Preventing Voltage-dependent Gating of Anthrax Toxin Channels Using Engineered Disulfides

Cellular injury from anthrax toxin results from the action of two distinct soluble, enzymatically active subunits: lethal factor (LF) and edema factor (EF). During intoxication, protective antigen (PA; 83 kD)—anthrax toxin's pore-forming subunit—is cleaved and assembled into a heptameric “prepore” complex, and then internalized and trafficked to the late endosome with LF and EF bound to it (Young and Collier, 2007). Acidification of the endosome leads to three events that are crucial to intoxication: (1) transformation of the soluble prepore (PA₆₃)₇ into a heptameric transmembrane channel (Blaustein et al., 1989; Miller et al., 1999), (2) unfolding of EF and LF (Krantz et al., 2004), and (3) establishing a pH gradient across the endosomal membrane that helps drive the passage of LF and EF through the (PA₆₃)₇ channel (Zhang et al., 2004b; Krantz et al., 2006).

Structures of soluble conformations of PA have been determined by x-ray crystallography in both monomeric (Petosa et al., 1997) and heptameric prepore forms (Petosa et al., 1997; Lacy et al., 2004). Although the structure of the transmembrane channel form of the PA₆₃ heptamer has not been solved, a large body of biochemical and electrophysiological data (summarized in Young and Collier [2007]) suggests that it resembles the mushroom-like structure of the heptameric pore-forming toxin α-hemolysin from Staphlococcus aureus, which has a large soluble “cap” and a 14-stranded β-barrel “stem” that forms its transmembrane domain (Song et al., 1996).

An interesting property of anthrax channels that is shared by other members of the pore-forming toxin and bacterial porin families is that they open and close in response to changes in transmembrane voltage (Donovan et al., 1981; Kagan et al., 1981; Finkelstein, 1985, 1990; Hoch et al., 1985; Blaustein et al., 1987, 1989; Gambale and Montal, 1988; Bainbridge et al., 1998). Although this process has been studied in detail in some cases, the mechanisms underlying the gating of channel-forming toxins remain largely unknown. The residues responsible for sensing the transmembrane voltage have not been identified, and the sites and extents of the actual physical gates have not been determined for these channels.

In this paper, we investigate the voltage-dependent gating of anthrax channels in planar lipid bilayers. We show that reaction of pore-lining cysteine residues in the β-barrel stem region with methanethiosulfonate (MTS) reagents leads to the formation of intersubunit disulfide bonds that lock channels in a conducting state. Because this effect occurs throughout the entire length of the β-barrel, we hypothesize that anthrax channel gating involves either a global constriction and expansion of the barrel, or a series of smaller conformational changes that propagate along the stem to a gate in the cap region. Because an open (PA₆₃)₇ channel appears to be a prerequisite for LF and EF translocation, a detailed understanding of the gating mechanism in anthrax channels may prove useful in the development of therapeutic strategies to thwart infection.

Expression plasmids encoding PA₈₃ wild-type (WT), WT containing a C-terminal hexahistidine (His₆) tag, single cysteine mutants (S312C, F313C, F314C, G317C, and G323C), and single cysteine mutants containing a C-terminal His₆ tag (S290C, S329C, S330C, D335C, and E343C) were provided by A. Finkelstein (Albert Einstein College of Medicine, Bronx, NY) and R.J. Collier (Harvard Medical School, Boston, MA). The S312C-His₆ construct was generated by subcloning the BamHI-XhoI fragment from pET-PAWTHis₆ into identical sites in the pET-PAS312C plasmid. The pET-LF_N plasmid encoding the 263–amino acid N-terminal domain of LF (LF_N) containing an N-terminal His₆ tag was provided by A. Finkelstein. Escherichia coli XL1-Blue cells were used in cloning, and E. coli BL21-DE3 cells were used for protein expression.

