Identification of a Novel Pharmacophore for Peptide Toxins Interacting with K+ Channels*

κM-conotoxin RIIIK blocks TSha1 K+ channels from trout with high affinity by interacting with the ion channel pore. As opposed to many other peptides targeting K+ channels, κM-RIIIK does not possess a functional dyad. In this study we combine thermodynamic mutant cycle analysis and docking calculations to derive the binding mode of κM-conotoxin RIIIK to the TSha1 channel. The final model reveals a novel pharmacophore, where no positively charged side chain occludes the channel pore. Instead the positive-charged residues of the toxin form a basic ring; κM-RIIIK is anchored to the K+ channel via electrostatic interactions of this basic ring with the loop and pore helix residues of the channel. The channel amino acid Glu-354 is likely to be a fundamental determinant of the selectivity of κM-RIIIK for the TSha1 channel. The Cγ-OH of Hyp-15 is in contact with the carbonyls of the selectivity filter, disturbing the charge distribution pattern necessary for the coordination of K+ ions. This novel, experimentally based pharmacophore model proves the existence of diverse binding modes of peptidic toxins to K+ channels and underlines the role of intermolecular electrostatic interactions involving channel loop side chains in determining the selectivity of toxins for specific K+ channel types.

M-conotoxin RIIIK blocks TSha1 K ؉ channels from trout with high affinity by interacting with the ion channel pore. As opposed to many other peptides targeting K ؉ channels, M-RIIIK does not possess a functional dyad. In this study we combine thermodynamic mutant cycle analysis and docking calculations to derive the binding mode of M-conotoxin RIIIK to the TSha1 channel. The final model reveals a novel pharmacophore, where no positively charged side chain occludes the channel pore. Instead the positivecharged residues of the toxin form a basic ring; M-RIIIK is anchored to the K ؉ channel via electrostatic interactions of this basic ring with the loop and pore helix residues of the channel. The channel amino acid Glu-354 is likely to be a fundamental determinant of the selectivity of M-RIIIK for the TSha1 channel. The C␥-OH of Hyp-15 is in contact with the carbonyls of the selectivity filter, disturbing the charge distribution pattern necessary for the coordination of K ؉ ions. This novel, experimentally based pharmacophore model proves the existence of diverse binding modes of peptidic toxins to K ؉ channels and underlines the role of intermolecular electrostatic interactions involving channel loop side chains in determining the selectivity of toxins for specific K ؉ channel types.
Potassium (K ϩ ) channels are a very diverse group of proteins that are key elements for a variety of different physiological functions including the electrical excitability of cells (1). During evolution a great variability of peptide toxins targeting different K ϩ channels have been evolved by several venomous organisms including snakes, spiders, and scorpions. The venoms of the marine cone snails are known to contain different families of peptides, the so-called conotoxins, which target ligandgated and voltage-gated ion channels with great specificity. Meanwhile several conotoxins have been identified that have been shown to interact with voltage-activated K ϩ channels. This includes the -conotoxins and the M-conotoxins (2).
Recently the structure of M-conotoxin RIIIK (M-RIIIK), 1 a peptide obtained from a venom duct library from Conus radiatus, was solved by NMR analysis (3). M-RIIIK is known to block Shaker and mammalian Kv1.2 K ϩ channels, whereas the highest affinity target so far identified is TSha1, a Shakerrelated K ϩ channel from trout (4,5). Despite the entirely different pharmacological specificity, a high structural similarity has been observed between this peptide and -conotoxin GIIIA, which specifically blocks Na ϩ channels, or -conotoxin PIIIE, which is a non-competitive blocker of nicotinic acetylcholine receptors.
