|Home | About | Journals | Submit | Contact Us | Français|
The integral membrane protein, M2, of influenza virus forms pH-gated proton channels in the viral lipid envelope1. The low pH of an endosome activates the M2 channel prior to hemagglutinin-mediated fusion. Conductance of protons acidifies the viral interior and thereby facilitates dissociation of the matrix protein from the viral nucleoproteins – a required process for unpacking of the viral genome2. In addition to its role in release of viral nucleoproteins, M2 in the trans-Golgi network (TGN) membrane prevents premature conformational rearrangement of newly synthesized hemagglutinin during transport to the cell surface by equilibrating the pH of the TGN with that of the host cell cytoplasm3. Inhibitng the proton conductance of M2 with the anti-viral drug amantadine or rimantadine inhibits viral replication4–7. We have determined by NMR the structure of the tetrameric M2 channel in complex with rimantadine. In the closed state, four tightly packed transmembrane (TM) helices define a narrow channel, in which a “tryptophan gate” is locked by inter-molecular interactions with aspartic acid. A C-terminal, amphipathic (AP) helix oriented nearly perpendicular to the TM helix, forms an inward facing base. Lowering the pH destabilizes the TM helical packing and unlocks the gate, admitting water to conduct protons, while the C-terminal base remains intact, preventing dissociation of the tetramer. Rimantadine binds at four equivalent sites near the gate on the lipid facing side of the channel and stabilizes the closed conformation of the pore. Drug-resistance mutations are predicted to counter the effect of drug binding by either increasing the hydrophilicity of the pore or weakening helix-helix packing, thus facilitating channel opening.
M2 is a 97-residue single-pass membrane protein with its N- and C-termini directed toward the outside and inside of the virion, respectively; it is a homotetramer in its native state8, 9. The four TM helices form a channel in which His37 is the pH sensor and Trp41, the gate6, 10, 11. The adamantane-based drugs, amantadine and rimantadine, which target the M2 channel, have been used as first-choice antiviral drugs against community outbreaks of influenza A viruses for many years, but resistance to the adamantanes has recently become widespread. Many structural models of this channel have been built, based on sequence analysis, mutagenesis, and solid-state NMR8, 11, 12. Many of these studies have been done on inherently unstable TM-only constructs, however, leading to conflicting structural conclusions.
While the TM-only peptide fails to form a stable tetramer, a construct of residues 18–60 [M2(18–60)], that includes 15 residues of the C-terminus in addition to the TM region, forms a stable tetramer in dihexanoyl-phosphatidyl-choline (DHPC) detergent micelles and yields high-resolution NMR spectra (Supplementary Fig. S1). In the closed conformation at pH 7.5, M2(18–60) is a homotetramer, in which each subunit has an unstructured N-terminus (residues 18–23), a channel-forming TM helix (residues 25–46), a short flexible loop (residues 47–50), and a C-terminal AP helix (residues 51–59). The TM helices assemble into a four-helix bundle with a left-handed twist angle of ~23° and a well-defined pore (Fig. 1). A ring of methyl groups from Val27 constricts the N-terminal end of the pore to ~3.1 Å (inner diameter). In agreement with proposed models11, 13, His37 and Trp41 are inside the pore. A 3-bond, 15N-13Cγ scalar coupling (3JNCγ) value of 1.5 Hz (Supplementary Table 1), shows the His37 χ1 rotamer to be predominantly trans, but with significant rotameric averaging. The χ1 of Trp41 is essentially locked in the trans position, as determined by 3JNCγ of 2.6 Hz, while the χ2 is also fixed at around −120° by the sidechain Hε1 – Nε1 dipolar coupling and NOEs. The Trp41 indole rings are at van der Waals (VDW) distance from each other, prohibiting passage of water or ions (Fig. 1c). The indole Hε1 of one subunit is on average 3.5 Å from the Asp44 carboxyl carbon of the adjacent subunit. The two residues can form an intermolecular hydrogen bond that stabilizes the closed Trp41 gate. The sidechain of Arg45 likely participates in an intermolecular interaction with Asp44. These findings are consistent with the increased pH-modulated activity of channels in which asparagine has replaced Asp4414. The C-terminal end of the channel extends into a loop (residues 47–50) that connects the TM domain to the C-terminal AP helix. RDCs and intra- and inter-monomer NOEs show that the AP helices lie roughly perpendicular (~82°) to the TM helices and assemble head-to-tail with a right-handed packing mode to form the base of the channel. The orientation and amphipathic character of the AP helices suggest that the C-terminal base lies on the surface of the membrane.
