Introduction

Anthrax toxin (AT) is the major virulence factor secreted by the bacterium Bacillus anthracis and consists of three components: edema factor (EF, 89 kDa), lethal factor (LF, 83 kDa) and protective antigen (PA, 83 kDa). EF (PDB code 1K8T) (1) and LF (PDB code 1J7N) (2) are the enzymatic components of anthrax toxin. EF is an adenylyl cyclase (3) and LF is a zinc metalloprotease (4). While EF and LF may be responsible for the lethality of anthrax, the cellular activities of both enzymes depends upon the ability of PA to deliver the two enzymes into the cell?s cytosol. The general mechanism of cell intoxication is shown in Figure 1. Following the release of the tripartite toxin by Bacillus anthracis, PA binds to AT receptors on the cell surface (5). The proteolytic cleavage of PA by furin or a furin-like protease releases the 20 kDa N-terminal segment (6) and allows the 63 kDa C-terminal fragment (PA63) to heptamerize into a ring-like structure (the PA63 prepore) (7). The oligomerization of PA63 forms the binding sites for the protein ligands, EF and LF (8, 9). After EF and LF bind to the PA63 prepore, the PA63 prepore-EF/LF complex is trafficked to the endosome (10), where the acidic environment induces conformational changes in the PA63 prepore to form a pore structure (the PA63 pore) which traverses the bilayer. Simultaneous to or following pore formation, EF and LF are released into the cell?s cytosol, where their enzymatic activity ultimately results in septic shock and death (11).

Figure 1: General mechanism of cell entry by anthrax toxin. PA, PA63 and PA20 are represented by yellow ovals, EF/LF by green wedge-shapes, and the furin-like protease by purple free-form. (top) PA binds to AT receptors at the cell surface, where furin or a furin-like protease releases a 20 kDa N-terminal fragment. PA63 (63 kDa) assembles into a ring-like heptamer (the PA63 prepore), which binds EF and LF. The PA63 prepore-EF/LF complex is trafficked to the endosome, where the acidic environment triggers the formation of a pore structure (the PA63 pore). EF and LF are released into the cell?s cytosol. (bottom) Pore formation involves the unfolding of the Greek-key motif (strands 2β1-2β4) in the PA63 prepore to produce a 14-strand, membrane-inserted β-barrel.

Because of PA?s central role in cell intoxication by anthrax toxin, it has been the subject of intense study. The X-ray structure of monomeric PA (PDB code 1ACC) at 2.1 Å has been reported (12). Protective antigen is a long and flat molecule with dimensions of 100 Å × 50 Å × 30 Å (Figure 2). As described by Petosa et al. (12), PA consists of four domains. Domain 1 (residues 1-258) has the twin calcium-binding site (13) as well as the cleavage site for furin or the furin-like protease. Enzymatic cleavage converts domain 1 into domain 1' (residues 168-258) and opens up the EF/LF binding sites on domain 1'. Domain 2 (259-487) forms most of the protein-protein interfaces in oligomeric PA63 and features a Greek-key motif formed by strands 2β1-2β4. This Greek-key motif is positioned at the interface between domains 2 and 4 and is integral to pore formation. Domain 3 (488-595) is the smallest of the four domains and may play a role in oligomerization. Finally, domain 4 (596-735) forms limited contacts to the other three domains and is responsible for binding the cell-surface receptor (Figure 2).

Figure 3: X-ray structures of the PA63 prepore and heptameric α-hemolysin (α-HL). The two structures are rendered in ribbons with different colors representing the different subunits of each heptamer. A) The PA63 prepore structure has a height of ~85 Å and an outside diameter of ~160 Å. B) α-HL has a mushroom-like structure. Its length measures ~100 Å while its ring-like structure has an outside diameter of ~100 Å.

Unlike oligomerization, pore formation involves major conformational changes in the PA63 prepore. Experimental data indicate that pore formation involves the unfolding of the Greek-key motif formed by strands 2β1-2β4 (Figure 1). As evident by crystallographic analysis of the PA63 prepore structure at pH 6.0 and pH 7.5, which shows that the loops connecting strands 2β2-2β3 are more disordered under acidic conditions, the unraveling of this Greek-key motif is pH-dependent (12). Under acidic conditions of either the endosome and in vitro, the seven 2β2-2β3 loops presumably peel away from the ring-like structure of the PA63 prepore and insert into the cell membrane to produce a 14-stranded β-barrel (14). Accordingly, the proposed structure of the PA63 pore is mushroom-like, formed by a ring-shaped ?cap? domain and a cylinder-like ?stem? domain which consists of the 14-strand, membrane-inserted β-barrel.

Besides the unfolding of the Greek-key motif, conformational changes at other regions of the protein structure may be critical to pore formation and EF/LF translocation. Recently, high-throughput scanning mutagenesis of PA63 indicated that mutations at 36 residues reduced PA63?s ability to mediate toxicity by at least 100-fold (15). While a majority of these mutations, 22 of the 36, were in domain 2, the remaining 14 mutations were in the other three domains, suggesting that conformational changes outside of the pore-forming domain 2 are critical to pore formation and EF/LF translocation. Although it is clear that these conformational changes are necessary for pore formation and EF/LF translocation, there is little experimental evidence to indicate the nature of these conformational changes. At present, the three-dimensional structure of the PA63 pore has not been determined and the atomic details of the prepore-pore transition can not be described. In addition, there remains no clear correlation between pore formation and the conductance of EF and LF across the cell membrane. Experimental evidences indicate that the channel of the PA63 pore is cation-selective (16) and has a minimum channel diameter of ~12 Å (17). How this channel facilitates the release of EF and LF into the cell?s cytosol remains unanswered.