Structurally and phylogenetically unrelated toxins that interact with voltage-activated K ϩ channels usually share a dyad motif composed of a lysine and a hydrophobic amino acid residue (Tyr or Phe). In particular, there is evidence that the lysine residue of this functional dyad occludes the K ϩ channel pore (6 -8). This dyad has been proposed to be the minimal core domain of the K ϩ channel binding pharmacophore (9 -12). A systematic mutational analysis conducted for all non-cysteine residues of M-RIIIK showed that M-RIIIK does not contain a functional dyad (3), in contrast to most K ϩ channel-targeting peptidic neurotoxins. In M-RIIIK all the residues that are most relevant for function contain a positive charge; these residues define a basic ring separating two peptide surfaces. The peptide residues that are moderately relevant for binding cluster on one of the two surfaces, whereas those that are irrelevant for binding occupy the other surface. Based on the lack of the dyad and on the observation that all the functionally relevant residues are clustered on one surface of the peptide delimited by a ring of positive charges, a novel pharmacophore model was hypothesized for the M-RIIIK-TSha1 complex. In this model the ring of positive charges is used as an anchor to residues of the K ϩ channel loops, and M-RIIIK blocks the channel by covering the pore as a lid (3). Additionally, no positively charged side chain penetrates the channel pore. The importance of a ring of basic residues for potassium channel binding was also proposed for the scorpion toxin Pi1 by Mouhat et al. (13), indicating that a similar pharmacophore might have independently evolved for cone snails and scorpions.
In this study we investigate the interaction of M-RIIIK with the pore of the TSha1 K ϩ channel. We use the data from an extensive mutant cycle analysis to calculate the orientation of the peptide within the pore of the TSha1 channel, whose structure was derived by homology to the known KcsA structure (14). Previous work (7,15) has demonstrated the validity of a computational approach based on intermolecular distance restraints, which can be derived by mutant cycle data, for obtaining models of the atomic interactions in bimolecular complexes. In this study the solution structure of M-RIIIK is docked to the TSha1 channel, allowing full flexibility of the side chains of both the peptide and the protein and using 17 intermolecular distance restraints derived from experimental mutant cycle data. The resulting model for the M-RIIIK-TSha1 complex indicates that M-RIIIK indeed occludes the permeation pathway as a lid, which is anchored by the electrostatic interaction of a ring of basic residues on the peptide surface with the outer vestibule of the ion channel pore. All functionally relevant residues belong to the interaction surface with the channel, whereas functionally irrelevant side chains are mostly projected toward solution. Specific electrostatic interactions between the positively charged side chains of the toxin and channel residues of the loop and outer helix regions are likely to be responsible for the selectivity of M-RIIIK for the TSha1 channel. Furthermore, our model suggests that the trans-hydroxyproline 15 of M-RIIIK, which is the homologous residue of Arg-13 of -conotoxins GIIIA, interacts with the carbonyl groups within the selectivity filter of the channel, thereby occupying the permeation pathway. Arg-13 of -conotoxins has been demonstrated to be essential for blocking the pore of Na ϩ channels (2,16,17). Our finding indicates that pharmacologically diverse but structurally similar peptides might interact with their specific targets in a analogous way despite the large differences in the amino acids sequence.

MATERIALS AND METHODS
Molecular Biology-The original DNA clone of the Shaker-related K ϩ channel (TSha1) from the central nervous system of rainbow trout (Onchorychus mykiss) was kindly provided by Prof. Dr. G. Jeserich, University of Osnabrü ck, Germany. In vitro site-directed mutagenesis of TSha1 was performed following standard procedures (18,19) and according to manufacturer protocols when commercial kits were used. The vector used throughout this study was pSGEM (3118 bp), a modified version of pGEMHE, which was a generous gift of Prof. Dr. Michael Hollmann (Bochum University, Bochum, Germany).
Electrophysiological Recordings and Data Analysis-The two-electrode voltage clamp technique in the Xenopus oocyte heterologous expression system was used to investigate the affinities of the toxin mutants for the different channel mutants. Oocyte injection and maintenance was performed as already described (20). The WT and mutant analogs of M-RIIIK were synthesized, folded, and tested as already described (3,4). The IC 50 values for the block of TSha1 channels were calculated from the peak currents at a test potential of 0 mV according to IC 50 ϭ fc/(1 Ϫ fc) ϫ [Tx], where fc is the fractional current, and [Tx] is the toxin concentration. Data are given as the mean Ϯ S.D. The calculation of the interaction free energy (⌬⌬G) from the electrophysiological data were performed as described earlier (20).
