Abstract
Voltage dependent potassium channels (Kvs) gate in response to changes in electrical membrane potential by coupling a voltage-sensing module with a K+-selective pore. Animal toxins targeting Kvs are classified to “pore-blockers” that physically plug the ion conduction pathway and “gating modifiers” that disrupt voltage sensor movements. A third group of toxins blocks K+ conduction by an unknown mechanism via binding to the channel turrets. Here we show that Cs1, a peptide toxin isolated from cone snail venom, binds at the turrets of Kv1.2 and targets a network of hydrogen bonds that govern water access to the peripheral cavities that surround the central pore. The resulting ectopic water flow triggers an asymmetric collapse of the pore by a process resembling that of inherent slow inactivation. Pore modulation by animal toxins exposes the peripheral cavity of K+ channels as a novel pharmacological target and provides a rational framework for drug design.
Introduction
Voltage-gated potassium channels (Kvs) shape the electrical properties of excitable cells by facilitating the selective flow of K+ ions across the cell membrane in response to changes in membrane potential. The Kv channel protein is composed of a pore domain that enable ion translocation, surrounded by four voltage sensing domains that gate the pore in a voltage dependent manner (Kuang et al., 2015). The selectivity for K+ is contributed by a 5-residue signature sequence (TVGYG) termed the “Selectivity Filter” (SF) that forms a constriction of the pore at the extracellular face of the channel (Heginbotham et al., 1994). To pass through the SF, K+ ions must shed their hydration shells. Having their backbone carbonyl oxygen atoms facing the pore and creating four consecutive coordination sites (s1-s4) for a dehydrated K+ ion, the signature residues catalyze this reaction (Zhou et al., 2001). At the SF, the alternate binding of K+ ions and water molecules in a single-file enables a “knock-on” mechanism, wherein an incoming ion exerts electrostatic repulsion that displace a neighboring ion along the pore, resulting in conduction (Roux et al., 2011).
In addition to its role in setting the preference for K+ ions, the SF serves as a gate that down-regulates ion flow via a process known as ‘C-type’ or ‘slow’ inactivation (SI). SI is triggered during prolonged depolarizations, and it is antagonized by ion occupancy in the pore (Cordero-Morales et al., 2006; Cuello et al., 2010; Yellen et al., 1994). While it is widely accepted that SI entails a conformational change that renders the SF non-conductive, the exact molecular description of this state is still under debate. The crystal structure of the bacterial channel KcsA at low external K+ has revealed a “pinched” SF in which the protein backbone was bent at Gly77 and ion coordination sites s2 and s3 were unoccupied (Zhou et al., 2001). This channel conformation was suggested to be sterically stabilized by a network of water molecules buried in cavities behind the SF (“peripheral cavities” hereafter, Ostmeyer et al., 2013). Residues situated near these cavities that serve as hydrogen-bond donors/acceptors were proposed to shape the rate of interconversion between the open and the inactivated channel states (Pless et al., 2013). This proposition is supported by the recent structure of the Kv1.2–2.1 paddle chimera in lipid nanodiscs in which these hydrogen bonds are compromised. (Matthies et al., 2018)
Many venomous organisms carry in their arsenal peptide toxins that block Kv channels, thus lowering the threshold for an action potential at the affected nerve or muscle, leading to paralysis. These toxins have traditionally been classified into two groups based on their mode of action (Kalia et al., 2015). Gating modifier toxins block the channel by binding its voltage-sensing domain, thereby altering the stability of the closed state. Pore-blocker toxins bind at the pore module and physically occlude the ion permeation pathway. Classical Kv pore-blockers from scorpion venom are short polypeptides, typically 30-40 residue long, with a rigid core composed of an alpha helix and an anti-parallel 3-stranded beta sheet. Early experiments, in which the interactions of the scorpion neurotoxins Charybdotoxin (CTX) and Agitoxin2 with Kvs were probed, revealed that a bound toxin can be dissociated by K+ ions entering the channel from the cytoplasmic side, a phenomenon termed “trans-enhanced dissociation” (MacKinnon and Miller, 1988). The latter phenomenon was completely abolished upon neutralization of the highly conserved Lys27 toxin residue (Hidalgo and MacKinnon, 1995; Park and Miller, 1992). A model in which the Lys27 ε-amino moiety of the bound toxin competes with K+ ions on a binding site at the extracellular face of the conduction pathway was put forward (Goldstein and Miller, 1993). This model was validated when the crystal structure of CTX in complex with a mammalian Kv channel was solved, revealing that the toxin projects the sidechain of Lys27 into the channel pore, and as a result K+ binding at the s1 site is lost (Banerjee et al., 2013).
In addition to these pore-blocking toxins, numerous studies have illuminated another diverse family of toxins that target the pore domain of K+ channels and block the ionic current, without physically occluding the pore. Instead, these toxins exhibit a pharmacophore made of a ring of positively charged residues and their binding sites are confined primarily to the channel turrets (Hu et al., 2014; Rodríguez De La Vega et al., 2003; Verdier et al., 2005; Xu et al., 2003; Zhang et al., 2003). The precise molecular mechanisms by which toxins of this family block their targets remain obscure, although a general concept (“turret-block”) whereby the toxin acts as a lid above the pore entry was proposed (Verdier et al., 2005; Xu et al., 2003).
Here we describe a novel block mechanism utilized by Conkunitzin-S1 (Cs1), a 60-residue peptide toxin isolated from Conus Striatus, which bind to the drosophila Shaker turrets. We show that instead of directly plugging the ion conduction pathway, Cs1 modifies the permeation of water molecules into the peripheral cavities, thus creating highly asymmetric water distributions around the SF that trigger a local collapse of the channel pore, analogous to slow inactivation. In addition to the description of the “pore-modulatory” action of the toxin, the data provides general new insights into the usage of the peripheral cavities of potassium channels as an important therapeutic target and a framework for rational drug design to affect channel function.