WT and single-cysteine mutants of PA (PA₈₃) were expressed and purified as described previously (Miller et al., 1999). His₆-tagged PA₈₃ proteins were purified on a HisTrap (GE Healthcare) Ni²⁺-IDA column by equilibration in 20 mM Tris-HCl, 500 mM NaCl, 5 mM imidazole, pH 8, washing with 40 mM imidazole, and elution in 20 mM Tris-HCl, 500 mM NaCl, 250 mM imidazole, pH 8, followed by dialysis into 20 mM Tris-HCl, 150 mM NaCl, pH 8.5. To generate the heptameric prepore form of PA, PA₈₃ was subjected to limited proteolysis by incubation with a 1:1,000 solution of trypsin (Sigma-Aldrich) for 30 min at room temperature, followed by quenching with 10-fold excess soybean trypsin inhibitor (Sigma-Aldrich). Nicked PA₈₃ was loaded onto a MonoQ sepharose column (GE Healthcare) and subjected to a 10–500 mM NaCl gradient in 20 mM Tris-HCl, pH 8.5. The (PA₆₃)₇ heptameric prepore eluted at ∼400 mM NaCl and was immediately aliquoted, frozen in liquid N₂, and stored at −80°C. 2 mM dithiothreitol (DTT; Sigma-Aldrich) was included in dialysis and storage buffers during preparation of the cysteine mutant proteins. The N-terminal domain of LF (LF_N) was purified as described previously (Zhang et al., 2004a). The His₆ tag was removed by treatment with thrombin protease (1 unit/mg His₆-LF_N; EMD) for 4 h at 25°C in 20 mM Tris-HCl, 150 mM NaCl, 25 mM CaCl₂, pH 8.4.

1, 2-Dioleoyl-sn-glycero-2-phosphocholine (DOPC) and 1, 2-Dioleoyl-sn-glycero-3-[(N-(5-amino-1-carboxypentyl) iminodiacetic acid) succinyl] (nickel salt) (DOGS-NTA) were purchased from Avanti Polar Lipids, Inc. Liposomes were prepared by mixing DOPC with DOGS-NTA in chloroform at a 10:1 molar ratio. The mixture was dried under N₂ gas, washed once in hexane to remove residual chloroform, and immediately resuspended in 20 mM Tris-HCl, 200 mM NaCl, pH 8.5, at a total concentration of 20 mg/ml. The lipid suspension was sonicated using a bath sonicator (Laboratory Supplies Company, Inc.) to generate small unilamellar lipid vesicles 15–50 nm in size. S312C-His₆ or S330C-His₆ prepore was mixed with small unilamellar lipid vesicles at a 1:20 (wt/wt) ratio and incubated for 60 min at 25°C in 20 mM Tris-HCl, 200 mM NaCl, 1 mM MgCl₂, 250 μM tris(2-carboxyethyl)phosphine (TCEP), pH 8.5. After incubation, proteoliposomes were collected by centrifugation at 150,000 g for 60 min at 4°C. Pellets were washed, resuspended in TCEP-free buffer (20 mM MES, 200 mM NaCl) with a reduced pH of 6.5 to facilitate pore formation, incubated at 25°C for 60 min in the presence of hydrogen peroxide (0.3%) to facilitate disulfide formation, and maintained at 4°C overnight. Samples were then exposed to 10 mM N-ethyl maleimide (NEM) for 20 min at 25°C before denaturation to quench any free thiols, solubilized in nonreducing sample buffer in the presence of NEM (60 mM Tris-HCl, 1% SDS, 5% glycerol), and either loaded directly or boiled in 8 M urea ± 10 mM DTT at 100°C for 10 min. Protein bands were resolved by SDS-PAGE on a 4–20% Tris-HCl gradient gel (Bio-Rad Laboratories) and stained with Coomassie blue. All manipulations with S312C and S330C protein samples were performed in tandem, and resultant protein samples were run on the same gel. Gels were digitally imaged using an Image Station 2000 (Kodak).