Docking Calculations-The model of the TSha1 channel was obtained by homology to the crystal structure of the KcsA channel (14) with a procedure similar to that described by Eriksson and Roux (7) to obtain the Shaker channel model. A large number of spatial restraints, extracted from the KcsA PDB file (1BL8), were optimized using the program MODELLER Version 6.2 to generate similarity-based models of TSha1 (21). The ideal 4-fold symmetry of the tetrameric channel was imposed on all the models. Of the 100 generated models, the one with the lowest MODELLER restraint energy was kept and refined by energy minimization with the program CNS solve 1.1 (22) using the "proteinallhdg" forcefield.
The docking protocol was based on a similar procedure as that used by Eriksson and Roux (7). 875 docking models of the complex between the TSha1 channel model structure and the NMR structure of Mconotoxin RIIIK were generated with the program CNS solve 1.1 (22) and using mutant cycle derived intermolecular distance restraints. In the thermodynamic double mutant cycle analysis, two interacting molecules A and B are mutated at specific sites a and b to give two molecules A* and B*. The binding free energies ⌬G for the wild type and single and double mutant complexes A-B, A*-B, A-B*, and A*-B* are combined to obtain the interaction free energy, The ⌬⌬G int is a measure of the non-additivity of the mutations a* on A and b* on B and reveals a direct interaction of the sites a and b (23,24). In these studies the ⌬⌬G values of Fig. 3 were translated into intermolecular distances as follows: (a) 2 kJ/mol Ͻ ⌬⌬G Յ 3 kJ/mol corresponds to a very weak intermolecular distance (d) restraint between sites a and b (4 Å Ͻ d Ͻ 6 Å); (b) 3 kJ/mol Ͻ ⌬⌬G Յ 4 kJ/mol corresponds to a weak intermolecular distance restraint (3.5 Å Ͻ d Ͻ 5.5 Å); (c) 4 kJ/mol Ͻ ⌬⌬G Յ 5.5 kJ/mol corresponds to a medium distance restraint (2.5 Å Ͻ d Ͻ 4.5 Å); ⌬⌬G Ͼ 5.5 kJ/mol corresponds to a strong distance restraint (1.5 Å Ͻ d Ͻ 3.5 Å). If a and b are polar or charged residues, the minimum distance between the heavy atoms of the polar or charged groups has to fulfill the intermolecular distance restraint; for non-polar residue pairs the minimum distance between all heavy atoms of the side chains is used. Because of the 4-fold symmetry of the TSha1 channel we used ambiguous distance restraints; the distance d ij between a residue of the toxin (i) and a residue of the channel (j) is described as, , where A, B, C, and D are the four identical subunits of TSha1.
To avoid redundant identical complexes, all distances from the residue Arg-10 of M-RIIIK were assigned to the subunit A of the channel. The NMR structure of M-conotoxin RIIIK shows that Arg-10 and Lys-18 are on opposite sides of the peptide; to ensure that the toxin is stretched over the pore, we impose that Lys-18 is close to the extracellular loop of the channel unit opposite to that in contact with Arg-10 (unit C). A total of 16 distance restraints were used: three very weak ). An additional distance restraint was added between the N terminus NH 3 ϩ of the toxin and any electronic acceptor of the channel, following the dramatic decrease in the toxin affinity upon acetylation of the terminal amine (3). All distances were restrained with a harmonic potential.
The docking protocol consisted of a series of short torsion angle Molecular Dynamics trajectories. In the first step of the simulated annealing protocol the temperature was increased to 1000 K (500 steps of 10 fs for a total of 5 ps); during this phase the van der Waals radius of the atoms was decreased from 1 to 0.1. In the second step the temperature was slowly lowered to 300 K (1000 steps of 15 fs for a total of 15 ps), and the van der Waals radius was increased from 0.1 to 1. At the beginning of the trajectories, the toxin was positioned randomly at a distance of 15 Å from the channel. All docking trajectories were started with different initial conditions by changing the initial velocities and the initial orientation of the toxin with respect to the channel pore axis. During the simulated annealing the electrostatic energy was turned on toward the end of the dynamics. A continuum solvent model was used with a dielectric constant ⑀ ϭ 12 (7). The distance restraint force was fixed to 300 kcal mol Ϫ1 Å Ϫ2 during the dynamics.