Results
Molecular dissection of the Conckunitzin-S1 – ShakerKD complex
Conckunitzin-S1 (Cs1) is a 60-residue toxin isolated from the venom of the fish hunting cone snail Conus striatus (Bayrhuber et al., 2005). The toxin molecule has a conserved Kunitz domain fold, composed of an N-terminal 3-10 helix, two-stranded beta sheet and a C-terminal alpha helix reticulated by two disulfide bridges (Dy et al., 2006). Our initial analysis of the toxin has indicated that it is highly toxic to fish larvae (Danio rario, LD50=500nM in bath application), practically non-toxic to mice (LD50 > 50μg/gr by subcutaneous injection), and that it has a considerable preference for insect over mammalian Kv channel isoforms (Fig. S1A, see also Schmidt et al., 2014). Aimed to decipher the structural basis for this preference, we carried a molecular dissection of the toxin-channel complex. First, the solvent exposed residues of Cs1 were substituted by alanine, the resulting toxin mutants were expressed and purified, and their potency on the high affinity K427D mutant of the drosophila shaker channel (dmKv1.2 K427D, ShakerKD hereafter) expressed in Xenopus oocytes was assayed. This analysis has pinpointed 5 toxin residues important for toxicity, four of which were confined to the C-terminal α-helix (Fig S1B). Site directed mutagenesis targeted to the extracellular residues of the channel pore domain revealed three residues with critical contributions to Cs1 binding – the aromatic Phe425, and the nearby negatively charged Asp427 and Asp431 (Fig. S1E,F). We concluded the molecular dissection with a double-mutant cycle analysis (Hidalgo and MacKinnon, 1995; Horovitz, 1996) to expose putative toxin-channel amino acid pairwise interactions (Fig. 1C). This analysis revealed substantial coupling energies between Arg34 of the toxin and all three channel residues, as well as between Phe425 of the channel to Arg49 and Tyr59 of the toxin. We set to obtain putative structures of the channel-toxin complex by employing a complementary, independent approach of non-constrained rigid-body docking. Toward this aim we crystalized and determined the structures of two toxin mutants at 1.3Å resolution (Fig S1D, Materials and Methods) from which we obtained high quality model of Cs1. The toxin molecule was docked onto a channel model that was based on the 2.4Å crystal structure of the closely-related kv1.2-kv2.1 paddle chimera (Long et al., 2007). Clustering the obtained putative structures, we observed a dominant toxin binding mode that accounted for 836 of the top 1000 solutions. This model was compatible with the distance restrains provided by the experimental coupling energies (Fig 1B), and it was stable during prolonged molecular dynamics (MD) simulations (>0.4μs). The docked toxin model revealed a network of hydrogen bonds and salt bridges between the side chains of ShakerKD:Asp447 from three of the four channel domains and the C-terminal residues of the toxin (Fig 1 C,D). These interactions could not be deduced by molecular dissection since substitutions at ShakerKD:Asp 447 which resides at the SF entrance, results in a non-conductive channel mutants (Yifrach and MacKinnon, 2002, and not shown). The positively-charged side chain of Cs1:Arg34 interacts with a negative binding pocket formed by ShakerKD: Asp427 and Asp431, rationalizing the weak binding of the toxin to the wild-type channel bearing a lysine at position 427 (Fig. S1A). The structural basis for the preference of Cs1 for insect over mammalian Kv channels can be readily inferred – the side chain of Phe425 from channel subunits B, C and D makes high impact interactions with Tyr59, Arg49 and Arg34 of Cs1, respectively. The high diversity of these interactions – π− π, cation- π, and aromatic-aliphatic stacking interactions mandates an aromatic residue at this position for high affinity binding. This criterion is frequently met by KV1.2 isoforms isolated from crustaceans, fish and insects, but it is not observed in mammals (Fig. S1G), consistent with the preferential toxin action on the natural prey of the cone snail.
Conkunitzin-S1 does not directly block the ion conduction pathway
A distinctive feature of the modeled Cs1-ShakerKD complex that became evident during early MD refinements was that the toxin does not physically occlude the channel pore. Cs1 is bound slightly off the central pore axis, and none of the toxin residues interact with the SF backbone carbonyls, as do classical Kv pore blockers (Goldstein and Miller, 1993). In MD simulations conducted with bound Cs1, water exchange between the outer vestibule of the channel pore and the bulk was observed, albeit at reduced rate compared to a toxin-free channel (Fig 2A,B). The gap between the toxin molecule and the channel pore was often occupied by a fully hydrated K+ ion (movie S2). In stark contrast, MD simulations of the classical pore blockers CTX and ShK (Lanigan et al., 2002) bound to their respective receptors, revealed a tight seal formed by the toxins around the channel pore (Fig. 2A,B, movie S2). This disparity between the simulated behavior of Cs1 and the canonical pore blockers lends itself to experimental scrutiny using trans-enhanced dissociation experiments (MacKinnon and Miller, 1988). In these experiments, we assayed toxin dissociation during strong depolarizations, which drive K+ ions to compete with a bound toxin on the S1 ion coordination site. We have elected to utilize for these experiments the classical pore-blocker ShK and not the wildly known Charybdotoxin, for which a complex structure is available. The reason for our choice was that CTX does not tolerate well an aromatic residue at position 425 of the Shaker channel (Goldstein and Miller, 1992), while Cs1 mandates it (present work). In contrast, ShK is quite tolerant to substitutions at position 425 (Fig. 5B). While ShK could be readily knocked off its binding site by prolonged depolarizations, Cs1 remained bound to the channel throughout these experiments, consistent with the proposed docking model in which none of Cs1 residues is bound at the s1 site (Fig 2C,D).
