The EAG gene family consists of the EAG (ether à go-go), ERG (ether à go-go-related gene), and ether à go-go-like potassium channel subfamilies (Bauer and Schwarz, 2001), all of which encode depolarization-activated potassium (K⁺) channels. hERG1 channels are predominantly expressed in the heart and give rise to the repolarizing I_Kr current that is responsible for action potential termination (Sanguinetti et al., 1995). Genetic defects in the herg1 gene can cause the long-QT2 syndrome, which sometimes leads to sudden cardiac death. More common, however, are acquired forms of the long-QT syndrome, caused by the pharmacological blockade of cardiac ion channels. In this context, hERG1 channels seem to be the most important target for those substances that induce the undesired side effect of QT prolongation (Keating and Sanguinetti, 2001). Therefore, substantial effort is spent in elucidating the structural requirements that make hERG1 channels particularly susceptible to various agents, whereas most other voltage-gated K⁺ channels (Kv channels) are insensitive. In a pioneering work by Mitcheson and colleagues (2000), an alanine scan of the inner pore region of hERG1 channels identified several residues that are of particular importance for stabilizing antiarrhythmic drugs inside the pore cavity. In the first place, these were two aromatic residues in the sixth transmembrane segment (Tyr652 and Phe656; see Fig. 2a) and two more residues facing the cavity at the apex of the pore helix (Thr623 and Val625). In particular the aromatic (S6) residues have been shown to be of high importance for hERG1 inhibition by a broad range of substances, including dofetilide, quinidine, MK-499, terfenadine, cisapride, vesnarinone, and chloroquine (Lees-Miller et al., 2000; Mitcheson et al., 2000; Kamiya et al., 2001; Sánchez-Chapula et al., 2002). However, they may constitute a necessary, but not a sufficient criterion for high-affinity block (nanomolar range) of antiarrhythmic agents.

This becomes evident in that the mentioned residues within S6 are conserved among EAG channels, although these channels differ strongly in their pharmacological properties. Schönherr et al. (2002) have shown that hEAG1 and hEAG2 channels differ in their sensitivity toward the antiarrhythmic agent quinidine by a factor of ∼100. Ju and Wray (2002) have shown that these channels differ in sensitivity toward terfenadine. Moreover, recent data indicate that the aromatic S6 residues Tyr652 and, in particular, Phe656 only partially affect hERG1 inhibition by fluvoxamine (Milnes et al., 2003).

Hence, the pharmacology of K⁺ channels from the EAG family is not simply determined by residues of the central pore cavity. In this respect, hEAG1 and hEAG2 provide an ideal pair of highly conserved proteins with different pharmacological properties. Given the strong sensitivity of hEAG1 channels toward the quaternary ammonium derivative clofilium and its tertiary analog LY97241 (Gessner and Heinemann, 2003), we systematically investigated the differences of pharmacological properties of these two channels. Employing site-directed mutagenesis and electrophysiological characterization of the effects of various drugs on these channels we identified residues inside the channel's pore that are essential for high-affinity blockade. In addition to two sites in the cavity (Ser436 and Val437), the residue Ala453 was found to be important for high-affinity binding of clofilium by hEAG1. Our experimental findings could be qualitatively reproduced in a ligand docking approach, making use of hEAG homology models based on the crystal structure of the bacterial KcsA channel (Doyle et al., 1998). The docking simulations with partial conformational relaxation of the hEAG model structures suggest a high-affinity binding mode with the alkane tail of clofilium located between selectivity filter and S6 helix. This binding mode can also qualitatively explain the effect of amino acid differences between hEAG1 and hEAG2 on clofilium binding based on differences in stereo-chemical fit between ligand and receptor. We propose a binding mode of “anchored cavity block,” which assumes that blocking molecules in the cavities can be anchored by hydrophobic chains in the space between the pore helix and S6, leading to increased binding affinity.

