Divergence of Ca²⁺ selectivity and equilibrium Ca²⁺ blockade in a Ca²⁺ release-activated Ca²⁺ channel

Ca²⁺ is a multifunctional signaling messenger crucial for diverse biological processes. Among the various ways by which cellular Ca²⁺ signals are generated, store-operated Ca²⁺ release-activated Ca²⁺ (CRAC) channels are recognized as a widespread mechanism for regulating transcription, motility, and proliferation in many cells (Feske, 2009; Hogan et al., 2010; Lewis, 2011). CRAC channels produce sustained intracellular Ca²⁺ elevations and are implicated in a growing list of human diseases including immunodeficiency (Feske, 2009), allergy (Di Capite et al., 2011), cancer (Prevarskaya et al., 2011), thrombosis (Varga-Szabo et al., 2011), and inflammatory bowel disease (McCarl et al., 2010). The broad expression of CRAC channels and their involvement in many physiological processes has produced intense interest in CRAC channels as targets for drug development. Yet, our understanding of how CRAC channels operate at a mechanistic level is still rudimentary and, in particular, the molecular and structural mechanisms of ion permeation and channel gating are only now beginning to be elucidated.

A distinguishing feature of CRAC channels is high Ca²⁺ selectivity (P_Ca/P_Na ≈ 1,000; Hoth and Penner, 1993). Current thinking about the origin of this selectivity is rooted in the idea of preferential Ca²⁺ binding to a high affinity binding site (K ≈ 20 µM) at the selectivity filter, which occludes Na⁺ flux through the pore (Prakriya, 2009). In support of this idea, a mutation at the predicted CRAC channel selectivity filter (E106D in Orai1) diminishes both Ca²⁺ selectivity as well as the affinity of Ca²⁺ blockade of Na⁺ flux (Prakriya et al., 2006; Vig et al., 2006; Yeromin et al., 2006; Yamashita et al., 2007), which would be expected if Ca²⁺ selectivity is primarily determined by the K_i of Na⁺ current blockade. Such a selection-through-affinity mechanism was originally described for voltage-gated Ca²⁺ channels, which, like CRAC channels, display exquisite Ca²⁺ selectivity (P_Ca/P_Na > 1,000; Sather and McCleskey, 2003). However, the apparent affinity of Ca²⁺ block is roughly 20-fold higher in voltage-gated Ca²⁺ channels than CRAC channels (K_d of ∼1 vs. 20 µM, respectively; Almers and McCleskey, 1984; Bakowski and Parekh, 2002; Su et al., 2004; Prakriya and Lewis, 2006), raising the possibility that the biophysical mechanisms of how these channels achieve high Ca²⁺ selectivity may differ. Importantly, although appealing in its simplicity, it remains uncertain whether an equilibrium binding model fully explains how CRAC channels achieve Ca²⁺ selectivity under physiological nonequilibrium conditions.

A convenient CRAC channel system in which this and related questions of ion selectivity can be investigated is the Orai3 channel. When overexpressed in HEK293 cells, Orai3 channels produce either Ca²⁺-selective or nonselective currents, depending on whether they are activated by the ER Ca²⁺ sensor, STIM1, or the small molecule 2-aminoethoxydiphenyl borate (2-APB; DeHaven et al., 2008; Peinelt et al., 2008; Schindl et al., 2008; Zhang et al., 2008; Yamashita et al., 2011). The selection-through-affinity model predicts that Ca²⁺ binding affinity in the poorly selective 2-APB–activated Orai3 channels should be lower, a possibility that is directly tested in this study.

In addition to Ca²⁺ selectivity, much attention has surrounded the gating mechanism of CRAC channels. STIM1 interacts with two distinct sites on each Orai subunit and these interactions contribute both to the accumulation of Orai channels at the ER–plasma membrane junctions and channel gating (Li et al., 2007; Muik et al., 2008; Park et al., 2009; McNally et al., 2013). However, our understanding of how STIM1 binding is coupled to channel gating is only now emerging. Biophysical studies using noise analysis have found that activation of CRAC channels after store depletion occurs through stepwise recruitment of closed channels to a high open probability (P_o) state (Prakriya and Lewis, 2006; Kilch et al., 2013). It is tempting to speculate that in the context of an activation mechanism involving reversible binding of multiple STIM1 molecules to CRAC channels (Li et al., 2010; Hoover and Lewis, 2011), the abrupt opening of single CRAC channels follows the concerted binding of the required number of STIM1 molecules to Orai1. A recent study has suggested 2-APB gating may also occur through a similar mechanism, wherein ligand binding (STIM1 or 2-APB) leads to a series of graded conformational changes culminating in stepwise opening of Orai channels (Amcheslavsky et al., 2013). However, whether 2-APB gating in fact occurs through stepwise channel opening has not been directly examined.

In this study we compared permeation, block, and gating of Orai3 channels activated by 2-APB and STIM1 to gain insights into the biophysical mechanisms that shape ion selectivity and gating of Orai channels. Our results indicate that the distinct Ca²⁺ selectivity of STIM1- and 2-APB–gated channels cannot be explained in terms of equilibrium Ca²⁺ binding at the selectivity filter set as defined by the K_d of Ca²⁺ blockade. Rather, we suggest that the kinetic rates of Ca²⁺ and Na⁺ entry/exit contribute to the lower Ca²⁺ selectivity of 2-APB–gated channels. Our results also indicate that despite different ion conduction properties, STIM1- and 2-APB–gated channel activation states exhibit modal gating to a high P_o state, suggesting that the allosteric mechanisms that open the pore in response to ligand binding are operationally similar between the two gating modes. Collectively, these results provide new insights into the mechanisms of ion selectivity and gating in Orai channels.

HEK293 cells were grown in medium consisting of 44% Dulbecco’s modified Eagle’s medium (Corning) and 44% Ham’s F12 (Corning), supplemented with 10% fetal calf serum (HyClone), 1% 200 mM glutamine, 1% 5000 U/ml penicillin, and 5,000 µg/ml streptomycin. The cells were maintained in log-phase growth at 37°C in 5% CO₂.

The CFP-Orai3 plasmids used here have been previously described (Yamashita et al., 2011). Site-directed mutagenesis to generate the indicated Orai3 mutants was performed using the QuickChange site-directed mutagenesis kit (Agilent Technologies) according to the manufacturer’s instructions and the results were confirmed by DNA sequencing. Orai3 and STIM1 were cotransfected using Transpass D2 (New England Biolabs, Inc.), with 200 ng Orai3 and 300 ng STIM1 per 12-mm coverslip when coexpressed or 200 ng when CFP-Orai3 was expressed alone.

The standard extracellular Ringer’s solution contained 130 mM NaCl, 4.5 mM KCl, 20 mM CaCl₂, 10 mM D-glucose, and 5 mM Na-HEPES, pH 7.4. The divalent-free (DVF) Ringer’s solution contained 150 mM NaCl, 10 mM HEDTA, 1 mM EDTA, and 10 mM HEPES, pH 7.4. pH was adjusted to 7.4 with NaOH or CsOH. 10 mM TEA-Cl was added to all extracellular solutions to prevent contamination from voltage-gated K⁺ channels. The standard internal solution contained 135 mM caesium aspartate, 8 mM MgCl₂, 8 mM BAPTA, and 10 mM Cs-HEPES, pH 7.2. For experiments examining block of Na⁺-I_CRAC by Ca²⁺, CaCl₂ was added to the standard DVF solution at the appropriate amount calculated from the MaxChelator software (WEBMAXC 2.10, available at http://www.stanford.edu/∼cpatton/webmaxc2.htm). The 300- and 600-µM [Ca²⁺]_o solutions were made by adding the indicated amount of CaCl₂ to a nominally Ca²⁺-free solution containing 150 mM NaCl and 10 mM HEPES, pH 7.4. For the pore-sizing studies described in Fig. 1 B, the following organic compounds were substituted for sodium methanesulfonate in the external solution: hydroxylamine HCl (NH₂OH-HCl), hydrazine HCl (NH₂NH₂-HCl), methylamine HCl (CH₃NH₂-HCl), dimethylamineHCl ((CH3)₂NH-HCl), trimethylamineHCl ((CH₃)₃N-HCl), and tetramethylammonium chloride ((CH₃)₄NCl). These chemicals were purchased from Sigma-Aldrich. pH was adjusted to 7.4 with NMDG except in the case of hydrazine HCl (pH 6.4) and hydroxylamine HCl (pH 6.2), which were studied at acidic pH to increase the ionized concentration of the test ion.