Conkunitzin-S1 induces an asymmetric constriction of the selectivity filter
The results presented thus far strongly negate a physical block of the channel pore by Cs1 and suggest that water molecules and ions from the extracellular milieu may access the outer channel vestibule in the presence of a bound toxin. Yet, ion flow is blocked completely when a saturating toxin concentration is applied (Fig S1A). To resolve this apparent conundrum, we simulated the behavior of the toxin bound channel during a series of 200ns time windows. In 5 out of 7 simulations carried, we observed a non-symmetric collapse of the channel pore at the SF region (Fig. 3). The observed collapse occurred typically after few tens of nanoseconds into the production run, and followed a common pattern of molecular events, depicted in Figure 3. In all simulations, the SF was initially found at 2,4 ionic configuration, with a K+ ion bound at s2 and additional ion fluctuating between s4 and the intracellular cavity of the channel (Fig 3B). At this stable ionic configuration (in the absence of an electric field), the SF assumed a symmetric conformation with a cross-subunit distance of ~8Å (Fig 3C). The initial trigger for pore collapse was the flip of the backbone carbonyl at Gly444 or Tyr445 away from the pore into the peripheral cavity (Fig 3A). This event was closely followed by the dissociation of the K+ ion bound at s2, leaving the SF with a single ion occupying s4, and water molecules at the remaining sites (Fig. 3B). This configuration of the SF was highly unstable, and triggered a rapid asymmetric collapse of the channel pore, in which one pair of opposing subunits assumed a pinched conformation typified by a 5.5Å cross-subunit distance (Fig. 3C). This phenomenon was unique to Cs1 simulations, and was not observed in control simulations using either toxin-free channels (Fig 3), that were obtained using the very same starting configuration with the toxin molecule removed, or complexes of the classical pore blockers, CTX and ShK, bound to Kv channels (Fig. S3). On the contrary, in the latter case the constant occupancy of ion coordination site s1 by the ε-amino group of the conserved toxin lysine stabilized the SF conformation, manifested in decreased fluctuations of the backbone atoms and the bound ions (Fig 3, S3). These observations strongly suggest that the asymmetric collapse observed at the selectivity filter region in our simulations is linked to the presence of the bound toxin rather to the initial configuration of the system.
Conkunitzin-S1 modifies water permeation into the peripheral cavities
The observed collapse of the channel pore in the presence of Cs1 was initiated by a flip of a backbone carbonyl at Gly444 (Fig. 3A). This pore residue is positioned ~11Å away from the nearest toxin residue (Arg55) and ~15Å away from the channel turrets, where all high impact interactions with the toxin take place. How can Cs1 remotely trigger a cascade of events that leads to the collapse of the pore? A long-range allosteric effect mediated by the turret residues, which contribute most of the Cs1 binding site, seemed unlikely since (i) substitutions at these residues had no apparent effect on the stability of the open state (Ranganathan et al., 1996; Yifrach and MacKinnon, 2002) and (ii), comparison of the conformational dynamics of the channel backbone during system equilibration phase with and without bound toxin did not reveal any signs for induced fit (backbone RMSD between the initial and the equilibrated structures were 1.9±0.07Å and 1.93±0.16Å (n=5) for toxin free and toxin bound simulations, respectively). We therefore focused on the interactions between the bound toxin and channel residues in the vicinity of the pore region. The central pore of K+ channels is surrounded by water-filled cavities, which were proposed to shape the rate and extent of the slow inactivation process (Ostmeyer et al., 2013). In Kv1.2, the confinement of water molecules within these peripheral cavities is achieved by two sets of channel residues. The “aromatic cuff” (Pless et al., 2013, Fig. S4A), composed of the side chains of Tyr445, Trp434 and Trp435 alongside Val443 form a hydrophobic barrier at the base of the cavity (“lower barrier”). At the top of the cavity, the side chains of Asp447, Met448 and Trp434 from one channel subunit and Thr449 from a neighboring subunit form a barrier that limits the exchange of water molecules between the cavity interior and the extracellular bulk (“upper barrier”, Fig S4B, movie S4A). In simulations of toxin free channels, the space between the two barriers is stably populated by 2-3 water molecules that exchange with the bulk at a typical average rate of ~30 molecules per 100ns (Fig. S4G). Since this exchange involves the crossing of the upper barrier, its rate depends on barrier dynamics. In particular, we find high correlation between the rate of exchange of cavity water and the dynamics of the hydrogen bond formed between the side chains of Asp447 and Trp434 (Fig. S4G), which was previously dubbed a “molecular timer” that sets the pace of slow inactivation (Pless et al., 2013, Fig. S4A). Since we find a strong correlation between its integrity to the rate of water exchange via the upper barrier, we refer to it hereafter as the D447:W434 gate, or simply the “D-W gate”. Analysis of ShakerKD trajectories in the presence of Cs1, reveal a major impact of the toxin on the dynamic behavior of the D-W gate. Asp447, positioned directly above the selectivity filter, interacts electrostatically with toxin residues in three of the four channel subunits (Fig.1, Fig 4A, S4D). These contacts allow the toxin to modify the D-W gate in a non-symmetrical fashion. In channel subunit C, a hydrogen bond and a salt bridge between Asp447 to Tyr51 and Arg55 of the toxin, respectively, stabilize the D447:W434 hydrogen bond, leading to a “closed” conformation of the D-W gate (Fig. 4B). Conversely, in subunit D of the channel, a salt bridge between Cs1:Arg49 and Asp447 “pulls” the later away from Trp434, leading to a constantly “open” gate conformation (Fig. 4B, S4H). In addition, we observed strong effects exerted by the interactions of Cs1:Tyr59 with the upper barrier residues in channel subunit B (Fig S4D, movie S4C) on the permeation of water into the peripheral cavity.
