An Extracellular Cu²⁺ Binding Site in the Voltage Sensor of BK and Shaker Potassium Channels

Copper and zinc are essential components of many enzymes and other proteins and, aside from iron, are the most abundant trace elements in the human body. Both metals are present in plasma at ∼15 μM and accumulate in the brain at concentrations on the order of 100 μM (Osterberg, 1980; Linder and Hazegh-Azam, 1996; Takeda, 2000; Mathie et al., 2006). While most of this metal is protein bound, Cu²⁺ and Zn²⁺ may function as signaling molecules that are concentrated in synaptic vesicles and coreleased with neurotransmitters (Frederickson et al., 2005). Nerve terminals in certain areas of the brain contain copper or zinc that is histochemically stainable and therefore present in a free or loosely bound form (Slomianka et al., 1990; Sato et al., 1994; Ono and Cherian, 1999; Takeda, 2000). Brain slice and synaptosome preparations release Cu²⁺ and Zn²⁺ in a Ca²⁺-dependent manner upon membrane depolarization (Howell et al., 1984; Hartter and Barnea, 1988), suggesting that total concentrations in the synaptic cleft may approach 100–250 μM Cu²⁺ (Kardos et al., 1989) or 300 μM Zn²⁺ (Assaf and Chung, 1984). The fraction of this release that can interact with synaptic proteins has not been clearly established, but indicator dye measurements suggest free synaptic concentrations on the order of 3 μM Cu²⁺ (Hopt et al., 2003) or 10 μM Zn²⁺ (Li et al., 2001; Thompson et al., 2002; Frederickson et al., 2006) are possible. Consequently there is considerable interest in understanding how low μM concentrations of Cu²⁺ or Zn²⁺ may interact with synaptic proteins such as ion channels. Such interactions could potentially play a role in normal synaptic physiology as well as disease states with neurological symptoms including inherited disorders of Cu²⁺ transport (Wilsons's and Menke's disease) and neurodegenerative diseases such as Alzheimers, Parkinsons, Creutzfeldt-Jakob disease, and amyotrophic lateral sclerosis, which are associated with altered Cu²⁺ and/or Zn²⁺ homeostasis (Bush, 2000; Frederickson et al., 2005; Mathie et al., 2006). In a broader sense, the ability of ion channels to detect trace metal ions is also relevant to understanding transition metal toxicity (Kiss and Osipenko, 1994).

Extracellular metal ions are well known to influence ion channel function by altering gating or blocking the pore (Elinder and Arhem, 2003). A variety of channels are extremely sensitive to Cu²⁺ and/or Zn²⁺ at concentrations <10 μM including subtypes of glutamate, glycine, Gaba(A), P2X, and acetylcholine receptors (Mathie et al., 2006; Huidobro-Toro et al., 2008), two-pore-domain K⁺ (K2P) channels (Gruss et al., 2004), and voltage-gated K⁺ (Kv1.3) (Teisseyre and Mozrzymas, 2006), Na⁺ (Nav1.5) (Mathie et al., 2006), and Ca²⁺ channels including a number of high voltage–activated types (Castelli et al., 2003) and Cav3.2 (Jeong et al., 2003). While the number of voltage-dependent channels identified as extremely sensitive to Cu²⁺ or Zn²⁺ is small, many voltage-gated channels exhibit responses to higher concentrations of metals (e.g., ≥100 μM) that have been studied in detail and provide a mechanistic basis for understanding of metal action (Elinder and Arhem, 2003).

Many transition metals at millimolar concentrations can shift the voltage dependence of channel activation in a nonspecific manner by screening surface charge on the channel and lipid bilayer. Similarly, metal occupancy of specific sites on channels outside of the voltage sensor can influence gating through electrostatic interaction with the voltage sensor (Elinder and Arhem, 2003; Yang et al., 2007). Metals can also bind directly to the voltage sensor and have been proposed to inhibit voltage sensor activation in squid axon K⁺ and Na⁺ channels by stabilizing the resting conformation (Gilly and Armstrong, 1982a,b). Acidic residues in the S2 and S3 segments of the voltage sensor domain of dEAG and hERG channels have been identified as putative coordinating sites for a variety of divalent cations (Silverman et al., 2000; Silverman et al., 2004; Fernandez et al., 2005). A key histidine in the S3–S4 loop of the voltage-sensor is also critical for the extreme metal sensitivity of Cav3.2 (Kang et al., 2006; Nelson et al., 2007). Finally, extracellular metals can enhance inactivation (Yellen et al., 1994), and the ability of Zn²⁺ to reduce the maximal conductance (G_KMAX) of Kv1.5 channels has been attributed to such a mechanism, involving putative coordinating residues in the pore domain (Kehl et al., 2002).

Although various mechanisms and sites of metal action in voltage-gated channels have been described, many fundamental questions remain unanswered. Most of the above examples involve channels with relatively low (mM) affinity and selectivity for metals. Which of these mechanisms, if any, apply to channels that are sensitive to metals at concentrations on the order of 1 μM? To what extent can channels distinguish between closely related metal ions like Cu²⁺ and Zn²⁺, and what mechanisms underlie selectivity? How are metals coordinated by voltage sensors, and how do such sites change with voltage sensor activation to interact with metals in a state-dependent manner?

Large conductance BK-type K channels (Slo1) are activated by voltage and intracellular Ca²⁺, play an important role in synaptic physiology (Faber and Sah, 2003), and are present in parts of the brain (Misonou et al., 2006) that contain synaptic Cu²⁺ and Zn²⁺ (Kozma et al., 1981; Slomianka et al., 1990; Sato et al., 1994; Ono and Cherian, 1999; Takeda, 2000). A previous study reported that BK channels from rat skeletal muscle are insensitive to 1–10 μM Cu²⁺ or 100 μM Zn²⁺, but their activity is reduced slowly and irreversibly by prolonged exposure to ≥20 μM Cu²⁺(Morera et al., 2003). This inhibitory effect of Cu²⁺ occurred with a substantial 20-min delay at 20 μM, was partially reversed by the reducing agent dithiothreitol (DTT), was reduced by mutation of extracellular Cys residues, and was attributed to a pro-oxidant effect of copper on amino acid sulfhydryl groups. By contrast, we demonstrate here an acute inhibitory effect of extracellular Cu²⁺ on the activation of heterologously expressed mSlo1 BK channels, characterized by an IC₅₀ of only 2 μM. The inhibition of G_K by Cu²⁺ is rapid, readily reversible, unaffected by removal of extracellular Cys residues, and therefore unlikely to involve channel oxidation. Instead, our results suggest that Cu²⁺ binds between transmembrane segments S1, S2, and S4 of the resting voltage sensor to inhibit voltage sensor activation. The binding site appears to be remarkably selective for Cu²⁺ over other transition metals (Zn²⁺, Cd²⁺, Mn²⁺, Fe³⁺, Co²⁺, and Ni²⁺). The state dependence and selectivity of Cu²⁺ action may explain why the extreme metal sensitivity of the BK channel has not been described previously. We also demonstrate a response of Shaker K⁺ channels to micromolar Cu²⁺ involving some of the same voltage sensor positions that are important in mSlo1. The ability of K⁺ channel voltage sensors to coordinate Cu²⁺ and Zn²⁺ has implications for understanding not only mechanisms of metal action in BK and related voltage-gated channels but also the conformational changes associated with voltage sensor activation.

BK channels experiments were performed with the mbr5 clone of the mouse homologue of the Slo1 gene (mSlo1) (Butler et al., 1993). The clone was propagated and cRNA transcribed as described previously (Cox et al., 1997). The wild-type (WT) Shaker channel was Shaker H4 with residues 6–46 deleted to remove N-type inactivation and T449V to inhibit C-type inactivation. Xenopus oocytes were injected with ∼0.5–50 ng cRNA, incubated at 18°C, and studied 2–7 d after injection. Site-directed mutagenesis was performed with the QuikChange XL site-directed mutagenesis kit (Stratagene) and confirmed by sequencing.