Patch-clamp recordings were performed using an Axopatch 200B amplifier (Molecular Devices) interfaced to an ITC-18 input/output board and an iMac G5 computer (Apple). Currents were filtered at 1 kHz with a 4-pole Bessel filter and sampled at 5 kHz. Stimulation and data acquisition and analysis were performed using routines developed on the Igor Pro platform by R.S. Lewis (Stanford University, Palo Alto, CA). The holding potential was +30 mV unless otherwise indicated. Two types of stimuli were used: (1) a 100-ms step to –100 mV followed by a 100-ms ramp from –100 to +100 mV usually applied every 1 s and (2) continuous recording holding potential at −100 mV. Current amplitudes were typically analyzed at −100 mV unless indicated otherwise.

200-ms sweeps were acquired at the rate of 4 Hz at a constant holding potential of −100 mV, digitized at 20 kHz, low-pass filtered using at 10 kHz using the amplifier’s built-in Bessel filter, and recorded directly to hard disk. The mean current and variance were calculated from each sweep. For spectral analysis, data were low pass filtered using a 20 Hz Gaussian filter and power spectra were computed from either 256- or 1,024-point segments using a Hamming window (Igor Pro; Wavemetrics) and averaged from 3–10 sweeps.

Unless noted otherwise, all data were corrected for leak currents collected in 20 mM Ca²⁺ + 50–100 µM La³⁺. Averaged results are presented as the mean value ± SEM at −100 mV unless indicated otherwise. All curve fitting was done by least-squares methods using built-in functions in Igor Pro 5.0. The minimal pore diameter was estimated as previously described (Prakriya and Lewis, 2006; Yamashita et al., 2007). In brief, relative permeabilities to various organic cations estimated from the bionic GHK equation were plotted against ion size as in Fig. 1 B and fit to the hydrodynamic relationship:

where P_x/P_Na is the relative permeability of the cation being tested, d_ion is the diameter of the test cation, and d_pore is the minimal pore diameter. The relative permeabilities for the organic cations were determined from changes in reversal potential induced by replacing extracellular Na⁺ in the standard DVF solution with the test cation.

To analyze the voltage dependence of block of Na⁺-I_CRAC by Ca²⁺_o, the model described by Guo and Lu (2000) was used. Unlike the older Woodhull (1973) model that assigns the voltage dependence of block solely to the valence of the blocking particle, the Guo and Lu (2000) model expresses voltage dependence in terms of an apparent valence, an empirical factor that encompasses the effects of blocker valence and the coupled movements of conducting ions displaced by blocker binding.

In this scheme, Ca²⁺ block is described in terms of second-order binding to a single site:

where Ch is the channel, Ca²⁺_o and Ca²⁺_i are extracellular and intracellular Ca²⁺, k₁ and k₋₁ are the binding and unbinding rates from the extracellular side, k₂ is the unbinding rate from the intracellular side, and each z_i represents the apparent valence for the corresponding transition. Block from the intracellular compartment is assumed to be negligible because of low (nanomolar) intracellular Ca²⁺ concentrations. With these assumptions, the fraction of unblocked current is given by (Guo and Lu, 2000):

where K₁ = k₋₁/k₁ is the equilibrium dissociation constant at 0 applied voltage, and Z₁ and z_i are the apparent valences. Z₁ (= z₁ + z₋₁) provides a measure of the overall voltage dependence of Ca²⁺ block and arises from the movement of the charged blocker (Ca²⁺) within the field as well as the possible displacement of permeant ions (Na⁺) within the pore. k₂/k₋₁ is the ratio of the rates of Ca²⁺ escaping into the cytoplasm versus returning to the extracellular solution from the pore, and thus provides a measure of Ca²⁺ permeation. The quantities k₂/k₋₁ and z₋₁+z₂ were treated as single adjustable parameters for fitting the data. To avoid complications arising from activation of Na⁺-I_CRAC during hyperpolarizing steps in 2-APB–gated channels (see Fig. S2), we measured block from the ratio of the steady-state currents in the presence and absence of extracellular Ca²⁺, rather than from the ratio of the steady-state to peak currents in Ca²⁺ as was done previously (Prakriya and Lewis, 2006; Yamashita et al., 2007).

T. Begenisich (University of Rochester, Rochester, NY) provided the program that we used to calculate current–voltage relations from a four-barrier, three-site Eyring rate model (Begenisich and Cahalan, 1980; Dang and McCleskey, 1998). In this model, the rate at which an ion moves from one site to another equals the product of the rate constant for this transition and the probability of occupancy of the source site by that ion. The individual rate constants are governed by the barrier heights and well depths. Current for a given ion is determined from the net ion flux rate over the second energy barrier (rate constant × probability of occupancy of that state). The details of the model have been previously described (Dang and McCleskey, 1998). The energies of the binding sites and wells in the model are adjustable parameters and were determined or constrained by known data. These values are justified in the Discussion and indicated in Fig. 9. The extracellular and intracellular Na⁺ concentrations were set to 150 mM, extracellular Ca²⁺ was 20 mM, and the intracellular Ca²⁺ concentration was zero in the model.

Fig. S1 shows the current–voltage relationships of WT and mutant (E81D, E85A/D87A/E89A, and E165A) Orai3 currents in a 20-mM Ca²⁺ Ringer’s solution. Fig. S2 shows the slow activation of the Na⁺-CRAC current in DVF solution during hyperpolarizing steps to −100 mV. Fig. S3 shows the power spectrum analysis of Orai3 currents gated by STIM1 or 2-APB.

Orai3 channels can be activated in a store-dependent manner by the ER Ca²⁺ sensor STIM1 or directly in a store-independent manner by high doses of 2-APB. In contrast to STIM1-activated Orai3 channels, however, 2-APB–activated Orai3 channels exhibit low Ca²⁺ permeability and readily conduct Cs⁺ ions (Schindl et al., 2008; Zhang et al., 2008; Yamashita et al., 2011). Here, we sought to understand differences in the interaction of conducting ions to the pores of STIM1- and 2-APB–activated Orai3 channels and compare their gating behaviors. To carry out these studies, we expressed CFP-Orai3 either alone or together with unlabeled STIM1 in HEK293 cells and studied their functional properties by patch-clamp electrophysiology. STIM1-activated Orai3 currents were obtained by depleting ER Ca²⁺ stores in cells expressing STIM1 and CFP-Orai3 with 1 µM thapsigargin before patch-clamp recordings, whereas 2-APB–activated currents were elicited by administering 2-APB to cells overexpressing CFP-Orai3. A dose–response experiment indicated that 2-APB activates Orai3 channels with an EC₅₀ of ∼24 µM and a high Hill coefficient of ∼8 (Fig. 1 A). Therefore, unless otherwise indicated, we used a concentration of 50 µM to elicit 2-APB–gated Orai3 currents.