To study the conformational changes associated with the prepore-pore transition and to provide the basis for elucidating the molecular mechanism of translocation, the PA63 pore was modeled using the X-ray structures of monomeric PA (12) (Figure 2) and heptameric α-hemolysin (α-HL, PDB code 7AHL) (Figure 3B) (18) as templates. In this paper, the all-atom model of the PA63 pore is presented. The crystal structure of monomeric PA was the building block for the cap domain of the PA63 pore model, while the X-ray structure of α-HL which features a 14-stranded β-barrel was used to construct the stem domain of the PA63 pore model (Figure 3). After the modeled cap and stem domains were linked, molecular dynamics and distance constraints were used to hydropathically and energetically refine the PA63 pore model to be consistent with published experimental data.

Materials and Methods

General Methods

A Silicon Graphics Octane 2 workstation with Insight II (Accelrys, San Diego, CA) was used to build and visualize the models. The quality of the models was assessed using Procheck (19) and HINT (eduSoft, Richmond, VA). Procheck was used to determine the geometric quality of the models. Residues with inverted chirality or unfavorable phi and psi torsional angles were corrected by manual adjustments and tethered minimization. HINT was used to identify unfavorable hydrophobic-polar and base-base interactions in the model. HINT was employed using a exp(-1r) distance dependence for hydropathic constants on atom pair wise interactions. The distance function was set to 50% hydropathic and 50% steric, and the steric term is a Lennard-Jones 6-9 function which is compatible with the cff91 force field (20) used throughout this study. Lone pairs for basic atoms were vector focused by a factor of 10. Hydropathic values in the models that were greater than the highest values observed in the monomeric PA crystal structure were defined as hydropathic violations and were resolved using constrained molecular dynamics. After the backbone atoms of the structure were superimposed, the rms deviations values were calculated using MOLMOL (21). The figures were rendered using Molscript (22) and Insight II. The Discover 3.0 program (Accelrys, San Diego, CA) was employed for all simulations, which were performed on a cluster of high-performance Silicon Graphics Origins. All simulations were performed using the cff91 force field with the nonbonded interaction limited to within 13 Å with a distance-dependent dielectric constant.

Tethered Minimization

A tethered minimization protocol was used to optimize structures without significantly changing their conformations and involves two steps (23). Minimiza-tion involved up to 5000 steps of Fletcher-Powell optimization until the norm of the gradient was < 1.0 kcal/mol Ų. In the first step, the backbone atoms were fixed and a tethering force of 2000 kcal/mol Ų was applied to the side chain heavy atoms. After each cycle of minimization, tethering force was stepped off by a factor of 0.30 until all external force was removed. In the second step, a tethering force of 2000 kcal/mol Ų was applied to all heavy atoms and again stepped off by a factor 0.30.

Molecular Dynamics Using Distance Constraints

Molecular dynamics incorporating distance constraints were used to bring together and move away atoms or pseudoatoms. Molecular dynamics were performed for 500 fs in 500 steps with a time step of 1 fs at 300K. This was followed by minimization which involved up to 5000 steps of Fletcher-Powell optimization until the norm of the gradient was < 1.0 kcal/mol Ų. For each cycle, the upper or lower limits of the distance constraints were decreased or increased, typically by increments of 0.01 Å, since this provided uniform and smooth translations of the domains.

Distance Constraints

For all simulations, the force constant for the distance constraints was 100 kcal/mol and the restraint function was a flat-bottomed well. For distance constraints used to maintain secondary structure, the lower and upper distance limits were to set ± 0.5 Å of the distances between the hydrogen bonded heavy atoms measured from the crystal structures or the homology models. The structures of the seven calcium-binding sites in the PA63 pore were maintained during molecular dynamics simulations by 1323 distance constraints. For the distance constraints used to anneal regions of the protein structure, the upper distance limits were incrementally decreased by 0.01 Å until the target distance was reached. These distance constraints usually tie together pseudoatoms that represented the backbone atoms of specified peptide sequences. Distance constraints were also used to ?push? apart the two atoms involved in the unfavorable interactions as determined by HINT.

The atomic coordinates (code 1V36 [PDB]) have been deposited in the Protein Data Bank, Research Collaboratory for Structural Bioinformatics, Rutgers University, New Brunswick, NJ (http://www.rcsb.org/).

Results

While there is sufficient published biochemical and structural data to build a structurally reasonable PA63 pore model, there is no definitive molecular methodology established in the literature to simultaneously address the multiple challenges of modeling a protein of the large size and structural complexity of the PA63 pore. Since the PA63 pore consists of nearly 64,000 atoms and the prepore-pore transition involves major conformational changes, the computational cost of modeling the PA63 pore is enormous. Because of this, modeling the PA63 pore directly from the PA63 prepore is prohibitive. Simultaneously unfolding seven large flexible loops consisting of nearly 78 residues each to form a highly-ordered β-barrel with an experimentally-consistent topology is incredibly difficult, if not impossible given the current computational tools. Therefore, in order to achieve the molecular model of the PA63 pore in a reasonable amount of time, access to a cluster of powerful computers is necessary. To enhance computational efficiency, the PA63 pore model was obtained via the end-stage construction of four separate modeling projects.

