Divalent Cations Regulate Connexin Hemichannels by Modulating Intrinsic Voltage-dependent Gating

It is now evident that connexin hemichannels (connexons) are functional when unapposed or undocked, operating in the plasma membrane like conventional ion channels. Analogous to their cell–cell or gap junction channel counterparts, hemichannels constitute large ion channels that are capable of mediating fluxes of a variety of signaling molecules. Reported functions of hemichannels include mediating the release of glutamate and ATP (Stout et al., 2002; Ye et al., 2003; Zhao et al., 2005; Genetos et al., 2007; Schock et al., 2008), the spread of Ca²⁺ waves (Weissman et al., 2004), and negative feedback from horizontal cells to cones in the vertebrate retina (Kamermans et al., 2001; Kamermans and Fahrenfort, 2004). Recently, hemichannels formed of the functionally related pannexin genes were reported to act as ATP release channels when coupled to purinergic P2Y receptors (Locovei et al., 2006; Pelegrin and Surprenant, 2006; Huang et al., 2007) and to contribute to neuronal excitotoxicity after ischemia, suggesting a role in mediating loss of cell metabolites and/or entry of extracellular ions during stroke (Thompson et al., 2006).

Due to the large size of the aqueous pore, hemichannel opening in the plasma membrane ought to be highly regulated. A relatively small number of open hemichannels can have a large impact on a cell's electrical properties, and unregulated opening of hemichannels can be deleterious to a cell's integrity due to excessive loss of intracellular constituents and influx of extracellular ions. Initial studies of hemichannels expressed in Xenopus oocytes were those of Cx46 and showed large, time and voltage-dependent currents that activated with depolarization; extracellular divalent cations strongly inhibited the currents and shifted activation to more positive potentials (Ebihara and Steiner, 1993). Subsequent studies of hemichannels formed of a variety of connexins have shown qualitatively similar properties, i.e., voltage and extracellular divalent cations are potent regulators with hyperpolarization and elevated levels of divalent cations generally promoting closure. However, it was reported that voltage-dependent gating in Cx46 hemichannels is lost in solutions containing no added divalent cations, suggesting that divalent cations are required for gating (Ebihara et al., 2003). Similarly, loss of gating was reported in Cx37 hemichannels upon removal of divalent cations from the bath solution and a model was proposed whereby extracellular divalent cations produce voltage-dependent block by traversing the pore and binding to a site located at the cytoplasmic end of the hemichannel (Puljung et al., 2004). In both cases, it was concluded that there is no intrinsic voltage gate and that voltage dependence can be explained solely by divalent cation block/unblock.

Thus far, studies reporting loss of voltage gating in hemichannels have relied solely on macroscopic recordings of connexins expressed in Xenopus oocytes where there is the possibility of series access resistance errors (Baumgartner et al., 1999) resulting from the large membrane conductances that typically accompany removal of divalent cations. These studies also did not consider the complexity of connexin hemichannel gating, specifically that there are two principal types of gating, one that closes hemichannels to a long-lived substate, termed V_j or fast gating, and one that closes them fully, termed loop or slow gating (Bukauskas and Verselis, 2004). V_j gating is the mechanism originally described in gap junction channels in response to transjunctional voltages, V_j. Loop gating was named provisionally because of the plausible involvement of the extracellular loops in closing hemichannels (Trexler et al., 1996). Both forms of gating appear to be properties that are common to all connexin channels and hemichannels (Bukauskas and Verselis, 2004).

Here we sought to examine the relationship between divalent cations and both V_j and loop gating by making use of single hemichannel recordings. We focused on Cx46 hemichannels in which V_j and loop gating have been shown to operate at opposite voltage polarities, positive and negative, respectively (Trexler et al., 1996). Also, single hemichannel recordings permitted unambiguous distinction between V_j and loop gating events, circumvented series access resistance problems, and allowed examination of hemichannels in isolation without contaminating endogenous Xenopus channels. We demonstrate that both forms of voltage gating are intrinsic voltage-dependent mechanisms in Cx46 hemichannels and that the action of external divalent cations is inconsistent with open channel block. Rather, divalent cations act as modifiers of intrinsic gating, selectively acting on loop gating by promoting stable, long-lived closures with membrane hyperpolarization.