Unlike STIM1-activated Orai3 channels, 2-APB–gated Orai3 channels readily conduct many large cations (Schindl et al., 2008). In accordance with previous results postulating that an enlarged pore is responsible for this feature (Schindl et al., 2008), 2-APB–activated Orai3 channels display a wider apparent pore width compared with STIM1-activated Orai3 channels (Fig.1 B). As a result of (or associated with)this structural change, 2-APB–gated Orai3 channels exhibit lower Ca²⁺ selectivity than STIM1-gated channels. This is reflected in the well-described leftward shift of the reversal potential (V_rev), which changed from 71 ± 6 mV (n = 7) for STIM1-gated channels to 25 ± 2 mV (n = 8) for 2-APB–gated channels in a Ringer’s solution containing 20 mM Ca²⁺. 2-APB–gated Orai3 channels also showed large outward currents carried by intracellular Cs⁺ (Fig. 1 C). Moreover, replacing extracellular Na⁺ with choline, a large cation that is impermeable through most cationic channels, revealed significant Na⁺ conduction at −100 mV (Fig. 1 D), reaffirming that 2-APB–gated channels are poorly Ca²⁺ selective under these experimental conditions.

Current models postulate that the high Ca²⁺ of CRAC channels originates from tight binding of Ca²⁺ to the CRAC channel selectivity filter (Prakriya and Lewis, 2003; Prakriya, 2009; McNally and Prakriya, 2012). On this basis, the widely different Ca²⁺ selectivities of STIM1- and 2-APB–activated channels should arise from differences in the energetics of Ca²⁺ binding to the Orai3 selectivity filter. We examined this question using several approaches: by estimating the thermodynamic stability of Ca²⁺ binding to the Orai3 channel pore from Ca²⁺ block measurements, from the voltage dependence of Ca²⁺ block, and from the rates of Ca²⁺ blockade of the monovalent current. Like voltage-gated Ca²⁺ channels, Orai channels readily conduct a variety of small monovalent cations including Na⁺ upon removal of extracellular divalent cations. Therefore, we first applied a DVF solution to elicit Na⁺ currents through Orai3 channels and examined the ability of micromolar concentrations of extracellular Ca²⁺ to block the Na⁺ currents (Fig. 2).

At a membrane potential of −100 mV, these tests indicated that Na⁺ currents through STIM1-gated Orai3 channels are blocked dose dependently by extracellular Ca²⁺ with an K_i of ∼25 µM (Fig. 2, A and F). This K_i is nearly identical to the sensitivity of native CRAC channels (Bakowski and Parekh, 2002; Su et al., 2004; Prakriya and Lewis, 2006) and Orai1 channels overexpressed in HEK293 cells (Yamashita et al., 2007). Thus, when gated by STIM1, the thermodynamic stability of Ca²⁺ binding is similar in Orai1 and Orai3 channels, as expected given the high homology of the primary sequence in the pore regions of these paralogous proteins. Furthermore, as previously described for Orai1 channels (Prakriya et al., 2006; Yamashita et al., 2007), mutating the predicted selectivity filter formed by the conserved amino acid Glu81 to Asp (E81D) lowered both Ca²⁺ block (Fig. 2, B and F; K_i = 111 µM) and the Ca²⁺ selectivity of STIM-activated Orai3 channels (Fig. S1 A), as expected if Ca²⁺ selectivity is directly regulated by Ca²⁺ binding at the selectivity filter.

Given that 2-APB–activated Orai3 channels are poorly Ca²⁺ selective, we hypothesized that these channels should be less sensitive to Ca²⁺ blockade than STIM1-gated Orai3 channels. Yet, much to our surprise, Na⁺ currents through 2-APB–activated Orai3 channels were blocked with similar Ca²⁺ sensitivity as STIM1-activated channels (Fig. 2, D and F; and Table 1). Moreover, as with STIM1-gated Orai3 channels, blockade of the monovalent current in the 2-APB gating mode was markedly lowered by the E81D mutation (Fig. 2, E and F; and Table 1; K_i = 154 µM), indicating that a pointed disruption of the Ca²⁺ selectivity filter formed by E81 destabilizes Ca²⁺ binding in both gating modes. Introducing the E165A mutation in the TM3 segment of Orai3 did not significantly alter the sensitivity of Ca²⁺ blockade (Fig. 2 F) or Orai3 selectivity (Fig. S1, A and B), effectively ruling out a contribution for E165 and TM3 for Ca²⁺ binding to the 2-APB–activated Orai3 pore. Collectively, these results indicate that the affinity of Ca²⁺ binding to the Orai3 pore at the test potential used here (−100 mV) is not different between the two gating modes, despite the lower Ca²⁺ permeability of 2-APB–gated channels (Fig. 1 D).

In many types of ion channels, the sensitivity and rate of open-channel block is affected by the concentration of permeant ions, either because of competition between blocker and permeant ions for pore binding sites or displacement of the blocker by permeant ions (Armstrong, 1971; Neyton and Miller, 1988b; Antonov and Johnson, 1999). In addition to revealing putative interactions between permeant ions and blockers in the pore, results from such experiments are often useful to delineate the position of pore sites for blockers and permeant ions (Neyton and Miller, 1988a). To approach this issue, we investigated the effects of varying the external Na⁺ concentration on Ca²⁺ blockade of inward Orai3 Na⁺ currents. These experiments revealed an interesting difference between STIM1- and 2-APB–gated channels. Lowering the extracellular Na⁺ concentration from 150 to 75 mM did not significantly alter the sensitivity of Ca²⁺ blockade of STIM1-gated Na⁺ fluxes (Fig. 3 A). In 2-APB–gated channels, however, the K_i of inhibition of Na⁺ currents in 2-APB–gated channels was reduced threefold to ∼9 µM (Fig. 3 B). Thus, decreasing the Na⁺ occupancy of external pore sites by lowering [Na]_o enhances Ca²⁺ binding to its high affinity site. Consistent with this interpretation, the blocking rate also increased twofold at the lower Na⁺ concentration (Fig. 3, D and E). These results indicate that competition between Na⁺ and Ca²⁺ ions in the pore influences the ability of Ca²⁺ to interact with its high affinity site in 2-APB–gated channels. One possibility is that Ca²⁺ has to dislodge Na⁺ ions from pore sites before it gains access to its high affinity site. This competition could occur either at the high affinity Ca²⁺ site (i.e., the selectivity filter) itself or at a more superficial location in the pore.

The outer vestibule of all Orai channels contains acidic residues that have been implicated in La³⁺ binding and that are postulated to facilitate cation accumulation at the mouth of the channel (Fig. 3 F; Yeromin et al., 2006; McNally et al., 2009). To examine the role of these residues for the observed competition between Na⁺ and Ca²⁺ ions, we mutated these acidic sites (E85A/D87A/D89A) and tested how this maneuver affects Ca²⁺ blockade. Unexpectedly, these tests revealed that block is considerably more complex in the triple mutant than that observed in WT Orai3 channels. Here, Ca²⁺ block exhibited a double sigmoid dependence on [Ca²⁺]_o at the normal concentration of [Na⁺]_o. The double sigmoid block could be well fit with the sum of two Hill equations with K_is of 33 and 1694 µM (Fig. 3 C). I–V relations showed that the reversal potential of the current progressively shifts rightward at high [Ca²⁺]_o (V_rev = 10 ± 1 mV in 600 µM Ca²⁺, 13 ± 1.5 mV in 2 mM Ca²⁺_o, and 25 ± 1 mV in 20 mM Ca²⁺_o) but is unchanged in the lower range of [Ca²⁺]_o (V_rev = 8 ± 1 mV at [Ca²⁺] of 0–300 µM). Thus, we interpret the double sigmoid behavior as the sum of block at low [Ca²⁺]_o and mixed Ca²⁺-Na⁺ conduction at high [Ca²⁺]_o. Precisely why Ca²⁺ block and Ca²⁺ conduction are so well separated in this mutant remains unclear and additional studies are required to examine this issue. Nonetheless, the block observed at low [Ca²⁺]_o permits a test of the Na⁺ dependence of this behavior. Reducing external [Na⁺]_o to 75 mM eliminated the double sigmodicity of blockade in this mutant and block here could be well fit with a single Hill relation (K_i of ∼64 µM). Importantly, the K_i of Ca²⁺ blockade at 75 mM [Na⁺] is not smaller than the K_i of blockade at high Na⁺ (150 mM; Fig. 3 C), indicating that neutralization of the acidic sites in the vestibule eliminates the dependence of block on the extracellular Na⁺ concentration. The most straightforward interpretation of these results is that mutation of the external TM1–TM2 loop acidic residues decreases Na⁺ binding and accumulation in the outer vestibule, thereby diminishing competition between Na⁺ and Ca²⁺ ions and permitting Ca²⁺ to interact with its high affinity site with equal ease irrespective of the Na⁺ concentration. The acidic residues of the outer vestibule thus contribute to the accumulation and binding of Na⁺ ions in WT Orai3 channels in the 2-APB–gating mode.

