Direct Evidence That Receptor Site-4 of Sodium Channel Gating Modifiers Is Not Dipped in the Phospholipid Bilayer of Neuronal Membranes*

In a recent note to Nature, R. MacKinnon has raised the possibility that potassium channel gating modifiers are able to partition in the phospholipid bilayer of neuronal membranes and that by increasing their partial concentration adjacent to their receptor, they affect channel function with apparent high affinity (Lee and MacKinnon (2004) Nature 430, 232–235). This suggestion was adopted by Smith et al. (Smith, J. J., Alphy, S., Seibert, A. L., and Blumenthal, K. M. (2005) J. Biol. Chem. 280, 11127–11133), who analyzed the partitioning of sodium channel modifiers in liposomes. They found that certain modifiers were able to partition in these artificial membranes, and on this basis, they have extrapolated that scorpion β-toxins interact with their channel receptor in a similar mechanism as that proposed by MacKinnon. Since this hypothesis has actually raised a new conception, we examined it in binding assays using a number of pharmacologically distinct scorpion β-toxins and insect and mammalian neuronal membrane preparations, as well as by analyzing the rate by which the toxin effect on gating of Drosophila DmNav1 and rat brain rNav1.2a develops. We show that in general, scorpion β-toxins do not partition in neuronal membranes and that in the case in which a depressant β-toxin partitions in insect neuronal membranes, this partitioning is unrelated to its interaction with the receptor site and the effect on the gating properties of the sodium channel. These results negate the hypothesis that the high affinity of β-toxins for sodium channels is gained by their ability to partition in the phospholipid bilayer and clearly indicate that the receptor site for scorpion β-toxins is accessible to the extracellular solvent.

Neurotoxins have been used extensively as molecular probes for defining the functional architecture of voltage-gated sodium and potassium channels (Na v s and K v s) (1)(2)(3)(4). Voltagegated ion channels are composed of two functionally linked yet structurally independent domains (or modules) (2,5). The pore domain, encompassing trans-membrane segments S5-S6, is responsible for selective ion permeation, whereas a voltagesensing domain (trans-membrane segments S1-S4) alters the conformation of the channel in response to changes in the membrane potential enabling channel activation. Structural and mechanistic models derived from the crystal structures of the archaebacterial channel K v AP (6,7) and the mammalian channel K v 1.2 (8,9) led to the suggestion that the positively charged S4 segment and parts of S3 form a stable hairpin (voltage-sensor paddle) located at the periphery of the channel within the membrane (6). Comparison of K v AP and K v 1.2 crystal structures (8,9) has suggested that the two channels are similar in structure with a very modest difference in the orientation of their voltage sensor (10). Although K v s are tetramers of four subunits, each containing six trans-membrane segments, sodium and calcium channels consist of a single large protein of four homologous domains (D1-D4, pseudo-subunits), each containing S1-S6 segments (2).
Regions of these channels involved in the gating mechanism have been shown to be associated with receptor sites for gating modifier toxins derived from venoms of scorpions (11), sea anemones (11,12), and spiders (6,13,14). Interestingly, these gating modifier toxins appear to interact with an equivalent region (S3-S4 loop), albeit at different domains of voltagegated ion channels (1,11,14,15). Analysis of the activity of VSTX1, a novel potassium channel gating modifier from the spider Grammostola spatulata, on the archaebacterial channel K v AP has led Lee and MacKinnon (16) to suggest that its apparent high affinity is gained by partitioning in the phospholipid bilayer, leading to an increase of toxin concentration near its receptor site (17,18). Based on the fact that such a modifier interacts with the S3-S4 loop (13) and relying on the x-ray structure of K v AP (6), they claimed that this loop is dipped within the phospholipid bilayer. This hypothesis has attracted the minds of other pharmacologists (19,20) and motivated experiments in which the ability of toxins that affect Na v gating to partition in liposomes was examined (20). The results of these experiments led to generalization of the MacKinnon hypothesis (16) with the claim that if such toxins were able to partition in liposomes, they might also be able to partition in neuronal membranes and thus reach their target receptor site. However, this suggestion has never been verified experimentally on intact neuronal membranes.