Flexibility was allowed for the conformation of the toxin backbone of the amino acids 1-3 and of all the toxin side chains. The N terminus backbone of the peptide was allowed to change conformation with respect to the NMR structure following the observation that this region is highly flexible in solution (3). The peptide backbone of residues 4 -24 was restrained with an harmonic potential based on the root mean square deviation from the NMR structure with a force of 500 kcal mol Ϫ1 Å Ϫ2 . This allows rotation and translation of the toxin with respect to the channel while keeping the overall folding close to that experimentally determined. The conformation of the trans-membrane helices of the channel was kept fixed, the backbone of residues 336 -386 was restrained to that of the initial model with an harmonic potential and a force of 300 kcal mol Ϫ1 Å Ϫ2 , whereas the side chains of residues 336 -386 were allowed full flexibility.
The MD were followed by 6 cycles of 500 steps of energy minimizations where the backbone restraint function was turned off and the distance restraints were decreased to 75 kcal mol Ϫ1 Å Ϫ2 . During the minimization the electrostatic potential was turned on.
The trajectories were analyzed using the g_cluster module of the GROMACS 3.1 program (25,26). The clusters were generated using distances r.m.s.d. with the GROMOS method (27) and a cutoff of 0.2 nm. The r.m.s.d. used to define the clusters is an intermolecular r.m.s.d. containing distances between toxin residues Leu-1, Ser-6, Arg-10, Hyp-13, Hyp-15, Lys-18, and Arg-19 and amino acids Glu-348, Pro-349, Glu-354, Gly-371, and Val-376 of the turret and pore helix region of the four units of the channel.

RESULTS AND DISCUSSION
We have performed a mutant cycling analysis to identify which amino acids of M-RIIIK interact with specific residues in the vestibule of the ion channel pore of TSha1. For this purpose we have generated several mutants of TSha1 with single amino acids changes in the pore region of the channel, where M-RIIIK is likely to interact (4). Fig. 1A shows the alignment of the amino acid sequences of the pore region of TSha1 and Shaker together with the one of KcsA, for which a high resolution structure is available (14). Several mutations within this area have been constructed and functionally assayed using the Xenopus expression system and two-electrode voltage-clamp experiments. The addition of 1 M of M-RIIIK leads to different reductions of the measured currents at a test potential of 0 mV for wild type and mutated ion channels (Fig.  1B), demonstrating that the affinity of the peptide for the mutated channels is reduced. For M375K channels, hardly any block can be observed at this toxin concentration. This indicates a very low affinity of the toxin for the mutated channel, which was estimated to be about 10 M (k D Ϸ 76 nM for the wild type channel). The changes in the affinity for the different channel mutants are summarized in Table I and confirm that M-RIIIK indeed interacts with the pore region of the channel. The relative changes induced by the different mutations are summarized in Fig. 2. The mutations E348S, P349K, S351K, M375L, and V376T led to a 2-fold reduction in the affinity of M-RIIIK. Interestingly the mutation E348K did not have a significant effect despite the drastic change in the electrostatic properties of the side chain. The mutations E354K and E354Q resulted in a decrease of the affinity of 6-and 4-fold, respectively. The mutation M375I resulted in a much stronger affin-  ity change than M375L (12-versus 2-fold), whereas the most dramatic effect was observed for M375K (Ͼ100-fold).
To investigate which amino acids of the TSha1 channel pore interact with specific amino acids of M-RIIIK, we measured the affinity of several toxin mutants to different ion channel mutants. In this analysis we concentrated on the amino acid residues of the toxin that had been shown to be functionally relevant (Leu-1, Arg-10; Lys-18, Arg-19) (3). The IC 50 values for the different combinations of channel and toxin mutations are summarized in Table II, whereas Fig. 3 shows the corresponding changes in binding energy (⌬⌬G) values. Assuming that a ⌬⌬G value of 2 kJ/mol indicates that the distance between the two mutated residues is less than 6 Å (24), the data of Fig. 3 indicate a quite complex network of interactions between the toxin and amino acids from different domains of the ion channel target.
