INTRODUCTION

Top Abstract Introduction Results & Discussion Concluding Remarks References

Nicotinic acetylcholine receptors (nAChRs) are ligand-gated ion channels involved in signal transduction in interneuron and neuromuscular junctions (Devillers-Thiery et al., 1993; Galzi and Changeux, 1994). They incorporate five subunits, each subunit having a large extracellular domain at the N end and four transmembrane segments, M1-M4. Mutagenesis and physiological studies indicate that the pore is lined by M2 segments (M2s) from each of the five subunits (Changeux et al., 1992; Galzi and Changeux, 1995). Solid-state NMR spectroscopy demonstrated -helical conformation and transmembrane orientation of M2 peptide (Bechinger et al., 1991). Fourier transform infrared spectroscopy of the purified nAChR reconstructed into lipid vesicles indicated predominantly -helical structure of the receptor transmembrane domains (Baenziger and Methot, 1995). Synthetic channels composed of four or five M2s exhibit some electrophysiological and pharmacological properties of nAChRs (Oblatt-Montal et al., 1993a,b). Periodicity of the residues responsible for the channel functions (Table 1) suggests an -helical structure for M2s that form a five-helix bundle (see Hille, 1992, for a review). The bundle is believed to have a funnel-like shape with a narrow cytoplasmic end, and six rings of homologous residues facing the pore at the levels of Glu¹ (intermediate ring), Thr⁴ (threonine ring), Ser⁸ (serine ring), Leu¹¹ (equatorial ring), Val¹⁵ (valine ring), and Leu¹⁸ (outer leucine ring). The five-helix bundle was visualized in structural models of nAChR pore (Oiki et al., 1988, 1990; Furois-Corbin and Pullman, 1989; Eisenman et al., 1990; von Kitzing, 1995). The structure of other transmembrane segments remains unknown; -sheets may be present in some of them (Hucho et al., 1994). N-termini of M1s may contribute to the synaptic end of the pore (Akabas and Karlin, 1995; DiPaola et al., 1990); the data support the funnel-like model of the channel. Assuming a 3-D homology of nAChR with the heat-labile enterotoxin, Ortels and Lunt (1996) proposed a model of nAChR with a central pore composed of slightly bent -helical M2s and -strands dominating in the remainder of the structure.

Electron microscopy studies suggest that M2s are kinked approximately in their midpoints (Unwin, 1993, 1995, 1996). The rotation of the kinked helices was suggested to cause a transition between the open and the closed states of the channel (Unwin, 1995). Using the electron microscopy data, Sankararamakrishnan et al. (1996) created models of the open and the closed channel with the kinked helices. The open-channel model predicts the minimum diameter of the pore to be 12 Å, the value conflicting with the data on permeability of the channel for organic cations (Dwyer et al., 1980; Nutter and Adams, 1995). To explain this disagreement, the authors of the model suggested that water molecules bound to the channel walls would narrow the pore. A network of H-bonds between the hydrophilic residues of the channel and the intrapore waters could be stable enough to diminish the effective diameter of the pore for transient penetrating cations. However, the water molecules may be displaced by amphiphilic ligands whose dwelling time inside the pore reaches tens of milliseconds. The well-known noncompetitive blockers of nAChR are rather small to bind tightly in the pore of 12 Å in diameter. In the absence of a high-resolution structure of nAChR, the actual diameter of the pore remains unknown. On the other hand, the recently resolved crystal structure of the five-helix bundle in the cartilage oligomeric matrix protein, a possible prototype ion channel, presents an example of a narrow pore (Malashkevich et al., 1996).

The suggestion on the channel gating via rotations of entire M2s does not agree with the cysteine-scanning data that both open and closed states of nAChR allow sulfhydryl reagents to reach nearly the same residues in M2s (Akabas et al., 1994; Akabas and Karlin, 1995). Dimensions of the pore and the gate should correlate: essential conformational changes (e.g., rotations of entire M2s) are necessary to close a wide pore, whereas even a slight rearrangement of hydrophobic residues at a channel constriction comparable in dimensions with a hydrated cation may block the permeation (Zhorov and Ananthanarayanan, 1996). Further modeling studies are necessary to explain seemingly conflicting data from various approaches.

A sensitive touchstone for models of nAChR pore is their ability to explain structure-activity relationships of noncompetitive blockers. The pore of nAChR is blocked by various organic cations: chlorpromazine (Benoit and Changeux, 1993), local anesthetics QX-222 and QX-314 (Neher and Steinbach, 1978), procaine (Adams, 1977), triphenylmethylphosphonium (Hucho et al., 1986), tubocurarine (Large and Sim, 1986), phencyclidine (Aguayo and Albuquerque, 1986; Papke and Oswald, 1989), bis-ammonium compounds (Zhorov et al., 1991; Antonov et al., 1995), MK-801 (Amador and Dani, 1991), etc. The binding sites for chlorpromazine (Revah et al., 1990) and triphenylmethylphosphonium (Hucho et al., 1986) were determined by the affinity labeling method. The mutation experiments determined residues responsible for the binding of QX-222 (Leonard et al., 1988; Charnet et al., 1990). The data on the channel blockade were used by Furois-Corbin and Pullman (1989) to create a funnel-like model of nAChR that allows chlorpromazine (CPZ) to pass the bulky Leu¹⁸ and Val¹⁵ rings and reach the Ser⁸ ring. An analysis of conformation-activity relationships of bis-ammonium blockers of the neuronal and muscle nAChRs had led us to predict dimensions of the pore at Ser⁸ and Thr⁴ rings (Zhorov et al., 1991; Brovtsyna et al., 1996; Tikhonov et al., 1996). In the present work, we employed the restrained Monte Carlo minimization method to create several five-helix-bundle models of nAChR with various geometrical profiles of the pore. We further simulated docking of structurally different blockers in the bundles to select an optimal model that predicts the binding sites of the blockers in agreement with the experimental data. The optimal model obtained is consistent with the available data on the permeation of organic cations, affinity labeling, cysteine scanning, and mutagenesis experiments. This model also agrees with some (but not all) suggestions of Unwin (1993, 1995) deduced from the electron microscopy images of nAChR.

