The selectivity filter of the voltage-gated sodium channel is involved in channel activation.

Amino acids located in the outer vestibule of the voltage-gated Na+ channel determine the permeation properties of the channel. Recently, residues lining the outer pore have also been implicated in channel gating. The domain (D) IV P-loop residue alanine 1529 forms a part of the putative selectivity filter of the adult rat skeletal muscle (mu1) Na+ channel. Here we report that replacement of alanine 1529 by aspartic acid enhances entry to an ultra-slow inactivated state. Ultra-slow inactivation is characterized by recovery time constants on the order of approximately 100 s from prolonged depolarizations and by the fact that entry to this state can be reduced by binding to the pore of a mutant mu-conotoxin GIIIA, suggesting that ultra-slow inactivation may reflect a structural rearrangement of the outer vestibule. The voltage dependence of ultra-slow inactivation in DIV-A1529D is U-shaped, with a local maximum near -60 mV, whereas activation is maximal only above -20 mV. Furthermore, a train of brief depolarizations produces more ultra-slow inactivation than a single maintained depolarization of the same duration. These data suggest that ultra-slow inactivation emanates from "partially activated" closed states and that the P-loop in DIV may undergo a conformational change during channel activation, which is accentuated by DIV-A1529D.

Upon depolarization voltage-gated Na ϩ channels first open and then enter one or more inactivated states. These inactivated states can be separated by their contribution to the time course of recovery. Upon repolarization from brief (millisecond time scale) depolarizations recovery is characterized by a single kinetic phase with a time constant of a few milliseconds ("fast inactivation"). If adult rat skeletal muscle (1) Na ϩ channels, heterologously expressed in Xenopus oocytes, are inactivated for Ն20 ms and then repolarized, the channels recover from inactivation with three distinct time constants, implying that there are at least three distinct inactivated states. These time constants are in the order of several ms (fast inactivation), several hundred ms ("intermediate inactivation"), and several thousand ms ("slow inactivation") (1,2). When the channels are inactivated for even longer periods, a component of recovery can be identified with a time constant in the range of 30 -100 s (3)(4)(5)(6). We refer to this inactivated state, from which channels recover with time constants in the order of ϳ100 s, as "ultraslow inactivation" (7).
Prolonged inactivation may be of substantial significance in a broad variety of physiological and pathological settings. In neurons, accumulation of Na ϩ channel prolonged inactivation may influence activity-dependent neuronal excitability, especially under pathological conditions of intense discharge such as epilepsy (8). In skeletal muscle, differences in Na ϩ channel prolonged inactivation may underlie differences in fast and slow twitch muscle excitability (9). Several genetic skeletal muscle diseases are a result of defects in inactivation (10). In the heart, myocardial infarction is associated with a delay in recovery from inactivation of Na ϩ currents (11,12). This effect may produce inhomogeneities of cardiac impulse conduction, setting the stage for reentrant arrhythmias and predisposing to sudden cardiac death. Understanding the mechanisms of prolonged inactivation also could define new targets for drug development and therapeutic strategies against neurologic, neuromuscular, and cardiac disorders.
The molecular basis of fast inactivation is thought to be a "ball and chain" mechanism involving a cytoplasmic loop between the third and fourth domain (the III-IV linker) (13)(14)(15), but little is known about the mechanism of the slower forms of inactivation.
We have shown previously that the mutations 1-DIII-K1237E 1 and 1-DIV-A1529D favor entry to the ultra-slow inactivated state (6,16). Residues Lys 1237 in DIII and Ala 1529 in DIV are predicted to line the outer channel pore and, together with residues Asp 400 in DI and Glu 755 in DII, are presumed to form a part of the selectivity filter of the channel (17)(18)(19)(20)(21)(22).
Recent work from other laboratories suggests that the voltage sensors in DIII and DIV play a unique role in coupling fast inactivation to voltage-dependent activation (23,24). Also, DIV voltage sensors have been shown to play a role in slow inactivation (25). In the present study we explore in detail the pro-perties of ultra-slow inactivation in the mutant DIV-A1529D. We find that ultra-slow inactivation in DIV-A1529D has a U-shaped voltage dependence. It can be induced more efficiently by depolarizing voltage trains, suggesting that ultraslow inactivation in DIV-A1529D occurs from intermediate closed states that are occupied on the way to the open state. This suggests that the P-loop in DIV may be involved in the activation mechanism of the channel.

EXPERIMENTAL PROCEDURES
Mutagenesis of the 1-The oligonucleotide-directed point mutation A1529D was introduced using four primer polymerase chain reaction. An oligonucleotide containing the mutation was designed with a change in a silent restriction site to allow rapid identification of the mutant. A vector consisting of the 1 coding sequence flanked by Xenopus globin 5Ј-and 3Ј-untranslated regions was provided as a gift by R. Moorman. This was used as the template for mutagenesis, and polymerase chain reaction fragments were isolated and subcloned into this template using directional ligations. Incorporation of the mutation was confirmed by DNA sequencing of the entire polymerized regions. The vector was linearized by SalI digestion and transcribed with SP6 DNA-dependent RNA polymerase using reagents from the mCAP RNA capping kit (Stratagene, La Jolla, CA). The rat brain ␤ 1 subunit of the Na ϩ channel was also subcloned into pAlterXG, and transcription was prepared from a BamHI-linearized template using SP6 RNA polymerase.
Stage V and VI Xenopus oocytes were isolated from female frogs (NASCO, Ft. Atkinson, WI), were washed with Ca 2ϩ -free solution (90 mM NaCl, 2.5 mM KCl, 1 mM MgCl 2 , 1 mM NaHPO 4 , and 5 mM HEPES titrated to pH 7.6 with 1 N NaOH), were treated with 2 mg/ml collagenase (Sigma) for 1.5 h, and had their follicular cell layers manually removed. Approximately 50 -100 ng of cRNA was injected into each oocyte with a Drummond micro-injector (Broomall, PA). The oocytes were incubated at 17°C for 12 h to 3 days before examination.