Modeling the Stem Domain

Experimental data indicates that the stem domain in the PA63 pore is a 14-strand β-barrel with a similar channel diameter (17) and β-sheet topology (24, 25) as the 14-strand β-barrel from the X-ray structure of heptameric α-hemolysin (Figure 3B). Because of their structural similarity, the β-barrel in the X-ray structure of α-HL was selected as the template for the stem domain. Figure 4 shows the continuous sheet topology of the model stem domain. Since the stem domain is considerably longer than the β-barrel of α-HL, it was modeled by stacking three smaller β-barrels, which were derived from the β-barrel of α-HL. The three smaller β-barrels (stem-A (Ser294-Val332), stem-B (Thr279-Ser296 and Ser330-Ala347), and stem-C (Glu275-Arg287 and Leu338-Leu352)) feature overlapping amino acid sequences to allow for their proper alignment when connected. The modeled stem domain is consistent with experimentally-determined topology of the stem domain except that it does not feature β-bulges at Asn281-Gln285 and Leu340-Gly342.


Figure 4: Topology of the stem domain in the PA63 pore model. The topology of the modeled stem domain is consistent with experimental results except for Thr282, Ser284 and Ala341 (25). The aromatic belts formed by Phe313/Phe314 and by Phe324 demarcate the trans and cis face of the bilayer, respectively. The modeled stem domain has twice as many residues as the α-HL β-barrel and was constructed from three smaller β-barrels. The partitioning of modeled stem domain into stem-A, stem-B and stem-C is shown.

Modeling the Cap Domain

There are two different approaches to building the cap domain. The first approach is to model the cap domain from the PA63 prepore crystal structure. Converting the PA63 prepore crystal structure into the cap domain of the PA63 pore model involves major conformational changes within the protomers as well as at the protomer-protomer interfaces. Orchestrating such large protein movements in the PA63 prepore crystal structure is extremely challenging. Additionally, the established protein-protein interfaces in the PA63 prepore crystal structure may limit the range of the conformational changes. The second approach is to convert the monomeric PA crystal structure into a single subunit of the cap domain and subsequently, use the seven-fold rotational symmetry of the PA63 pore to assemble this single subunit into the cap domain. Because the latter approach is computationally more efficient and provides increased flexibility in modeling the protein-protein interfaces, the monomeric PA crystal structure was selected as the template.

The monomeric PA crystal structure (12) was converted into the cap domain of the PA63 pore in two stages. In the first stage, the monomeric PA crystal structure was converted into a single subunit of the cap domain. The calcium atoms and water molecules in the X-ray structure were removed in the monomeric PA crystal structure, and residues were deleted or appended to produce a PA63 subunit with an amino acid sequence of 168-274 and 353-735. In vivo, residues 1-167 are removed by proteolytic cleavage, while residues 275-352 unfold to form the stem domain. The monomeric PA63 has a large opening, measuring nearly 10 Å between domains 2 and 4 (Figure 5A). Molecular dynamics using distance constraints were used to close this interdomain opening. As a result, the closed PA63 subunit (Figure 5B) is a much more compact structure than the initial PA63 subunit.


Figure 5: Comparison of the 63 kDa C-terminal of the monomeric PA crystal structure (monomeric PA63) and the average protomer of the PA63 pore model. Both the crystallographic monomeric PA63 (A) and the average protomer model (B) have amino acid sequence of 168-274 and 353-735. Both structures are depicted as surface renderings, in which domain 1' is colored green, domain 2 cyan, domain 3 yellow, and domain 4 purple. Evident in the crystallographic monomeric PA63 subunit (A) is a large, ~10 Å opening between domains 2 and 4, while the average protomer model (B) offers a tightly packed molecular structure.

In the second stage, seven copies of this closed PA63 subunit were arrayed into a ring-shaped heptamer. While in the same plane, the seven PA63 subunits were translated towards the ring center until van der Waals violations between the surfaces of <0.25 Å were registered. Once the subunits were aligned, van der Waals violations were resolved by manually adjusting side chain torsion angles until all residues at the interfaces were positioned in a hydropathically favorable manner.

Joining the Cap and Stem Domains

An initial PA63 pore model was produced by linking the model cap and stem domains and refining the interface between the two domains. Unfortunately, there is little biochemical or structural data to suggest the relative orientation of the cap and stem domains to one another. If the PA63 pore was similar to α-HL, the stem domain would tuck into the cap domain. Alternatively, the stem domain may connect to the base of the cap domain, which is consistent with the connective strands of the cap domain being located at its base. Because the latter involves the least conformational change in the PA63 prepore, the stem domain was linked to the base of the cap domain.