Residues 47–50 give no NOE peaks and have no stable, hydrogen-bonded structure in the detergent micelles used in our work. We believe that this segment adopts a more stable conformation in the viral membrane because Cys50, which we mutated to serine to avoid disulfide formation, is normally palmitoylated15. Modeling shows that extending the TM helix to Phe48 would place residue 50 facing the membrane, allowing for insertion of the palmitoyl acyl chain into the lipid bilayer. This minor rearrangement would also move the AP helices closer to the TM domain.
Drug binding has the effect of stabilizing the closed conformation. Upon addition of drug, the resonances of residues 43–46 at the C-terminus of the channel, which are severely exchange-broadened in the drug-free sample, became significantly sharper and more homogeneous (Supplementary Fig. S2). The protein-drug NOEs collected from four different NOESY spectra (Supplementary Fig. S3) place the binding site between adjacent helices at the C-terminal end of the TM domain near the Trp41 gate, on the membrane side of the channel (Fig. 1d). We could not detect drug NOEs in other parts of the protein, including the widely proposed drug-binding site in the pore of the channel. The amine head-group of rimantadine is in contact with the polar sidechains of Asp44, Arg45, and the indole amine of Trp41. The sidechains of Ile42 from one helix, and Leu40 and Leu43 from another helix form the hydrophobic walls of the binding pocket that interact with the adamantane group of rimantadine. Thus, rimantadine covers a unique polar patch in the otherwise hydrophobic environment of the TM domain. Interactions between rimantadine and the channel are consistent with structure-activity relationships of the adamantane group16. In particular, the basic nitrogen group and size limits at the methyl site are critical. These requirements are the result of the interactions with Asp44 and the small hydrophobic pocket around Ile42, respectively.
Water NOEs measured in the 110 ms 15N-separated NOESY experiments give a clear picture of water distribution relative to the channel (Fig. 2). The lipid-facing surface of the TM region is largely protected from water by the DHPC micelle. In the closed channel pore, the Val27 ring at the N-terminus and the Trp41 gate at the C-terminus essentially block water from freely diffusing into the pore from either side of the membrane. Within the TM region, only the amides of Ser31 and Ile32 have NOE crosspeaks at the chemical shift of water, probably corresponding to the hydroxyl proton of Ser31 in exchange with water. A polar residue is present at position 31 in all sequenced variants of M2, suggesting that proton conduction requires water bound to this site. This water may serve to bridge the proton relay from the N-terminal end of the pore to the His37 pH sensor. Water was detected at the C-terminus of the TM region, beginning at Arg45. The Hε1 of the Trp41 indole ring, which points toward the C-terminal side of the pore, also has a strong NOE to water, indicating that the base of the channel is accessible to bulk water.
Lowering the pH from 7.5 to 6.5 broadens most of the NMR resonances corresponding to the TM helix (Fig. 3a). The resonance broadening could not be attributed to protein aggregation, as the self-diffusion coefficients were essentially unchanged between pH 7.5 and 6.5. Thus, activation of the channel is coupled to increased conformational exchange in the TM domain. In contrast, the resonances of the AP helices are essentially unaffected by lowering the pH, indicating that the C-terminal base of the tetramer remains intact as the channel opens.
In addition to destabilizing helix-helix packing in the TM domain, channel activation must also correlate with increased dynamics of the Trp41 gate. Since the indole amide resonance of Trp41 remained intense as the pH was lowered from 7.5 to 6.0, it serves as a useful NMR probe for monitoring channel opening. We compared the ms timescale dynamics of the Trp41 indole ring between the closed and open states by carrying out relaxation-compensated Carr-Purcell-Meiboom-Gill (CPMG) experiments17 at pH 7.5, 7.0, and 6.0. A two-site exchange model fits the dependence of 15N relaxation due to chemical shift exchange on the frequency of refocusing (1/τcp) of chemical shift evolution (Fig. 3b), implying that the gate switches between two configurations at any given pH. As the pH was lowered from 7.5 to 6.0, the rate of fluctuation increased by more than four-fold (Fig. 3b), indicating that channel activation “unlocks” the gate. Adding rimantadine to the channel at an intermediate pH of 7.0 slowed the timescale of the gate motion to nearly that of a drug-free gate at pH 7.5 (Fig. 3c). These results confirm that the reconstituted channels in the NMR sample are pH-gated, and are consistent with the location of the rimantadine site proximal to the gate.