Since in their experiments the scorpion ␣-toxin Lqq5 did not partition in phospholipid vesicles, whereas the ␤-toxin Css2 did, Blumenthal and colleagues (20) have suggested that ␤-toxins access their D2/S3-S4 binding site via a similar process proposed for K v gating modifiers, namely partition in the phospholipid bilayer and then lateral diffusion toward the binding site on the sodium channel. The unusual number of hydrophobic residues that appear on the surface of site-4 gating modifiers seemingly supports this assumption (37). This suggestion constitutes the underpinning of the idea that receptor site-4, which is associated with D2/S3-S4, is dipped within the membrane, whereas receptor site-3, which is associated with D4/S3-S4, faces the extracellular solvent (20).
We have challenged this revolutionary idea by analyzing the mode of interaction of a number of radiolabeled ␤-toxins with receptor site-4 on insect and mammalian neuronal membrane preparations and bring direct evidence that negates this hypothesis. We show that scorpion ␤-toxins bind with a fast association rate to a single channel receptor site and that most of them do not partition in neuronal membranes. We show that although some ␤-toxins, such as the depressant toxin LqhIT2, partition in insect neuronal membranes, this partition is not related to the toxin binding to receptor site-4 and effect.
Binding Experiments-Locust synaptosomes, prepared from dissected brains and ventral nerve cords of adult Locusta migratoria, were prepared by established methods (40). Neuronal membranes from cockroach were prepared from whole heads of adult Periplaneta americana according to a previously described method (34). Mammalian brain synaptosomes were prepared from adult albino Sprague-Dawley rats (ϳ300 g, lab-oratory-bred), as was described previously (41). Membrane protein concentration was determined using a Bio-Rad protein assay by using bovine serum albumin as standard. All toxins were radioiodinated by lactoperoxidase (Sigma; catalog number L8257; 7 units/60 l of reaction mix) using 10 g of toxin and 0.5 mCi of carrier-free Na 125 I (Amersham Biosciences) following a published protocol (42). The monoiodotoxin was purified using a Resource Analytical reverse phase high pressure liquid chromatography column (6.4 ϫ 100 mm, 15-m particle size; Amersham Biosciences). The concentration of the radiolabeled toxin was determined according to the specific activity of the 125 I corresponding to 2500 -3000 dpm/fmol monoiodotoxin, depending on the age of the radiotoxin and by estimation of its biological activity (usually 70 -80% (43)). The composition of the media used in the binding assays and termination of the reactions were described previously (41,43). Nonspecific toxin binding was determined in the presence of 1-30 M unlabeled toxin and is indicated in the figure legends. Equilibrium competition binding assays were performed using increasing concentrations of unlabeled toxin in the presence of a constant low concentration of 125 I-toxin and analyzed by the computer program KaleidaGraph (Synergy Software, Reading, PA) using a non-linear Hill equation (for IC 50 determination). The K i values were calculated by the equation where L* is the concentration of radioiodinated toxin and K d is its dissociation constant. Each experiment was performed in duplicate and repeated at least three times as indicated (n). Data are presented as mean Ϯ S.D. of number of independent experiments (41,43).
The kinetic data for ligand association and dissociation rates were subjected to analysis by LIGAND (Elsevier Biosoft, Cambridge, UK) by using Kinetic Analysis. Each curve was subjected to multislope analysis to determine whether there are one or two slopes. For LqhIT2 binding, kinetic analysis of single and double exponent curves was performed by IGOR Pro (wave-Metrics Inc., Lake Oswego, OR). Toxin dissociation was induced by the addition of excess cold toxin, and the dissociation rate constant (k off ) was determined directly from a first order plot of toxin dissociation versus time. The time constant ( ϭ ln2/k off ) was estimated directly from the binding curves.  (45,46)) and were injected into Xenopus laevis oocytes as was described previously (28).
Two-electrode Voltage Clamp Recording-Currents were measured 1-2 days after injection using a two-electrode voltage clamp and a GeneClamp 500 amplifier (Molecular Devices, Downingtown, PA). Data were sampled at 10 kHz and filtered at 5 kHz. Data acquisition was controlled by a Macintosh PPC 7100/80 computer equipped with an ITC-16 analog/digital converter (Instrutech Corp., Port Washington, NY) utilizing Synapse (Synergistic Systems). Capacitance transients and leak currents were removed by subtracting a scaled control trace utilizing a P/6 protocol (46). Bath solution contained (in mM): 96 NaCl, 2 KCl, 1 MgCl 2 , 2 CaCl 2 , 5 HEPES, pH 7.85. Oocytes were washed with bath solution flowing from a BPS-8 perfusion system (ALA Scientific Instruments, Westbury, NY) with a positive pressure of 4 p.s.i. Toxins were diluted with bath solution and applied directly to the bath to the final desired concentration.