Currents were recorded using the patch clamp technique in the outside-out configuration (Hamill et al., 1981), at room temperature (20–22°C). For BK channels, the internal solution contained (in mM) 110 KMeSO₃, 20 HEPES, and 40 μM (+)-18-crown-6-tetracarboxylic acid (18C6TA) to chelate contaminant Ba²⁺ (Diaz et al., 1996; Neyton, 1996). In addition, “0 Ca²⁺” solution contained 5 mM EGTA, reducing free Ca²⁺ to an estimated 0.8 nM, and Ca²⁺ solutions contained 5 mM HEDTA. [Ca²⁺] reported as 1, 3, 5, and 10 μM correspond to concentrations of 0.87, 2.4, 3.98, and 8.7 μM measured with a Ca²⁺ electrode (Orion Research Inc.). Ca²⁺ was added as CaCl₂ and [Cl⁻] was adjusted to 10 mM with HCl. The standard external solution contained 110 KMeSO₃, 2 MgCl₂, 6 HCl, 10 MES, 10 MOPS. MES and MOPS buffers were used rather than HEPES, which is known to chelate Cu²⁺ (Mash et al., 2003; Sokolowska and Bal, 2005). For Shaker channels, the internal solution contained (in mM) 160 KCl, 11 EGTA, 20 HEPES, and the standard external solution contained 155 NaCl, 5 KCl, 3CaCl₂, 1 MgCl₂, 10 MES, 10 MOPS. The pH of all solutions were adjusted to 7.2, except for pH experiments where the external pH was adjusted from 5.2 to 8.14. Solutions were exchanged by washing the chamber with 20 volumes of solution (∼5 ml), with the exception of experiments in Fig. 1 B and Fig. 3 (C and D), which used a rapid perfusion system (SF-77B Perfusion Fast-Step, Warner).

Metal stock solutions of 0.1 M or 1 M nitrate salt were prepared in distilled water, stored at 4°C, and diluted to their final concentration in standard external solution immediately before use. Metals were not buffered and are described by their nominal concentrations. All chemicals were Ultra-grade from Sigma-Aldrich including MES and MOPS (>99.5%) and transition metal nitrate salts (>99.99%) to avoid metal contamination. Metals were washed out using a 5 mM EGTA solution prepared from the standard external solution with a total of 2.43 mM Mg²⁺ to maintain free Mg²⁺ at 2 mM. Following application of EGTA solution, the chamber was washed with standard external solution before recording currents. We did not observe any effect on mSlo1 or Shaker channel function of switching between standard extracellular solutions and EGTA solution, suggesting any metal contamination is below the channel's detection limit.

Free metal concentrations can be influenced by the formation of chloride complexes and the limited solubility of metal hydroxides at neutral pH. Chloride is primarily a concern for Cd²⁺, estimated to reduce [Cd²⁺]_free to 49% of its nominal concentration in mSlo1 solution (10 Cl⁻) based on equilibrium constants reported by Pivovarov (2005). By contrast, free Cu²⁺ and Zn²⁺ should only be reduced to 99% and 99.5% of their nominal concentration in mSlo1 solution or 86% and 90% in Shaker solution (160 Cl⁻) (Pivovarov, 2005). The poor solubility of copper hydroxide (Cu(OH)₂) can limit [Cu²⁺]_free to equilibrium concentrations on the order of 100 μM, but hydroxide formation is slow and can require many hours to equilibrate (Hidmi and Edwards, 1999). Consequently, metal solutions were prepared and used within 30 min to minimize hydroxide formation, and discarded if any precipitate was observed. To evaluate the free copper concentration in our experimental solutions, [Cu²⁺] measured with a Cu²⁺ electrode (Orion Research Inc.) was compared in equivalent solutions at pH 7.2 and an acidic pH 5.8 where hydroxide formation is minimal. The results were indistinguishable after correcting for a constant shift in electrode potential due to pH, indicating that the concentrations used in experiments are very close to the nominal concentrations. In both cases, electrode potentials varied linearly with log[Cu²⁺] over a range of 0.1 to 1000 μM with slopes (∼30 mV/10-fold change) within the range specified by the manufacturer.

Data were acquired with an Axopatch 200B amplifier (Molecular Devices Corp.) in patch mode with Axopatch's filter set at 100 kHz. Currents were filtered by an 8-pole Bessel filter (Frequency Device, Inc.) at 20 kHz and sampled at 100 kHz with an 18-bit A/D converter (Instrutech ITC-18). A P/4 protocol was used for leak subtraction (Armstrong and Bezanilla, 1974) with a holding potential of −80 or −90 mV for mSlo1 or Shaker channels, respectively. Electrodes were made from thick-walled 1010 glass (World Precision Instruments, Inc.). The electrode's resistance in the bath solution (1.0–2.5 MΩ) was used as an estimate of series resistance (R_S) for correcting the voltage at which macroscopic I_K was recorded. Series resistance error was <15 mV for all data presented. A Macintosh-based computer system was used in combination with Pulse Control acquisition software (Herrington and Bookman, 1995) and Igor Pro for graphing and data analysis (WaveMetrics, Inc.). A Levenberg-Marquardt algorithm was used to perform nonlinear least-squared fits. Data are presented as mean ± SEM.

For mSlo1 channels, open probability (P_O) was estimated over a wide voltage range by recording macroscopic I_K when NP_O was high (>5), and single channel currents in the same patch when NP_O was low (<10). Macroscopic conductance (G_K) was determined from tail currents at −80 mV following 30-ms voltage pulses (at 5-s intervals), and was normalized by G_Kmax in the absence of metals. At more negative voltages, NP_O was determined from steady-state recordings of 1–60-s duration that were digitally filtered at 5 kHz. NP_O was determined from all-points amplitude histograms by measuring the fraction of time spent (P_K) at each open level (k) using a half-amplitude criteria and summing their contributions NP_o = ΣkP_k. P_o was then determined by estimating N from G_Kmax (N = G_Kmax/γ_K, where γ_K is the single channel conductance at −80 mV).

G_K-V and dose–response relations were characterized by Boltzmann functions and Hill equations, respectively, as data descriptor functions. Boltzmann functions were of the form G_K(V) = G_Kmax [1 + exp(−z_APP(V − V_0.5)/kT)]⁻¹, where z_APP is the apparent charge movement. Hill equations were of the form Y([L]) = Y₀ + (Y_Max − Y₀)*[L]^nH/([L]^nH + IC₅₀^nH), where [L] is the ligand concentration, Y₀ and Y_Max are values of the dependent variable (Y) at zero and saturating [L], and n_H is the Hill coefficient.

Macroscopic BK channel potassium current (I_K) evoked by membrane depolarization is reduced markedly and rapidly by 100 μM extracellular Cu²⁺ (Fig. 1, A and B). I_K was recorded in the absence of intracellular Ca²⁺ (0 Ca²⁺) from outside-out patches expressing mSlo1 channels (Fig. 1 A). Steady-state inhibition of I_K was achieved within a fraction of a second of applying 100 μM Cu²⁺ to a patch (Fig. 1 B) and completely reversed by washing out Cu²⁺ with 5 mM EGTA (Fig. 1 A). Thus I_K inhibition does not reflect oxidation of BK channels by Cu²⁺, which reportedly proceeds with a 1-min delay in 100 μM Cu²⁺ and is not reversed by removal of Cu²⁺ (Morera et al., 2003). We did not observe any irreversible effects even after more than 20 min in 100 μM Cu²⁺.

Both outward current during a depolarization and inward tail current following the pulse were similarly reduced by 100 μM Cu²⁺ (Fig. 1 A), and the single channel current–voltage relation was unaffected (Fig. 1 C). Thus Cu²⁺ does not reduce I_K by acting like many intracellular divalent cations to rapidly block the BK channel pore and reduce single channel conductance in a voltage-dependent manner (Oberhauser et al., 1988; Ferguson, 1991). Instead, Cu²⁺ inhibits BK channel activation, shifting the half-activation voltage (V_0.5) of the G_K-V relation to more positive voltages by +82 mV and decreasing G_Kmax by ∼18% in 1000 μM Cu²⁺ as estimated by fits to Boltzmann functions (Fig. 1 D). Increasing [Cu²⁺] from 100 to 1000 μM produced no additional reduction in G_K (Fig. 1 D), indicating that 100 μM is a saturating concentration and further supporting that Cu²⁺ does not act by occluding the pore. I_K kinetics were also altered by 100 μM Cu²⁺ in a manner consistent with a positive shift in the G_K-V relation, slowing the activation time constant 1.6-fold at +220 mV (Fig. 1 E) and speeding deactivation 1.4-fold at −80 mV (Fig. 1 F).

The sensitivity of mSlo1 channels to Cu²⁺ is indicated by mean G_K-V relations measured in different [Cu²⁺], from 0 to 200 μM (Fig. 2 A). Currents were recorded in the presence of 3 μM intracellular Ca²⁺ to shift activation to more negative voltages and facilitate measurements in the presence of Cu²⁺. Similar to the results in 0 Ca²⁺, 100 μM Cu²⁺ shifted the G_K-V by +74 ± 1 mV and reduced G_Kmax by 27 ± 1% (Fig. 2 A). Normalized G_K-Vs in Fig. 2 B better illustrate the changes in V_0.5. I_K kinetics in 3 μM Ca²⁺ were altered by 100 μM Cu²⁺ more than in 0 Ca²⁺ (slowed 3.3 ± 0.1-fold at +220 mV, speeded 2.5 ± 0.2-fold at −80 mV). However, activation remained fast compared with the 30-ms voltage pulse duration (e.g., τ(I_K) = 3.1 ± 0.1 ms at +220 mV), indicating that G_K-V relations in Fig. 2 A should accurately represent the steady-state gating of Cu²⁺-bound channels.