	METHOD

The protocol of the Monte Carlo minimization method (MCM) (Li and Scheraga, 1988) and an ECEPP/2 force field (Momany et al., 1975; Nemethy et al., 1983) were used to search for optimal conformations. Energy was minimized in the space of generalized coordinates (Zhorov, 1981, 1983): torsions of M2s and blockers, bond angles of blockers, positions of M2s and blockers (Cartesian coordinates of their root atoms), and orientations of M2s and blockers (Euler angles of the local coordinate systems centred at the root atoms). Calculations were carried out with a ZMM program package (Zhorov, 1981, 1993). Geometrical profiles of the pore and positions of the blockers along the pore were restrained with flat-bottom penalty functions (Brooks et al., 1985) with a bottom width of 1 Å. Initial assembly of the models, their visualization, and manipulation were performed with a molecular graphics program elaborated by one of us (Tikhonov, to be published).

Trajectories were calculated at T = 600 K. A subsequent starting point in a trajectory was obtained by changing a randomly selected generalized coordinate (excluding bond angles) of the preceding point by a random increment. From a given starting point, energy was minimized in the space of all generalized coordinates until the norm of energy gradient became less than 1 kcal·mol¹·rad¹. The resulting minimum-energy conformation (MEC) was accepted in the trajectory if its energy (E) was less than that of the preceding point of the trajectory (E_p) or if a random number n  (0, 1) was less than exp((E  E_p)/RT). MCM trajectories were terminated when the last N_u energy minimizations did not lower the energy of the best MEC found. In different experiments, the parameter N_u varied between 100 and 3000 (see results).

Atom-atom interactions were calculated with a cutoff distance of 7.5 Å. The partial charges at the atoms of blocking molecules were calculated by the CNDO/2 method. The lowest-energy MECs of lone blockers were used as starting approximations for the docking of the blockers in the five-helix bundle. During the docking, the main-chain torsions and positions of M2s, as well as bond angles and rings of the blockers, were kept rigid, and the other generalized coordinates were varied. Minimum dimensions of the blockers that may correspond to the cross-ectional dimensions of the pore at their binding sites were calculated, taking into account van der Waals radii of atoms (Zhorov et al., 1991). Other details of the calculations are described elsewhere (Zhorov and Ananthanarayanan, 1996).

	RESULTS AND DISCUSSION

Top Abstract Introduction Results & Discussion Concluding Remarks References

General assumptions

-Helical M2s are believed to be the major pore-lining segments in nAChRs (Unwin, 1995; Galzi and Changeux, 1995; Bertrand et al., 1993), whereas N-end parts of M1s contribute to the pore only partially (DiPaola et al., 1990; Akabas and Karlin, 1995). We composed the pore of M2s only. Because M2s are conserved in various nAChRs (Le Novere and Changeux, 1995), we used a consensus sequence (Table 1) that would corresponds to the Ser¹² to Ala¹² mutant of the M2 segment from the -subunit of Torpedo californica nAChR. Nicotinic acetylcholine receptors from different sources have diverse permeabilities and sensitivities to blockers (Sands and Barish, 1991; Vernino et al., 1992; Zhorov et al., 1991; Cuevas and Adams, 1994). However, many compounds (e.g., bisammonium blockers) demonstrate analogous tendencies of structure-activity relationships in muscular and neuronal nAChRs (Brovtsyna et al., 1996; Tikhonov et al., 1996). This justifies our use of the consensus M2 for a general model of nAChR pore.

Helices between Thr⁴ and Ser⁸ are not tilted radially to the pore axis

First we studied conformational properties of the single M2. The -helical structure was biased by restraining H-bonds NH_i···O_i4 in the range of 1.7-2.0 Å. Conformations of the side chains were MCM-optimized (N_u = 500), their starting torsions being as in the optimal -helical conformations of the corresponding monopeptides. The latter were calculated earlier (Zhorov, unpublished data) with the ZMM program and were found to coincide with those reported by Vasques et al. (1983). During the search, only side-chain torsions were randomized, but all of the torsions were varied in energy minimizations.