RESULTS AND DISCUSSION
We have examined the hypothesis that the binding site of scorpion ␤-toxins on the Na v s resides in the phospholipid milieu of neuronal membranes by using different experimental approaches. We analyzed the effects of a number of toxins on the gating properties to assess whether they develop in rates that could be correlated with the rates of partitioning and lateral diffusion presented by Lee and MacKinnon (16) and compared them with the kinetic parameters of their binding interaction with their channel receptor site. We have also analyzed directly whether these toxins are capable of partitioning in neuronal membranes.
Effects of Scorpion ␤-Toxins on the Activation Properties of Na V s-We compared the effects of the anti-insect selective depressant toxin LqhIT2 on the insect para DmNa v 1 channel with those induced by the anti-mammalian ␤-toxin Css4 on the rat brain rNa v 1.2a channel (27), as well as analyzed the effects of the multipotent toxin Lqh␤1 on both channels expressed in Xenopus laevis oocytes. The three ␤-toxins induced the typical effect of shift in voltage dependence of channel activation to more negative membrane potentials ( Fig. 1) (27, 30)). The apparent association rates of LqhIT2 and Lqh␤1 to DmNa v 1 were determined by the currents induced by the depolarization step to Ϫ40 mV ( Fig. 1), as was determined for Css4 (27). The currents developed in a similar fast rate ( ϭ 1.53 Ϯ 0.29 and 14.3 Ϯ 3.3 s, for LqhIT2 and Lqh␤1, respectively; Fig. 1), as was shown for Css4 on rNa v 1.2 (27). Thus, the ␤-toxin association rate to receptor site-4, which leads to the modification of the Na v gating properties, is 2 orders of magnitude faster than the relatively slow rates of partitioning and lateral diffusion of VSTX1 to its channel receptor in artificial phospholipid membranes (16).
Kinetic Parameters of the Binding of ␤-Toxins to Neuronal Membranes-The kinetic parameters of Bj-xtrIT, Css4, and LqhIT2 binding to neuronal membranes were determined using radiolabeled ligands. This analysis revealed that the association rate constant (k on ) of the 125 I-Bj-xtrIT to cockroach neuronal membranes was 20.4 Ϯ 5.3 ϫ 10 6 M Ϫ1 s Ϫ1 (n ϭ 3), and the k on of 125 I-Css4 to rat brain synaptosomes was 12.6 Ϯ 3.2 ϫ 10 6 M Ϫ1 s Ϫ1 (Fig. 2), which were comparable with the k on of the classical scorpion ␣-toxin Lqh2 to rat brain synaptosomes (12.0 Ϯ 4.0 10 6 M Ϫ1 s Ϫ1 (41)). The dissociation rate constants (k off ) of Bj-xtrIT and Css4 from their respective receptor sites were also similar, 0.91 Ϯ 0.25 ϫ 10 Ϫ3 s Ϫ1 (n ϭ 4) and 1.66 Ϯ 0.4 ϫ 10 Ϫ3 s Ϫ1 (n ϭ 3), respectively. The equilibrium dissociation constants (K d ) calculated from the kinetic parameters were in agreement with those obtained in equilibrium binding stud- Scorpion ␤-Toxin Receptor on Na V s Channels Is Extracellular JULY 28, 2006 • VOLUME 281 • NUMBER 30 ies (Fig. 2, insets; K d ϭ 0.09 Ϯ 0.03 nM (n ϭ 4) and K d ϭ 0.73 Ϯ 0.11 nM (n ϭ 4) for Bj-xtrIT and Css4, respectively). These results indicate that these toxins bind in a reversible manner to a single receptor site in neuronal membranes.