Asymmetric water permeation into the peripheral cavities promotes pore-collapse
The asymmetric effect of Cs1 on the open probability of the D-W gate in discrete channel subunits directly translated in MD simulations into an asymmetric rate of water exchange through the upper barriers. In toxin-free simulations, the rate of water exchange between the peripheral cavities and the bulk solvent through the upper cavity barriers was 31.75±2.4 molecules per 100ns, evenly distributed between the four channel subunits (Fig S4E). In Cs1-bound simulations, the rate of water exchange through the upper barrier in subunit D, having a permanently open D-W gate, was nearly four-fold higher compared to subunits B and C, having their D-W gates in predominantly closed conformations (Fig S4E, S4H). This resulted in a high thermal agitation of water molecules within the peripheral cavity of subunit D, compared to the other subunits (Fig 4C). These highly disordered water molecules could not be stably confined within the peripheral cavity boundaries, and instead they spread to neighboring channel regions by triggering a flip of the aromatic ring of Tyr445, thereby breaching the hydrophobic barrier at the bottom of the cavity (Fig. 4D, movies S4B,S4C). The breach of the lower barrier allowed the exchange of water with the neighboring cavities (movie S4A, Fig. S4G), or with the intracellular vestibule through a path contributed by the hydrophilic side chains underneath (movie S4B,S4C, Fig S4F). In both routs, the flow of water behind the selectivity filter could be diverted into the central pore by a flip of a backbone carbonyl at Gly444 or Tyr445, displacing the bound ion from S2 and leading to pore collapse (Fig 3, movie S4C). In turn, collapse of the central pore increased the diameter of the peripheral cavity, allowing for faster water flow behind the selectivity filter into the intracellular vestibule of the channel pore (movie S4C).
A silent binding mode of Conkunitzin-S1 to a non-inactivating ShakerKD mutant
The described mechanism of Cs1 action is highly reminiscent of the inherent slow inactivation, as both processes involve the constriction of the channel pore in response to alternating hydration patterns at the peripheral cavities. A mutation at the peripheral cavity upper barrier, T449Y, previously shown to eliminate slow inactivation (López-Barneo et al., 1993) also abolished block by Cs1 (Fig. 5A). Yet, we did not assign this residue to the Cs1 binding site, as both mutagenesis (Fig. 5A) and modeling data did not reveal any contribution of Thr449 for toxin binding. The finding that the T449Y mutation has not affected the binding of ShK (Fig 5B), which makes close contacts with the pore region, suggested that the substitution has not induced major rearrangements within the channel protein. To gain structural insight into the loss of function of Cs1 on ShakerKD T449Y, we introduced the mutation into the docked toxin model and subjected the resulting complex to MD simulations. This complex retained all high impact toxin-channel contacts and exhibited novel favorable interactions between the toxin and the substituted tyrosine (Fig. 5C). Simulation therefore predicts a tight binding of Cs1 to the ShakerKD T449Y mutant, despite its apparent lack of activity. We tested this prediction by a series of functional binding competition assays in which the ability of Cs1 to inhibit the binding of ShK to ShakerKD derivatives expressed in oocytes was examined (Fig. 5D). Indeed, we found that the rate of ShK association with ShakerKD T449Y was reduced upon pre-application of Cs1 (Fig. 5D, Fig S5A,B). This effect was specific for this channel mutant - we have not observed competition between the two toxins on a channel mutant with low affinity for Cs1 (F425A, Fig 5D, middle), or upon pre-application of a toxin mutant with low affinity for the channel (R49D. Fig 5D, right). How could the apparent high-affinity binding of Cs1 to ShakerKD T449Y be reconciled with its apparent lack of activity? MD simulations of the Cs1-bound T449Y channel mutant revealed high stability of the channel pore (Fig. S5C). Analysis of these trajectories offered two non-redundant mechanisms (Fig. S5) that rationalize the pore stability in the presence of bound toxin. The first is the stabilization of the aromatic-cuff barrier by the newly introduced tyrosine rings. In subunits A and D of the channel, the Tyr449 rings rotated downwards and formed multiple contacts with residues of the aromatic-cuff barrier. This has stabilized the barrier and allowed it to resist the highly disordered water molecules allowed into the peripheral cavities by the bound toxin (Fig. S5C, movie S5A). A second mechanism involved a structural rearrangement within the channel-toxin complex, in which the critical interactions between the C-terminal residues of the toxin and the D-W gates at chains C and D were replaced by a novel set of interactions with the newly introduced tyrosine rings (Fig. S5C-E, movie S5B). This rearrangement has diminished toxin effect on D-W gate dynamics and restored normal water permeation into the peripheral cavities. In summary, the silent binding of Cs1 to ShakerKD T449Y strongly support our supposition that the toxin does not directly plug the ion conduction pathway, and simulations of the channel mutant with a bound toxin reveal specific modifications focused at the proposed action sites of the toxin.
