Structure-guided Transformation of Charybdotoxin Yields an Analog That Selectively Targets Ca2+-activated over Voltage-gated K+ Channels*

We have used a structure-based design strategy to transform the polypeptide toxin charybdotoxin, which blocks several voltage-gated and Ca2+-activated K+channels, into a selective inhibitor. As a model system, we chose two channels in T-lymphocytes, the voltage-gated channel Kv1.3and the Ca2+-activated channel IKCa1. Homology models of both channels were generated based on the crystal structure of the bacterial channel KcsA. Initial docking of charybdotoxin was undertaken with both models, and the accuracy of these docking configurations was tested by mutant cycle analyses, establishing that charybdotoxin has a similar docking configuration in the external vestibules of IKCa1 and Kv1.3. Comparison of the refined models revealed a unique cluster of negatively charged residues in the turret of Kv1.3, not present in IKCa1. To exploit this difference, three novel charybdotoxin analogs were designed by introducing negatively charged residues in place of charybdotoxin Lys32, which lies in close proximity to this cluster. These analogs block IKCa1with ∼20-fold higher affinity than Kv1.3. The other charybdotoxin-sensitive Kv channels, Kv1.2 andKv1.6, contain the negative cluster and are predictably insensitive to the charybdotoxin position 32 analogs, whereas the maxi-KCa channel, hSlo, lacking the cluster, is sensitive to the analogs. This provides strong evidence for topological similarity of the external vestibules of diverse K+channels and demonstrates the feasibility of using structure-based strategies to design selective inhibitors for mammalian K+channels. The availability of potent and selective inhibitors ofIKCa1 will help to elucidate the role of this channel in T-lymphocytes during the immune response as well as in erythrocytes and colonic epithelia.

Potassium channels are a diverse superfamily of ϳ80 integral membrane proteins that play crucial roles in many different physiological processes and are widely recognized as ther-apeutic targets. Venoms from spiders, scorpions, snakes, bees, and marine extracts have yielded polypeptide inhibitors of mammalian K ϩ channels, many of which bind with high affinity to a vestibule at the external entrance of the channel pore. Some of these polypeptide toxins have been used as molecular calipers to estimate the dimensions of K ϩ channel vestibules (1)(2)(3)(4)(5)(6)(7). Recently, the structure of the bacterial K ϩ channel KcsA from Streptomyces lividans has been determined by x-ray crystallography (8). The turret region and the external pore of this channel correspond to the external toxin-binding vestibule in eukaryotic K ϩ channels, and the crystallographic dimensions of the KcsA channel vestibule are remarkably similar to those estimated by toxin-mapping methods for eukaryotic K ϩ channels (9). The convergence of these two approaches raises the possibility of exploiting structure-based strategies to design specific inhibitors that target pharmacologically relevant K ϩ channel targets.
To test the feasibility of this approach, we have used charybdotoxin (ChTX), 1 a polypeptide that potently blocks the voltage-gated channel Kv1.3 and the Ca 2ϩ -activated channel IKCa1, both present in human T-lymphocytes (10,11) to design an analog that selectively targets IKCa1. These two channels have been chosen as our model system since they are widely regarded as therapeutic targets. Both channels regulate the membrane potential of resting and activated T-cells and modulate the calcium signaling response that is essential for their activation (12). Inhibitors of these channels block the activation of human T-lymphocytes (13)(14)(15). Several potent and selective peptide and non-peptide inhibitors are available for Kv1. 3. However, there is a dearth of selective blockers of the Ca 2ϩ -activated IKCa1 channel, and the most selective inhibitor of this channel, clotrimazole, also inhibits cytochrome P450-dependent enzymes (16 -19). The IKCa1 gene also encodes the "Gardos" channel in erythrocytes and is thought to encode the IK Ca channel in colonic epithelial cells, platelets, and pancreatic islets (20 -23). IKCa1 inhibitors are currently being evaluated for prevention of chloride and water loss in diarrhea and for the treatment of erythrocyte dehydration in sickle cell disease (19,24,25). Highly specific blockers of the IKCa1 channel may therefore have clinical use in both these ailments as well as a potential use as immunosuppressants.