Materials and Methods

Results

Inhibition of hEAG1 and hEAG2 Channels by LY97241. hEAG1 channels were shown to be very sensitive to the antiarrhythmic agent clofilium and its sensitivity was similar to that of hERG1 channels (Gessner and Heinemann, 2003). In addition, hEAG1 channels were sensitive to the antiarrhythmic quinidine, whereas hEAG2 channels showed considerably lower sensitivity (Schönherr et al., 2002). Given the potency of clofilium to inhibit hEAG1 channels, we wanted to compare this with hEAG2 to gain a deeper insight into the molecular mechanisms of channel block by antiarrhythmics. Because clofilium is permanently charged and therefore does not readily permeate through the membrane to access the channel pore from the cytoplasmic face, we used its tertiary ammonium analog LY97241, which has been shown to inhibit hEAG1 channels expressed in mammalian cells (Gessner and Heinemann, 2003).

In Fig. 1a, current traces from HEK 293 cells expressing hEAG1 (top) and hEAG2 (bottom) channels are shown for the indicated concentrations of LY97241. hEAG1 channels are already strongly blocked at 10 nM. In contrast, inhibition of hEAG2 channels requires much higher LY97241 concentrations, suggesting a structural difference in the binding sites underlying this difference. The dose-response curves in Fig. 1b yielded apparent IC₅₀ values of 10.2 and 1520 nM for hEAG1 and hEAG2, respectively. Thus, under these conditions, hEAG1 channels are inhibited by LY97241 about 150-fold more potently than hEAG2 channels, giving rise to a systematic assessment of the underlying molecular mechanisms.

Chimeras of hEAG1 and hEAG2 Channels. To localize the major molecular determinants for the differences described above, chimeras between hEAG1 and hEAG2 (see Materials and Methods and Fig. 2a) were assayed for LY97241 block upon expression in HEK 293 cells. Figure 2b shows dose-response curves for these constructs; the resulting IC₅₀ values are presented in Fig. 2c. Introducing the entire core domain of hEAG2 (i.e., all transmembrane segments, including the pore loop) into hEAG1 makes the resulting E1-c2 construct even less sensitive toward LY97241 than hEAG2 (factor 234 with respect to hEAG1). In the reverse case, E2-c1 is not quite as sensitive as hEAG1 (factor 57 with respect to hEAG2). Exchange of the pore regions (E1-p2 and E2-p1) confers the respective properties of the other channel type to a lesser degree (factors 45 and 27, respectively). In contrast, introducing the outer pore region of hEAG2 in hEAG1 (E1-S5p2) had almost no effect (factor 1.6 with respect to hEAG1). For hERG1 it was shown that residues Thr623, Ser624, and particularly Val625, which are conserved in hEAG1 (Thr435, Ser436, and Val437) but differ in part in hEAG2 (Thr431, Thr432, and Ile433) are important for stabilizing the channel blockers MK-499, terfenadine, and cisapride (Mitcheson et al., 2000). Therefore, we also tested the double mutant hEAG1 S436T/V437I (E1-SV/TI) and the reverse mutant hEAG2 T432S/I433V (E2-TI/SV). On a logarithmic scale, both mutants yield roughly intermediate sensitivities between hEAG1 and hEAG2 (factors 33 and 18). Thus, the pore region is the major determinant for LY97241 sensitivity. Residues Ser436 and Val437 in hEAG1 have a strong impact, but they are not the only residues important for stabilization of the ligand. The aromatic residues Tyr464 and Phe468, which were identified to be crucial of MK-499 binding in hERG1 channels (Mitcheson et al., 2000), are conserved among hEAG1, hEAG2, and hERG1 and are thus probably required for drug stabilization but cannot solely account for the high-affinity block of hEAG1 obtained by LY97241.

Voltage Dependence and Activation Kinetics. The block of hEAG1 channels by LY97241 is strongly use- and voltage-dependent (Gessner and Heinemann, 2003). To test whether effects on block efficiency correlate with differences in gating between the EAG variants, we also characterized the voltage dependence and kinetics of activation of all chimeras and mutants. As shown in Fig. 2d, a higher mean open probability is achieved in hEAG1 compared with hEAG2 channels. This could potentially account for the stronger inhibition by the open-channel blocker LY97241. In Fig. 2e, the estimated half-maximal activation voltages (V_n) and the slope factors (k_n) characterizing the voltage dependence of activation are shown for all channel constructs. Chimera E1-c2 has a strongly left-shifted activation (V_n = -42.7 ± 2.2 mV compared with -4.7 ± 2.2 mV for hEAG1). Nevertheless, this left-shift in activation was not accompanied by an increase in block efficiency. In fact, block was even weaker than for hEAG2. The strongest right-shift is observed in chimera E2-p1 (V_n = 49.6 ± 1.8 mV compared with 32.0 ± 3.1 mV for hEAG2). However, block efficiency was intermediate between E2-c1 and E2-TI/SV, both characterized by less depolarized V_n values. Thus, the mutants span a V_n range of about 90 mV, but there is no clear correlation to the block efficiency. A similarly inconsistent picture is obtained for the slope factors, k_n.