Refinement of the cap-stem domain interface involves translating the stem domain towards the cap domain until the two domains were tightly packed. The residues in the cap domain were fixed and a methane molecule (serving as a dummy atom and used as a constraining system) was placed at the center of its vestibule. The methane molecule does not form interactions with the protein structure and therefore, would not affect the results of the molecular dynamics simulations. At the end of this procedure, the methane molecule is removed. The position of the methane was fixed and a distance constraint was created between the methyl carbon atom and a pseudoatom representing the center of the stem domain. Decreasing the upper limit of this distance constraint during molecular dynamics translated the stem domain towards the cap domain by ~2.5 Å. This procedure required >100 hours of CPU time. Iterations of tethered minimization and manual changes to the orientation of the stem domain were used to refine the interface between the two domains. Refining the interface between the cap and stem domains resulted in conformational changes at the interface between domains 2 and 4. This necessitated additional refinement of the domain 2-4 interface.

Modeling the PA63 Pore

The initial PA63 pore model was refined to fit crucial mutational data. It was shown that mutations of amino acids 510-518, particularly Asp512, in PA63 disrupted oligomerization, while mutations of Lys397, Asp425 and Phe427 in PA63 blocked translocation (26, 27). An examination of the X-ray structures of monomeric PA and the PA63 prepore shows that the side chains of Lys397, Asp425 and Asp512 are solvent-exposed and not hydrogen bonded. In building a model of the PA63 prepore, an opportunity was afforded to define potentially important intersubunit interactions for these residues. During molecular dynamics simulations, distance constraints were used to form the Arg252/Asp512, Lys395/Asp425 and Lys397/Glu398 salt-bridges at the protomer-protomer interfaces. The side chain atoms of these residues are separated by as much as 10 Å in the initial PA63 pore model. Therefore, in order to form these salt-bridges, it is necessary to shift the protomers of the cap domain closer to one another. As a consequence, in parallel with the distance constraints used to form the three salt-bridges, distance constraints were used to pull the subunits of the cap domain closer to each other. Pseudoatoms representing residues 254-267 and 362-371 were created for each subunit and linked to adjacent pseudoatoms by distance constraints. During the molecular dynamics simulations, decreasing the upper limits of these 7 distance constraints translated the subunits of the cap domain in a rigid-body fashion. In the end, the seven pseudoatoms were brought from an initial separation of ~30.5 Å to a final distance of ~27.5 Å. Periodically, during molecular dynamics simulations, the hydropathic quality of the seven interfaces was checked using HINT and unfavorable interactions were resolved by ?push? distance constraints to separate the two atoms involved. Because the PA63 pore model consists of a large number of atoms (nearly 64,000 atoms) and features extensive protein-protein interfaces (>2200 Ų of buried surface area per monomer), this procedure required over one month of CPU time on a cluster of high performance SGI Origins. The large amount of computational time required to build the PA63 pore model reiterates the necessity of access to a cluster of powerful computers in order to achieve the model in a reasonable timeframe, particularly when one considers that many of the structural features of the PA63 pore can not be established a priori and have to be examined in trial simulations. Once the protomer-protomer interfaces were optimized, the re-incorporation of the 14 calcium atoms completed the PA63 pore model.

Overall Structure of the PA63 Pore Model

As shown in Figure 6A, the final PA63 pore model has a mushroom-like structure, measuring ~180 Å in length. The cap and stem domains of the model are distinct. The cap domain is ~80 Å long and ~120 Å wide, while the stem domain is ~100 Å long and ~28 Å wide, as measured Cα to Cα. Molecular dynamics studies of the PA63 pore model indicate that the channel has a minimum diameter of ~12 Å - ~17 Å, depending on the side chain conformations and is consistent with solute exclusion experiments, showing a minimum channel diameter of ~12 Å (17).


Figure 6: Comparison of the PA63 pore model and the crystal structure of α-HL. Both structures are shown in ribbon; cyan ribbon representing sheet structure, red ribbon helical structure and green ribbon loops. The two structures have a high sheet content. (A) The PA63 pore model has a mushroom-like structure. Its cap domain is ~80 Å long and ~120 Å wide, while the stem domain is ~100 Å long and ~28 Å wide, as measured Cα to Cα. (B) The crystal structure of α-HL is also mushroom-like. However, it measures only ~100 Å long with an outside ring diameter of ~100 Å and a β-barrel only ~50 Å long. The β-barrel of α-HL is partially buried inside the ring-like structure.

Consistent with experimental data, the transmembrane segment of the stem domain of the PA63 pore model is amphipathic (25). While the polar residues line the channel interior, the nonpolar residues line the channel exterior, enabling the formation of favorable interactions with the alipathic tails of the bilayer (Figure 7A). Beyond the transmembrane segment, the exterior of the stem domain is both basic and acidic. On the channel exterior, the middle of the stem domain is strongly basic due to the presence of Arg287, Lys291, Arg297, His299, His336 and Arg344, while the top of the stem domain is acidic due the presence of Asp283 and Glu348 (Figure 7A).

Further, consistent with the cation selectivity of the PA63 pore (16), channel interior in the PA63 pore model is strongly acidic (Figure 7B). There are patches of negatively-charged residues at the top of the vestibule in the modeled cap domain (Glu465 and Asp472), near the center of the channel (Asp276, Glu343, Asp425 and Asp426), and at the transmembrane segment (Glu302, Glu308 and Asp315). Table I shows the approximate relative distance of the Cα of each of these negatively-charged residues from the trans face of the bilayer.

