A Variable Residue in the Pore of Kv1 Channels Is Critical for the High Affinity of Blockers from Sea Anemones and Scorpions*

Animal toxins are associated with well defined selectivity profiles; however the molecular basis for this property is not understood. To address this issue we refined our previous three-dimensional models of the complex between the sea anemone toxin BgK and the S5-S6 region of Kv1.1 (Gilquin, B., Racape, J., Wrisch, A., Visan, V., Lecoq, A., Grissmer, S., Me´nez, A., and Gasparini, S. (2002) J. Biol. Chem. 277, 37406–37413) using a docking procedure that scores and ranks the structures by comparing experimental and back-calculated values of coupling free energies (cid:1)(cid:1) G int obtained from double- mutant cycles. These models further highlight the interaction between residue 379 of Kv1.1 and the conserved dyad tyrosine residue of BgK. Because the nature of the residue at position 379 varies from one channel subtype to another, we explored how these natural mutations influence the sensitivity of Kv1 channel subtypes to

Molecular recognition and specific association of protein ligands and protein targets are central to most biological processes. Understanding the molecular basis of these interactions is critical for engineering novel protein-protein interactions. In particular, understanding how protein ligands bind with high affinity to only a subset of closely related receptors may help to design ligands with novel selectivities.
A number of studies have been carried out to identify the sites used by protein ligands to bind to several related receptors. These sites were shown to be composed of a core formed by conserved hot spot residues together with target-specific residues (1)(2)(3)(4)(5). However, how these sites accommodate the different receptors subtypes is still poorly understood. In many cases the molecular determinants responsible for protein ligand discrimination remain to be identified.
Toxins from sea anemones and scorpions that block currents through Kv1 voltage-gated potassium channels are particularly appropriate to investigate the molecular basis of selectivity of protein-protein interactions since each toxin binds to only a subset of Kv1 channel subtypes (6). We have previously studied in detail BgK, a 37-amino acid peptide isolated from the sea anemone Bunodosoma granulifera (7), which binds with similar high affinity to Kv1.1, Kv1.2, and Kv1.6 (8) but not to Kv1.4 and Kv1.5 channels. The sites used by BgK to bind to its different targets were identified by alanine scanning (9). These sites share three critical residues (Lys-25, Tyr-26, and Ser-23), conserved in all Kv1-blocking toxins from sea anemones, which have also been shown to be functional in ShK, another sea anemone toxin (for review, see Ref. 10). Thus, these three residues form the functional core of Kv1-blocking sea anemone toxins. Similarly, a comparison of the functional sites of Kv1-blocking scorpion toxins ␣KTx1-3 (11) suggested a functional core formed by four residues (for review, see Ref. 10). Furthermore, comparison of the functional cores of Kv1-blocking sea anemone and ␣KTx1-3 scorpion toxins revealed that they commonly contain a pair of residues formed by a lysine and a hydrophobic residue (for review, see Ref. 10). This functional dyad (12), which is the smaller common functional denominator of a variety of Kv1-blocking toxins (1), likely reflects a common binding feature of these toxins (10). However, as suggested by a study with a cone snail toxin (13), this binding mode may not be the only one adopted by Kv1-blocking toxins.
Recently, structural models of the complex BgK⅐S5-S6 region of Kv1.1, based on distance restraints derived from doublemutant cycles (9), revealed that residues from the BgK functional core interact with both conserved and non-conserved residues of Kv1 channels and suggested a role of the latter residues in the selectivity of the toxin for a subset of Kv1 subtypes. In particular, these models emphasized the importance of Kv1.1 residue 379, a variable position in Kv1 channels that is critical for binding of external tetraethylammonium ion (14 -17).
