A Phenylalanine Clamp Catalyzes Protein Translocation Through the Anthrax Toxin Pore

Abstract

The protective antigen component of anthrax toxin forms a homoheptameric pore in the endosomal membrane, creating a narrow passageway for the enzymatic components of the toxin to enter the cytosol. We found that, during conversion of the heptameric precursor to the pore, the seven phenylalanine-427 residues converged within the lumen, generating a radially symmetric heptad of solvent-exposed aromatic rings. This “φ-clamp” structure was required for protein translocation and comprised the major conductance-blocking site for hydrophobic drugs and model cations. We conclude that the φ clamp serves a chaperone-like function, interacting with hydrophobic sequences presented by the protein substrate as it unfolds during translocation.

Anthrax toxin is composed of three nontoxic proteins, which combine on eukaryotic cell surfaces to form toxic, noncovalent complexes. [See (1) for a review.] Protective antigen (PA), the protein translocase component, binds to a cellular receptor and is activated by a furin-family protease. The resulting 63-kD receptor-bound fragment, PA₆₃, self-assembles into the prepore, which is a ring-shaped homoheptamer (Fig. 1A). The prepore then forms complexes with the two ~90-kD enzymatic components, lethal factor (LF) and edema factor (EF). These complexes are endocytosed and delivered to an acidic compartment (2). There, the prepore undergoes an acidic pH-dependent conformational rearrangement (3) to form an ion-conducting, cation-selective, transmembrane pore (4), allowing bound LF and EF to translocate into the cytosol.

Fig. 1.

Structural models of a lumen-facing phenylalanine heptad. (A) A ribbons rendering of the PA₆₃ prepore (27), viewed axially, where domain 4 is proximal. Domains are colored: D1′ (magenta), D2 (green), D3 (gold), and D4 (blue). F427 (red, space filling) is modeled into the structure. (B) Hypothetical cross section of the PA₆₃ channel, or pore, colored as in (A). The membrane-spanning tube is the 14-stranded β barrel from domain 2 (5, 6). (C) Illustration of the effect of MTS-ET modification on Cys-substituted mutants of PA₆₃ in macroscopic conductance studies. Conductance, g, is determined from the current, I, and Δψ as g = I/Δψ. (D) Fraction of conductance blocked ( f_block) by MTS-ET modification (28) in domain 2 cap residues [as in (C), where f_block = 1 − g_block/g] (table S2). Error bars show means + σ_SE (n = 3). (E) EPR spectra of PA₆₃ heptamers uniformly labeled at F427C with a Cys-reactive nitroxide spin label in the prepore state at pH 8.5 (upper spectrum) and the pore state at pH 6 (lower spectrum). Approximate luminal diameters, d, are based on the observed spin-spin interactions. (F) Unitary conductance, γ, of single PA₆₃ channels, with indicated substitutions at F⁴²⁷. Channels formed by F427G PA₆₃ (*) initially opened to a conductance of 90 pS, but, unlike any of the other channels, flickered to 60 and 30 pS substates. γ values are accurate to at least ±10%, except for F427L and F427W, which are accurate to ±20%.

The PA₆₃ pore (Fig. 1B) is believed to consist of a mushroom-shaped structure, with a globular cap connected to a β-barrel stem that is ~100 Å long (5, 6). A model of the 14-strand β barrel reveals its lumen, which is ~15 Å wide and can only accommodate structure as wide as an α helix (7). The narrow pore creates a structural bottleneck, requiring that the catalytic factors, LF and EF, unfold in order to be translocated (8, 9). The destabilization energy required to unfold the tertiary structure of LF and EF originates partly from the acidic pH in endosomes, which causes their N-terminal domains (LF_N and EF_N) to become molten globules (MG) (7). A positive membrane potential [+Δψ (10)], when coupled with these acidic pH conditions, is sufficient to drive LF_N through PA₆₃ pores formed in planar lipid bilayers (9). To enter the narrow confines of the ~15-Å-wide lumen, LF_N must shed its residual tertiary structure and convert from the MG form to an extended, “translocatable” conformation (7). How does a solvent-filled pore mediate the disassembly of an MG protein, packed, albeit loosely, with hydrophobically dense stretches of polypeptide? We surmised that an interaction surface inside the pore might facilitate further unfolding of the MG to the extended, translocatable form.