Cx46 DNA was cloned from rat genomic DNA using PCR amplification with primers corresponding to amino- and carboxy-terminal sequences as described previously (Trexler et al., 1996). Synthesis of RNA and preparation and injection of oocytes have been described previously (Trexler et al., 1996, 2000). Injected oocytes were kept at 18°C in a modified ND96 solution containing (in mM) 88 NaCl, 1 KCl, 1 MgCl₂, 1.8 CaCl₂, 5 glucose, 5 HEPES, 5 pyruvate pH 7.6.

To record macroscopic hemichannel currents, Xenopus oocytes were placed in a polycarbonate RC-1Z recording chamber (Warner Instruments) with a slot-shaped bath connecting inflow and outflow compartments to allow for rapid perfusion. A suction tube was placed in the outflow compartment and a separate reservoir was used for grounding, connected to the main chamber with an agar bridge. At the start of each experiment, oocytes were bathed in the modified ND96 solution (MND96). Perfusion solutions consisted of MND96 to which Ca²⁺ and or Mg²⁺ concentrations were adjusted to desired levels. 0 Ca²⁺ and 0 Mg²⁺ solutions refer to nominal conditions in which no CaCl₂ or MgCl₂ was added. I-V curves at each concentration of Ca²⁺ and Mg²⁺ were obtained by applying slow voltage ramps (3 min) from +50 to −100 mV from a holding potential of −40 mV. The data were acquired at 1 kHz. Conductance was calculated as I/(V_m − V_rev), where V_rev is the reversal potential. For the G-V plots, the digitized data were decimated and plotted as points; data points around V_rev, which represents a discontinuity, were discarded, giving the curves a continuous appearance. All recordings were obtained with a GeneClamp 500 two-electrode voltage clamp (Axon Instruments). Both current-passing and voltage-recording pipettes contained 1 M KCl. To record single hemichannel currents, Xenopus oocytes were manually devitellinized in a hypertonic solution consisting of (in mM) 220 Na aspartate, 10 KCl, 2 MgCl₂, 10 HEPES and then placed in MND96 solution for recovery. Oocytes were then individually moved to the recording chamber containing the patch pipette solution (IPS), which consisted of (in mM) 140 KCl, 1 MgCl₂, 5 HEPES, 1 CaCl₂, 3 EGTA, and pH adjusted to 7.6 with KOH. The recording compartment was connected via a 3 M agar bridge to a ground compartment. To examine the effects of divalent cations, the Mg²⁺ concentration was altered. For experiments using excised patches, changes in the bath solution for any given patch were accomplished through an inflow tube. Bath volume was ∼0.3–0.4 ml and total volume exchange occurred within 5–10 s. Using MaxChelator (http://maxchelator.stanford.edu), free Ca²⁺ and Mg²⁺ levels in our IPS solution were calculated to be ∼2 nM and 1 mM, respectively. IPS solutions in which Mg²⁺ was omitted or raised to 10 mM did not appreciably change free Ca²⁺.

Single channel records from voltage steps and ramps were leak subtracted by measuring leak conductance of a given patch from full closing transitions and extrapolating linearly with voltage. Single hemichannel I-V curves were obtained by applying 8-s voltage ramps from −70 to +70 mV. To measure open probability in patches containing single hemichannels, currents from patches held at a constant voltage were idealized using the Sublevel Hinkley Detector, SHD (Draber and Schultze, 1994; Trexler et al., 1999). The SHD was set to consider only two levels, fully open and closed, which results in the SHD acting as a half-height threshold algorithm. The SHD was set at low sensitivity to avoid detection of brief, incomplete, noisy events from the closed state. Thus, the closed dwell times incorporated not only the time spent fully closed, but also the time spent flickering around the closed state. Dwell times in the two states were compiled and Po was calculated by summing the dwell times in the open state and dividing by the total recording time. All currents in cell-attached and excised-patch configurations were recorded using an Axopatch 1D amplifier (Axon Instruments Inc.). Currents were filtered at 1 kHz and data were acquired at 5 kHz. All recordings, both macroscopic and at the single hemichannel level were acquired using an AT-MIO-16X D/A board from National Instruments and our own acquisition software (written by E. Brady Trexler). All chemicals were purchased from Sigma-Aldrich.