In this study, we have constructed homology models of IKCa1 and Kv1.3 based on the crystal structure of the KcsA channel and performed preliminary docking of ChTX for heuristic purposes. The accuracy of these docking configurations was tested by mutant cycle analysis that measures the strength of coupling between interactive pairs of toxin and channel residues (26). Using this approach, we have determined the ChTX docking configuration in both channels and identified a structural feature unique to the ChTX-IKCa1 interaction surface. This paper describes the guided design and electrophysiological characterization of three novel ChTX analogs that specifically target this unique IKCa1 motif.  (11) and the PCR-generated (28) IKCa1-Asn 239 and IKCa1-Lys 239 mutants (7) have been described. The hSlo (BK Ca ) expression construct was a kind gift from Dr. L. Toro (UCLA). Endogenously expressed IK Ca and Kv1.3 currents were studied in phytohemagglutinin-activated human T-lymphocytes, and endogenous SK Ca currents were studied in the human Jurkat T-cell line (29,30). Fetal calf serum, L-glutamine, penicillin, and streptomycin were obtained from Life Technologies, Inc.

Reagents
Polypeptide Synthesis-Fmoc-derivatives were obtained from Bachem AG (Bubendorf, Switzerland). Solid-phase assembly was initiated with t-butyl-Fmoc-Ser resin. Automated stepwise assembly was carried out entirely on an ABI 431A peptide synthesizer (Applied Biosystems, Foster City, CA). The ChTX position 32 analogs were solubilized, oxidized, and purified by reversed-phase HPLC, and fractions were pooled and lyophilized. The structure and purity of the peptides were confirmed by reversed-phase HPLC, amino acid analysis, and electrospray ionization mass spectrometry analysis. Samples were weighed and adjusted to account for peptide content prior to bioassay. The ChTX-Asp 25 analog has been described (4). Recombinant ChTX-Gln 31 , ChTX-Glu 31 , and ChTX-Orn 27 were kind gifts from Dr. C. Miller (Brandeis University).
Homology Models of Kv1.3 and IKCa1- Fig. 1 shows the amino acid sequence of the turret region, pore, and parts of the inner helix of the KcsA channel aligned with the corresponding regions of IKCa1 and Kv1.3. These channel regions interact with polypeptide toxin inhibitors, and mapping studies with these toxins have shown that the external vestibules of the Kv1.3 and IKCa1 channels are topologically similar to that of KcsA (7,9,31). Based on this alignment (Ͼ50% sequence identity) and published structural data for the KcsA channel (8), homology models of the IKCa1 and Kv1.3 vestibules were constructed and energyminimized. Coordinates for KcsA (Protein Data Bank code 1BL8) were kindly supplied by Dr. MacKinnon (Rockefeller University). Residues in each subunit that were not defined in the crystal structure (Arg 27 , Ile 60 , Arg 64 , Glu 71 , and Arg 117 ) were inserted using the Biopolymer module of Insight98 (Molecular Simulations Inc., San Diego, CA). Models of Kv1. 3 and IKCa1 were generated from the corrected structure by mutating appropriate KcsA residues (between positions 23 and 119) in Biopolymer, thus simulating the S5-P-S6 regions of the two larger mammalian channels (see Fig. 1). These models were energy-minimized in the CVFF force field of the Discover module of Insight98. 10,000 iterations were performed using the conjugate gradient algorithm with a 25-Å cutoff for non-bonded atoms and a distance-dependent dielectric in place of explicit water molecules (10,000 iterations were sufficient to bring each model to steady state).
K ϩ channels possess a conserved selectivity filter, Gly-Tyr-Gly-Asp, located in the channel pore. In KcsA, the backbone carbonyl groups of this selectivity filter are arranged in the pore lumen so as to form a series of oxygen rings, the dimensions of which allow the coordination of desolvated K ϩ ions. The Tyr side chains are oriented away from the pore and form hydrogen bonds with surrounding Trp side chains, suggesting the possibility that this acts like a spring to hold the pore open to the appropriate dimensions (8). In KcsA, this hydrogen bonding is observed between Tyr 78 and Trp 68 . Although a Trp 67 -Trp 68 diad is conserved in Kv1. 3 and Shaker, only the equivalent of Trp 67 is found in IKCa1 (Trp 242 ). Assuming that the spring mechanism exists in IKCa1 as it does in KcsA, it is presumably mediated by Tyr 253 -Trp 242 hydrogen bonding. In an attempt to preserve the overall architecture of each channel model while allowing any mutated residues to find their respective local energy minima, the C ␣ atoms of all channel residues were fixed in Cartesian space during the course of energy minimization (and during subsequent molecular dynamics simulations). Our Kv1.3 model is similar to that described previously (31), although the earlier model was constructed without access to the KcsA coordinates.