Planar lipid bilayers were formed using the brush technique of Mueller et al. (1963) by painting a lipid solution of 3% diphytanoyl phosphatidylcholine (DPhPC; Avanti Polar Lipids) in decane over a 250- or 500-μm aperture in a polysulfone cup. Bilayers separated two compartments (1 ml) containing buffered salt solutions (10 mM MES, 100 mM KCl, 1 mM EDTA, pH 6.6) that were stirred using small magnetic bars. Experiments were performed under voltage-clamp conditions using either a List EPC-7 or Warner BC-535 patch clamp amplifier coupled to a single pair of Ag/AgCl electrodes bathed in 3 M KCl, with electrical contact to the bath solutions made via 3-M KCl agar (3%) salt bridges. Bilayer thinning was monitored either visually or by capacitance. All voltages reported are those of the cis (protein-containing) side with respect to trans. Current responses were fed through an Axon 1322A analogue/digital converter, and recorded and analyzed on a PC using Axon pClamp software (version 8). Recordings were sampled at 10 kHz and low pass filtered at 1 kHz.

Experiments were performed after a stable membrane conductance was obtained, typically 12–15 min after the addition of (PA₆₃)₇ to the bilayer chamber. To measure the instantaneous and “quasi” steady-state I-V relationship of WT anthrax channels, bilayers were pulsed for 2.5 s to voltages ranging from −100 to +100 mV, in 10-mV increments. Before each test pulse, the voltage was held at 0 mV to reopen any channels that may have closed during the previous pulse. Instantaneous I-V measurements were taken immediately after the start of each test pulse (after allowing for the 10–20 msec capacitance transient), and quasi steady-state I-V measurements were taken at the end of the 2.5-s test pulse.

A tail current protocol was used to obtain the apparent open probability (P₀) as a function of voltage in WT and Cys mutant channels. Bilayers were subjected to a 5-s prepulse at −100 mV, followed by a 1-s test pulse, to voltages ranging from −100 to +100 mV (in 10-mV increments), and then a 1-s tail pulse to −20 mV. The apparent open probability was obtained from the normalized tail current: P₀ = I/I_max, where I is the current 10 msec into the tail pulse and I_max is the maximum tail current measured. Open probability (P₀) was plotted as a function of voltage and fitted to a Boltzmann equation:

where V₀ is the midpoint of the voltage-activation curve, z is the effective gating charge, P_min is the minimum open probability, V is the applied voltage, T is the temperature in degrees Kelvin, and R and F have their usual meaning.

[2-(Trimethylammonium)ethyl] MTS chloride (MTSET) and N-(b-D-Glucopyranosyl)-N'-[(2-methanethiosulfonyl)ethyl] urea (MTS-glucose; Toronto Research Chemicals) were prepared as 100-mM stocks in ice-cold H₂O and were aliquoted and stored at −20°C until use. NEM (Sigma-Aldrich) was prepared in H₂O as a 200-mM stock and stored as described above. DTT (Sigma-Aldrich) was made fresh daily as a 200-mM stock in H₂O. TCEP (Thermo Fisher Scientific) was prepared in H₂O as a 250-mM stock and stored at −20°C until use. Cysteine-substituted (PA₆₃)₇ was thawed on ice and incubated with 10 mM DTT for 30 min at 25°C before addition to the bilayer chamber. In a typical recording, 1–10 fmol of protein was added to the 1-ml cis compartment of the bilayer chamber containing 250 μM TCEP to maintain cysteine thiols in reduced form during channel formation and membrane insertion. After the membrane conductance had reached a steady-state, MTSET, MTS-glucose, or NEM was added to the trans side of the bilayer while stirring the chamber. In oxidation experiments, the cis compartment was perfused with 10 volumes of DTT/TCEP-free buffer, and in certain experiments, 0.15% H₂O₂ (Sigma-Aldrich) was added to both cis and trans to promote oxidation.

LF_N (final concentration, 7.2 nM) was added to the cis side of bilayers containing either reduced or oxidized S312C channels. This led to a >95% block in conductance within 2 min under conditions of symmetric pH 6.6 and +20 mV transmembrane potential. Once conductance had reached a stable minimum, the cis compartment was perfused with 10 volumes of buffer, and the pH was adjusted to 5.6 by adding 1/30 volume 1 M MES, pH 5.2, to generate a 1-U pH gradient (cis_5.6/trans_6.6) while holding at +1 mV. Translocation of LF_N through the channel to the trans compartment was then initiated by stepping the voltage to +20 mV, and its rate was determined by following the rise in membrane conductance.