Fig. 1 shows representative Cx46 hemichannel current–voltage relations obtained by applying slow voltage ramps to oocytes in a two-electrode voltage clamp recording configuration. The curves in Fig. 1 A were obtained in external Ca²⁺ concentrations ranging from 1.8 to 0 mM (nominal). External Mg²⁺ was held constant at 1 mM. The corresponding conductance–voltage relations at each Ca²⁺ concentration are plotted below. Overall, the magnitude of the conductance progressively increased as the external Ca²⁺ concentration decreased. At each Ca²⁺ concentration, conductance declined robustly with hyperpolarization, but sensitivity to voltage shifted substantially such that the voltage at which conductance reached half-maximal shifted from approximately −10 mV in 1.8 mM Ca²⁺ to approximately −60 mV in 0 mM (nominal) Ca²⁺. Conductance also declined at positive voltages, but more modestly than at negative voltages and with no appreciable shift in voltage sensitivity with changes in external Ca²⁺. Lowering external Mg²⁺ while maintaining external Ca²⁺ constant at 1.8 mM had little effect on Cx46 hemichannel currents (unpublished data). However, upon lowering external Ca²⁺ to nominal levels (0 added), the effects of external Mg²⁺ were clearly evident and were qualitatively similar to those of external Ca²⁺, i.e., causing changes in the magnitude of the conductance and large shifts in the G-V relations at negative voltages and little or no effect at positive voltages (Fig. 1 B). Representative currents obtained from a family of voltage steps are shown in Fig. 1 C. Raising extracellular Ca²⁺ or Mg²⁺ substantially reduced the magnitude of the Cx46 currents, slowed activation, and sped up deactivation, consistent with previous reports (Ebihara and Steiner, 1993; Ebihara et al., 2003). Thus, both divalent cations appear to regulate Cx46 hemichannels similarly with Ca²⁺ being considerably more potent as previously reported (Ebihara and Steiner, 1993; Ebihara et al., 2003).

Placing Cx46-expressing oocytes in a solution containing no added Ca²⁺ and Mg²⁺ produced very large membrane currents that displayed little voltage dependence, even with hyperpolarizing voltages exceeding −100 mV (unpublished data). Concerned that series access resistance could lead to an apparent weakening or loss of voltage dependence and that activation of other endogenous currents could contribute to the appearance of altered voltage dependence, we recorded from cell-attached and excised patches where series access problems were circumvented and where the hemichannels could be observed in isolation; patches containing other endogenous channel activity were discarded. In some cases expression was high enough that patches solely contained multiple Cx46 hemichannels as shown in Fig. 2. Under conditions in which external Mg²⁺ was maintained at 1 mM and external Ca²⁺ was brought to submicromolar levels by addition of EGTA, Cx46 hemichannels closed robustly with hyperpolarization. In the example shown in Fig. 2 A, Cx46 hemichannels were essentially open at small negative membrane potentials and hyperpolarization beyond −40 mV initiated closure; at a voltage of −70 mV, closure was robust. Positive membrane voltages also elicited gating of Cx46 hemichannels, but the characteristics differed substantially. Most notably, conductance did not decline to zero, leaving a substantial residual conductance. As previously described, closure at positive and negative voltages occur by different gating mechanisms. At positive voltages there is gating to a residual subconductance state by a mechanism we have termed V_j gating; at negative voltages Cx46 hemichannels gate fully closed by a mechanism we have termed loop gating (Trexler et al., 1996; Bukauskas and Verselis, 2004). In Cx46, these gating mechanisms are distinguished not only by voltage polarity and the degree of hemichannel closure, but also by characteristically different gating transitions. V_j gating transitions are rapid and not resolved, typical of ion channel gating, whereas loop gating transitions consist of passage through a series of transient subconducting states en route to full closure/opening. Examples are shown in Fig. 2 C at −70 and +40 mV.

