Search tips
Search criteria 


Logo of nihpaAbout Author manuscriptsSubmit a manuscriptHHS Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
Science. Author manuscript; available in PMC 2010 October 2.
Published in final edited form as:
PMCID: PMC2869078

A Gating Charge Transfer Center in Voltage Sensors


Voltage sensors regulate the conformations of voltage-dependent ion channels and enzymes. Their nearly switch-like response as a function of membrane voltage comes from the movement of positively charged amino acids, arginine or lysine, across the membrane field. We used mutations with natural and unnatural amino acids, electrophysiological recordings and X-ray crystallography to identify a charge transfer center in voltage sensors that facilitates this movement. This center consists of a rigid cyclic ‘cap’ and two negatively charged amino acids to interact with a positive charge. Specific mutations induce a preference for lysine relative to arginine. By placing lysine at specific locations the voltage sensor can be stabilized in different conformations, thus enabling a dissection of voltage sensor movements and their relationship to ion channel opening.

Voltage sensors are membrane proteins that change their conformation in response to voltage differences across the cell membrane. They are best known as components of voltage-dependent ion channels, in which voltage sensor conformational changes regulate channel opening. Voltage-dependent K+ (Kv) and Na+ (Nav) channels produce nerve impulses and voltage-dependent Ca2+ (Cav) channels initiate muscle contraction and many other cellular processes (1). Certain enzymes also have voltage sensors, which allow the membrane voltage to regulate catalytic activity (2, 3).

Voltage sensors can exhibit a nearly switch-like dependence on membrane voltage, changing from ‘off’ to ‘on’ over a few hundredths of a volt. Such a steep dependence on voltage exists because a large quantity of electric charge – known as gating charge – is transferred across the membrane voltage difference when a sensor switches off to on (4). In the Shaker Kv channel the total gating charge is 12 to 14 elementary charges per channel, or 3.0 to 3.5 elementary charges from each of four voltage sensors (57). The charges originate from positive charge carrying amino acids, mostly arginine but occasionally lysine, located on the fourth membrane-spanning helix (S4) of the voltage sensor (5, 6).

Crystal structures of voltage sensors from Kv channels show that they consist of only four transmembrane helices surrounded by the lipid membrane (811). The seemingly simple structure of the voltage sensor presents an apparent paradox: how does the voltage sensor transfer a large quantity of charge across the low dielectric, charge destabilizing interior of the membrane when there are only four helices to ‘shield’ the charges? Some explanations have posited that the sensor does not physically move its charges very far across the membrane, but rather that it somehow restructures the electric field relative to the charges (1217). On the other hand, other studies suggest that charged amino acids on S4 move 15 Å to 20 Å across the membrane (1820). Such large movements would seem to require a specific mechanism to stabilize the charges as they cross the membrane center.

Voltage sensors of Kv channels

Kv channels consist of a central ion conduction pore surrounded by four voltage sensors (Fig. 1A) (811). Each voltage sensor contacts the pore through two functionally important interfaces: one located near the extracellular (top) surface of S1 and another near the intracellular surface at the S4–S5 linker (21). The segments of the voltage sensor colored red in Figure 1A (S3b, S4 and the S4–S5 linker) are proposed to move relative to the more static segments colored grey (S1, S2 and S3a) (18, 22). The crystal structure of the closed conformation has not been determined, but in a proposed mechanism a hinge between S3a and S3b would allow the voltage sensor paddle (S3b and S4) to undergo movement, which would exert force on the S4-S5 linker to close or open the pore as modeled in Figure 1B (10, 11, 21). The voltage difference across the membrane drives the conformational change by acting on positive charged residues on S4 (Fig. 1, C and D).

Fig. 1
An occluded cation binding site is highly conserved among voltage-sensor containing proteins

In the crystal structure of the Kv1.2–Kv2.1 chimera (paddle chimera), the pore is open and the voltage sensors adopt a membrane depolarized (positive inside relative to outside) conformation (Fig. 1, A to C) (11). We refer to this depolarized voltage sensor conformation as the open voltage sensor and the hyperpolarized conformation as the closed voltage sensor. In the open crystal structure, positive charged amino acids are labeled according to their position on S4 from outside to inside, 0 to 5 (Fig. 1, C and D). The outermost charged amino acids, labeled R0–R4, are located either near the phospholipid headgroup layer (R0 and R1) or in an aqueous cleft created because the paddle is tilted away from S1 and S2 (R2 to R4) (Fig. 1C). R0–R4 are therefore in or near an extracellular surface-exposed environment. The next positive charged amino acid, K5, is different because it is isolated from the external aqueous surface by a phenylalanine side chain, Phe233 (Fig. 1C, green side-chain). The Phe is highly conserved among voltage sensors from many different proteins, including Kv, Nav, Cav, voltage-dependent H+ (Hv) channels and voltage-dependent phosphatase (VSP) enzymes (Fig. 1E) (23, 24). What role does this Phe play?

