Chemoselective tarantula toxins report voltage activation of wild-type ion channels in live cells

We chose to synthesize VST reporters from the tarantula peptide guangxitoxin-1E (GxTX), a potent modulator that selectively binds Kv2 channels (29, 30). In combination with patch-clamp electrophysiology, GxTX pharmacology is currently the most stringent test of whether Kv2 channels contribute to electrical signaling (13). In neurons and pancreatic islet cells, GxTX inhibits delayed rectifier current with minimal effects on other current types and has been recently used to reveal unexpected contributions of Kv2 channels to electrical signaling (13, 31, 32). Based on the effects of GxTX on Kv2 channel gating and the behavior of radiolabeled GxTX (30), we suspected GxTX might bind much more weakly when the channels become activated by voltage. To test this, we synthesized fluorescently labeled GxTX to determine if this VST could optically report Kv2 channel localization and activation.

In an initial attempt at generating GxTX variants, we expressed GxTX with an N-terminal hexahistidine tag and a C-terminal protease cleavage site in bacteria (SI Materials and Methods), but this peptide had little activity against Kv2.1 channels (Fig. S1A). This suggested that attachment of reporters to side chains, rather than the termini, was most likely to result in bioactive GxTX. We turned to chemical synthesis of GxTX to exploit nonnative amino acids that could serve as chemoselective conjugation points. We synthesized a GxTX variant with its sole methionine replaced by an oxidation-resistant norleucine isostere and found it altered Kv2.1 function by producing a strikingly positive shift in activation voltage (Fig. 1B), similar to wild-type GxTX (27, 29, 33). To avoid methionine oxidation complications, norleucine 35 GxTX was used as the background for further variants. To identify GxTX residues that can be substantially modified without disrupting toxin folding or activity, we introduced chemoselective amino acids into a series of variants. Mutagenesis studies of homologous VSTs (28, 34) guided selection of residues on the hydrophilic face of GxTX (33) that might have minimal impacts on channel binding. We substituted orthogonally protected cysteine residues [Cys(Acm)] at seven hydrophilic sites (Fig. 1C) and tested for activity against Kv2.1 channels. Of these seven Cys(Acm) mutants, five acted similar to GxTX at 100-nM concentration, shifting the voltage midpoint of Kv2.1 conductance to greater than +40 mV (Fig. S1A). When Lys27 was replaced with Cys(Acm), more than 300 nM was required to induce a perceptible gating shift; GxTX with Asn32 replaced by Cys(Acm) was inactive at concentrations up to 1 μM. Regions of GxTX that were tolerant to perturbations were suggested by mapping substitutions onto a GxTX NMR structure (33) (Fig. 1 D and E).

We generated a speculative model of GxTX bound to its Kv2.1 channel binding site (35) using Rosetta computational methods (36–39). The lowest energy resultant docked structure served as a structural hypothesis for how GxTX interacts with each subunit of Kv2.1 (Fig. 1F). Although GxTX stabilizes resting conformations of the voltage sensor (35), this model represents GxTX binding to what is proposed to be an activated conformation of the voltage sensor (40). If the orientation of the toxin toward the channel remains similar in different states, it could be useful for predicting points of attachment for reporter molecules. This docking configuration places intracystine loop 2 most distal from the channel and membrane, suggesting that bulky reporter conjugates attached to loop 2 might be less likely to interfere with channel binding (Fig. 1G).

To create probe attachment sites on loop 2, we incorporated chemoselective groups, either an alkyne or a free thiol, for conjugation of fluorescent and affinity probes. To add an alkyne, we synthesized GxTX with the unnatural amino acid propargylglycine, for labeling with azide-bearing probes using Cu(I)-catalyzed Click reactions. Whereas many Click variations have been reported, Cu(I) salts are known to catalyze the disproportionation of thiol compounds (41), and we sought to identify conditions that would neither reduce nor scramble the cystine knot disulfides. We found that addition of the water-soluble Cu(I)-chelator bis[(tertbutyltriazoyl)methyl]-[(2-carboxymethyltriazoyl)methyl]-amine (BTTAA) (42) yielded GxTX Click conjugates that retained function against Kv2.1 channels. We attached a monofunctional 5-kDa polyethylene glycol (PEG5K) by its azide group to GxTX Pra13 (Fig. S1 B and D). The product (GxTX-PEG5K) was active against Kv2.1 channels, thereby inhibiting channel opening by shifting the midpoint of conductance to voltages similar to wild-type GxTX (Fig. 1H).