Recordings were made in the two-electrode voltage clamp configuration using a Dagan CA-1 voltage clamp (Dagan, Minneapolis, MN) or a TEC 10CD clamp (NPI Electronic, Tamm, Germany). Both clamp amplifiers had a series compensation circuit. All recordings were obtained at room temperature (20 -22°C). The oocytes were placed in recording chambers in which the bath flow rate was about 100 ml/h, and the bath level was adjusted so that the total bath volume was less than 500 l. The electrodes were filled with 3 M KCl and had resistances of less than 1 M⍀. Using pCLAMP6 (Axon Instruments, Foster City, CA) software, data were acquired at 71.4 kHz after low pass filtration at 2 kHz (Ϫ3dB). Curve fitting was performed using ORIGIN 3.5 (MicroCal Software, Inc., Northampton, MA). Recordings were made in a bathing solution that consisted of 90 mM NaCl, 2.5 mM KCl, 1 mM BaCl 2 , 1 mM MgCl 2 , and 5 mM HEPES titrated to pH 7.2 with 1 N NaOH. BaCl 2 was used as a replacement for CaCl 2 to minimize Ca 2ϩ -activated Cl Ϫ currents, which arise as a consequence of Ca 2ϩ entry via some of the mutant channels. To test for Ca 2ϩ permeation in DIV-A1529D oocytes were bathed in the following solution: 85 mM CaCl 2 , 2.5 mM KCl, 10 mM HEPES-Ca(OH) 2 , pH 7.2.
The mutant -conotoxin GIIIA R13Q (-CTX R13Q) was synthesized by solid state synthesis on a polystyrene-based Rink amide resin on an Applied Biosystems 431A synthesizer, as described previously (26). The crude linear peptide was initially desalted on Sephadex G-10/20% acetic acid and then purified by preparative HPLC to ϳ90% homogeneity as determined by analytical HPLC. Following folding of the peptide by air oxidation, -CTX R13Q was purified to near homogeneity by HPLC.
Data Evaluation-The time courses of recovery from inactivation of normalized peak inward currents were fit with the following exponential functions. or where I 2 is the peak inward Na ϩ current of the test pulse during recovery, I 1 is the peak inward Na ϩ current of a test pulse under fully available conditions, 1 , 2 , and 3 are the time constants of distinct components of recovery, A 1 , A 2 , and A 3 are the respective amplitudes of these time constants, and C is the final level of recovery. The time course of development of the ultra-slow inactivated state was best fit with a single exponential function.
where d is the time constant of development of ultra-slow inactivation and S ϱ is the final level of ultra-slow inactivation expressed as a fraction of total inactivation. The data are expressed as the means Ϯ S.E. Statistical comparisons were made using the two-tailed Student's t test. A p Ͻ 0.05 was considered as being significant. Unless otherwise stated n ϭ 6.

RESULTS
Recovery from Ultra-slow Inactivation in 1 and DIV-A1529D-DIV-A1529D shows a slowly recovering component of inactivation. Fig. 1A shows the growth of inward currents through DIV-A1529D channels with subsequent pulses at 20-s intervals. From a holding potential of Ϫ120 mV, the channels were first inactivated by a 300-s depolarizing prepulse to Ϫ50 mV. Recovery from inactivation after returning to Ϫ120 mV was monitored by repetitive test pulses to Ϫ10 mV. The test pulse duration was 30 ms. About 50% of the current recovered within 20 s, whereas the remaining fraction took several minutes to complete recovery.
DIV-A1529D shows a larger component of a slowly recovering current in comparison with 1. Fig. 1B summarizes the time course of recovery from inactivation, produced by a 300-s depolarizing prepulse to Ϫ50 mV in native 1 and in constructs with the mutation DIV-A1529D (n ϭ 6 for each data point). To ensure complete recovery from fast inactivation between test pulses, the test pulse duration was decreased to 20 ms. The data were normalized to the final current level after full recovery. Clearly, wild type 1 currents recovered completely within ϳ50 s, whereas the time course of recovery of DIV-A1529D channels was substantially slower. The data points were well fitted with two exponentials (Equation 1) reflecting two channel populations recovering from distinct inactivated states, which we refer to as slow and ultra-slow inactivation (6). Recovery from slow inactivation had a mean time constant of 7.9 Ϯ 1. Thus, significantly more channels recovered from ultra-slow inactivation in DIV-A1529D than in 1.
Development of Ultra-slow Inactivation in DIV-A1529D-To determine the time course of development of ultra-slow inactivation in DIV-A1529D, prepulses to Ϫ50 mV of variable duration were applied from a holding potential of Ϫ120 mV, and the time course of recovery at Ϫ120 mV was monitored for each prepulse duration by subsequent 30-ms test pulses to Ϫ10 mV at 20-s intervals. The time course of recovery for each prepulse duration was then fitted with two exponentials that yielded time constants and amplitudes of slow and ultra-slow inactivation (Equation 1). In Fig. 1C the amplitude of the ultra-slow exponential component of recovery, reflecting the fraction of channels recovering from ultra-slow inactivation (A 2 in Equation 1 ϭ F inactivating ), is plotted as a function of the respective durations of the inactivating prepulse. The connecting line is the result of a single exponential fit (Equation 3) to the data points. The time constant of development of ultra-slow inactivation was 81 Ϯ 17.9 s. Thus, both entry to and exit from ultra-slow inactivation in DIV-A1529D channels had time constants similar to those previously found in DIII-K1237E channels (6).