Channels formed by (PA₆₃)₇ are gated by voltage; they open within 50 ms at positive voltages (Fig. 1 A) and close within seconds at voltages more negative than −20 mV (Fig. 1 C). The tendency to close at negative voltages is also evident in the region of negative slope conductance in the steady-state I-V relation compared with the instantaneous I-V, which is fairly linear (Fig. 1 D). Although not the focus of this paper, anthrax channels also exhibit a small amount of deactivation gating (e.g., closing) at larger positive voltages; this is more readily seen when pulsing from a holding potential that keeps channels open (Fig. 1 C) (Blaustein and Finkelstein, 1990; Finkelstein, 1994). To further characterize the steady-state gating properties, we applied a tail current protocol and determined the relative open probability as a function of voltage in DPhPC bilayers separating symmetric 0.1 M KCl, pH 6.6, solutions (Fig. 1, A and B). Fits of the normalized tail currents to a Boltzmann function (Eq. 1) reveal a midpoint of activation (V_1/2) at −12 mV and a slope factor (z) of 1.7 elementary charges.

Substitution of cysteine for pore-lining residues in the stem region does not significantly alter the gating properties of the channel under reducing conditions; a typical example is seen with S312C channels (Fig. 2 A, upper trace, and B). (We suspect that the small nonzero open probability [0.05] at large negative voltages in the unreacted S312C channel is a result of a small amount of oxidation occurring despite the presence of a reducing agent, as discussed in greater detail below.) Exposure of such a channel to the sulfhydryl-reactive reagent MTSET, however, dramatically perturbs gating—reacted channels no longer exhibit any gating at negative potentials, but instead remain essentially locked in a conducting state throughout the voltage range studied (Fig. 2 A, lower trace, and B). This can be seen both as a failure of channels to close at negative voltages, and by a lack of any turning on at positive voltages (compare upper and lower traces of Fig. 1 A). Indeed, this effect induced by MTSET unmasks the small amount of deactivation gating observed at larger positive voltages that is typically obscured by the concurrent activation gating seen in pulsing from −100 mV to the test pulse in the protocol in Fig. 1 A.

The record shown in Fig. 2 C illustrates several other features of the MTSET reaction. MTSET exerts a large effect on the conductance of cysteine-substituted anthrax channels, as first shown by Benson et al. (1998). Addition of 400 μM to the trans side of the bilayer leads to an ∼75% decrease in conductance over several seconds. Reacted channels also rectify, passing more current at positive voltages than at negative voltages. This rectification, as well as the locking open effect, can be seen more clearly in Fig. 2 D, in which the current trace following MTSET exposure (solid black line) has been rescaled (solid gray line) by renormalizing it to the instantaneous current at −60 mV before MTSET addition (dotted line). Although not shown, the addition of DTT or TCEP to the chamber restores both the conductance and the voltage-gating behavior exhibited before MTSET exposure. The elimination of gating is not due to a simple shift in the voltage-activation relation to more negative potentials; as seen in the normalized I-V plot in Fig. 2 B, the relative open probability remains very high throughout the entire voltage range examined. (Although not shown, channels exhibit minimal gating even when pulsed to voltages in the −100 to −150 mV range; however, bilayers were not stable enough at these voltages for the durations necessary to construct a full normalized I-V plot.) Reaction of S312C channels with MTS-glucose, a polar, uncharged reagent, also prevents gating, although its effect on conductance is smaller and it reacts more slowly than MTSET; 400 μM MTS-glucose irreversibly reduces current by ∼40% after ∼20 s (Fig. 3).

Probing cysteines substituted at other sites along the β-barrel stem of the channel with MTS reagents reveals an interesting and altogether unexpected result: their ability to lock channels in a conducting state is not restricted to S312C or to any one particular region of the barrel. We see this same effect at every pore-lining cysteine substitution we examined throughout the entire stem region, with the sole exception of F313C, predicted to lie near the turn of the barrel (Fig. 4). Furthermore, the ability of an MTS reagent to eliminate gating is not correlated with its effects on conductance, and this is evident upon comparing the effect of a single agent at all of the sites tested, or comparing MTSET and MTS-glucose at any one site.