When external Mg²⁺ was omitted from the external solution, gating in response to voltage was substantially altered, but only at negative membrane voltages (Fig. 2 B). V_j gating at positive membrane voltages remained unchanged whereas the robust closure due to loop gating at negative membrane potentials did not occur. This effect on gating is also illustrated in a different patch recording containing multiple Cx46 hemichannels in which application of a large positive voltage to +90 mV produced the characteristic decrease to a residual level and was essentially the same when external Mg²⁺ was present or absent (Fig. 3). Application of a large negative membrane voltage of −70 mV produced robust closure in the presence of 1 mM external Mg²⁺, but little or no closure in the absence of added Mg²⁺. However, when a very large hyperpolarizing voltage to −160 mV was applied, current decreased robustly. Notably, however, the current at steady state, although substantially reduced, remained very noisy in contrast to the relatively quiet baseline typically observed with closure in the presence of divalent cations, in this case Mg²⁺. These results demonstrate that V_j gating is not affected when the external divalent cation concentration is reduced and further suggests that loop gating, although modulated by external divalent cations, remains robust in their absence or at very low (submicromolar) concentrations.

To better examine the action of divalent cations at negative membrane potentials, we recorded from patches containing only a single Cx46 hemichannel. Fig. 4 shows 30-s traces from an excised, outside-out patch recorded at various negative membrane potentials with and without Mg²⁺ added to the bath; free Ca²⁺ was buffered to submicromolar levels with EGTA. When no Mg²⁺ was present on the extracellular side, the hemichannel still clearly exhibited voltage dependence with hyperpolarization. Although very large negative membrane potentials were required to reduce open probability (P_o), it nonetheless decreased in a voltage-dependent manner evident as increasingly less time spent in the fully open state (see all points histograms to the right of each trace). Interestingly, the hemichannel did not spend much time in a clearly defined closed state either, displaying nearly continuous noisy fluctuations in current from the baseline that varied in amplitude, consistent with the noisy baseline observed in patches containing multiple active hemichannels (Fig. 3 B). When viewed at an expanded time scale, the noisy fluctuating current was occasionally interrupted by full closures to a quiet baseline (Fig. 4 C, quiet baseline indicated by arrows and full openings by asterisks). Changing the Ca²⁺/EGTA ratio so that the calculated free Ca²⁺ varied between ∼2 and 10 nM did not appear to change the number of full closures to a quiet baseline (unpublished data).

In the presence of 1 mM Mg²⁺, the behavior of the hemichannel differed significantly (Fig. 4 B). First, the hemichannel closed at more modest hyperpolarizations (compare histograms at −90 mV with and without Mg²⁺), consistent with the shift observed in macroscopic recordings, and second, the noisy baseline current was transformed into a more stable, quiet baseline current. Although there were still noisy fluctuations from the baseline, they were intermittent and reduced in extent. Plots of P_o vs. V_m calculated from 60–120-s recordings of the same patch illustrate the large shift in activation by external Mg²⁺ (Fig. 4 D). Mean values for V₀, the voltage at which P₀ is 0.5, in the presence or absence of extracellular Mg²⁺ were −59 ± 7 mV (n = 5 patches) and −105 ± 11 (n = 6 patches). Although shifted, voltage dependence with no added external Mg²⁺ remained comparably steep.

The results, thus far, suggest that divalent cations selectively modify loop gating rather than acting as open channel blockers or gating particles. To further test the viability of open channel block, we examined the sidedness of the divalent cation effect, specifically whether Mg²⁺ was effective at closing the hemichannel from either side of the membrane as would be expected for a blocking site in the aqueous pore. Fig. 5 (A and B) shows recordings of single Cx46 hemichannels in excised patches. Bath and pipette solutions contained EGTA to chelate Ca²⁺_. Mg²⁺ (1 mM) was present in the bath solution, but not the pipette solution, resulting in exposure of only one side of the hemichannel to divalent cations upon excision of the patch. In an outside-out patch, where the extracellular face of the hemichannel was exposed to the bath and hence 1 mM Mg²⁺, open probability was low at a membrane potential of −70 mV (Fig. 5 A). Upon hyperpolarizing further to −90 mV, openings became very brief and infrequent. In dramatic contrast, in an inside-out patch, when the cytoplasmic face of the hemichannel was exposed to the bath and hence 1 mM Mg²⁺, the hemichannel remained largely open at −70 mV (Fig. 5 B). Further hyperpolarization to −90 mV and then −130 mV progressively reduced open probability. Hyperpolarization of the membrane, in this case, required depolarization of the patch pipette, which would tend to keep divalent cations that are on the bath (cytoplasmic) side from entering the pore. Thus, the voltage-dependent reduction in open probability observed with membrane hyperpolarization and Mg²⁺ present only on the cytoplasmic side is the wrong polarity to be caused by voltage-dependent entry of Mg²⁺ into the pore leading to block.