As an alternate strategy for chemoselective conjugation, we revealed a free “spinster” thiol on the surface of the peptide (43). Removal of the Acm protecting group has typically been catalyzed by Ag⁺ and facilitated with excess reductant (44), or oxidant (45), but these conditions are incompatible with preserving existing disulfides while obtaining a reduced surface thiol. To achieve the required mixed redox state, we removed Acm after GxTX folding and disulfide formation by displacing the Acm group with Ag⁺ followed by precipitation of Ag⁺ with Cl⁻ under acidic conditions, breaking the Ag–S bond without reducing or scrambling the disulfides (Fig. S1C). Maleimide-functionalized fluorophores, including tetramethylrhodamine and dylight550, were condensed with the free thiol, and Cys13-fluorophore GxTX conjugates retained activity against Kv2.1 (Fig. 1I). These results demonstrate that variants of GxTX can be folded and chemoselectively ligated to reporter molecules while retaining bioactivity against ion channels.

To determine whether these VST conjugates could identify ion channels in live cells, we tested whether GxTX–fluorophores would colocalize specifically with Kv2.1 ion channels. When two mammalian CHO-K1 cell lines, one expressing Kv2.1 and the other expressing a blue fluorescent protein (BFP), were briefly treated with 100 nM of a GxTX–fluorophore conjugate, the GxTX fluorescence remained associated only with Kv2.1-expressing cells (Fig. 2A). When CHO-K1 cells were transiently transfected with Kv2.1-GFP, the channels appeared in clusters on the cell surface (Fig. 2B). GxTX localized tightly with these clusters (Pearson’s overlap coefficient r = 0.796), whereas surrounding untransfected cells had little GxTX fluorescence. In living neurons, a similar phenomenon was seen (Fig. 2C), although the colocalization with Kv2.1-GFP was not as complete (Pearson’s overlap coefficient r = 0.589), possibly owing to the presence of endogenous Kv channels. Whether tarantula VSTs such as GxTX have a high affinity for channels is a point of active debate (27, 46, 47). GxTX, like many other tarantula peptides, partitions into the outer leaflet of the plasma membrane, and it has been difficult to discern how much of its affinity for channels is due to membrane partitioning versus channel binding interactions (27). The colocalization of GxTX with Kv2.1 in cultured cells and neurons demonstrates that a strong, specific interaction occurs between the tarantula peptide and the ion channel voltage sensor. This precise colocalization makes GxTX–fluorophores viable ion channel probes.

We hypothesized that GxTX binding is coupled to channel activity: when the membrane voltage is depolarized, voltage sensors adopt activated conformations and the binding of the inhibitory VST is weakened (Fig. 3A). Studies have concluded that the affinity of ion channels for certain VSTs is voltage dependent and that depolarizing pulses can release inhibitory VSTs from the channels (18–21), but the precise changes in affinity have never been measured. To measure the degree to which GxTX affinity depends on voltage activation, the fraction of Kv2.1 channels bound by GxTX was deduced from K⁺ current at different holding potentials (SI Materials and Methods). At a holding potential of −100 mV, GxTX was found to inhibit K⁺ current during brief test pulses with a half-maximal inhibitory concentration (IC₅₀) of about 2 nM (Fig. 3B, black). In cells held at 0 mV, the IC₅₀ rose beyond 100 nM (Fig. 3B, blue). This shift of GxTX affinity with voltage indicates that GxTX indeed binds to Kv2.1 in a conformation-dependent manner.

To better understand the thermodynamics of GxTX interactions with Kv2.1, we further analyzed the dose dependence of electrophysiological responses. Whereas four tarantula VSTs may bind a Kv channel tetramer (Fig. 1F) (48), only one toxin is required to eliminate current at neutral voltage (27). Due to this 4:1 stoichiometry, conductance at 0 mV is related to the GxTX dissociation constant (K_d) by the fourth power of the probability of a subunit being bound (SI Materials and Methods, Eq. S2). When this established analysis (48) is applied to our dose–response data set, it reports a 65-fold shift in the K_d for GxTX at −100 mV compared with 0-mV holding potential (Fig. 3B, curves). Voltage activation by depolarization to 0 mV induces Kv2.1 channels to reach a new distribution between closed, open, and inactivated conformations. Toxin-treated channels do not appreciably open at 0 mV (Fig. 1B), suggesting the toxin binding will be even weaker to more activated conformations at positive voltages. Whereas the relative affinities of the many closed, open, and inactivated channel conformations could not be distinguished with these experiments, our results clearly indicate that voltage activation lowers the affinity of GxTX for Kv2.1.