Channel block by Conkunitzin-S1 is slowed by heavy water
The proposed blocking mechanism for Cs1 postulates two stages of toxin action. The first is channel binding, driven mainly by aromatic and cation-π interactions between the toxin and the channel turrets. This stage by itself does not block the channel pore. The second is the interference of the toxin with the normal behavior of the D-W gates via polar interactions, which triggers pore-collapse. This second step crucially depends on the flow of water molecules through the peripheral cavities that surround the selectivity filter. A prediction derived from the proposed mechanism would be that the rate of channel block by the toxin is a function of the basal exchange rate of water molecules between the cavity interior and the bulk solvent. To examine the validity of this prediction, we have sought an experimental design that would allow us to separate between the two steps of Cs1 action, namely toxin binding and pore-collapse. To this end, we have employed the ShakerKD M448K mutant, which exhibit accelerated slow inactivation, while retaining high affinity for toxins (Fig S6A; Koch et al., 2004). While Cs1 binding to unmodified ShakerKD was voltage-insensitive (Fig. 2C,D), we could partially reverse toxin block of the M448K mutant using prolonged depolarizations (Fig. S6B). This voltage-induced reversal of channel block was distinct from classical trans-enhanced dissociation (Fig. S6B). The toxin effect could be restored by the subsequent incubation at a potential where the channel is closed, at a rate that was toxin concentration dependent, giving rise to a bell-shaped curve, with a rising phase that follows recovery from slow inactivation, and falling phase that follows re-establishment of toxin block (Fig 6A, S6C). The rate constants associated with these two opposing processes typically differed by an order of magnitude allowing for a straightforward isolation of Cs1 block onset rate with high temporal resolution (Fig.6). To further isolate the rate component contributed by the water exchange at the peripheral cavities, a series of experiments in which we compared the function of the toxin in H2O and D2O based solutions was performed. D2O was chosen for these experiments since, (i) it is compatible with electrophysiological measurements; (ii), it has a smaller diffusion coefficient compared to H2O (Weingärtner, 1984), and (iii), deuterium bonds are more stable compared to hydrogen bonds (Scheiner and Čuma, 1996). We reasoned that the latter two factors should slow down the D2O-exchange rate through the upper barrier and allow us to challenge directly the proposed action mechanism of Cs1. Since the effect of D2O on the rate of cavity water exchange in ShakerKD is not readily amenable for direct experimental determination, we sought to examine it indirectly, taking advantage on the established link between the water occupancy at the peripheral cavities and the macroscopic rate of recovery from slow inactivation (Ostmeyer, Chakrapani, Pan, Perozo, & Roux, 2013). We reasoned that while D2O may affect channel gating via a non-specific mechanism – resulting from its increased viscosity (Kestin et al., 1985) or its effect on gating transitions (Fig. S6E, see also Díaz-Franulic et al., 2018), since recovery from slow inactivation takes place at resting membrane potential, at which minimal gating transitions takes place, it is unlikely to be affected by these non-specific effects. Comparison between the rates of recovery from slow inactivation of the M448K mutant in D2O and H2O based solutions revealed a small (18%) yet highly significant reduction in the fast component upon transition from H2O to D2O (Fig. S6D), consistent with a slower rate of exchange of D2O molecules buried behind the selectivity filter. Cs1 block resettling curves revealed a clear decrease in rate upon transition from H2O to D2O (Fig. 6B). This effect was specific, as similar experiments conducted with the classic pore-blocker ShK did not reveal any effect of D2O on block onset (Fig. 6C). The specific effect of H2O to D2O exchange on the development of block by Cs1 strongly suggest the existence of a step involving the motion of solvent molecules, independent from toxin binding, in the action mechanism of Cs1, as predicted from MD simulations.
Discussion
Animal toxins acting on voltage gated ion channels were forged by evolution to rapidly alter their targets, endowing venomous organisms with powerful tools for predation and defense. These toxins have traditionally been categorized into two classes – “gating modifier” toxins, which alter ion conduction by interfering with the voltage dependent motions of the channel, and “pore blocker” toxins that bind the channel pore and physically occlude the ion permeation pathway. Here we present a novel mechanism utilized by animal toxins to block K+ channels, which defies the traditional classification and expose a new pharmacological target present in most ion channels.
Pore-modulation - a novel block mechanism of K+ channels
Classical pore blockers isolated from venomous organisms have played pivotal roles since the early days of ion channel research, facilitating the initial purification of novel channel proteins and providing the first insights into their subunit arrangement and the overall topology of the channel pore (Mackinnon, 1991; MacKinnon and Miller, 1989; Ranganathan et al., 1996). Canonic K+ pore blockers display high phylogenetic diversity and variable backbone folds; yet, they all share a strictly conserved lysine residue that physically plugs the channel pore (Dauplais et al., 1997). This mode of action was initially deduced from classical biophysical experiments (Park and Miller, 1992) and later gained strong support upon the elucidation of a solid-state NMR (Lange et al., 2006) and crystal (Banerjee et al., 2013) structures of channel-toxin complexes. Alongside the canonic K+ blockers, exists a diverse group of animal toxins that binds at the turret regions of K+ channels and blocks the ion conductance without physically occluding the pore (Hu et al., 2014; Rodríguez De La Vega et al., 2003; Verdier et al., 2005; Xu et al., 2003; Zhang et al., 2003). The precise molecular mechanisms by which toxins of this family block their targets have remained thus far elusive. Cs1, described herein, clearly belongs to this second group of K+ blockers and its novel mode of action, which we term “pore-modulation”, sheds a new light on the block mechanisms exploited by these toxins.
MD simulations of ShakerKD in the presence of bound Cs1 revealed that in lack of tight binding between the toxin and the channel pore, water molecules and hydrated ions can access the extracellular vestibule (Fig. 2A). The demonstration that the non-inactivating ShakerKD T449Y mutant exhibit normal ion conduction while bound to Cs1 (Fig. 5) further negates a “pore-lid” mechanism for this toxin. Instead, we suggest a mechanism in which the contacts between Cs1 and the channel turrets serve primarily to coordinate its critical interactions with the D-W gates. These gates consist of a network of hydrogen bonds that govern water permeation into the water-filled peripheral cavities that surround the central ion pore. Toxin interference with the D-W gates creates imbalanced water traffic at the peripheral cavities and triggers pore collapse. These three principal properties of Cs1 – turret binding, requirement for intact slow-inactivation and interaction with the D-W gates are shared by multiple K+ blocker families, for which a clear block mechanism has not been described. First. the cone-snail toxins of the κM family that exhibit many pharmacological similarities to Cs1, despite their radically different backbone fold. κM-conotoxin RIIIK of this group binds the turret region of the Shaker channel, does not block the non-inactivating T449Y mutant (Ferber et al., 2003), and exhibit a ring of positively charged residues directed towards the channel pore (Verdier et al., 2005). Second, the HERG blockers of the γ-KTX family. These scorpion toxins form multiple interactions with the extensive turrets of the HERG channel, yet, their block is strongly affected by substitutions at the upper-barrier that inhibit channel inactivation (Hu et al., 2014; Rodríguez De La Vega et al., 2003; Zhang et al., 2003). The recent structure of one such channel mutant, HERG:S631A, clearly demonstrate that the substitution prevents a tilt of the aromatic ring of the signature motif (GFG) and stabilize the canonic pore conformation (Wang and MacKinnon, 2017). Our analysis argues strongly that these toxins, alongside other “turret-binding” toxins, block the channel using the water-mediated pore-modulation mechanism described herein. Combined with recent reports of small molecules that bind at the peripheral cavities and modulate channel conductance (Lolicato et al., 2017; Wang and MacKinnon, 2017), pore-modulating toxins highlight these channel regions as an attractive pharmacological target and provides necessary cues for the rational design of compounds directed at these sites.