Kinetics of channel activation from hyperpolarized voltages to +50 mV was estimated using a binomial activation model as described previously (Schönherr et al., 1999). This model takes into account the fast and slow gating steps described for the activation of EAG channels. For the time constant of the rapid activation step, hEAG1 (4.15 ± 0.26 ms) and hEAG2 (4.33 ± 0.27 ms) channels did not show significant differences. For the slow activation kinetics, which was about two times faster in hEAG1 (15.1 ± 1.0 ms) than in hEAG2 (27.3 ± 1.7 ms), the core domain was the major determinant (data not shown). The pore domain had an intermediate effect and the double mutation in the pore did not have a big impact. However, these alterations of the gating parameters of the channel constructs did not directly lead to a conclusion regarding residual differences between the block efficiency of hEAG1/E2-c1 and hEAG2/E1-c2, for example.

Thus, differences in susceptibility toward LY97241 between human EAG isoforms cannot be simply explained by their gating characteristics but must result from differences in the binding site within the channel's pore.

Mapping of the Pore by Site-Directed Mutagenesis. Model building of hEAG1 and hEAG2 K⁺ channels based on the known structure of the bacterial K⁺ channel KcsA indicates that most of the amino acids that differ in both channels do not point toward the interior of the channel (Fig. 3b). Only residues Ser436 (Thr432 in case of hEAG2; see arrow in Fig. 3b) and Val437 (Ile433 for hEAG2) contact the interior of the pore in the models. Replacement of these residues in hEAG1 by the corresponding residues of hEAG2 and vice versa had a strong effect on channel block by LY97241 when expressed in HEK 293 cells.

To gain a better insight into the block mechanism we aimed to measure channel block by direct application of blocker from the cytosolic side. For this purpose, we had to employ X. laevis oocytes as an expression system. To avoid any possible ambiguity regarding the binding site, we used clofilium instead of LY97241 for this purpose. The alignment of the pore region of KcsA, hERG1, hEAG1, and hEAG2 (Fig. 3a) shows that only eight residues are different between hEAG1 and hEAG2 in the inner pore loops. In addition to the double mutant E1-SV/TI we generated all other mutants individually and tested the effect of 10 nM clofilium in inside-out macro-patches. The results are shown in Fig. 3c. Although hEAG2 channels are barely blocked (remaining current, 87.9 ± 2.2%), hEAG1 channel current is reduced to 6.8 ± 0.5% of the control. Most individual mutations, including S491N (data not shown), which in addition to the pore residues differed between hEAG1 and E1-p2, had minor effects when introduced in the background of hEAG1. However, the mutants indicated in red strongly reduced the block by clofilium: S436T (42.9 ± 1.3%), V437I (16.9 ± 1.1%), S436T/V437I (72.2 ± 0.6%), and A453S (41.9 ± 5.6%). Therefore, we can conclude that in addition to the residues Ser436 and Val437 in hEAG1, residue Ala453 contributes to the stabilization of clofilium in the pore.

Comparison of Different Blocking Agents. The impact of Ala453 on channel block by clofilium is not obvious when looking at the structure of the pore model. Previous models in which hERG-blocking substances were docked inside the channel's cavity have considered a contact to sites homologous to Ser436 and Val437 in hEAG1 (Mitcheson et al., 2000). Ala453, however, is not pointing into the pore as reproduced in a molecular modeling approach (see Fig. 3b).