By cysteine-scanning mutagenesis coupled to [2-(trimethylammonium) ethyl] methanethiosulfonate (MTS-ET) modification (5) (Fig. 1C), we identified residues that line the lumen of PA₆₃ in the globular cap portion of domain 2, the pore-forming domain (Fig. 1D). F⁴²⁷ (11) was the most hydrophobic residue identified in the otherwise hydrophilic pore lining. It is absolutely conserved in homologous toxins (fig. S1), and mutating it blocks protein translocation (12, 13). F⁴²⁷→C⁴²⁷ (F427C) channels were most strongly affected by MTS-ET modification, implying that F⁴²⁷ is prominent and solvent-exposed within the lumen.

To address how the seven F⁴²⁷ residues were arranged within the lumen of the PA₆₃ pore, we used electron paramagnetic resonance spectroscopy (EPR) to measure the proximity of nitroxide spin labels attached to F427C. The EPR spectrum of the spin-labeled prepore showed weak spin-spin interactions (Fig. 1E). This observation was consistent with the crystal structure of the prepore, in which F⁴²⁷ lies in a disordered loop (2β10 to 2β11) near the lumen, such that neighboring F⁴²⁷ residues are 15 to 20 Å apart (Fig. 1A). Upon conversion to the pore state by acidification to pH 6, a saturating spin-spin interaction appeared, indicating that the spin probes had converged and were separated by less than 10 Å (Fig. 1E). Consistent with this, single-channel ion conductances were roughly inversely proportional to the size of the substitution at 427 (Fig. 1F), as predicted from a cylindrical pore conductance model (14). Channels with large aliphatic (Leu) or aromatic (Trp) residues at 427 showed smaller conductances than channels with Ala or Phe, their smaller respective counterparts. Thus, as the pore forms, the seven phenylalanines create a narrow aromatic iris, or “ring of rings,” within the lumen (Fig. 1B and supporting online material text).

In planar lipid bilayers, LF_N binds to wild-type (WT) PA₆₃ pores, blocking ion conductance by >95% (at +20 mV and ~10 nM LF_N; Fig. 2C); the first 21 residues of LF_N’s cation-rich, flexible N terminus are essential for this macroscopic blocking effect (15). At the single-channel level, the N terminus of LF_N bound stably within the PA₆₃ channel and prevented passage of hydrated K⁺ ions, as manifested by a continuously closed state (Fig. 2A). When F427A PA₆₃ was similarly assayed, LF_N was only able to block macroscopic conductance by ~50% (Fig. 2C). In single-channel recordings, the partial block observed for F427A PA₆₃ channels resulted from a dynamic “flickering” between an open state, multiple partly closed substates, and a fully closed state (Fig. 2A). Histogram analysis of a single-channel recording of F427A PA₆₃ (Fig. 2B) was consistent with macroscopic experiments, in that the average conductance was ~50% of the open channel conductance (Fig. 2C). We infer that LF_N is bound to the domain 1′ surface of the F427A channel, but its flexible N terminus, which enters the channel first (15), is not stably bound within the lumen and is unable to block conductance effectively. LF_N blocked PA₆₃ pores more effectively with large aromatic or aliphatic residues at position 427 than with small or hydrophilic residues (fig. S2). Thus the heptad of F⁴²⁷ residues is integral to binding LF_N’s N terminus, leading us to term the site the “φ clamp.”

Fig. 2.