The optimal MEC found is a linear -helix with Ser⁸ and Thr⁴ hydroxyls H-bonding to the main-chain carbonyls of Ser⁴ and Glu¹, respectively. Ser⁸ and Thr⁴ side chains are known to face the pore. In these residues, O_Ser⁸ and H-C_Thr⁴ are the atoms farthest from the helix axis (3.9 and 5.3 Å, respectively). Hence, in the five-helix bundle, the Ser⁸ ring would be wider than the Thr⁴ ring, even if the helical axes between Ser⁸ and Thr⁴ were not tilted radially to the pore axis. Various experimental and theoretical data indicate that the Ser⁸ ring is indeed wider than the Thr⁴ ring. The latter forms the narrowest constriction in the open channel (Hucho and Hilgenfeld, 1989; Villarroel et al., 1991; Villarroel and Sakmann, 1992). A Thr⁴ to Ser⁴ mutation increases the channel permeability, probably because of an enlargement of the pore (Imoto et al., 1991). Electron microscopy images of the open channel (Unwin, 1995) and the model by Sankararamakrishnan et al. (1996) suggest that the Thr⁴ ring is narrower than the Ser⁴ ring. In the series of biscationic blockers N(Et)₃-(CH₂)₅-NR¹R²R³, the compounds with quaternary ammonium groups NR¹R²R³ are more active than the compounds of similar dimensions with tertiary or secondary amino groups (Brovtsyna et al., 1996). The smaller cationic head of these blockers supposedly binds to the Thr⁴ ring. The low activity of the blockers with the tertiary/secondary amino group indicates that the latter does not form H-bonds with Thr⁴ oxygens, implying that the CH₃_Thr⁴ rather than the HO_Thr⁴ group protrudes into the pore. The predicted dimensions of the pore at Ser⁸ and Thr⁴ rings are, respectively, 6.1 × 8.3 Å and 5.5 × 6.4 Å (Zhorov et al., 1991; Brovtsyna et al., 1996). The different dimensions of the Ser⁸ and Thr⁴ rings correlate with the nonequal protrusions of these residues into the pore, suggesting that the helices between the two rings do not have a significant radial slope to the pore axis.

A kinked-helices bundle would allow blockers to fit the Ser⁸ ring

CPZ labels Thr⁴ at the -subunit and Ser⁸ at all of the subunits of T. californica nAChR (Revah et al., 1990). The fact that CPZ reaches the Ser⁸ ring indicates that the Leu¹¹/Ala¹² ring may accommodate the tricyclic moiety of CPZ, the minimum profile dimensions of which are 5.7 × 12.2 Å. The fact that the Leu¹¹/Ala¹² ring is much wider than the Ser⁸ ring (6.1 × 8.3 Å) means that the helical axes between Leu¹¹/Ala¹² and Ser⁸ are radially tilted to the pore axis. The tilt between Leu¹¹/Ala¹² and Ser⁸ and the absence of the tilt between Ser⁸ and Thr⁴ means that M2 helices are kinked between Ser⁸ and Leu¹¹/Ala¹². This agrees with the suggestion of Unwin (1993, 1995) that M2s are kinked near the Leu¹¹ ring. The results of cysteine scanning (Akabas et al., 1994) also imply a kink near Leu¹¹: in the closed channel, positions 10, 11, and 12 are accessible to sulfhydryl reagents (Table 1), which is inconsistent with the helical structure.

Similar considerations that CPZ has to pass the rings of the bulky Val¹⁵ and Leu¹⁸ residues to reach a narrow Ser⁸ ring had led Furois-Corbin and Pullman (1989) to a funnel-like model of nAChR composed of straight (not kinked) helices. Given the dimensions of the Thr⁴ ring (5.5 × 6.4 Å) and the Leu¹¹/Ala¹² ring (at least 5.7 × 12.2 Å), the dimensions of the Ser⁸ ring in the straight-helices bundle would exceed 6.1 × 8.3 Å. Mutations of Ser⁸ affect the pore blockade by QX-222 (Charnet et al., 1990), suggesting that these residues contribute to the binding site of QX-222. The minimum profile dimensions of QX-222 (6.0 × 8.6 Å) are close to the predicted dimensions of the Ser⁸ ring (6.1 × 8.3 Å). In the model by Furois-Corbin and Pullman (1989), the side chains of Ser⁸ residues extend into the pore. In our model, side chains of Ser⁸ do not protrude into the pore, but are folded to form H-bonds with the main-chain carbonyls (Fig. 1 e). To retard QX-222 and bis-ammonium compounds at the Ser⁸ ring, the helical axes of M2s with the folded Ser⁸ side chains should be even closer to the pore axis than the helices with the extended Ser⁸ side chains. The bundle of the tilted but straight M2s does not render the Ser⁸ ring constricted enough to retard QX-222 and bis-ammonium compounds. In other words, the data on ligand binding suggest the dimensions of Thr⁴, Ser⁸, and Leu¹¹/Ala¹² rings to be 5.5 × 6.4 Å, 6.1 × 8.3 Å, and at least 5.7 × 12.2 Å, respectively. A model composed of the straight helices would not satisfy all the three requirements simultaneously. These are the "pharmacological" reasons for the kinked M2s.

View larger version (58K): [in this window] [in a new window]   FIGURE 1   C Tracing (A, B), space-filled (C, D), and ribbon (E, F) presentations of the optimal model of the nAChR channel (MP8) in the open (A, C, E) and the closed (B, D, F) states viewed from the extracellular (A-D) and membrane-exposed (E, F) sides. Along with C tracing, A and B show side chains of Thr4, Ser8, and Leu11 residues as thin lines. In the space-filled representation, Thr4 and Val15 rings are shown in dark grey, the Ser8 ring in light grey, and Leu11 in black. In the open state, Thr4 and Ser8 rings are accessible from the outside. In the closed state, Leu11 residues protrude into the pore, making it too narrow for the permeation of hydrated cations. The ribbon diagrams show two nonadjacent M2s with the pore-lining residues (positions 4, 8, 11, 12, 15, and 18) and highlight the intrasegment side chain to main-chain H-bonds.