Docking clofilium to the hEAG1 cavity in front of the selectivity filter keeping the protein rigid yielded solutions in which the ligand filled part of the cavity. A typical solution is shown in Fig. 4a (left). The aromatic ring of clofilium is approximately located at the center of the channel contacting the aromatic Phe468 side chains. The corresponding residue (Phe656) has been found previously to be important for high-affinity binding in hERG channels (Mitcheson et al., 2000). However, rigid docking did not produce any placements with close contacts between the ligand and residues that differ between hEAG1 and hEAG2 and that have been found to be important for clofilium binding. In contrast, the semiflexible docking procedure that allowed for conformational relaxation of the protein upon ligand docking (see Materials and Methods) resulted in a complex with clofilium in an extended conformation and the alkane tail located in a space between the S6-helix and the selectivity filter. This region is accessible upon small conformational readjustment of the channel structure (Fig. 4a, right). Note that in the crystal structure of an open conformation of the bacterial MthK channel, the space between S6 and the selectivity filter is well accessible (Jiang et al., 2002). In the closed channel form, that is represented by the hEAG1 model, the alkane chain of clofilium is well packed and in van der Waals contact with several hydrophobic groups, among them also Ser436, Val437, and even Ala453 is not very far from the end of the alkane chain (∼4 Å). The aromatic part is still in contact with the Phe468 residues located approximately at the center of the channel.

Results from molecular modeling suggest that the alkane tail of clofilium is stabilized in the space between the pore loop and S6, near Ala453. This implies that other blockers without such a tail should not respond strongly to changes at that position. Therefore, we tested the block of hEAG1, hEAG1 S436T, hEAG1 V437I, E1-SV/TI, hEAG1 A453S, and E1-c2 in inside-out patches of X. laevis oocytes for block by clofilium, quinidine, and TBA. The chimera E1-c2 was chosen to represent the hEAG2 phenotype, because hEAG2 was expressed only weakly in X. laevis oocytes, compromising quantitative measurements in the inside-out mode. The resulting dose-response curves are shown in Fig. 4, a-c (top). The corresponding IC₅₀ values are listed in Table 1. In all three cases of blockers, hEAG1 channels are much more strongly inhibited than hEAG2 channels. For clofilium, the difference between hEAG1 and E1-c2 inhibition was smaller by a factor of ∼2 than determined in HEK 293 whole-cell measurements, indicating differences in channel gating or block efficiency depending on expression system and recording mode. In addition, in all cases, the double mutant E1-SV/TI strongly reduces the block, almost restoring hEAG2 properties. This is not the case for mutant A453S, however. Although this mutation has a strong impact for clofilium (Fig. 4a), the impact is smaller for quinidine (b) and there is basically no impact at all for TBA (c). With respect to E1-c2, the triple mutant E1-c2-TIS/SVA substantially increases the block efficiency of all substances tested (Table 1).

Subsequently, we attempted to correlate these differences to structural considerations using the docking approach described above. Quinidine can also be docked at a similar position in the cavity, with its aromatic part in contact with the central Phe468 residues. However, it is bulkier than clofilium, and its shorter ethene group cannot penetrate as deeply into the space between the S6-helix and the selectivity filter as clofilium (see Fig. 4b). The interaction between the channel blockers clofilium and quinidine with the hEAG1 variants was calculated after conformational relaxation of the mutated model complexes (Table 1). Mainly because of an increased sterical strain caused by the larger side chains Thr432 and Ile433 (hEAG2 numbering) in two of the variants, a reduced ranking compared with wild-type hEAG1 was found in qualitative agreement with the experimental observation. In the reverse case the double mutation in the background of hEAG2 (E2-TI/SV) enhanced clofilium binding by about the same energy difference (-9.2 kcal/mol) than obtained for E1-SV/TI (+10.8 kcal/mol; see Table 1). The distance (>10 Å) between the ligand and amino acids 445, 448, 451, 456, and 459 excluded any sterical influence on clofilium binding in these calculations. No effect or only a very small calculated effect was found for the A453S mutant, indicating that there may not be sterical reasons for the observed reduction in binding of this variant. Ligand docking studies, including explicit water molecules, are computationally too demanding to allow for a systematic search for putative ligand binding positions at the channel pore region. Therefore, polar and electrostatic contributions to the calculated ranking of docking solutions have been included approximately by applying an implicit electrostatic screening model with an effective dielectric constant ϵ = 20 (representing the mixed protein and aqueous environment in the pore). The nonpolar alkane chain of clofilium forms a well packed complex with the “anchor” region in the docked complexes such that the relative ranking is mainly determined by the sterical fitness between channel mutants and ligand.