Figure 7: Surface renderings of the PA63 pore model. For A and B, the surfaces are colored according to the negative (red) and positive (blue) electrostatic surface potential. (A) The exterior surface of the PA63 pore model has mostly positive electrostatic potential. While the transmembrane segment is nonpolar, the portion of the stem domain that extends beyond the membrane features mostly positive electrostatic potential. (B) A slice through the center of the model shows the channel interior. To provide perspective to the channel, the heavy atoms that line the channel lumen are rendered as green sticks. Consistent with the cation-selectivity of the PA63 pore, the channel interior of the model is strongly negative. (C) A protomer in the PA63 pore model is rendered as a solid purple surface while other six protomers are shown as transparent green surfaces. The protomer molecule consists of two structural motifs. The first is a flat rectangular structure, measuring ~80 Å × ~45 Å × ~30 Å. The second is an extended hairpin which has a 180° twist and 45° tilt.

Structural Features of the Protomers of the PA63 Pore

The protomers in the PA63 pore model are flat rectangles with dimensions of ~80 Å × ~45 Å × ~30 Å. The extended hairpins of the protomers in the PA63 pore model have a 180° twist and a 45° tilt (Figure 7C). Although the protomers in the PA63 pore model have similar overall structures, they are not identical and have a backbone rmsd of 2.4 Å between them. Ideally, the protomers in the PA63 pore should be identical. However, that may not necessarily have to be the case for PA63 pore in solution or on the cell surface, where the weak association between the domains, particularly domain 4, and the large size of the protein complex, may lead to similar but not identical protomers. Consistent with the X-ray structure of monomeric PA (12), the PA63 pore model has a high β-sheet content. In general, domain 1' of the PA63 pore model consists of three small helices, domain 2 has a large β-hairpin, a β-sheet core and two helices, domain 3 has several β-sheets and 4 helices, and domain 4 consists of a β-sandwich and two small helices.

Discussion

Comparison of Model to the X-ray Structure of Hemolysin

While the PA63 pore model and the α-HL crystal structure both have a high content of β-sheet structure, the PA63 pore model (Figure 6A) dwarfs the α-HL crystal structure (Figure 6B). The length of the PA63 pore model and the α-HL crystal structure are ~180 Å and ~100 Å, respectively. Two factors contribute to a majority of this 80 Å length difference. The first factor is that the stem domain of the PA63 pore model is twice the length of the β-barrel in the α-HL crystal structure, which explains 50 Å of the 80 Å length difference. The second factor is the relative orientation of the stem and the cap domains in the α-HL crystal structure and the PA63 pore model. While the stem domain is joined to base of the cap domain in the PA63 pore model, the β-barrel of α-HL is tucked into its ring-shaped structure. This accounts for a ~30 Å length difference.

Comparison of Model to X-ray Structure of the PA63 Prepore

The outside diameters of the cap domain in the PA63 pore model and the ring-shaped structure of the PA63 prepore crystal structure are 120 Å and 160 Å, respectively. This ~40 Å difference mostly reflect the translation of domain 4 during the prepore-pore transition. The Greek-key motif in domain 2 forms extensive contacts to the protein surface of domain 4. During pore formation, the unfolding of the Greek-key motif necessitates the refolding of domain 4 onto a new surface on domain 2. This shifts domain 4 towards the channel center. As the PA63 pore model illustrates, domain 4 can not only translate in the direction of the channel center but also along a direction normal to the outer circumference of the cap domain. Domain 4 is weakly associated with the other three domains. As a measure of this displacement, the average displacement of the Cα atom of Asp650, which is positioned at the outer edge of domain 4 in the PA63 prepore X-ray structure and the PA63 pore model, is ~30 Å. This value is higher than the average displacement of 20 Å expected for the direct path of domain 4 towards the channel center.

In addition, a slightly reduced channel diameter is observed in the PA63 pore model relative to the PA63 prepore X-ray structure. The channel in the PA63 prepore X-ray structure has an average diameter of ~35 Å and narrows to ~20 Å in some regions. The minimum channel diameter within the cap domain of the PA63 pore model is slightly narrower at 17 Å and occurs near loop residues 473-474. This 3 Å difference is a consequence of the translation of the subunits of the cap domain closer to one another to form the intersubunit salt-bridge interactions, which also moves each subunit closer to the channel center. Therefore, it is not surprising that all the atoms in the PA63 pore model occupy novel Cartesian space relative to the PA63 prepore crystal structure.

The PA63 pore model features Arg252/Asp512, Lys395/Asp425 and Lys397/Glu398 salt-bridge interactions that are not evident in the PA63 prepore X-ray structures. Mutations of Lys397 and Asp425 were shown to block PA?s ability to form a pore and translocate but had no effect on PA?s ability to bind to the AT receptor, undergo enzymatic cleavage and oligomerize (26, 27). This suggests that Lys397 and Asp425 facilitate conformational changes in the PA63 prepore that are necessary for pore formation. The formation of the Lys397/Glu398 and Asp425/Lys395 salt-bridges in the PA63 pore model predictably alters the conformations of the 2β7-2β8 and 2β10-2β11 loops relative to their positions in the monomeric PA and the PA63 prepore X-ray structures. These conformational changes may be crucial to pore formation, since they contribute to the tight packing at the interface of the cap-stem domains in the model. Further experimental work will be needed to characterize the precise intermolecular interactions formed by Lys397 and Asp425 side chains.