Effects of Cu²⁺ on G_K were detectable at submicromolar concentrations and an approximate half-maximal response was observed in 2.5 μM Cu²⁺ (filled squares, Fig. 2, A and B). A dose–response relation determined by plotting the shift in half-activation voltage (ΔV_0.5) versus [Cu²⁺] (filled symbols, Fig. 2 C) is fit by a Hill equation (see Materials and methods) with an IC₅₀ of 1.97 ± 0.25 μM and Hill coefficient (n_H) of 1.01 ± 0.09. Dose–response curves were also obtained by plotting G_K-[Cu²⁺] relations at single voltages as illustrated in Fig. 2 C (open symbols) for 110, 150, and 190 mV. The IC₅₀ and Hill coefficient obtained by fitting G_K-[Cu²⁺] relations at different voltages are plotted in Fig. 2, D and E, respectively (filled symbols), and were similar to those derived from ΔV_0.5 (Fig. 2, D and E, dotted lines) with IC₅₀ values of 0.53–2.24 μM and n_H ranging from 0.9 to 1.2. Similar results were also obtained from G_K-[Cu²⁺] relations in 0 Ca²⁺ (Fig. 2, D and E, open symbols). The voltage dependence of IC₅₀ and n_H may reflect that the steepness of G_K-V curves was also reduced by Cu²⁺ as illustrated in Fig. 2 F by plotting the apparent charge from Boltzmann fits (z_APP) versus [Cu²⁺].

Several different mechanisms could potentially account for a shift in the G_K-V relation by Cu²⁺, including inhibition of either the closed to open conformational change or voltage sensor activation. To help distinguish among these possibilities we examined the effects of saturating 100 μM Cu²⁺ on steady-state open probability (P_O) over a wide voltage range in 0 or 5 μM intracellular Ca²⁺ and plotted the results on a semi-log scale in Fig. 3 A. P_O was measured to values as low as ∼10⁻⁷ by recording unitary current activity in macropatches (see Materials and methods). The mean log(P_o)-V relations in Fig. 3 A reveal several important features of Cu²⁺ action and suggest that Cu²⁺ acts in large part to inhibit voltage sensor activation.

At extreme negative voltage, log(P_o) achieves a weakly voltage-dependent limiting slope (Fig. 3 A, dotted lines) because BK channels can open even when voltage sensors are not activated (Horrigan and Aldrich, 1999; Horrigan et al., 1999). At these voltages, P_o reflects the intrinsic stability of the gate and is increased almost 100-fold by 5 μM Ca²⁺ (Fig. 3 A), illustrating that Ca²⁺ acts independent of voltage sensor activation to stabilize the open conformation (Horrigan and Aldrich, 2002). By contrast, 100 μM Cu²⁺ has little or no effect on P_o at extreme negative voltages in the presence or absence of Ca²⁺, implying that Cu²⁺ does not act either by stabilizing the closed conformation nor by interfering with Ca²⁺ action. Instead, Cu²⁺ reduces P_o in a voltage-dependent manner, with maximal effect at intermediate voltages. This response is unlikely to reflect an intrinsic voltage dependence to Cu²⁺ binding since if Cu²⁺ enters a binding site in the membrane electric field from the extracellular solution it should bind best at more negative voltages where the impact of Cu²⁺ on P_O is least. Therefore, the voltage dependence of Cu²⁺ action is likely to arise from interaction with the voltage sensor. The major effect of 100 μM Cu²⁺ can be approximated by shifting the log(P_o)–V relation along the voltage axis by +45 mV (Fig. 3 B), consistent with an inhibition of voltage sensor activation. This shift is less than the +74 mV shift in V_0.5 (Fig. 2 B) because Cu²⁺ also causes a slight reduction in the slope of the log(P_o)-V relation (Fig. 3 B).

Cu²⁺ could potentially inhibit voltage sensor activation by binding in a state-dependent manner to the voltage sensor to stabilize the resting conformation, similar to the proposed mechanism of Zn²⁺ action on squid K⁺ channels (Gilly and Armstrong, 1982a). If so, then Cu²⁺ should interact better with the resting than the activated state. To test this hypothesis, we examined the effect on BK channel activation of applying Cu²⁺ at different membrane potentials (Fig. 3, C and D). I_K was recorded in response to brief test pulses to +120 mV before and after application of 100 μM Cu²⁺ (Fig. 3 C) with [Ca²⁺]_i = 5 μM. Cu²⁺ was applied during a 500-ms prepulse immediately preceding the second test pulse. When Cu²⁺ was applied at prepulse voltages (V_PRE) that do not activate the channel (−80 to +40 mV), the steady-state current recorded during the second test pulse (I_P2) in the presence of Cu²⁺ was reduced to ∼40% of the control (I_P1) (Fig. 3 C, green traces). However, at more depolarized V_PRE the inhibitory effect of 100 μM Cu²⁺ was reduced, and no difference between I_P1 and I_P2 was observed at V_PRE ≥ +180 mV (Fig. 3 C, red traces). The dependence of Cu²⁺ inhibition on prepulse voltage was analyzed by estimating the fraction of channels not inhibited as f_NI = (I_P2[V_PRE] − I_P2[−80])/(I_P1 − I_P2[−80]) and plotting this quantity versus V_PRE (f_NI[100 Cu²⁺], Fig. 3 D). These data superimpose on the mean P_o-V relation obtained in 0 Cu²⁺ (P_O, Fig. 3 D), suggesting that Cu²⁺ interacts poorly and/or is poorly accessible to its binding site when channels are activated. Indeed the failure of traces in Fig. 3 C to converge to the same current level at the end of the second test pulse suggests that Cu²⁺ binding does not equilibrate within 30 ms at the test pulse condition (+120 mV, 100 μM Cu²⁺), favoring the possibility that access of Cu²⁺ to its binding site is slow in the activated channel.

It should be noted that the results in Fig. 3 (C and D) do not distinguish whether the state dependence of Cu²⁺ action reflects a failure to interact with the activated voltage sensor or the open conformation. In BK channels the voltage dependence of voltage sensor activation (Q-V relation) and P_o-V relation differ in 0 Ca²⁺ but are superimposable in saturating 70 μM Ca²⁺ (Horrigan and Aldrich, 2002). We did not measure the Q-V relation under the experimental conditions in Fig. 3 C (5 μM Ca²⁺). However, an allosteric gating scheme that reproduces the 0 and 70 μM Ca²⁺ data (Horrigan and Aldrich, 2002) predicts the half-activation voltages of Q-V and P_O-V in 5 μM [Ca²⁺] should differ by only 3 mV. Thus the similar voltage dependence of f_NI[100 Cu²⁺] and P_O in Fig. 3 D is compatible with a dependence of Cu²⁺ inhibition on either channel opening or voltage sensor activation.

To further characterize the metal sensitivity of mSlo1, we compared the effect on V_0.5 of different transition metals (Cu²⁺, Zn²⁺, Fe³⁺, Mn²⁺, Ni²⁺, Co²⁺, and Cd²⁺) at 100 μM (Fig. 4 A). Cu²⁺ produced by far the greatest shift in V_0.5 (+74.3 ± 2.7 mV), followed by Cd²⁺ (+19.0 ± 1.9 mV) and Zn²⁺ (+8.5 ± 1.5 mV). None of the other metals tested significantly altered V_0.5. These results indicate that, among the metals tested, the BK channel is primarily sensitive to Cu²⁺ at physiologically relevant concentrations.

The weak response of mSlo1 to metals other than Cu²⁺ at 100 μM suggests either that these metals bind to the channel with much lower affinity than Cu²⁺, or their occupancy of the binding site has little effect on gating. To distinguish these possibilities we examined the response to higher metal concentrations (Fig. 4, B and C). As Cd²⁺ was increased beyond 100 μM, G_K-V relations continued to shift to more positive voltages, producing a response in 4000 μM Cd²⁺ similar to that of 100 μM Cu²⁺ (Fig. 4 B). Thus Cd²⁺ can alter gating to the same extent as Cu²⁺, but at much higher concentrations. The dose–response relation for Cd²⁺ is similar in shape to that of Cu²⁺ but characterized by an IC₅₀ of 0.70 mM (Fig. 4 C). This IC₅₀ may be overestimated by approximately twofold owing to the ability of Cd²⁺ to complex with chloride (see Materials and methods), but is still more than two orders of magnitude greater than the IC₅₀ of Cu²⁺. Partial dose–response relations were also obtained for Zn²⁺ (IC₅₀ = 1.2 mM), Mn²⁺, and Ni²⁺ (Fig. 4 C), indicating that these metals are also capable of inhibiting BK channel activation at much higher concentrations than Cu²⁺. Therefore the metal binding site appears to be highly selective for Cu²⁺.