In our previous work (3) we proposed a new pharmacophore model for the M-RIIIK-TSha1 complex based on the fact that the toxin lacks the dyad motif composed of a lysine and a hydrophobic amino acid residue and on the observation that all the functionally relevant residues are clustered on one surface of the peptide delimited by a ring of positive charges. In our model the ring of positive charges is used as an anchor to residues of the K ϩ channel loops, and M-RIIIK blocks the channel by covering the pore as a lid. Additionally, no positively charged side chain penetrates the channel pore. The new pharmacophore is fully supported by the mutant cycling analysis data. Arg-10, Lys-18, and Arg-19 have the largest interactions with residues Glu-348, Pro-349, Ser-351, and Glu-354 of the turret and pore helix regions of the K ϩ channel, thus excluding that any of the positively charged side chains penetrates the channel pore. Instead, the positively charged residues interact with the loops of the four channel units, anchoring the peptide to the channel.
The mutant cycling data have been used as intermolecular distance restraints in docking calculations with the aim of obtaining a model for the complex of M-RIIIK with the TSha1 channel. The structure of M-conotoxin RIIIK, determined in aqueous solution by NMR, was docked to the model structure of the TSha1 channel using the distance restraints derived from the mutant cycle analysis, as described under "Materials and Methods." According to what observed by Eriksson and Roux (7) for the Ag2Tx-Shaker complex, we find that one or two distance restraints are insufficient to achieve a selection among the different orientations of the toxin with respect to the channel pore. Therefore, we chose to use all the available distance restraints in the docking calculations. The 875 generated models can be subdivided into 4 clusters, which have been numbered from I to IV according to increasing total energy (Fig. 4). Details on the criterion used for clustering are given under "Materials and Methods." 90% of the generated models belong to cluster I, which also contains the structure with the lowest energy; 6% of the models belong to cluster II, 3% to cluster III, and 1% to cluster IV. No intermolecular distance restraint violations are found in clusters I and II, whereas all structures of clusters III and IV violate the distance restraints Glu-354 -Arg-19 and Met-375-Leu-1, respectively. The backbone conformation of the toxin is well defined in all four clusters (toxin backbone r.m.s.d. are 1, 1.4, 1.1, and 1 Å in clusters I, II, III, and IV, respectively). The conformation of the toxin backbone is quite close to the one determined by Cluster II-The positively charged side chain of Arg-10 contacts the backbone carbonyls of Pro-349, Glu-350, and Ser-351 and the side chains of Glu-348 and Glu-354 of unit A, similarly to cluster I. Lys-18 interacts with the backbone carbonyls of Pro-349, Glu-350, and Ser-351 and with the side chains of Glu-348, Ser-351, and Glu-354 of unit C; Arg-19 contacts the side-chain carbonyls Glu-354 of the same unit C. As for cluster I, the electrostatic interactions of the side chains of Arg-10, Lys-18, and Arg-19 with loop residues of unit A and C stretch the peptide over the channel pore. In cluster II, Leu-1 is situated close to the loop residues of unit D, unlike in cluster I, where Leu-1 contacts the extracellular loop of unit B (Fig. 4). The positively charged N terminal group contacts the side chains of Gln-352 and Glu-354, whereas the Leu side chain has hydrophobic contacts with Pro-349. The distance between the side chain of Leu-1 and the side chains Met-375 and Val-376 in this model is higher than in model I.

Overall Orientation of the Toxin with Respect to the Channel
Given its small size, M-RIIIK cannot interact with the extracellular loops of all four units. In all four clusters the toxin contacts three of four units of the K ϩ channel. In model I, units A, B, and C bind the toxin, whereas in models II, III, and IV units A, C, and D are involved in contacts with the peptide (Fig.  4). According to the NMR studies (3), the toxin has a discoid shape; in model I the flat surface of the disc is out of the plane defined by the four channel loops, whereas in models II, III, and IV the toxin is flattened on the channel like a pancake. Although the toxin contacts only three of the four units, the accessibility for a potassium ion to the selectivity filter of the channel is reduced by more than 85% in all models.