In this study we have refined our previous models of the complex BgK⅐S5-S6 region of Kv1.1 using a previously developed docking procedure (18) that screens the structures by comparing experimental (9) and back-calculated values of coupling free energies ⌬⌬G int from double-mutant cycles. These models provide a detailed description of the interactions involving the residues of the BgK functional core. Interestingly, one of these interactions, involving the carbonyl of a glycine residue from the channel selectivity filter, appears to be common to the different binding modes used by toxins whether or not they contain a functional dyad. Furthermore, our model strengthens the putative importance of Kv1.1 residue 379. We have investigated the importance of this residue using binding and electrophysiology experiments on different Kv1 channels mutated at position 379. Our results show that mutations at position 379 are sufficient to abolish or enhance sensitivity to toxins, indicating that this single position critically controls the affin-* The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. ity of BgK and other sea anemone and scorpion toxins for Kv1 channel subtypes.
The channel was positioned such that the cavity was centered at z ϭ 0 and the pore was aligned with the z axis, the extracellular side on the positive side. The atoms of all channel residues for which z Ͻ 10 were kept fixed. For the other channel residues the backbone atoms were restrained relative to the initial model using a harmonic potential. The strength of the backbone restraints was progressively decreased from 100 to 6 kcal/mol. For the turret residues (residues 345-359), the strength was reduced and decreased from 50 to 0.5 kcal/mol. An energy restraint allowing complete rotation and translation was applied to the toxin backbone (100 kcal/mol) and to the C␤ (10 kcal/mol).
The docking procedure started from a random position and orientation of the toxin. In a first step hydrogen atoms were not included, and electrostatic interactions were ignored. The best structures in terms of van der Waals interaction energy were refined. In a second step hydrogen atoms were introduced, and the system was annealed from 800 to 400 K in 10,000 steps. During these two steps distance restraint force constants were set to 20, 10, 2 kcal/mol for the strong, medium, and weak constraints, respectively. In a final step the structures were refined by slow cooling from 800 to 300 K in 8000 steps during which the distance restraint force constants were reduced to 5, 2, and 0.5 kcal for the strong, medium, and weak constraint, respectively. The equations of motion were integrated using a time step of 2 fs, and the length of all the bonds involving hydrogen atoms were kept rigidly fixed using SHAKE (20). The structures with high levels of energy restraint (Ͼ40 kcal/mol) were rejected. The best structures in terms of van der Waals interaction energy were selected. For these structures the ⌬⌬G int from double-mutant cycles were back-calculated using a continuous implicit solvent model based on the Poisson-Boltzmann equation (18). This equation was solved numerically using the PBEQ module (21) implemented in the program CHARMM (22). A set of atomic Born radii, calibrated and optimized to reproduce the electrostatic free energy of the 20 amino acids in molecular dynamics simulations with explicit water molecules, was used (23). The nonpolar contribution to the binding free energy was empirically written as a fraction of the van der Waals interactions ⌬E vdW upon formation of the complex, with ϭ 0.17 (18). The dielectric constant of the protein was set to ⑀ prot ϭ 12 (18). The structures that gave the best agreement between back-calculated and experimental values (9) were selected. All the calculations were performed using the CHARMM program version c28a3 (22).
Heterologous Expression of Kv1 Channels in Mammalian Cells-TsA-201 cells were maintained in 10-cm-diameter tissue culture dishes as previously described (8). When near confluency, medium was replaced by antibiotic-free medium, and cells were transfected using 25-30 g of DNA and 60 l of Lipofectamine 2000 (Invitrogen) according to the manufacturer's instructions. Cells were collected 24 h after transfection, and membranes were prepared as previously described (8).
Western Blots-Proteins from membrane preparations were separated on 12.5% SDS-PAGE and transferred to a nitrocellulose membrane (Optitran, Schleicher & Schuell) using a semidry transfer apparatus and Tris-glycine-SDS-methanol buffer. Membranes were saturated overnight at 4°C with TBS buffer (20 mM Tris-HCl, pH 7.5, 500 mM NaCl) containing 3% (w/v) bovine serum albumin (BSA), washed once with TBS-Tween (0.05% (v/v) Tween 20), and incubated 1 h at room temperature with a rabbit antibody specific for Kv1 subtypes (Sigma) in TBS-Tween buffer containing 0.1% (w/v) BSA. After 3 washes, the membrane was incubated for 1 h at room temperature with a peroxidase-conjugated goat anti-rabbit IgG (Jackson ImmunoResearch), and after three wash steps, the peroxidase reaction was initiated by the addition of 3,3Ј diaminobenzidine (Sigma) in 100 mM Tris-HCl, pH 7.4, 0.2% (v/v) H 2 O 2 to visualize the hybridized probes. Pre-stained molecular weight markers from Biolabs were used.