The dose–response curves for Mn²⁺ and Ni²⁺ in Fig. 4 C are steeper than those of Cu²⁺, Cd²⁺, or Zn²⁺, possibly reflecting a different site or mechanism of action or an additional contribution to G_K inhibition from surface charge screening at millimolar concentrations. Charge screening has little effect on BK channel activation in neutral bilayers (MacKinnon et al., 1989), and is estimated in cell membranes to shift V_0.5 by only ∼3 mV in response to an increase in divalent cation concentration from 1 to 10 mM, and ∼9 mV from 10 to 100 mM (Hu et al., 2006). Thus, surface charge effects should be small compared with the 75-mV G-V shift caused by Cu²⁺ and can only account for a fraction of the G_K decrease by Mn²⁺ and Ni²⁺.

To locate the Cu²⁺ binding site we mutated individual amino acid residues to identify those that influence the IC₅₀ for Cu²⁺. Any side chain containing oxygen, nitrogen, or sulfur atoms can potentially interact with metal ions. However Cys and His are by far the most common coordinating residues in transition metal binding sites (Rulisek and Vondrasek, 1998). mSlo1 contains three Cys (C14, C141, and C277) and one His (H254) that are potentially accessible to the extracellular solution. A triple mutant lacking all three Cys (C14A/C141A/C277A) exhibited a G_K-V relation and inhibition by 100 μM Cu²⁺ similar to the WT (Fig. 5 A). Likewise, H254A shifted the G_K-V by approximately +60 mV but exhibited a similar V_0.5 shift as the WT in response to 100 μM Cu²⁺ (Fig. 5 B). Dose–response relationships for H254A and C14A/C141A/C277A were indistinguishable from the WT (Fig. 5 C). Therefore, we conclude that native Cys and His residues do not participate in Cu²⁺ binding.

Acidic amino acids (Asp and Glu) are often important in coordinating metal ions. To test for the participation of such groups in Cu²⁺ binding, we examined the ability of 100 μM Cu²⁺ to inhibit G_K at different extracellular pH (Fig. 5 D). The reduction in G_K at +220 mV was maximal from pH 7.2 to 8.2, but inhibition was relieved under more acidic conditions and almost eliminated at the lowest pH tested (5.2). This relief of inhibition was not due to a change in channel gating because reducing pH from 7.2 to 5.2 produced only a small +20-mV shift in V_0.5. Thus the effect of pH on Cu²⁺ action likely reflects a decrease in Cu²⁺ binding at acidic pH.

The pH dependence of 100 μM Cu²⁺ action was fit by a Hill equation (Fig. 5 D, solid line) with half-relief of inhibition at pH 6.0 (n_H = 1.1), suggesting Cu²⁺ is coordinated by one or more protonatable side chains with a pKa near 6. Since we have already ruled out the participation of histidine, this suggests that acidic residues may be involved. Although the mean pKa values for Asp and Glu in proteins are 3.4 and 4.1, respectively, values >6 have been reported, usually in enzyme active sites and ligand-binding sites (Forsyth et al., 2002).

To identify acidic residues that coordinate Cu²⁺ we focused initially on the S2 and S3 segments of the voltage sensor domain because acidic residues in these segments of dEAG and hERG channels have been implicated in metal binding. An amino acid sequence alignment of dEAG and hERG with mSlo1 and several other voltage-dependent channels (Fig. 6 A) illustrates that dEAG and hERG contain an unusually large number of acidic residues in S2 and S3, three of which (boxed) contribute to divalent cation sensitivity (Silverman et al., 2004; Fernandez et al., 2005). Only one of these putative coordinating residues is conserved in other Kv channels, corresponding to D153 in the S2 segment of mSlo1. Mutation of D153 in mSlo1 greatly alters the dose–response curve for Cu²⁺ (Fig. 6 B). Substitution of Lys at this position (D153K) reduced Cu²⁺ sensitivity (increasing IC₅₀ eightfold to 15.9 μM), whereas a His increased Cu²⁺ sensitivity (decreasing IC₅₀ eightfold to 0.26 μM). In both cases, mutant dose–response curves were shifted relative to the WT (dotted curve) with little change in shape, consistent with a change in the affinity of a single binding site. Thus D153 in S2 appears to contribute to Cu²⁺ binding in BK channels.

Fig. 6 B illustrates two important principals that we used to identify potential Cu²⁺-coordinating sites. First, we measured dose–response curves for each mutant to obtain a quantitative indication of changes in Cu²⁺ sensitivity (IC₅₀), related to binding affinity. Measurements at a single Cu²⁺ concentration cannot distinguish changes in affinity from changes in efficacy; and efficacy can in principal be altered by modifications in gating that are not directly related to Cu²⁺ binding. We also measured IC₅₀ because we do not expect single mutations to abolish Cu²⁺ binding in an all or none fashion. The high sensitivity and selectivity of mSlo1 for Cu²⁺ suggests that Cu²⁺ may be coordinated by multiple residues, no one of which is critical for binding. In metalloprotein and small molecule binding sites of known structure, Cu²⁺ is usually coordinated by three to eight groups (Rulisek and Vondrasek, 1998). Second, at each site we compared the effects of multiple mutations, including those expected to inhibit or enhance binding. A correlation between changes in IC₅₀ and the expected ability of different side chains to interact with metals, as in Fig. 6 B, provides support that a candidate position interacts directly with Cu²⁺. The use of multiple substitutions also helped to control for effects of mutation on channel gating. Mutations in the voltage sensor often shift or alter the shape of the G_K-V relation (Ma et al., 2006). We controlled for G_K-V shifts to some extent by measuring dose–response curves for all mutants at their V_0.5 (Table I); and small shifts are not expected to greatly effect IC₅₀ determination since similar values were obtained for the WT at different voltages (Fig. 2 C). However, because Cu²⁺ action is state dependent (Fig. 3, C and D), large changes in channel gating can potentially influence IC₅₀. Therefore, when possible, we compared mutants at each site that exhibit very similar gating properties. One such case is D153, which has previously been identified as contributing to gating charge in mSlo1 (Ma et al., 2006). The mutations at this site substantially reduce gating charge and shift V_0.5 to more positive voltages, but the mutants in Fig. 6 B (D153K and D153H) have similar G_K-V properties as indicated by Boltzman fit parameters in Table I.

Although D153K substantially increased IC₅₀, it only slightly reduced the ability of 100 μM Cu²⁺ to inhibit G_K (Fig. 6 B), and therefore does not recapitulate the response of the WT to low pH. This suggests that additional acidic residues may be present in the Cu²⁺ binding site. mSlo1 contains a highly conserved Asp in S3 (D186) (Fig. 6 A). Mutation of D186 to Ala in mSlo1 has substantial effects on channel gating (Ma et al., 2006). However, D186A and D186H have similar gating properties (Table I) and indistinguishable dose–response curves (Fig. 6 C), implying that D186 is not involved in Cu²⁺ binding.

The only other acidic residue in the S1–S4 segments of mSlo1 that is potentially accessible to Cu²⁺ is D133, at the extracellular end of S1. This position is highly conserved in most voltage-gated channels (Fig. 6 A), and the residues equivalent to D133 and D153 contact each other in the crystal structure of Kv1.2 (Long et al., 2005b), suggesting a likely role for D133 in Cu²⁺ binding. Consistent with this prediction, D133A increased IC₅₀ almost 30-fold relative to the WT, whereas D133C and D133H decreased IC₅₀ 1.4- and 3-fold (Fig. 6 D). Mutations at this site have similar small effects on gating (Table I) (Ma et al., 2006), supporting a direct role for D133 in Cu²⁺ binding.

To further characterize the Cu²⁺-binding site, we mutated 10 additional positions in S1–S4 segments plus E139 in the S1–S2 linker (Fig. 6 A). We tested all potential metal-coordinating residues in transmembrane segments that are close to D133 or D153 based on primary sequence and the structure of Kv1.2, as well as loci in S2 or S3 that interact with metals in dEAG and hERG. The effects of different mutations on the IC₅₀ for Cu²⁺ are summarized in Fig. 7 A.

