Mapping the Anthrax Protective Antigen Binding Site on the Lethal and Edema Factors*

Entry of anthrax edema factor (EF) and lethal factor (LF) into the cytosol of eukaryotic cells depends on their ability to translocate across the endosomal membrane in the presence of anthrax protective antigen (PA). Here we report attributes of the N-terminal domains of EF and LF (EF N and LF N , respectively) that are critical for their initial interaction with PA. We found that deletion of the first 36 residues of LF N had no effect on its binding to PA or its ability to be translocated. To map the binding site for PA, we used the three-dimensional structure of LF and sequence similarity between EF and LF to select positions for mutagenesis. We identified seven sites in LF N (Asp-182, Asp-187, Leu-188, Tyr-223, His-229, Leu-235, and Tyr-236) where mutation to Ala produced significant binding defects, with H229A and Y236A almost completely eliminating binding. Homologous mutants of EF N displayed nearly identical defects. Cytotox- icity assays confirmed that the LF N mutations impact intoxication. The seven mutation-sensitive amino acids are clustered on the surface of LF and form a small convoluted patch with both hydrophobic and hydro-philic character. We propose that this patch constitutes the recognition site for PA. to permit the appropriate truncation and also introduce an Nde I site. In all constructs, 5 (cid:2) -CTAGGATCCT-TACCGTTGATCTTTAAGTTCTTCC-3 (cid:2) was used to introduce a Bam HI site and act as the reverse primer. Sequences were amplified from a pET15b-LF template by PCR. The PCR products and pET15b (Nova- gen) were gel-purified and digested with Nde I and Bam HI. The digested fragments and nicked vector were gel-purified again, ligated, and trans- formed into Escherichia coli DH5 (cid:1) . Transformants were screened by digestion and verified by sequencing.

Bacillus anthracis releases three discrete monomeric proteins that assemble at the host cell surface into toxic complexes (1). These proteins are collectively referred to as anthrax toxin (ATx) 1 and include the edema factor (EF) and lethal factor (LF) enzymes and the protective antigen (PA). PA mediates the delivery of EF and LF across the host cell membranes so that they can access their cytosolic substrates. EF is an 89-kDa adenylate cyclase that impairs phagocytosis in macrophages (2). LF is a 90-kDa zinc-dependent protease that cleaves mitogen-activated protein kinase kinases in macrophages (3)(4)(5). At high concentrations, the combination of PA and LF can result in death of the host cell macrophages and even the host (6).
ATx intoxication involves binding of PA (83 kDa) to a specific mammalian cell-surface receptor and proteolytic activation by furin or a furin-like protease (7). Cleavage results in the release of the N-terminal 20-kDa fragment of PA, and the release in turn allows the receptor-bound 63-kDa domain (PA 63 ) to hep-tamerize and bind EF and/or LF (8). The complex of heptameric PA 63 ((PA 63 ) 7 ) bound to both receptor(s) and catalytic factor(s) on the cell surface is then internalized by receptor-mediated endocytosis (9). The low pH environment of the endosome is thought to trigger a structural change in (PA 63 ) 7 , allowing it to form a pore and mediate translocation of EF/LF across the endosomal membrane into the cytosol (10). Translocation is also likely to involve unfolding and refolding of the large catalytic moieties before and after traversing the membrane, respectively (11).
EF and LF bind to (PA 63 ) 7 competitively (12) and with high affinity (K d ϳ1 nM (13)). These proteins do not share sequence similarity with other proteins in the database but do share significant sequence identity (35%) and similarity (55%) with each other over the first ϳ250 residues (14). Given their different catalytic activities, this N-terminal sequence was hypothesized to contain a common domain for binding PA. Indeed, it was shown that the N-terminal domain of LF (LF N ) was sufficient for binding PA and could act as a carrier for delivery of heterologous proteins across membranes in the presence of PA (15). The threedimensional crystal structure of LF shows that LF N forms a distinct structural domain and explains how the binding function can be separated from the rest of the molecule (16).
In the current study we have identified attributes of LF and EF that are crucial for their binding to PA. First, we showed by deletion mutagenesis that the first 36 residues of LF N are dispensible for binding to PA 63 and translocation. Next we identified seven positions in LF N (Asp-182, Asp-187, Leu-188, Tyr-223, His-229, Leu-235, and Tyr-236) and the corresponding positions in EF N where mutation to Ala significantly impairs binding to PA and toxin action. These residues are located in three different ␣-helices that interact to form a surface-exposed patch on one face of the LF structure. These patches on EF and LF are proposed to represent the PA 63 recognition sites of these proteins.