Docking of IKCa1 and Kv1.3 with ChTX-Energy-minimized channel models were juxtaposed with the closest-to-average conformation of ChTX in such a way as to preclude steric contact (typically this resulted in the closest possible contacts between ligand and receptor being Ն10 Å). ChTX was then positioned manually so that Lys 27 was facing the pore and Arg 25 was oriented toward channel residues, consistent with earlier studies (4). Further mutant cycle analyses were performed to test the ChTX-IKCa1 interactions predicted by the docking model. To preserve the overall architecture of ChTX and the channel models during docking, various constraints were placed on both molecules prior to performing molecular dynamics. The C ␣ atoms of all channel residues were fixed in Cartesian space, and distances measured from the structure of ChTX (33 d ␣␣ (i,iϩ4) distances along the backbone of ChTX and 3 d ␣␣ distances, representing the disulfide bonds at positions 7-28, 13-33, and 17-35 were applied as constraints in Discover with a tolerance of Ϯ0.1 Å and a force constant of 1000 kcal⅐mol Ϫ1 . Docking simulations were performed by first energy-minimizing the restrained ligand-receptor complex as outlined above, followed by 250 ps of molecular dynamics at 300 K with a 1.0-fs time step, a 25-Å cutoff for non-bonded atoms, and a distance-dependent dielectric. A 25-Å cutoff distance was used even though it lengthens the computation time because initial docking simulations with cutoff distances of 15 and 20 Å did not reproduce some expected interactions between charged side chains of the toxin and channel. After allowing ϳ50 ps for equilibration, the lowest van der Waals energy conformation was further energyminimized as described above. Models of ChTX docked to Kv1.3 and IKCa1 were analyzed using Insight98. An alternative docking simulation was tried in which the turret region of the Kv1.3 model (residues 373-379 of each subunit) was left unrestrained to determine what effect this might have on potential electrostatic interactions between it and the Lys 32 region of ChTX. In this simulation, the turret moved outward (further away from ChTX) relative to the simulation in which it was restrained.
Electrophysiological Analysis-Each K ϩ channel expression construct was specifically linearized and transcribed in vitro. As described earlier, the cRNA, together with a fluorescent fluorescein isothiocyanate dye, was injected into rat basophilic leukemia cells. Fluorescent cells were visualized after 2-6 h of incubation, and specific currents were measured using the patch-clamp technique (32,33). Cells measured in the whole-cell configuration were bathed in mammalian Ringer's solution containing 160 mM NaCl, 4.5 mM KCl, 2 mM CaCl 2 , 1 mM MgCl 2 , and 10 mM HEPES (adjusted to pH 7.4 with NaOH) with an osmolarity of 290 -320 mosM. The internal pipette solution for Kv recordings contained 134 mM potassium fluoride, 1 mM CaCl 2 , 2 mM MgCl 2 , 10 mM HEPES, and 10 mM EGTA (adjusted to pH 7.2 with KOH) with an osmolarity of 290 -310 mosM, and these currents were measured following 200-ms depolarizing pulses to 40 mV from the holding potential every 30 s. The internal pipette solution for IK Ca and SK Ca recordings contained 135 mM potassium aspartate, 2 mM MgCl 2 , 10 mM HEPES, 10 mM EGTA, and 8.7 mM CaCl 2 (adjusted to pH 7.2 with KOH) with an osmolarity of 290 -310 mosM (free [Ca 2ϩ ] i ϭ 10 Ϫ6 M). Kv currents were activated by a 200-ms voltage step from a holding potential of Ϫ80 to 40 mV every 30 s. K Ca currents were activated with 1 M free Ca 2ϩ i and 200-ms voltage ramps from Ϫ150 to 50 mV applied every 5 s (holding potential ϭ Ϫ80 mV). For hSlo expression, 25-50 ng of cRNA was injected into Xenopus laevis oocytes 3 to 6 days prior to recording. The outside-out patch-clamp configuration was used to record macroscopic hSlo currents. For these experiments, the external solution contained 140 mM NaMeSO 3 , 10 mM HEPES, 2 mM MgCl 2 , and 2 mM KCl (adjusted to pH 7.4 with NaOH), whereas the internal solution contained 140 mM KMeSO 3 , 10 mM HEPES, 2 mM MgCl 2 , 2 mM KCl, and 1 mM H-EDTA, and CaCl 2 was added to obtain a free [Ca 2ϩ ] i of 5 ϫ 10 Ϫ5 M (adjusted to pH 7.2 with KOH). hSlo currents were activated with 50 M free Ca 2ϩ i and 200-ms voltage ramps from Ϫ80 to 80 mV applied every 5 s. Series-resistance compensation (80%) was used if the current exceeded 2 nA. Leak currents were subtracted using the P/8 procedure for Kv currents. K d values were calculated using the equation K d ϭ ([toxin]/((1/y) Ϫ 1)) (y ϭ unblocked fraction of current) and are shown as mean Ϯ S.D. (n Ն 4).
