Investigation of the Drug Resistance Mechanism of M2-S31N Channel Blockers through Biomolecular Simulations and Viral Passage Experiments

Abstract

Recent efforts in drug development against influenza A virus (IAV) M2 proton channel S31N mutant resulted in conjugates of amantadine linked with aryl head heterocycles. To understand the mechanism of drug resistance, we chose a representative M2-S31N inhibitor, compound 3, as a chemical probe to identify resistant mutants. To increase the possibility of identifying novel resistant mutants, serial viral passage experiments were performed with multiple strains of H1N1 and H3N2 viruses in different cell lines. This approach not only identified M2 mutations around the drug-binding site, including the pore-lining residues (V27A, V27F, N31S, and G34E) and an interhelical residue (I32N), but also a new allosteric mutation (R45H), in addition to L46P previously identified, located at the C-terminus of M2 that is more than 10 Å away from the drug-binding site. The effects of each mutation were next investigated using electrophysiology, recombinant viruses, and molecular dynamics (MD) simulations. The reduced sensitivity in channel blockage correlated with increased drug resistance in antiviral assays using recombinant viruses. The MD simulations show that the V27A, V27F, G34E, and R45H mutations increase the diameter and hydration state of the pore in complex with compound 3. The Molecular Mechanics Generalized Born (MM-GBSA) calculations result in more positive binding free energies for the complexes of resistant M2 (V27A, V27F, G34E, R45H) with compound 3 compared to the stable complexes (S31N and I32N). Overall, this is the first systematic study of the drug resistance mechanism of M2-S31N channel blockers using multiple viruses in different cell lines.

M2 is an acid-activated proton channel essential for influenza A virus (IAV) replication.¹ The IAV M2 forms homotetramers where each subunit consists of a small, unstructured N-terminal domain, a transmembrane (TM) helix constituting the channel pore, and an intraviral region consisting of an amphiphilic helix and an unstructured tail. After viral entry, endosomal acidification activates M2 via protonation of histidine residues at position 37, opening the tryptophan gate at residues 41 to allow shuttling of protons to the virion core (Figure 1A). M2 also neutralizes the pH in the Golgi network in order to prevent premature conformational changes in the viral proteins. The channel has also been suggested to play a role in driving membrane curvature for viral budding.²

Figure 1.

(A) M2 is a pH-activated channel that conducts protons into the virion after acidification of endosomes. (B) Structure of M2TMC (TM plus cytoplasmic tail) (PDB ID 2L0J) and sequence of the WT M2TM from A/Udorn/1972 strain with residues associated with drug 3 resistance colored in red. (C) Chemical structure of Amt (1) and the isoxazole-containing M2-S31N inhibitor M2WJ332 (2) and compound 3. (D, Left) Snapshot of M2TMC-S31N in complex with 3 after 100 ns of MD simulation (ligand carbons are shown in orange, protein in gray ribbon, main chain of residues V27, A30, N31, H37 in magenta ribbon and side chain in sticks). (D, Right) 2D diagram of the most important interactions between compound 3 and M2TMC-S31N observed from the 100 ns MD simulations trajectory (hydrogen-bond contacts between NH₂⁺ and isoxazole ring of the ligand and N31 side chains are shown with blue spheres and between hydroxyl of the ligand and A30 backbone carbonyl with green spheres; gray spheres indicate van der Waals contacts).

Antivirals are an essential component in the management of IAV outbreaks, but the emergence of drug resistance is an ongoing challenge.³⁻⁵ One of the first FDA-approved antiflu drugs available was the WT M2 inhibitor amantadine (Amt, (1)) (Figure 1), but this drug was discontinued due to the rise of resistance within circulating strains. Resistance to WT M2 drugs is caused by mutations in the TM domain of the M2 protein (M2TM). The fine-tuned sequence of M2 places constraints on the types of drug-resistant mutations allowed.^6,7 Amino acid substitution L26F, V27A, A30T, G34E, and S31N confer cross-resistance to amantadine (Figure 1B).^6,8−14 However, the vast majority (95%) of resistant viruses bear the S31N substitution in M2, while 1% is V27A, L26F, A30T, and G34E.^4,8