Coexpression of DIV-A1529D with the Rat Brain ␤ 1 Subunit-Coexpression of the ␤ 1 subunit with the ␣ subunit of the 1 Na ϩ channel speeds current decay and accelerates recovery from fast and slow inactivation (6,(27)(28)(29)(30). We wanted to ex-plore whether the ␤ 1 subunit affected recovery from ultra-slow inactivation in addition to well known modulating effects of ␤ 1 on fast and slow inactivation. We found that, compared with DIV-A1529D ␣ alone, longer prepulse durations were required to drive DIV-A1529Dϩ␤ 1 channels into ultra-slow inactivation. Fig. 1 (D and E) shows the time course of recovery of DIV-A1529Dϩ␤ 1 channels from ultra-slow inactivation produced by a 1200-s inactivating prepulse to Ϫ50 mV. Note that the prepulse duration was four times as long as in the experiment with DIV-A1529D ␣ alone, shown in Fig. 1A. Inward currents during recovery of DIV-A1529Dϩ␤ 1 were slowly increasing, suggesting that the channels recovered from ultra-slow inactivation. Comparison of the current traces in Fig. 1 (A and D) demonstrates that coexpression with ␤ 1 substantially accelerated current decay. However, the time course of recovery from ultraslow inactivation of DIV-A1529Dϩ␤ 1 was similar to that of DIV-A1529 ␣ alone (Fig. 1, B and E). Fig. 1F shows that increasing the prepulse duration from 300 to 1200 s significantly increased the fraction of DIV-A1529Dϩ␤ 1 channels recovering from ultra-slow inactivation. In contrast, entry into ultra-slow inactivation at Ϫ50 mV in DIV-A1529D ␣ alone channels was completed after ϳ300 s (Fig. 1C). Thus, coexpres-sion of DIV-A1529D ␣ with the ␤ 1 subunit did not abolish ultra-slow inactivation but delayed entry into the ultra-slow inactivated state.
Voltage Dependence of Ultra-slow Inactivation-The fraction of channels recovering from ultra-slow inactivation was strongly voltage-dependent. To examine the voltage dependence of ultra-slow inactivation in DIV-A1529D, the oocytes were depolarized for 300 s from Ϫ120 mV to prepulse voltages in the range of Ϫ90 mV to ϩ30 mV. After each prepulse the potential was returned to Ϫ120 mV, and recovery was monitored by 20-ms test pulses to Ϫ10 mV at 20 s intervals (n ϭ 4 -12). The time course of recovery for each prepulse potential was then fit with a double exponential function (Equation 1) to estimate the fraction of channels recovering from ultra-slow inactivation (A 2 ϭ F inactivating ).
To facilitate comparison with standard availability curves, the fraction of channels not recovering from ultra-slow inactivation was plotted as a function of prepulse voltage in Fig. 2. The voltage dependence of ultra-slow inactivation was U-shaped. At

FIG. 1. Ultra-slow inactivation in DIV-A1529D.
A, growth of inward current during recovery from ultra-slow inactivation in DIV-A1529D channels. From a holding potential of Ϫ120 mV, the channels were inactivated by a 300-s depolarizing step to Ϫ50 mV. Thereafter, the potential was returned to Ϫ120 mV, and recovery from inactivation was monitored by repetitive 30-ms test pulses to Ϫ10 mV at 20-s intervals. B, comparison of the time courses of recovery from inactivation, produced by a 300-s depolarizing prepulse to Ϫ50 mV in native 1 (filled circles) and in DIV-A1529D (filled squares; n ϭ 6 for each data point). The time course of recovery was monitored by successive 20-ms test pulses to Ϫ10 mV, applied at 20-s intervals. Peak inward currents were normalized to the final current level attained after full recovery. Recovery was substantially slower in DIV-A1529D channels than in wild type 1. The time course of recovery was best fit with double exponential functions (Equation 1; connecting lines). C, the time course of development of ultra-slow inactivation in DIV-A1529D. The membrane potential was depolarized from Ϫ120 to Ϫ50 mV for variable durations, and the time course of recovery at Ϫ120 mV was monitored for each prepulse duration as described for B. The time course of recovery from ultra-slow inactivation for each prepulse duration was then fitted with two exponentials (Equation 1). The amplitude of the ultra-slow exponential component of recovery (A 2 in Equation 1 ϭ F inactivating ) is plotted as a function of the respective duration of the inactivating prepulse. The connecting line is the result of a single exponential fit (Equation 3) to the data points. The time constant of development of ultra-slow inactivation was 81 Ϯ 17.9 s. D, growth of inward current during recovery from ultra-slow inactivation in DIV-A1529D, coexpressed with the rat brain ␤ 1 subunit. From a holding potential of Ϫ120 mV, the channels were inactivated by a 1200-s depolarizing step to Ϫ50 mV. Thereafter, the potential was returned to Ϫ120 mV, and recovery from inactivation was monitored by repetitive 30-ms test pulses to Ϫ10 mV at 20-s intervals. E, peak inward currents of the experiment shown in D were normalized to the final current level attained after full recovery. The time course of recovery was best fit with a double exponential function (Equation 1; connecting line). The time constant of the slower component representing recovery from ultra-slow inactivation was 130 s, which is similar to the mean value for DIV-A1529D ␣ only channels (110.9 Ϯ 11.5 s; see text). F, comparison of the amplitude of ultra-slow inactivation in DIV-A1529Dϩ␤ 1 channels following inactivating prepulses to Ϫ50 mV. Prepulse duration was 300 s (n ϭ 5) and 1200 s (n ϭ 4). Prolonging prepulse duration from 300 to 1200 s significantly increased the amplitude of ultra-slow inactivation to a value of ϳ0.6. This amplitude was similar to the amplitude of ultra-slow inactivation in DIV-A1529D ␣ alone channels, following prepulses of 300-s duration. Thus, coexpression with the ␤ 1 subunit slowed development of ultra-slow inactivation.
prepulse potentials of approximately Ϫ60 mV F noninactivating reached a minimum of ϳ 0.3, whereas this fraction was substantially increased as prepulse potentials were set at more positive and at more negative values.