We attempted to inhibit Cu²⁺ binding by substituting Ala at each position (Fig. 7 A, open circles). In a few positions (D153, R207, and R210) where Ala mutations failed to express functional channels or expressed poorly, alternative “low-affinity” substitutions were used (Lys, Gln, and Asn). In many cases, mutants exhibited IC₅₀s indistinguishable from the WT (dotted line). At all positions tested in S1–S4 segments we also made “high-affinity” substitutions (His or Cys) in an attempt to enhance Cu²⁺ sensitivity. A total of six positions were identified where low-affinity substitutions increased IC₅₀, but only four of these showed a difference in IC₅₀ between low- and high-affinity substitutions, indicated by vertical bars in Fig. 7 A. In addition to D133 and D153, these putative Cu²⁺-coordinating sites are Q151 in S2 (Fig. 7 B) and R207 in S4 (Fig. 7 C). Two sites (R210 and Y130) showed decreases in Cu²⁺ sensitivity independent of the substituting residue. In the case of R210, this is likely to reflect an effect of mutation on gating as discussed below. Y130 could conceivably act as an electron-withdrawing group to enhance interaction of Cu²⁺ with the putative coordinating residue D133 located one helical turn away.

Substitution of putative Cu²⁺-coordinating sites with Cys or His produced a substantial eightfold increase in Cu²⁺ sensitivity in the case of D153H, but a more modest 1.4–4-fold increase at positions 133, 151, and 207 (Fig. 7 A). Such a result is not unexpected because Cu²⁺ is not strongly selective for Cys or His over other side chains. However, to confirm that D133, Q151, and R207 participate directly in metal coordination we also looked at their Cd²⁺ sensitivity. Cd²⁺ interacts strongly with Cys or His, therefore Cys or His substitutions should have a much greater effect on the sensitivity of the channel for Cd²⁺ than Cu²⁺. Consistent with this prediction, D133C, Q151H, and R207C mutations produced more than 1,000-fold decreases in the IC₅₀ for Cd²⁺ relative to the WT (Fig. 7 A, filled symbols). Indeed, these mutants were all more sensitive to Cd²⁺ than Cu²⁺. The ability of Cys or His substitutions to enhance Cd²⁺ sensitivity and switch the selectivity of the channel for Cd²⁺ versus Cu²⁺ supports that D133, Q151, and R207 participate directly in metal binding.

The identification of R207 in S4 as a potential Cu²⁺-coordinating residue is notable because this site plays an important role in voltage sensor activation (Ma et al., 2006) and because the presence of arginine in metal binding sites is rare. Therefore we investigated the contribution of R207 in more detail (Fig. 8).

The participation in metal binding of the Arg side chain, which is protonated and therefore positively charged at physiological pH, is unusual but not unprecedented (Bewley et al., 1999; Ferraroni et al., 2002; Berkovitch et al., 2004; Di Costanzo et al., 2006). Unlike Lys, the Arg side chain is bifurcated such that the proton is delocalized and two nitrogens are potentially available to interact with a metal ion. Nevertheless, two issues must be addressed to confirm the involvement of R207 in Cu²⁺ binding. First, the possibility that differences in IC₅₀ among mutant and WT channels reflect differences in gating rather than Cu²⁺ binding is of particular concern in the case of R207. Mutations of this site greatly facilitate voltage sensor activation (Ma et al., 2006), which could reduce Cu²⁺ sensitivity, given the state dependence of Cu²⁺ action (Fig. 3, C and D). Second, our analysis assumes that changes in IC₅₀ reflect modification of a native binding site; but the possibility must also be considered for any putative coordinating residue that high-affinity (Cys and His) substitutions modify Cu²⁺ sensitivity by introducing a new binding site.

The R207Q mutation shifts voltage sensor activation (i.e., the Q-V relation) to more negative voltages by >200 mV (Ma et al., 2006), such that the P_o-V relation for R207Q in 0 Cu²⁺ (Fig. 8 A) is left shifted and shallow compared with the WT (Table I). The possibility that this change in gating contributes to the increased IC₅₀ of R207Q relative to the WT is supported by the observation that mutations at position 210 (R210N and R210C), which also enhance voltage sensor activation (Ma et al., 2006), exhibit a P_o-V similar to R207Q (Table I) and increase IC₅₀ (Fig. 7 A). Thus the reduced Cu²⁺ sensitivity of R207Q cannot be taken as conclusive evidence that an Arg at this position normally interacts with Cu²⁺. That said, R207 is clearly in a position to interact with Cu²⁺ because the steady-state activation of R207Q and R207C are almost identical in the absence of Cu²⁺ (Fig. 8 A), implying that the 170-fold difference in IC₅₀ for these mutants (Fig. 7 A) must reflect a difference in Cu²⁺ binding rather than gating.

The assumption that R207 mutations modify a native binding site is supported by the dose–response relations of these mutants for Cu²⁺ (Fig. 7 C) or Cd²⁺ (Fig. 8 C). If a mutation enhances metal sensitivity by introducing a new binding site with effects on gating independent of the native site, then the mutant should exhibit a biphasic dose–response relation. In no case was this observed. For all four putative coordinating sites, Cu²⁺ dose–response relations were shifted by mutation with little change in shape and were well fit by Hill equations with n_H near 1. The G_K-[Cu²⁺] relations for R207Q and R207C at +180 mV were best fit by Hill coefficients of 0.75 ± 0.03 and 0.77 ± 0.07, respectively (Fig. 7 C). Similarly, R207C decreases the IC₅₀ for Cd²⁺ by more than four orders of magnitude relative to the WT (Fig. 7 A) with little effect on the shape of the dose–response curve (Fig. 8 C). Thus, the dose–response of R207C appears monotonic and saturated over the concentration range where R207Q or the WT respond to Cu²⁺ or Cd²⁺, respectively, consistent with modification of a single binding site. It is conceivable that a second phase of inhibition for R207C might exist at concentrations >100 μM but be difficult to detect from the dose–response curve because G_K at +180 mV is already reduced by >90% in 100 μM Cu²⁺ or Cd²⁺. However, we can show that this is not the case by comparing G_K-V relations for R207C in 100 and 1000 μM Cu²⁺ (Fig. 8 A) or Cd²⁺ (Fig. 8 D). In both cases there is no appreciable change in the G_K-V relation over this concentration range, indicating that the response of R207C is indeed saturated at 100 μM Cu²⁺ or Cd²⁺.

If Cu²⁺ inhibits BK channel activation by binding in a state-dependent manner to the voltage sensor, then mutations in the binding site may alter not only IC₅₀ but also efficacy. Efficacy can be defined as the extent to which a saturating concentration of Cu²⁺ shifts the G_K-V relation (ΔV_0.5MAX), and is ultimately determined by the relative affinity (i.e., K_ds) of Cu²⁺ for different states (Colquhoun, 1998).

Why do we expect mutations to alter efficacy, and what do such effects reveal about the interaction of Cu²⁺ with coordinating residues? If Cu²⁺ binds in a state-dependent manner then the sum its energetic interactions with coordinating groups must be state dependent. Thus Cu²⁺ interaction with some individual coordinating groups must be state dependent while others may be state independent. Mutations of a state-dependent group that presumably abolish interaction with Cu²⁺, such as the substitution of a coordinating residue by Ala, must alter Cu²⁺ efficacy because a state-dependent component of the net binding energy will be eliminated. Mutations that enhance interaction with Cu²⁺ are also likely to alter efficacy, but are not certain to do so unless the site only interacts with Cu²⁺ in a single state such as the resting state. By contrast, Ala substitution at a state-independent site should reduce apparent affinity (increase IC₅₀) without altering efficacy, a so-called pure binding effect (Colquhoun, 1998). A high-affinity substitution at such a site is also likely but not certain to leave efficacy unchanged unless coordination geometry is identical in different states.

The effects of mutation on Cu²⁺ efficacy for mSlo1 (Fig. 9 A) suggest that some cooridinating residues interact with Cu²⁺ in a state-dependent manner while others do not.Fig. 9 A plots ΔV_0.5MAX measured with saturating 1000 μM Cu²⁺ for various mutations of the four putative Cu²⁺-coordinating sites. Mutations in S2 (Q151, D153) that alter IC₅₀ (Fig. 7 A) have no effect on efficacy (Fig. 9 A), suggesting that S2 interacts with Cu²⁺ in a state-independent manner. By contrast, mutation of R207 (S4) or D133 (S1) had effects on efficacy that parallel their impact on IC₅₀ (Fig. 7 A). That is, mutations that increased apparent affinity (decreasing IC₅₀) also increased ΔV_0.5MAX, suggesting that D133 and R207 interact with Cu²⁺ in a state-dependent manner. Although the magnitude of ΔV_0.5MAX can potentially be influenced by the steepness of the G_K-V relation (Cui and Aldrich, 2000), mutants tested at each site had similar G_K-V shapes (Table I, Fig. 8 A), implying that variations in ΔV_0.5MAX do not reflect effects of mutation on gating. In addition, R207C exhibits a much larger Cu²⁺-dependent shift in the log(P_o)-V relation (Fig. 8 B) than the WT (Fig. 3 A), providing more direct evidence for an enhanced shift in voltage sensor activation for this mutant.