Kinked-helix conformations of the lone M2 correspond to local energy minima

It is well known that an -helix may be terminated or kinked because of H-bonds between Ser/Thr side chains and the main-chain carbonyls in the preceding helical turn (Levitt, 1978; Bolin et al., 1982). To test whether such H-bonds would stabilize kinks in the isolated -helix, we carried out MCM calculations of the entire M2 and its undecapeptide models (Ala)₄-X-(Ala)₄, where X = Ile-Ser-Val and X = Leu-Thr-Leu. Both main-chain and side-chain torsions were randomized in these calculations. Relatively short MCM trajectories (N_u = 500) starting from the ideal -helical conformations yielded many MECs with various kinks, and Ser and Thr side chains H-bonding to the main-chain carbonyls. Similar MCM calculations of the undecapeptides with X = Ile-Ala-Val and X = Leu-Leu-Leu did not yield kinks. A longer MCM trajectory of M2 (N_u = 3000) yielded many MECs within 7 kcal/mol of the apparent global minimum, with Ser and Thr side chains H-bonding to the main-chain carbonyls. None of these MECs were kinked in the middle of the helix, but many MECs were kinked at the helical ends. Thus the linear helices are more stable than the kinked ones, implying that the kinks are not an inherent property of M2s.

These results are in good agreement with the solid-state NMR studies in micelles, demonstrating that a 25-residue peptide with the sequence of T. californica -M2 is helical throughout its length, except for a few residues at the N- and C-terminal ends (Opella et al., 1997). Qualitatively similar results were obtained by molecular dynamics calculations of lone M2s (Sankararamakrishnan and Sansom, 1994). In nAChR, the kinks may be stabilized by interactions of M2s with other transmembrane segments and with intrapore waters. Noteworthy is the observation that Ser⁸ has conservative neighbors Ile/Val, the bulky -branched side chains of which should avoid the hydrophilic environment of the pore and hence may cause backbone deformation.

The model of the open channel

The above considerations suggest that M2 helices of nAChR should diverge at the synaptic end of the pore and that intersegment interactions should stabilize the kinks initially proposed by Unwin (1993). In the absence of other transmembrane segments, a five-helix bundle model with the diverging M2s is not expected to correspond even to a local minimum of energy. Therefore, restraints are necessary to keep together the helices diverging from the kinks to the synaptic end of the pore. On the other hand, the cytoplasmic parts of M2s (from the kinks to the N ends) may come in contact with each other. Contacts between blockers and the channel would depend on conformations and orientations of M2s and, hence, on the restraints supporting the diverging parts of M2s. We considered the restraints as additional variables in the search for a model of the pore that would be optimal energetically and agreeable with the available experimental data.

We created 10 models of the pore (designated MP1-MP10) by restraining C atoms of the pore-facing residues 4, 8, 15, 18, and 19 at various distances from the pore axis (Table 2). The axis was represented by a set of fixed pseudo-atoms (that do not interact with the channel and the blockers) separated from each other by 1 Å. Starting structures were assembled as symmetrical bundles of the straight helices (corresponding to the optimal MECs of the lone M2) with the helical axes parallel to the pore axis. MCM trajectories (N_u = 500) were calculated for each model, with the main-chain torsions in the residues 1-8 and 12-21 kept rigid. The obtained MECs with energies below 7 kcal/mol from the apparent global minimum were further minimized, with all of the torsions allowed to vary. Although each MCM trajectory started with the straight helices, the requirements for Thr⁴ and Ser⁸ to be equally distant from the pore axis and for Val¹⁵ and Leu¹⁸ to be increasingly distant from the pore axis yielded kinks in the flexible region Val⁹-Ala¹².

View this table: [in this window] [in a new window]   TABLE 2   Models of the pore (MP1-MP10) obtained with restraints (D, Å) between the five C atoms in the ring of the given residues and the nearest pseudo atom (X) at the pore axis

Models MP2, MP3, and MP5 (Table 2) were rejected after a visual inspection because they had small kinks inconsistent with the requirements for the binding of the blockers. For other models, the energy profiles of the six blockers shown in Table 3 were calculated. A starting point of each profile was obtained by positioning the blocker at the extracellular extension of the pore axis. At each point of the profile, atom N⁺ of the blocker (in bisammonium blockers, atom N⁺ in the smaller cationic head) was restrained to a pseudo-atom at the given level of the pore axis, and the energy was minimized with M2 positions, orientations, and backbone torsions kept fixed. A realistic model was expected to provide a free access for all of the blockers to the pore, but impose energy barriers of at least 7 kcal/mol that would retard the blockers at the binding sites known from experiments.

Most of the models failed to yield energy profiles satisfying the above requirements. For instance, the profile for CPZ in MP4 showed a high energy barrier at the level of Leu¹¹ that would not let CPZ approach the Ser⁸ ring. The profile for compound (III) in MP10 did not have a barrier at the Thr⁴ ring that would retard the thrimethylammonium head of the blocker. In such a channel, compound (III) would leak through the pore. The profiles for compound (II) in MP1 and MP6 had high barriers at the Ser⁸ ring that would not let the trimethylammonium head reach the Thr⁴ ring. The best agreement between the calculated energy profiles and the experimental data on the location of the binding sites of the compounds (I)-(VI) was found for MP8. Further refinement of MP8 (see below) decreased its energy without significant changes in the geometry. The C tracing, space-filled, and ribbon presentations of the refined model MP8 are shown in Fig. 1, a, c, and d, respectively. Some geometrical parameters of model MP8 are given in Table 4.