LF_N conductance block and translocation studies with PA₆₃ mutated at F⁴²⁷. (A) Current record for a single F427A PA₆₃ channel (upper panel) exhibiting partial block with 7 nM LF_N in the cis compartment. A similar single-channel record for a WT channel (lower panel) shows complete block with 7 nM LF_N in the cis compartment [Δψ = +20 mV (10), pH 5.5]. A diagram (right) showing the “blocked” channel corresponds to LF_N’s N terminus binding within the pore. (B) Histogram of the single-channel conductances for the LF_N block in (A) for WT (black) and F427A PA₆₃ (gray). (C) Concentration dependence of the LF_N conductance block for WT (○) and F427A PA₆₃ channels (•) under macroscopic conditions (Δψ = +20 mV, pH 5.5). Normalized conductance (g_norm) curves are fitted to a single- and two-site binding model forWT and F427A channels, respectively, where the WT equilibrium dissociation constant (K_D) is ~150 pM. (D) Planar lipid bilayer macroscopic conductance records of LF_N translocation through WT and the indicated F⁴²⁷ mutant channels [where pH_cis = 5.5, pH_trans = 6.5 (29)]. After PA₆₃-induced channel formation reached steady state, 20 nM LF_N was added to the cis compartment (under a Δψ of +1 to +10 mV). When the conductance block reached steady state, the cis compartment was perfused, and translocation (as manifested by the rise in conductance) was initiated by increasing Δψ to +30 mV. Plotted records were normalized as the fraction translocated, excepting F427A and F427D channels, in which no appreciable translocation occurred. (E) Cellular translocation assay of F427 PA mutants (12), where translocation is reported by detecting the inhibition of protein synthesis by the domain from diphtheria toxin, which is fused to the C terminus of LF_N. Mutants are classified as active (+) or inactive (−).

To probe the role of the φ clamp in polypeptide translocation, we measured the rate of LF_N translocation through PA₆₃ channels formed in planar lipid bilayers (9). LF_N was added to the cis compartment, blocking conductance. After perfusing to remove unbound LF_N, we stepped Δψ to a higher positive voltage and monitored the rate of translocation by the increase in channel conductance as LF_N traversed the pore (Fig. 2D). With the most active substitutions at F⁴²⁷, LF_N translocated at rates that trended as Phe > Leu ≈ Trp > Tyr (Fig. 2D). Other small aliphatic or hydrophilic substitutions, including Asp (an isosteric, hydrophilic control for Leu), were inefficient at promoting translocation. The most active residue, therefore, was Phe; the more hydrophilic aromatic, Tyr, was five times less active. These in vitro translocation studies recapitulated results of cellular assays of toxin action (Fig. 2E). Preference at the φ-clamp site for the γ-branched Leu over its β-branched isomer, Ile, correlates with the fact that aromatic residues also place more hydrophobic surface nearer to the center of the lumen. Thus the φ-clamp structure represents an active site, requiring either an aromatic surface or a more centrally oriented aliphatic surface to catalyze translocation efficiently.

Considering the hydrophobicity of the φ clamp, we hypothesized that this site may also be the binding site for hydrophobic cations, such as tetrabutylammonium (TBA) (Fig. 3A) (16). This hypothesis is consistent with studies suggesting that the TBA site is within the cap of the PA₆₃ channel, cis to the extended β barrel (6). TBA’s affinity for F427A channels was greatly reduced from that of WT (4000- fold, or ~5 kcal mol⁻¹; Fig. 3D). Large aliphatic residues at 427 were unable to recover the TBA block (Fig. 3E). F427L channels, for example, had a ~4 kcal mol⁻¹ reduction in affinity for TBA. The least defective substitution was F427Y, which reduced the TBA block fourfold, or ~0.8 kcal mol⁻¹. The recovery of the TBA block in F427Y channels was notable, because Tyr is substantially more hydrophilic than Phe. We infer that the TBA blocking mechanism also includes cation-π interactions, which occur when aromatic residues interact with cations through their delocalized, negative π-electron clouds (17).