The EF/LF binding sites in the PA63 prepore crystal structure and the pore model are largely hydrophobic and strongly basic. Mutagenesis studies mapped out the EF/LF binding sites in oligomeric PA63 (28). Arg178, Lys197, Arg200, Pro205, Ile207, Ile210 and Lys214 form the core of the EF/LF binding site, while Glu190, Asp195, Asn198, Lys199, Phe202, His211, Lys213, Leu216 and Lys218 outline its periphery. Although the seven EF/LF binding sites in the pore model are not structurally identical, they share common features to one another and to the EF/LF binding site in the prepore crystal structure. Since the crystal structures of EF and LF have been determined and the PA binding site in LF has been mapped using mutational analysis (29), the PA63 pore model may prove useful in the development of peptide-based or small molecule inhibitors against EF/LF binding.

Comparison of the Model to the X-ray Structure of Monomeric PA

An average structure was calculated for the seven protomers of the PA63 pore model and compared to the 63 kDa C-terminal segment of monomeric PA X-ray structure (monomeric PA63). Since strands 2β1-2β4 (residues 275-352) experience major conformational change during the prepore-pore transition, amino acids 275-352 were not considered in this comparison. Figures 5A and 5B shows a surface representation of the monomeric PA63 and the average modeled protomer in the absence of residues 275-352, respectively. While the average modeled protomer is a closed and compact structure, a large interdomain groove is apparent in the monomeric PA63 structure. From the superimposition of their backbone atoms, the two structures display a backbone rmsd of 2.4 Å for residues 175-735, excluding residues 275-352. Domains 1' and 3 displayed lower backbone rmsd values between the two structures of 2.0 Å and 1.1 Å, respectively. Reflecting their major conformational changes during pore formation, domains 2 and 4 have higher rmsd values of 2.8 Å and 3.2 Å, respectively.

Binding to the Cell Surface

Given that the typical membrane has a height of ~50 Å, the stem domain of the PA63 pore model extends ~50 Å above the membrane surface (Figure 8). Since it is unlikely that nature would assemble a β-barrel longer than would be necessary for it to function, the additional length of the stem domain may be used to span the length of the AT receptor molecule. Several published experimental results point to this type of interaction between the PA63 pore and the AT receptor molecules. In the endosome, the PA63 prepore and the PA63 pore are bound to AT receptor molecules. Published studies indicate that PA binds to the AT receptor through domain 4 (30). Recently, alanine-scanning mutations of PA revealed that a small loop (amino acids 679-693) in domain 4 is critical to PA?s binding to the AT receptor (31). In the PA63 pore model, this small loop is located at the base of the cap domain and is mostly solvent exposed. Further, Tyr681 and Asn683 which had the largest effect on cell binding and toxicity are located near the interface between domains 2 and 4. These results suggest that the AT receptor molecules are positioned between the cap domain and the membrane surface and are closely packed against the stem domain, as illustrated in Figure 8, in which gray and white space-filling renderings represent generic AT receptor molecules. Two cellular proteins, tumor endothelium marker 8 (TEM8) (32, 33) and capillary morphogenesis gene 2 (CMG2) (34), were shown to bind PA via their extracellular von Willibrand factor A (VWA) domains and therefore, may function as AT receptors.


Figure 8: Proposed PA63 pore-AT receptor complex on the membrane surface. The PA63 pore model is rendered as a green solid surface. The lipid bilayer is shown as purple and cyan sticks. Generic protein structures, representing the AT receptors (structures undetermined), are drawn as white and grey surfaces. To visualize the full length of the stem domain, a portion of the bilayer and two of the AT receptor molecules were removed. Given that the average height of the cell membrane is ~50 Å, the stem domain of the model extends ~50 Å beyond the membrane surface. It is proposed that the AT receptor occupies the span between the cap domain and the membrane surface.

Recently, Santelli et al. reported the X-ray structure of monomeric PA complexed with the CMG2 I domain (PDB code 1T6B) (35). PA is a long and flat molecule and CMG2 I domain binds to domains 2 and 4 of PA at its base. Notably, a pocket in CMG2 I domain formed by Tyr119, His121 and Tyr158 is occupied by the β3-β4 loop in domain 2 of PA. During the prepore-pore transition, strands β3 and β4 peel away from the PA63 prepore to form the 14-stranded, membrane-inserted β-barrel. Based on the X-ray structures of the PA63 prepore and the PA-CMG2 I domain complex, Santelli et al. presented a hypothetical model of the receptor-bound, membrane-inserted PA63 pore. Their model consists of three components: i) a ?cap domain? structure derived directly from the PA63 prepore crystal structure, ii) seven CMG2 I domains arrayed at the base of the ?cap domain? structure in a similar fashion to the PA-CMG2 I domain complex, and iii) a 14-stranded β-barrel was positioned below the channel of the ?cap domain? structure. There is a large gap between the membrane surface and the CMG2 I domain, which Santelli et al. attributes to the second domain of CMG2, the three-dimensional structure of which has not been determined. Using the β-barrel as a measure, Santelli et al. suggests a height of ~40 Å for the second domain of CMG2.