However, it might be possible that the A453S mutation stabilizes or traps water molecules (not present in the calculations) in the space between the S6-helix and the selectivity filter that must be expelled by hydrophobic groups of the ligands. Such effects are not included in the calculated ranking of ligand binding. The calculated ranking in case of a docking position obtained after rigid docking (with no close contacts between Ser436 or Val437 and clofilium) (Fig. 4a) showed a much smaller effect of the mutations on binding with no correlation to the experimental results (Table 1).

Based on the X-ray structure of the KcsA channel in complex with TBA, a model for the hEAG1-TBA complex was generated (see Materials and Methods). In this model, the quaternary ammonium group of TBA is located at the entrance of the selectivity filter and the butyl groups partly filling the cavity but without close contacts to Ser436, Val437, or Ala453 (Fig. 4c). For this binding geometry, no significant sterical influence was found on mutation of the above residues on binding (data not shown). However, more accurate electrostatic energy calculations using the finite-difference Poisson-Boltzmann method (Madura et al., 1995) indicate a higher electrostatic penalty for introducing TBA into an S436T cavity compared with wild-type hEAG1. In the finite-difference Poisson-Boltzmann method, a dielectric boundary between molecule and surrounding aqueous solvent is introduced, allowing estimation of the change in electrostatic solvation upon complex formation. In case of TBA binding to the smaller S436T cavity, the calculated change in electrostatic solvation was 2.1 kcal/mol less favorable than binding to the wild-type hEAG1 cavity assuming an internal dielectric constant ϵ_in = 4. This difference increased to 4.9 kcal/mol in case of using ϵ_in = 2 for the protein (ϵ_w = 78 for water). Such energy changes correspond to changes of the binding constant of between 30 (ϵ_in = 4) and 3500 (ϵ_in = 2) in qualitative agreement with the experimental observation of strongly reduced TBA binding to hEAG2 or hEAG1 variants that contain an S436T mutation.

Thus, with the semiflexible docking approach, we were able to qualitatively explain the molecular mechanisms underlying the differences in efficiency of block by clofilium, quinidine, and TBA. The data suggest that a possible anchoring of hydrophobic chains of blocking molecules in the cavities between the pore helix and S6 can result in additional stabilization yielding high-affinity block.

Because the impact of mutation A453S on the effect of clofilium could be circumstantial, we tested as additional hERG1 channel blockers MK-499 and terfenadine for their efficiency to inhibit currents mediated by hEAG1, hEAG1 A453S, and E1-c2. In inside-out patches, the following IC₅₀ values were determined: MK-499, 102 ± 14 nM (n = 4, hEAG1), 358 ± 30 nM (n = 5, A453S), 22.7 ± 1.5 μM (n = 5, E1-c2); terfenadine, 15.8 ± 2.6 nM (n = 6, hEAG1), 177 ± 14 nM (n = 3, A453S), 1.03 ± 0.24 μM(n = 3, E1-c2). Thus, both substances distinguish well between hEAG1 and hEAG2 pores (factor of 222 and 65, respectively) and in both cases, mutation A453S has a significant effect on channel block (factor of 3.5 and 11.2, respectively).

The obvious effect of mutation A453S on the channel block by clofilium, quinidine, MK-499, and terfenadine suggests that neighboring residues may also affect drug stability in the pore. Therefore, we attempted to assay mutation M457A, a structural neighbor of Ala453 located closer to the pore cavity. This mutant resulted only in very small currents in X. laevis oocytes that could not be assayed with patch-clamp methods. In addition, the small currents measured in whole oocytes displayed inactivation such that a faithful interpretation of block studies would be compromised. Residue Ser433 is located on the pore helix in relation to the center between Ala453 and M457. When mutated to either alanine or cysteine, channels with largely normal kinetics were obtained. The block by clofilium in the inside-out configuration was affected only marginally (less than 2-fold; see Table 1). This result is compatible with model considerations because neither S433A nor S433C has a strong sterical impact on the alkane chain of clofilium when located in the pore pocket. Channel block by quinidine and terfenadine was strongly increased (4- to 13-fold reduced IC₅₀ values; see Table 1), supporting the importance of the pore pocket region in stabilizing drugs inside the hEAG pore. However, with the modeling approach applied here, the increased binding of TBA and quinidine is not readily explained.