NEM, another thiol-reactive reagent, also reacts with pore-lining residues throughout the stem region, although its effect on conductance is smaller than that of MTSET at each site studied (Fig. 4), with some sites showing only a minimal effect (e.g., 312, 314, 323, and 335). NEM does label these sites because washout and subsequent exposure to MTSET has little effect (not depicted). Although the overall pattern of NEM reactivity matches that of MTS reagents, its behavior is strikingly different in one important way: NEM has no effect on gating at any of the sites that we examined (Figs. 4 and 5).

Our observation that two different MTS reagents are capable of eliminating gating, whereas NEM is not, argues that the details of the thiol chemistry play an important role in this phenomenon. Reaction of a thiol with an MTS reagent generates a mixed disulfide bond, whereas maleimides react via a Michael addition that generates a carbon–sulfur bond. This led us to investigate the notion that MTS reagents might be catalyzing thiol exchange to generate intersubunit (and intramolecular) disulfides between neighboring cysteines in the pore, as suggested previously (Zhang et al., 2004b). Consistent with this hypothesis is our observation that at certain sites in the pore, individual substitution for cysteine results in channels that are susceptible to gradual rundown in current if a reducing agent like DTT or TCEP is washed out of the chamber, an effect that can be accelerated dramatically by the addition of the oxidant H₂O₂ (Fig. 6 A). Furthermore, the residual current exhibits minimal gating at negative voltages (Fig. 6 B). Both the decline in current and the loss of gating could always be reversed by the readdition of reducing agent (Fig. 6). We also found that if these cysteine-substituted (PA₆₃)₇ channels are not maintained under reducing conditions, their channel activity upon addition to a bilayer bathed in DTT- or TCEP-free solutions is substantially reduced, and the addition of reducing agent leads to a large increase in current. Such observations also argue against the generation of higher oxidation states of sulfur because these are not readily reversed by these reducing agents.

To test the possibility that oxidation reduces conductance by decreasing the diameter of the channel lumen, we examined its effect on the rate of protein translocation through the channel. We chose for our assay LF_N, the 263-residue N-terminal fragment of LF, because the effects of voltage and pH on its rate of transport through anthrax channels have been characterized previously (Zhang et al., 2004b; Krantz et al., 2006). When LF_N is added to the cis side of a bilayer containing S312C channels held at +20 mV in the presence of cis TCEP, it enters the channel and blocks the flow of the current-carrying ion (K⁺); this leads to the large decrease in current seen in the first portion of Fig. 7 A. If the cis pH is lowered from 6.6 to 5.6, LF_N is driven completely through the channel into the trans compartment. This translocation, which can be monitored by the increase in current that reflects relief of block, takes several seconds (Fig. 7 A). In contrast, translocation of LF_N is much slower through S312C channels that have been exposed to H₂O₂; relief of block of these channels, whose conductance is much lower, requires several minutes (Fig. 7 B). Subsequent addition of DTT to the cis compartment restores both the conductance and gating properties of the channels (Fig. 7 C), as well as the rate of LF_N translocation. The effects of LF_N on oxidized channels are reassuring in another regard—they provide evidence that this nongated conductance is due to properly assembled channels that possess an intact LF_N docking site (Phe clamp) (Krantz et al., 2005), and not the result of some nonspecific leak induced by oxidized protein.

