A model of anthrax toxin lethal factor bound to protective antigen

Materials and Methods

Modeling the Structure of the LF_N-PA Dimer Complex. To reduce computational time, the PA₆₃-PA₆₃ dimer was truncated to include residues 177-260 (the subdomain that contains the EF/LF-binding site) and 458-595 (a subdomain that mediates oligomerization). The manually docked models were used as a starting point for sampling the surrounding free energy landscape by using many independent Monte Carlo minimization trajectories according to a Rosetta-Dock protocol described in refs. 17 and 18. Briefly, the rigid degrees of freedom of the starting model are randomly perturbed, and the perturbed model is subjected first to low-resolution refinement and then to high-resolution refinement. In the high-resolution refinement step, the side-chain and backbone degrees of freedom are optimized simultaneously in the context of a detailed all-atom energy function dominated by short-range hydrogen bonding, van der Waals interactions, and desolvation. The rms deviation (rmsd) values were calculated over the LF_N molecule after superposition of the PA dimer with the starting model.

Preparation of PA, LF_N, and Mutants. Oligonucleotides were synthesized by Integrated DNA Technologies (Coralville, IA). Mutations in PA and LF_N were made in the pET22b-PA (1-735) (19) and pET15b-LF_N (1-263) (15) constructs, respectively, by using the QuikChange method (Stratagene) of site-directed mutagenesis. ³⁵S-labeled LF_N proteins were produced by in vitro transcription/translation by using a TNT coupled reticulocyte lysate system (Promega). Otherwise, PA and LF_N were expressed and purified from Escherichia coli as described in refs. 20 and 21. PA was activated by using a trypsin-to-PA ratio of 1:1,000 (wt/wt). The mixture was incubated at room temperature for 30 min and quenched with a 10 M excess of soybean trypsin inhibitor.

Disulfide Crosslinking. Activated PA was incubated with an equimolar amount of LF_N in the presence of excess DTT for 1 h on ice to allow for binding. The PA-LF_N mixture was applied to a Sephadex G-25 column (Amersham Pharmacia) to remove DTT and exchange the complex into a buffer containing 50 mM NaCl and 20 mM Tris·Cl (pH 8.0). Samples were allowed to oxidize for 10 min before adding N-ethylmaleimide to quench any remaining free cysteines. The proteins were precipitated with trichloroacetic acid, resuspended in SDS buffer, and then analyzed by SDS/PAGE using a 7.5% acrylamide/SDS gel. Gels were stained with Coomassie blue.

Binding Assay for Charge-Reversal Mutations. PA-mediated binding of ³⁵S-labeled LF_N was performed on CHO-K1 cells (CCL-61, American Type Culture Collection) as described in ref. 22. Supplies for cell culture media were from Invitrogen. Cells were grown in Ham's F-12 medium supplemented with 10% calf serum, 500 units/ml penicillin G, and 500 units/ml streptomycin sulfate and were maintained as monolayers in a humidified atmosphere of 5% CO₂. The cells (2 × 10⁵ cells per well) were incubated on ice with 2.4 × 10^-8 M trypsin-nicked PA for 1 h. The cells were washed with PBS and incubated on ice with ³⁵S-labeled LF_N for 1 h. Cells were washed three times with PBS and treated with lysis buffer, and the radioactive content was determined by scintillation counting.

Results

Creating a Model of LF_N Bound to a PA Dimer. LF_N was manually docked on a truncated PA dimer in two orientations that differed by ≈180° (Fig. 1c). In the first orientation (Fig. 1c Left), the PA-binding site on LF_N was aligned to subsite I of PA to maximize both the charge complementarity between the negatively and positively charged amino acids and the overlap of hydrophobic residues. In the second orientation (Fig. 1c Right), the negatively charged residues of the LF_N PA-binding site were aligned to PA subsite II. Both models were then subjected to refinement by using the Rosetta-Dock protocol. To sample the free-energy landscape in the vicinity of the manually docked models, we carried out 3,000 independent refinement trajectories starting from random perturbations of the starting models (see Materials and Methods). Whereas the refinement of models from the second orientation did not result in an energetic minimum (data not shown), the energy landscape produced by the refinement from the first orientation (Fig. 2) contained a pronounced energy minimum. Alignment of the PA dimers indicates that, in these low-energy models, LF_N differs from the starting model by ≈20 Å rmsd. Dramatic energy funnels, such as this one, were found post facto to be strong indicators of the correctness of a prediction in the double-blind Critical Assessment of Predicted Interactions (CAPRI) protein-protein docking experiment in which a number of the predictions made by using the Rosetta-Dock protocol turned out to have close to atomic-level accuracy (18, 23). We chose the lowest energy model generated, the lowest energy point in Fig. 2, as our prediction. It should be emphasized that the model was selected based on energy criteria alone, and that no experimental information was used other than that implicit in the manually docked starting structures.