As shown in Fig. 1, unlike STIM1-gated channels, 2-APB–activated Orai3 channels are poorly Ca²⁺ selective and display significant outward conduction of intracellular Cs⁺ ions at positive voltages. In principle, the outward conduction of intracellular Cs⁺ ions could arise if Ca²⁺ blockade of monovalent conduction at positive voltages is weaker in 2-APB– than in STIM1-gated channels. To test this possibility, we measured the voltage dependence of Ca²⁺ blockade in the two gating modes from a series of test potentials in 20 µM Ca²⁺_o (Fig. 4, A and B). These experiments indicated that Na⁺ currents in both gating modes are blocked by extracellular Ca²⁺ in a voltage-dependent fashion, with negligible block at voltages positive to 0 mV (Fig. 4 D). However, the voltage dependence of Ca²⁺ block of 2-APB–activated channels was markedly steeper, and V₅₀ more depolarized compared to STIM1-gated channels (Fig. 4 D). The anomalously high voltage dependence of Ca²⁺ block in 2-APB–gated Orai3 channels (Z₁ = 3.4) cannot be explained exclusively in terms of blocking site location because the required site depth is far greater than that of a single Ca²⁺ (z = 2) moving all the way through the electric field (Woodhull, 1973). Instead, as described previously for K⁺ channels (Martínez-Francois and Lu, 2010), the most straightforward explanation is that Ca²⁺ block in this case is coupled to the displacement of permeant Na⁺ ions. This conclusion is further supported by a striking shift in the voltage dependence of block in the E85A/D87A/E89A triple mutant, where Na⁺ ion binding in the vestibule is expected to be absent. Here, the extent of blockade was essentially invariant in a large range of voltages tested (−120 to −20 mV), but declined gradually between 0 and +100 mV (Z₁ = 0.8; Fig. 4 D). These results indicate that the steep voltage dependence of Ca²⁺ blockade in WT 2-APB–gated Orai3 channels is likely driven by the concurrent movement of Na⁺ ions that are pushed into the pore from the outer vestibule. When considered together with the finding that external Na⁺ ions affect Ca²⁺ binding to the 2-APB–gated pore (Fig. 3 B), these observations lead us to conclude that Ca²⁺ access to the high affinity site is hindered in 2-APB–gated channels as a result of the occupancy of external pore sites by Na⁺ ions. Ca²⁺ can block the pore but has to displace several Na⁺ ions into the pore to gain access to its binding site.

In contrast to 2-APB–gated channels, the lack of effect of extracellular Na⁺ concentration on Ca²⁺ block in STIM1-gated channels (Fig. 3 A) and the shallower voltage dependence Ca²⁺ block (Fig. 4 ) suggest that far fewer (if any) Na⁺ ions have to be displaced for Ca²⁺ to access its site when the channel is gated by STIM1. This doesn’t mean that Ca²⁺ access to its high affinity site is energetically more favorable in STIM1-gated channels; in fact, as described in the next section, the rate constants indicate otherwise. Nonetheless, these results reaffirm the multi-ion nature of the Orai3 pore and are qualitatively consistent with the suggestion that pore occupancy by Na⁺ ions is greatly enhanced in 2-APB–gated Orai3 channels than in STIM1-gated Orai3 channels.

Although interesting, the distinct voltage dependencies of STIM1- and 2-APB–gated channels do not readily explain the inability of the 2-APB–gated pore to discriminate between Ca²⁺ and Na⁺ ions at negative voltages. This conclusion is supported by at least two aspects of the data. First, at a negative voltage (−100 mV) where STIM1-gated channels are Ca²⁺ selective, 2-APB–gated channels are blocked to a nearly identical extent as STIM1-gated channels, yet the experiments in Fig. 1 clearly show that a large fraction of the current at −100 mV is carried by Na⁺. Second and conversely, STIM1-gated channels do not readily conduct monovalent ions (e.g., Cs⁺) even at positive voltages (>0 mV), where Ca²⁺ block in these channels is negligible. These characteristics of the voltage dependence indicate that the differing Ca²⁺ selectivities of STIM1- and 2-APB–gated channels cannot be explained directly in terms of the voltage dependence of Ca²⁺ binding to the pore.

In addition to differences in voltage dependence of blockade, STIM1- and 2-APB–activated Orai3 channels showed markedly different kinetics of Ca²⁺ blockade. At −100 mV, STIM1-activated Orai3 currents were blocked by external Ca²⁺ (20 µM) whose initial time course was fit with a single exponential function (Fig. 5, A and B). The time constant of blockade, 7 ± 1 ms (n = 5), closely matches the block rate previously seen in STIM1-activated Orai1 currents (Yamashita et al., 2007) and native CRAC channels in Jurkat T cells (Prakriya and Lewis, 2006), suggesting that the blocking mechanism is operationally similar in both channel types. For 2-APB–gated channels, measurements of the time constant of Ca²⁺ blockade was complicated by a small degree of slow activation of the monovalent current during steps to −100 mV (Fig. S2). However, lowering the holding potential from +50 to −30 mV diminished the hyperpolarization-induced activation of the 2-APB–gated monovalent current (Fig. S2), allowing us to directly examine the kinetics of Ca²⁺ blockade. These experiments showed that block kinetics were significantly faster in 2-APB–gated channels compared with STIM1-activated channels (Fig. 5, A and B; τ = 1.2 ± 0.1 ms; n = 7 cells).

We used the time constants of Ca²⁺ blockade from these experiments to ascertain the apparent second-order on and off rates of Ca²⁺ binding to the block site. If Ca²⁺ accesses a single binding site from the extracellular side at a rate k_on and exits the site at a rate k_off, then the time constant (τ) of blockade can be described by the relation:

Further, the blocked current fraction is given by the relationship:

Fig. 5 C plots the reciprocal of the averaged time constant at −100 mV against different Ca²⁺ concentrations. These plots could be well fit with straight lines whose slope represents the apparent second-order on rate constant (k_on) and the intercept the off rate constant (k_off) for Ca²⁺ binding (Eq. 2). For STIM1-gated channels, the values of these parameters from the fit were k_on = 4 × 10⁶ M⁻¹s⁻¹ and k_off = 33 s⁻¹. These rate constants are similar to previously measured values for overexpressed Orai1 channels and native CRAC channels in Jurkat T cells (Prakriya and Lewis, 2006; Yamashita et al., 2007), indicating that the rates of Ca²⁺ entry and exit from the selectivity filter do not differ appreciably between STIM1-gated Orai1 and Orai3 channels. In contrast, k_on and k_off estimated from the data for 2-APB–gated currents were 3 × 10⁷ M⁻¹s⁻¹ and 246 s⁻¹, respectively, or approximately eightfold faster than the rate constants for STIM1-activated channels. These substantially faster rate constants imply that the energy barriers for Ca²⁺ entry and exit into/from the pore are lower in 2-APB–gated channels. Further, the faster off rate implies that the dwell time of Ca²⁺ occupancy at the block site (1/k_off) is significantly lower in 2-APB–activated channels compared with STIM1-activated channels (Fig. 5 E).