Structure–activity relationship (SAR) studies coupled with electrophysiological (EP) and antiviral assays led to the discovery of potent blockers of M2-S31N mutant such as M2WJ332 (2) and compound 3 (Figure 1), most of which contain an aryl headgroup linked to an Amt analogue through a methylene bridge (Figure 1C).¹⁵⁻²² These compounds are positioned inside the M2-S31N pore with the aryl head facing toward the N-terminus of the channel according to solution and solid-state NMR (ssNMR) as well as MD simulations (Figure 1D).^15,17,23 These studies show that compounds are trapped by the V27 side chains at the N-terminus of the M2-S31N pore.²³ Recent findings from MD simulations and ssNMR spectra revealed that channel blockers bind to the M2TM-WT and M2TM-S31N mutant in flipped orientations; Amt binds to M2TM-WT with its positively charged ammonium facing down toward the C-terminus of the channel, while M2TM-S31N inhibitors bind to the channel with its positively charged ammonium facing up toward the N-terminus of the channel.²⁴⁻²⁶

Understanding the mechanism of drug resistance helps guide the development of antivirals and treatment strategies. Although previous studies have investigated the drug resistance mechanism of M2 channel blockers, most of these studies were performed using one virus strain in one cell line. It is unknown whether resistance might be viral strain-dependent or cell type-dependent and to our knowledge, no such systematic studies have been reported. Accordingly, we chose the representative M2-S31N inhibitor 3, and performed drug resistance selection experiments with a panel of clinically relevant Amt-resistant H1N1 and H3N2 IAV strains in two cell lines (MDCK and A549). In addition to the allosteric mutation L46P that was selected with A/California/07/2009 (H1N1) virus in MDCK cells that we studied previously,²⁷ this study revealed six additional mutations that cause resistance to 3 located at three distinct regions of the M2 channel: pore-lining residues (V27A, V27F, N31S, and G34E), an interhelical residue (I32N), and a C-terminal residue (R45H). Using EP, antiviral assays, MD simulations, and MM-GBSA calculations, we subsequently evaluated the functional and structural properties of these mutant channels as well as their sensitivity toward drug inhibition. In general, mutations at the drug-binding site had the most profound effect on drug efficacy, whereas interhelical and C-terminal mutations moderately perturbed drug binding and inhibition, but was able to obstruct effective blockage by causing a conformational change that led to opening of the pore. Our approach provides a general strategy of studying drug resistance against influenza A M2 by using the selective pressure of drug application in multiple viral strains and cell types to increase the probability of identifying novel drug-resistant mutants.

Methods

Virus Strains and Antiviral Assays

IAV strains A/California/07/2009 (H1N1), A/Washington/29/2009 (H1N1), A/WSN/1933 (H1N1), A/Wisconsin/67/2005 (H3N2), and A/Victoria/361/2011 (H3N2) were used in this study.

Serial passages were performed as previously described²⁸ in either MDCK or A549 cells. For MDCK and A549 cells, a multiplicity of infection (MOI) of 0.001 and 0.01 or 0.1 was used, respectively. All viruses were amplified in either MDCK cells or MDCK cells overexpressing β-galactoside α-2,6-sialyltransferase 1 (ST6Gal I).

Plaque reduction assays were used to determine viral titer or antiviral activity. MDCK cells (100% confluent) were washed with PBS containing magnesium and calcium and infected with virus diluted in Dulbecco’s modified Eagle’s medium (DMEM) with 0.5% BSA. Virus infection was synchronized at 4 °C, then placed in a 37 °C, 5% CO₂ incubator for 1 h. Virus was then removed and a prewarmed media overlay (DMEM, 2 μg/mL N-acetyl trypsin, 1.2% Avicel microcrystalline cellulose (FMC Biopolymer, Philadelphia, PA) was gently added to each well. After 2–3 days of incubation, overlay was removed and cells were stained with 0.2% crystal violet dye solution. Plaque area in each well was quantified using ImageJ.

Recombinant A/Udorn/1972 (rH3N2) were generated using pHH21 plasmids containing each of the eight gene segments (NS, M, NA, NP, HA, PA, PB1, and PB2) as well as four pcDNA plasmids containing the polymerase complex proteins (PB1, PB2, PA, and NP). Mutations were introduced into the M segment pHH21 plasmid using the QuikChange Site-Directed Mutagenesis Kit (Agilent, Santa Clara, CA). HEK293T cells were transfected with all 12 plasmids using TransIT reagent (Mirus Bio LLC) in Opti-MEM (Gibco) and incubated overnight at 37 °C. The next day, the media of transfected cells was exchanged with Opti-MEM supplemented with 2 μg/mL N-acetyl trypsin and incubated overnight. Cells were then resuspended and cocultured with MDCK cells. Harvested recombinant viruses were amplified in MDCK cells.

Two-Electrode Voltage Clamp (TEVC) Assay

Xenopus laevis frog oocytes were microinjected with RNA transcripts of M2-S31N or its mutants as reported before.²⁸ Percent inhibition of compound 3 for each mutant was tested in triplicates by recording the current after a 2 min drug incubation of 100 μM at pH 5.5. For kinetic values (K_d) a pH 5.5 pulse was applied to limit the duration each oocyte spent at low pH.