Double-mutant Cycle Analysis-Toxin-channel interactions, predicted from the docking configuration of ChTX in IKCa1, were tested experimentally by mutant cycle analysis. This method evaluates the strength of the interaction between any given pair of channel and toxin residues. For each mutant cycle, we measured the potency (K d ) of ChTX and its analogs on IKCa1 and its mutants. The change in coupling energy (⌬⌬G) for a given pair of ChTX-IKCa1 residues and their mu-tants was calculated using the formula ⌬⌬G ϭ kTln⍀, as described (7). Based on earlier studies (3,26), ⌬⌬G values Ն0.5 kcal⅐mol Ϫ1 indicate that a pair of toxin and channel residues are likely to lie within 5 Å of each other.

Overall Strategy
Our approach to design a selective inhibitor relies on the identification of a unique feature in the ChTX-IKCa1 interface that is absent from the ChTX-Kv1.3 binding surface. Therefore, the success of our comparative structure-based design strategy depends on understanding how ChTX interacts with the IKCa1 and Kv1.3 vestibules. To accomplish this, we developed homology models of the Kv1. 3 and IKCa1 pore and vestibule regions (S5-P-S6 segments) based on the amino acid sequence alignment shown in Fig. 1 and the crystal structure of KcsA (see "Materials and Methods"). Initial docking of ChTX was performed based on published data for Kv1.3 and then carried out for IKCa1 under the assumption that ChTX may interact in a similar way. Both docking models were then tested by mutant cycle analyses, and these results were used to generate final refined docking models (Fig. 2). Comparison of the two docking models revealed a unique motif in the ChTX-Kv1.3 interface that is absent in the ChTX-IKCa1 binding surface, and three related ChTX analogs were designed to target this difference.

Docking of ChTX in Kv1.3
Several well characterized ChTX-Kv1.3 interactions, previously identified by mutant cycle analyses and electrostatic compliance experiments (4), were used to dock ChTX in the vestibule of the Kv1.3 model. In this configuration, Lys 27 of ChTX protrudes into the pore in close proximity to Kv1.3-Tyr 400 located in the selectivity filter. ChTX-Arg 25 interacts with Kv1.3-Asp 386 (in one subunit) and Kv1.3-His 404 (from two adjacent subunits). The docking model based on these data also predicts an interaction between ChTX-Lys 31

Docking ChTX in IKCa1 Based on Its Docking with Kv1.3
The polypeptide toxin ChTX blocks Kv1.3 (K d ϭ 2 nM) and IKCa1 (K d ϭ 5 nM) with almost identical potency (10,11,30), and another toxin, Stichodactyla helianthus toxin, utilizes a similar core-binding domain to bind to both channels (7) (Fig. 1), lies in close proximity to ChTX-Arg 25 and ChTX-Lys 31 , but not to ChTX-Lys 32 . In this model, the critical ChTX residue, Lys 27 , protrudes into the channel pore and lies in the vicinity of Tyr 253 in the selectivity filter. These docking models were tested experimentally by evaluating the coupling energy of specific residues predicted to be in close proximity and then refined. Specifically, we determined the coupling energies for the following pairs of specific toxin/channel residues using mutant cycle analyses: ChTX-Arg 25 /IKCa1-Asp 239 , ChTX-Lys 31 /IKCa1-Asp 239 , ChTX-Lys 32 /IKCa1-Asp 239 , and ChTX-Lys 27 /IKCa1-pore.