In contrast to the acceleration of k_on seen in 2-APB–gated Orai3 channels, perturbation of the selectivity filter by the E81D mutation decreased k_on tenfold (k_on = 4 × 10⁵ M⁻¹s⁻¹ and k_off = 130 s⁻¹). This is similar to the previously described behavior of E106D Orai1 mutant channels, which also exhibit substantial slowing of the on-rate constant (Yamashita et al., 2007). Together with the differences in Ca²⁺ block K_i described in Fig. 2, these results indicate that the mechanisms underlying diminished Ca²⁺ selectivity are likely quite different between Orai3 channels gated by 2-APB and E81D Orai3 channels gated by STIM1. Specifically, the lower Ca²⁺ selectivity of E81D Orai3 channels can be rationalized in terms of diminished Ca²⁺ binding to the selectivity filter, in turn related to diminished rate of Ca²⁺ entry into the pore. However, this does not readily explain the lower Ca²⁺ selectivity of 2-APB–gated WT Orai3 channels, which exhibit the same K_i of Ca²⁺ blockade at −100 mV as STIM1-gated Orai3 channels (Fig. 2 E) but accelerated binding rate constants (Table 1). We will return to this issue in the Discussion.

To directly examine unblock kinetics, we depolarized the membrane to +100 mV, where blockade is expected to be nonexistent after induction of block at −100 mV. As shown in Fig. 5 D for STIM1-gated channels, a step to +100 mV initiates current recovery with a time constant of ∼10 ms, reflecting unbinding of Ca²⁺ from the block site. This recovery time specifies a k_off (= 1/τ) of 100 s⁻¹. Measurements of current recovery in 2-APB–activated channels were complicated by a small degree of deactivation of the control current (i.e., in the absence of blocking Ca²⁺ ions). When channels were blocked by 20 µM Ca²⁺, however, the deactivation phase was not seen, presumably because it is masked by recovery of current from Ca²⁺ blockade. The time course of current recovery in this phase occurred with a time constant of ∼1 ms. The true current recovery likely occurs with a faster time constant because, as noted above, current recovery in this case would be slowed by channel deactivation occurring simultaneously. Nevertheless, even this overestimated recovery time course specifies a k_off of 1,000 s⁻¹, significantly faster than the k_off estimate for STIM1-gated channels. Taken together, these results complement findings observed at −100 mV and indicate that Ca²⁺ binding to the 2-APB–gated pore is much more labile than that seen in STIM1-gated channels, and the bound Ca²⁺ comes off at a significantly higher rate.

The results presented in Figs. 3 and 4 indicate that Na⁺ occupancy at external pore sites is enhanced in 2-APB–gated Orai3 channels relative to STIM1-gated channels. Moreover, tests with the E85A/D87A/E89A triple mutant suggest that this effect is at least partially explained by Na⁺ accumulation in the outer vestibule where these residues are positioned. These findings signify that the molecular and structural features of the outer vestibule formed by the TM1–TM2 loops differ in the two gating modes. We examined this issue further by assessing the sensitivity of 2-APB–activated Orai3 channels to blockade by the trivalent lanthanide ion La³⁺. CRAC channels are potently blocked by low concentrations of lanthanides (Mason et al., 1991; Yeromin et al., 2006). Electrophysiological studies have indicated that high affinity La³⁺ blockade of Orai channels occurs primarily through binding of the trivalent ions to acidic residues in the TM1–TM2 loop segments (Fig. 3 F; Yeromin et al., 2006; McNally et al., 2009). Hence, we rationalized that La³⁺ blockade may allow us to gauge probable alterations in the structure of the TM1–TM2 loops in 2-APB–gated Orai3 channels.

These tests revealed that 2-APB–activated Orai3 channels are significantly less sensitive to La³⁺ blockade than STIM1-activated channels (Fig. 6; apparent K_i = 470 nM in STIM1-gated channels and K_i = 6 µM in 2-APB–gated Orai3 channels, both at −100 mV). Somewhat puzzlingly, the apparent Hill coefficient was also reduced from 1.1 in STIM1-activated currents to 0.7 in 2-APB–activated currents (Fig. 6 C). A Hill coefficient other than 1 is suggestive of multiple binding sites and could arise either because of negative cooperativity between multiple binding sites or progressive occupancy of multiple noninteracting sites with different binding affinities (Prinz, 2010). In the case of Orai channels, although electrophysiological studies indicate that high affinity lanthanide block is explained by blocker binding at the external vestibule (Yeromin et al., 2006; McNally et al., 2009), the recent crystal structure of Drosophila melanogaster Orai depicts a Gd³⁺ electron density in close proximity to the Glu selectivity filter in crystals soaked with 1 mM GdCl₃ (Hou et al., 2012). Thus, these studies raise the prospect that there are two lanthanide binding sites in the ion conduction pathway: a high affinity site formed by the TM1–TM2 loops, and a second, low affinity site formed by the Glu selectivity filter. Structural alterations in the TM1–TM2 loop residues may reduce but not eliminate high affinity binding entirely, increasing blocker occupancy of the second, low affinity binding site and causing an apparent reduction in the overall Hill coefficient. Consistent with this possibility, mutating the TM1–TM2 loop acidic residues (E85A/D87A/E89A) significantly reduced the sensitivity of La³⁺ blockade, but preserved low affinity blockade. This residual blockade could be well fit with a Hill relation of n = 1 for both STIM1- as well as 2-APB–gated Orai3 channels, which is consistent with the presence of a lower affinity La³⁺ site deeper in the pore. Intriguingly, the La³⁺ sensitivity of the triple mutant was considerably lower when gated by 2-APB (K_i = 189 µM) than when gated by STIM1 (K_i = 10 µM). Assuming that the residual La³⁺ binding occurs at the selectivity filter (E81), this result reveals that the affinity of ion binding (La³⁺ in this case) at the selectivity filter does in fact differ between 2-APB– and STIM1-gated channels, even though this is not obviously detected in Ca²⁺ at −100 mV (Fig. 2). Based on these observations, we conclude that the TM1–TM2 loops, which form the outer vestibule, adopt a different conformation in 2-APB–gated Orai3 channels, which diminishes La³⁺ binding in the vestibule, but enhances Na⁺ occupancy.

The faster second-order Ca²⁺ binding rate constants and increased Na⁺ ion pore occupancy in 2-APB–gated channels led us to next consider whether the energy barriers for ion flow are diminished in this gating mode. To test this possibility, we estimated the unitary conductance of STIM1- and 2-APB–gated Orai3 channels using nonstationary fluctuation analysis (Sigworth, 1980). Orai3 channels were activated either by 2-APB (50 µM) or by STIM1 after store depletion, and monovalent current noise was analyzed in 200-ms sweeps acquired at −100 mV (Sigworth, 1980). The relationship between current variance (σ²) and the P_o of the channel can be described by the relation:

where, i is the unitary current amplitude and N is the number of channels. Eq. 4 indicates that variance is related to P_o by the well-described parabolic relationship, with variance reaching a maximum when P_o is 0.5. Substituting the term for N in terms of the total current (I = iNP_o), we get the relation:

A representative σ²/I plot of Orai3 current activated by 2-APB is illustrated in Fig. 7 B. The data show that Orai3 channel activation increases σ² as expected. Yet, the σ² versus I plots did not show a curvature expected from the canonical parabolic σ²/I relationship. Instead the data could be well fit with straight lines with a mean slope of 56 ± 6 fA (n = 12 cells; Fig. 7 B). As described previously (Prakriya and Lewis 2006), a linear σ²/I relationship can arise either because channel P_o is always <<1 when the macroscopic current changes because of an increase in P_o (Eq. 5) or, alternately, because of a change in channel number N at any constant value of P_o (Eq. 4). To distinguish between these possibilities and properly interpret the linear σ²/I relation, an estimate of P_o is needed. For this, we used blockade of Orai3 currents by micromolar concentrations of extracellular Ca²⁺ to define the position along the parabolic σ²/I relationship where Orai3 channels operate.