As shown in Fig. 1F, coexpression of the ␤ 1 subunit with DIV-A1529D ␣ slowed entry into the ultra-slow inactivated state. Hence, the fraction of DIV-A1529Dϩ␤ 1 channels that had entered the ultra-slow inactivated state after a prepulse duration of 300 s was smaller than if only the ␣ subunit was expressed. However, the voltage dependence of ultra-slow inactivation of DIV-A1529Dϩ␤ 1 channels was similar to DIV-A1529D ␣ alone, exhibiting a U shape with minimum availability at ϳ-60 mV (Fig. 2). As shown in Fig. 1D, coexpression with the ␤ 1 subunit strongly reduced slow current decay but still left a substantial amount of ultra-slow inactivation. Thus, ultra-slow inactivation was not confined to the abnormally slow gating fraction of 1 channels expressed in Xenopus oocytes. Because entry into ultra-slow inactivation was faster in DIV-A1529D ␣ alone channels, further experiments were performed without coexpression of ␤ 1 .
U-shaped voltage dependence of inactivation is well known in calcium channels, where it is frequently taken as evidence for Ca 2ϩ current-dependent inactivation (31). To test for Ca 2ϩ current-dependent inactivation, oocytes expressing DIV-A1529D channels were bathed in a high Ca 2ϩ solution (85 mM CaCl 2 ; see "Experimental Procedures"). Upon exchange of the standard 90 mM Na ϩ with the high Ca 2ϩ solution, all inward current was lost, indicating a lack of measurable permeation of Ca 2ϩ ions through DIV-A1529D channels. This argues against Ca 2ϩ current-dependent inactivation in DIV-A1529D channels.
Ultra-slow inactivation occurred at more hyperpolarized voltages than the ionic current. As shown in the inset of Fig. 2, maximum inward current in DIV-A1529D occurred at potentials positive to Ϫ20 mV, whereas F noninactivating was minimal at approximately Ϫ60 mV. Similarly, integrated inward currents representing charge entry (Fig. 2, inset, Q) peaked at Ϫ20 mV and declined to 0 at Ϫ50 mV. These values are consistent with published data from wild type 1 expressed in Xenopus oocytes (32,33) and in mammalian cells (34). The discrepancy between the voltage yielding maximal ultra-slow inactivation and the voltage of maximum inward current and of maximum charge entry strongly argues against current-dependent inactivation.
Voltage Dependence of Na ϩ Channel Availability Following Brief Conditioning Prepulses-Given the U-shaped voltage dependence of ultra-slow inactivation, it was important to deter-mine whether other inactivated states in DIV-A1529D also had nonmonotonic dependence on prepulse voltage.
The following protocol was designed to obtain an estimate of the voltage dependence of inactivated states that were elicited by conditioning prepulses short enough to avoid entry of channels into ultra-slow inactivation. 1-s conditioning prepulses to various voltages were applied, and the available, noninactivated fraction of channels was gauged by a subsequent test pulse (Fig. 3). Both in wild type 1 and in DIV-A1529D channel availability after 1-s conditioning prepulses decreased monotonically with depolarization, asymptotically approaching zero at potentials positive to Ϫ40 mV. The absence of U-shaped voltage dependence of inactivation produced by a 1-s prepulse suggested that the ultra-slow inactivated state in DIV-A1529D was unlikely to be connected to those inactivated states that were produced by 1-s conditioning prepulses.
Recovery from a 1-s Conditioning Prepulse-Prepulses of 1-s duration have been shown to recruit a minimum of three inactivated states in 1 (1). Although the experiment shown in Fig.  3 suggested a monotonic voltage dependence of other than ultra-slow inactivated states, it failed to provide information regarding the relative partitioning among faster inactivated states at a given prepulse potential. Thus, we sought to explore whether conditioning prepulses of 1-s duration would be able to recruit a minimum number of three distinct states of inactivation in DIV-A1529D and, if so, whether any one of these states had a nonmonotonic voltage dependence similar to ultra-slow inactivation. Therefore, we examined the time course of recovery of DIV-A1529D channels after a 1-s conditioning prepulse to Ϫ50 mV (i.e. F noninactivating was at minimum) and to Ϫ10 mV (where F noninactivating at maximum). As shown in Fig. 4 the time course of recovery in DIV-A1529D was adequately fitted with three exponential decay functions (Equation 2), suggesting the presence of at least three inactivation processes (1).
The amplitudes of recovery from inactivation, reflecting the fraction of channels recovering from anyone distinct state were similar for both prepulse potentials (see legend to Fig. 4). This result indicates that within the range of examined prepulse voltages, the mutation DIV-A1529D did not affect the voltage dependence of inactivated states produced by brief prepulse durations. Thus, it is unlikely that brief prepulses could produce inactivated states showing U-shaped voltage dependence.