Does One of the Four Models Better Fit the Mutational Data?
To discriminate between the four models, we analyzed the location of the functionally relevant residues of the toxin with respect to the channel pore (Fig. 6). In our previously published mutational analysis (3), we identified Leu-1, Arg-10, Lys-18, and Arg-19 as being fundamental for binding to the K ϩ channel (red in Fig. 6); Ser-3, Asn-8, Leu-11, Hyp-13, Val-14, Hyp-15, and Asn-20 resulted to be of medium importance (yellow in Fig.  6), whereas mutations of Hyp-2, Ser-6, Leu-7, Leu-9, Hyp-21, and Thr-24 to alanine did not significantly reduce the activity of the toxin (gray in Fig. 6). Model I is in very good accord with the mutational data (Fig. 6A). All the functionally relevant residues with the exception of Asn-8 face the channel pore or are in contact with the loops, whereas no residue that resulted in being unessential for function strongly interacts with the channel. The only gray residue facing the channel pore is Ala-16, for which no mutational data are available. Model II fits the mutational data worse than model I. Three residues of medium importance for binding (Ser-3, Asn-8, and Leu-11) are not in contact with the channel, whereas the C terminus of the toxin, whose side chain was found to be irrelevant for binding, contacts the loop of unit C (Fig. 6B). In model III the residues 21-24 at the C terminus of the toxin are involved in several contacts with units C and D, whereas the stretch 11-15, which contains several residues of medium importance for function, points toward solution (Fig. 6C). In model IV, the C terminus of the toxin deeply penetrates the channel, and a great portion of the residues that are unimportant for binding interacts with the protein (Fig. 6D).
The fact that a whole stretch of functionally relevant toxin residues (residues 11-15) is not in contact with the channel in model III, and IV cannot be taken as the sole argument to exclude these models, as mutations at this site might destabilize the overall structure of the toxin. However, it is rather improbable that amino acids profoundly interacting with the protein (stretch 21-24) can be mutated to alanine without influencing affinity. All in all the mutational analysis data are severely contradicted by models III and IV, allowing us to exclude them as valid representations of the complex between M-RIIIK and the TSha1 channel. Model II is retained at this stage, as it does not contradict the mutational data as extensively as model III and IV.

Interaction of M-RIIIK with the TSha1 Pore
To further discriminate between cluster I and II, we analyzed which part of the toxin is in close proximity to the selectivity filter. For different toxins that interact with voltagegated K ϩ channels there is evidence that a positively charged side chain occludes the channel pore (6 -8). This positively charged amino acid is part of the functional dyad together with an amino acid containing a hydrophobic side chain, usually a tyrosine. In our previous work we demonstrated that M-conotoxin RIIIK does not possess a functional dyad. Moreover, as a result of the mutant cycling analysis, all positively charged amino acids interact with the turret or pore helix regions of the channel and, therefore, cannot penetrate the channel pore. Which toxin residues then face the channel pore in our models? In model I the Hyp-15 is very close to the selectivity filter of units C and B, and the hydroxyl oxygen at the C␥ is at a hydrogen bond distance (2.8 Å) from the carbonyl of the Gly-373 of the selectivity filter of unit B (Fig. 7A). In model II the Hyp-13 is in proximity of the selectivity filters of unit A and D but makes no tight direct contact with them (Fig. 7B). To distinguish between the two models, mutant cycle data involving the residues of the GYG selectivity filter would be required. This is unfortunately not feasible as any mutation of residues in the selectivity filter destroys the channel activity. Nevertheless, model I and model II can be distinguished by considering the distance of Hyp-13 or Hyp-15 from the amino acid Val-376 of the TSha1 channel. In model I Hyp-15 is expected to be closer to Val-376 (of unit C) than Hyp-13, while in model II both Hyp-13 and Hyp-15 are at a similar distance from Val-376 of units D and C, respectively. The mutant cycle analysis performed using TSha1 V376T resulted in a ⌬⌬G of 0.6 kJ mol Ϫ1 for O13A and of 2.51 kJ mol Ϫ1 for O15A, which reflects the geometry of model I much better than that of model II. To obtain further evidence for the interaction of Hyp-15 with the amino acids of the selectivity filter, we assayed the affinity of the O15R and O15K mutants, following the idea that a positive charge should be well tolerated but the bulky arginine should not be easily accommodated in proximity of the tight channel pore. The affinity of the mutant O15K for the wild type TSha1 channel was found to be 430 Ϯ 20 nM (n ϭ 4) against a value of 930 Ϯ 160 nM (n ϭ 3) for the mutant O15A. This indicates that a positive charge in position 15 is preferred to the absence of any polar group, as would be expected if the side chain of Hyp-15 interacts with the carbonyls of the selectivity filter. On the other hand O15R binds the channel with a k D of 2940 Ϯ 540 nM (n ϭ 4), suggesting that Hyp-15 is situated close to the channel pore, where there is no space to accommodate the bulky arginine side chain. Although these data do not prove directly the interaction of Hyp-15 with residues of the selectivity filter, they strongly support it.