If 2-APB activates Orai3 channels to a P_o > 0.5, then partial blockade of current with a channel blocker should result in an increase in current variance, as the variance-mean current plot moves leftward on the parabolic σ²/I plot. Only after further blockade, when P_o is attenuated to values <0.5, should variance be expected to decline. Consistent with this prediction, modest blockade of the monovalent current by a relatively low dose of extracellular Ca²⁺ resulted in increase in current noise (Fig. 8 A). In contrast, strong blockade at higher Ca²⁺ concentrations decreased current variance. This result indicates that the 2-APB–activated current has a P_o > 0.5. In fact, estimates of P_o, i, and N from fits of the data similar to that illustrated in Fig. 8 B revealed values of P_o = 0.7 ± 0.05, i = 283 ± 61 fA, and n = 21,444 ± 4,521 (n = 4 cells). Power spectrum analysis of control and Ca²⁺-blocked traces indicated that no high frequency components of current noise were missed under the 10-kHz low-pass filtering conditions used for these recordings (Fig. S3 B). These results indicate that 2-APB–activated Orai3 channels have a high P_o.

Measurements of STIM1-gated Orai3 channel variance revealed that these channels also have a high P_o (Fig. 8, C and D). Estimates of P_o, i, and N from fits of the data similar to that illustrated in Fig. 8 D revealed values of P_o = 0.73 ± 0.03, i = 71 ± 1 fA, and n = 22,537 ± 5,670 (n = 5 cells). Power spectrum analysis revealed that STIM1-gated currents exhibit lower frequency noise components than 2-APB–gated currents (Fig. S3), suggesting that despite a similar P_o, 2-APB–gated channels exhibit flickery openings. Importantly, the estimated unitary current of STIM1-gated channels was significantly smaller (71 fA) than unitary current of 2-APB–gated channels (∼280 fA).

These results indicate that in the absence of extracellular blocking divalents, 2-APB–gated channels exhibit a significant increase (fourfold) in the rate of Na⁺ conduction compared with STIM1-gated channels, which is indicative of a substantial decrease in the energy barriers for Na⁺ ion conduction. Moreover, the high P_o and linear σ²/I plots seen for both 2-APB as well as store-operated Orai3 channels indicate that both modes of activation (at least in the time scales of the 200-ms voltage sweeps used here) occur through a mechanism involving stepwise recruitment of channels to a high P_o mode, rather than a monotonic increase in P_o. This suggests that the gating mechanisms in both modes of channel activation may be operationally similar.

In the prevailing model of CRAC channel selectivity, high Ca²⁺ selectivity is achieved through preferential binding of Ca²⁺ at the channel selectivity filter formed by a ring of Glu residues, causing occlusion of Na⁺ flux through the pore (McNally and Prakriya, 2012). Central to this idea is the notion that selectivity is governed by high affinity of Ca²⁺ binding to the selectivity filter. In agreement with this possibility, a previous study found that mutation of the selectivity filter (E106D Orai1) results in parallel decreases in both Ca²⁺ selectivity and the affinity of Ca²⁺ blockade of Na⁺ fluxes (Yamashita et al., 2007), as would be expected if the strength of Ca²⁺ binding directly influences the ability of bound Ca²⁺ ions to impede Na⁺ flux. Such a thermodynamically driven view of Ca²⁺ selectivity predicts that channels with lower relative Ca²⁺ selectivity should display diminished Ca²⁺ blockade of Na⁺ flux. Yet, here we find that at a voltage (−100 mV) where 2-APB–gated channels exhibit significant monovalent conduction, inhibition of Na⁺ Orai3 current by micromolar concentrations of extracellular Ca²⁺ is similar between the Ca²⁺-selective STIM1-gated and the nonselective 2-APB–gated Orai3 channels. This result indicates that equilibrium blockade of Na⁺ flux, as measured by K_i of Na⁺-CRAC inhibition, cannot account for the differing Ca²⁺ selectivity of the two gating modes.

Several explanations can, in principle, be considered for the different Ca²⁺ selectivity of STIM1- and 2-APB–gated channels. One explanation is that selectivity is influenced not only by the steady-state Ca²⁺ binding affinity to the selectivity filter but also by the kinetic properties of ion entry. Many studies have considered the idea that selectivity is governed by the rates of ion entry into the channel, with less selective channels exhibiting slower rates of ion entry into the selectivity filter (Grabe et al., 2006; Nimigean and Allen, 2011). Applying this idea to Orai3 channels, one scenario is that the weakly Ca²⁺-selective 2-APB–gated Orai3 channels exhibit a slower rate of Ca²⁺ entry into the high affinity binding site than the highly Ca²⁺ selective STIM1-gated channels. This appears to the case for E81D Orai3 mutant channels, which exhibit a 10-fold slower k_on (10⁵ M⁻¹s⁻¹), reaffirming the importance of high affinity binding at the selectivity filter for Ca²⁺ selectivity. However, this explanation cannot account for the lower Ca²⁺ selectivity of 2-APB–gated WT Orai3 channels because our estimates of the second-order on rates suggest that k_on is actually faster in 2-APB–gated channels (∼10⁷ M⁻¹s⁻¹) than in STIM1-gated channels (∼10⁶ M⁻¹s⁻¹).

A closely related kinetic possibility is that channels with lower Ca²⁺ selectivity exhibit transient (rather than stable) binding compared with channels with high Ca²⁺ selectivity. Indeed, estimates of Ca²⁺ dwell times extrapolated from the k_off measurements indicate that 2-APB–gated channels exhibit considerably shorter dwell times than STIM1-gated channels. If this is also true at millimolar concentrations of extracellular Ca²⁺ where Ca²⁺ permeation occurs (which admittedly is an unsubstantiated assumption), the decreased dwell time of Ca²⁺ occupancy could lead to diminished Ca²⁺ selectivity by allowing a higher level of net Na⁺ permeation between Ca²⁺ binding events at the block site. Because each bound Ca²⁺ ion would be severalfold slower to leave in STIM1-gated channels, it could prevent many more Na⁺ ions from passing through than in 2-APB–gated channels. However, it is also possible that the decreased Ca²⁺ block dwell time is a consequence of enhanced Na⁺ permeation driven by repulsion from Na⁺ ions destabilizing Ca²⁺ binding (Fig. 3 B). In this case, a faster entry rate for Na⁺ would naturally bias selectivity toward Na⁺.

A third possibility is that the CRAC channel pore contains additional Ca²⁺ binding sites distinct from the single block site examined here. In this scenario, new Ca²⁺ binding sites that are not readily detected at low (micromolar) concentrations of extracellular Ca²⁺ used to gauge block come into play at millimolar concentrations to confer high Ca²⁺ selectivity of STIM1-gated channels. This argument is undermined by the fact that molecular and structural studies reveal only a single locus for Ca²⁺ binding in the pore composed of conserved Glu residues in TM1 (E106 in Orai1 and E81 in Orai3). The reminder of the pore (TM1) is lined with hydrophobic and basic residues that are not supportive for Ca²⁺ binding. However, a key limitation of the current study is the experimental disconnect between permeability measurements, which can only be performed at high (millimolar) Ca²⁺ concentrations and Ca²⁺ block measurements, which are performed at low (micromolar) Ca²⁺ concentrations. This disconnect is exemplified by the vast difference in Ca²⁺ affinities seen at low and high extracellular Ca²⁺ concentrations: the K_d of Ca²⁺ blockade of Na⁺ flux is ∼20 µM, but saturation of I_CRAC occurs at millimolar Ca²⁺ concentrations (K_m = 1–3 mM; Hoth and Penner, 1993; Premack et al., 1994; Fierro and Parekh, 2000). In fact, this is a common theme in many classes of ion channels: as multi-ion occupancy of the pore increases at high ionic strengths, the apparent affinity of Ca²⁺ binding to the pore declines (Hille 2001), likely through repulsive interactions between closely spaced ions in the pore. One plausible scenario through which this could occur is if the side chains of the six Glu residues in TM1 cluster into two groups along the axis of the pore to form two binding sites. Because these sites would be in close proximity, electrostatic repulsion between the closely spaced Ca²⁺ ions could reduce the apparent affinity of Ca²⁺ binding at the selectivity filter (Almers and McCleskey, 1984; Sather and McCleskey, 2003), potentially explaining the much lower dependence of I_CRAC on extracellular Ca²⁺.