These results suggest that the relative position of side chains may change during voltage sensor activation, such that S1 and S4 coordinate Cu²⁺ better in the resting than the activated state, while S2 maintains a constant interaction with Cu²⁺. A speculative model that illustrates this principal is depicted in Fig. 9 B, and is consistant with our data, but does not represent a unique solution.

Three of the four putative Cu²⁺-coordinating residues in mSlo1 are charged (D133, D153, and R207) and highly conserved among voltage-gated K⁺, Na⁺, and Ca²⁺ channels (Fig. 6 A). This raises the possibility that many voltage-gated channels contain a metal binding site homologous to that in mSlo1. To test this possibility we examined the Cu²⁺ and Zn²⁺ sensitivity of Shaker K⁺ channels (Fig. 10).

Previous studies have shown that 300 μM Zn²⁺ slows the activation kinetics of Shaker and shifts steady-state activation to more positive voltages (Boland et al., 1994). As in the case of mSlo1, we found that inactivation-removed Shaker (WT) was on the order of 100-fold more sensitive to extracellular Cu²⁺ than Zn²⁺ (Fig. 10, A–D). At −10 mV, the half-time to steady-state activation (τ_0.5) of I_K was slowed approximately threefold and 15-fold in 5 μM and 20 μM Cu²⁺, respectively, and 100 μM Cu²⁺ inhibited the current during a 50-ms pulse almost completely (Fig. 10 A). G_K-V relations (Fig. 10 B) were shifted progressively to more positive voltages by 5, 20, and 100 μM Cu²⁺ and exhibited a decrease in G_Kmax similar to mSlo1 (Fig. 2 A). Zn²⁺ had effects on Shaker activation kinetics (Fig. 10 C) and the G_K-V relation (Fig. 10 D) that were qualitatively similar to those of Cu²⁺ but occurred at much higher concentrations. Thus metal binding in Shaker, as in mSlo1, appears very selective for Cu²⁺ over Zn²⁺.

Although Shaker can respond to low μM Cu²⁺ (e.g., 5 μM) it is evident from the small response to 1 μM Cu²⁺ (Fig. 10, A and B) that Shaker is less sensitive to Cu²⁺ than mSlo1. The G_K-V shift of Shaker by 100 μM Cu²⁺ (64 mV, Fig. 10 B) may be overestimated because the effect of Cu²⁺ on activation kinetics prevented achievement of steady-state activation during a 50-ms pulse at all but the most positive voltages tested. Given this caveat, the IC₅₀ of Shaker approximated by fitting V_0.5-[Cu²⁺] from Fig. 10 B with n_H = 1 was 18 μM. This difference compared with the BK channel is consistent with the fact that mSlo1 contains a putative coordinating residue Q151 that is not conserved in Shaker, and that the Q151A mutation increases mSlo1's IC₅₀ to 19.7 μM (Fig. 7 A).

To demonstrate that the metal binding site in Shaker is homologous to that in BK channels we examined the effects on Cu²⁺ and Zn²⁺ sensitivity of mutating S1 (E247) and S4 (R365) residues, corresponding to two of the three conserved coordinating sites in mSlo1 (Fig. 10, E–J). We tested for effects on activation kinetics since the kinetic changes complicated determination of steady-state activation. Mutation of E247 (D133 in mSlo1) to Ala (E247A) almost abolished the response to 5 μM Cu²⁺ and 1000 μM Zn²⁺ (Fig. 10, E and G), whereas E247C enhanced sensitivity to both Cu²⁺ (Fig. 10 F) and Zn²⁺ (Fig. 10 H). A similar result for Cu²⁺ was observed when R365 in S4 (R207 in mSlo1) was mutated to Ala (Fig. 10 I) or Cys (Fig. 10 J). In addition, similar to mSlo1, we observed that the response of WT Shaker to 5 μM Cu²⁺ was abolished at extracellular pH 5.2 (unpublished data). We did not investigate E283 in S2 (equivalent to D153 in mSlo1), because mutations at this position in Shaker expressed poorly in outside-out patches. Thus, we confirmed that two of the conserved Cu²⁺-coordinating residues in mSlo1 also contribute to the Cu²⁺/Zn²⁺ sensitivity of Shaker.

This study describes a previously unreported, rapid, and reversible inhibitory effect of extracellular Cu²⁺ on BK channel (mSlo1) activation at low micromolar and submicromolar concentrations. Cu²⁺ acts primarily to shift the voltage dependence of steady-state activation, together with lesser effects on G_Kmax and I_K kinetics. The G_K-V relation is shifted by up to 75 mV (ΔV_0.5) with an IC₅₀ of 2.0 μM Cu²⁺ and a Hill coefficient of 1.0 under our experimental conditions, and G_K can be reduced by >90% at voltages less than V_0.5. Thus BK channels are members of a small group of voltage-dependent channels that are extremely sensitive to physiologically relevant concentrations of Cu²⁺; and mSlo1 may be the most Cu²⁺-sensitive K⁺ channel identified so far.

Biophysical and mutagenesis results indicate that Cu²⁺ binds in a state-dependent manner to the voltage sensor domain where it interacts with transmembrane segments S1, S2, and S4, including acidic and basic residues that are highly conserved among most voltage-gated channels including Shaker. Native Cys and His residues do not contribute to the Cu²⁺ sensitivity of mSlo1. Likewise, native cysteines are not required for the Zn²⁺ sensitivity of Shaker (Boland et al., 1994). The binding site appears very selective for Cu²⁺ over closely related transition metals (Zn²⁺, Cd²⁺, Mn²⁺, Fe³⁺, Co²⁺, and Ni²⁺). Thus the voltage sensor can form a selective metal binding site with micromolar apparent affinity without requiring Cys or His residues that are often critical for the extreme metal sensitivity of other channels. We demonstrate here that Shaker K⁺ channels can also respond to low micromolar Cu²⁺ and that some of the putative coordinating sites in mSlo1 are important for the Cu²⁺ and Zn²⁺ sensitivity of Shaker. Thus it is possible that our results define a metal binding site of general importance in voltage-gated channels.

Despite extensive investigation of metal coordination in enzymes, relatively little is known about metal binding sites in ion channels outside of the permeation pathway. In many cases, individual residues critical for the extreme sensitivity of particular channel subtypes have been identified, such as Cav3.2 and Nav1.5 (Satin et al., 1992; Jeong et al., 2003). In some cases, metal bridges that coordinate Cd²⁺ or Ni²⁺ between a pair of Cys and/or His residues have been defined in native channels (Gordon and Zagotta, 1995) or introduced by mutation (Holmgren et al., 1998; Webster et al., 2004). However, in few cases have potential coordinating residues been screened in sufficient detail to characterize a multivalent binding site. Importantly, we screened many residues in mSlo1, making multiple substitutions at most sites to inhibit and enhance Cu²⁺ sensitivity and to control for effects of mutations on gating. To confirm the direct participation of some positions in metal binding we also demonstrated effects of Cys or His substitution on Cd²⁺/Cu²⁺ selectivity. In all, we identified four putative coordinating residues and 14 residues that do not contribute to Cu²⁺ sensitivity; both findings place constraints on the possible structure of the binding site. This detailed characterization together with the fact that putative coordinating sites are in transmembrane segments of the voltage sensor and in most cases highly conserved among Kv channels allows us to use Kv channel structures as a framework for understanding what the binding site may look like and how it may change with gating. These and other implications of our results are discussed below.

One reason that BK channels have not previously been recognized as extremely metal sensitive probably reflects their Cu²⁺ selectivity. Unlike Cav3.2, which is similarly sensitive to a number of transition metals, with IC₅₀s for Cu²⁺, Zn²⁺, and Ni²⁺ of 0.9, 2.1, and 4.9 μM (Kang et al., 2006), the BK channel responds poorly to metals other than Cu²⁺. Another factor that may have masked the effect of Cu²⁺ is the state dependence of its action. Contrary to our results, a previous study failed to observe an acute response of single BK channels from rat skeletal muscle to 1–100 μM Cu²⁺ (Morera et al., 2003). This difference is unlikely to reflect that Morera et al. used a native channel reconstituted in lipid bilayers because they reported observing similar preliminary results with cloned hSlo1 channels expressed in Xenopus oocytes; the same preparation used here with a Slo1 homologue that is virtually identical to mSlo1. However, Morera et al. tested for the effect of Cu²⁺ under steady-state recording conditions where channels were maximally activated with P_o ∼0.9 before Cu²⁺ was applied (V = +40 mV, [Ca]_i = 60 μM). It is possible that these recording conditions prevented the acute effect of Cu²⁺ since mSlo1 channels that are maximally activated for 0.5 s are also insensitive to 100 μM Cu²⁺ (Fig. 3, C and D).