Pore-modulation and slow inactivation
MD simulations of Cs1-bound ShakerKD reveal a sequence of events that culminate in an asymmetric collapse of the channel pore, triggered by the bound toxin. Several recent studies conducted with unmodified and chemically modified channels suggest that the molecular events captured by our simulations faithfully describe channel dynamics. First, chemical modifications at the SF of KcsA that enhance slow inactivation were shown to trigger asymmetric collapse of the channel pore, by altering the turnover of water molecules bound behind the SF. Basins at the free energy landscape of unmodified KcsA, corresponding to the asymmetrically constricted pore conformation were detected (Li et al., 2017). Second, a chemical modification of Kv1.2:Trp434 was shown to enhance slow inactivation by increasing water traffic through the D-W gate (Lueck et al., 2016). A key event that re-occurred in all Cs1-bound simulations was the collapse of the hydrophobic cuff barrier that enabled water exchange between the peripheral pore and the central vestibule of the channel and between pockets of neighboring subunits (Fig. S4 F,H, movies S4B,C). Flips of the pore backbone carbonyls then diverted this water stream into the main pore, leading to ion loss and rapid collapse of the SF. Flipped backbone carbonyls were captured in the crystal structures of inactivated ion channels (Cordero-Morales et al., 2006). The recent Cryo-EM structures of the HERG channel and its non-inactivating S631A mutant further establish a direct link between a compromised cuff-barrier and the inactivated state of the channel (Wang and MacKinnon, 2017). Water conduction behind the collapsed SF of KcsA was suggested to explain the high water permeation of its closed state (Furini et al., 2009; Hoomann and Jahnke, 2013). Our simulations reveal an opening of a hydrophilic path connecting the peripheral cavity with the intracellular vestibule that involves two conserved threonine sidechains (Fig. S4F). Alanine substitutions at the second of these Thr residues were recently shown to impair slow inactivation in Kv1.2, Kv1.5 and KcsA and to alter the occupancy of water molecules buried behind the SF of crystalized KcsA mutants (Labro et al., 2018). Combined, these studies portray a molecular machinery composed of water paths and gates behind the K+ channel pore that set the rate and timing for slow inactivation. Pore modulating toxins efficiently exploit this machinery to impose ectopic channel gating.
Competing interests
The authors has no financial competing interests.
Materials and Methods
Animals
Female wild-caught Xenopus leavis frogs were purchased from NASCO. Experiment were performed according to the guidelines of the Weizmann Institute Animal Care and Use Committee permit #37070717-3.
Method details
Expression and purification of Conkunitzin-S1 and ShK
E. coli BL21 (DE3) were transformed with the appropriate toxin construct cloned into the SapI/BamHI sites of pTwin1 (New England Biolabs). Freshly transformed colonies were used to inoculate a 10-ml starter in LB containing 0.1mg/ml Carbenicillin and 0.03mg/ml chloramphenicol. Following 12 hours of growth at 37°C, the starter was diluted into a baffled flask containing 1 liter LB supplemented with antibiotics and grown to OD600 of 0.6. IPTG was added to a concentration of 0.4 mM and culture growth was continued for additional 5 hours. Cells were harvested by centrifugation (10 min at 5000 x g at 4°C) and re-suspended in 50 ml H2O. The cell suspension was frozen and thawed at 50°C for 45 minutes to disrupt cell walls and total lysis was achieved by sonication. The cell lysate was centrifuged for 15 min at 20,000 x g and the inclusion bodies pellet was washed and precipitated twice in washing solution containing 25% (W/V) sucrose, 5 mM EDTA, 1% (V/V) Triton X-100, in PBS pH 8.0. The inclusion bodies were dissolved and incubated for 1 hour in 10 ml denaturation buffer containing 6 M Guanidinium-HCl, 0.1 M Tris-HCl pH 8.0, 1 mM EDTA and 10mM reduced glutathione. The denatured protein suspension was diluted 20-fold into cold (4°C) renaturation solution composed of 0.2 M ammonium acetate pH 8.0, 0.5M Sucrose and 0.2 mM oxidized glutathione, and incubated at 4°C with gentle stirring for 24 hours. The renaturation mixture was filtered through 1 mm Whatman paper to eliminate insoluble proteins and the soluble material was precipitated in 50% ammonium sulfate at 4°C for 4 hours. The precipitate was collected by filtration through a GF/C membrane and re-suspended in 20ml phosphate buffer (50mM, pH 6.2) to activate intein cleavage which releases the toxin from the tag. Following 24 hours incubation at room temperature, the soluble protein was isolated by centrifugation and 0.2μM filtration. The protein sample was then subjected to RP-HPLC, using a 10ml tricorn column packed with source 15RPC on an ACTA-Pure HPLC system. Proteins were eluted with a linear gradient of acetonitrile containing 0.1% TFA (buffer B). Cs1 typically eluted at 28% B and ShK at 23% B, enabling clear separation from the intein moiety that eluted at 70% B. RP-purified samples were dried by lyophilization and kept at −80°C.