We also examined the effect of increasing the concentration of Mg²⁺ beyond 1 mM. Exposure of the extracellular side of Cx46 hemichannels to 10 mM Mg²⁺ rapidly and effectively closed them (Fig. 5 C). In contrast, exposure of the cytoplasmic side to 10 mM Mg²⁺ did not have an appreciable effect except for a small reduction in the magnitude the unitary current (Fig. 5 D). In the example shown, amplitude histograms taken from the indicated segments illustrate the reduction in current magnitude and little change in the distribution of states. The effect of 10 mM Mg²⁺ over a large voltage range is illustrated in single hemichannel I-V relations obtained by applying voltage ramps from −70 to +70 mV to excised patches (Fig. 6 A). The left traces were obtained from an inside-out patch containing a single hemichannel in which the Mg²⁺ concentration was 1 mM on both sides. The middle and right traces were obtained from excised patches with Mg²⁺ elevated to 10 mM on the cytoplasmic and extracellular sides, respectively. Ensemble currents obtained from multiple ramps and patches are shown in Fig. 6 B. The presence of 10 mM Mg²⁺on the cytoplasmic side had little effect on the single channel I-V relation. Both loop gating at negative membrane voltages and positive membrane voltage remained essentially similar. In contrast, the same concentration of Mg²⁺ applied to the extracellular side had a dramatic effect, keeping hemichannels closed at inside negative voltages; opening usually required depolarization to inside positive voltages, consistent with a large positive shift in activation. The apparent effect on V_j gating at positive voltages can be explained by interactions between the two gates, which lie in series along the channel pore. When external divalent cation concentrations are high, the loop gates can close or remain closed at more positive voltages, thereby changing the voltage sensed by the V_j gates and altering their tendency to close (Srinivas et al., 2005). For completeness, we also examined patches in which no added Mg²⁺ was present on the outside so that the only source of Mg²⁺ was on the cytoplasmic side. Again, exposure of the cytoplasmic side to 10 mM Mg²⁺ had no effect on the single hemichannel I-V relation (unpublished data). The small reduction in unitary conductance and the small shift in reversal potential (∼−2.1 mV) observed with 10 mM Mg²⁺ on the cytoplasmic side are consistent with Mg²⁺ being permeable to Cx46 hemichannels.

Connexin hemichannels constitute a less well-studied class of membrane ion channel that is strongly regulated by membrane voltage and divalent cations. Typically, hemichannels tend to open with depolarization and/or reduction in the levels of external divalent cations. More recently, however, it has been suggested that voltage dependence is not an intrinsic property of hemichannels, but rather is solely attributable to block/unblock by divalent cations. In Cx46 and Cx37 hemichannels, voltage-dependent gating was reported to be absent in solutions containing no added divalent cations (Ebihara et al., 2003; Puljung et al., 2004). A model to account for the effects of divalent cations in Cx46 hemichannels contains two binding sites, one in the pore that when bound blocks or closes the hemichannel in a voltage-dependent manner (Ebihara et al., 2003). Binding to a second site, accessible only from the extracellular side, stabilizes the blocked or closed state. Thus, it was deemed that there is no intrinsic voltage gating mechanism. In Cx37 hemichannels, polyvalent cations, including extracellular Ca²⁺, Mg²⁺, and Gd³⁺, were proposed to traverse the length of the electric field and bind at the cytoplasmic end of the hemichannel pore (Puljung et al., 2004).

