example, for compounds 4, 7, 9, 12, and 15, the IC50 values detected by TEVC are substantially higher than EC50 values by PRA. In contrast, the IC50 values detected for compounds 13, 24, and 25 by TEVC are lower than their PRA EC50 values. Discrepancies in both cases might arise from different binding kinetics of various M2 channel blockers (Wang et al., 2018). The IC50 values from TEVC or SEVC measurements are normally plotted using percentage inhibition at a given time point after compound perfusion, therefore the IC50 values may not accurately reflect the true binding potency of the compounds. Instead, kinetic measurements of the K d values are a more stringent way to determine the true binding potency of M2 channel blockers. However, kinetic measurements are labor intense and are often reserved as a secondary assay to characterize lead candidates that have already shown potent antiviral activity such as compound 16. Nevertheless, it is understandable that the lack of a straight linear correlation between the results from the in vitro TEVC or SEVC assay and the cellular CPE or PRA assay might also be additionally due to other factors such as a compound's membrane permeability, virus strain differences, or other off-target effects. However, this is not a specific issue related to M2 channel blockers per se, as other drug candidates commonly face this challenge as well. In some cases, discordance also extends to M2 proteins and viruses containing different mutations. For example, compound 9 inhibits M2 Val27Ala with lower IC50s than M2 WT (0.3 and 18 μM, respectively), while the EC50s in PRA are the opposite (i.e. 1.8 and 0.3 μM, respectively). Similar results are also seen for compounds 7, 10, and 20 and could reflect, for example, differences in virus strains used and/or subtle changes in the binding kinetics of these compounds against different M2 forms. Thus, as with other drug development efforts, interpretation of specific results for individual compounds must be performed carefully and on a case-by-case basis.

Emerging approaches for new M2 inhibitor discovery and development

The advent and improvement of electrophysiological techniques have substantially advanced our ability to directly measure the membrane currents from single cells expressing M2 or other viroporins of interest. The optimization of methods including TEVC and SEVC has also allowed us to study the fundamental biophysical properties of M2 variants and how small molecules engage and inhibit these channels. However, the inherent limitations of traditional electrophysiology such as low throughput, technical demands, and inaccessibility of the technique to many research groups have largely restricted screening and discovery of novel M2 inhibitors to a few specialized laboratories (Wang et al., 2015). Toward addressing these limitations, recently developed systems with the capacity or potential of automated electrophysiology screening could substantially reduce the time, complexity and cost associated with manual traditional patch clamp experiments, particularly in the context of SEVC. Notably, the companies Sophion (QPatch), Nanion (Patchliner) and Fluxion (IonFlux) have developed automated systems where screening of tens or hundreds of compounds could, in principle, be performed by SEVC in one day (Priest et al., 2017).

To enable high-throughput screening of M2 inhibitors with ready availability to more research groups, S. cerevisiae yeast strains have also been developed which are induced to express WT or adamantane-resistant M2 proteins in the presence of galactose. In these assays, M2 expression inhibits yeast growth, as measured by culture turbidity, which in turn is restored by administration of M2 inhibitors. As shown by Balgi and Roberge (2009), amantadine restored the growth of yeast expressing WT M2 from 27% of control to 95% at 1 μM. This assay was further optimized for high throughput screening: for example, the authors subsequently screened ~250,000 compounds and identified 21 new and potent inhibitors of WT M2 (Balgi et al., 2013). These reports demonstrate that growth restoration assays are a sensitive, economical and technically simple technique for high throughput screening for inhibitors of M2 and presumably other viroporins. A similar expression system using E. coli has also been used to assess the properties of random M2 mutations (Santner et al., 2018b), although results from these assays do not consistently agree with results obtained by TEVC (Musharrafieh et al., 2019; Santner et al., 2018a).