Toxin crystallization, X-ray data collection and structure determination
Crystals of Cs1 Q54A and Cs1 R55D were grown using the hanging-drop vapor-diffusion method. The crystals of Cs1 Q54A were grown from 0.2 M Ammonium sulfate, 0.1 M Sodium acetate trihydrate pH 4.6 and 25% w/v PEG 4,000. The crystals formed in the hexagonal space group P63, with one molecule per asymmetric unit. Crystals of Cs1 R55D were grown from 0.05M Potassium phosphate monobasic and 20% w/v PEG 8,000. The crystals formed in the trigonal space group P3221, with two copies per asymmetric unit. Complete datasets to 1.3Å resolution of both protein crystals were collected at 100K using ADSC Q4 CCD detector at ID14-EH4 of the European Synchrotron Radiation Facility (ESRF, Grenoble, France).
Molecular replacement: The structures of the toxins were solved by molecular replacement using the program MOLREP (Vagin and Teplyakov, 2000). The search model was based on a low resolution structure of Conkunitzin-S1 previously determined at 2.4 Å resolution (PDB code 1Y62). All steps of atomic refinement of both structures were carried out with the CCP4/REFMAC5 program (Murshudov et al., 1997) and by Phenix refine (Afonine et al., 2012). The models were built into 2mFobs - DFcalc, and mFobs - DFcalc maps by using the COOT program (Emsley and Cowtan, 2004). Details of the refinement statistics of the Cs1 Q54A and Cs1 R55D structures are described in Table 1. The coordinates of Cs1 Q54A and Cs1 R55D were deposited in the RCSB Protein Data Bank with accession codes 6Q61 and 6Q6C respectively. The structures will be released upon publication.
Circular dichroism spectrometry
CD-spectra were recorded at 25°C using a model 202 circular dichroism spectrometer (Aviv Instruments, Lakewood, NJ, USA). Toxins (140 μM) were dissolved in 5 mM sodium phosphate buffer, pH 7.0, and their spectrum measured using a quartz cell of 0.1 mm light path. Blank spectrum of the buffer was run under identical conditions and subtracted from each of the toxin spectra.
Electrophysiology – oocytes handling
Plasmids encoding for the Drosophila Shaker and its mutants clone in pBlueScript were linearized using EcoRI and purified using phenol/chloroform. The linearized templates were used for mRNA synthesis using the T7 mMESSAGE mMACHINE Transcription Kit (Ambion) and stored as stock solutions at −80°C. Xenopus leavis female frogs were anesthetized and subjected to surgery by incision in the lower half of the belly. The oocytes were pulled out from the incision and placed in sterile calcium free ND96 solution containing 96mM NaCl, 2mM KCl, 1mM MgCl2 in 5mM HEPES pH 7.5 and then incubated in collagenase solution (3mg/ml) for 2 hours in order to defoliculate them. After the collagenase treatment, the oocytes were incubated in ND96 solution supplemented with 1.8 mM CaCl2, 2.5 mM sodium pyruvate, and 100 mg/ml gentamicine (NDE). Selected defolliculated oocytes were injected with 1ng mRNA of interest using a Drummond 510 microdispenser. Injected oocytes were incubated at 18°C for 1-2 days in NDE solution prior the electrophysiological experiments.
Electrophysiology – data acquisition
K+ currents from the injected Xenopus oocytes were measured by two-electrode voltage clamp using a Gene Clamp 500 amplifier (Axon Instruments, Union City, CA, USA). Oocytes were placed in a 100 μl fiberglass bath and perfused with ND96. Toxin solutions were freshly made by the re-suspension of lyophilized toxin in ND96 supplemented with 1 mg/ml BSA, and then diluted and applied directly to the bath. Some measurement that required high external K+ were conducted in a 90K solution containing 90mM KCl, 2mM MgCl2, 10mM HEPES at pH 7.4. Experiments in which the effects of D2O were examined were conducted in ND96 or 90K solutions freshly prepared using D2O as the solvent. Data were sampled at 10 kHz and filtered at 5 kHz using a Digidata 1550A device controlled by pCLAMP 10.5 (Axon Instruments, Union City, CA). Capacitance transients and leak currents were removed by subtracting a scaled control trace utilizing a P/4 protocol.
Electrophysiology – Data analysis
Dose-response curves were acquired by the application of increasing toxin concentrations into the measurement bath. Each curve was constructed from at least 5 toxin concentrations, an adequate period of incubation (> 1 min) at each toxin concentration ensured steady-state response. Experiments were carried in triplicates. Data points were fit using a Hill equation with the hill coefficient restricted to unity:
Were I0 is the unmodified current measured before toxin application, C is toxin concentration and EC50 is the half maximal toxin concentration. Dose-response curves of Cs1 were acquired on the background of the K427D mutation, unless otherwise noted, and this channel mutant is referred throughout the manuscript as ShakerKD. The high affinity binding of the toxin to this channel allowed for exact potency determinations by achieving effect saturation and facilitated double-mutant cycle analysis using channel mutants with low affinity for the toxin.
Double-mutant cycle analysis – Toxin residues that were found important for toxicity (R34, R49, Y51, R55 and Y59) were substituted to alanine, and the potencies of the resulting mutants were assayed on each of the high impact channel mutants (F425A, D427K, D431A). Coupling energies between toxin and channel residues were extracted from the experimental data as described by Hidalgo and MacKinnon (1995), assuming a linear relation between toxin binding and channel block (ie, EC50 = Kd). For each pair of channel/toxin mutants a coupling coefficient was calculated: from which a coupling energy was extracted by where R is the gas constant, and T is the temperature (25°C).