Using single hemichannel recordings in excised patches where we could examine hemichannels in isolation and test the effectiveness of divalent cations from either side of the membrane, we find that a pore-blocking model is unlikely, at least for Cx46 hemichannels. In solutions containing no added Mg²⁺ and EGTA to reduce Ca²⁺ to submicromolar levels, we found that robust voltage dependence persisted and that the action of divalent cations was to selectively modulate one of two forms of intrinsic voltage-dependent gating that we have provisionally termed loop gating (Trexler et al., 1996; Verselis and Bukauskas, 2002; Bukauskas and Verselis, 2004). Although this selective action on loop gating was evident both in macroscopic and single hemichannel recordings, we wish to point out that when we examined the effects of Ca²⁺ and Mg²⁺ macroscopically, the results depended on the level of Cx46 expression. When currents were large (exceeding 5–10 μA), because of high levels of expressed connexin and/or exposure of oocytes to low levels of divalent cations, voltage dependence indeed weakened or even disappeared. In solutions containing no added Ca²⁺ or Mg²⁺, when membrane conductances were usually elevated to very high levels (>10 μS), we typically did not observe voltage dependence at all over a large voltage range, negative and positive. Such effects in oocytes can be attributed to series access resistance whereby the effective voltage drop across the membrane is rendered nonuniform and reduced in magnitude due to significant contributions of cytoplasmic and extracellular resistances (Baumgartner et al., 1999). In addition, tight clustering of channels and possibly hemichannels can produce overlapping access pathways for each aqueous pore, creating the characteristics of a single, large diameter pore in which access resistance makes a larger relative contribution (Wilders and Jongsma, 1992; Hille et al., 1999). Weakening or loss of voltage dependence has been documented in recordings of gap junctional currents from oocyte cell pairs exhibiting high junctional membrane conductances (Verselis et al., 1994). Thus, we conclude that we could not entirely rely on macroscopic recordings to assess the effects of low concentrations of divalent cations on connexin hemichannel gating. It is possible that series access resistance may be the basis for the apparent loss of hemichannel gating reported by Ebihara et al. (2003) and Puljung et al. (2004).

Both the persistence of gating in solutions containing no added Mg²⁺ and EGTA to chelate Ca²⁺ and the sidedness of the Mg²⁺ effect argue strongly against pore block as the mechanism by which divalent cations rapidly and reversible modulate Cx46 currents, at least in the concentration range we examined. Altering Mg²⁺ between 1 and 10 mM on the extracellular side had profound effects on open probability, whereas the same changes in Mg²⁺ on the cytoplasmic side had little effect, indicating that the binding site is certainly not on the cytoplasmic side of the hemichannel. In particular, when Mg²⁺ was elevated to 10 mM on the cytoplasmic side, the expectation for a pore-blocking mechanism would be that voltage dependence should reverse in sign as depolarization should now drive Mg²⁺ into the channel. We did not see evidence of such reversal of polarity, even when the extracellular side was exposed to a solution containing no added Mg²⁺ so that the only source of Mg²⁺ was through the pore from the cytoplasmic side. It is possible that access from the cytoplasmic side of the hemichannel is hindered in some way. However, this possibility would be difficult to reconcile for a channel with such a large pore that has been shown to be cation selective and permeable to large fluorescent dyes and large thiol reagents added to either side of the membrane (Pfahnl and Dahl., 1998; Trexler et al., 2000; Kronengold et al., 2003a,2003b) The V_j gate, which has been suggested to involve narrowing of the cytoplasmic vestibule (Ri et al., 1999), could conceivable block access when in the substate conformation, but there is little gating at positive voltages up to +35 mV and block is still absent from the cytoplasmic side. We also observed that addition of Mg²⁺ to the cytoplasmic side reduced unitary conductance and shifted the reversal potential in a direction consistent with Mg²⁺ permeability. Thus, it is unlikely that access to a blocking site in the pore is hindered from the cytoplasmic aspect of the hemichannel.