Binding Assays-All binding assays and data analyses were carried out as previously described (8). For measuring dissociation rate constants (k off ), dissociation was initiated by adding a 4000-fold molar excess of BgK. Aliquots of the binding reaction were diluted at different times into ice-cold wash buffer and filtered as previously described (8).
Kv1.4 and Kv1.5 currents were recorded at room temperature using the IonWorks HT (Molecular Devices) multichannel whole-cell voltage clamp instrument (24). Hole resistances in the planar 384-well electrode array were ϳ3 megaohms. Electrical access to the cytoplasm was achieved by perforation in 0.13 mg/ml amphotericin B for 4 min. The test pulse consisted of a 150-ms step from a holding potential of Ϫ80 mV to ϩ50 mV for Kv1.4 currents and of a 100-ms step from a holding potential of Ϫ80 mV to ϩ50 mV followed by 50 ms at Ϫ40 mV for Kv1.5 currents. In both cases the pulse was performed before and after 5 min of incubation with BgK, during which the cells were not voltageclamped, and leak conductances were measured during a 160-ms step from Ϫ80 mV to Ϫ70 mV preceding the test pulse. Only cells with membrane resistances of Ͼ70 megaohms were included in the analysis. Data were acquired at 10 kHz. For mutated Kv1.4 channels, the am-plitude of the peak currents in the presence of BgK was normalized to the peak current in control plotted against peptide concentration and fit to the Hill equation of the form is the peptide concentration, and IC 50 is the peptide concentration resulting in 50% inhibition.
In summary, agreement between back-calculated and experimental ⌬⌬G int values was correlated with two structural characteristics in the complexes. First, BgK(Phe-6) is in close contact to Kv1.1(Tyr-379), and second, Kv1  Fig. 3B.
Position 379 in Kv1 Channels Is Critical for BgK Binding-The refined models of the complex BgK⅐S5-S6 region of Kv1.1 strengthen the importance of Kv1.1 residue 379, a variable residue in Kv1 channels (Fig. 4) that was previously suggested to be important for binding of BgK to a subset of Kv1 channel subtypes (8,9). Indeed, we showed that replacing this single residue in Kv1.3 by the equivalent residue in Kv1.1 (mutant Kv1.3(H399Y)) was sufficient for enhancing BgK affinity by 33-fold, as assessed by competition binding experiments with 125 I-HgTX 1 (A19Y/ Y37F) (8). Furthermore, although no specific binding could be obtained with membranes from tsA-201 cells expressing Kv1.3, the radiolabeled analog of BgK, 125 I-BgK(W5Y/Y26F), could bind to Kv1.3(H399Y) with a K d of 40 Ϯ 3 pM (Table II) (8).
Wild-type Kv1.5 and mutants Kv1.5(R487Y) and Kv1.5-(R487V) showed similar expression levels, as indicated by Western blots; a single band corresponding to a 80-kDa protein was revealed in each case (not shown). No specific binding could be observed with wild-type Kv1.5, whereas saturable and reversible binding of 125 I-BgK(W5Y/Y26F) was observed in membranes from cells expressing either Kv1.5(R487Y) or Kv1.5(R487V). These data indicate that replacement of Kv1.5 residue Arg-487 by tyrosine or valine increases the affinity for 125 I-BgK(W5Y/ Y26F). However, we were not able to measure these affinities accurately because of very high rates of ligand dissociation (Ͼ3 or 5 ϫ 10 Ϫ2 s Ϫ1 ), which prevented successful separation of free from bound 125 I-BgK(W5Y/Y26F).