Fig. 3.

QAP conductance block at the φ-clamp site. (A) TBA, (B) TPP, and (C) a hydrophilic analog of TBA (2-acetylamino-2,2-bis-ethoxy-carbonyl- ethyl)-trimethyl-ammonium, colored by atom: C (black), N (blue), O (red), H (white), and P (orange). (D) QAP ions were added symmetrically to the cis and trans compartments (Δψ = +20 mV, pH 5.5), and g_norm was recorded once QAP binding reached equilibrium. For TPP block, single-site binding models were fit for WT PA₆₃ (red ○), K_D = 46 ± 2 nM (±SD), and F427A PA₆₃ channels (red •), K_D = 1.7 ± 0.2 mM. For TBA block, a single-site binding model was fit for WT PA₆₃ (black □), K_D = 7.3 ± 0.3 μM, but a two-site binding model was required for F427A PA₆₃ (black ▪), such that the K_D values of the major amplitude (88%) and minor amplitude (12%) were 30 ± 2 mM and 1.5 ± 0.7 μM, respectively. (E) Binding free energy change, $\mathrm{\Delta }\mathrm{\Delta }G=RT\hspace{0.17em}ln{K}_{\text{D}}^{\text{Mut}}/K_{\text{D}}^{\text{WT}}$, of TBA (white bars) and TPP (black bars) block for F⁴²⁷ mutants. (F) Model compound blocking studies of WT PA₆₃ channels, performed as in (D). Experimental binding energies (ΔG_exp = RT ln K_D) were correlated to the expected binding energies (ΔG_theo) composed of solvation (ΔG_s) (18) and aromatic enhancement (ΔG_aro) energies as well as an offset, c, where ΔG_theo = ΔG_s + ΔG_aro + c. ΔG_aro = αn, where α was fit to 0.7 ± 0.3 kcal mol⁻¹ per aromatic ring, using the number of aromatic rings per compound, n. c was 0.8 ± 0.4 kcal mol⁻¹ (table S1). The linear fit, ΔG_exp = ΔG_theo × m (35 compounds), had a slope, m, of 0.95 ± 0.03 and total error, σ_SD, of 1.2 kcal mol⁻¹. Arrows indicate compounds in (A), (B), and (C).

We next examined a library of 35 quaternary ammonium and phosphonium ion (QAP) compounds to establish the nature of the φ-clamp binding interaction. Aliphatic QAP compounds (table S1) blocked conductance according to a solvent-accessible surface area solvation energy model (18); i.e., compounds with more hydrophobic surface blocked more effectively (Fig. 3F). Thus, a hydrophilic analog of TBA that is comparable in size, but functionalized with hydrophilic amide and ester groups (Fig. 3C), blocked 140-fold more weakly than TBA (table S1). We found that WT channels preferred tetraphenylphosphonium (TPP) (Fig. 3B) to TBA by 160-fold (Fig. 3D). Broadly across the QAP library, the φ clamp preferred aromatic moieties by ~0.7 kcal mol⁻¹ per aromatic ring (Fig. 3F and table S1). Specifically, the conductance block observed for the polyaromatic, 4-aminoquinolone drug, quinacrine (19), was reduced ~1000-fold in F427A channels (fig. S3), indicating the φ-clamp site may be exploited in the development of channel-blocking drugs. Thus the correlation observed for the diverse QAP compound library (Fig. 3F and table S1) indicates that the φ clamp does not, in and of itself, recognize specific geometric or steric features of the substrates. Instead, the WT φ clamp recognizes substrates primarily by nonspecific hydrophobic interactions (20), although its negative π-clouds also contribute electrostatically through aromatic-aromatic, π-π and cation-π interactions.