Our result differs from that of Pfahnl and Dahl (1999) who examined the effects of Ca²⁺ on Cx46 hemichannels in excised patches. Exposure of Cx46 hemichannels to high concentrations of Ca²⁺ (2.4 mM added) on either side of the membrane led to robust closure. In their hands, the voltage polarity that produced closure reversed when the side of Ca²⁺ application was reversed. These results led to the conclusion that Ca²⁺ enters and binds to a site inside the pore. An extension of having a binding site in the connexin pore is that cytoplasmic Ca²⁺ and Mg²⁺ would likely be effective at closing gap junction (cell–cell) channels as well. Although cytoplasmic Ca²⁺ has been shown regulate gap junction channels at concentrations as low as ∼150–300 nM, there is evidence that this modulation requires soluble intermediaries (Peracchia, 2004). For Cx43, intracellular Ca²⁺ elevated to the μM range did not cause uncoupling when calmodulin was inhibited (Lurtz and Louis, 2007), suggesting that Ca²⁺ itself is not a potent blocker. Also, Ca²⁺ permeability through gap junctions may contribute to the propagation of Ca²⁺ waves (Saez et al., 1989; Paemeleire et al., 2000). Elevating Ca²⁺ to the mM range on the cytoplasmic side has not been tested in gap junctions, but such high concentrations would be difficult to interpret because of secondary effects and deleterious effects on cell viability. Elevating Mg²⁺, which is normally 1–3 mM inside cells, to as high as 10 mM did not produce block in Cx37 gap junction channels, but did alter gating kinetics more than predicted based on screening suggestive of binding (Ramanan et al., 1999). Permeability of hemichannels to Ca²⁺ has not been rigorously tested. In any case, the high concentration of cytoplasmic Ca²⁺ used in the studies of Pfahnl and Dahl (1999) may produce effects that are unrelated to loop gating. Thus, Ca²⁺ and Mg²⁺ could conceivably share a common extracellular site that functions to regulate loop gating and additional effects of high concentrations of Ca²⁺ may promote hemichannel closure by independent means. We were unable to obtain stable patches in solutions that did not contain EGTA or BAPTA and therefore could not reliably test the effects of μM concentrations of Ca²⁺ on single Cx46 hemichannels. Resolution awaits examination of the effects of Ca²⁺ at the single hemichannel level over a wide range of concentrations.

In Cx32 hemichannels, a ring of 12 Asp residues in the second extracellular loop, E2, consisting of D169 and D178 was reported to be the basis of Ca²⁺ regulation (Gomez-Hernandez et al., 2003). However, the reported affinity for external Ca²⁺ in this hemichannel is low compared with Cx46, with 50% reduction in current occurring at a concentration of 1.3 mM; affinity for Mg²⁺ is only modestly lower. Also, the proposed mechanism of action was reported to involve effects both on gating and conduction centered around a subconducting state. Pore block was proposed as one possible mechanism of action, but in Cx46, the first extracellular loop, E1, has been shown to contribute to the pore. Also, both D169 and D178 were necessary to maintain Ca²⁺ regulation in Cx32, but negative charges at both positions is not widely conserved among connexins. Thus, it may be that there are connexin-specific differences in divalent cation regulation and/or a multiplicity of actions over a wide concentration range.

Our results indicate that loop and V_j gating are both intrinsic voltage-dependent mechanisms, with loop gating being that which is selectively modulated by extracellular divalent cations. The V_j gating mechanism has been proposed to involve movement of the N-terminal domain at the cytoplasmic end the channel, mediated by a proline-kink motif in the second transmembrane domain (Verselis et al., 1994; Ri et al., 1999). The result is a narrowing of the pore at the cytoplasmic end, i.e., the residual subconducting state. Consistent with this mechanism being intrinsically voltage dependent, charged amino acid substitutions in NT can reverse gating polarity, suggesting reversal of the sign of a voltage sensor (Verselis et al., 1994; Purnick et al., 2000). The molecular determinants of loop gating remain unknown, but there is the suggestion that this mechanism operates at the opposite end of the hemichannel, i.e., at the extracellular end. First, loop gating is robustly modulated by external, not internal, divalent cations. Second, studies using the substituted cysteine accessibility method in Cx46 hemichannels reported accessibility of residue L35 in the first transmembrane domain, TM1 (Zhou et al., 1997; Kronengold et al., 2003b). Closure of the loop gate appeared to block accessibility from the cytoplasmic side, suggesting the gate or the constriction that occurs with gating is located extracellular to this position (Pfahnl and Dahl, 1998). Finally, conformational changes at the surfaces of hemichannels examined using atomic force microscopy indicate that the extracellular pore entrance is significantly decreased in diameter upon addition of Ca²⁺ (Muller et al., 2002; Thimm et al., 2005). Whether this structural change is related to loop gating is unknown.