The structure of the M2 proton channel thus reveals a simple yet effective gating. The tight packing of the four TM helices brings the bulky indole rings of Trp41 into VDW contact to form the channel gate. The gate is further stabilized by inter-subunit hydrogen bonds with Asp44. Lowering the pH protonates the imidazole rings of His37, destabilizing helix-helix packing by electrostatic repulsion. This conformational rearrangement breaks interactions between Trp41 and Asp44 and allows the gate to flip open. A pair of conserved N-terminal cysteines have been shown to form intermolecular disulfides in vivo9. Thus, the TM helices are tethered at one end by N-terminal disulfides and at the other end by the C-terminal base, ensuring that destabilization of the four-helix bundle during channel activation does not cause dissociation of the tetramer (Fig. 4). Indeed, truncation of the AP helix results in channels that rapidly lose channel activity18.
The external drug-binding site was unexpected. Drug-resistance mutations seemed to suggest a site inside the pore, because in early models of the channel, residues that lead to drug resistance were predicted to be pore-lining. The known mutations that confer drug resistance are L26F, V27A, A30T, S31N, G34E, and L38F. Mapping these residues onto the structure (Supplementary Fig. S7) reveals that Val27, Ala30, and Gly34 are pore-lining, but Leu26, Ser31 and Leu38 are in the helix-helix packing interface. Moreover, these mutations are spread out over more than three turns of the TM helix, covering a distance much larger than the dimensions of amantadine or rimantadine. Pinto and Lamb have pointed out that having a cork-plugging-the-bottle model is insufficient to explain all the results of electrophysiology studies19. For example, drug inhibition is more effective when applied to the closed channel than the open channel, which is not expected of a pore-blocking mechanism5. Several of the drug-resistance mutations in pore-lining residues have been shown to retain drug binding20. Although a pore blocker is expected to fit tightly in the pore, channel inhibition is unusually tolerant of modifications to the adamantane scaffold16. Together, the above observations suggest an allosteric inhibition mechanism.
Is the external binding site consistent with all drug-resistance mutations? Although the exact structural effects of resistance mutations are difficult to predict, what they do have in common is that they either perturb the helix-helix interface (L26F, V27A, S31N, L38F), or increase the hydrophilicity of the pore (A30T, G34E). From this observation, and from detection of a conformational exchange among multiple states at lowered pH, we propose an allosteric inhibition mechanism that can account for all of the mutations. In our model, drug binding makes the closed channel harder to open, while drug-resistance mutations destabilize the closed channel, making it easier to open. Replacing Val27 with alanine enlarges the N-terminal opening and weakens helix-helix packing, and therefore may facilitate channel opening. Ala30 and Gly34 are inside the pore, and replacing them with threonine and glutamate, respectively, may facilitate pore hydration, and in turn, channel opening. Leu26, Ser31 and Leu38 are helix-helix interface residues; their mutations likely perturb helix-helix packing and lower the energetic cost of channel opening.
Why then, are no drug-resistant mutations observed near the drug-binding site? In fact, few mutations are ever observed in this region of the channel (drug-resistant or otherwise), which is not surprising due to the functional constraints placed on these residues in proximity to the channel pH sensor (His37) and the channel gate (Trp41). As the structure illustrates, intermolecular contacts between Asp44 and Arg45 form an integral part of the channel gate, along with Trp41. The residues that form the hydrophobic walls of the binding pocket – Leu40, Ile42, and Leu43 – are on the lipid face of the channel and must retain hydrophobicity for membrane partitioning. In order to accommodate the Trp41 indole rings within the channel, the helices splay slightly at the C-terminus of the TM domain, and interhelical contacts below Leu38 – with the exception of the Trp41 sidechains – are no longer important for channel assembly. Thus, residues essential to channel assembly are in the N-terminal half of the TM helix, exactly where drug-resistance mutations occur.
Binding from the membrane side is consistent with the high membrane partition coefficient of adamantane drugs, which effectively concentrates them in the membrane and lowers their level in the aqueous phase21, 22. Adamantanes interact with a number of other ion channels, including viroporins from hepatitis C23, the potassium channel Kcv of the chlorella virus PBCV-124, and the human NMDA receptors25. Hanatoxin, an allosteric inhibitor of voltage gated K+ channels with a high membrane partition coefficient, also has an external binding site26, 27. Membrane-side binding may thus be a feature of many channel inhibitors. This mode of inhibition could be advantageous for drug design because drug molecules are typically much larger than hydrated ions selected by ion channels, and therefore the energy barrier for drug to find a blocking site inside the channel pores would be much higher than targeting a functional site from the membrane side of the channel.