Ultra-slow Inactivation in DIV-A1529D Is Favored by Pulse Trains-The fact that inactivation produced by 1-s prepulses did not exhibit U-shaped voltage dependence argues against a serial Markovian kinetic scheme, where ultra-slow inactivation FIG. 2. Voltage dependence of ultraslow inactivation in DIV-A1529D and DIV-A1529D؉␤ 1 . The membrane potential was depolarized from Ϫ120 mV to the indicated prepulse voltages for 300 s, and the time course of recovery at Ϫ120 mV was monitored for each prepulse potential as described for Fig. 1B (n ϭ 4 -12). The time course of recovery from ultra-slow inactivation for each prepulse potential was then fit with a double exponential function (Equation 1) to estimate the fraction of channels recovering from ultraslow inactivation (A 2 ). Channel availability defined as the fraction of channels not recovering from ultra-slow inactivation (1 Ϫ A 2 ϭ F noninactivating ) is plotted as a function of prepulse voltage. Inset, Current-voltage relationship in DIV-A1529D, determined by 20-ms test pulses to the indicated potentials, from a holding potential of Ϫ120 mV (open squares). Open circles, integrated inward currents, reflecting charge entry, versus voltage. is reached via faster inactivated states. The voltage range where ultra-slow inactivation reached a local maximum was around Ϫ60 to Ϫ50 mV. This is the voltage range over which Na ϩ channels are activated, i.e. the channels pass through a number of closed states to reach a final open state. Therefore, we reasoned that ultra-slow inactivation could be connected to the kinetic processes of activation and/or deactivation. To test this we examined whether ultra-slow inactivation could be produced by repetitive brief depolarizations that enhance the probability of channels to undergo transitions between closed and open states. Fig. 5A shows the effect of 3333 repetitive step depolarizations to Ϫ20 mV, applied from a holding potential of Ϫ120 mV. Each depolarization had a duration of 2 ms, followed by an interpulse interval of 28 ms, at Ϫ120 mV (33.3 Hz). The entire duration of the pulse train was 100 s. Each data point in Fig. 5A indicates the inward current elicited by one out of 50 applied pulses. Clearly, the pulse train was associated with a slowly developing decline in peak inward current, reflecting cumulative channel inactivation. After 100 s the 33.3 Hz train was stopped, and recovery from cumulative inactivation was monitored by repetitive 20-ms test pulses to Ϫ20 mV, applied at 20-s intervals from a holding potential of Ϫ120 mV. The time course of recovery was very slow, and the initial current level (before the 33.3 Hz train) was not reached until Ͼ200 s had elapsed at Ϫ120 mV. Fitting a double exponential function to the normalized time course of recovery after the train (Fig. 5B, line) revealed that ϳ36% of the channels recovered from ultraslow inactivation ( ϭ 85 s). Table I presents a quantitative comparison between the fraction of channels recovering from ultra-slow inactivation, produced by a single 15-s prepulse to Ϫ50 mV, by a 33.3-Hz pulse train, and by a 20-Hz pulse train (n ϭ 6 for each protocol). During the trains 2-ms step depolarizations to Ϫ20 mV were applied for a total duration of 100 s. The cumulative amount of time the channels spent at depolarized potentials was 6.6 and 4 s during the 33.3-and 20-Hz trains, respectively. After each train the time course of recovery was examined as described for the train in Fig. 5. The data in Table I demonstrate that the fraction of channels recovering from ultra-slow inactivation (A 2 ) was significantly greater after each train than after the single prepulse, even though channels spent substantially less time at depolarized potentials during each train than during the constant voltage. Previously we showed that during a single 300-s depolarization to Ϫ20 mV only ϳ10% of channels entered the ultra-slow inactivated state, whereas following a 300-s depolarization to Ϫ60 mV, ϳ70% of channels became ultra-slow inactivated. During the depolarizing trains channels were either at the holding potential of Ϫ120 mV or at the depolarized potential of Ϫ20 mV. Neither of these potentials should recruit a substantial amount of ultra-slow inactivation. The fact that multiple repeated depolarizations drove substan-

FIG. 3. Voltage dependence of steady-state availability probed by 1-s conditioning prepulses.
From a holding potential of Ϫ120 mV the membrane potential was changed to the level displayed on the abscissa. Subsequently, the membrane potential was stepped to Ϫ20 mV for 20 ms to assay the available peak inward currents. The available peak inward currents were normalized to 1 by the current measured at the holding potential of Ϫ120 mV. Both in wild type 1 and in DIV-A1529D channel availability decreased monotonically with depolarization, reaching a minimum at potentials positive to Ϫ40 mV. The solid lines represent fits to a Boltzmann function I/I max ϭ 1/(1 ϩ exp ((V Ϫ V1 ⁄2 )/k)) where V1 ⁄2 is the midpoint of the curve and k is the slope factor.

FIG. 4. Time course of recovery from a 1-s conditioning prepulse in DIV-A1529D.
DIV-A1529D channels were inactivated by a 1-s conditioning prepulse to Ϫ10 mV (A, n ϭ 11) and to Ϫ50 mV (B, n ϭ 6). Recovery from inactivation at Ϫ120 mV was assessed by 20-ms test pulses to Ϫ20 mV at variable intervals following the inactivating prepulse (because of clamp speed limitations of the two electrode voltage clamp in Xenopus oocytes intervals Ͻ10 ms were not tested). The peak inward currents were normalized to the maximum recovery (1) and to the availability of the test pulse with the shortest coupling interval (0), to correct for differences in channel availability between prepulses at Ϫ10 mV and prepulses at Ϫ50 mV. The solid lines show least square fits by a third order exponential decay function (Equation 2). The time constants were obtained by repetitive fitting sessions and then fixed, whereas the amplitudes were allowed to float for the final fitting (84). The fitting parameters of time constants were 1 ϭ 100 ms, 2 ϭ 2000 ms, and 3 ϭ 12000 ms, for both prepulse potentials. The fitting parameters of amplitudes at potential Ϫ10 mV were A 1 ϭ 0.05 Ϯ 0.01, A 2 ϭ 0.59 Ϯ 0.02, and A 3 ϭ 0.36 Ϯ 0.02. The fitting parameters at potential Ϫ50 mV were A 1 ϭ 0.05 Ϯ 0.01, A 2 ϭ 0.61 Ϯ 0.02, and A 3 ϭ 0.33 Ϯ 0.01. The amplitudes were statistically not different at potentials Ϫ10 and Ϫ50 mV (p Ͼ 0.05). tially more channels into ultra-slow inactivation than a single prolonged depolarization to the same potential suggests that ultra-slow inactivation is favored by the repeated transitions between open and closed states. It is unlikely that ultra-slow inactivation developed from open states because, as demonstrated by the current voltage relationship (Fig. 2, inset), the threshold for channel opening was positive to Ϫ50 mV, and ultra-slow inactivation was maximal at approximately Ϫ60 mV. We conclude that ultra-slow inactivation most likely is attained via transitions from "near" open states. This hypothesis may also explain the U shape of ultra-slow inactivation; we assume that upon depolarization channels undergo a number of closed state transitions before the final open state occurs. The probability for channels to populate any specific closed state depends on the transmembrane potential. Ultra-slow inactivation may occur from closed states that have a high prob-ability to be recruited at approximately Ϫ60 mV. Holding channels at Ϫ60 mV may accumulate those closed states from which a kinetic pathway leads into ultra-slow inactivation, explaining why entry to ultra-slow inactivation is maximal at this voltage. Similarly, cycling channels between closed and open states during pulse trains will also accumulate channels in specific closed states that could be connected to the ultra-slow inactivated state.