Methods for MD simulations are in the Supporting Information.

Results and Discussion

Drug Resistance Selection

We selected a representative M2-S31N inhibitor, compound 3, to identify the novel drug-resistant mutants in M2. A panel of IAV H1N1 (A/California/07/2009; A/Washington/29/2009; A/WSN/1933) and H3N2 (A/Wisconsin/67/2005; A/Victoria/361/2011) were used for drug resistance selection. All viruses contained the S31N mutation, which are resistant to Amt (1) but can be potently inhibited by compound 3. EC₅₀ values for these P0 (Passage 0 of the ligand) viruses were between 0.4 and 1.3 μM, consistent with what has been reported previously.²⁰ Viruses were passaged in both MDCK and A549 cells, except for A/Washington/29/2009 (H1N1) and A/Victoria/361/2011 (H3N2), which were passaged in MDCK cells only because viral titers were too low in A549. Table 1 summarizes the results from the drug resistance selection experiments.

Table 1. Serial Viral Passages of M2-S31N Inhibitor Compound 3.

	cell lines
	MDCKa				A549b
IAV Strain	passage	[3] (μM)c	EC50 (μM)d,e	mutation	[3] (μM)c	EC50 (μM)d,e	mutation
A/California/07/2009 (H1N1)	0	N/A	0.8 ± 0.05	WT	N/A	0.8 ± 0.05	WT
	1	0.75	N.D.	N.D.	0.75	N.D.	N.D.
	2	1.5	1.5 ± 0.4	N.D.	1.5	N.D.	N.D.
	3	3	1.1 ± 0.3	N.D.	3	>30 (resistant)	V27A
	4	6	0.7 ± 0.4	N.D.
	5	12	>30 (resistant)	L46P
A/Washington/29/2009 (H1N1)	0	N/A	0.4	WT
	1	0.75	N.D.	N.D.
	2	1.5	N.D.	N.D.
	3	3	N.D.	N.D.		N.T.
	4	6	N.D.	N.D.
	5	12	2.0 ± 0.7	N31S + S31N
	6	24	N.D.	N31S
A/WSN/1933 (H1N1)	0	N/A	0.8 ± 0.1	WT	N/A	0.8 ± 0.1	WT
	1	0.75	N.D.	N.D.	0.75	N.D.	N.D.
	2	1.5	N.D.	N.D.	1.5	N.D.	N.D.
	3	3	N.D.	N.D.	3	>30 (resistant)	I32N
	4	6	N.D.	N.D.
	5	12	>30 (resistant)	G34E + G34
A/Wisconsin/67/2005 (H3N2)	0	N/A	1.2 ± 0.05	WT	N/A	1.20 ± 0.046	WT
	1	0.75	N.D.	N.D.	0.75	N.D.	N.D.
	2	1.5	N.D.	N.D.	1.5	N.D.	N.D.
	3	3	3.3 ± 0.2	N.D.	3	1.1 ± 0.2	N.D.
	4	6	N.D.	N.D.	6	N.D.	N.D.
	5	12	N.D.	N.D.	12	N.D.	N.D.
	6	24	>30 (resistant)	R45H	24	>30 (resistant)	V27F
A/Victoria/361/2011 (H3N2)	0	N/A	1.3 ± 0.1	WT
	1	0.75	N.D.	N.D.
	2	1.5	N.D.	N.D.
	3	3	1.9 ± 0.07	N.D.		N.T.
	4	6	N.D.	N.D.
	5	12	N.D.	N.D.
	6	24	>30 (resistant)	G34E

^aConfluent MDCK cells or MDCK cells overexpressing ST6Gal I were infected with the indicated virus at MOI 0.001.

^bConfluent A549 cells were infected with the indicated virus at MOI 0.1 for A/California/07/2009 (H1N1) and MOI 0.01 for A/WSN/1933 (H1N1) and A/Wisconsin/67/2005 (H3N2).

^cViruses were propagated for up to 3 days in the presence of compound 3 at the indicated concentrations (μM).

^dEC₅₀ = mean ± SD of two biological duplicates.

^eDrug sensitivity was determined by plaque assay as previously described. N.D., not determined; N.T., not tested.