All of the experiments presented thus far in support of intersubunit disulfide formation have been electrophysiological. To further test this hypothesis, we turned to a biochemical approach that would allow us to detect disulfides by assaying for dimer formation on a gel. Miller et al. (1999) have previously shown that a reduction in the pH of a solution of purified (PA₆₃)₇ from 8.5 to 6.5 leads to an SDS-resistant form of this heptamer. To more closely mimic the state of an anthrax channel embedded in a bilayer, we reconstituted His₆-tagged heptameric S312C (PA₆₃)₇ prepore complexes at high pH under reducing conditions into liposomes doped with DOGS-NTA, a lipid that contains a nickel-NTA moiety at its head group. This approach has been shown to help orient the protein at the bilayer and improve the yield of properly inserted heptameric channels after pH reduction (Sun et al., 2007). When this prepore form of the protein is exposed to denaturing conditions, it dissociates into PA₆₃ momoners (Fig. 8, lanes 1 and 2); however, if the pH of the sample is first lowered to 6.5, a substantial portion transforms from the soluble prepore form to the heptameric SDS-resistant species (Fig. 8, lanes 3 and 6). If this transformation proceeds under oxidizing conditions (removal of TCEP, and in 0.3% H₂O₂) and the sample is subsequently boiled in urea, the heptameric band is no longer resolvable; however, two other bands are now present: one consistent with a monomeric species, and a higher molecular weight band consistent with dimer formation (Fig. 8, lane 4). (Although these data were generated in the presence of H₂O₂, similar results are obtained in its absence.) If this same sample is boiled in urea and exposed to DTT, the dimer band is no longer evident (Fig. 8, lane 5). Could the dimer band seen in lane 4 result from disulfide bond formation between exposed cysteines of denatured monomers rather than from cysteines of neighboring subunits in the SDS-resistant heptamer? There are two reasons why we think this is unlikely: (1) a high concentration of NEM is present just before and during denaturation, and this would likely quench any exposed thiols; and (2) we do not observe dimer formation in S330C mutant (PA₆₃)₇, where the cysteine is predicted to face away from the pore (Fig. 8, lanes 6–8).

Over the past two decades, substantial progress has been made in understanding the role played by the PA component of anthrax toxin in the pathogenesis of Bacillus anthracis. Much of the focus has been on the assembly of PA₆₃ into a heptameric channel, and on the function of this pore as a conduit for the membrane translocation of LF. One aspect of the behavior of anthrax channels—their opening and closing in response to changes in transmembrane voltage—has been little studied. In the course of exploring the effects of thiol-reactive compounds on pore-lining cysteine residues in these channels, we discovered an unexpected phenomenon that provides a window into this gating process: cysteine-substituted channels exposed to MTS reagents remain locked in a conducting state even at large negative voltages that would otherwise fully close unreacted channels. Three features of this gating perturbation are noteworthy: (1) The mechanism by which MTS reagents eliminate gating—catalyzing the formation of intersubunit disulfide bonds via disulfide exchange—has not been demonstrated in an ion channel, and it is different from that generally proposed for these compounds (Karlin and Akabas, 1998); (2) Reaction of cysteines at sites all along the ∼100 Å length of the channel's β-barrel generates channels that do not gate. This contrasts with results in other voltage-gated channels in which reaction at a specific site or region locks channels open (Holmgren et al., 1998; Sukhareva et al., 2003); and (3) The geometric constraints for disulfide bond formation suggest that the there must be enough flexibility along the length of the β-barrel to allow significant conformational changes to occur.

The rationale for using an MTS reagent as a probe to study structural features of an ion channel is that reaction with a solvent-exposed cysteine introduces an adduct whose charge or bulk will affect the channel's conductance if that cysteine is coincident with or near the channel's ion conduction pathway (Karlin and Akabas, 1998). Typically, the effect of introducing a charged moiety like ethyltrimethylammonium is different than that of a neutral group like glucose, although exceptions to this have been described (Bera et al., 2002; Reeves et al., 2005). Because we find that two different MTS reagents affect gating in the same way, regardless of charge, we wondered whether the lock-open effect might arise from an action of the MTS reagent that is independent of the introduced adduct. Additional support for this notion was the observation that reaction of a single anthrax channel with either MTS-glucose or MTSET results in up to four to five successive step-wise drops in conductance (Benson et al., 1998; Zhang et al., 2004b). Because it is difficult to envision, given estimates of the size of the anthrax channel pore coupled with the size and/or charge of an -ET or -glucose adduct, that the four to five distinct chemical reactions underlying these conductance drops are the successive additions of ET or glucose groups into the channel's pore (Fig. 9 A), Zhang et al. (2004b) speculated that some of the observed step-wise drops in conductance represent disulfide exchange reactions between neighboring subunits. This mechanism is illustrated in Fig. 9 B. An initial reaction of a cysteine thiol with an MTS-X reagent generates the mixed disulfide PA-S-S-X (where PA represents one of the seven channel subunits), which then participates in thiol exchange with a cysteine on an adjacent subunit to generate an intersubunit disulfide bond between the neighboring cysteines. The addition of the S–X group into the pore generates a channel with lower unitary conductance because of steric and/or electrostatic effects. Why would generation of an intersubunit disulfide lower the conductance? Although we discuss the structural implications of disulfide formation in greater detail below, we suspect that the lumen of a channel containing an intersubunit disulfide is narrower because the cysteines on adjacent subunits that participate in a disulfide bond will be significantly closer than they would be in an open, reduced channel.