Molecular Mechanism of Ultra-slow Inactivation in DIV-A1529D-Previously, we have reported that the mutation DIII-K1237E in 1 enhanced the probability of entry to ultra-slow inactivation (6). We hypothetized that ultra-slow inactivation may result from a conformational change of the outer channel vestibule. This hypothesis was supported by the observation that binding of the mutated -conotoxin GIIIA R13Q to the outer channel pore dramatically reduced the likelihood of DIII-K1237E channels entering ultra-slow inactivation. One explanation for this effect was that binding of -CTX R13Q to the outer channel vestibule protected channels from entry to ultraslow inactivation by physical hindrance of the underlying conformational change of the outer channel vestibule. Assessment of channel gating kinetics in the toxin-bound state was possible because -CTX R13Q only partially occludes the outer vestibule, resulting in a residual current of about 25-30% of that in control (35).
We tested whether -CTX R13Q reduced the likelihood of DIV-A1529D channels to enter the ultra-slow inactivated state, as has been demonstrated with DIII-K1237E channels. Fig. 6A shows the time course of recovery of DIV-A1529D channels from a 300-s prepulse to Ϫ50 mV. Recovery at Ϫ120 mV was monitored by repetitive 20-ms test pulses at 20-s intervals. The currents were assessed during control and during superfusion with 27 M -CTX R13Q. Binding of -CTX R13Q substantially reduced the amount of ultra-slow recovery from inactivation.
These results suggest that binding of -CTX R13Q protects a fraction of channels from entry to ultra-slow inactivation in DIV-A1529D as well as in DIII-K1237E. However, -CTX R13Q has been shown to shift voltage-dependent channel gating, perhaps by an electrostatic effect on the voltage sensors (35). To test whether the reduction of ultra-slow inactivation by -CTX R13Q may have resulted from a voltage shift of ultraslow inactivation, we investigated the effect of -CTX R13Q on the voltage dependence of ultra-slow inactivation. As demonstrated in Fig. 6B, -CTX R13Q reduced ultra-slow inactivation over the voltage range where ultra-slow inactivation was maximal, i.e. between Ϫ60 and Ϫ40 mV, without shifting its voltage dependence. Thus, the toxin-mediated reduction in the probability of entry to ultra-slow inactivation did not result from a mere shift of the voltage dependence of ultra-slow inactivation.
These results support the notion that ultra-slow inactivation is produced by a conformational change in the outer vestibule of the channel. -CTX R13Q may act as a "splint in the vestibule," thereby preventing channels from entry to the ultra-slow inactivated state. DISCUSSION The major findings of the present study are: (i) the mutation DIV-A1529D in 1 increased the propensity of Na ϩ channels to enter an ultra-slow inactivated state, (ii) the voltage depend-

FIG. 5. Ultra-slow inactivation in DIV-A1529D can be produced by a pulse train.
A, from a holding potential of Ϫ120 mV, 3333 step depolarizations to Ϫ20 mV were applied. Each depolarization had a duration of 2 ms, followed by 28 ms at Ϫ120 mV. Only one out of 50 pulses is shown. During the pulse train the peak inward current declined progressively, indicating cumulative inactivation. After 100 s the pulse train was stopped, and recovery from inactivation was monitored by repetitive 20-ms depolarizations to Ϫ20 mV from a holding potential of Ϫ120 mV, at 20-s intervals. B, normalized time course of recovery after the pulse train shown in A. The data were fit by a double exponential function (Equation 1; solid line).

TABLE I
Parameters of recovery from ultra-slow inactivation produced by a single 15-s prepulse to Ϫ50 mV, by a 100-s 33.3 Hz train of 2 ms depolarizations to Ϫ20 mV, and by a 100-s 20 Hz train of 2-ms depolarizations to Ϫ20 mV The holding potential between step depolarizations was Ϫ120 mV. The time course of recovery was assessed as described in the legends to Figs. 1B and 5B. Parameters 2 and A 2 refer to the result of double exponential fits (Equation 1) to the time courses of recovery.

Sodium Channel Selectivity Filter and Activation
ence of the ultra-slow inactivated state was U-shaped, (iii) entry to the ultra-slow inactivated state was promoted by repeated brief depolarizing pulses, and (iv) entry to the ultraslow inactivated state was reduced by blocking the outer channel vestibule with a mutant -CTX GIIIA.

Kinetic Effects of Mutations in the Selectivity Filter Region-
Recently, we reported that replacement of the lysine at site 1237 in 1 by serine and by glutamic acid dramatically increased the likelihood of entry to an ultra-slow inactivated state (6). However, mutating residue K1237 not only altered channel gating but also caused a substantial loss of ionic selectivity (22,36). This dual effect of amino acid replacements at site 1237 raised the question of whether ultra-slow inactivation was produced by variations in species or in number of permeating ions rather than by an alteration in channel structure per se. The finding that the charge conserving mutation K1237R was not associated with substantial amounts of ultra-slow inactivation (6) despite having dramatic effects on ionic selectivity (22) supported the idea that some kind of structural effect of the mutation was the basis for the enhanced likelihood of entry to ultra-slow inactivation. The present study adds further support to the latter contention because we were able to demonstrate that substantial amounts of ultra-slow inactivation can be recruited by the mutation DIV-A1529D. Mutations at this site or homologous sites in other Na ϩ channel isoforms exhibit only minor effects on ionic selectivity (19,21,36,37). The fact that ultra-slow inactivation can be observed in a mutant with severely altered permeation properties (DIII-K1237E) and in a mutant with highly preserved permeation properties (DIV-A1529D) argues against the notion that the disruption of ionic selectivity forms the basis of ultra-slow inactivation. Ultraslow inactivation appears to occur by a mechanism unrelated to changes in permeation.