Identification of a Disulfide Crosslink Between LF_N Y108C and PA N209C. Because neither LF nor PA contains cysteine, introducing cysteines by site-directed mutagenesis represents a straightforward method to test for formation of disulfide crosslinks with binding. First, heptameric, wild-type PA₆₃ or PA₆₃ N209C was incubated with wild-type LF_N or one of seven LF_N mutants: Y108C, K110C, Y118C, Q132C, S134C, D136C, and Q228C. Binding was allowed to occur in the presence of excess DTT to prevent the formation of nonspecific disulfides. The DTT was then removed, and the samples were allowed to oxidize briefly before being treated with N-ethylmaleimide. The formation of a disulfide crosslink between LF_N Y108C and PA₆₃ N209C was visible as a slow-mobility band on an SDS gel that could be disrupted in the presence of DTT (Fig. 3). This band was shown by Western blotting with anti-PA and anti-LF_N antibodies to contain both PA and LF_N (data not shown). The crosslink formed selectively, because LF_N K110C, Y118C, Q132C, S134C, D136D, and Q228C did not form crosslinks when incubated with PA₆₃ N209C (data not shown). Likewise, LF_N Y108C did not crosslink with PA₆₃ S186C (data not shown).

Identification of Complementary Charge-Reversal Mutations That Can Rescue Binding. The LF_N- and PA-binding surfaces are notable in that they contain a large number of negatively and positively charged residues, respectively. The idea that electrostatics might play an important role in the binding interaction suggested that it might be possible to exchange a specific negative residue in LF_N and a positive residue in PA in a way that would not compromise the binding interaction.

Reversing the charge in six of the positively charged residues of PA (R178D, K197D, R200E, K213E, K214E, and K218E) inhibited binding of wild-type LF_N, (binding was observed at 0-63% of wild-type levels, Fig. 4). Similarly, substituting a lysine for LF_N D187 reduced PA-binding to 3% of wild-type levels. Pairing the LF_N D187K mutant with each of the six PA charge-reversal mutants restored binding to 120% of wild-type levels in the case of the LF_N D187K-PA K213E pair but had no significant effect on the other five PA mutants (Fig. 4). After identifying the LF_N D187K-PA K213E pair, LF_N D187K also was tested against PA K213D and shown to restore binding to 130% of wild-type levels (Fig. 4).

A similar experiment was conducted in which a LF_N E142K mutant was paired against the seven PA charge-reversal mutants. Although the LF_N E142K mutant on its own was not defective in binding wild-type PA (it binds at 116% of the wild-type level), the mutation rescued the binding defect in K218E, such that binding for the pair was at 128% that of wild-type (Fig. 4). E142K failed to complement the binding defects in the other six PA mutants tested.

Discussion

The goal for this study was to generate a model for how LF_N binds to and is oriented on PA. Simple docking was initially confounded by the large discrepancy in size of the two putative binding sites (Fig. 1 a and b). To circumvent this problem, we adopted three approaches to further explore this molecular interface.

The first was to model the complex in two extremely different orientations and then select a model based solely on their energies. The flat, rectangular shape of the surfaces containing the LF_N- and PA-binding sites, along with the clusters of negatively and positively charged residues within these binding sites, suggested that, to a first approximation, LF_N would bind the PA dimer in one of two orientations (Fig. 1c). We docked LF_N to a PA dimer in orientations that differed by ≈180° and submitted both models to energetic minimization using a Rosetta-Dock protocol. We observed a dramatic energy funnel for the model in which the PA-binding site of LF_N was docked to PA subsite I (Fig. 2).

The second approach was to identify one or more points where a disulfide crosslink could be effected between LF_N and PA. We chose residues in LF_N and PA that were located near the binding sites but where alanine substitution had been shown not to affect binding (8, 15). We introduced cysteine mutations in these positions and observed a specific disulfide crosslink between LF_N Y108C and PA N209C (Fig. 3). In the energetically favorable model, the distance between LF_N Y108 and PA N209 is consistent with a disulfide being able to form if cysteines were substituted at these positions (Fig. 5a).

The final approach was designed to identify pairs of charged residues in the two proteins that could be reversed without inhibiting the interaction; this approach also was expected to yield insights into the importance of electrostatics in the LF_N-PA interaction. We found that reversing the charge of certain residues in LF_N or PA could inhibit binding to the wild-type partner protein and reasoned that pairing these charge-reversal mutants so that they could maintain an electrostatic interaction might rescue binding for these otherwise defective mutants. We identified two such pairs of charge-reversal mutants: LF_N D187K-PA K213D/K213E and LF_N E142K-PA K218E, suggesting that the LF_N D187-PA K213 and LF_N E142-PA K218 residues are close in the LF_N-PA complex (Fig. 4). In the low-energy model, the charge pairs are located on either side of the disulfide crosslink (Fig. 5a). Although the Rosetta-Dock protocol does not emphasize electrostatics, the model suggests that, with modest rearrangement of side chain rotamers, these pairs of residues should be close enough to form favorable electrostatic interactions.