Discussion

The acquired form of the long-QT syndrome is a relatively common side effect of therapeutically different drugs (Keating and Sanguinetti, 2001). The hERG1 potassium channels, which give rise to the repolarizing I_Kr currents in cardiac myocytes, are the primary molecular targets. Several studies have shown that, compared with Kv channels, the outstanding pharmacological sensitivity of the hERG1 channels is mainly based on a larger volume of the inner cavity forming the binding site (del Camino et al., 2000; Mitcheson et al., 2003), and the lining of this binding site in particular by two aromatic residues (Tyr652 and Phe656) within S6 (Mitcheson et al., 2000). Moreover, inactivation of hERG1 channels strongly enhances the block efficiency of several drugs (e.g., Suessbrich et al., 1997; Ficker et al., 1998, 2001), although the underlying mechanism is not fully understood (Chen et al., 2002). In this work, we studied the inhibition of a pair of channels, hEAG1 and hEAG2, that, in contrast to hERG1 channels, do not inactivate but share the important aromates and the large size of the inner cavity with hERG1 channels. Although several drugs block hEAG1 and hERG1 channels in a similar fashion, hEAG2 channels are often much less sensitive. Therefore, the couple hEAG1/hEAG2 constitutes an ideal pair of ion channels suited to study the molecular details involved in high-affinity block of EAG channels in general but in the absence of inactivation.

Channel Inhibition by LY97241. hEAG1 channels expressed in HEK 293 cells are effectively blocked by LY97241, a tertiary analog of the antiarrhythmic agent clofilium (Steinberg and Molloy, 1979). The IC₅₀ value of 10.2 nM obtained in this study agrees well with previous studies on hEAG1 channels expressed in Chinese hamster ovary cells (4.9 nM) and HEK 293 cells (8.3 nM), measured with a similar pulse protocol, but in the presence of physiological Mg²⁺ concentrations (Gessner and Heinemann, 2003). hEAG2 channels, which like hEAG1 channels do not inactivate, are much less effectively blocked by LY97241. The slightly weaker inhibition of hEAG1 compared with hERG1 channels (2.2 nM) might reflect the lack of inactivation, which was shown to affect hERG inhibition by LY97241 (Suessbrich et al., 1997) and other blockers like dofetilide (Ficker et al., 1998, 2001). The difference in LY97241 sensitivity between hEAG1 and hEAG2 (150-fold), which clearly is not attributable to differences in channel gating, is slightly larger than for quinidine [100-fold (Schönherr et al., 2002)]. In both cases, the IC₅₀ values for the inhibition of hEAG1 channels are less than 5-fold higher than for hERG1 channels (Paul et al., 2002; Gessner and Heinemann, 2003). Thus, the pharmacological differences between the two noninactivating channels hEAG1 and hEAG2 are much more pronounced than between hEAG1 and the inactivating channel hERG1. The same holds true for the two hERG blockers terfenadine and MK-499, which block hEAG1 with a potency intermediate to those of clofilium and quinidine. Introduction of the hEAG2 core into the hEAG1 channel reduced the effects of all substances tested by at least 50-fold.

The potency of LY97241 to inhibit either channel is about 100-fold higher than for quinidine. Previous work on EAG channels has been hampered by the lack of potent and selective blocking agents. Therefore, the finding that LY97241, clofilium, and terfenadine are high-affinity blockers for hEAG1 is of high practical relevance. Because EAG1 channels have been implied to increase the oncogenic potential of tumor cells (Pardo et al., 1999), pharmacological tools are urgently needed to test this hypothesis and to examine the usefulness of these channels as pharmacological targets in cancer treatment. In addition, these blockers may be useful tools to distinguish EAG1 and EAG2 channels in the central nervous system to elucidate their physiological role in neuronal signaling.

Molecular Mechanism of Clofilium Block. Building on the similarities between LY97241 and clofilium in structure and their binding to hEAG1 channels (Gessner and Heinemann, 2003), we tested the effect of single-site mutations in hEAG1 on clofilium block. To gain direct access of the blocker to the intracellular side, channels were expressed in X. laevis oocytes, and clofilium was applied in the inside-out configuration. Note that the stronger effect of clofilium in these experiments, compared with the channel inhibition by LY97241 in HEK 293 cells, can be explained by the different recording modes (whole-cell versus inside-out) and a slightly higher potency of clofilium compared with LY97241 (Gessner and Heinemann, 2003).