Origin of U-shaped Inactivation-One of the unexpected findings of the present study was that the voltage dependence of ultra-slow inactivation in DIV-A1529D was U-shaped, whereas ultra-slow inactivation in DIII-K1237E had a monotonic dependence on prepulse voltage (Fig. 6A in Ref. 6). Ushaped voltage dependence of inactivation is a well recognized entity in Ca 2ϩ channels, but there are only few reports of this phenomenon in Na ϩ channels. Chandler and Meves (38) observed U-shaped voltage dependence of inactivation in Na ϩ channels of squid giant axons internally perfused with NaF. In a more recent report, U-shaped voltage dependence of inactivation was observed in BTX-modified Na ϩ channels (39). In both reports the upturn of voltage-dependent availability at positive prepulse potentials was assumed to result from a second nonabsorbing inactivated state.
In Ca 2ϩ channels U-shaped inactivation is widely considered to result from Ca 2ϩ current-dependent inactivation (31). In Na ϩ channels, [Ca 2ϩ ] i has been demonstrated to modulate inactivation (40). Thus, it is reasonable to speculate that mutant Na ϩ channels, which allow permeation of Ca 2ϩ ions, might exhibit Ca 2ϩ current-dependent inactivation. In case of the mutant DIV-A1529D Ca 2ϩ current-dependent inactivation seems unlikely because we were unable to detect Ca 2ϩ permeation when the oocytes were bathed in a solution containing 85 mM Ca 2ϩ . This is consistent with the finding that the mutation A1741E (in rat brain II, corresponding to A1529E in 1) was blocked by Ca 2ϩ (IC 50 ϭ 119 M), and no permeation of Ca 2ϩ or Ba 2ϩ was detected (36). In yet another report the mutation A1529C did not exhibit Ca 2ϩ permeability (37).
Furthermore, the fact that U-shaped inactivation reached a maximum near Ϫ50 mV, whereas maximum inward current and maximum charge entry occurred at potentials positive to Ϫ20 mV, strongly argues against current-dependent inactivation.
Recently, it was shown that U-shaped voltage dependence in delayed rectifier K ϩ channels and in N-type calcium channels arises from purely voltage-dependent mechanisms (41)(42)(43). It was suggested that U-shaped inactivation occurs preferentially from partially activated closed states, i.e. nonconducting states that have some or all voltage sensors in the activated position. This model was supported by the observation that a train of depolarizing pulses cycling channels through closed and open states produced more inactivation than a single depolarizing pulse of the same duration, despite the fact that the net time during which channels were depolarized was longer with a single conditioning prepulse (41). The observation that accumulation of intermediate closed states during repetitive pulses produced more inactivation than single pulses of the same duration suggests that inactivation occurred mainly from intermediate closed states (41,42). Additional strong support for preferential closed state inactivation as the basis for U-shaped voltage dependence of inactivation came from the observation in voltage-gated Ca 2ϩ channels that the inactivation rate was fastest at a voltage where only one-third of the total gating charge had moved (43).
"Cumulative inactivation" (44), i.e. inactivation that builds up during repetitive brief depolarizations, has been reported in a subset of K ϩ channels (44 -52). Aldrich (53) suggested that during pulse trains, where the pulse duration is sufficiently short to avoid inactivation within a pulse, channels inactivate predominantly from closed states. Upon repolarization, after each pulse, recovery from inactivation is sufficiently slow that little recovery occurs during the short interpulse interval.
In the present study, we found that DIV-A1529D channels could be driven into ultra-slow inactivation by repetitive brief depolarizations. Furthermore, a steady depolarization to Ϫ50 mV of the same duration as the applied pulse train produced substantially less ultra-slow inactivation despite the fact that channels spent substantially more time at depolarized potentials. These results are consistent with the idea that ultra-slow inactivation is reached preferentially via closed states. Hence, DIV-A1529D channels are most likely to undergo ultra-slow inactivation at the voltage range of Ϫ60 to Ϫ50 mV because this voltage range has the highest probability to accumulate intermediate closed states, which may provide a pathway for entry to the ultra-slow inactivated state.
-CTX R13Q Interferes with Ultra-slow Inactivation in DIV-A1529D-Na ϩ channel block by -CTX R13Q is incomplete, leaving residual single channel current, which allows the examination of channel gating in the blocked state (6,35,54,55).
Recently, we presented evidence that -CTX R13Q is capable of destabilizing the ultra-slow inactivated state by a mechanism unrelated to simple electrostatic interaction with the gating process (6), suggesting that -CTX R13Q interacted with the outer channel vestibule in a way that prevented ultraslow inactivation. We proposed that ultra-slow inactivation most likely reflected a rearrangement of the outer pore, similar to C-type inactivation in voltage-gated K ϩ channels and that binding of -CTX R13Q to the outer pore stabilized the structure of the outer vestibule, thereby protecting channels from ultra-slow inactivation. In the present study ultra-slow inactivation in DIV-A1529D was substantially reduced when -CTX R13Q was bound to the outer vestibule. Furthermore, as shown in Fig. 6B, superfusion with -CTX R13Q did not result in a shift of the voltage dependence of ultra-slow inactivation, but the U-shaped dependence on prepulse voltage appeared blunted relative to the unblocked state. This is consistent with the hypothesis that the reduction of ultra-slow inactivation by -CTX R13Q did not result from electrostatic interaction with the voltage-sensing channel structures, but most likely resulted from an impediment of entry to the ultra-slow inactivated state, similar to the interaction of -CTX R13Q with DIII-K1237E (6). The molecular mechanism of the protection from ultra-slow inactivation by -CTX R13Q remains to be elucidated. In theory, a part of -CTX R13Q protruding into the outer pore could act as a splint in the vestibule, preventing a pore collapse as the molecular event underlying ultra-slow inactivation. Alternatively, the toxin might, by virtue of multiple interactions with the outer surface of the channel (56, 57), act as a molecular scaffold, thereby stabilizing the structure of the outer vestibule.