EXPERIMENTAL PROCEDURES
Reagents and Chemicals-Oligonucleotides were synthesized by Integrated DNA Technologies. Supplies for cell culture media were from Invitrogen. Sigma supplied all chemicals unless noted otherwise.
Preparation of LF N Constructs-The plasmid pET15b-LF contains the entire LF gene, except for the portion that encodes the signal sequence. The oligonucleotide 5Ј-GGAGGAACATATGGCGGGCGGT-CATGGTGATG-3Ј was used to introduce an NdeI site and serve as a forward primer in the amplification of LF N -(1-263). The forward primers for LF N -(28 -263), LF N -(33-263), LF N -(37-263), LF N -(40 -263), and LF N -(43-263) were constructed to permit the appropriate truncation and also introduce an NdeI site. In all constructs, 5Ј-CTAGGATCCT-TACCGTTGATCTTTAAGTTCTTCC-3Ј was used to introduce a BamHI site and act as the reverse primer. Sequences were amplified from a pET15b-LF template by PCR. The PCR products and pET15b (Novagen) were gel-purified and digested with NdeI and BamHI. The digested fragments and nicked vector were gel-purified again, ligated, and transformed into Escherichia coli DH5␣. Transformants were screened by digestion and verified by sequencing.
Preparation of LF N and Mutants-Mutations were made using the QuikChange method of site-directed mutagenesis using a protocol supplied by the manufacturer (Stratagene). All mutants were made in pET15b-LF N -(28 -263). DNA for each mutant was fully sequenced and then used to create 35 S-LF N protein by in vitro transcription/translation using the TNT-coupled Reticulocyte Lysate System (Promega).
Preparation of EF N and Mutants-The plasmid pET15b-EF N -(1-254) was prepared similarly to the LF N constructs above. Oligonucleotides 5Ј-GGAGGAACATATGAATGAACATTACACTGAG-3Ј and 5Ј-CTAG-GATCCTTAACCTTCTTTCTTCAAACTTTC-3Ј were used as forward and reverse primers, respectively, to PCR-amplify EF N -(1-254) from a pET15b-EF template containing the entire EF gene. The oligonucleotides also introduced NdeI and BamHI sites to facilitate the ligation back into a pET15b vector. Mutants of this construct and their corresponding 35 S-labeled proteins were made as described above for LF N .
Preparation of LF N -DTA Mutants-Mutations in LF N -DTA were made using the same primers and protocol used for constructing the LF N mutants. Each mutant was transformed into E. coli BL21(DE3) (Novagen) and purified using the protocol described previously for wildtype LF N -DTA (11).
Preparation of PA and LF N Proteins-PA, LF N , and LF N (Y236A) were purified from E. coli as described previously (11,17). Protein concentrations were determined using the Bradford protein assay reagent (Bio-Rad). A 0.2 mg/ml solution of PA was activated using a trypsin:PA ratio of 1:2000 (w/w). The mixture was incubated at room temperature for 30 min and quenched with a 10 M excess of soybean trypsin inhibitor. Wild-type LF N -(28 -263) and a Y236A mutant were incubated with trypsin at trypsin:LF N ratios of 1:10, 1:100, 1:1000, and 1:10,000 (w/w) for 60 min. Proteolysis was quenched with soybean trypsin inhibitor, and digestion profiles of wild-type and mutant LF N were compared by SDS-PAGE to assess protein stability.
Cell Culture-CHO-K1 cells (ATCC CCL-61) were grown in Ham's F-12 medium supplemented with 10% calf serum, 500 units/ml penicillin G, and 500 units/ml streptomycin sulfate. Cells were maintained as monolayers and grown in a humidified atmosphere of 5% CO 2 .
Cell Surface Binding and Translocation Assay-The protocol for measuring PA-mediated binding and translocation of 35 S-labeled LF was previously described (11). Briefly, CHO-K1 cells (2 ϫ 10 5 cells/well) were incubated on ice with trypsin-nicked PA for 1 h. The cells were washed with PBS and then incubated on ice with 35 S-labeled EF N or LF N for 1 h. Cells were washed once with PBS and then treated with a low pH buffer at 37°C for 1 min. The cells were exposed to Pronase or a no-Pronase control for 8 min at 37°C, and then protease inhibitors were added. Cells were treated with lysis buffer, and the radioactive content was determined by scintillation counting. To test the level of binding at high pH, the cells were washed three times with PBS after the incubation with EF or LF. The cells were lysed and assayed for radioactive content as above. All of the mutants were assayed for their cell surface binding at high pH.