Sankararamakrishnan et al. (1996) used restraints derived from the electron microscopy data by Unwin (1993, 1995) to create a kinked-helices model of nAChR with a rather wide pore. To account for the experimental data on the permeation of organic cations through the pore and on the blockade of nAChR, the authors suggested that the real dimensions of the pore are smaller because of wall-bound waters. Our model agrees less with the electron microscopy images, but it does not require wall-bound waters to explain the well-known data on channel permeability and blockade. In the electron microscopy image of the open channel, the intracellular parts of M2s slope tangentially to the pore axis by ~45° and form a right-handed barrel. Unwin (1995) noted that the tangential slope of the helices provides a larger exposure of Ser⁸ and Thr⁴ in the pore. In our model, the requirement for Thr⁴ and Ser⁸ side chains to face the pore caused a right tangential slope of ~10° in the cytoplasmic parts of M2s. The synaptic parts of M2s in our model have a radial slope of ~40°, which is similar to that observed in the closed channel (Unwin, 1993). Because of this slope, the parts of M2s from the kinks to the C-ends diverge, so that M1s would penetrate partially between M2s. Because M1s and M2s span the membrane in opposite directions, only N-parts of M1s may be accessible from the pore. This agrees with the prediction of Furois-Corbin and Pullman (1989) and with the experimental data (DiPaola et al., 1990; Akabas and Karlin, 1995).

Blockers in the pore

The energy-minimized profiles for the blockers (I), (II), and (IV) in model MP8 are given in Fig. 2. The total conformational energy (Fig. 2 c) and its electrostatic (Fig. 2 a) and nonbonded (Fig. 2 b) components are calculated for different positions of the ligands along the pore. The electrostatic profiles for all of the blockers are similar. They are determined by the interactions of Thr⁴ and Ser⁸ hydroxyls with the positive charge at the blockers' ammonium group. For each of the effective blockers (II), (III), (V), and (VI), the deepest minimum of nonbonded energy (Fig. 2 b) coincides with the minimum of electrostatic energy. For the compounds (I) and (IV), the minima of nonbonded and electrostatic energy do not coincide. These compounds are poor blockers of nAChR (Brovtsyna et al., 1996).

View larger version (16K): [in this window] [in a new window]   FIGURE 2   Energy-minimized profiles of the compounds I (bold lines), II (thin lines), and IV (dashed lines) in the open-channel model of nAChR (MP8). (A) Electrostatic energy. (B) Nonbonded energy. (C) Total energy. A coordinate of a blocker specifies the position of atom N+ in the blocker's smaller cationic head relative to the RMS plane drawn via C atoms of the Thr4 ring. The negative coordinates correspond to the extracellular direction. The energies are normalized to be zero at the entry of the pore, where the blockers practically do not interact with the channel. The electrostatic energy profiles for all of the blockers have wide and deep minima between Leu11 and Thr4 rings. The nonbonded energy profiles have relatively shallow minima with the steep energy barriers between Ser8 and Thr4 rings. The positions of these barriers correlate with the dimensions of the compounds (the smaller blockers penetrate deeper into the pore). The N+Et3 group of compound (I) cannot pass the Ser8 ring. The N+Me3 group of compound (II) meets sterical hindrances at the level of the Thr4 ring. The ammonium group of compound (IV) can pass through the channel.

An energy-minimized profile strongly depends on the starting structure and, hence, provides only a rough estimate of the interactions between a blocker and the channel. To eliminate the influence of rather arbitrary starting conformations, we calculated MCM-optimized profiles for all of the blockers in MP8. Each profile was obtained by calculating 35 MCM trajectories (N_u = 100) for 35 positions of the blocker. The starting point in each trajectory was the MEC obtained in the corresponding position of the energy-minimized profile. In each trajectory, atom N⁺ of the blocker was restrained to a pseudo-atom in the pore axis, and conformations and orientations of the blocker and M2s were MCM-optimized. Fig. 3 shows the energy-minimized and MCM-optimized profiles of CPZ in the pore. The MCM optimization yielded a significant decrease in the conformational energy (Fig. 3 c) and its electrostatic (Fig. 3 a) and nonbonded (Fig. 3 b) components as compared to the energy-minimized profile. The wide minimum in the MCM-optimized profile suggests a certain mobility of CPZ in the pore. Both energy-minimized and MCM-optimized profiles have similar coordinates of the nonbonded energy barriers that retard CPZ and other blockers (profiles are not shown) at the corresponding binding sites. The coordinates of these barriers at the energy-minimized profiles were the major criteria for selecting the optimal model MP8. The fact that MCM-optimized profiles yielded similar coordinates of the barriers is evidence of the validity of MP8.