The IC₅₀ for Cu²⁺ inhibition of mSlo1 G_K varied with voltage (0.53–2.24 μM, Fig. 2 D) but was in any case comparable to that reported previously for the most Cu²⁺-sensitive Kv channel Kv1.3 (IC₅₀ = 5.3 ± 0.5 μM) (Teisseyre and Mozrzymas, 2006) or the K2P “leak” channels Task-3 (IC₅₀ = 2.7 ± 0.4 μM) and Trek-1 (EC₅₀ = 3.0 ± 1.0 μM) (Gruss et al., 2004). However, the Cu²⁺ dose–response relation for mSlo1, characterized by a Hill coefficient ∼1.0, is much shallower than that of Kv1.3 (n_H = 3.8), Trek-1 (n_H = 1.7 ± 0.8), or Task-3 (n_H = 1.8 ± 0.4). Therefore, with the caveat that these channels have not been studied under identical conditions, the BK channel is potentially the most Cu²⁺-sensitive K⁺ channel in that it has a lower threshold for response to Cu²⁺ (submicromolar), owing to its shallow dose–response relation.

We identified D153 as a coordinating site in mSlo1 because the corresponding residue in hERG is important for Cd²⁺ sensitivity (Fernandez et al., 2005). However, additional acidic coordinating sites in the S2 and S3 segments of hERG and dEAG are not conserved in mSlo1 (Fig. 6 A), suggesting the binding site in hERG and dEAG may be distinct from that in Slo1. Our results provide further support for this conclusion. First, we tested sites in S2 and S3 close to the positions that are important in hERG/dEAG and none were important for Cu²⁺ sensitivity in mSlo1. This included substituting Cys and Ala residues at a position in S2 (N158) that corresponds to a coordinating residue in hERG and dEAG. Conversely, the coordinating sites we identified in mSlo1 are largely absent from hERG or dEAG. The equivalent of D133 and Q151 in mSlo1 are aliphatic (I, V) in dEAG and hERG, and R207 is replaced by a lysine. Thus the binding sites of mSlo1 and dEAG may be distinct while mSlo1 and hERG share only a single acidic residue in S2.

The physiological importance of BK channel inhibition by Cu²⁺ is unknown. However, it is certainly intriguing that Slo1 appears poised to respond specifically to Cu²⁺ at concentrations estimated to be achieved in the synaptic cleft (Hopt et al., 2003). BK channel activation in nerve terminals is important for limiting action potential duration, Ca²⁺ entry, and neurotransmitter release (Faber and Sah, 2003). BK channels and Cu²⁺ are both present in glutamatergic synapses in the hippocampus (Kozma et al., 1981; Mathie et al., 2006; Misonou et al., 2006; Schlief and Gitlin, 2006). Thus the possibility exists that Cu²⁺ release could exert a positive feedback effect to enhance neurotransmitter and Cu²⁺ release by inhibiting BK channels in the same nerve terminal. Whether or not this could occur may depend on several factors including the metal sensitivity of Ca²⁺ channels in the nerve terminal as well as the time course of Cu²⁺ elevation. Because BK channels are less sensitive to Cu²⁺ when fully activated than at rest, it is possible they would respond less to release during an action potential than residual Cu²⁺ from preceding synaptic events. Another factor that could conceivably be important to the physiological response of BK channels is the effect of accessory β subunits. However, preliminary experiments indicated no appreciable effects on Cu²⁺ sensitivity of coexpressing mSlo1 with an excess of bovine β₁ (IC₅₀ = 2.8 μM, n_H = 0.84, +220 mV, 0 Ca²⁺) or human β₄ (IC₅₀ = 2.7 μM, n_H = 0.95, +240 mV, 0 Ca²⁺).

Shaker K⁺ channels were found to be roughly 10-fold less sensitive to Cu²⁺ than mSlo1 in terms of IC₅₀. However the response of Shaker could also be physiologically relevant because, unlike mSlo1, Shaker activation kinetics were greatly slowed by Cu²⁺. For example, 5 μM Cu²⁺ slowed activation kinetics threefold at −10 mV (Fig. 10 A) and might therefore greatly reduce the response to a brief action potential even though the effect of this concentration on steady-state activation was modest.

Several lines of evidence suggest that Cu²⁺ binds to the voltage sensor in a state-dependent manner to inhibit voltage sensor activation. First, Cu²⁺ shifts steady-state activation (e.g., V_0.5) to more positive voltages, consistent with a shift in voltage sensor activation. Conversely, Cu²⁺ does not affect P_o at extreme negative voltages in the presence or absence of intracellular Ca²⁺ (Fig. 3 A), indicating that Cu²⁺ has little or no direct effect on channel opening or Ca²⁺-dependent activation. Second, mutagenesis results define a site of action in the voltage sensor domain (Fig. 7). Third, mutation of some putative coordinating sites (D133 and R207) alter Cu²⁺ efficacy (ΔV_0.5MAX), indicating that Cu²⁺ action does not depend merely on occupancy of its binding site, and consistent with a state-dependent interaction that favors Cu²⁺ binding to the resting state (Fig. 9). Fourth, the insensitivity of fully activated channels to Cu²⁺ (Fig. 3 B) also implies a state-dependent change in the affinity and/or access of the binding site. Finally, the dose–response relation for Cu²⁺ can be reproduced by a model that assumes Cu²⁺ binds in a state-dependent manner to the voltage sensor, as discussed below.

To a first approximation, the effect of Cu²⁺ on steady-state activation can be reproduced by a gating scheme illustrated in Fig. 11 A (Scheme 1), which assumes that Cu²⁺ binds with higher affinity to the resting than activated state of the voltage sensor. The best fit of this model to the ΔV_0.5-[Cu²⁺] relation (Fig. 2 C, thick dotted curve) is almost identical to the Hill equation fit (solid curve) and yields Cu²⁺ dissociation constants for resting and activated states of 0.75 μM and 6.03 μM, respectively (see Fig. 2 legend for parameters).

Scheme 1 represents an extension of a well-established model that describes the effects of voltage and Ca²⁺ on BK channel activation (HA model) (Horrigan and Aldrich, 2002). The HA model (colored black in Scheme 1) asserts that a channel can undergo a closed to open (C-O) conformational change that is allosterically regulated by four independent and identical voltage sensors and Ca²⁺ sensors. Voltage sensors in each subunit equilibrate between resting (R) and activated (A), and Ca²⁺ sensors equilibrate between Ca²⁺ free (X) and Ca²⁺ bound (XCa²⁺). Equilibrium constants for channel opening, voltage sensor activation, and Ca²⁺ binding are L, J, and K, respectively, where L and J are voltage dependent with partial charges z_L and z_J, and K = [Ca²⁺]/K_D. The coupling between voltage sensor and gate is represented by an allosteric factor D such that the C-O equilibrium constant increases D-fold for each activated voltage sensor. Interactions of Ca²⁺ binding with the gate and voltage sensor are represented by allosteric factors C and E, respectively. Scheme 1 also assumes that each voltage sensor can equilibrate between Cu²⁺ free (Y) and Cu²⁺ bound (YCu²⁺) with equilibrium constant K_Cu = [Cu²⁺]/K_dCu in the resting state and E_CuK_Cu in the activated state. E_Cu is an allosteric factor analogous to that describing the interaction between of Ca²⁺ and voltage sensors (E). The steady-state open probability predicted by Scheme 1 is defined by Eq. 1 (see next page).

Scheme 1 requires only an 8.0-fold difference in affinity between resting and activated states (E_cu = 0.124) to account for the G_K-V shift by Cu²⁺. Thus, the estimated K_ds of both the resting and activated states are much less than 100 μM, suggesting that the inability of 100 μM Cu²⁺ to inhibit fully activated channels in Fig. 3 C may reflect poor access to the binding site rather than a weak interaction with the activated state. This conclusion is supported by the observation that mutation of sites in S2 (Q151 and D153) altered IC₅₀ but not efficacy (ΔV_0.5MAX, Fig. 9 A), suggesting these residues can interact with Cu²⁺ in the activated state.

Although Scheme 1 reproduces the ΔV_0.5-[Cu²⁺] relation it does not account for all effects of Cu²⁺. Scheme 1 predicts that the G_K-V in 3 μM Ca²⁺ should shift without decrease in slope (Fig. 2 F, curve A) or G_Kmax. However we observed that 100 μM Cu²⁺ reduced G_Kmax by 27% and reduced the slopes of the G_K-V and log(P_o)-V relations by 39% and 12%, respectively. Several mechanisms that could account for these effects are discussed below, but cannot currently be distinguished based on our data.

