A Novel Extracellular Calcium Sensing Mechanism in Voltage-Gated Potassium Ion Channels

Direct observation of HERG inactivation: effect of extracellular Ca²⁺ on channel gating

The HERG K⁺ channel was originally described as an inwardly rectifying channel based on the current–voltage relationship (Trudeau et al., 1995). Subsequent studies indicated that this apparent rectification results from inactivation gating (Schönherr and Heinemann, 1996; Smith et al., 1996; Spector et al., 1996). In this regard, HERG K⁺ channels seemed to be atypical, compared with other six transmembrane family K⁺ channels that are either delayed rectifiers or A-type channels with rapid channel activation followed by somewhat slower inactivation. Figure1A illustrates HERG K⁺ currents measured during brief (50 msec) steps between +30 and +180 mV. At these membrane potentials, the rate of channel activation became rapid relative to inactivation, and the inactivation process was directly observed. Although the rate of activation remained highly voltage-dependent, the rate of inactivation was nearly constant in this voltage range.

Changes in extracellular Ca²⁺ modify the voltage dependence of opening of this channel with little effect on channel inactivation (Johnson et al., 1999). Figure1B demonstrates this directly. Superimposed K⁺ current tracings from a single cell are shown in different extracellular Ca²⁺concentrations ranging from 3 μm to 10 mm. In the lowest concentrations of Ca²⁺, the voltage step to +70 mV elicited a rapidly activating current that quickly inactivated like an A-type K⁺ channel. As extracellular Ca²⁺ was increased, channel opening was delayed, but the rate of inactivation remained unchanged. The result was a truncation of the observed current peak. Note that marked modulation occurred between 0.3 and 3 mm. The specific interaction of Ca²⁺ with the voltage dependence of channel activation, but not inactivation, led us to investigate the role of amino acids in the vicinity of the channel voltage sensor in determining channel Ca²⁺sensitivity.

Extracellular negatively charged amino acids and the effect of [Ca²⁺]_o

Ca²⁺ typically interacts with negatively charged amino acid side chains of proteins (Cowan, 1993;Eismann et al., 1994; Schreiber and Salkoff, 1997). This is attributable to the fact that ionized Ca²⁺has a full outer electron orbital, giving it a noble gas-like unreactive chemistry. As a result, Ca²⁺tends to interact with other molecules largely through electrostatic attraction and repulsion (Cowan, 1993). Likely candidates for Ca²⁺ interaction are extracellularly exposed negatively charged amino acids such as glutamate and aspartate. Because Ca²⁺ does not permeate HERG channels and the intracellular solution always contained a high affinity Ca²⁺ chelator (5 mmBAPTA), the hydrophilic extracellular loops of the channel are candidate regions for Ca²⁺ interaction. The S4 transmembrane segment has been identified as a voltage sensor of channels in the six transmembrane voltage-gated K⁺ channel family (Papazian et al., 1987;Tempel et al., 1987; Perozo et al., 1994; Mannuzzu et al., 1996). The extracellular S3–S4 linker of HERG contains only two acidic amino acids, adjacent glutamates at positions 518 and 519, and they are located just above the S4 transmembrane segment. These negatively charged amino acids were mutated individually and together to probe their role in the response to extracellular Ca²⁺. The phenotypes of the WT and mutant channels are shown in Figure 2. K⁺ current from cells expressing WT, E518A, E519A, or the double mutant (change of E518 and E519 to alanines: EEAA) are shown at numerous test potentials. Each mutation significantly altered the gating of HERG channels in control extracellular solutions. Interestingly, the E518A mutant appeared very similar to the WT channel in elevated [Ca²⁺]_o, e.g., removal of the negative charge caused a decrease in outward current during depolarizations and enhanced the rates of tail current decay. Thus, removal of this negative charge seemed to resemble addition of positive charges (Ca²⁺) in the WT channel. In all three cases (E518A, E519A, and EEAA), outward currents during depolarizations were decreased, and the rates of decay of tail currents were increased. Further evaluation of these mutant channels compared with WT revealed unique differences (described below). First, we evaluated other channel properties to determine whether inactivation or permeation properties were altered.

Effect of negatively charged amino acids, mutations, or Ca²⁺ on inactivation

Our previous study demonstrated a significant effect of Ca²⁺ on HERG channel activation gating in the absence of an effect on inactivation (Johnson et al., 1999). Therefore, we wished to determine whether these amino acids that appeared to influence channel activation gating had an effect on channel inactivation. Despite the distinct current phenotypes of the mutant channels, their inactivation was not affected. Neutralization of E518, E519, or both together did not change the HERG channel inactivation time constants measured between 0 and +100 mV (p > 0.2; n = 4–8) (Fig.3). Thus, HERG inactivation gating is insensitive both to external Ca²⁺ and mutation of E518 and E519, in direct contrast to activation gating.