View larger version (17K): [in this window] [in a new window]   FIGURE 3   Energy-minimized (thin lines) and MCM-optimized (bold lines) energy profiles of chlorpromazine (CPZ) in the open-channel model of nAChR (MP8) with the fixed backbone torsions. (A) Electrostatic energy. (B) Nonbonded energy. (C) Total energy. A coordinate of CPZ specifies the position of its atom N+ relative to the RMS plane drawn via C atoms of the Thr4 ring. The negative coordinates correspond to the extracellular direction. The profiles are normalized to have a zero energy at the entry of the pore, where CPZ practically does not interact with the channel.

Fig. 4 shows the lowest-energy complexes of the blockers with nAChR that were found in calculations of MCM-optimized energy profiles. The optimal mode of binding of compound (I) shows one of two NEt₃ groups retarded at the Ser⁸ ring without reaching the Thr⁴ ring (Fig. 4 a). This agrees with the prediction by Brovtsyna et al. (1996). The week electrostatic interactions of the blocker with the Ser⁴ and Thr⁴ rings (Fig. 2 a) may explain its low activity. In compound (II), NMe₃ group binds between the Ser⁸ and Thr⁴ rings, whereas the bulky NEt₃ group interacts effectively with Ala¹² and Leu¹¹ residues (Fig. 4 b). This binding mode disagrees with our earlier prediction that the small and the large cationic heads of the compound (II) would interact with the Thr⁴ and Ser⁸ rings, respectively (Brovtsyna et al., 1996). Our present calculations confirm the prediction by Tikhonov et al. (1996) that the NH₃ group of compound (IV) would pass through the Thr⁴ ring. The minima of nonbonded and electrostatic interactions of compound (IV) with MP8 do not coincide with each other. This explains the low activity of compound (IV). Both energy-minimized and MCM-optimized profiles for QX-222 predict that the ligand's NMe₃ group and aromatic moiety bind, respectively, at the Ser⁸ ring and the Leu¹¹/Ala¹² rings (Fig. 4 c). Substitution of Ala¹² by Ser¹² would weaken the binding of the hydrophobic ring of QX-222. This agrees with the experimental data of Charnet et al. (1990). [H³]CPZ labels position 8 in five M2s, position 11 in two M2s, and position 4 in only -M2 of the T. californica nAChR channel (Revah et al., 1990), suggesting an asymmetry of the pore not considered in our model. The optimal binding mode of CPZ (Fig. 4 d) shows the phenothiazine moiety interacting with Leu¹¹, Ala¹², and carbonyl oxygens of the Ser⁸ ring; the amino group of CPZ interacts with the side-chain oxygens of the Ser⁸ ring and approaches the Thr⁴ ring. Because the photoreactive phenothiazine moiety of CPZ is far from the Thr⁴ ring, the model does not explain the labeling of Thr⁴ in -M2. Another low-populated binding mode of CPZ may exist with the alkylamino chain folded up to let one of the aromatic rings approach Thr⁴. An intercalation of the tricyclic moiety of CPZ between M2 segments also cannot be ruled out.

View larger version (101K): [in this window] [in a new window]   FIGURE 4   Optimal complexes of nAChR with the blockers. Only two nonadjacent M2 segments are shown. The Thr4 ring is shown in dark grey, the Ser8 ring in light grey, and Leu11 in black. (A) A bulky N+Et3 group of compound (I) cannot pass the Ser8 ring and reach the Thr4 ring. (B) Compound (II) binds to the Ser8 and Thr4 rings by the N+Me3 group, whereas the N+Et3 group binds to the Leu11 and Ala12 rings. (C) The aromatic group of QX-222 interacts with Leu11 and Ala12 rings, and the N+Me3 group with the Ser8 ring, in agreement with mutagenesis experiments by Charnet et al. (1990). (D) The phenothiazine moiety of the CPZ molecule interacts with Leu11 and Ala12 rings, whereas the cationic head interacts with the Ser8 and Thr4 rings, in agreement with affinity labeling experiments by Revah et al. (1990).

Various noncompetitive blockers of nAChR demonstrate a tissue selectivity. For example, procaine blocks muscle and neuronal nAChRs with different voltage dependencies (Cuevas and Adams, 1994). It is unclear whether the difference is due to peculiarities of the pore-lining residues or of M2 backbone conformations. Mutagenesis experiments may rule out one of these possibilities. For example, if an engineered muscle receptor with the pore-lining residues from a neuronal receptor would mimic the latter by its pharmacological properties, it would indicate that the two receptors have similar backbone conformations of M2s.

In the present model, Ser⁸ and Thr⁴ hydroxyls form H-bonds with the main-chain carbonyls at the previous helical turns. Hydroxyl oxygens of Thr⁴ and Ser⁸ protrude into the pore and form two hydrophilic rings important for the permeation of cations and for the binding of the blockers' positively charged groups (note wide minima of electrostatics energy between the Ser⁸ and Thr⁴ rings, Fig. 2 a). The kinks release Ser⁸ main-chain oxygens from H-bonds with HN  Ala¹² groups. Lone electron pairs of these oxygens face the pore, forming an additional hydrophilic ring (Fig. 5 a) whose location between the wide extracellular part of the channel and the narrow part should be important for the permeation. This may be a "physiological" reason for the kinks.

View larger version (71K): [in this window] [in a new window]   FIGURE 5   Space-filled (A, B) and ribbon (C, D) presentations of the kinked region of nAChR pore in the open (A, C) and closed (B, D) conformations. In the open state, main-chain oxygens (dark grey) of Ser8 residues (light grey) protrude into the pore and form a polar ring that may be important for the permeation of cations. In the closed state, this ring is shielded by the hydrophobic side chain of Leu11 (black). The ribbon diagrams show the side chains of Thr4, Ser8, and Leu11 residues and highlight H-bonds that stabilize the kinks.