One mechanism that cannot account for effects of 100 μM Cu²⁺ on G_Kmax and G_K-V shape is the possibility raised by Fig. 3 C that Cu²⁺ cannot equilibrate with its binding site during a 30-ms depolarization. First, it is important to note that Cu²⁺ is equilibrated with the resting channel by the start of each voltage pulse, since the effect of 100 μM Cu²⁺ reaches a steady state within 1 s of application at −80 mV (Fig. 1 B) while the interval between pulses is 5 s. Second, 100 μM Cu²⁺ has a maximal effect on V_0.5 (Fig. 1 D and Fig. 2 C), implying that channels have a maximal number of Cu²⁺ bound when the G_K-V is measured, regardless of when or how fast Cu²⁺ reaches its binding site. Thus we conclude that decreases in G_Kmax and G-V slope in 100 μM Cu²⁺ reflect an equilibrium gating property of Cu²⁺-bound channels.

Although Cu²⁺ equilibration kinetics cannot account for changes in G_K-V shape observed in saturating Cu²⁺, they can affect the predictions of Scheme 1 and our estimate of K_dCu. To show this, we considered the extreme case that Cu²⁺ equilibrates at the holding potential but cannot bind or dissociate during a voltage pulse. Thus, the probability that a voltage sensor is Cu²⁺ bound (P_Cu[Cu²⁺]) is determined only by the K_d of the resting state (K_dCu), and the distribution of channel species with different numbers of Cu²⁺ bound (i = 0–4) is determined by a binomial distribution where each species has a different open probability (P_O[i]) specified by Scheme 1. Mean open probability in this case is given by Eq. 2 (see below).

Eq. 2 can fit the ΔV_0.5-[Cu²⁺] relation (Fig. 2 C, thin dotted curve) with a value of E_Cu = 0.129 similar to that determined from Eq. 1 (0.124). However, the value of K_dCu from Eq. 2 (2.1 μM) is comparable to the IC₅₀ and therefore greater than predicted by Eq. 1 (0.75 μM). An interesting property of Eq. 2 is it predicts a decrease in G_K-V slope (z_APP) at intermediate [Cu²⁺] (Fig. 2 F, curve B). However, this slope decrease is a consequence of heterogeneity in the number of Cu²⁺ bound to each channel rather than a property of any one species (P_o[i]), and therefore does not persist in saturating Cu²⁺.

The decreased slope of the log(P_o)-V relation observed in saturating Cu²⁺, according to the HA model, can only be caused by a decrease in gating charge or a decrease in voltage sensor/gate coupling (D-factor) (Ma et al., 2006). Either of these mechanisms is possible. One of the putative coordinating sites for Cu²⁺, D153, is thought to be a voltage-sensing residue in Slo1 in part because a charge reversal at this position (D153K) reduces the slope of log(P_o)-V by almost 50% (Ma et al., 2006). If Cu²⁺ remains associated with D153 during voltage sensor activation it might in effect reverse the net charge of this residue and reduce gating charge. It would also be possible for Cu²⁺ to reduce voltage sensor/gate coupling if its binding depends on the open state of the channel as well as the activation state of the voltage sensor. Intracellular ligands in Slo1 are known to have such effects on voltage sensor/gate coupling (Horrigan et al., 2005; Horrigan and Ma, 2008).

The decrease in G_Kmax by Cu²⁺ was observed at all [Ca²⁺] tested (0–10 μM) and is not caused by a decrease in single channel conductance (Fig. 1 C) or a failure to measure steady-state activation since Cu²⁺ had a relatively minor effect on I_K kinetics in mSlo1 (Fig. 1 F). These results cannot be accounted for by Scheme 1 or the HA model because any mechanism that decreases the C-O equilibrium constant (L) to reduce G_Kmax in 0 Ca²⁺ should be overcome by the ability of Ca²⁺ to increase L more than 100-fold at 5–10 μM concentrations (Horrigan and Aldrich, 2002). That Cu²⁺ reduced G_Kmax by a similar amount in different [Ca²⁺]_i suggests that a fraction of channels enter a nonconducting state that is not included in our model. Metal binding to the pore domain of Kv1.5 has been proposed to reduce G_Kmax by stabilizing a slow inactivated state (Kehl et al., 2002). Such a mechanism seems unlikely since Slo1 is not known to inactivate in the absence of β subunits. However, BK channels have been reported to enter brief “flicker” closed states that are not included in the HA model and limit the maximum open probability of BK channels to ∼0.9 in all [Ca²⁺]_i(Rothberg and Magleby, 1998). Therefore it is conceivable such states could be stabilized by Cu²⁺ to reduce G_Kmax in a Ca²⁺-independent manner.

The detailed structure of the BK channel is unknown. However, the putative Cu²⁺-coordinating residues in mSlo1 are for the most part conserved in other voltage-gated channels, including Kv channels of known structure (Fig. 6 A). We have also confirmed that S1 and S4 residues, corresponding to D133 and R207 in mSlo1, are important for the Cu²⁺ and Zn²⁺ sensitivity of Shaker (Fig. 10). Thus it is likely we can gain insight into the properties of this site by examining the structure of Kv channels.

The crystal structures of several Kv channels have been solved in the activated conformation (Jiang et al., 2003; Long et al., 2005a; Long et al., 2007). In Fig. 11 (B and C) we mapped the putative Cu²⁺-coordinating residues from mSlo1 onto the highest resolution example, a Kv1.2/Kv2.2 chimera (Long et al., 2007), based on the sequence alignment in Fig. 6 A. Although our results suggest this structure in the activated conformation should not be ideal for Cu²⁺ binding, it does provide some important insight. First, a cavity is present between the S1–S4 transmembrane segments that contains water molecules (dotted spheres) and therefore can presumably accommodate metal ions. Second, the three most highly conserved putative Cu²⁺-coordinating residues in S1(D133), S2(D153), and S4(R207) face each other in the crystal structure to form what looks like a reasonable binding pocket (Fig. 11 C). Although the fourth putative Cu²⁺-coordinating residue (Q151) faces away from the others in the crystal structure (toward lipid), it is conceivable that S2 in mSlo1 might adopt a different orientation to allow both Q151 and D153 to participate in Cu²⁺ binding (as illustrated in Fig. 9 B). Third, all four putative coordinating residues are distributed around the cavity, near the top of the voltage sensor, and roughly define a plane when viewed from the side (Fig. 11 B). This distribution may help explain the Cu²⁺ selectivity of the BK channel since the most favored coordination geometry for Cu²⁺ is square planar (Rulisek and Vondrasek, 1998). That many sites above and below the putative coordinating residues did not contribute to Cu²⁺ sensitivity in mSlo1 (backbone colored dark blue) is also consistent with a planar coordination geometry. However it should be noted that other than Cys, His, and acidic residues, extracellular loops were not tested in detail and could potentially participate in metal binding through backbone interactions. Finally, the acidic coordinating residues D133 and D153 are present on antiparallel helices (S1 and S2) and contact each other (in Kv1.2), consistent with the possibility that their pKa's are perturbed to account for the pH sensitivity of Cu²⁺ action (Fig. 5 D). Such an interaction would tend to increase the pKa of one partner and decrease the other. Additional factors that could increase pKa include the presence of D133 at the C terminus of a helix and the location of these residues near an aqueous crevice (Forsyth et al., 2002).

Our results suggest that Cu²⁺ should bind better in the resting than the activated conformation. Therefore it is reasonable to expect that the relative position of the four putative coordinating residues may change during activation. Recent studies based on a variety of approaches suggest that S4 may undergo a large motion relative to S1 and S2 during activation in Kv channels (Campos et al., 2007; Long et al., 2007; Pathak et al., 2007). Such a motion is qualitatively consistent with the formation of a state-dependent binding site between S1, S2, and S4. However, it should be noted that this gating motion is likely larger in Kv channels than in BK channels, which are weakly voltage dependent. Indeed we have suggested previously that the resting conformation of mSlo1 may be similar to an intermediate activated state of Shaker (Ma et al., 2006). It is possible that Cu²⁺ slowed the activation of Shaker (Fig. 10 A) more than mSlo1 (Fig. 1 E) by stabilizing an intermediate state in Shaker. However another reason for this difference is that voltage sensor activation is rate limiting for I_K activation in Shaker, whereas the closed–open transition is rate limiting in Slo1 (Horrigan and Aldrich, 2002). Thus a slowing of voltage sensor activation by Cu²⁺ could increase the time constant of I_K activation in Shaker while altering primarily the delay in activation of Slo1 (Horrigan and Aldrich, 1999).

This work was supported by a grant from the National Institutes of Health (NS42901) to F.T. Horrigan. An abstract of this work was presented at the 50th Annual Meeting of the Biophysical society (2006).

Olaf S. Andersen served as editor.