Mutations did not change K⁺ selectivity

In contrast to changes in activation gating, we observed no effects on the ion permeation properties, suggesting that other channel properties were not affected by these changes. The ionic selectivity of all the mutant channels was normal, because there was no change in the K⁺ current reversal potential measured in 4 mm extracellular K⁺: WT, −83 ± 0.9 mV, n = 10; E518A, −81 ± 0.3 mV, n = 9; E519A, −82 ± 0.3 mV,n = 10; EEAA, −82 ± 0.6 mV, n = 10 (NS; p > 0.2).

Effect of negatively charged amino acids or Ca²⁺on channel activation

In Figure 4, voltage–activation curves are shown to compare the shifts in the voltage dependence caused by changes in [Ca²⁺]_o in the WT and mutant HERG channels. Increasing Ca²⁺ caused progressive shifts of channel voltage-dependent activation relationships to more depolarized potentials in all channels, but the magnitude of the Ca²⁺ response and the relative position of the voltage dependence curves differed among the WT and charge-neutralized mutant channels. Note the differences in degree of separation of the curves in high and low Ca²⁺ and the relative shifts caused by change in Ca²⁺ among the different channels. There did not appear to be changes in the slope of the relationships, but the entire family of curves in a given channel (WT vs each mutant) was variously translated along the voltage axis. The degree to which elevation of Ca²⁺ was able to shift the curves was different among the channels as well. Initially it seemed paradoxical that channels with such similar and modest changes in amino acid sequence could behave so differently. A channel with the simple removal of a negative charge behaved very differently than when the adjacent charge was neutralized. These distinctions were very nicely accounted for by considering the allosteric nature of the channels. The solid curves through the data were generated by the allosteric gating model described below.

The Ca²⁺-independent voltage dependence of the channels (i.e., that measured in very low calcium: 3 μm Ca²⁺) differed among the channels as well. This can be seen more clearly in Figure5 in which the half maximal voltages of channel activation (V_1/2) measured from data in Figure 4 are plotted as a function of the Ca²⁺ concentration. The double mutant (EEAA) and the E518A single mutant had a minimalV_1/2 of approximately +30 mV more depolarized than that of the WT or E519A mutant channels (V_1/2 in 3 μm[Ca²⁺]_o: WT, −44 ± 6 mV; E518A, −17 ± 3 mV; E519A, −48 ± 5 mV; EEAA, −17 ± 8 mV). For all channels, theV_1/2–[Ca²⁺] curve saturated before reaching the lowest tested [Ca²⁺] (3 μm). Therefore, the minimal V_1/2 can be considered Ca²⁺ independent. Neutralization of glutamate 518, but not 519, resulted in channels with an altered Ca²⁺-independent voltage dependence. Channels in which E519 was neutralized had an altered Ca²⁺ response, but their Ca²⁺-independent voltage dependence was like that of WT channels.

The Ca²⁺ responses of the WT and mutant channels for currents measured at membrane potentials between −60 and +30 mV were fit with a Hill equation (Eq. 2; same data as in Fig. 4analyzed as a function of [Ca²⁺]_o; data not shown). The fitted parameters (K_D and Hill coefficient, b) were highly voltage dependent. The Hill coefficients from these fits varied from a minimum near one to a maximum greater than three depending on the membrane potential, suggesting multiple binding sites and cooperativity. Thus, a single Ca²⁺ binding site seems unlikely for several reasons. First, the Hill coefficient (b > 1, Eq. 2) values indicate increased complexity and suggest multiple Ca²⁺ binding sites. Secondly, HERG, like other six transmembrane domain voltage-gated K⁺ channels, is believed to form rotationally symmetric homotetrameric channels when expressed alone in heterologous systems. If all four subunits cooperated to create a single Ca²⁺ binding site on the extracellular face of the channel, this site would necessarily be centrally located and would therefore likely be in the channel pore. However, we know that Ca²⁺ does not block HERG channels as would be expected for a binding site in the pore lumen (Johnson et al., 1999). In the absence of a single centrally located Ca²⁺ binding site, a likely possibility for a homotetrameric channel is that each subunit has its own Ca²⁺ site(s). The similarities to hanatoxin modification of Kv2.1 and omega-Aga-IVA modification of calcium channels are also consistent with four Ca²⁺ sites and certainly more than one (Swartz and MacKinnon, 1995, 1997a,b; Li-Smerin and Swartz, 1998;Winterfield and Swartz, 2000).