A model of nAChR is also expected to explain the permeability of the channel for organic cations. Different experiments predict different dimensions of the pore. Dwyer et al. (1980) approximated the minimal cross-sectional dimensions of the open pore in the frog muscle receptor by a rectangle of 6.5 × 6.5 Å. Cohen et al. (1992) estimated the pore diameter to be as large as 8.4 Å. Nutter and Adams (1995) predicted the minimal diameter of the neuronal receptor to be 7.6 Å. Taking into account the conformational flexibility of the organic cations and measuring the minimum silhouettes of their high-populated conformations, Brovtsyna et al. (1996) estimated the minimum dimensions of the neuronal receptor to be 5.5 × 6.4 Å. The pore dimensions in our present model are close to those predicted by Dwyer et al. (1980) and Brovtsyna et al. (1996). The essential difference in the predicted dimensions of the pore may be due to the peculiarities of the receptor subtypes studied. It should be noted, however, that a circle or a rectangle is a rough approximation for the lumen cross section, and the channel walls as well as organic ions are conformationally flexible. Calculation of MCM-optimized profiles for permeating organic cations may help correlate their chemical structure with the permeation properties. This interesting problem was not addressed in the present study.

Channel gating

Mutation of Leu¹¹ residues affects desensitization and gating of nAChR (Revah et al., 1991; Filatov and White, 1995; Labarca et al., 1995). A possible mechanism of the gating assumes notable rotations of entire M2s, with Leu¹¹ residues exposed in the pore, precluding the permeation of the hydrated cations and the kinked regions acting as molecular swivels (Unwin, 1995). Indeed, noncompetitive antagonist 3-(trifluoromethyl)-3-(m-[¹²⁵I]iodophenyl)diazirine (TID) labels Val¹⁶ and Leu¹¹ rings in the closed channel and Val⁶, Leu¹¹, Thr⁴, and Ser⁸ rings in the open channel (White et al., 1991; Cohen et al., 1992). However, cysteine scanning did not reveal an essential difference in the accessibility of individual residues of M2s in the open and closed states of the channel (Akabas et al., 1994; Akabas and Karlin, 1995). If pore-facing hydrophobic residues form a constriction whose dimensions are comparable with those of hydrated cations, even minor conformational changes in the residues may dramatically affect the permeation, thus causing an elementary phase of the gating (Zhorov and Ananthanarayanan, 1996). Leu¹¹ residues are located at the kinks, where the split system of H-bonds provides a high flexibility to the backbones. Some conformations of the flexible backbones may represent the closed states of the channel, whereas other conformations may correspond to the open states, suggesting that rotations of entire M2s may not be necessary for channel gating.

To explore this possibility, we calculated a long MCM trajectory (N_u = 2000) of the five-helix bundle. Positions and orientations of M2s and the torsions of the residues 1-8 and 13-21 were kept fixed, the torsions of the residues 9-12 being randomized. The search yielded 45 different structures with energies up to 7 kcal/mol from the apparent global minimum. In these structures, the kinked regions of M2s had a split network of -helical H-bonds. However, each kinked region was stabilized by at least one intrasegment H-bond. We found conformers with the -helical H-bonds Ser⁶··Leu¹⁰ and Ile⁷···Leu¹¹ (Fig. 5 c), as well as conformers with the 3₁₀-helical H-bonds Ser⁶···Val⁹ and Ile⁷···Leu¹⁰ (Fig. 5 d). In the conformers with the -helical H-bonds, Ala¹² faced the pore, suggesting that such conformers represent the pore-opening states of M2s. In the conformers with the 3₁₀-helical H-bonds, Leu¹¹ faced the pore, suggesting that these conformers may represent the pore-closing states of M2s. The 45 collected structures of the five-helix bundle comprised 225 different conformers of M2s, among which 52 conformers corresponded to the pore-closing states.

At the next stage, we selected two structures of the bundle corresponding to the open and to the closed channel, and energy-minimized them, with all of the torsions allowed to vary. The obtained models are shown in Fig. 1 and the torsions of one of the M2s in Fig. 6, and energies are given in Table 5. The structure shown in Fig. 1, a, c, e, represents the open channel with Ala¹² facing the pore. The structure shown in Fig. 1, b, d, f, represents the closed channel with Leu¹¹ residues protruding into the pore and diminishing its radius up to 1.75 Å. This pore is too narrow to let through the hydrated Na⁺, whose radius is 2.4 Å (Moore, 1972).

View larger version (17K): [in this window] [in a new window]   FIGURE 6   Backbone torsions versus residue numbers of M2 in the pore-opening (A) and pore-closing (B) conformations.

In the closed channel, we also found a candidate for another 3₁₀-helical H-bond, Leu¹¹_NH···O_Ser⁸, with a distance of 2.54 Å (Fig. 5 d). Our model with the rigid valence geometry did not allow further approach by Leu¹¹_NH of O_Ser⁸. However, this H-bond may occur in the real channels. The hydrophobic side chain of Leu¹¹ would stabilize this H-bond by protecting it from the intrapore waters (Fig. 5 b), whereas in M2 with Ser¹¹, the waters would destabilize it and hence the closed state of the channel. Indeed, Leu¹¹ to Ser¹¹ mutation does destabilize the closed state of nAChR (Labarca et al., 1995). Thus the gating of the nAChR channel may be due to conformational changes in the short flexible chains (involving Leu¹¹ residues) that connect more stable -helical parts of M2s. This model does not require energetically expensive rearrangements of the transmembrane segments; the latter may be necessary to transfer the channel to the long-lasting desensitized states (Feltz and Trautmann, 1982).