Previously, A/California/07/2009 (H1N1) virus was passaged in MDCK cells which led to the identification of the drug-resistant L46P mutation.²⁷ Interestingly, when A/California/07/09 (H1N1) virus was passaged in A549 cells, a different mutation V27A emerged at passage 3. This cell type-dependent resistance evolution was similarly obtained with two other influenza viruses: A/WSN/1933 (H1N1) and A/Wisconsin/67/2005 (H3N2). For A/WSN/1933 (H1N1) passage in MDCK cells, resistance was observed at P5 and the emergence of G34E was discovered. When this virus was passaged in A549 cells, we observed resistance at P3 and identified the I32N mutation in the M2 gene segment. A/Victoria/361/2011 (H3N2) appeared resistant at P6, and the G34E mutation was identified at this passage. A/Wisconsin/67/2005 (H3N2) developed drug resistance at P5 in both cell lines, and sequencing revealed the C-terminus mutation R45H and the pore-lining V27F mutation from MDCK and A549 cell passages, respectively.

The H1N1 strain A/Washington/29/2009 was passaged 6 times in MDCK cells, and the EC₅₀ value at P5 increased more than 4-fold. The sequences of the M2 gene segment from P5 and P6 indicated the emergence of the N31S mutation. This is interesting as it suggests that M2-S31N can revert to M2-WT (N31S) and become drug-resistant. Overall, six mutations V27A, V27F, N31S, I32N, G34E, and R45H were identified during the current passage experiments.

Functional Inhibition of Drug-Resistant M2 Channels

To determine the functional properties of these drug-resistant mutants, we expressed each mutant M2 channel in Xenopus laevis frog oocytes for two-electrode voltage clamp (TEVC) measurements. Mutations were introduced into the M2-S31N channel gene of A/California/07/2009 (H1N1) using site-directed mutagenesis, and the expression levels of M2 on the oocyte plasma membrane were detected by immunofluorescence using anti-M2 antibody.

M2 conductance for each mutant was determined by analyzing the inward current after oocytes were bathed in buffer at pH 5.5 to activate the channel (Figure 2). The mutations V27A, V27F, and I32N showed similar functional properties as M2-S31N (Figure 2A–D). Interestingly, it was previously found that the single mutation V27F (S31) resulted in nonfunctional channels,⁷ and the introduction of S31N mutation may therefore be compensatory. Conversely, the expression of the G34E mutation resulted in a lower resting potential as well as aberrant morphologies in the oocytes, possibly caused by increased leakiness at high pH (8.5). Nevertheless, we were able to obtain one recording trace after several attempts (Figure 2E). Finally, the R45H mutation had comparable current recordings as the M2-S31N when the channel was activated at pH 5.5 (Figure 2F). However, when the bathed solution was exchanged to pH 8.5 (high pH outside the oocyte, low pH inside) we observed outward or reverse proton current (Figure 2F). This may be due to the weakening of a salt bridge between residue D44 and R45 that helps create a tighter seal in the W41 gate. This finding is also observed when D44 is mutated to either alanine, cysteine, or asparagine, but not when R45 is mutated to cystine.²⁹

Figure 2.

Representative electrophysiology recording traces of M2 mutant variants with the application of compound 3. On the left side of each trace, the M2 variant is labeled; on the top of each trace, the scheme of solution applied to the testing oocyte are depicted. The decrease of inward current in the presence of 100 μM compound 3 at pH 5.5 showed the inhibition of different M2 variants by compound 3. (A) M2-S31N. (B) M2-S31N/V27A. (C) M2-S31N/V27F. (D) M2-S31N/I32N. (E) M2-S31N/G34E. (F) M2-S31N/R45H. In the M2-S31N/R45H trace (F), two arrows show the outward current when the perfusion solution is switched from pH5.5 to pH8.5.

Consistently, compound 3 showed 54.5 ± 2.6% inhibition against M2-S31N channel after 2 min of incubation at 100 μM drug concentration (Table 2) (Figure 2A). M2 variants with mutations at the pore-lining residues V27A, V27F, and G34E were less susceptible to compound 3 blockage, having 8.1 ± 2.0%, 20.4 ± 5.4%, and 28.2 ± 27.5% inhibition, respectively (Figure 2B,C,E). Interestingly, M2 variants with the interhelical mutation I32N and C-terminal R45H mutant remained susceptible, with 72.2 ± 2.2 and 33.7 ± 1.3% inhibition, respectively (Figure 2D,F).

Table 2. Effects of M2 Mutants on Drug Binding Investigated by Reverse Genetics, EP, Antiviral Assays, and MD Simulations.