Voltage-dependent toxin dissociation assays - Oocytes expressing ShakerKD were exposed to 2nM ShK or 5nM Cs1 for 5 minutes until a stable block of ~80% was achieved. Voltage dependent toxin-dissociation was assayed using a two-pulse voltage protocol (Fig. 2C, inset) in which two test pulses (P1,P2) of 50ms to −10mV applied from a holding potential of −90mV were separated by a strong depolarizing pulse (+100mV) of variable duration. Voltage-dependent dissociation of the bound toxin leads to an elevated amplitude of the current recorded during the second test pulse and an increase of the ratio P2/P1.
Channel re-block assays – oocytes expressing ShakerKD M448K with current densities of 5-7μA were subjected to toxin dose inducing 80-90% block and the toxin effect was allowed to settle for at least 1 minute. Bell-shaped curves of current amplitude during channel recovery from inactivation and re-block by the toxin were acquired as described in fig. 6. The data were fit by a bi-exponential function:
Where I1 and I2 are the current amplitudes at P1 and P2, respectively, and τ1 and τ2 are the rate constants for recovery from inactivation and re-block by the toxin, respectively.
MD simulations
Modeling of the ShakerKD-Cs1 and ShakerKD-ShK complexes
A structural model of ShakerKD pore domain (residues Lys376-Asp488) was constructed based on the published crystal structure of the Kv1.2-Kv2.1 paddle chimera (PDB ID 2R9R, 86% identity, 94% similarity) using MODELER (Webb and Sali, 2017). For docking Cs1, the 1.3Å crystal structure of the highly active (EC50 = 3.57±0.74nM, n=3) Q54A mutant (table 1, PDB ID 6Q61) was utilized. For the docking of ShK, we have utilized the published solution structure of the toxin (Tudor et al., 1996). The resulting models were submitted for a rigid-body docking without any constraints using the CLUSPRO web service (Kozakov et al., 2017), and the solutions were ranked using the weighting coefficients of the Van der Waals+electrostatics docking mode. For both toxins, the majority of the top 1000 solutions were divided between four clusters, which could be converged into a single unique solution by rotation about the pore axis. These unique solutions were further validated by cross-referencing against the available experimental restraints (for Cs1, see Results, for ShK see Lanigan et al., 2002; Rashid and Kuyucak, 2012). For modeling the interaction of CTX with the KV1.2-Kv2.1 paddle chimera, we utilized its available crystal structure (4JTA). The simulation system for the ShakerKD T449Y:Cs1 was constructed by the sequential substitution of T449 in all four subunits of the pre-assembled ShakerKD:Cs1 system using the MUTATOR plugin of VMD.
MD System setup
The simulation systems employed in this study were assembled using the VMD software suite (Humphrey et al., 1996). The channel protein was embedded in a lipid bilayer containing 190 POPC molecules and solvated with 11621 (toxin free system) to 13773 (toxin bound systems) water molecules. Two K+ ions were initially placed at sites s2 and s4 and a water molecule was placed at site s3 of the selectivity filter based on the crystallographic coordinates to obtain an initial 2,4 configuration. Additional K+ and Cl− ions were added to neutralize the system and set the KCl concentration at 150mM. Final system dimensions were roughly 95 X 95 X 80 Å2 for toxin free and 95 X 95 X 100 Å2 for toxin bound systems. We have used the CHARMM36 parameter set for proteins, lipids and salt ions, and the CHARMM TIP3P model for water (Best et al., 2012).
MD Simulation protocol
Simulations were carried either on GPU using the Win64/CUDA, or on CPU using the Linux-x86_64-ibverbs versions of NAMD (Phillips et al., 2005). All simulations were performed under NPT conditions at 300K and 1atm. Periodic boundary conditions and electrostatic interactions were treated using the particle mesh Ewald method with ~1Å grid size and a 12Å cutoff distance. A time step of 2fs was employed, and bonds involving hydrogen atoms were fixed using the SHAKE algorithm. Assembled systems were minimized using 2000 steps of a conjugated gradient and line search algorithm and then equilibrated as follows: initially the protein coordinates were fixed and the system was equilibrated for 0.5ns to allow the lipids and the solvent to assume proper densities. Then, a series of 0.2ns equilibration steps with decreasing harmonic restrains of 10, 5, 2, 1, 0.6, 0.3 and 0.1 Kcal/mol/Å2 was applied first to the sidechains, and then to the backbone of the protein. System equilibration was concluded by an unconstrained 20ns run. Six unique systems were simulated in the scope of the current study. Each production run was 200ns long. Seven different production runs were carried for the Cs1: ShakerKD complex; the other systems, namely, Cs1: ShakerKD T449Y, ShK: ShakerKD, CTX: Paddle Chimera, free ShakerKD and free ShakerKD T449Y, were each simulated in 3 independent production runs.
Analysis protocols for MD simulations
Trajectory snapshots were saved every 100ps during production simulations. Custom written MATLAB code was used for trajectory analysis (see supplementary file). The peripheral cavity of a given channel subunit was empirically defined as the volume formed by the intersection of four spheres, with centers coordinates at the backbone nitrogen atoms of Val438, Tyr445, Met448 and Val451 and with radii of 10, 7, 10 and 14Å respectively. This definition effectively filtered out pore-waters and water molecules from the neighboring cavities. For RMSF calculations, the mean position of each water molecule during its residence period in a given cavity was used as the reference distance. The radius of gyration for a backbone carbonyl of a given residue was calculated as the distance of its oxygen atom from the geometrical center of the Cα atoms contributed by that residue in each of the four subunits. Hydrogen bonds were scored based on a donor-acceptor distance of less than 3Å and a cutoff angle of 120°.
Quantification and Statistical Analysis
All statistical analyses were conducted using the MATLAB built-in functions ttest or ttest2 for one sample/paired sample T-tests. All data are expressed as the mean ± standard deviation, unless otherwise stated in the figure legend.