At the synaptic end of the pore, M1s may fill the space between the diverging M2s. Cysteine mutants in the N end part of M1 are labeled by sulfhydryl reagents, with accessibility of residues highly dependent on the presence of acetylcholine (Akabas and Karlin, 1995). M1s are close to the acetylcholine recognition site of nAChR (Devillers-Thiery et al., 1993) and may link this site with the effector part of the gating mechanism. For example, agonist-induced conformational changes in M1s may shift an equilibrium between the open and the closed states in the kinked regions of M2s. This mechanism would explain why mutations in the inner, intermediate (Glu¹) and Thr⁴ rings affect mainly the ion permeation via the channel (Imoto et al., 1988, 1991; Konno et al., 1991; Villarroel et al., 1991). These rings are at the cytoplasmic part of the pore, where M2s come in contact with each other, so that their conformations do not depend on the channel's gating. Contrarily, in the synaptic part, M2s interact with M1s that may transmit the signal from the acetylcholine recognition site to the gate. Indeed, mutations at this region affect the gating and desensitization of nAChR (Revah et al., 1991; Labarca et al., 1995).

Essentially different roles were proposed for Leu¹¹ residues: occlusion of the pore during desensitization (Revah et al., 1991; Bertrand et al., 1993), stabilization of the closed state of the channel by interacting with each other (Unwin, 1993), governing channel gating (Labarca et al., 1995), and setting the mean open time through interactions with other regions of the channel rather than forming the gate itself (Filatov and White, 1995). Our model visualizes the former two roles. Below we propose a possible explanation of the experiments that question the gating role of Leu¹¹ residues.

Based on the finding that Leu¹¹ substitution by Thr in only one of the five M2s stabilized the open state of the channel and assuming that the remaining four Leu¹¹ residues would still effectively occlude the pore, Filatov and White (1995) concluded that Leu¹¹ residues do not form the gate. However, a permeation of a hydrated cation via a ring of hydrophobic residues may be highly sensitive, even to small changes in the dimensions of the ring (Zhorov and Ananthanarayanan, 1996). Cations should permeate via the open hydrophobic gate with the water entourage. If the dimensions of the open gate match those of the hydrated cation, exposing only one Leu¹¹ residue to the agonist-bound channel would transfer it in a nonconducting state. In our model, conformational transitions in each M2 are independent. Brief channel closings during a burst can be explained as stochastic transitions of one of M2s to the "closed" conformation. Consecutive mutations of each Leu¹¹ by Ser or Thr may progressively decrease the probability of the nonconducting states of the agonist-bound channel and, hence, increase the probability of the open states. If only a portion of the agonist-bound channels are open simultaneously, these mutations should also increase the population of the open channels, i.e., their apparent affinity to agonists. These effects were observed in the experiments by Labarca et al. (1995) and Filatov and White (1995). These experiments also demonstrated a gating in the channels with all five Leu¹¹ residues substituted by Ser or Thr, suggesting that residues at other positions may protrude into the pore and block the ion permeation.

Akabas et al. (1994) demonstrated that the closed cannel is more accessible to a sulfhydryl reagent ethylammonium derivative of methanethiosulfonate (MTSEA) than the open channel (see Table 1). Based on these data and on the fact that the replacement of the extracellular Na⁺ by MTSEA did not increase the leak current, Akabas et al. (1994) suggested that the gate of the channel is at least as cytoplasmic as Glu¹. However, MTSEA may pass the closed Leu¹¹ gate in the deprotonated state. In this state, the amino group is poorly hydrated, and the minimum silhouette of MTSEA is rather small.

Our interpretations of the experiments on the channel gating do not rule out the original interpretations. Various conducting and nonconducting states of nAChR may be possible, and our model may visualize only some of the possibilities. Further studies are necessary to prove or disprove the proposed mechanism of the gating.

	CONCLUDING REMARKS

Top Abstract Introduction Results & Discussion Concluding Remarks References

The model of the nAChR pore composed of the consensus M2s explained general properties of the nAChR pore and peculiarities of its blockade by various compounds. Like any atomic-scale computational model, our model presents details with a resolution not achievable by contemporary experimental methods. Of course, not all of these details are equally reliable. A single model cannot accommodate all of the available experimental data with their original interpretations. Thus our model reproduces kinks in M2s observed in electron cryomicroscopy studies, but it is not consistent with the large diameter of the pore implied by the same experiments. Our model supports the idea that Leu¹¹ residues are directly involved in the channel gating, but it is inconsistent with other possible interpretations of mutagenesis experiments involving Leu¹¹. We have proposed alternative interpretations that may be tested in further experiments. For example, Leu¹¹ to Ile or Val substitutions may affect the channel gating, particularly equilibrium between the conducting and nonconducting states of agonist-activated channels.

Present calculations ignored other transmembrane segments, intrapore waters, and entropy contributions to the free energy. Considering these and other factors should improve the capacity of the models to explain the existing experimental data and make it possible to design new experiments.