M2 variantsa	% inhibition (2 min, 100 μΜ)b	Kd (μΜ)c	EC50 (μΜ)d	RMSDproteine	RMSDligandf	res27–res27	ΔGeffg
S31N	54.5 ± 2.6	5.5 ± 0.3	0.9 ± 0.1	3.17 ± 0.33	2.37 ± 0.40	5.94 ± 1.19	–51.44 ± ± 2.49
S31N V27A	8.1 ± 2.0	240.7 ± 61.8	>30	3.42 ± 0.38	2.14 ± 0.35	8.29 ± 0.94	–44.02 ± 4.30
S31N V27F	20.4 ± 5.4	>300	>30	3.43 ± 0.58	2.62 ± 0.33	8.71 ± 0.91	–47.40 ± 3.59
S31N I32N	72.2 ± 2.2	20.6 ± 5.7	4.9 ± 2.7	2.28 ± 0.22	2.83 ± 0.56	6.68 ± 1.43	–52.46 ± 4.17
S31N G34E	28.2 ± 27.5	n.a.	>30	4.31 ± 0.54	4.31 ± 0.54	12.69 ± 2.13	–47.10 ± 3.48
S31N R45H	33.7 ± 1.3	118.3 ± 22.2	6.2 ± 2.4	3.52 ± 0.60	3.11 ± 0.59	8.39 ± 3.82	–43.51 ± 3.39

^aRecombinant viruses were generated using A/Udorn/1972 (H3N2) background via reverse genetics.

^bEP studies use A/California/07/2009 (H1N1) M2 (S31N protein) variants; mean ± SEM calculated at 2 min of drug perfusion.

^cCalculated from the ratio, k_off/k_on.

^dPlaque reduction assay.

^eMean ± SD (Å). Protein RMSD is calculated for the C_α atoms of the α-helices, for the last 50 ns of the trajectories. Frame 0 is used as reference structure.

^fMean ± SD (Å). Ligand RMSD is calculated after superposition of each protein–ligand complex to that of frame 0 (reference structure) based on the Cα atoms of the protein, for the last 50 ns of the trajectories.

^gEffective binding free energy calculated as ΔG_eff = ΔE_MM + ΔG_sol (see Methods section).

To characterize the binding kinetics of compound 3 for each mutant, we obtained K_d values using a washout protocol as previously described.²⁸ As expected, the K_d value for M2-S31N was found to be 5.5 ± 0.3 μM (Table 2). The pore-lining mutations V27A and V27F had significantly higher K_d values of 240.7 ± 61.8 μM and >300 μM, respectively. We were not able to obtain K_d values for G34E mutation because expression of this mutant destabilized oocytes. The interhelical mutation I32N had a K_d value of 20.6 ± 5.7 μM, a 3.7-fold increase compared to the M2-S31N blockage. For the C-terminal mutation R45H, a K_d value of 118.3 ± 22.2 μM was obtained, which was lower than the values obtained for the pore-lining mutations V27A and V27F. Overall, the functional studies reveal differences in the binding properties for pore-lining, interhelical, and C-terminal mutations in the M2-S31N channel.

To confirm that these M2 mutations are indeed associated with phenotypical drug resistance in cell culture, each mutant was introduced into reverse engineered viruses with the A/Udorn/1972 (H3N2) background. Briefly, V27A, V27F, I32N, G34E, and R45H were introduced into a pHH21 plasmid containing the M2-S31N using site-directed mutagenesis and transfected in order to generate infectious recombinant viruses. EC₅₀ values were determined for each recombinant virus by titrating compound 3 in plaque reduction assays (Table 2). V27A, V27F, and G34E were completely resistant, with EC₅₀ values greater than 30 μM. The mutations I32N and R45H were still partially susceptible to drug inhibition with EC₅₀ values 4.9 ± 2.7 and 6.2 ± 2.4 μM, respectively. Overall, there is a positive correlation between the K_d values from EP and the in vitro cell culture antiviral EC₅₀ value (Table 2), suggesting that the phenotypical resistance in cell culture is indeed a result of a single mutation at the M2 proton channel.

Insights into the Mechanism of Resistance Elucidated by Molecular Dynamics Simulation

The apo-proteins M2TMC-S31N, M2TMC-S31N/V27A, M2TMC-S31N/V27F, M2TMC-S31N/I32N, M2TMC-S31N/G34E, and M2TMC-S31N/R45H were simulated for 300 ns MD simulation in hydrated POPC bilayers using OPLS2005,³⁰⁻³² CHARMM36,³³ and amber99sb^34,35 force fields which have been reported³⁶ that perform well in positioning transmembrane peptides inside bilayers. CHARMM36,³³ and amber99sb are widely applied when an accurate description of the secondary and tertiary protein structure is needed.³⁷ It was found that amber99sb and CHARMM36 were superior than OPLS2005 in describing the α-helix conformation and performed similarly with an RMSD_protein value for the protein Ca carbons smaller than 2.5 Å. OPLS2005 was selected as more suitable for the description of drug–protein interactions.³⁸ The resulted conformations of the apo-proteins were used to simulate the complexes of mutant M2TMC-S31N with 3 for 100 ns.