Mutant Cycle Analyses of Predicted Docking Configuration
ChTX-Arg 25 and ChTX-Lys 31 Interactions with IKCa1-Asp 239 -Replacement of the negatively charged residue Asp 239 in IKCa1 with the neutral asparagine (Asn 239 ) reduced the K d for native ChTX by ϳ14-fold. A more substantial reduction in affinity of ϳ180-fold was observed when a positively charged lysine (IKCa1-Lys 239 ) was introduced at this channel position (Fig. 3, A and B). Neither mutation altered the biophysical properties of the channel. Charge reversal mutations at ChTX positions 25 (Arg 25 3 Asp) and 31 (Lys 31 3 Glu) also dramatically decreased toxin affinity by ϳ1325and ϳ47-fold, respectively, for the wild-type IKCa1 channel (Fig. 3, A-C). In contrast, a similar mutation at ChTX position 32 (Lys 32 3 Asp) decreased affinity by only 9-fold (Fig. 3D).
Thermodynamic double-mutant cycle analyses were performed to measure the change in coupling energy between IKCa1-Asp 239 and each of the three ChTX residues Arg 25 , Lys 31 , and Lys 32 . The ⌬⌬G values for the mutant cycles shown in Fig. 3B are 2.25 and 3.1 kcal⅐mol Ϫ1 , respectively, indicating that IKCa1-Asp 239 and ChTX-Arg 25 lie within 5 Å of each other. Strong coupling was also seen between IKCa1-Asp 239 and ChTX-Lys 31 , with the coupling energies being 1.9 and 2.4 kcal⅐mol Ϫ1 for the two cycles shown in Fig. 3C. In contrast, the ⌬⌬G of 0.02 kcal⅐mol Ϫ1 for the cycle in Fig. 3D indicates that IKCa1-Asp 239 and ChTX-Lys 32 are not energetically coupled.
ChTX-Lys 27 Interactions with the Pore of IKCa1-Two approaches were used to test the proximity of ChTX-Lys 27 to the IKCa1 pore, the first showing that it lies close to a K ϩ -binding site and the second demonstrating that a shorter analog at this position (ChTX-Orn 27 ) interacts with a residue at the outer mouth of the pore.
All K ϩ channel toxin inhibitors contain a critical lysine residue (in this case, ChTX-Lys 27 ) that projects into the channel pore (34) and lies close to a K ϩ -binding site near the conserved tyrosine (Tyr 400 in Kv1. 3) in the selectivity filter (5-7, 9, 31). Occupancy of this site by K ϩ ions destabilizes toxin interactions with Kv1.3 via electrostatic repulsion of this critical lysine (5-7). If ChTX-Lys 27 lies close to a K ϩ -binding site near Tyr 253 in the selectivity filter of the IKCa1 pore, then occupancy of this site by K ϩ ions should reduce the affinity of the  , IKCa1, Kv1.3, Kv1.2,  Kv1.6, and hSlo showing sequence similarities >50%. The positions of the turret region, pore, and inner helix are indicated, and critical amino acids are shown in boldface. The negatively charged cluster present in Kv channels (shaded), and homologous residues in K Ca channels are boxed.
toxin for the channel. We tested this by examining the effect of increasing the external K ϩ concentration from 4.5 to 164 mM on the affinity of the IKCa1 channel for ChTX-Lys 27 . As a control, we also performed this experiment using the ChTX analog ChTX-Dap 27 , in which Lys 27 was replaced by the shorter (2.5 Å) positively charged non-natural amino acid diaminopropionic acid. Earlier studies on Kv1.3 have shown that Dap 27 interacts with His 404 at the entrance of the pore (5), rather than with the K ϩ -binding site in the selectivity filter, and the same may be true for IKCa1. Therefore, the interaction of Dap 27 with the IKCa1 pore should be insensitive to changes in the external K ϩ concentration. Consistent with our prediction, the mutant cycle shown in Fig. 4A yields a strong coupling energy (⌬⌬G ϭ 0.76 kcal⅐mol Ϫ1 ), indicating that ChTX-Lys 27 lies within 5 Å of a K ϩ -binding site located in the IKCa1 pore. Interestingly, the critical lysine at position 22 in the sea anemone S. helianthus toxin also interacts with a K ϩ -binding site within the pore of the IKCa1 channel, and the affinity of S. helianthus toxin for IKCa1 is reduced as the external K ϩ concentration is increased (7).