The Hill fit-derived parameters must be interpreted cautiously, and the apparent voltage dependence of the Hill-derived K_Ds can be interpreted in several ways. First, this nonglobal analysis (i.e., fitting only a single independent variable at a time) neglects relevant information about the system, for example membrane potential. Although the Ca²⁺ response is obtained at several membrane potentials, each fit is independent and not constrained by data at other membrane potentials. The Hill fit analysis also lumps all interactions (possibly multiple K_Ds) into a single K_D. In contrast, the global fit (discussed below) is constrained by information obtained across all membrane potentials and calcium concentrations. The voltage-dependent apparent K_D values from the Hill fit analysis suggest several possible interpretations. First, the accessibility of the ligand to a single binding site of constant affinity could change as a function of membrane potential, because of the voltage-dependent channel conformational changes that must occur for channel function, i.e., the binding site is partially guarded. This would cause an apparent change in the affinity, but cannot account for the Hill coefficients that differ from unity. Another possible explanation for the voltage-dependent K_D is that Ca²⁺ binding is influenced by the electric field through a direct effect on the positively charged ligand. In the Hill equation analysis, the fitted K_Ds ranged from ≤10⁻⁴m at −50 mV to 5 × 10⁻³m at +25 mV. These changes are consistent with a positively charged ligand binding with higher affinity at negative membrane potentials. This cannot readily explain the voltage-dependent changes in Hill slopes either.

In addition and not mutually exclusive, these observations could indicate that the affinity of the binding site for Ca²⁺ changes as a function of membrane potential because of a change in the protein conformation. For example, different channel conformations (closed or activated) may have distinct Ca²⁺ affinities. The data indicate that the resting closed channels bind Ca²⁺ with a higher affinity than the activated channel. Membrane potential determines the fraction of channels that are in the resting or activated states and the apparent K_D increases at more depolarized potentials in which more channels are activated.

The voltage dependence of the Hill fits led us to consider an allosteric model for the Ca²⁺-channel interactions. Many multisubunit proteins including ion channels exhibit cooperativity, and conformational changes that confer differential ligand affinity have been described for many allosteric proteins (Monod et al., 1965; Hille, 1977; Hondeghem and Katzung, 1977; Marks and Jones, 1992; Galzi et al., 1996; Cox et al., 1997; Changeux and Edelstein, 1998; Horrigan and Aldrich, 1999; Rothberg and Magleby, 1999).

Allosteric modulation of gating by Ca²⁺

To refine the analysis of the Ca²⁺-HERG interaction, we considered voltage-dependent allosteric models of channel-gating and Ca²⁺ interactions. Because HERG is a voltage-gated channel (Sanguinetti et al., 1995; Trudeau et al., 1995), numerous states must be considered to account for channel voltage dependence (Patlak, 1991; Bezanilla et al., 1994; S. Wang et al., 1997). These channels are homotetramers when expressed heterologously. The four-fold symmetry of the channel naturally leads to at least five states to describe channels with 0–4 voltage sensors activated. Channel inactivation requires additional states. If only a single Ca²⁺ binds to each homotetrameric channel, then all four subunits must create a single Ca²⁺ binding site on the extracellular face of the channel. This site would necessarily be centrally located, as with tetraethylammonium (TEA) and certain toxins (Hurst et al., 1992), and would therefore likely be in the channel pore. However, Ca²⁺ does not block HERG channels (Johnson et al., 1999). The radial symmetry of the channel, the apparent cooperativity seen in the Ca²⁺ concentration–response curve Hill fits (data not shown), and the unlikelihood of a single central Ca²⁺ binding site suggest that each subunit has its own Ca²⁺ site(s). Minimally, to accommodate five voltage-dependent states and five calcium-liganded states requires a 25 state model. To include Ca²⁺ and voltage dependence and the allosteric changes from closed to activated channel conformations requires at least a two-tiered 50 state allosteric modelRothberg and Magleby, 1999). Nonindependent voltage sensors or Ca²⁺ binding would lead to even larger models (Horrigan et al., 1999).

As the complexity of these models increases, the number of free parameters increases, and confidence in the meaning of individual parameters decreases. Results shown below and in our previous work (Johnson et al., 1999) indicate that the inactivated states cannot be distinguished from open states with regard to Ca²⁺; hence, combining open and inactivated states that are occupied during depolarizations can be justified. Because the primary goal of this work was to understand the Ca²⁺ modulation of HERG, we can further simplify the analysis by combining the voltage sensing steps into a single concerted transition. There is reasonable evidence that in other voltage-gated K⁺ channels (Bezanilla, 2000), Ca²⁺-activated K⁺ channels (Cox et al., 1997; Horrigan et al., 1999; Rothberg and Magleby, 1999), and Ca²⁺ channels (Marks and Jones, 1992) a final cooperative step is indeed responsible for channels leaving the closed state.