Compound 3 in complex with each M2TMC-S31N mutant was simulated to determine the drug orientation within the binding site as well as its atomic interactions with the pore-lining residues. For the M2TMC-S31N in complex with 3, the aryl head moiety of the drug has an outward orientation (toward N-end) in agreement with the experimental findings.¹⁵ In the binding site, the adamantane ring of the ligand is inward to N31, close to A30 and G34, and the ligand forms hydrogen-bond interactions between its NH₂⁺ group and isoxazole nitrogen with N31 and between the hydroxyl group of compound 3 with backbone A30 carbonyls (Figure 3A).

Figure 3.

Overlay of the average structure of 3–M2TMC-S31N complex (ligand carbons in green and protein in pink ribbon) with average structures of 3–M2TMC-S31 N/mutant complexes (ligand carbons are shown in yellow and protein in gray ribbon; carbons of a few side chains are shown in gray sticks) after 100 ns MD simulation. (A) M2TMC-S31N alone in complex with 3 (ligand carbons in green and protein in pink ribbon) after 100 ns MD simulation. (B) Overlay structures for the complexes of 3–M2TMC-S31N with 3–M2TMC-S31N/V27A. (C) Overlay structures for the complexes of 3–M2TMC-S31N with 3–M2TMC-S31N/V27F. (D) Overlay structures for the complexes of 3–M2TMC-S31N with 3–M2TMC-S31N/I32N. (E) Overlay structures for the complexes of 3–M2TMC-S31N with 3–M2TMC-S31N/G34E. (F) Overlay structures for the complexes of 3–M2TMC-S31N with 3–M2TMC-S31N/R45H, after 100 ns MD simulation. For the mutant proteins ligand carbons are shown in yellow and protein in gray ribbon; carbons of a few side chains are shown in gray sticks.

For the M2TMC-S31N/V27A mutant channel, the compound is able to bind, and retains hydrogen bonding interactions with N31 side chains but waters were able to pass between the ligand and the residue 27 gate, probably due to the weakened van der Waals interactions between 3-isopropyl-isoxazolyl of 3 and residue 27 due to the V27A mutation (Figure 3B). Overall, the root-mean-square-deviation (RMSD) of the protein and ligand for the M2TMC-S31N/V27A complex is similar to that of the M2TMC-S31N–3 complex (Table 2).

Next, in the M2TMC-S31N/V27F–3 complex (Figure 3C) the ligand still binds through hydrogen bond interactions with N31 and occasionally with waters, because the van der Waals interactions between 3-isopropyl-isoxazolyl of 3 and residue 27, being phenylalanine instead of valine, become repulsive due to the steric crowding and waters. The values for RMSD_lig and RMSD_protein do not significantly deviate during the simulation relative to M2-S31N (Table 2).

The I32N mutation also does not produce any significant conformational change compared to the M2TMC-S31N–3 complex (Table 2) during the 100 ns MD simulation (Figure 3D). Overall, this is consistent with the EP electrophysiological and antiviral findings for this mutation.

The substitution of the G34 position with glutamic acid (G34E) causes repulsion of adamantane of 3, losing the critical hydrogen-bonding interactions with N31 residues (Figure 3E); N31 side chains turn and form hydrogen bonds with the G34 side chain of the same TM helix chain. The ligand translocates significantly according to the RMSD_lig of ca. 4.2 Å (Table 2) also perturbing the M2AH conformation resulting in RMSD_protein value of ca. 4.3 Å (Table 2).

In the case of R45H, the 300 ns MD simulation of the apo protein shows that the substitution of R45 with histidine leads to a more open structure of the M2 pore. The MD simulation trajectory shows that H45 also has a hydrogen-bonding interaction with D44 of an adjacent helix, but this is weaker compared to the stronger ionic interaction between R45 and D44. The MD simulations suggest that interaction D44–R45 keeps the whole M2 pore more tightened. According to the 100 ns MD simulations of the M2TMC-S31N R45H complex with 3, the ligand and the protein did not deviate from the structure of the M2TMC-S31N complex (Table 2). The ligand inside this mutant channel retains the same interactions with M2TMC-S31N but since the V27 gate is less tight, waters were able to pass between the ligand and V27 isopropyl side chains.