We then constructed mutants of Kv1.6 in which the variable residue Tyr-429 ( Fig. 4) was replaced with either arginine or lysine. Kv1.6 was chosen because of its high affinity for 125 I-BgK(W5Y/Y26F) (Table II). Western blot analysis revealed that expression of wild-type and mutant channels was similar; in all cases a single band corresponding to a 70-kDa protein was detected (data not shown). Specific binding of 125 I-BgK(W5Y/ Y26F) could not be detected to either Kv1.6 mutant, indicating that replacement of Kv1.6 residue Tyr-429 by arginine or lysine decreases the affinity for 125 I-BgK(W5Y/Y26F).
Kv1 Channel Position 379 Is Critical for Binding of Other Toxins-Competition experiments were carried out to measure the ability of two scorpion toxins, ChTX (␣KTx1.1) and KTX (␣KTx3.1) (11), to inhibit 125 I-BgK(W5Y/Y26F) binding to different Kv1 channels bearing either a valine or a tyrosine residue at position 379 (Table III). We found that ChTX binds with 2 or 3 orders of magnitude higher affinity to channels possessing a valine residue than to channels possessing a tyrosine residue. Reciprocally, KTX binds with 2 orders of magnitude higher affinity to channels possessing a tyrosine residue than to channels possessing a valine residue. These results agree with previous studies carried out with mutated Kv1.3 channels (30,31). Furthermore, both toxins are able to bind with nM affinity to Kv1.4 channels bearing a single mutation (Kv1.4(K532V) for ChTX and Kv1.4(K532Y) for KTX) (Table III).
We also assessed the ability of the sea anemone toxin ShK to inhibit 125 I-BgK(W5Y/Y26F) binding to Kv1.4(K532V) and Kv1.4(K532Y) (Table III). Although this toxin does not block Kv1.4 channels (32), it inhibits 125 I-BgK(W5Y/Y26F) binding to Kv1.4(K532V) and Kv1.4(K532Y) (Table III). Furthermore, it binds to channels possessing a tyrosine residue with 60-fold higher affinity than to channels possessing a valine residue.   and Kv1.5 (right) currents were recorded in control and 5 min after application of BgK, as indicated. Kv1.4 currents were activated by a 150-ms step from a holding potential of Ϫ80 mV to ϩ50 mV, and Kv1.5 currents were activated by a 100-ms step from Ϫ80 mV to ϩ50 mV followed by 50 ms at Ϫ40 mV. For the dose-response curve, mean values were calculated from 3 to 13 determinations, except only one determination was made for 33 nM BgK applied to Kv1.4(K532V). Currents in BgK are plotted as % of control currents. The IC 50 values determined were 87 nM for Kv1.4(K532V) and 84 nM for Kv1.4(K532Y). wt, wild type.

DISCUSSION
One of the main goals of studying protein-protein interactions is to understand the molecular basis of selectivity or, in other words, how a protein ligand can display large differences in affinity for closely related receptors. Sea anemone and scorpion toxins ␣KTx1-3 that block Kv1 voltage-gated potassium channels offer a well documented model system to address this issue. Indeed, these toxins bind with high affinity to different subsets of Kv1 channels, and none of them bind to Kv1.4 and Kv1.5 channels (Refs. 6 and 32 and this study). It was previously shown that association of these toxins with their different targets is mediated by residues from a conserved functional core and by variable target-specific residues (for review, see Refs. 1 and 10). Furthermore, the presence in the functional cores of these toxins of a common denominator formed by two residues (the functional dyad) suggests that toxins sharing this feature have undergone convergent evolution, with the dyad acting as a common anchor in each complex (9, 12) (for review, see Ref. 10). However, although these toxins possess such a similar binding anchor, they display distinct selectivity toward the different Kv1 channels subtypes, and the molecular features responsible for these distinct profiles of selectivity are not yet understood.