The M2(18–60) polypeptide construct was expressed as a C-terminal fusion to trpLE with an N-terminal 9-His tag in the pMM-LR6 vector28. The M2(18–60)tetramer was reconstituted by dissolving peptide in solution containing 50 mM sodium phosphate, 6 M guanidine HCl, 150 mM DHPC, dialyzing against a solution containing 40 mM sodium phosphate (pH 7.5) and 30 mM glutamate, and concentrating. Rimantadine was added to the reconstituted protein. The final NMR sample used for structure determination contained 0.75 mM M2(18–60) (monomer), ~300 mM DHPC, and 40 mM rimantadine. Given the DHPC has an aggregation number of 2729 and the strong partition coefficient of rimantadine in phospholipids (rimantadine aqueous solubility is very low, ~ 50 μM), locally, there are about 4 rimantadine molecules per micelle compartment in which the channel resides.
The NMR protocol employed was similar to that previously described by Oxenoid & Chou30. An extensive set of structural restraints including 230x4 intra- and 27x4 inter-molecular distance restraints derived from nuclear Overhauser enhancements (NOEs), 27x4 orientation restraints from residual dipolar couplings (RDCs), and 23x4 sidechain rotamers from 3-bond scalar couplings were used to generate an ensemble of 15 low energy structures with a backbone rmsd of 0.30 Å for the channel region, and 0.89 Å for all structured regions (Fig. 1a). The refinement statistics and NMR-derived restraints are summarized in Supplementary Table S1. Structure calculation was accomplished in two steps, in which the overall tetramer conformation was first defined by NOE-derived distance restraints and J coupling-derived dihedral restraints using a high-temperature simulated annealing protocol, and subsequently refined against RDCs at low temperature. More experimental details are given in the METHODS section below.
M2(18–60) was expressed into inclusion bodies as a fusion to (His)9-trpLE31. The M2(18–60) peptide was released from the fusion protein by CNBr digestion in 70% formic acid (2 hr, 0.2 g/ml). The digest was dialyzed to water, lyophilized, and loaded onto a C4 column (Grace-Vydac) in 2:1:2 hexafluoroisopropanol:formic acid:water and separated on a gradient of 3:2 isopropanol:acetonitrile. The lyophilized peptide was refolded at 250 μM by dissolving in 6 M guanidine and 150 mM DHPC and dialyzing against the final NMR buffer containing 40 mM sodium phosphate and 30 mM glutamate. The sample was concentrated to a final M2(18–60) concentration of 0.75 mM (monomer). Rimantadine was added after concentrating. The concentration of DHPC was determined from 1H NMR spectroscopy to be around 300 mM. Given the DHPC has an aggregation number of 2729 and the strong partition coefficient of rimantadine in phospholipids (rimantadine aqueous solubility is very low, ~ 50 μM), locally, there are about 4 rimantadine molecules per micelle compartment in which the channel resides.
NMR experiments were conducted at 30°C on spectrometers equipped with cryogenic probes (Bruker, Billerica, MA). Sequence specific assignment of backbone 1HN, 15N, 13Cα, and 13Cβ chemical shifts were accomplished using the TROSY versions of the HNCA and HNCACB experiments on a 15N-, 13C, and 85% 2H-labeled protein. Sidechain χ1 and χ2 rotamers were obtained from measurements of the 3-bond scalar couplings including 3JNCγ, 3JC’Cγ and 3JCαCδ32, 33. 1H-15N RDCs were measured for the protein-detergent complex weakly aligned in radially-compressed polyacrylamide gel34. Intramonomer NOEs involving both backbone and sidechain protons were assigned using the 3D 15N-edited and 13C-edited NOESYs recorded with NOE mixing times of 110 and 150 ms, respectively, on a sample containing 15N-, 13C-labeled protein, rimantadine, and deuterated DHPC (D35-DHPC) (Avanti Polar Lipids, Inc.). For identifying contacts between adjacent monomers, intramonomer NOEs were first assigned to completion. This is possible owing to the low complexity of the NOESY spectra of M2(18–60) (see Supplementary Fig. S4&5). The remaining NOEs involving residues of the TM and AP helices, which could not be explained by intramonomer distances based on the known secondary structures, were identified to be intermonomer NOEs. The assignment of intermonomer distance restraints and structure calculation were carried out iteratively until all NOE crosspeaks in the NOESY spectra were self-consistent in the tetrameric structure. Protein-drug NOEs were first identified using the 15N-edited and 13C-edited NOESYs described above, and subsequently confirmed by a 15N-edited NOESY recorded with 500 ms mixing time on a sample containing uniform 15N- and 2H-labeled protein, rimantadine, and D35-DHPC.