Recordings at the single hemichannel level in solutions containing no added Mg²⁺ and EGTA revealed interesting features of loop gating. Although progressive hyperpolarization induced progressively less time spent in the fully open state, as expected for a voltage-dependent mechanism, Cx46 hemichannels did not spend much time fully closed either, but rather flickered nearly continuously between what may be poorly resolved subconductance states and/or a closed state. At the largest membrane voltage (−130 mV) that we could maintain a stable patch recording for an extended period of time, it was difficult to decipher clear conductance states, except for occasional and brief full openings (Fig. 4 C, asterisks) and occasional full closures to the baseline during which flickering was interrupted (Fig. 4 C, arrows). Typically, the full closures were brief, lasting tens of milliseconds or a few hundred milliseconds at best. Addition of extracellular divalent cations, in this case 1 mM Mg²⁺, promoted full closures that lasted much longer, many seconds. Flickering still occurred within these long closures but appeared as intermittent bursts in contrast to the nearly continuous flickering in the absence of added Mg²⁺. We suggest that the full closures under very low divalent cation conditions represent spontaneous entry into a fully closed state rather than occasional binding resulting from trace amounts of divalent cations, because varying the Ca²⁺/EGTA ratio such that the calculated free Ca²⁺ varied from ∼2 to ∼10 nM did not increase the frequency of the full closures (unpublished data).

We suggest the following regulatory mechanism in Cx46 hemichannels that combines effects of voltage and divalent cations. Hemichannels undergo a complex series of conformational changes from the fully open state to the fully closed state that constitutes loop gating. Some of the intermediate conformational states in loop gating are resolvable as short-lived subconductance states. This process is intrinsically voltage dependent, with hyperpolarization tending toward closure. Flickering from the fully closed state plausibly represents attempts at opening or partial opening. Thus, voltage drives the hemichannel to close, perhaps by initiating movements in the extracellular loops, but residence in a stable, fully closed state requires the binding of divalent cations to an extracellular site. In the absence of extracellular divalent cations, the hemichannel exhibits continuous transiting between the short-lived states that characterize loop gating transitions.

Possibilities for hemichannel function, both physiological and pathological, are growing in a number of tissues (Evans et al., 2006). Voltage and divalent cations are both involved in regulating hemichannels and it appears that changes in extracellular Ca²⁺ and membrane potential will likely combine to rapidly and reversibly regulate hemichannel opening in robust fashion. This study shows that an intimate association between divalent cations and voltage gating occurs through loop gating, a gating mechanism that is intrinsically voltage dependent as well as Ca²⁺/Mg²⁺ dependent. In this view, loop gating is the mechanism responsible for opening hemichannels in the plasma membrane by depolarization, lowered extracellular Ca²⁺, and, perhaps, by other conditions known to modulate hemichannel activity such as metabolic inhibition, intracellular pH and pharmacological agents (John et al., 1999; Trexler et al., 1999; Contreras et al., 2002; Trelles, M.P., and M. Srinivas. 2007. Biophysical Society Meeting Abstracts, Biophysical Journal. Supplement: 442a, Abstr. 2117). In addition, other factors, such as elevated extracellular K⁺, have been shown to modulate Ca²⁺ sensitivity, thereby changing the dynamic range over which hemichannels can respond to divalent cations (Srinivas et al., 2006). The importance of having divalent cations as modulators rather than as gating particles is that the fundamental mechanism of closure is an intrinsic gating mechanism. Thus, differences in loop gating among connexins may underlie apparent differences in Ca²⁺/Mg²⁺ regulation. Also, without loop gating, divalent cations may not effectively regulate hemichannels, leading to cell death through influx of Ca²⁺and/or efflux of ATP and other vital intracellular contents.

We wish to thank Terrance Seals for technical assistance.

This work was supported by National Institutes of Health grants EY13869 to M. Srinivas and GM54179 to V.K. Verselis.

Lawrence G. Palmer served as editor.