A final possibility is that the change in selectivity arises directly as a consequence of a change in the charge/volume ratio because of structural enlargement of the pore. Some models of Ca²⁺ selectivity have concluded that Ca²⁺ selectivity in L-type voltage-gated Ca²⁺ channels is driven by a charge/space competition mechanism wherein selectivity arises from a balance of electrostatics and the excluded volume of ions in the crowded selectivity filter (Nonner et al., 2000; Sather and McCleskey, 2003; Malasics et al., 2009). Ca²⁺ is selected over Na⁺ because its higher charge more effectively neutralizes the negative charge within the narrow confines of the selectivity filter than the monovalent charge of Na⁺ ions. These models predict that squeezing the volume of the selectivity filter by narrowing the pore should generally result in preference for Ca²⁺ over Na⁺ ions. Conversely, increasing the pore size (as seen in the 2-APB–gated channels) should result in greater monovalent occupancy because of relaxation of the excluded volume and space/charge constraints. Such a model is certainly qualitatively consistent with our experimental observations and more work is needed to illuminate the extent to which volume exclusion and the higher charge/size ratio contributes to Ca²⁺ selectivity in CRAC channels.

To get an initial understanding of how alterations in the energetics of ion conduction influence Ca²⁺ selectivity, we considered a simple barrier model of the pore based on Eyring rate theory. Rate theory descriptions of the pore in which the ion conduction pathway is depicted as a series of discrete energy wells separated by barriers that limit ion conduction have been widely used to understand the major forces that regulate selectivity and conduction in ion channels (Begenisich and Cahalan, 1980; Hille, 2001). Such models admittedly have limitations in their ability to capture the molecular attributes of the pore such as locations and chemistry of binding sites and the physical basis of energy barriers (Hille, 2001). Nonetheless, they often yield useful insights on the energetic influences of binding sites and barriers and can provide qualitative forecasts of the effects of their changes on ion selectivity. We first considered the simplest model possible: a two-barrier, one binding site model to examine blockade of Na⁺ ions by micromolar concentrations of Ca²⁺. The Gibbs free energy of the Ca²⁺ site can be calculated from the relation:

Substituting the observed values of K_i at −100 mV (Table 1), the relationship specifies a well depth of 11–12 RT units (depending on the extracellular Na⁺ concentration) for the Ca²⁺ site. G_Ca declines to 9 RT units in the E81D Orai3 mutant channels (K_i of ∼110 µM).

The rate at which an ion moves from one site to another equals the product of the rate constant for this transition and the probability of occupancy of the source site by that ion. For the energy profile diagrammed in Fig. 9 A, the rate constant k₁ is given by:

indicating that entry rate declines exponentially with increasing barrier height. v represents the maximum possible reaction rate set by the thermal vibrational frequency and has the value of 5.8 × 10¹² s⁻¹ at room temperature (v = kT/h, where k is the Boltzmann constant and h is the Planck’s constant). For STIM1-gated channels, if the second-order k_on is approximated as the rate of Ca²⁺ entry into the pore, then the energy of this transition is given by relation:

yielding an outer barrier height of ∼14 RT. Thus, a Ca²⁺ ion lodged at the well faces a total energy barrier of 12 + 14 = 26 RT units to escape back outside. To estimate the height of the cytoplasmic barrier, we used k_off estimated from the single-site model (33 s⁻¹). Under our experimental conditions (V_m = −100 mV), if we assume that this is due escape of Ca²⁺ into the intracellular compartment, this rate specifies an energy of 26 RT units, reflecting the total energetic barrier faced by Ca²⁺ ions lodged at the binding site. Thus, the height of the inner barrier is 26 − 12 or 14 RT units, roughly identical to the external barrier. The values of k_on and k_off increase to 3 × 10⁷ M⁻¹s⁻¹ and 246 s⁻¹, respectively, in 2-APB–gated channels. From Eq. 8, these correspond to energies of 12 and 24 RT units for the entry and exit of Ca²⁺ ions from the binding site. Thus, barrier heights for Ca²⁺ decline by roughly 2 RT units in 2-APB–gated channels compared with STIM1-gated channels.

The single-site model yields Ca²⁺ unbinding rates (33 s⁻¹ in STIM1-gated channels and 245 s⁻¹ in 2-APB–activated channels) that are vastly lower than the Ca²⁺ flux rates observed at millimolar extracellular Ca²⁺ concentrations (∼11,000 s⁻¹ calculated from a unitary current of ∼3.7 fA at −80 mV [Zweifach and Lewis, 1993]). This paradox, well described in voltage-gated Ca²⁺ channels (Sather and McCleskey, 2003), indicates that single-site models cannot model permeation of Ca²⁺ at high Ca²⁺ concentrations. We therefore considered a four-barrier three-site model similar to the one previously described for voltage-gated Ca²⁺ channels with a single high affinity binding site flanked by two low affinity sites, allowing occupancy of the pore by multiple ions (Dang and McCleskey, 1998). Like voltage-gated Ca²⁺ channels (Sather and McCleskey, 2003), CRAC channels exhibit two Ca²⁺ affinities—a relatively high affinity of 25 µM (−11 RT units) for Ca²⁺ blockade of Na⁺ currents (Su et al., 2004; Prakriya and Lewis, 2006; Yamashita et al., 2007) and a low affinity (we used a value of 3.3 mM or −6 RT units [Hoth and Penner, 1993]) for saturation of the Ca²⁺ current. Thus, as extracellular Ca²⁺ is raised from micro- to millimolar concentrations, the apparent affinity of the pore declines by several orders of magnitude, ensuring that ions are not permanently trapped in the pore. In studies of voltage-gated Ca²⁺ channels, this change has been modeled as arising either because of electrostatic repulsion between two closely spaced Ca²⁺ ions located at the selectivity filter (Almers and McCleskey, 1984; Friel and Tsien, 1989) or because of the presence of multiple shallow, low affinity sites flanking a single high affinity site (Dang and McCleskey, 1998), conferring in effect a staircase of multiple rising steps at the exits of the free-energy profile. We used the latter model to evaluate whether changes in barrier heights influence the relative Ca²⁺ selectivity of CRAC channels.

Values of the barrier heights (∼14 RT units) and the central well depth (∼12 RT units) were kept close to estimates in the single-site model. The validity of these energies for the high Ca²⁺_o condition was confirmed from analysis of the rates of permeation at high Ca²⁺ concentrations. At high (millimolar) Ca²⁺ concentrations under which Ca²⁺ permeation occurs, the off rate for Ca²⁺ conduction is k_off = k_on × K_M. Substituting the values of k_on (4 × 10⁶ M⁻¹s⁻¹) and K_M (3.3 mM) yields a k_off of ∼12,000 ions s⁻¹. This rate is very close to the observed flux rate of ions from the unitary Ca²⁺ current amplitude determined from noise measurements (∼3–4 fA at −80 to −100 mV, or ∼11,000 s⁻¹; Zweifach and Lewis, 1993; Prakriya and Lewis, 2002), indicating that the k_on estimate obtained from the single-site model provides a reasonable rate of Ca²⁺ entry into the pore even at high Ca²⁺ concentrations. Likewise, a flux rate of 12,000 ions/s corresponds to a total energy barrier of ∼20 RT units from Eq. 8. As noted in the previous paragraph, Ca²⁺ flux saturates with a K_M = 3.3 mM, which specifies a well depth of −6 RT units during Ca²⁺ permeation (Eq. 6). With a total energy barrier of 20 RT units from the deepest to the highest point of the energy landscape during Ca²⁺ permeation and an apparent well of −6 RT units, the inner barrier height should be 20 − 6 = 14 RT units, identical to value estimated from the single-site model. We placed the wells toward the extracellular region of the pore (at electrical distances of 0.15, 0.2, and 0.3 units) to account for the known positions of E81 and the TM1–TM2 acidic residues (E85/D87/E89) in the outer vestibule (Fig. 3 F).