MM-GBSA binding free energy calculations, which sample MD simulations trajectories, resulted in a more negative binding free energies (ΔG_eff) by more than 4 kcal mol^–1 for the complexes between 3 and M2TMC-S31N or M2TMC-S31N/I32N (Table 2) compared with the mutant complexes M2TMC-S31N/V27A, M2TMC-S31N/V27F, M2TMC-S31N/G34E, and M2TMC-S31N/R45H. This result suggests less stable binding of the ligand in the latter complexes which is consistent with a weaker blocking effect. M2TMC-S31N/R45H was between the less stable complexes as revealed by the most positive ΔG_eff of −43.51 ± 3.39 kcal mol^–1. This is probably due to the weaker hydrogen-bonding interaction between H45 and D44 from an adjacent helix, compared to the stronger ionic interaction between R45 and D44 in the case of M2TMC-S31N complex.

Next, we investigated whether drug resistance is a consequence of pore expansion caused by M2 mutations. X-ray crystal structures and MD simulations reveal a wider N-terminal V27A (S31) M2 compared with WT M2.³⁹ Widening of the pore weakens the hydrophobic interactions that are observed between amantadine and V27. In complexes between 3 and M2TMC-S31N the substitution of V27 with alanine in the M2TMC-S31N channel also increases the opening of the gate of the pore as suggested by the distance between the center of masses (COM) of Val27 from ca. 9.6 Å in the M2TMC-S31N to 11.1 Å in M2TMC-S31N/V27A complex (Figure 4A). In the M2TMC-S31N/V27F channel, a slightly more pronounced pore expansion is observed compared to the M2TMC-S31N/V27A variant, possibly caused by repulsive forces of the bulky phenylalanine side chain with the ligand’s 3-cyclopropyl-isoxazole (Figure 4B). The G34E mutation caused the largest increase in the overall pore diameter (Figure 4B). Conversely, the interhelical mutant I32N did not cause expansion of the pore relative to S31N complex alone (Figure 4A). For the R45H mutant, the distance between V27 Ca carbons increases from ca. 11.6 Å in M2TMC-S31N to ca. 13.8 Å in M2TMC-S31N/R45H complex (Figure 4B). This opening in the N-end causes a less tight binding of 3 leading to the moderate blocking of M2-S31N. Overall, the channel pore expansion at V27 ranks in the following order: G34E > V27F > V27A > R45H > I32N ∼ S31N (res27–res27 distance in Table 2), which is consistent with the degree of drug resistance as shown by the K_d and EC₅₀ values (Table 2). In other words, the wider the channel pore is, the more pronounced the drug resistance is for a given M2 mutant. A direct result of the opening of the N-terminal channel in these mutants is the abolishment or weakening of the critical hydrophobic interactions between the cyclopropyl group from compound 3 and the hydrophobic side chains from residue 27, leading to drug resistance.⁶

Figure 4.

Effect of mutations on the M2 pore size in complex with compound 3. (A) Overlay of TM pores (residues 25–46) of M2TMC-S31N (in gray sticks), S31N/I32N (in red sticks), and S31N/V27A (in green sticks), which shows from a top view the increase of pore diameter according to the sequence V27A > I32N ∼ S31N. (B) Overlay of TM pores (residues 25–46) of M2TMC-S31N/V27A (in green sticks), S31N/G34H (in yellow sticks), S31N/V27F (in light blue sticks), and S31N/G34E (in magenta sticks), which shows from a top view the increase of pore diameter according to the sequence G34E > V27F > R45H ∼ V27A. For the quantitative measurement of the distance between the center of masses (COM) of V27 of each mutant please refer to Table 2.

Next, we analyzed the pore hydration associated with drug-resistant mutant channels as resulted from the MD simulations. In the apoprotein M2TMC-S31N, a network of water molecules hydrate the N-terminal pore spanning the N-end to the His37 tetrad. M2TMC-S31N blockers disrupt this water network by dehydrating the pore at the drug-binding site (see the yellow density in Figure 5A). In the M2TMC-S31N/V27A, the increased distance between the 3-cyclopropyl isoxazole of compound 3 and residue 27 allow for the passage of waters, resulting in the overall loss of blockage (Figure 5B). Indeed, water densities are observed between the isoxazole ring and Ala30 and Asn31 residues. Water densities are also found occupying the pore vestibule between the adamantane cage and His37 tetrad. The M2TMC-S31N/V27F channel shows similar water distribution as the M2TMC-S31N/V27A channel, but is overall more hydrated (Figure 5C), consistent with the increased expansion of the pore illustrated in Figure 3. Additionally, the M2TMC-S31N/V27F mutation produces an influx of waters at the C-terminal end that extends through the Trp41 gate below the His37 tetrad. The M2TMC-S31N/G34E channel also has water densities throughout the pore, but the helical secondary structure is significantly unfolded as a result of the mutation (Figure 5D). The M2TMC-S31N/I32N in complex with 2 do not increase the hydration of the pore relative to M2TMC-S31N in complex with 3 (Figure 5E), indicating that this mutant might remain drug sensitive. Collectively, the M2TMC-S31N/V27A, M2TMC-S31N/V27F, M2TMC-S31N/G34E, and M2TMC-S31N/R45H produce a more hydrated channel compared to M2TMC-S31N and M2TMC-S31N/I32N. Although compound 3 can still bind to the mutated channels, drug binding does not completely stop proton conductance through the water wires, which may explain the drug-resistant phenotype.