Effects of mutating Phe233

To address this question we mutated this Phe to 19 other amino acids (Fig. 2, A and B). These experiments were carried out with the Shaker K+ channel instead of the paddle chimera channel because Shaker expresses to high levels in Xenopus oocytes and it is the most extensively studied Kv channel with respect to gating function (2529). Even though the crystal structure of the Shaker K+ channel has not been determined, the paddle chimera channel shares high sequence identity and therefore should serve as an accurate model for designing and analyzing experiments on the Shaker channel. The mutants fall into four groups according to the level of expressed current and the midpoint voltage (Vm) of the activation curve (Fig. 2, A and B). Only two substitutions, Tyr and Trp, produced currents near wild type levels with negative Vm (Fig. 2B).

Fig. 2
A rigid cyclic side chain is important at the position of Phe233

The three amino acids with a permanent aromatic side chain (Phe, Tyr, Trp) support the highest current levels and negative Vm. To test the importance of aromaticity and the possible existence of a cation- π interaction we substituted Phe with cyclohexylalanine (Cha), which has a rigid cyclic side chain but is not aromatic. This mutant yielded functional channels with a negative Vm, much like channels with Phe, Tyr and Trp (Fig. 2C). In addition, a variety of unnatural amino acid analogues of Phe, including 3,5-F-Phe (F2Phe), 4-methyl-Phe (MePhe), 4-cyano-Phe (CNPhe), 4-bromo-Phe (BrPhe) consistently produced negative Vm values, but did not show a systematic correlation with the negative electrostatic potential on the surface of the aromatic ring (fig. S1). These results suggest that a rigid cyclic side chain is important at this position but aromaticity is not.

The channel containing Trp activates with the most negative Vm (Fig. 2, A and B). The channel kinetics underlying the more negative Vm in the Trp mutant can be seen with a voltage protocol in which the membrane is depolarized to open the channels and then hyperpolarized to close them: the rate of closure is extremely slow (Fig. 3, A and B). In general the rate of closure is voltage dependent, becoming faster as the membrane voltage is made more negative (right side of Fig. 3, A and B). In the wild-type channel [R1K5(F), corresponding to Arg at position 1 and Lys at position 5 on S4 (Fig. 1D), and Phe at the equivalent position of Phe233 on S2 (Fig. 1E)] closure is so rapid that a voltage of −65 mV is sufficient to close channels in a brief period of time (Fig. 3A). But when Phe233 is substituted by Trp [R1K5(W)] even at −110 mV – a voltage at which closure is expected to be much faster than at −65 mV – the channels take much longer to close (Fig. 3B). The negative Vm and slow closure suggests that the Phe to Trp mutation energetically favors the open channel relative to the closed channel.

Fig. 3
Effects of the Phe to Trp mutation depend on the amino acid at position 5

Effects of the Phe to Trp mutation depend on the K5 amino acid

The effects of mutating Phe to Trp depend on the nature of the amino acid at position 5 (Fig. 3, A to C). This dependence can be seen through a comparison of double mutants: the R1K5(W) channel has a more negative Vm and slower rate of closure compared to the R1K5(F) channel (Fig. 3, A, B and D), but the R1R5(W) channel has a more positive Vm and faster rate of closure than the R1R5(F) channel (Fig. 3C, D and fig. S2A). Thus, the Trp residue slows closure in the background of Lys and speeds it in the background of Arg.