The Best Model
Model I provides a novel pharmacophore for the interaction of toxin peptides with K ϩ channels that best explains the mutational and mutant cycle data for the M-RIIIK-TSha1 complex. The peptide is anchored to the extracellular loops of three of the four channel units via a basic ring provided by the side chain of Arg-10, Lys-18, Arg-19, and the protonated N terminus of Leu-1, and no positively charged residue occludes the channel pore.
Unlike several other peptides that target voltage-activated K ϩ channels, including charybdotoxin from a scorpion, BgK from a sea anemone, and dendrotoxin from snakes, M-conotoxin RIIIK does not contain a functional dyad of a hydrophobic and a positively charged side chain, which was proposed to be the minimal functional core for K ϩ channel-blocking peptides (9 -12). The dyad model has been recently challenged by Mouhat et al. (13), who demonstrated by mutational analysis and docking calculations that a functional dyad is not strictly required for binding of the scorpion toxin Pi1 to the voltage-gated Kv1.2 potassium channel. Instead, a basic ring of four positively charged side chains was proposed to be a fundamental recognition element by the residue Asp-355 of each of the four Kv1.  (5), which like Kv1.2 contain an uncharged amino acid at position 359, indicates that channel amino acids other than Pro-359 play a comparably important role in toxin recognition. For the toxin it is reasonable to hypothesize that the spatial distribution of the positive charges of the basic ring exerts a fundamental role in channel recognition and provides a solid framework for the high selectivity of the K ϩ channelblocking toxins among the diverse targets.
As shown in the NMR-derived solution structure of the peptide (3), the basic ring defined by Leu-1, Arg-10, Lys-18, and Arg-19 separates the peptide in two faces; all functionally relevant residues, identified by systematic alanine mutations, cluster on one face, which is in contact with the channel in model I.
The mutant cycle analysis excludes that one of the positively charged side chains of M-RIIIK penetrates the channel pore, as they all interact with amino acids of the extracellular loop and pore helix. In model I the C␥-hydroxyl of Hyp-15 tightly interacts with the carbonyl of Gly-373 of the unit C selectivity filter. It has been proposed that the carbonyls of the selectivity filter exert an essential role in the coordination of K ϩ ions at the entrance of the channel pore (14). Dehydration of the ions must occur at the selectivity filter, as a hydrated K ϩ is too large to enter the channel pore. By acting as electron donors, the carbonyls of the selectivity filter replace the water molecules usually associated with K ϩ ions in solution, thus compensating for the large energetic penalty of the dehydration process. A H-bond between the C␥-hydroxyl of Hyp-15 would block one of the selectivity filter carbonyls, seriously disturbing the energetic of the K ϩ dehydration process.