Inhibition of Protein Synthesis-The protein synthesis inhibition assay was used to measure the ability of PA to deliver mutants of LF N -DTA to the cytosol and was performed as described previously (18). PA was incubated with CHO-K1 cells for 4 h at 37°C in the presence of various amounts of LF N -DTA. The cells were incubated with leucinedeficient medium supplemented with 1 Ci of [ 3 H]leucine/ml for 1 h at 37°C and then washed twice for 10 min each with 5% trichloroacetic acid. The protein precipitate was solubilized with 100 l of 0.2 M KOH. An equal volume of 0.1 M HCl was added, and the amount of tritiated protein was determined by scintillation counting.

Cell Surface Binding and Translocation of LF N Constructs-
Constructs of LF N (LF N -(1-263) and LF N -(28 -263)) were made to correlate with the observed crystal structure of LF. In both constructs the C terminus was extended from amino acid 255 to 263 to complete the 1␣12 helix and the structural domain of LF N (16). Comparison to protein made from a prior construct (pET15b-LF N -(1-255) (17)) showed no change in activity as assessed by a cell surface binding and translocation assay (data not shown). A previous report (15) showing that a 40-residue N-terminal deletion in LF is incapable of killing macrophages suggested that the LF N N terminus, or a portion of it, is required for efficient ATx intoxication. Truncation of the first 27 residues (which were not visible in the structure) had no effect on binding and translocation (Fig. 1). To delineate further how much of a truncation could be tolerated, the LF N -(33-263), LF N -(37-263), LF N -(40 -263), and LF N -(43-263) proteins were tested for PA-dependent binding and translocation. LF N -(33-263) and LF N -(37-263) showed wild-type levels of binding and translocation. Binding of both LF N -(40 -263) and LF N -(43-263) was reduced by ϳ50%, and translocation of that which bound occurred at wild-type levels of efficiency (Fig. 1).
Mapping the Sequence Conservation between EF and LF to the LF Structure Suggests a PA-binding Site-The EF N and LF N sequences were aligned with ClustalW (19) using the BLOSUM62 matrix (20). Strictly conserved residues between the two proteins were mapped to a surface representation of the LF structure ( Fig. 2A). Although the alignment showed significant homology throughout the first 250 residues, analysis with respect to the structure revealed only one significant cluster of conserved surface-exposed amino acids. This cluster (colored red in Fig. 2A) is formed by the intersection of two ␣-helices (1␣6 and 1␣10) and a tyrosine residue (Tyr-223) projecting from a third buried helix (1␣9). We hypothesized that this cluster represented the PA-binding site for LF.
Defining the PA-binding Site Using 35 S-LF N Mutants and a Cell Surface Binding Assay-The conserved residues of the cluster shown in Fig. 2A were individually mutated to alanine in the LF N -(28 -263) construct, and the mutant proteins were 35 S-labeled using in vitro incorporation of [ 35 S]Met at the five methionine sites of the construct. Each 35 S-LF N mutant was visualized on a 4 -20% SDS gradient gel using a PhosphorImager and/or film (data not shown). The concentration for each mutant was measured by scintillation counting and normalized such that each 35 S-LF N sample was applied at 10 Ϫ10 M. Samples were tested for their ability to bind cells in the presence and absence of PA. The difference in binding was then compared with that for LF N -(28 -263) and displayed as a fraction of wild-type binding.
Mutations at Glu-142, Ser-183, Asp-184, Glu-196, and Val-232 showed 79 -100% of the binding observed with wild-type LF N -(28 -263) (Fig. 3), and those residues are shown in green in Fig. 2B. Reduced binding was observed in D182A, D187A, L188A, Y223A, and L235A ( Fig. 3 and Table I). These residues are shown in purple in Fig. 2B to signify a significant although not complete binding defect. The H229A and Y236A mutants showed almost undetectable levels of binding ( Fig. 3 and Table  I) and are shown in red in Fig. 2B. Additional mutations were made in some of the non-conserved residues surrounding the cluster (Y108A, K110A, E139A, T141A, N179A, S181A, Q186A, Q228A, D231A, Q234A, E239A, D245A, and E249A). All showed high (Ͼ65% of wild-type) levels of binding (Fig. 3) and are shown in green in Fig. 2B. One mutation (Y125A) was made on the opposite side of the molecule and bound at 74% of that seen for wild-type (Fig. 3). Finally, the His-229 position was mutated to both glutamine and asparagine to further characterize the role the nitrogen atoms of this residue might play in binding. As with H229A, binding for both mutants, H229N (0 Ϯ 1%) and H229Q (9 Ϯ 5%), was severely impaired.