Assuming four binding sites for the tetrameric channel, there are at least five types of states corresponding to the channel with 0–4 Ca²⁺ bound. The open to inactivated transitions of HERG are Ca²⁺ insensitive (Johnson et al., 1999), which permitted us to combine the open and inactivated states (referred to as activated). Making the assumption that only the voltage-dependent conformation of the channel (closed or activated) determines the affinity for Ca²⁺ and that each subunit binds Ca²⁺ independently from the others, the degree of Ca²⁺ binding then biases the fundamental voltage-dependent transition. The result is the simplified two-tiered, 10 state MWC model [see Cox et al. (1997) for in-depth discussion of this approach shown below in scheme 1 and described in Eq. 3] if Ca²⁺ only distinguishes closed and activated states (i.e., all closed state dissociation constants, Kc, are equivalent, as are all activated state dissociation constants, Ka). In this model, the receptor (here the HERG channel) protein exists in two possible conformations, tense or relaxed, corresponding in this case to the closed or activated channel (Monod et al., 1965; Changeux and Edelstein, 1998).

This voltage-dependent allosteric model (see Eq. 3) was fit to all of the data simultaneously as a function of the known membrane potentials and Ca²⁺ concentrations. Figure6 shows the voltage- and Ca²⁺-response surface that was fitted to these data sets. The shaded gray bands indicate the height along the vertical axis. Round symbols show the mean data points, and the vertical lines on the symbols show the distance of the data points from the fitted plane, illustrating the quality of the fit.

This simple model fit the observed data remarkably well. The smooth curves in Figures 4 and 5, as well as the shaded gray surface in Figure 6, are all from the model fit. The fitted parameters are shown in Figure 7. The sensitivity of the channels to changes in membrane potential was similar in all channels and at all Ca²⁺ concentrations, as reflected by the fitted Q (apparent gating charge), which was the same for all the channels (mean ± SE; WT = 2.3 ± 0.05 e⁻, E518A = 2.5 ± 0.04 e⁻, E519A = 2.4 ± 0.05e⁻, EEAA = 2.5 ± 0.002 e⁻). The constancy of Q among the different channels suggests that the E518 and E519 do not constitute part of the gating charge and are not directly responsible for voltage sensing. The change in free energy required to move the channels from closed to activated states in the absence of Ca²⁺ and voltage (L₀) differed considerably among the channels. This is apparent not only from fittedL₀ values (Fig. 7), but also by the increased voltage threshold for channel activation in the E518A and EEAA mutants visible in Figures 4 and 5. In contrast, the E519A mutant had an L₀ similar to that of the wild-type channel. Removal of the negative charge at position 518 had little effect on the Ca²⁺ dissociation constant of the closed channel (Kc; Fig. 7), but channels bearing an E519A mutation had a Kc twice that of the WT or E518A mutant, indicating that neutralization of this negative charge decreased the affinity of the closed channel for Ca²⁺.

We found that the E518A mutant had a reduced Ka, indicating an enhanced affinity of activated channels for Ca²⁺, despite the fact that Kawas not reduced in the EEAA double mutant. This observation suggests that the 518 and 519 amino acids are not entirely independent with respect to their influence on activated channel Ca²⁺ binding. To quantify the degree of energetic coupling between these amino acids, we used thermodynamic mutant cycle analysis (see Scheme FS1 and Eq. 4) (Carter et al., 1984;Sali et al., 1991; Hidalgo and MacKinnon, 1995; Schreiber and Fersht, 1995; Horovitz, 1996; Ranganathan et al., 1996; Frisch et al., 1997). The distinct Ca²⁺ dissociation constants for the closed (Kc) and activated (Ka) states and the equilibrium constant (L₀) for the distribution among closed and activated states for each channel were used to estimate the energetic interactions between these charged amino acids in the different conformational states. Using these parameters, a coupling coefficient (Ω) was calculated (Eq. 4). AnΩ of unity indicates independence and a lack of energetic coupling. Deviations of Ω from unity indicate energetic coupling between the amino acids. The Ω values were:L₀, 21.2; Kc, 1.1;Ka, 2.4. The large Ω for theL₀ indicates that there is energetic coupling between these amino acids in determining the equilibrium change in free energy between the closed and activated states, although clearly E518 was more critical in determining theL₀ value (Fig. 7). The coupling coefficient for the closed channel Ca²⁺dissociation constant was near unity, indicating a lack of coupling between these residues in the closed states. In contrast,Ω for activated channels was 2.4, indicating that an interaction between these amino acids influences Ca²⁺ binding on channel opening. ThisΩ value is relatively small, but it is noteworthy because these adjacent residues did not display coupling in the closed state. This implies a change in the microenvironment around E518 and E519 between the closed and activated channels. The negative charge at position 518 influences Ca²⁺ association when the channel is in the activated state but not when the channel is closed.