To determine whether the voltage-dependent shift in the GxTX K_d was due to a change in the binding rate, dissociation rate, or both, we measured kinetics of Kv2.1 inhibition and recovery at different voltages, and calculated microscopic binding–dissociation rates. The time course of inhibition and recovery were fit with a function that extracts a microscopic binding rate (k_on) and dissociation rate (k_off) (SI Materials and Methods, Eq. S3). When cells were held at −100 mV, GxTX inhibition occurred faster than at 0 mV (Fig. 3C), consistent with the toxin's weaker affinity at this depolarized voltage. This suggests that access of the VST to its binding site is compromised when the voltage sensors are activated. At both voltages, the time course of inhibition accelerated as GxTX concentration increased (Fig. 3D). Association rate rose linearly with GxTX concentration, consistent with its expected first-order dependence (Fig. 3D). k_on became 13-fold slower when the voltage was raised to 0 mV (Table 1). To determine k_off accurately, dissociation was measured after toxin washout. The return of Kv2.1 current after GxTX washout was accelerated when long pulses to 0 mV were included in the voltage stimulus protocol (Fig. 3E; SI Materials and Methods). The calculated k_off during these 0-mV pulses was 70-fold faster than at −100 mV (Fig. 3F). Thus, toxin binding and dissociation are both affected by voltage, which indicates that GxTX's state specificity is due to its slow binding and rapid dissociation from voltage-activated channels. Fundamentally, the ratio of k_off/k_on determines the K_d of a bimolecular process. At both voltages tested, the ratio of k_off/k_on (calculated from kinetics, Eq. S3) is within an order of magnitude of K_d (calculated from equilibrium measurements, Eq. S2), but the values do not precisely match. This difference suggests the thermodynamic model that forms the basis of parameter extraction requires further refinement. Our calculations assume that toxin binds to each of four channel subunits independently and that toxin affinity is static over time. The independence of a tarantula VST binding to Kv2.1's four subunits has been experimentally validated (16). However, the voltage-dependent dissociation of tarantula VSTs from mutant Kv2.1 channels was found to be dependent on pulse duration and frequency (21). Progression through multiple resting, activated, and inactivated conformations is expected to result in affinity change over time. Although our model is oversimplified, it captures the basic features of activation-driven affinity change. The parameters extracted from kinetic measurements with this model show that channel activation alters the rates of association and dissociation in rough agreement with equilibrium measurements, demonstrating that the toxin has high affinity for resting channels and low affinity for activated channels.

We next tested whether we could optically measure channel activity with VST–fluorophore conjugates. The optical reporting ability of a GxTX–fluorophore was tested on CHO cells expressing Kv2.1 in whole-cell voltage-clamp fluorometry experiments. GxTX-dy550 was chosen for these experiments because it was found to have electrophysiological properties similar to GxTX (Fig. 1I and Fig. S2). To optically measure VST binding to voltage-clamped cells, we quantitated the fluorescence that remained after application and washout of 100 nM GxTX-dy550 (Fig. 4A). When cells were held at −100 mV, fluorescence slowly decayed. When cells were depolarized to 0 mV to accelerate toxin dissociation, fluorescence loss markedly accelerated (Fig. 4B). By fitting an exponential function (Eq. S4) to the fluorescence decays, rates of fluorescence change, k_{−100 mV} and k_{0 mV}, were extracted (Fig. 4C). The similar kinetics of k_{−100 mV} fluorescence decay and k_off of GxTX-dy550 at −100 mV (Fig. S2C) is consistent with fluorescence decay resulting from VST dissociating from channels. A clearly significant difference in fluorescence decay was seen after voltage change, with k_{0 mV} being 53-fold faster than k_{−100 mV}. We conclude that the voltage-driven fluorescence change results from GxTX dissociation due to Kv2.1 voltage activation.