In the current study we addressed the above issue by establishing refined models of the complex between the sea anemone toxin BgK and the S5-S6 region of Kv1.1. These models were generated with a procedure (18) that screens the structures by back-calculating the ⌬⌬G int from double-mutant cycles. This approach led us with well defined complex structures with a r.m.s.d. of 1.3 Å around the average BgK position. The average distance between BgK C␣ of mean structures from this procedure and the previous one (9) is 3.0 Å but is reduced to 1 Å for the functional core residues of BgK (Lys-25, Tyr-26, and Ser-23). The interactions involving these residues were determined in more detail with this procedure since, in addition to those previously determined for the functional dyad residues Lys-25 and Tyr-26 of BgK (9), it identified interactions between the third residue of the BgK functional core (Ser-23) and Kv1.1 residues Tyr-379 and Gly-376.
Therefore, the interactions involving the BgK functional core residues appear to be of two types. Some involve the mainchain atoms of the channel and, thus, may be conserved in each complex. They correspond to previously described interactions between the dyad lysine side chain and the carbonyl oxygen atoms of the selectivity filter (9) and to hydrogen bonds between the Ser-23 side chain and the carbonyl oxygen atoms of Gly-376. Interestingly, a recent study based on molecular docking (33) identified interactions between the conserved functional residue Ser-10 of scorpion toxins from subfamilies ␣KTx1 and ␣KTx3 (11) and the carbonyl of Kv1.3(Gly-396), equivalent to Kv1.1(Gly-376). Furthermore, a very recent model of the complex (34) between the cone snail M-conotoxin RIII-K that does not contain a functional dyad (13), and the TSha1 channel, a Kv1 channel from the rainbow trout, identified a similar interaction between the side chain of the hydroxyproline O15 from the toxin and the carbonyl of TSha1(Gly-373), equivalent to Kv1.1(Gly-376). Therefore, this interaction appears to be common to the binding of toxins whether or not they possess a functional dyad. The second interactions involving the binding core residues of BgK are established with the side chain of a variable residue of Kv1 channels (Tyr-379 in Kv1.1). This residue interacts with the aromatic residues Tyr-26 (from the dyad) and Phe-6, and the spatial organization of these aromatic rings is frequently conserved in proteins (35). The side chain of Tyr-379 also forms hydrogen bonds with the side chain of the BgK binding core residue Ser-23 and with Asn-19.
That the binding core residues of BgK interact with a nonconserved residue of the channel suggested a critical role of the position of this channel for BgK selectivity (9). We further investigated this idea by determining the affinity of BgK and 125 I-BgK(W5Y/Y26F), a radiolabeled analog of BgK (8), for different Kv1 channels mutated at that position using binding and electrophysiology experiments. Although our experiments did not formally rule out that the absence of specific binding with some mutated channels could be due to other reasons than low affinity of 125 I-BgK(W5Y/Y26F), such as lack of channel tetramerization, we believe that this is unlikely to be the case since most of these mutated channels (Kv1.1(Y379H), Kv1.6(Y429K), Kv1.4(K532Q)) have been previously shown to be functional (28,36,37). Position 379 of Kv1 channels appears to be a critical determinant for BgK binding since mutations at this position are sufficient to alter the affinity for BgK. Thus, we could generate Kv1.4 and Kv1.5 channels sensitive to 125 I-BgK(W5Y/Y26F) and BgK by solely introducing a valine or a tyrosine at this position. In the complex BgK⅐S5-S6 region of Kv1.1, Tyr-379 establishes both hydrophobic interactions with BgK residues Phe-6 and Tyr-26 and hydrogen bonds with residues Ser-23 and Asn-19. However, our results indicate that these hydrogen interactions are not critical since replacing Tyr-379 in Kv1.1 by phenylalanine does not reduce the affinity for BgK. Furthermore, we found that all the residues at position 379 that could confer high affinity to 125 I-BgK(W5Y/Y26F) and BgK (Tyr, Phe, Thr, Val, Cys) allow hydrophobic interactions. Therefore, in agreement with our previous suggestion (9), the hydrophobic component of the interactions involving residues at position 379 is likely to be critical for BgK to bind Kv1 channels with high affinity.