There are seven potential cysteine targets in a heptameric cysteine-substituted channel, and we do not yet know how many disulfides and/or –X adducts remain in each reacted channel upon exposure to MTSET or MTS-glucose. The nature and number of reactions that take place will likely be influenced by the size and charge of the MTS reagent. We do suspect, at least for MTSET, that the final reacted state is one in which, in addition to the disulfides, an -ET group is present in the pore because such channels exhibit an outward rectification not seen in unreacted channels or channels reacted with MTS-glucose, and reaction with MTSET and MTS-glucose have different effects on conductance.

Further evidence that the effect of MTS reagents on gating is not due to the addition of an adduct in the pore is our observation that at several locations, NEM decreases conductance but never affects gating. Although MTS reagents and maleimides each react specifically with cysteine thiols, they do so using very different mechanisms. Maleimides generate a carbon–sulfur bond that cannot participate in disulfide exchange (and therefore cannot be reduced by DTT). In addition, reaction with a maleimide leaves a bulky group in the pore, yet that action does not affect gating at any of the sites we have tested.

Oxidation experiments lend additional support to our hypothesis that locking into a conducting state involves the formation of disulfide bonds. In the absence of any reducing agent, certain cysteine-substituted channels spontaneously lose their ability to gate, and this occurs much more rapidly in the presence of H₂O₂. Oxidized channels translocate K⁺ ions and proteins more slowly, demonstrating that oxidation narrows the channel's lumen. All of these effects are fully reversed by the readdition of a reducing agent, which suggests that oxidation is not decreasing conductance by generating higher oxidation states of pore-lining cysteines.

The final line of evidence in support of disulfide bond formation is our biochemical detection of dimer formation. When we analyze cysteine-substituted anthrax channels that have been reconstituted into lipid vesicles and incubated under oxidizing conditions, we are able to detect a fraction whose migration pattern on a gel is consistent with a dimer. Furthermore, this fraction is no longer resolvable if exposed to DTT, and it is not seen when reconstituting a cysteine-substituted channel whose cysteine is predicted to face away from the pore.

From a structural perspective, the most compelling conclusion emerging from our findings is that the channel's entire β-barrel stem region participates in the gating process. Absent a high resolution structure of the pore form of the anthrax channel in both the closed and open states, we cannot draw a detailed picture of the gating motions from our data. Nevertheless, we can gain some insight into the magnitude of the conformational changes that might occur during gating by considering the x-ray crystal structure of α-hemolysin, which likely shares structural features with the stem domain of the anthrax channel (Song et al., 1996; Benson et al., 1998; Nassi et al., 2002; Krantz et al., 2004; Nguyen, 2004).