The Mutations in LF N That Affect Binding Also Show Defects in Cytotoxicity-Mutations that showed a significant defect in ability to bind PA were further characterized by a cytotoxicity assay using LF N -DTA. LF N -DTA is a protein fusion between LF N and the catalytic domain of diphtheria toxin, which inhibits protein synthesis by catalyzing the ADP-ribosylation of elongation factor-2. CHO-K1 cells were incubated with PA and LF N -DTA and then checked for ability to incorporate tritiated leucine into proteins. With a chosen set of conditions, including a 10 Ϫ10 M concentration of PA, the concentration of wild-type LF N -DTA required to inhibit leucine incorporation by 50% (EC 50 ) was 8 Ϯ 2 pM (Fig. 4). The EC 50 value for each of the seven LF N -DTA mutants (D182A, D187A, L188A, V223A, H229A, L235A, and Y236A) was higher than that of wild-type LF N -DTA, and all except that of D187A differed by at least a factor of 10 ( Fig. 4 and Table I).
Mutations in LF N and LF N -DTA Do Not Affect Structural Stability-To ensure structural stability, we expressed and purified the LF N mutant with the most dramatic PA-binding defect (Y236A) to homogeneity and compared its tryptic digestion profile to that of wild-type LF N . The mutant expressed at wild-type levels (100 mg/liter) and eluted as a discrete symmetric peak from an anion exchange column. Incubation of the mutant and wild-type proteins with trypsin for 1 h at room temperature resulted in only slight digestion with a 1:100 w/w trypsin:LF N ratio, and a 1:10 (w/w) ratio was required for full digestion. At both ratios the digestion product profiles on SDS-PAGE were identical between mutant and wild-type (data not shown). Each of the seven mutant proteins with a PA-binding defect was expressed and purified in the LF N -DTA construct. The mutations had no effect on expression yield, and in all cases the proteins eluted as a single symmetric peak off an anion exchange column.
Homologous Mutations in EF N Show Similar Defects in Binding PA on Cells-Mutations were made in EF N at positions corresponding to the seven LF N positions where mutation caused a defect in binding PA. A cell surface binding assay was performed, and binding was calculated as a percentage of that observed for 35 S-EF N . Each mutant (D169A, D174A, L175A, Y214A, H220A, L226A, and Y227A) displayed reduced binding relative to wild-type EF N (Fig. 5 and Table I). DISCUSSION We have used the recently solved crystal structure of LF to guide new experiments probing the relationship between structure and function in ATx intoxication (16). The structure reveals that the bulk of LF N , residues 40 -263, forms a compact structural domain, whereas the ␣-helix containing residues 28 -39 (1␣1) projects out, away from the rest of the molecule (Fig. 2A). The first 27 residues of the structure were not visible, indicating either lack of structure or significant motion. Constructs that take advantage of these structural properties (LF N -(1-263) and LF N -(28 -263)) showed enhanced E. coli expression as compared with LF N -(1-255), with no effect on cell surface binding and translocation. In fact, LF N with a 27-, 32-, or 36-residue N-terminal truncation showed wild-type levels of binding and translocation, whereas a 39-or 42-residue truncation showed a ϳ50% reduction in binding but wild-type translocation efficiency.
The data suggest that the alterations in function observed by us and others with N-terminal deletions are due to a defect in FIG. 2. A, a surface representation of LF depicting the N terminus beginning at residue 28 and the conserved cluster hypothesized to represent the PA-binding site. The colored region corresponds to LF N and displays its sequence conservation with EF N with non-conserved residues in yellow and conserved residues in red. Rotation of the molecule revealed that the only significant patch of three-dimensional sequence conservation on the surface of LF N was the cluster of red residues observed in the center of this view. B, the same surface representation of LF as shown in A is re-colored to represent the results from the cell surface binding assay. All colored residues were individually mutated to alanine, labeled with 35 S, and tested for the ability to bind cells in a PA-dependent fashion. Residues colored in green displayed wild-type levels of binding. The seven residues observed to affect binding are labeled and colored purple to represent an intermediate effect and red to indicate a dramatic effect. All mutations were made in the LF N -(28 -263) construct. binding rather than in a later step in the process. It should be noted that the method of in vitro transcription/translation does not allow for removal of the N-terminal hexahistidine tag and thrombin cleavage site which adds an additional 20 residues to the N terminus of each protein. Our data indicate that the first 36 residues are not specifically required for binding and trans-location, but we do not address whether a flexible "tail" of any sequence is required for function. If a flexible N terminus is required, it is unable to fully compensate after 39 residues have been deleted. Starting at residue 37, the side chains of the N terminus begin to interact with the main body of the LF N structure. Therefore, these N-terminal residues are likely to affect binding indirectly by providing structural support to the rest of the molecule. Other researchers have reported that the LF mutants Y148A, Y149A, I151A, and K153A are unable to lyse macrophages (21), but structural information now indicates that these residues are part of 1␣4, the central helix of the LF N fold with no to relatively low surface exposure. Mutation at these sites is likely to affect binding indirectly by altering structural stability.