We found that three residues in hEAG1 (Ser436, Val437, and Ala453), when mutated to the corresponding residues in hEAG2, attenuated clofilium block. The other residues have only marginal impact because they are apparently not pointing toward the pore pocket (e.g., I456M and I459V). Two of the important residues (Ser436 and Val437) located at the apex of the pore helix and, hence, lining the inner cavity are conserved in hERG1 channels and also affected hERG1 channel inhibition by MK-499 when mutated to alanine (Mitcheson et al., 2000). In agreement with our results for LY97241 block, a double mutation of both residues does not fully confer clofilium sensitivity between hEAG1 and hEAG2 channels. The third residue, Ala453, remote from the inner cavity, also attenuates clofilium block, intermediate to the single-site mutations at positions 436 and 437. Residue Ala453 also strongly affects binding of MK-499 and terfenadine although channel block by TBA is not altered.

To interpret the experimental findings on a molecular and structural level we employed docking studies of clofilium to homology models of hEAG1 and hEAG2 channels, based on the crystal structure of the bacterial KcsA channel. The modeling attempt is based on the assumption that the channel structure of KcsA is generally similar to that of the hEAG1 and hEAG2 structures. Although the resulting structural models showed good stereochemical quality, such homology models may contain structural inaccuracies (e.g., small misplacements of main chain and side chain elements), possibly limiting its usefulness for docking studies. In fact, using the rigid model structures during docking forced clofilium to adopt a kinked conformation (a typical solution is shown in Fig. 4a) to fit into the space in front of the selectivity filter, and the resulting model was unable to explain the experimental findings. To cope with the possible inaccuracies of the homology models and the possibility of protein conformational changes during ligand binding, we employed a semi-flexible docking procedure. This procedure allowed for full relaxation of protein side chains and at least partial adjustment of main chain and whole secondary structural elements during docking. The semiflexible docking procedure resulted in a clofilium-channel complex with intimate contacts between clofilium and residues 436 and 437, and the alkane tail of clofilium located in a space between S6 and pore helix, a working model referred to as anchored cavity block. In addition to the larger number of contacts with the protein, the clofilium ligand adopts an energetically more favorable straight conformation with most dihedral torsion angles of the alkane chain in the trans configuration compared with the conformations observed in case of rigid docking. The ligand placement found assuming a semiflexible protein during docking is further supported by the qualitative agreement of calculated ranking versus experimental data for amino acid substitutions at residues 436 and 437 (see below). Moreover, the aromate ring of clofilium is in close contact to the aromatic Phe468 side chain, presumably determining the enhanced block efficiency compared with Kv channels (Malayev et al., 1995). The corresponding residue (Phe656) has been shown previously to be important for high-affinity binding in hERG channels (Mitcheson et al., 2000). Phe468, however, cannot account for the difference in clofilium block between hEAG1 and hEAG2 channels because it is conserved between both channel isoforms.

The ligand docking performed in this study assumed an ideal fit of the ligand in the closed channel conformation, because there is ample of evidence that high-affinity blockers, particularly clofilium (Gessner and Heinemann, 2003), are trapped in the closed channel. However, it should be noted that the open-channel conformation provides a much greater room for potential blocking molecules such that the initial association to the docking site can be much facilitated.

Anchored Cavity Block. The strong effect of the A453S mutation on clofilium block of hEAG1 channels may be explained either as a direct effect on the proposed alkane tail binding site or an indirect gating effect, caused by an altered flexibility of the S6 helices. To discriminate between these possibilities, we compared the effect of mutations at positions 436, 437, and 453 on EAG inhibition by clofilium, quinidine, and TBA. In addition, we tested the effect of A453S on MK-499 and terfenadine block. Although anchored binding was expected to be weaker for quinidine, because of a bulkier and shorter alkane side chain, no anchored binding should be observable for TBA. Moreover, docking simulations supported the experimentally testable hypothesis that the strength of attenuation of block by mutations at positions 436 and 437 should be in the order 436 > 437 for clofilium, 436≈437 for quinidine, and 436 >> 437 for TBA, although in the latter case attributable more to electrostatic than sterical reasons. The experimental results were in good agreement with these presumptions. In addition, MK-499 and terfenadine might be able to protrude into the pore pocket. However, as modeled for MK-499 to a closed channel structure (data not shown), more space is required between S6 and the pore helix to accommodate the bulkier tails. It is readily conceivable, however, that this space is provided in a fully flexible channel given the large conformational changes expected during channel opening (Jiang et al., 2002).