Previous studies have shown that position 379 is also important for scorpion toxins since agitoxin does not bind to Shaker channels containing lysine or glutamine residues at this position (449, using Shaker numbering) (38), and ChTX, KTX, and margatoxin (MgTX) do not bind to Kv1.3 channels containing arginine residue at this position (30). In our study we also demonstrated that this position is an essential component for high affinity binding of the scorpion toxins ChTX and KTX and the sea anemone peptide ShK. Altogether, these results indicate that position 379 is critical for the selectivity of sea anemone and scorpion toxins and differentiates between two groups of Kv1 channels. First, the channels possessing a lysine, arginine, or glutamine residue at position 379, such as Kv1.4 and Kv1.5, are not recognized by these toxins. Second, channels possessing a valine, tyrosine, phenylalanine, threonine, histidine, or cysteine residue at position 379 bind some of these toxins. However, a number of these toxins can bind to only a subset of channels from this second group, and again, it seems that this selectivity is dictated by the nature of the residue at position 379. Thus, the low affinity of BgK for Kv1.3 and of ShK for Kv1.2 channels (39) seems to reflect the presence in these channels of a histidine and a valine residue, respectively, whereas the low affinity of ChTX for Kv1.1 channels (8) seems to reflect the presence of a tyrosine residue. Moreover, it was recently shown that this position also contributes significantly to the selectivity of the scorpion toxin maurotoxin (40). Therefore, each toxin seems to "sense" the nature of the side chain of residue 379 in a unique way. Other proteins also bind to their different targets using highly conserved residues that form a "functional core" and target-specific variable residues (2)(3)(4)(5). This binding mode, called "two components binding mode" (1) or "dual recognition" (2, 3), has been described in large detail for bacterial immunity proteins, which bind with high affinity to cognate colicins and with low affinity to non-cognate colicins (2,3). It was shown that two of the three conserved hot spot residues of the functional core of immunity proteins interact with a variable residue of colicins and that the high affinity of a immunity protein for its cognate colicin is conferred by the capacity of some of its variable residues to interact with this colicin variable residue (2,3). Data reported in this paper also suggest a similar mechanism for toxins; the high affinity of a toxin for a particular Kv1 subtype could be conferred by the capacity of target-specific variable residues to interact with the variable channel residue at position 379. Two main arguments support this hypothesis. First, BgK(Phe-6), a target-specific residue (9), is in close contact to Kv1.1(Tyr-379). Second, ShK, a sea anemone toxin homolog to BgK, can bind to Kv1.3 with high affinity, whereas BgK cannot, and it was shown that ShK(Arg-11), a residue absent in BgK, interacts with Kv1.3(His-404), the equivalent residue in the channel to Kv1.1(Tyr-379) (41).
Obviously, residue 379 is not the only determinant for toxin selectivity, and other Kv1 channel regions may have some influence since a toxin does not necessarily bind with the same affinity to channels possessing the same residue in position 379. For instance, the scorpion toxin hongotoxin 1 binds to Kv1.6 with an affinity 2 orders of magnitude lower than to Kv1.1 (42). For BgK, there seems to be a correlation between the charge of the turret and the kinetics of dissociation of 125 I-BgK(W5Y/Y26F), which are slower when the turret contains more negatively charged residues. This was inferred from comparison of 125 I-BgK(W5Y/Y26F) dissociation rates for the channels Kv1.1, Kv1.3(H399Y), Kv1.4(K532Y), and Kv1.6, which all possess a tyrosine residue in position 379 and whose S5-S6 regions differ by the position 381, which is uncharged in all cases, and by their turret sequences and charges. Also, it was shown that the turret region influences ChTX binding since a single mutation in the Shaker channel turret increases its affinity by 3 orders of magnitude (43).
In conclusion, we demonstrated that position 379 is critical for the selectivity of BgK and other toxins possessing a functional dyad. This suggests that this residue could be a major actor in the "two binding components" or "dual recognition" of Kv1 channels by these toxins by interacting with both conserved residues from the functional core of the toxin and targetspecific variable residues.