The crystal structure of the open channel shows that the side chain of K5 is isolated from the external and internal solutions in an occluded binding site formed by Phe above and Glu and Asp below. Glu and Asp form ionized hydrogen bonds with the Lys amino group (Fig. 1C) (11). Because K5 is in direct contact with the Phe, it is understandable that the functional consequence of mutating Phe to Trp could depend on whether Lys or Arg is present at position 5. High conservation of all three amino acids, Phe, Glu and Asp, among Kv, Nav, Cav and Hv channels as well as VSP enzymes suggests that the occluded site is an important feature of voltage sensors in general (Fig. 1E). A crystal structure of the Phe to Trp mutant of the paddle chimera channel was determined to 2.9 Å (table S1) and shows that the open conformation of the voltage sensor is preserved and that the occluded site remains intact, with the K5 side chain present and Trp replacing Phe (Fig. 3E).

The observations described above lead us to the following hypothesis: the occluded site affects the conformation of the voltage sensor through the strength of its interaction with Lys relative to Arg at position 5. Mutation of either the binding site (Phe to Trp) or the Lys (Lys to Arg) influences gating. The more negative Vm and slow rate of closure observed with the mutation R1R5(W) to R1K5(W) would suggest that Lys binds more tightly relative to Arg in the presence of Trp in the occluded site, causing the voltage sensor to be stabilized in its open conformation (Fig. 3, B to D).

Lys substitutions stabilize specific conformations of the voltage sensor

If Lys is favored relative to Arg in the occluded site when Phe is mutated to Trp, and if the voltage sensor transfers its positive charged residues through this site when it undergoes a conformational change, then it should be possible to stabilize the voltage sensor in specific conformations by substituting Lys at different positions while Trp is present in the binding site. In the presence of Phe with Lys at position 1 [K1R5(F)], the channels open and close rapidly but in the presence of Trp with Lys at position 1 [K1R5(W)] the channels open slowly and close rapidly (Fig. 4A and B). That is, the Phe to Trp mutation slows opening (but not closing) when Lys is present at position 1, and it slows closing (but not opening) when Lys is present at position 5. When Lys is present at positions 1 and 5 simultaneously both channel opening and closing are slowed (Fig. 4C and fig. S2B). The activation curve midpoint voltage, Vm, is most negative for R1K5(W), most positive for K1R5(W), and intermediate for K1K5(W) (Fig. 4D). These results are consistent with the hypothesis that Lys at position 5 favors the open conformation and Lys at position 1 favors the closed conformation.

Fig. 4
Lys at position 1 and 5 stabilize the voltage sensor in its closed and open conformation respectively in the presence of Trp

Voltage sensor conformations assessed through gating currents

Ionic currents associated with pore opening and closing, as described above, provide an indirect measure of voltage sensor conformational change. Gating currents on the other hand provide a direct measure, because they result directly from the displacement of the charged amino acids in the voltage sensor when it changes its conformation (4). Gating currents are measurable as a component of the transient current that follows a membrane voltage step. (fig. S3) (6). Agitoxin2 was used to prevent ionic (K+) currents, because this toxin blocks the pore but does not interfere with voltage sensor movements (fig. S3B) (6, 30). Integration of the remaining transient current gives the total charge associated with each voltage step, which can be graphed as a function of membrane voltage in a Q–V plot (fig. S3, C and D). Xenopus oocytes not expressing Kv channels show a linear Q–V plot (fig. S3C), which reflects the charge required to bring the membrane to its new voltage (the linear capacitive charge). Oocytes expressing Kv channels show the same linear component as well as a nonlinear component. The nonlinear component – the gating charge – corresponds to the displacement of charged amino acids as the voltage sensor undergoes its conformational change (fig. S3D).

In Figure 5A to D transient currents during a single voltage step are shown for the four mutants analyzed in Figure 4D. In the R1R5(W) channel the transient currents associated with opening (voltage stepped from hyperpolarized to depolarized value) and closing (voltage stepped back to hyperpolarized value) are brief, and consequently the gating current is buried within the linear capacitive current (Fig. 5A). In contrast, the transient current associated with closing the R1K5(W) channel has a prolonged component that extends beyond the duration of the linear capacitive current (Fig. 5B). This is direct evidence that the voltage sensor is being delayed in its transition from its open to closed conformation. The K1R5(W) channel has a prolonged component associated with channel opening (Fig. 5C), and the K1K5(W) channel has a prolonged component associated with both opening and closing (Fig. 5D). Therefore, the voltage sensor is delayed in its closed to open transition when Lys is present at position 1, and delayed in its open to closed transition when Lys is present at position 5. The time course of the transient current during opening and closing in Figure 5D is consistent with a conformational change of the voltage sensor that involves multiple sequential steps following an initial delay (Fig. 6, A and B).