We also examined the sensitivity of Kv1. 3 and IKCa1 to the analog ChTX-Orn 27 , in which the shorter positively charged ornithine (side chain length of 5.0 Å) was substituted for the critical Lys 27 residue (chain length of 6.3 Å). As shown in Fig.  4B, this analog blocked Kv1.3 in the nanomolar range (K d ϭ 196 nM), but was significantly less effective against IKCa1 (K d ϭ 3300 nM). Since ChTX-Lys 27 protrudes into the pores of both IKCa1 and Kv1.3, why does ChTX-Orn 27 have lower affinity for IKCa1 than for Kv1.3? The outer pore regions of both channels are almost identical (Fig. 1), except for the presence of a histidine (His 404 ) at the entrance to the Kv1.3 pore in place of valine (Val 257 ) in IKCa1. To determine whether this difference might contribute to the differential sensitivity to ChTX-Orn 27 , we replaced Kv1.3-His 404 with valine and measured the affinity of ChTX-Orn 27 for this mutant channel, which more closely resembles the IKCa1 pore. In keeping with our hypothesis, ChTX-Orn 27 blocked Kv1.3-Val 404 with an affinity similar to that for IKCa1 and significantly less than that for wild-type Kv1.3 (Fig. 4B). We were unable to determine the effect of the reverse mutation (Val 257 3 His) on the sensitivity of IKCa1 to ChTX-Orn 27 since this mutant channel is nonfunctional.

Refinement of the Docking Models
Our mutant cycle studies demonstrate that ChTX-Arg 25 and ChTX-Lys 31 interact with IKCa1-Asp 239 in different subunits, whereas ChTX-Lys 27 lies close to a K ϩ -binding site in the pore. The docking configuration of ChTX with both models was recalculated using these experimental data as restraints in molecular dynamics simulations (Fig. 2). Where significant cou-pling was observed between toxin and channel residues (⌬⌬G Ͼ 0.7 kcal⅐mol Ϫ1 ), a target distance (as noted below) was applied between specific atoms from each residue with a 50 kcal⅐mol Ϫ1 force constant (31). Thus, docking constraints were applied between the following pairs of atoms: ChTX-Kv1.3 docking, Lys 27 N -Tyr 400 C ␤ (from all four subunits), Arg 25 C -His 404 N ␦1 (from two adjacent subunits, target distance of 8.0 Å), and Lys 31 N -Asp 386 C ␥ (from a single subunit, target distance of 6.0 Å); and ChTX-IKCa1 docking, Lys 27 N -Tyr 253 C ␤ (from all four subunits), Arg 25 C -Asp 239 C ␥ (from a single subunit), and Lys 31 N -Asp 386 C ␥ (from a single subunit, target distance of 6.0 Å).
The final docking configurations of ChTX with Kv1.3 and IKCa1 (Fig. 2) place the toxin in similar orientations in the two channels, highlighting the usefulness of homology modeling approaches in defining toxin-channel interactions. In the case of Kv1.3, the orientation of the toxin about the pore axis is guided by the proximity of Arg 25 to His 404 . Coupling of Arg 25 to two His 404 side chains in adjacent subunits brings the former in closer proximity to Asp 386 , an interaction alluded to by Aiyar et al. (5), than if Arg 25 is coupled to only one His 404 side chain. This docking configuration of ChTX with Kv1.3 matches closely the ChTX docking configuration with IKCa1 (Fig. 2). Thus, ChTX utilizes the same three key residues, Arg 25 , Lys 27 , and Lys 31 , to interact with homologous residues in the external vestibules of Kv1.3 and IKCa1. ChTX-Arg 25 and ChTX-Lys 31 , positioned at opposite ends of the toxin, interact with Kv1.3-Asp 386 and IKCa1-Asp 239 in diametrically opposite subunits of these channel tetramers. ChTX-Lys 27 , located at the center of the channel-binding surface in the toxin, projects into the pores and lies close to the selectivity filter of both channels. A similar toxin-channel interaction has also been shown for S. helianthus toxin, which utilizes a conserved core domain to interact with Kv1. 3 and IKCa1 (7) despite having a structural fold that bears no resemblance to ChTX (35).

Comparison of the Refined IKCa1 and Kv1.3 Models: Identification of a Unique Feature in IKCa1
The amino acid sequences and refined models of IKCa1 and Kv1.3 were compared to identify structural features unique to IKCa1. The sequence alignment in Fig. 1 shows a cluster of three negatively charged residues, Glu 373 , Asp 375 , and Asp 376 , in the turret region of Kv1.3 (shaded) that are not present in IKCa1. The Kv1.3 model (Fig. 2, A and B) shows the locations of these three acidic residues within the vestibule of Kv1.3. Glu 373 and Asp 376 are oriented toward the center of the channel pore, whereas Asp 375 is at the outer edge of the turret. The docking configuration of ChTX in Kv1. 3 indicates that ChTX-Lys 32 lies in the vicinity of Glu 373 and Asp 376 , although the terminal ammonium group of this Lys 32 is ϳ10 Å away. No intermolecular constraints were applied among any of these residues during the docking, but if a weak constraint was included between Asp 376 C ␥ and Lys 32 N, the distance from Lys 32 to this part of the turret decreased to ϳ8 Å, without any significant change elsewhere. Based on this observation, the introduction of a negatively charged residue at ChTX position 32 might therefore significantly reduce the affinity of such an analog for Kv1.3 via electrostatic repulsion.