To test whether Kv2.1 activation can be reversibly reported by this fluorescent VST, we bathed cells continuously in GxTX-dy550 and switched back and forth between voltages. In 10 nM GxTX, fluorescence was concentrated on the surface of Kv2.1-expressing cells (Fig. 4D) and remained constant over time (Fig. 4E) as the VST slowly bound to and dissociated from Kv2.1 channels. When voltage-clamped cells were held at a constant voltage, fluorescence intensity also remained constant. Varying voltage resulted in dynamic changes of cell fluorescence, indicating that the fluorescent VST optically reported the change in activation state of the ion channel (Fig. 4 D and E and Movie S1). The rate of fluorescence change (k_∆F) varied with voltage and saturated at positive voltages (Fig. 4F), consistent with GxTX localization reporting activation of Kv channels rather than the membrane voltage. Thus, the activity of VGICs can be reported by state-selective fluorescent VST localization to live cells.

To test whether our VST probe can report channel activation at trace concentrations, which inhibit a negligible fraction of channels, we measured fluorescence from cells bathed in probe at a concentration substantially below the K_d of its target VGIC. At 1 nM, GxTX-dy550 is 30-fold below its measured K_d for Kv2.1, and has little effect on the whole-cell K⁺ current (Fig. S2A). At this inefficacious concentration, fluorescence could be detected on Kv2.1-expressing cells. This fluorescence was modulated by epochs of action-potential-like stimuli: 2-ms steps to +40 mV from a holding potential of −80 mV at 100 Hz. During these stimulus epochs, fluorescence decayed toward a similar baseline; in the periods between stimulus epochs, fluorescence slowly recovered (Fig. 4G). These signals suggest that every action potential increases the odds of probe dissociation. The mean fluorescence decay rate during the initial stimulus epoch was k_∆F = 0.025 ± 0.005 s⁻¹ (fit with Eq. S4, F₀ unconstrained, n = 5 cells), intermediate to the dissociation rates seen at polarized or depolarized voltages (Fig. 4 C and F and Fig. S2C). This suggests that fluorescence change reports VGIC response to many action-potential-like stimuli integrated over time. The probe redistributes between channels and solution to reach a binding equilibrium dictated by the VGIC activation pattern. In this way, the state-dependent probe reports channel activation in response to action-potential-like voltage changes.

The conformation-specific binding of GxTX–fluorophores to Kv channels suggests that the VST could guide other visible probes similarly. To determine if GxTX could imbue very large molecules with its binding properties, we tested whether cells expressing Kv2.1 channels would bind to 100-µm microbeads coated with GxTX. We chose beads used for one-bead–one-compound (OBOC) combinatorial drug discovery programs (49–51) to establish whether GxTX-like molecules might be discoverable in an OBOC screen. These OBOC beads are composed of a solid-phase polystyrene dendrimer core with a PEG shell for water solubility and cell compatibility. We used copper(I)-catalyzed azide-alkyne Click cycloaddition to conjugate an alkyne-functionalized GxTX variant to azide-functionalized beads (Fig. 5A), as this chemistry successfully produced bioactive, PEGylated GxTX (Fig. 1E and Fig. S1 B and D). Beads with a GxTX coating adhered to Kv2.1-expressing cells (Fig. 5B). We found that bead–cell adhesion required both the VST peptide and a VGIC. Kv2.1 cells did not adhere well to control beads without GxTX, and cells lacking Kv2.1 did not adhere well to GxTX beads (Fig. 5 C and D). We next tested whether channel conformation impacted this GxTX-bead–Kv2.1-cell interaction. As studying adhesion of voltage-clamped cells to beads was infeasible, external [K⁺] was used to change voltage and activate channels. We first confirmed that increasing external [K⁺] depolarized cells to release GxTX from Kv2.1. Fluorescence decay resulting from GxTX-dy550 dissociation was measured in a control (5 mM) [K⁺] solution and after addition of a high (180 mM) [K⁺] solution. The high [K⁺] solution accelerated GxTX dissociation (Fig. 5 E and F), consistent with the high [K⁺] increasing resting voltage to activate the Kv2.1 channels. To assess the effects of channel activation on cell adhesion to GxTX-beads, Kv2.1-expressing cells were mixed with beads and allowed to adhere. These bead–cell complexes were then incubated in either a control or high [K⁺] solution. In blinded trials on five separate days, the fraction of beads with cells bound was reduced by incubation in high [K⁺] solution (Fig. 5G). Thus, the activity dependence of Kv2.1 binding was preserved when GxTX was attached to beads. The microbead results indicate that voltage sensor modulators can be identified by ion channel binding to OBOC beads. This suggests that new modulators of VGICs could be discovered in OBOC libraries. Taken together, these experiments show that VSTs can guide molecules of a wide range of sizes and compositions, from subnanometer fluorophores to 100-µm beads, and that these conjugates can act as reporters of VGIC activation.