Fig. 5
Voltage sensor movements assessed by gating currents
Fig. 6
State model of voltage sensor conformational change

Q–V plots with the linear capacitive component subtracted are shown in Figure 5E for the same mutant channels. The quantity Q/Qmax is the fraction of total gating charge following a voltage step. When Q/Qmax is small, corresponding to small displacements of the voltage sensor from its closed conformation, the four plots fall into two groups: one that shows charge movement at very negative voltages (around −100 mV) and another that begins to move charge at less negative voltages (around −50 mV) (Fig. 5E). The two groups correspond to voltage sensors with Arg (more negative, filled symbols) or Lys (less negative, empty symbols) at position 1. When Q/Qmax is large, corresponding to small displacements of the voltage sensor from its open conformation, again two groups are observed, but this time depending on the residue at position 5: voltage sensors with Arg at position 5 (Fig. 5E, circles) exhibit a distinct second component of charge movement compared to voltage sensors with Lys at position 5 (Fig. 5E, squares).

The Q–V plots are consistent with the hypothesis that Lys relative to Arg interacts more favorably with the occluded site (Fig. 5E). Near the voltage sensor’s closed conformation the occluded site presumably binds to positive charged S4 amino acids near position 1. If Lys is present at position 1 and binds more favorably than Arg then a more depolarized membrane voltage (i.e. less negative) should be required to drive the voltage sensor away from its closed, to its open conformation. Near the voltage sensor’s open conformation the occluded site presumably binds to positive charged S4 amino acids near position 5. Therefore, Lys at position 5 should stabilize the open conformation whereas Arg should destabilize it. As a consequence, in the presence of Arg at position 5, a more positive membrane voltage should be required to force the voltage sensor through its final transition to its open conformation. This final transition appears to correlate with a distinct second component of charge movement. The second component, visible in both the R1R5(W) and K1R5(W) channels, accounts for approximately 25% of the total gating charge movement. In both channels the pore opens and ions begin to conduct only in association with this second component (Fig. 5, F and G).


In summary the data support the following conclusions: 1. Phe233 forms the extracellular lid on an occluded binding site in voltage sensors into which S4 positive charges bind. 2. Mutating Phe to Trp alters the relative strength of interaction between Arg or Lys and this site. 3. Lys appears to interact more favorably compared to Arg in the presence of Trp: at position 5 Lys stabilizes the open voltage sensor conformation, at position 1 Lys stabilizes the closed voltage sensor conformation. 4. Given that Lys at position 1 and at position 5 appears to interact with the site in the closed and open conformations, respectively, we propose that the occluded site serves as a catalytic center to lower the energy associated with the transfer of each of the voltage sensor charges as they cross the membrane. 5. Arg at position 5 isolates the last 25% of gating charge as a separate component in the Q–V plot, which occurs concomitantly with pore opening. The association of this last component of gating charge with pore opening suggests that binding of the charged residue at position 5 into the occluded site is required for pore opening.

A simple state model of voltage sensor transitions is shown (Fig. 6A). Each of the five states, connected by four transitions, corresponds to conformations of a single voltage sensor associated with the positive charged residue at positions 1 through 5 inside the occluded site. Thus, state 1 represents the fully closed voltage sensor in which the positive charge of the position 1 amino acid binds to the occluded site, and state 5 represents the open voltage sensor in which the positive charge of the position 5 amino acid binds to the occluded site. The data do not uniquely define each transition in this model so the following simplifying assumptions are made. A single value for the forward rate constants and a single value for the backward rate constants are used except for the rate constants describing the fourth transition connecting states 4 and 5. Voltage-dependence is distributed over the four transitions equally, with 25% of the total gating charge in each. It is further assumed that the open pore corresponds to the condition in which all four voltage sensors are open (i.e. all four have reached state 5.) and that four voltage sensors undergo independent conformational changes.