IKCa1 contains two neutral residues (Ala 226 and Gln 229 ) and one basic residue (Arg 228 ) ( Figs. 1 and 2, C and D) in place of the acidic residues in the turret of Kv1.3. The only negatively charged residue in the turret region of IKCa1 (Glu 227 ) is located at the outer edge of the turret, pointing away from the center of the vestibule (Fig. 2,C and D), and is therefore unlikely to interact directly with any ChTX residues. In contrast to Kv1.3, a ChTX analog containing a negatively charged residue at position 32 would be expected to retain most of its potency against the IKCa1 channel.

Introduction of Negatively Charged Residues at Position 32 in ChTX Results in Analogs Selective for IKCa1 over Kv1.3
To test these predictions, we replaced Lys 32 in ChTX with glutamate and tested the affinities of this novel analog (ChTX-Glu 32 ) and native ChTX for the cloned IKCa1 and Kv1.3 channels. Native ChTX blocked both IKCa1 and Kv1.3 in the low nanomolar range with K d values of 5 and 2 nM, respectively (Figs. 5, A and B, and 6). As anticipated from our model, introduction of a negatively charged residue at ChTX position 32 reduced the affinity of this analog for Kv1.3 by ϳ350-fold while only minimally affecting (ϳ6-fold) its affinity for IKCa1. Thus, ChTX-Glu 32 exhibits a ϳ20-fold higher affinity for IKCa1 than for Kv1.3.
Since both channels are expressed endogenously in activated human T-lymphocytes, we also examined the effect of the ChTX-Glu 32 analog on native IKCa1 and Kv1.3 currents. Fig.  5C shows a ramp protocol eliciting K ϩ currents in activated human T-lymphocytes. IKCa1 was the main carrier of K ϩ currents at potentials more negative than Ϫ40 mV, whereas at depolarized potentials, K ϩ currents were carried by a combination of IKCa1 and Kv1.3 channels. Consistent with the results on the cloned channels, the voltage-dependent Kv1.3 current in activated T-lymphocytes was affected only minimally by 250 nM ChTX-Glu 32 (Fig. 5C), a concentration that blocked ϳ25% of the cloned Kv1.3 current, whereas this concentration of ChTX-Glu 32 almost completely inhibited IKCa1 currents. In contrast, native ChTX blocked both channels equally in the low nanomolar range (Fig. 5C). These results show that ChTX-Glu 32 is a selective and potent inhibitor of the cloned IKCa1 channel and its native counterpart in human T-lymphocytes compared with Kv1. 3.
Encouraged by the selective properties of ChTX-Glu 32 , we generated two additional ChTX analogs. The negatively charged residues aspartate (ChTX-Asp 32 ) and p-carboxyphenylalanine (ChTX-Cpa 32 ) were substituted for Lys 32 in ChTX to investigate the influence of the side chain length on potency and selectivity. The aspartate side chain is shorter (3.1 Å) than in glutamate (4.6 Å), whereas the p-carboxyphenylalanine side chain is longer (7.3 Å). ChTX-Asp 32 and ChTX-Cpa 32 blocked IKCa1 channels significantly more potently than Kv1.3 (Fig. 6), although ChTX-Glu 32 , the analog with the intermediate-sized side chain, was the most selective because it exhibited the greatest difference in affinity between IKCa1 and Kv1.3. Thus, using a structure-based homology modeling strategy, we predicted a novel toxin-channel interaction that was exploited in the design and engineering of three novel ChTX analogs (ChTX-Asp 32 , ChTX-Glu 32 , and ChTX-Cpa 32 ). Each of them contains a negatively charged residue at position 32 and selectively blocks IKCa1 channels while being significantly less effective against Kv1.3.