For Na⁺ ions, the unitary Na⁺ currents of STIM1- and 2-APB–gated channels from noise analysis are 0.08 and ∼0.3 pA, respectively, corresponding to off-rate energies of 17 and 14 RT units, respectively. Our tests indicate that the Na⁺ current saturates with a K_d of ∼90 mM (or −2.4 RT units; unpublished data). With the reasonable assumption that this binding is predominantly governed by the selectivity filter, the barrier heights thus can be determined to be ∼14 and 12 RT units in STIM1- and 2-APB–gated channels, similar to the values for Ca²⁺. The major difference in energy profiles experienced by Ca²⁺ and Na⁺ is thus the absence of a high affinity central binding site for Na⁺.

With these values of barrier heights and well depths, we next computed the I–V profiles using the four-barrier, three-well model (Dang and McCleskey, 1998). For STIM1-gated channels, the simulated I–V profiles qualitatively matched the experimental I–Vs seen in Orai3 channel activated by STIM1 (Fig. 9, B and C). As predicted from the deep well depth and high barriers for ion flow, the I–Vs showed strong inward rectification with V_rev at very positive voltages, typical of STIM1-gated Orai channels. The proportion of current carried by monovalent ions was negligible both at negative and positive voltages (Fig. 9 B). Decreasing the well depth by 3 RT units to mimic the diminished Ca²⁺ block affinity of the E81D Orai3 mutant reduced Ca²⁺ selectivity and resulted in greater monovalent permeation as seen experimentally (Fig. 9 B). This latter finding reaffirms the critical importance of the high affinity Ca²⁺ binding site for Ca²⁺ selectivity. Thus, the model qualitatively reproduces many key experimental features of STIM1-activated Orai3 currents.

To understand why the 2-APB–gated channels exhibit lower Ca²⁺ selectivity, we next altered the parameters of the model to mimic those seen in 2-APB–gated channels. Lowering the outer barriers by 3 RT units for both Na⁺ and Ca²⁺ ions while leaving the well depths unchanged elicited profound changes in ion selectivity (Fig. 9 C). The theoretical I–Vs revealed a large leftward shift in the V_rev and a significant outwardly rectifying monovalent current, in accordance with the properties observed in 2-APB–gated channels. Importantly, the fraction of inward current carried by Ca²⁺ was significantly diminished compared with the model for STIM1-gated channels. A shallower well depth resulted in even less Ca²⁺ selectivity and greater monovalent ion permeation (unpublished data), as seen in the E81D mutant, again reaffirming the importance but not the exclusive role of the high affinity site for Ca²⁺ selectivity. Thus, these findings reveal that in addition to the well-established influence of the well depth, barrier heights are critical determinants of Ca²⁺ selectivity in CRAC channels. In particular, the results indicate that structural changes that produce broad alterations in barrier heights for permeant ions can elicit robust effects on Ca²⁺ selectivity, even if the well depth for Ca²⁺ binding is unchanged. We consider a possible mechanistic basis of this feature in the following paragraph.

Ca²⁺ selectivity in voltage-gated Ca²⁺ channels, which have a relatively large pore, yet are highly selective for Ca²⁺, has been traditionally viewed as a prime example of selectivity by affinity. In these channels, the high pore affinity for Ca²⁺ (K_i of ∼0.7 µM) is believed to form the basis of the physiologically important selectivity of the channels for Ca²⁺ over the more prevalent Na⁺ and K⁺ ions (Sather and McCleskey, 2003). In CRAC channels, affinity of Ca²⁺ binding (K_i of Na⁺ current blockade) is at least 25-fold lower, yet these channels exhibit comparably high Ca²⁺ selectivity, suggesting that CRAC channels achieve selectivity through additional mechanisms. Based on the experimental and modeling results presented in this study, we propose that in addition to specific Ca²⁺ binding, high Ca²⁺ selectivity of CRAC channels may be attributed, at least in part, to the rejection of all ions (both preferred and non-preferred) by high energy barriers. Enhancing the Na⁺ and Ca²⁺ flux rates by lowering the entry and exit barriers paradoxically reduces Ca²⁺ selectivity as seen in the 2-APB–gated channels. This may be related to the decreased residence time of Ca²⁺ occupancy at the block site, thereby increasing opportunistic flow of Na⁺ ions between block events. Alternately, it could be related to enhanced Na⁺ ion entry rates into the pore, which could destabilize Ca²⁺ binding through a knockoff effect. More quantitative and modeling studies are needed to examine the mechanistic underpinnings of this effect, but the results presented in this study provide a basis for testing these and other models.

What is the physical basis of the wells and barriers in the Eyring model? It is straightforward to expect that the high affinity Ca²⁺ binding site is formed by the selectivity filter, and the superficial cation binding site can also be plausibly assigned to the vestibule acidic residues (E85/D87/D89; Fig. 3 F). As argued above, an inner cation binding site can also be envisioned to arise at high Ca²⁺ concentrations if the carboxylate side chains of the Glu residues at E81 cluster into two groups to form separate binding sites. It is more difficult, however, to envision the physical basis of the energy barriers, which may in fact not have any tangible physical correlate. The outer “barrier” may simply reflect the ease with which ions are dehydrated. The inner barrier may simply arise from a location in the pore where ions pause momentarily before moving to the next energy minimum. Still, these unknowns do not diminish the lessons of the Eyring rate analysis, for they illustrate the plausibility that both kinetic and thermodynamic factors shape Ca²⁺ selectivity of Orai channels.

Analysis of current noise indicates that activation of Orai3 channels by 2-APB occurs through a mechanism qualitatively similar to the stepwise recruitment of silent channels to a high P_o mode described for gating of native CRAC channels by store depletion (Prakriya and Lewis, 2006) and activation of Orai3 channels by STIM1 (this study). In both gating modes, channel opening occurs slowly over time scales of seconds and involves recruitment of closed channels to a long-lasting high P_o state (P_o of ∼0.7). One simple explanation is that the slow gating mode of Orai channels is a channel intrinsic behavior, independent of the activation stimulus. However, given the strong nonlinearity of 2-APB activation (Hill coefficient of ∼8; Fig. 1 A), an alternative possibility is that the abrupt opening of single channels to the high P_o state reflects the concerted action of multiple ligand molecules (two STIM1 molecules per Orai monomer for STIM1-gated channels [Li et al., 2010; Hoover and Lewis, 2011] and at least eight 2-APB molecules per channel for 2-APB–gated channels) that triggers stepwise channel opening. Either way, this operational similarity suggests that despite differences in the permeation and selectivity of open channels, STIM1 and 2-APB both use a similar molecular mechanism for gating the channel, possibly through related transduction mechanisms wherein the steps between ligand binding and pore opening are identical. This scenario is also consistent with a recent suggestion that 2-APB and STIM1 both use a graded activation mechanism to cause pore opening (Amcheslavsky et al., 2013). More studies are needed to elucidate the complete story of the concerted action of STIM1 and 2-APB on channel opening, taking into account the channel stoichiometry and the presence of multiple ligand binding sites on each Orai subunit.

The authors are grateful to Ted Begenisich for providing the computer program for the four-barrier, three-site Eyring rate model and to Ted Begenisich, Jon Sack, Chris Lingle, and Ed McCleskey for helpful consultations during the course of this work.

This research was supported by National Institutes of Health (grant NS057499).

The authors declare no competing financial interests.

Sharona E. Gordon served as editor.