For illustrative purposes the graphs below the state model depict chemical free energy as a function of voltage sensor reaction coordinate in a channel in which Phe has been mutated to Trp (Fig. 6A). The presence of Lys in the occluded site is represented as a deeper energy well relative to Arg. One physical interpretation is that Lys binds more tightly to the occluded site. Another interpretation is that the energy barriers and wells are higher for Arg relative to Lys. Equations for the scheme in Figure 6A (Supplementary Information) generate the main features observed in the electrophysiology data (Fig. 6, B to D). Notably, by changing only the well depth of state 5 by approximately 4.6 RT relative to all other barriers and wells, the qualitative features of both the gating current time course and the Q–V plots of R1K5(W) and R1R5(W) are recapitulated (Fig. 6, B to D). In particular, relative stabilization of state 5 with Lys in the presence of Trp [R1K5(W)] prolongs the gating current associated with channel closing and gives it the correct time-dependent shape (Fig. 6, B and C), whereas destabilization of state 5 with Arg in the presence of Trp [R1R5(W)] speeds the gating current and separates the Q-V plot into two components (Fig. 6, B and D).

The model captures an important feature of gating, most explicitly displayed in channels with Arg at position 5 [R1R5(W) and K1R5(W)]: multiple transitions occur within each voltage sensor (connecting states 1 through 4 in the model) prior to a final transition (state 4 to 5), which is closely associated with pore opening (Fig. 5, F and G, and Fig. 6D). Thus, according to the model, the closed channel is actually associated with a distribution of conformations (states 1 through 4 within each voltage sensor), which is a function of the degree to which the membrane is hyperpolarized: the more negative the voltage, the closer the distribution comes to state 1. The open channel on the other hand is associated with a specific conformation (state 5) that must be achieved by all four voltage sensors. In order to approximate the behavior of the electrophysiology data we found it necessary to use unique forward and backward rate constants for the fourth (final) voltage sensor transition (Supplementary Information). We speculate that this final transition is very different because it is associated with opening the pore.

In the context of the crystal structure the model can explain a discrepancy between the total gating charge per channel (approximately 14 elementary charges) and the sum of reduced charges when each of the five positions is mutated to a neutral amino acid one at a time (approximately 18 elementary charges) (6). Due to the presence of two carboxylate ligands, the occluded binding site probably requires a positive charge in it at all times – except during brief transitions between states driven by voltage. Therefore, mutations that remove a positive charge from a particular position should reduce measured gating charge for two reasons: because the number of charges transported by the voltage sensor is reduced, and because the distribution of states before and/or after the voltage step is altered (i.e. the starting and/or ending positions of all the charges is altered). The two reasons combined can account for the apparent discrepancy between total gating charge and the sum of reduced gating charge by mutation.

We do not know the physical basis of Lys versus Arg discrimination by the occluded site in the Phe to Trp mutant. The case in which Lys binds more favorably to the occluded site is indistinguishable from that in which Arg binds less favorably because the Boltzmann distribution of states depends only on the relative energy differences among the available states. We favor the idea that Trp causes Arg to bind in the site with lower affinity, and also perhaps raises the barrier for an Arg to enter the site because Trp is larger than Phe and would be expected to constrict the site, destabilizing the larger Arg guanidinium group relative to the smaller Lys amino group. A correlation between the size of substituted groups on the Phe side chain (Br > CN > ME > H) and the value of Vm is consistent with this idea (fig. S1).

The data imply a specific distance over which S4 charged amino acids move across the membrane with gating. In the crystal structures of the open conformation Lys at position 5 is located in the occluded binding site (Fig. 1C and and3E)3E) (11). The electrophysiological data suggest that Lys at position 1 binds in the occluded site in the fully closed (strongly hyperpolarized) conformation (Fig. 4). The α-carbon distance between position 1 and 5 is 21 Å along the S4 helix and 18 Å perpendicular to the membrane (fig. S4). This distance falls within the range 15 Å to 20 Å inferred independently by biotin-avidin accessibility studies on a prokaryotic Kv channel (18, 19).

This study identifies an occluded site, conserved in voltage sensors, that catalyzes the transfer of positive charges across the membrane field in the process of voltage sensing. By manipulating the structure of the site we have altered its selectivity between Arg and Lys. This selectivity is used to stabilize the voltage sensor in specific conformations and dissect the relationship between the voltage sensor and pore conformational changes.

Supplementary Material

Supplemental Data


We thank members of D. Gadsby’s laboratory (Rockefeller University) for assistance with oocyte preparation; members of the Mackinnon laboratory for assistance; the staff at beamline X29 (National Synchrotron Light Source, Brookhaven National Laboratory) for advice at the synchrotron; and A. Banerjee, J.A. Letts, G. von Heijne (Stockholm University), J. Chen (Purdue University) for helpful discussions. R.M. is an Investigator in the Howard Hughes Medical Institute. D.A.D. is a George Grant Hoag Professor of Chemistry. Supported by NIH GM43949 to R.M and NS 34407 to D.A.D. The X-ray crystallographic coordinates and structure factor files have been deposited in the Protein Data Bank with accession ID 3LNM.