The Outer Vestibule of the Na ϩ Channel May Be Involved in Activation Gating-For DIV-A1529D the dynamic rearrangement of the pore, which most likely forms the basis of ultraslow inactivation is preferentially reached through intermediate closed states, on the way to the open state of the channel. This suggests that the residue Ala 1529 may be involved in the process of opening the channel pore. We envision that the P-loop of domain IV, which contains residue A1529, undergoes some kind of movement during the opening process of the channel. The replacement of alanine 1529 by aspartic acid might interfere with this motion, thus rendering the channel susceptible to undergo the molecular rearrangement that is reflected by ultra-slow inactivation.
The notion that the P-loop in DIV may participate in channel gating is supported by the finding that this part of the channel is extraordinarily flexible (21,58,59). This flexibility may be mediated by two glycines in close proximity to residue Ala 1529 (Gly 1530 and Gly 1533 ). Glycine residues are considered to allow for a high degree of protein backbone flexibility (60). Contrary to P-loop DIV, P-loops in DII and III contain only one glycine (Gly 754 and Gly 1238 , respectively), and P-loop of DI does not contain any glycine. This underscores a potentially unique role of P-loop in DIV as a flexible part of the channel.
Complementary support for a possible role of the DIV P-loop as a part of the gating machinery of the channel comes from the demonstration that the adjacent DIV S6 segment plays an important role in fast inactivation (61-64), slow inactivation (64 -66), and in binding of batrachotoxin (BTX) and local anesthetics (64,67,68).
BTX is a potent neurotoxin that stabilizes the open state of Na ϩ channels. As a result the channels open persistently, activation is shifted in the hyperpolarized direction, deactivation is slowed, and inactivation is prevented (69). Furthermore, the ionic selectivity of BTX-modified channels is substantially reduced, which indicates that the selectivity filter may be involved in the action of the drug. Finally, BTX binding has been shown to produce U-shaped voltage dependence of Na ϩ channel availability (39). Thus, BTX-modified channels share a number of properties with channels carrying the mutation DIV-A1529D. This suggests that both Ala 1529 and the binding site of BTX may be mechanistically linked to structures involved in the gating machinery. This notion is supported by a recent report demonstrating that the residue DIV-S6 Val 1583 is implicated in BTX binding and in channel gating (64). According to a recent model of the Na ϩ channel, the DIV-S6 residue Val 1583 may be in reasonable proximity to the DIV-P-loop residue Ala 1529 to allow for interaction between the two residues during a rearrangement associated with channel gating (70,71). In this context it is noteworthy that the DIV P-loop may not only be extremely flexible, as discussed above, but may protrude farther into the pore than P-loops of domains I-III, based on electrical distance measurements (19). Thus, it is not unreasonable to assume an interaction of the P-loop residue Ala 1529 with residues located in DIV-S6.
Similar to BTX binding, the local anesthetic block is well known to be linked to channel gating. A mutual relationship appears to exist between the local anesthetic block and the conformation of the outer channel vestibule because outer pore mutations influence the access of local anesthetics to their binding pocket (71,72), and local anesthetic binding is associated with a structural rearrangement of the outer channel vestibule (73). Furthermore, recent studies suggest that the mechanism of action of the local anesthetic lidocaine involves transitions along the activation pathway (74 -76). These findings accord well with the idea that DIV-S6 and the DIV P-loop may be linked to activation gating.
Tetrodotoxin and saxitoxin are known to block the Na ϩ channel by binding to its outer vestibule/selectivity filter region and thereby occluding the pore (77). Makielski et al. (78) characterized a "post-repolarization" block by these toxins and proposed that the toxins had a higher affinity for a pre-open state that was accessed briefly during depolarization and for a more prolonged period during recovery (78). Satin et al. (79) further showed that the DI vestibule residue, which is the greatest determinant of isoform differences in toxin affinity (Tyr 401 in 1), influences the rate constants for recovery through the pre-open higher affinity state. This reinforces the idea that some residues in the channel vestibule are involved in the conformational changes associated with the pre-open state.
This idea gains further support from recent structural data regarding the bacterial KscA K ϩ channel. In KcsA the transmembrane helices, TM2, which are structurally equivalent to S6, undergo a structural rearrangement during activation (80). Specifically, TM2 rotates in a counterclockwise direction while swinging away from the permeation pathway, thus increasing the diameter of the inner vestibule. A similar model has been proposed for the mechanism of activation in voltage-gated K ϩ channels (81,82). In the voltage-gated Na ϩ channel the inner vestibule is considered to be lined by the S6 segments of domains I-IV (83). As mentioned above, it is plausible that the P-loop residue Ala 1529 may be in close proximity to the binding pocket for BTX and local anesthetics in S6. Thus, molecular motions of the DIV S6 segment during activation gating may well be transmitted to the adjacent P-loop residues and vice versa.
In summary, we propose that the mobility of the P-loop in DIV allows for participation of this structure in the complex rearrangement of S6 segments during channel opening. The mutation DIV-A1529D may interfere with this complex rearrangement prior to channel opening. As a result the outer channel vestibule undergoes a dynamic rearrangement that forms the basis of the propensity of DIV-A1529D to enter the ultra-slow inactivated state at voltages near the threshold for channel opening, thus accounting for the U-shaped voltage dependence of ultra-slow inactivation.