A thorough analysis of the binding mode of various scorpion toxins to diverse K ϩ channels (28, 29) revealed three possible binding modes that involve different faces of structurally sim- ilar toxins; they are the internal binding mode, where residues of the turret region, pore helix, and selectivity filter of the channel are involved, the intermediate mode, involving residues of the turret region, and the external mode, which involves residues far away from the selectivity filter. In the intermediate and external binding mode no functional dyad is observed on the peptide site. This analysis, triggered by the observation that the scorpion toxin BmTx3 can block both A-type K ϩ and HERG currents (30) using different functional faces, confirms the existence of different binding modes of peptidic toxins to K ϩ channels, some of which do not involve a direct obstruction of the pore by a positively charged side chain. Based on the solution structure and the mutational analysis of M-RIIIK, we proposed that this toxin could bind the TSha1 channel in the intermediate mode. However, the mutant cycle data and the docking model I prove the interaction of toxin residues with amino acids in the turret and pore helix regions and to some extent also with the selectivity filter. In the classification of Rodriguez de la Vega et al. (28), the position of the toxin in the M-RIIIK-TSha1 channel complex resembles the internal binding mode best. However, the side chain of Hyp-15 does not completely occlude the channel pore as the lysine of the functional dyad but rather disturbs the charge distribution at the selectivity filter by involving one of the carbonyls in a H-bond.
Despite the entirely different pharmacological specificity and many differences in the amino acid sequence, M-conotoxin RIIIK shows striking structural similarities to -conotoxin GIIIA (31)(32)(33), which specifically blocks NaV1.4 Na ϩ channels. M-conotoxin RIIIK and -conotoxin GIIIA belong to the same phylogenetic family and share a common pattern of cysteine bridges. Interestingly, Hyp-15 of M-conotoxin RIIIK is at the homologous position of Arg-13, the Na ϩ channel pore occluding residue in the peptide sequence of -conotoxin GIIIA. This fact together with the high structural similarity of the two toxins suggests that M-conotoxin RIIIK and -conotoxin GIIIA might block K ϩ and Na ϩ channels, respectively, with a similar geometry. However, because the O15R mutant of M-RIIIK is inactive on Na v 1.4 sodium channels (data not shown), additional parameters must be involved in determining the selec-tivity of the two peptides for Na ϩ or K ϩ channels. Similarly to M-RIIIK, a ring of positively charged side chains can be identified on -GIIIA as well (3). Most likely the overall shape of the two peptides and the charge distribution of the basic ring are key determinants for the Na ϩ or K ϩ channel selectivity of -conotoxin GIIIA and M-conotoxin RIIIK.

Conclusions
We have presented a combined approach based on thermodynamic mutant cycle data and docking calculations to derive the structure of the complex of M-conotoxin RIIIK with the K ϩ channel TSha1. This peptide shows a novel binding mode to the K ϩ channel that is not centered around a functional dyad. Instead, a ring of positive charges anchors the peptide to the turret and pore helix region of the channel; the electrostatic complementarity between this ring and the channel side chains is likely to be responsible for the selectivity of M-RIIIK toward the TSha1 channel, as it can be inferred from the important role of Glu-354 of different units in contacting the basic ring. This amino acid is not present in lower affinity targets of M-RIIIK, such as the Shaker channel. The intermolecular interactions involving the channel loops and the toxin positively charged side chains can be of general importance in determining the selectivity of peptidic toxins for ion channels.
The role of the positive charged side chain that in many peptide-ion channel complexes occludes the channel pore is partially exerted by Hyp-15, whose side chain, however, is too short to penetrate the pore. The C␥-OH of Hyp-15 contacts the carbonyls of the selectivity filter of one to two units, perturbing the electrostatic forces that regulate the dehydration of potassium ions at the pore entrance.
The binding mode of M-RIIIK to the TSha1 channel closely resembles that of -conotoxin GIIIA to the Na v 1.4 Na ϩ channel. The three-dimensional structure of the two peptides is highly similar in the C-terminal part despite the poor superposition of the amino acid sequence. Hyp-15 of M-RIIIK occupies the homologous position of Arg-13 of -GIIIA, which is known to occlude the Na ϩ channel pore. Both peptides contain a basic ring, although the form and the charge density at the surface differs (3). This observation demonstrates that the two conotoxins have developed a similar scaffold and binding geometry to their respective targets, whereas the pharmacological selectivity is determined by the exact charge distribution and three-dimensional shape.