Structures were calculated using the program XPLOR-NIH35. The secondary structure of the monomer was first calculated from random coil using intramonomer NOEs, backbone dihedral restraints derived from chemical shifts (TALOS)36, and side chain χ1 and χ2 restraints shown in Supplementary Table 2. A total of 20 monomer structures were calculated using a standard high-temperature simulated annealing (SA) protocol in which the bath temperature was cooled from 1000 to 200 K. To obtain an initial set of tetramer structures, four copies of the lowest-energy monomer structure calculated above were used. The same high-temperature SA run was performed in the presence of intermonomer NOEs and all other intramonomer restraints except RDCs. For each experimental intermonomer NOE between two adjacent subunits, four identical distance restraints were assigned respectively to all pairs of neighboring subunits to satisfy the condition of C4 rotational symmetry. During the annealing run, the bath is cooled from 1000 to 200 K with a temperature step of 20 K, and 6.7 ps of Verlet dynamics at each temperature step, using a time step of 3 fs. A total of 100 tetramer structures were calculated, and independently cross validated by 1H-15N RDCs. Fitting of RDCs to structures was done by singular value decomposition (SVD), using the program PALES37. The goodness of fit was assessed by Pearson correlation coefficient (r) and quality factor (Q). Among the 100 structural models, 15 structures of which the individual subunits have on average the best agreement with RDCs (r ~ 0.91 and Q ~ 0.25) were selected for a final, low-temperature refinement against RDCs in the presence of all NOE and dihedral restraints. During the refinement, the bath was cooled from 200 to 20 K with a temperature step of 10 K, and 6.7 ps of Verlet dynamics at each temperature step, using a time step of 3 fs. The force constants for NOE and experimental dihedral restraints were fixed at 100 kcal mol−1 Å−2 and 40 kcal mol−1 rad−2, respectively. RDC restraint force constant was ramped from 0.01 to 0.125 kcal mol−1 Hz−2. The force constant for RDCs was set to a small value to prevent violation of NOE restraints. For each of the 15 structures validated by RDCs, 10 RDC-refined structures were generated. From that set, the structure with the lowest total energy was added to the final ensemble to describe the structural diversity of the solution structure. The structure with heavy atom conformation closest to the mean was chosen to represent the final ensemble. Without violating any NOE restraints, the final subunit structures fit RDCs to r of 0.98 and Q of 0.15, with Da = 14.2 Hz and Rh = 0.24. Final backbone RMSD from the mean in the TM and AP domains were 0.30 Å and 0.56 Å, respectively. The structures have been deposited in the Protein Data Bank with accession code 2RLF.
The timescale of chemical shift exchange of the Trp41 sidechain was measured using a relaxation-compensated CPMG experiment17 in 1D mode at 1H frequency of 600 MHz. The dependence of 15N relaxation due to chemical exchange on the frequency of refocusing (1/τcp) of chemical shift evolution was fitted to a two-site exchange model given by Rex 1− (2τex/τcp)tanh(τcp/2τex), where Rex is the contribution to transverse relaxation due to chemical shift exchange, and τex is the correlation time of the process that is generating the chemical shift exchange38.
We thank Marcelo Berardi for many discussions and Stephen Harrison for assisting with the manuscript. This work was supported by NIH grant AI067438 and the Pew Scholars Program in the Biomedical Sciences awarded to J.J.C. J.R.S is supported by an NIH F32 postdoctoral fellowship.
Author contributionsJ.R.S. and J.J.C. designed research, performed research, analyzed data, and wrote the paper.
Conflict of interest statement
J.J.C. and J.R.S. declare competing financial interests. A provisional patent entitled “Systems and Methods for Studying Influenza” was filed on October 25, 2007 on behalf of Harvard Medical School by Wolf, Greenfield & Sacks, P.C.
The structures have been deposited in the Protein Data Bank with accession code 2RLF.