Selectivity Profile of the ChTX Position 32 Analogs
ChTX is reported to block three other K ϩ channels potently besides IKCa1 and Kv1.3, including the voltage-gated K ϩ channels Kv1.2 (27) and Kv1.6 (36) and the large-conductance Ca 2ϩactivated K ϩ channel hSlo (37,38). In an attempt to predict the behavior of the ChTX position 32 analogs on these channels, we compared the amino acid sequences of the turret regions of these channels with those of Kv1.3 and IKCa1. Kv1.2 and Kv1.6 channels both contain a negatively charged cluster at the position homologous to that in Kv1.3 (Fig. 1). If the vestibules of these channels have a similar architecture to that of Kv1.3, they would exhibit a lower affinity for the ChTX position 32 analogs compared with wild-type ChTX. In contrast, hSlo has only two negatively charged residues in the turret region, Glu 305 and Asp 302 (Fig. 1). If its vestibule has a comparable topology to that of IKCa1, hSlo-Glu 305 and hSlo-Asp 302 would be at the same positions as IKCa1-Glu 227 and IKCa1-Ser 224 , respectively (Fig. 2,C and D), and neither residue would lie in close proximity to ChTX-Lys 32 . Therefore, the introduction of negatively charged residues at this toxin position should have little effect on toxin potency for hSlo. In keeping with our predictions, the introduction of negatively charged residues at position 32 in ChTX significantly reduced the affinity of the three analogs for Kv1.2 in a similar fashion to Kv1.3 while only minimally affecting their affinity for hSlo. We found that the Kv1.6 channel was insensitive to ChTX, in contradiction to some published data (36), but in confirmation of other studies (39). Consequently, Kv1.6 was also resistant to the ChTX-Glu 32 analog (Fig. 7). Since our comparative modeling approach accurately predicted the sensitivity of the Kv1.2 and hSlo channels to the three ChTX position 32 analogs, it is likely that the external vestibules of Kv1.2, Kv1.3, IKCa1, and hSlo are structurally similar to that of the KcsA channel.

Concluding Remarks
In this study, we generated homology models of the pore regions of two prototypical mammalian K ϩ channels, the voltage-gated Kv1.3 channel and the Ca 2ϩ -activated IKCa1 channel, both present in human T-lymphocytes, based on their known structural similarity to the KcsA channel. Our Kv1.3 model differs in detail from the previous models (4, 5), for example, in the orientation of the key residue Tyr 400 in the selectivity filter. Guided by established ChTX-Kv1.3 interactions, we docked this toxin in the Kv1.3 model and used mutant cycle analysis to confirm the proximity of a pair of ChTX (Lys 31 ) and Kv1.3 (Asp 386 ) residues predicted to be close. Since ChTX blocks IKCa1 with roughly equivalent potency compared with Kv1.3, we hypothesized that ChTX might sit in the IKCa1 vestibule with a similar geometry to that in Kv1. 3. We therefore performed a docking simulation of ChTX in IKCa1 for heuristic purposes. Multiple sets of predicted interactions were confirmed by mutant cycle analyses and then used to generate refined models of the docking configurations. Comparison of the two toxin-channel interfaces suggested a unique structural motif, a cluster of negatively charged residues present only in the Kv channel, that was exploited in the design and generation of three novel ChTX analogs. These analogs, containing negatively charged residues at toxin position 32, exhibit specificity for the IKCa1 channel over Kv1. 3. They do not block other voltage-gated and small-conductance Ca 2ϩ -activated K ϩ channels that lack this unique feature, but they inhibit hSlo channels, which resemble IKCa1 in the turret region.
Our results strongly suggest that a polypeptide toxin with comparable affinities for different K ϩ channels, even those belonging to widely divergent subfamilies (e.g. Ca 2ϩ -activated IKCa1 and voltage-gated Kv1.3 channels), interacts with a topologically similar toxin-binding site in the external vestibule of these channels (Fig. 2). This has allowed us to start with a promiscuous polypeptide toxin (ChTX) that blocks multiple K ϩ channels and to apply a comparative homology modeling approach to design novel analogs that selectively target intermediate-and large-conductance Ca 2ϩ -activated K ϩ channels. These ChTX analogs might be useful tools in elucidating the role of the IKCa1 channel in cells that express ChTX-sensitive voltage-gated K ϩ channels, but not hSlo, e.g. activated T-lymphocytes, erythrocytes, and colonic epithelia. Finally, our studies provide support for the feasibility of using the structure of the KcsA channel to guide the design of selective and potent inhibitors for a large variety of mammalian K ϩ channels.