To identify the PA-binding sites of EF and LF, we assumed that the binding site would consist only of surface-exposed residues and that the site would be similar, if not identical, between EF and LF. We identified only one significant conserved cluster of surface-exposed residues on the LF structure and hypothesized that this cluster would represent the PAbinding site for LF. The results using alanine scanning and the cell surface binding assay support our hypothesis, because mutation of most of the conserved residues within the cluster caused a decreased ability to bind PA, whereas none of the mutations in non-conserved residues affected binding (Fig. 2). Mutation of both conserved and non-conserved residues surrounding the cluster did not affect binding but served to delimit the boundary of the binding site (green in Fig. 2B). Finally, the results indicate the relative importance of residues within the cluster to binding. In this assay, D187A and L235A showed a ϳ50% reduction in binding of PA, whereas the D182A, L188A, and Y223A mutants bound at only ϳ25-35% of that seen with wild-type (Table I). The H229A and Y236A mutations had the most dramatic effects on binding and indicate that these residues are critical in the binding interaction between LF and PA.
To test the role of these seven residues by another method, we made the same single alanine mutations in an LF N -DTA construct. Being able to perform a protein synthesis inhibition assay with LF N -DTA had several advantages. Using purified protein eliminated the possibility of contaminants affecting the results and also allowed us to accurately measure concentration. Most importantly, the assay serves as a measure of more than binding, as cytotoxicity depends on multiple factors. In addition to LF N -DTA needing to bind PA, the process requires proper endocytosis and trafficking, pH-induced structural changes in PA resulting in pore formation, and the proper unfolding and refolding of LF N -DTA to allow for its efficient translocation and catalysis.
The results from the protein synthesis inhibition assay reinforce the idea that the seven residues identified in the binding assay (Asp-182, Asp-187, Leu-188, Tyr-223, His-229, Leu-235, and Y236) form a PA-binding site. The EC 50 value for each of the seven LF N -DTA mutants was higher than that for wildtype LF N -DTA, and the relative cytotoxicity profile generally  correlated well with the profile of observed binding differences (Table I). For example, the Y236A mutation, as in the binding assay, showed the most dramatic defect, requiring a concentration 100 times that of wild type to achieve 50% inhibition. Slight profile differences included the H229A mutation, which had a drastic defect in the cell binding assay but a more modest defect when incorporated into LF N -DTA. In an effort to characterize further the role of His-229 in binding PA, we constructed two additional mutants, H229N and H229Q. These mutations are more similar to histidine than alanine in terms of steric bulk and can be used to probe the role, if any, of the two nitrogen atoms of histidine (22). In this case, however, both 35 S-LF N H229N and 35 S-LF N H229Q nearly ablated binding, supporting the idea that the dramatic reduction in cell surface binding observed for H229A is significant. Furthermore, although it does not prove that the histidine nitrogens are not involved, this result suggests that it could be unique properties of the imidazole ring and/or the titratable charge that forms the basis of this binding interaction.
Using the argument of sequence conservation to identify a binding site implies that the homologous seven mutations in EF will also show defects in binding. This was confirmed using the cell surface binding assay for each of the seven corresponding EF N mutants ( Table I). The fact that the profiles mirror those observed with the LF N -(28 -263) mutations is notable and may indicate that the profile differences between the binding and cytotoxicity assays indicate roles for some of these residues later in the intoxication process.
The seven residues identified in this report form a patch on one face of the LF structure. Although distant in primary sequence, the patch is close enough to the N terminus of the molecule to suggest that the structural integrity of the N terminus (starting after residue 36) would affect the properties of the binding site. As visualized in the uncomplexed LF structure, the patch has a net negative charge (largely due to Asp-182 and Asp-187) and contains both a pocket (composed of residues His-229, Tyr-223, and Leu-188) and a protruding loop (containing Leu-235 and Tyr-236). Although the charge and contour properties of the patch do not suggest an obvious binding site on the surface of (PA 63 ) 7 , we hope to use these mutants to guide the search for the LF/EF-binding site on (PA 63 ) 7 . A model of these molecules in complex will provide the framework for addressing the next steps in the intoxication pathway of anthrax toxin.