While the β-barrel presented by Santelli et al. is strikingly similar to the one presented in this paper, both measuring ~100 Å in height and featuring strands with a 180° twist and 45° tilt, there are significant differences between the two models. The first difference involves the interface between the β-barrel and the cap domain. In the model offered by Santelli et al., the β-barrel and the ?cap domain? structure are not covalently linked and there appears to be a large separation between the two domains. At first glance, it would seem that translating the β-barrel in the Santelli et al. model either 10 Å towards or 10 Å away from the ?cap domain? structure would produce equally valid models. In this sense, the height of the Santelli et al. pore model and as a corollary, the height of the second domain of CMG2 suggested by Santelli et al. is subjective. In contrast, the PA63 pore model presented in this paper features an unbroken structure between the cap and stem domains and has a well-defined height of ~180 Å. Given that the membrane has a height of 35 Å as in the Santelli et al. model, the second domain of CMG2 bound to the PA63 pore model presented in this paper has a height of ~30 Å, which may be a more reliable measure of the CMG2 dimension. A second major difference between the two models is the outside diameters of their ?cap domain? structures. Pore formation produces a large void space between domains 2 and 4 in each protomer and additionally, between domain 2 and the receptor CMG2. In the Santelli et al. model, these large gaps in the protein structure are not addressed. The ?cap domain? structure in the Santelli et al. model has essentially the same outside diameter and presumably the same general conformation as the PA63 prepore crystal structure, while the PA63 pore model presented in this paper forms a tight protein structure, as reflected in the hydropathically and energetically reasonable interface between the stem and cap domains as well as between domains 2 and 4.

Translocation of the Protein Ligands

Since cell entry by proteins is a critical biological process, elucidating the molecular mechanism of EF/LF translocation is of great interest. While there is compelling experimental evidence to indicate that EF/LF translocation depends upon pore formation, there is little evidence to suggest how the channel of the PA63 pore facilitate the release of EF/LF into the cell?s cytosol.

Based on results from solute exclusion experiments (17), the PA63 pore model was constructed to have a minimum channel diameter of ~12-17 Å, depending on side chain conformations of residues lining the channel interior. However, the minimum channel diameter of the PA63 pore model is significantly smaller than would be expected if EF and LF were to translocate through the channel. Based on their X-ray structures, native EF (1) and LF (2) would require a minimum channel diameter > 45 Å. Therefore, passing EF and LF through the channel of the PA63 pore requires major conformational changes in the X-ray structures of EF and LF and in the PA63 pore model and could include displacements of the AT receptors and phospholipids of the bilayer. Experimental evidence indicates that in vivo, EF and LF partially unfold upon translocation (36, 37), allowing EF and LF to translocate through a smaller pore than would be predicted based on their crystal structures. Whether partially-unfolded EF and LF can migrate through the channel of the PA63 pore is debatable. In addition, since there exists SDS-sensitive and SDS-resistance forms of the PA63 prepore at low pH (7), multiple conformations of the PA63 pore may also exist, an indication that different oligomeric states of the PA63 pore could have larger channel diameters.

Alternatively, the biological role of the PA63 pore may not be to directly transport EF and LF across the membrane but rather to facilitate their release by destabilizing the membrane. Peptides corresponding to the N-terminal segment of influenza virus hemagglutinin HA-2 were shown to be able to fuse with membranes, disrupt the structural integrity of the lipid bilayers and cause the leakage of endosomal and liposomal contents (38). In a similar fashion, the PA63 pore could be membrane-disruptive. The conductance of water molecules and counterions through the channel of the PA63 pore changes the dielectric constant of the local environment and could induce conformational changes in the lipid bilayer. In addition, EF and LF may play active roles in their escape from the endosome. Experimental evidence indicates that at low pH, both EF and LF inserts into the membrane (36, 37) and thus, destabilize the lipids of the bilayer system. In combination with the partially-unfolding of EF and LF, the destabilization of the membrane by the PA63 pore and membrane-insertion of EF and LF may allow EF and LF to escape the endosome.

Acknowledgements

The author thanks the Advanced Biomedical Computing Center at the National Cancer Institute for technical assistance and computing time on the SGI cluster. This work was supported by the U.S. Army Medical Research and Material Command Research Plan #02-4-3U-057 and IAA #Y3-CM-1005-05 (MRMC and NCI).

References and Footnotes

1. C. L. Drum, S. Z. Yan, J. Bard, Y. Q. Shen, D. Lu, S. Soelaiman, Z. Grabarek, A. Bohm, and W. J. Tang. Nature 415, 396-402 (2002).
   
2. A. D. Pannifer, T. Y. Wong, R. Schwarzenbacher, M. Renatus, C. Petosa, J. Bienkowska, D. B. Lacy, R. J. Collier, S. Park, S. H. Leppla, P. Hanna, and R. C. Liddington. Nature 414, 229-233 (2001).
   
3. S. H. Leppla. Proc. Natl. Acad. Sci. USA 79, 3162-3166 (1982).
   
4. G. Vitale, R. Pellizzari, C. Recchi, G. Napolitani, M. Mock, and C. Montecucco. Biochem. Biophys. Res. Commun. 248, 706-711 (1998).
   