The luminal diameter (C_α–C_α) of α-hemolysin's β-barrel is ∼26 Å (accounting for sidechain volume, which is relevant to LF or EF translocation, yields a diameter of ∼19 Å [Krantz et al., 2004]), with residues of adjacent subunits being an average distance (C_β–C_β) of ∼11 Å from each other. If the β-barrel of the anthrax channel is of similar dimensions, it must be flexible enough to accommodate a sizable conformational change because disulfide bond formation requires that β carbons of participating cysteines come within 4–5 Å of each other (Careaga and Falke, 1992). What can a disulfide-containing channel that appears locked open teach us about gating motions? By trapping the channel into a nongating lower conductance state whose lumen is narrower than that of the open state, yet wider than that of the closed state, intersubunit disulfides may be capturing a gating transition on the path to the final closed configuration. Their effect would be to “straitjacket” the channel by forming a buttress that either prevents full collapse of the entire barrel, or freezes the channel into a state that doesn't allow some more subtle conformational change from propagating to a gate. Is this state a grossly distorted one that bears no resemblance to the channel's native state(s)? Several observations of reduced and oxidized channels argue against this possibility. The cation selectivity following reaction of S312C channels with MTS-glucose is essentially unchanged (reversal potential of −25 mV pre- and −24 mV post-reaction, in a fivefold cis:trans KCl gradient). This suggests that an important feature of the pore remains intact. LF_N blocks reduced and oxidized channels with similar potency, which argues that oxidation does not disrupt the channel's Phe clamp, the binding site for LF located in the permeation pathway in the channel's cap domain. Finally, tetrabutylammonium blocks reduced and oxidized channels with equal potency from the cis side, lending further support to the notion that the Phe clamp is intact in oxidized channels. Thus, the only perturbation in channel structure and function that we can detect in reacted channels is a lack of appreciable gating.

K⁺ is the major permeant ion in our experiments, and given the size and chemical nature of the pore, it likely traverses the channel in its hydrated state (diameter of ∼8 Å). This implies that for the channel to gate shut, it must reduce its luminal diameter to less than this value somewhere along its permeation path. Although we have not yet pinpointed the exact location and extent of this gate, our preliminary examination of the state dependence of MTS-glucose reactivity suggests the presence of such a narrowing of both cis and trans to E343C (unpublished data). Because this residue appears to lie ∼3/4 of the way up the barrel (Fig. 4), and because all of the stem domain trans to this site is predicted to have β-barrel architecture, this would imply that the dynamics of gating involve a reversible change in diameter of more than a factor of 2 in some part of the barrel.

Given our observation that disulfides all along the barrel can perturb gating, and that there is gated access to residue 343 from both the cis and trans sides of the channel, we must consider the possibility that the entire barrel collapses and reexpands during gating. Although a motion of this magnitude has not been demonstrated previously in a β-barrel protein, it is worth noting that the high thermal factors in this region of the crystal structure of α-hemolysin suggest significant mobility there (Song et al., 1996). In addition, a β-barrel can change its diameter while maintaining a constant strand number if it changes its shear number (McLachlan, 1979) and perforce, the tilt of its β strands relative to the membrane normal. This phenomenon is seen in MspA, an octameric 16-stranded β-barrel porin found in the outer membrane of Mycobacterium smegmatis (Faller et al., 2004). The pore of MspA exhibits an abrupt reduction in diameter from 40 to 28 Å without changing either its β architecture or strand number; this change is associated with a drop in shear number from 32 to 16, and β-strand tilt angle from 56 to 37 degrees.

Of the MTS-sensitive residues that we have probed, only F313C shows no effect on gating upon reaction with an MTS reagent. This residue is predicted to lie at the protein–lipid interface at the very trans end of the barrel, and our failure to see a gating effect there may indicate that the trans end of the barrel is significantly wider.

Our data have allowed us to speculate about the mechanism of voltage gating in anthrax channels. A more thorough understanding of this process will ultimately require, as with any voltage-gated ion channel, that we (1) delineate the channel's gate, (2) identify the voltage-sensing residues, and (3) determine the structural connection between the sensor and the gate, goals that we are currently pursuing.

We thank Amy Sinor, Rachel Darman, Alan Finkelstein, and Dan Cox for helpful discussions, and Merritt Maduke and Chris Miller for critically reviewing the manuscript.

This work was supported by National Institutes of Health grants R01HL68985 (to R. Blaustein) and T32HL069770 (to D. Anderson).

Angus C. Nairn served as editor.