The interaction of the depressant toxin LqhIT2 with cockroach neuronal membranes was, however, more complex, and therefore the k on value could not be determined directly. LqhIT2 binds to two distinct, non-interacting binding sites (47) (Fig. 3), of which the high affinity site is of low capacity (48) (K i ϭ 0.9 Ϯ 0.66 nM; B max ϭ 2.4 Ϯ 0.02 pmol/mg, (n ϭ 3); Fig. 3, inset) and is similar to that of excitatory toxins, e.g. Bj-xtrIT (Fig. 2, inset) (48). The high affinity site was shown to be blocked by anti-peptide antibodies directed against external sodium channel regions (47), and its capacity was reported to be similar to that of saxitoxin receptors on the same neuronal membranes (47)(48)(49). These results indicate that the high affinity binding site for LqhIT2 resides on the sodium channel. The low affinity site of LqhIT2 (K i ϭ 307 Ϯ 88 nM, n ϭ 3) is of a capacity that is too high to be related to Na v s (B max ϭ 68 Ϯ 9.3 pmol/mg), and its nature is still unknown (47). Since the low affinity/high capacity binding site observed for LqhIT2 is evident when the high affinity site is saturated by an excitatory toxin, such as AahIT or Bj-xtrIT (34, 47) (Fig. 3), and the radiolabeled toxin can be displaced by LqhIT2 and Lqh␤1 (Fig. 3), it does not represent partition in the phospholipid membrane. Partition in the membrane should have infinite capacity (i.e. non-displaceable and not dependent on concentration) and relatively slow kinetics, as reported for the binding of the spider gating modifier toxin VSTX1 to phospholipid membranes (16).
LqhIT2 Partitioning in the Phospholipid Bilayer of Insect Neuronal Membranes-Blumenthal and colleagues (20) have shown that the ␤-toxin Css2 was able to partition in an artificial phospholipid bilayer (liposomes), whereas the scorpion ␣-toxin Lqq5 did not. Therefore we examined in a direct approach whether ␤-toxins are capable of partitioning in the phospholipid bilayer of neuronal membranes. We followed after the direct binding of LqhIT2 in low (0.1 nM) and high concentrations (30 M) to locust and cockroach neuronal membranes using tracer amounts of 125 I-LqhIT2. Since a high LqhIT2 concentration occupies all displaceable binding sites on these neuronal membrane preparations (Fig. 3), any measurable 125 I-LqhIT2 associated with the membranes would imply interaction with a site of infinite capacity, that is, the phospholipid bilayer. In a similar manner, we measured the binding interaction of 125 I-Bj-xtrIT and 125 I-Css4 with the neuronal membranes of cockroach and rat brain, and as a control, we used the ␣-toxin 125 I-Lqh␣IT, which binds to receptor site-3 on insect Na v s.  to its interaction with receptor site-4 or to its physiological effect.
Receptor Site-4 on the Na V Is Accessible from the Extracellular Side of Neuronal Membranes-Using different experimental approaches, we show that scorpion ␤-toxins affect Na v s gating independent of their ability to partition in the phospholipid bilayer. We first demonstrate that the ␤-toxins Css4 (very similar to Css2 (51)) and Bj-xtrIT do not partition in rat brain or insect neuronal membranes (Fig. 4) and that their rates of association to receptor site-4 ( Fig. 2) (27,52) are 2 orders of magnitude faster than that measured for VSTX1 partition and lateral diffusion (16). The difference between our data and those of Blumenthal and colleagues (20) may be attributed to the phospholipid composition and the presence of cholesterol in the neuronal membranes used in our study. Not only has the lipid composition of membranes been shown to influence the partition coefficient of polypeptides (53-55), but it was also shown that specific phospholipids were required for reconstitution of functional rat brain Na v s in liposomes and for binding of the scorpion ␣-toxin Lqq5 to receptor site-3 (56). Our results are corroborated by previous reports on 125 I-Css2 binding to Na v s in electroplaque membranes (57) and 125 I-Css4 binding to reconstituted rat brain sodium channels in liposomes (58), in which the nonspecific binding was time-independent, suggesting that the two toxins did not interact with the phospholipid bilayer. We further show that the slow partitioning of the ␤-toxin LqhIT2 in neuronal membranes (Figs. 4 and 5) is unrelated to its binding to receptor site-4, whereas its fast, high affinity binding correlates well with the fast development of toxin effect on channel activation ( Fig. 1) (27). Since ␣-toxins do not seem to interact with phospholipids upon binding to receptor site-3 (20) (Fig. 4), we conclude that both receptor site-4 and receptor site-3 of Na v gating modifiers are not dipped in the neuronal membrane. These receptor sites involve loops of S3-S4 in domains 2 and 4 of the channel, which suggests that at least in part, these loops protrude out of the phospholipid bilayer to the extracellular aqueous solution.