5. K. A. Bradley, and J. A. Young. Biochem Pharmacol 65, 309-314 (2003).
   
6. K. R. Klimpel, S. S. Molloy, G. Thomas, and S. H. Leppla. Proc. Natl. Acad. Sci. USA 89, 10277-10281 (1992).
   
7. J. C. Milne, D. Furlong, P. C. Hanna, J. S. Wall, and R. J. Collier. J. Biol. Chem. 269, 20607-20612 (1994).
   
8. J. Mogridge, K. Cunningham, D. B. Lacy, M. Mourez, and R. J. Collier. Proc. Natl. Acad. Sci. USA 99, 7045-7048 (2002).
   
9. J. L. Elliott, J. Mogridge, and R. J. Collier. Biochemistry 39, 6706-6713 (2000).
   
10. V. M. Gordon, S. H. Leppla, and E. L. Hewlett. Infect. Immun. 56, 1066-1069 (1988).
    
11. T. C. Dixon, M. Meselson, J. Guillemin, and P. C. Hanna. N. Engl. J. Med. 341, 815-826 (1999).
    
12. C. Petosa, R. J. Collier, K. R. Klimpel, S. H. Leppla, and R. C. Liddington. Nature 385, 833-838 (1997).
    
13. S. Gao-Sheridan, S. Zhang, and R. J. Collier. Biochem. Biophys. Res. Commun. 300, 61-64 (2003).
    
14. C. J. Miller, J. L. Elliott, and R. J. Collier. Biochemistry 38, 10432-10441 (1999).
    
15. M. Mourez, M. Yan, D. B. Lacy, L. Dillon, L. Bentsen, A. Marpoe, C. Maurin, E. Hotze, D. Wigelsworth, R. A. Pimental, J. D. Ballard, R. J. Collier, and R. K. Tweten. Proc. Natl. Acad. Sci. USA 100, 13803-13808 (2003).
    
16. Blaustein, R. O., Koehler, T. M., Collier, R. J., and Finkelstein, A. Proc. Natl. Acad. Sci. USA 86, 2209-2213 (1989).
    
17. A. Finkelstein. Toxicology 87, 29-41 (1994).
    
18. L. Song, M. R. Hobaugh, C. Shustak, S. Cheley, H. Bayley, and J. E. Gouaux. Science 274, 1859-1866 (1996).
    
19. R. A. Laskowski, M. W. MacArthur, D. S. Moss, and J. M. Thornton. J. Appl. Cryst. 26, 283-291 (1993).
    
20. J. R. Maple, T. S. Thacher, U. Dinur, and A. T. Hagler. Chem. Design Automat. News 5, 5-10 (1990).
    
21. R. Koradi, M. Billeter, and K. Wuthrich. J. Mol. Graphics 14, 52-55 (1996).
    
22. P. J. Kraulis. Journal of Applied Crystallography 24, 946-950 (1991).
    
23. J. C. Burnett, J. J. Schmidt, R. G. Stafford, R. G. Panchal, T. L. Nguyen, A. R. Hermone, J. L. Vennerstrom, C. F. McGrath, D. J. Lane, E. A. Sausville, D. W. Zaharevitz, R. Gussio, and S. Bavari. Biochem. Biophys. Res. Commun. 310, 84-93 (2003).
    
24. E. L. Benson, P. D. Huynh, A. Finkelstein, and R. J. Collier. Biochemistry 37, 3941-3948 (1998).
    
25. S. Nassi, R. J. Collier, and A. Finkelstein. Biochemistry 41, 1445-1450 (2002).
    
26. B. R. Sellman, M. Mourez, and R. J. Collier. Science 292, 695-697 (2001).
    
27. B. R. Sellman, S. Nassi, and R. J. Collier. J Biol Chem 276, 8371-8376 (2001).
    
28. K. Cunningham, D. B. Lacy, J. Mogridge, and R. J. Collier. Proc. Natl. Acad. Sci. USA 99, 7049-7053 (2002).
    
29. D. B. Lacy, M. Mourez, A. Fouassier, and R. J. Collier. J. Biol. Chem. 277, 3006-3010 (2002).
    
30. Y. Singh, K. R. Klimpel, C. P. Quinn, V. K. Chaudhary, and S. H. Leppla. J. Biol. Chem. 266, 15493-15497 (1991).
    
31. M. J. Rosovitz, P. Schuck, M. Varughese, A. P. Chopra, V. Mehra, Y. Singh, L. M. McGinnis, and S. H. Leppla. J. Biol. Chem. 278, 30936-30944 (2003).
    
32. K. A. Bradley, J. Mogridge, M. Mourez, R. J. Collier, and J. A. Young. Nature 414, 225-229 (2001).
    
33. S. Liu, and S. H. Leppla. J. Biol. Chem. 278, 5227-5234 (2003).
    
34. H. M. Scobie, G. J. Rainey, K. A. Bradley, and J. A. Young. Proc. Natl. Acad. Sci. USA 100, 5170-5174 (2003).
    
35. E. Santelli, L. A. Bankston, S. H. Leppla, and R. C. Liddington. Nature (in press).
    
36. X. M. Wang, M. Mock, J. M. Ruysschaert, and V. Cabiaux. Biochemistry 35, 14939-14946 (1996).
    
37. J. Wesche, J. L. Elliott, P. O. Falnes, S. Olsnes, and R. J. Collier. Biochemistry 37, 15737-15746 (1998).
    
38. E. Wagner, C. Plank, K. Zatloukal, M. Cotton, and M. L. Birnstiel. Proc. Natl. Acad. Sci. USA 89, 7934-7938 (1992).