Alignment of the PA molecules from the energetically favorable model and its starting model reveals that LF_N has shifted by a rmsd of 20 Å. This large departure from the starting model is because of a twist in LF_N that unexpectedly minimizes the interaction of LF_N with PA subsite II. As modeled, LF_N contacts the K197 residue of the PA₆₃-PA₆₃ interface but does not make any direct contacts with R178 and R200 (the two other residues of the second subsite suggested by the mutagenesis work done in the PA dimer; ref. 8) (Figs. 1b and 5c). We found that it was not possible to identify an alternate low-energy model in which LF_N could interact with these residues. One possibility is that there is a conformational change in LF_N and/or PA that could not be modeled by using rigid backbone structures. We now question, however, whether LF_N-dependent dimerization of the oligomerization-deficient PA mutants yielded an unambiguous map of binding defects. Because R178, K197, and R200 are located at the dimer interface, it is possible that mutation of these residues causes oligomerization defects and does not directly affect ligand binding. Because PA dimers are formed only in the presence of ligand, it is difficult to distinguish these two possibilities. Given that the subsite II data may not reflect LF_N binding and that the model recapitulates the independently identified disulfide and electrostatic pairs, we propose the low-energy model as a reliable prediction of the LF_N-PA dimer complex structure. The fact that a purely energy-based prediction can reproduce the experimental results quite well and even point at possible incorrect information is encouraging and demonstrates that high-resolution structure prediction can make useful contributions to the structural characterization of a protein-protein interface, particularly in conjunction with experimental data. The combination of experimental and computational methods in this study may represent the beginning of a new paradigm for structure determination as computational methods become more accurate and structural biologists seek to understand larger and more complex systems that are less amenable to traditional high-resolution structure-determination methods.

The energetically favorable model has LF_N spanning two neighboring PA₆₃ subunits with a buried surface area of 2,300 Ų (Fig. 5 a and c). Although no experiments were conducted on the EF_N-PA interaction for this study, the EF_N structure aligns to the LF_N of the refined model with an rmsd of 1.7 Ų for 191 Cα atoms, suggesting that EF_N and LF_N bind PA similarly. By contrast, the LF_N-binding site overlaps but is distinct from the PA₂₀-binding sites. PA cannot oligomerize in the presence of PA₂₀ because of steric clash. The model shows that a single LF_N molecule binds across two neighboring PA₆₃ subunits and displaces the PA₂₀ fragments of both subunits. This may explain why ligand binding is so important for PA oligomerization.

The bulk of the LF_N interactions are nonetheless with a single PA₆₃ subunit. There is excellent packing between the PA-binding site on LF_N (Fig. 1a) and the PA ligand-binding subsite I (Fig. 1b) with a significant number of electrostatic interactions (Fig. 5b). The interface also contains a buried His residue contributed by LF_N, H229. The prevalence of charged residues at the interface may be relevant to the pH dependence of the subsequent steps of translocation. The low pH of the endosome triggers conversion of the PA heptameric prepore to the pore and initiates the process of ligand translocation. Low pH also seems to aid the unfolding of LF_N, a process required to transport such a large molecule through the narrow pore lumen (10). Because the enzymatic ligand ultimately needs to be released from the heptamer surface to be translocated, there may be a pH dependence to the binding affinity as well. This pH dependence could be achieved by having a high number of charged and/or titratable residues at the interface.

An electrostatic interaction also may be involved in LF_N's contacts with the PA₆₃-PA₆₃ interface, because the model indicates that LF_N E135 and the K197 residue from the neighboring PA subunit will be in close proximity (Fig. 5a). An attempt to verify this interaction by pairing charge-reversal mutants was unsuccessful (data not shown) but may reflect the fact that PA K197 contributes to the binding interaction from both subunits (Fig. 1b). Despite the lack of direct contacts with R178 and R200 of subsite II, the model does suggest that LF_N spans an interface and structurally occludes the neighboring subunit of PA (Fig. 5c). This occlusion is consistent with the observations that only three molecules of EF/LF/LF_N can bind the heptamer at one time (7) and that the PA₆₃ dimer formed from two nonoligomerizing mutants binds only a single LF_N molecule (16). Finally, the model indicates that the N-terminal α-helix of LF_N is oriented over the luminal space of the PA heptamer (Fig. 5 c and d). This helix, corresponding to residues 27-43, represents the first visible part of the LF_N crystal structure, because the N-terminal 26 residues are presumably disordered (12). It has been shown that the N terminus initiates the translocation of LF_N through the lumen of the PA heptameric pore (24). Having LF_N bound such that the N-terminal helix is poised above this opening should facilitate this process and may mean that the N-terminal 26 residues can bind inside the prepore lumen before the beginning of pore formation and translocation (Fig. 5d).