Figure 5.

Water densities inside the pore shown representatively for the complex between compound 3 and (A) M2TMC-S31N; (B) M2TMC-S31N/V27A; (C) M2TMC-S31N/V27F; (D) M2TMC-S31N/I32N; (E) M2TMC-S31N/G34E; (F) M2TMC-S31N/R45H. Panels A and D show that in M2TMC-S31N and M2TMC-S31N/I32N no waters are passed, which is consistent with blockage from 3. Panels B, C, E, and F show that mutations V27A, V27F, G34E, and R45H allow waters passage, which is consistent with lack of blockage from 3. In G34E a significant unfolding of the α-helices is observed (ligand atoms are shown as van der Waals spheres, water as yellow density, protein in ribbon and side chains from residues 27, 31, 37, and mutated amino acid in sticks).

Conclusions

The conjugates of amantadine with heterocyclic aryl heads may represent useful leads for a new generation of anti-influenza A drugs. We showed previously that L46P mutation located outside the M2 pore causes a conformational change of M2-S31N which resulted in a broadening of the pore,²⁷ mainly at its N-end where the conjugate ligand binds, resulting in the loss of M2-S31N channel blockage. Additional mutations discovered herein include residues oriented either in the interior of the pore, that is, V27A, V27F, and G34E, or at the interhelical space I32N, or at the C-terminal allosteric site R45H. All five mutations were generated in recombinant viruses as well as expressed in oocytes for functional measurements. The efficacy of channel blockage and antiviral activity is reduced significantly in pore-lining mutations (V27A, V27F, and G34E), but partially reduced for interhelical I32N mutant and C-terminal allosteric mutant R45H. Our simulations show that 3 can still bind in V27A, V27F, or R45H but did not block proton conductance due to opening in residue-27 gate caused by conformational changes at the N-terminus or C-terminus of the pore, respectively. In the case of M2TMC-S31N/G34E the ligand loses binding due to strong repulsive forces. This is apparent when observing the hydration state of the mutant channel when bound to compound 3.

Since different influenza strains cocirculate in each influenza season, it is critical to investigate how different strains will mutate under the same drug selection pressure. This will help researchers prioritize lead compounds with a high genetic barrier to drug resistance for further translational development. With the aim of developing the next generation of influenza antivirals to combat drug-resistant viruses, we chose M2-S31N as a drug target as it is conserved in more than 95% of current circulating influenza A viruses. We have shown previously²⁰ that our rationally designed M2-S31N inhibitors such as compound 3 inhibit multiple contemporary circulating influenza A viruses (H1N1 and H3N2) in multiple cell lines. By performing resistance selection using multiple strains of influenza viruses in multiple cell lines, we have found that (1) different influenza strains can evolve the same or different drug-resistant mutants with variable degree of resistance in the same cell line, (2) the chances of identifying novel drug-resistant mutants is higher when resistance selection experiments are repeated with multiple influenza strains in multiple cell lines, and (3) one influenza strain can evolve different drug-resistant mutants in different cell lines under the drug selection pressure of the same drug. In light of this finding, the genetic barrier to drug resistance of other influenza drugs that are currently in development should be similarly evaluated using multiple viruses in multiple cell lines. It is also interesting to follow with in vivo mouse model studies to see the correlation between in vitro drug resistance and in vivo drug resistance. Collectively, this study provides a biochemical basis for the mechanisms of drug resistance to M2-S31N inhibitors and will contribute to the chemical refinement for these antiviral molecules.

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsptsci.0c00018.

• Protein preparation–docking calculations, MD simulations, MD simulations protocol, analysis of MD simulations trajectories, MM-GBSA calculations (PDF)

Author Contributions

^# Rami Musharrafieh and Panagiotis Lagarias contributed equally to this work. R.M. performed the drug resistance selection and reverse genetic studies with the help from R.H., and A.R. C.L.M. performed the electrophysiological assays. P.L. performed the molecular dynamics simulations with the help from G.L. J.W. and A.K. designed and supervised this study. R.M., J.W., and A.K. wrote the manuscript with contributions from other authors.

The authors declare no competing financial interest.