Supporting Online Material

Materials and Methods

Figs. S1 to S4

Table S1


References and Notes

1. Hille B. Ion Channels of Excitable Membranes. Sinauer Associates; Sunderland, MA: 2001.
2. Okamura Y, Murata Y, Iwasaki H, Sasaki M. Tanpakushitsu Kakusan Koso. 2006 Jan;51:18. [PubMed]
3. Murata Y, Iwasaki H, Sasaki M, Inaba K, Okamura Y. Nature. 2005 Jun 30;435:1239. [PubMed]
4. Armstrong CM, Bezanilla F. J Gen Physiol. 1974 May;63:533. [PMC free article] [PubMed]
5. Seoh SA, Sigg D, Papazian DM, Bezanilla F. Neuron. 1996 Jun;16:1159. [PubMed]
6. Aggarwal SK, MacKinnon R. Neuron. 1996 Jun;16:1169. [PubMed]
7. Schoppa NE, McCormack K, Tanouye MA, Sigworth FJ. Science. 1992 Mar 27;255:1712. [PubMed]
8. Jiang Y, et al. Nature. 2003 May 1;423:33. [PubMed]
9. Long SB, Campbell EB, Mackinnon R. Science. 2005 Aug 5;309:897. [PubMed]
10. Long SB, Campbell EB, Mackinnon R. Science. 2005 Aug 5;309:903. [PubMed]
11. Long SB, Tao X, Campbell EB, Mackinnon R. Nature. 2007 Nov 15;450:376. [PubMed]
12. Ahern CA, Horn R. Neuron. 2005 Oct 6;48:25. [PubMed]
13. Cha A, Snyder GE, Selvin PR, Bezanilla F. Nature. 1999 Dec 16;402:809. [PubMed]
14. Chanda B, Asamoah OK, Blunck R, Roux B, Bezanilla F. Nature. 2005 Aug 11;436:852. [PubMed]
15. Glauner KS, Mannuzzu LM, Gandhi CS, Isacoff EY. Nature. 1999 Dec 16;402:813. [PubMed]
16. Posson DJ, Ge P, Miller C, Bezanilla F, Selvin PR. Nature. 2005 Aug 11;436:848. [PMC free article] [PubMed]
17. Starace DM, Bezanilla F. Nature. 2004 Feb 5;427:548. [PubMed]
18. Ruta V, Chen J, MacKinnon R. Cell. 2005 Nov 4;123:463. [PubMed]
19. Jiang Y, Ruta V, Chen J, Lee A, MacKinnon R. Nature. 2003 May 1;423:42. [PubMed]
20. Larsson HP, Baker OS, Dhillon DS, Isacoff EY. Neuron. 1996 Feb;16:387. [PubMed]
21. Lee SY, Banerjee A, MacKinnon R. PLoS Biol. 2009 Mar 3;7:e47. [PMC free article] [PubMed]
22. Banerjee A, MacKinnon R. J Mol Biol. 2008 Sep 5;381:569. [PMC free article] [PubMed]
23. Ramsey IS, Moran MM, Chong JA, Clapham DE. Nature. 2006 Apr 27;440:1213. [PubMed]
24. Sasaki M, Takagi M, Okamura Y. Science. 2006 Apr 28;312:589. [PubMed]
25. Schoppa NE, Sigworth FJ. J Gen Physiol. 1998 Feb;111:313. [PMC free article] [PubMed]
26. Islas LD, Sigworth FJ. J Gen Physiol. 1999 Nov;114:723. [PMC free article] [PubMed]
27. Zagotta WN, Hoshi T, Aldrich RW. J Gen Physiol. 1994 Feb;103:321. [PMC free article] [PubMed]
28. Zagotta WN, Hoshi T, Dittman J, Aldrich RW. J Gen Physiol. 1994 Feb;103:279. [PMC free article] [PubMed]
29. Hoshi T, Zagotta WN, Aldrich RW. J Gen Physiol. 1994 Feb;103:249. [PMC free article] [PubMed]
30. Garcia ML, Garcia-Calvo M, Hidalgo P, Lee A, MacKinnon R. Biochemistry. 1994 Jun 7;33:6834. [PubMed]