Hence, in addition to the aromatic residues Tyr464 and Phe468, which were postulated to stabilize antiarrhythmic agents in the pore of hERG1 channels by a π-stacking mechanism, more requirements are necessary for high-affinity block of K⁺ channels of the EAG family. Our data suggest that a possible anchoring of hydrophobic chains of blocking molecules in the cavities between the pore helix and S6 can result in additional stabilization yielding high-affinity block. Such additional interactions between alkyl tails and parts of the pore wall have been postulated also for block of Kv channels by alkyl-triethylammonium derivatives (Choi et al., 1993), depending on the length of the alkyl chains. Future studies, using trifunctional, clofilium-based substances with different lengths and flexibilities of the alkyl-tail and the alkyl-linker to the aromate head, have to address the exact positioning of clofilium within the EAG channel cavity. This is important because 1) aromatic groups, as in dofetilide, may permit such an anchoring (Ficker et al., 2001) and 2) the homologous position of the S6 aromates between human ERG1 and Drosophila melanogaster EAG1 has been questioned and repositioning during activation/inactivation gating may occur (Chen et al., 2002).

Our proposed model of anchored cavity block for substances like clofilium explains the pharmacological differences between hEAG1 and hEAG2 channels. However, it cannot be excluded that allosteric effects induced by changes at position Ala453 contribute to the observed phenotypes. Nevertheless, regardless of the exact molecular binding mode, the results provided here highlight for the first time the importance of residues in the space between S6 and the pore helix for the stabilization of various drugs in the pore of EAG channels.

Given the clinical relevance, a natural question is whether or not the equivalent sites would also affect pore block of hERG1 channels. hERG1 channels harbor a serine at that position (Ser641). We generated mutation hERG1 S641A and assayed it for block by LY97241, quinidine, and MK-499 in HEK 293 cells. The block by LY97241 was increased by a factor of 2 (wt, 2.17 ± 0.44 nM, n = 5; mutant, 1.03 ± 0.04 nM, n = 6). The block of quinidine (1.27 ± 0.46 μM, n = 5 versus 1.38 ± 0.15 μM, n = 5) and MK-499 (7.78 ± 1.29 nM, n = 5 versus 11.6 ± 1.7 nM, n = 6) was not affected. This result qualitatively supports the impact of residue S641 for the stabilization of LY97241 in hERG1 channels. However, mutation S641A accelerates inactivation by about 4 to 5-fold and shifts the midpoint of voltage-dependence of recovery from inactivation by about -40 mV. Therefore, a faithful quantitative assessment of state-independent binding properties of the blockers cannot be achieved. In fact, this is one of the major advantages of using the hEAG1 channel as a model system for such studies, because its pore is highly homologous to that of hERG1 channels but does not undergo inactivation.

Because block of hERG1 channels is often the cause of undesired cardiac drug side-effects, pharmacophore models would be valuable for the estimation of hERG1 inhibition by compounds before expensive preclinical and clinical studies. However, recent attempts to create such models were based on the assumption that compounds in the pore cavity are bound by a common mechanism (Ekins et al., 2002). Our structural consideration for hEAG channels now imply that small compounds such as TBA are bound by the cavity, whereas affinity to alkyl-tailed compounds, such as clofilium, or substances with bulkier tails, such as MK-499 and terfenadine, can be increased by neighboring pore compartments. Regardless of whether this relies on direct anchoring of the drugs in the pore pocket, one can distinguish between (at least) two groups of blockers with different binding modes, being affected by either mutation A453S or S433A in hEAG1. Because different binding modes imply different drug-channel interactions, it seems likely that one cannot straightforwardly propose a universal pharmacophore model for hERG (or hEAG) channels. Predictive pharmacophore models for the EAG channel family need to account for different binding modes of complex blocker structures.