In a separate approach, M2 was incorporated into virus-like particles, and proton conductance was measured using a potentiometric fluorescent dye readout (Sulli et al., 2013). Using this technology, the authors successfully screened 107,572 compounds in 384-well format and discovered 19 new M2 WT-specific inhibitors with IC50 values ranging from 140 nM to 13 μM (Sulli et al., 2013), although no inhibitor was identified to target the drug resistant mutants of M2.

When designing drugs against viral targets, an inherent challenge is the higher degree of structural flexibility exhibited by viral proteins in comparison to many prokaryotic and eukaryotic targets (Das et al., 2010; Fischer and Hsu, 2011). Yet another challenge imposed against rational drug design is the dependence of such highly flexible proteins on their surrounding environment, which in turn affects their function. In contrast to thermostable proteins that encompass tight hydrophobic cores, viral proteins are frequently loosely packed, resulting not only in a smaller difference in energy levels between folded and unfolded states, but also a reduced contribution of random mutations or interactions with chemical inhibitors to overall stability (Paciaroni et al., 2002). This equips viruses with enormous adaptive mechanisms to buffer the deleterious effects of random mutations; even more in the case of RNA viruses which intrinsically mutate faster than DNA viruses (Tokuriki et al., 2009). These challenges also apply to M2, where until recently the lack of high-resolution structures had impeded the advancement of rational drug design. Other obstacles that have prevented the successful design of a universal M2 inhibitor include the constricted drug binding site, technical difficulties in effective measurement of drug inhibition, and absence of reliable protocols to assess protein drug interactions.

To overcome these challenges, computation now plays a critical role in the field of rational drug design, as computers now have sufficient processing power to run virtual screening of large chemical libraries against multiple targets. While these strategies can identify multiple antiviral leads including putative viroporin inhibitors (Radosevic et al., 2019; Tietjen et al., 2015), the likelihood of false-positive hits necessitates their validation by experimental techniques. Another role of computation is now the extended ability to run complex molecular dynamics simulations in shorter times and for larger complexes to predict and visualize the interaction of viral protein domains and putative inhibitors, complexes of proteins, and simulation of complexes of viral proteins and cellular factors.

Taken together, several emerging technologies are now available to design and assess the efficacy and activity of novel M2 inhibitors. However, it is important to note that both TEVC and PRA remain “gold standards” for these studies and should be applied as secondary assays to rule out false positives obtained from higher-throughput methods.

Concluding remarks

The next influenza virus pandemic, albeit unpredictable, is not a matter of “if” but “when.” New therapeutics, especially those that are distinct from current antivirals and with the ability to inhibit existing drug-resistant viral strains, may be essential toward mitigating the next inevitable outbreak. The M2 viroporin, a proton-dependent proton channel required for virus entry and egress, has historically been an effective antiviral target for drugs like amantadine and rimantadine. However, mutations in M2, particularly lining the pore where adamantanes interact, are now widespread and have rendered these therapies ineffective. Over the past decade, new inhibitors of drug-resistant M2 consisting of both adamantane derivatives and new chemical scaffolds have been found that inhibit M2 ion conduction, as demonstrated using both electrophysiological techniques in vitro influenza replication assays. The most promising leads also inhibit multiple adamantine-sensitive and resistant forms of M2 (e.g., compounds 7–10, 12, 17–18, 22, 25), inhibit viruses with major resistance to other IAV antivirals like oseltamivir (5, 8, 10, 16) and/or synergize with oseltamivir (5, 16, 28), exhibit high genetic barriers to resistance following long-term in vitro passaging (5, 8, 14–16), have supportive preclinical parameters such as good stability and low toxicity (9, 16, 23), and rescue mice from lethal infection (9). Going forward, further discovery of new chemical scaffolds and/or optimization of existing leads by medicinal chemistry may be enabled by emerging technologies to screen for and monitor M2 conductance. In addition, demonstration of clinical efficacy of new chemical leads remains to be achieved. The coming decade will likely show whether new M2 inhibitors can significantly add to the evolving armamentarium of antivirals to combat current and emerging seasonal and pandemic influenza strains.