In determining a model for translocation, we disfavored a priori relationships used to describe metal ion-conducting channels, because metal ion throughput (Fig. 1F) did not correlate with protein transport (Fig. 2, D and E), and the steric constraints imposed by the φ-clamp site did not impede protein translocation. The WT φ-clamp site should create a hydrophobic and steric energy barrier for bulky, hydrophilic, or charged residues in the protein substrate. Paradoxically, although channels containing narrower, aromatically lined φ-clamp sites were less ion conducting, they translocated LF_N orders of magnitude more rapidly than channels made wider, more hydrophilic, and more ion conducting at the φ-clamp site.

We therefore considered a chaperone model in which the φ-clamp’s phenyl rings directly interacted with the translocating polypeptide. This view is supported, because the φ clamp forms a narrow iris (Fig. 1E) that can effectively grasp the translocating polypeptide chain (Fig. 2A, diagram). Furthermore, model compound studies indicate that the φ clamp recognizes substrates through the hydrophobic effect, enhanced by aromatic-aromatic, π-π, and cation-π interactions (Fig. 3 and table S1). Similarly, in the potassium ion channel from Streptomyces lividans (KcsA), a hydrophobic cavity defines the TBA blocking site and corresponds to a docking site for a hydrophobically dense peptide sequence (21). In globular proteins, like LF, hydrophobically dense peptide segments occur periodically along the translocating chain (Fig. 4A). As these segments unwind from the MG protein, they would be expected to bind favorably to the φ clamp, causing translocation to be blocked kinetically by the unfavorable hydrophilic barrier in the Ser/Thr-rich β barrel (Fig. 4B). Thus the φ-clamp energy well might be expected to hinder translocation, causing polypeptide segments bound at the site to “pause.” However, as the φ-clamp site actually catalyzes polypeptide translocation, it must reduce some other larger barrier, such as the unfolding of the substrate protein. By analogy to the KcsA channel’s selectivity filter, which provides sequential rings of hydrophilic carbonyl oxygen atoms that mimic the inner hydration shell of a K⁺ ion (22), we propose that the φ-clamp site creates an environment that mimics the hydrophobic core of the unfolding MG protein. This would reduce the energetic penalty of exposing hydrophobic sequences to solvent or the hydrophilic lumen of the channel. The PA₆₃ pore would thereby function as a Brownian ratchet, enabling the unwound leading segment of a translocating protein to move through the channel, and the trailing part of the protein to more readily unfold (Fig. 4C).

Fig. 4.

A model of φ-clamp catalyzed protein translocation. (A) Chemical complexity of a PA₆₃ substrate, LF_N, in which residues are colored by functionality: hydrophobic (green), greater than −1.75 kcal mol⁻¹ in solvation energy (18) after applying a 10-residue running-window average; cationic (blue); and anionic (red). (B) The PA₆₃ pore with a luminal φ-clamp site (red) is structurally imposed on an energy-well diagram for a hydrophobic stretch of polypeptide sequence from LF_N [see (A)]. The φ-clamp site is the anticipated well, separating the unfolding barrier on the cis side in the solvent-filled cap from a translocation barrier on the trans side in the solvophilic, Ser/Thr-rich β barrel. Energy diagrams are for WT and F427A PA₆₃ (solid and broken lines, respectively), where the mutation simultaneously reduces the hydrophobic-mediated stabilization imparted by the φ-clamp site and raises the barrier to unfolding hydrophobic sequences from the protein substrate. (C) Intermediate states for a stepwise, Brownian ratchet unfolding and translocation mechanism, such that hydrophobically dense polypeptide segments (green) interact with the φ-clamp constriction.

The results presented here show that PA₆₃ is not merely a passive conduit through which proteins electrophorese; rather, it actively engages the translocating substrate via the φ clamp. This paradigm may be relevant to the functions of other polymer-translocating channels containing exposed hydrophobic sites, such as the “hydrophobic gasket” and “aromatic slide,” identified in the structures of the protein secretase (23) and maltoporin (24) channels, respectively.