Alanine-scanning Mutations in Domain 4 of Anthrax Toxin Protective Antigen Reveal Residues Important for Binding to the Cellular Receptor and to a Neutralizing Monoclonal Antibody*

A panel of variants with alanine substitutions in the small loop of anthrax toxin protective antigen domain 4 was created to determine individual amino acid resi- dues critical for interactions with the cellular receptor and with a neutralizing monoclonal antibody, 14B7. Substituted protective antigen proteins were analyzed by cellular cytotoxicity assays, and their interactions with antibody were measured by plasmon surface resonance and analytical ultracentrifugation. Residue Asp 683 was the most critical for cell binding and toxicity, causing an (cid:1) 1000-fold reduction in toxicity, but was not a large factor for interactions with 14B7. Substitutions in residues Tyr 681 , Asn 682 , and Pro 686 also reduced toxicity significantly, by 10–100-fold. Of these, only Asn 682 and Pro 686 were also critical for interactions with 14B7. However, residues Lys 684 , Leu 685 , Leu 687 , and Tyr 688 were critical for 14B7 binding without greatly affecting toxicity. The K684A and L685A variants exhibited wild type levels of toxicity in cell culture assays; the L687A and Y688A variants were reduced only 1.5- and 5-fold, respectively. Bacillus cell line, CHO FD11, were previously described (6, 31) and were grown in (cid:3) -min- imal essential medium supplemented with 10% fetal bovine serum, 10 m M HEPES, 2 m M GlutaMax I, and 50 (cid:2) g/ml gentamycin. Production and Purification of Small Loop Mutants— PA variants were constructed from parental plasmid pYS5 (32). For each variant, mutagenic primers encoding the desired amino acid substitution were made for each strand. The region between the change and the unique Bam HI site downstream of the PA coding region on pYS5 was amplified with one mutagenic primer and a primer matching the Bam HI site. The other mutagenic primer and a primer complementary to the unique Pst I site in the PA coding region were used to amplify the region between the desired mutation and the Pst I site. The products from the two reactions were then used to prime off each other in a third PCR whose product gave a fragment containing the desired change, having Pst I- and Bam HI-cut sites on the ends. The PCR products and pYS5 were each digested with Pst I and Bam HI prior to ligation. The construct for each variant was verified by sequencing. PA variants were produced and purified in a modification of previously described methods (25, 33). Briefly, PA was expressed and se- creted by B. anthracis strain BH445. Culture supernatants containing the desired protein were adjusted to 5 m M EDTA and filter-sterilized prior to the addition of ammonium sulfate to 2 M . Phenyl-Sepharose Fast Flow resin was then added in batch. After binding for 1 h with resin a funnel and with 600 ml cold (1.5 M m M m M EDTA). PA variants were eluted from the resin with 0.3 M 10 m M 7.4, m M EDTA. The resultant proteins were further purified by precipitation with ammonium sulfate (47 g/100 ml of eluted protein) followed by gel filtration tography Sephacryl S-100 column. Fractions eluted m M Tris, pH NaCl, m M EDTA. Pooled fractions containing the PA variants (cid:1) 90% pure, as on gels Coomassie Brilliant Blue and were frozen (cid:2) 80 °C. Cytotoxicity and Neutralization Assays— Cytotoxicity assays were with both RAW264.7 (mouse macrophage) and CHO-CL6 cell lines. For RAW264.7 assays, cells were plated 96-well plates at a of 30,000–35,000 cells/well and incubated overnight. The next morning growth medium was by aspiration and replaced with 125 (cid:2) l of new medium containing 200 ng/ml LF and a dilution series of PA or PA variants ranging from 0 to 1000 ng/ml, where each tested dilution was in triplicate wells. Cells were incubated at 37 for 4 h. Cell viability was by the addition of 3-[4,5-dimethyl-thiazol-2-yl]-2,5-diphenyltetrazolium bromide (MTT). Viable cells con- verted the MTT to an insoluble blue pigment, which was measured as (34). Results were plotted and with LF PA. EC regression the type EC Analytical Centrifugation— Sedimentation equilibrium experiments were conducted in a Beckman Optima XL-I/A analytical ultracentrifuge (Beckman Coulter, Fullerton, CA). Antibody, wild type PA or variants, and mixtures were dissolved in 150 (cid:2) l of phosphate-buffered saline at loading concentrations of 0.1–0.3 mg/ml, and equilibrium absorbance profiles were acquired at rotor speeds of 9,000, 13,000, and 17,000 rpm; at a rotor temperature of 4 °C; and at wavelengths of 230, 250, and 280 nm. Protein extinction coefficients at 280 nm were estimated using the software SEDNTERP, kindly provided by Dr. J. Philo. Buoyant molar mass values for the proteins were obtained experimentally but were consistent with estimated values derived from amino acid sequence. Binding constants were calculated by global nonlinear regression with the software SEDPHAT, based on equations for the radial concentra- tion distribution of ideally sedimenting species in mechanical and chemical equilibrium as described elsewhere (38, 39). The model for the analysis of the radial total protein distribution included four exponen-tial terms, accounting for free antibody, free PA, and 1:1 and 1:2 complexes, linked at each radius by mass action law and the statistical factors for independent binding to two identical sites. Sedimentation velocity experiments were performed at a rotor speed of 45,000 rpm and a temperature of 20 °C. Antibody, wild type PA or variants, and mixtures were diluted to a concentration of 0.2 mg/ml. 220 (cid:2) l of PA solutions were filled in the sample sector of Epon double-sector centerpieces, and mixtures of PA with antibody were filled in the reference sector. The light intensity transmitted through both sectors at 280 nm was measured with the absorbance scanning system in time intervals of 400 s. Using the software SEDFIT, the intensity data were transformed into pseudoabsorbance data (40) and analyzed with the model of diffusion-corrected sedimentation coefficient distributions, c ( s ) (41, 42). As a result, separate sedimentation coefficient distributions for the PA and the mixtures were obtained, which resolved free PA, anti- body, and the complex. Interpretation of the sedimentation coefficient distributions directly allowed both the determination of the active con- centration of PA variants and the semiquantitative assessment of the strength of the interaction between PA and the antibody from the sedimentation coefficient of the complex. shows broader distribution with some complex and some free 14B7. shows pseudoabsorbance versus radius at different times for the original velocity sedimentation data of the PA P686A/14B7 mixture, corrected for systematic noise components.

Bacillus anthracis secretes two toxins: edema toxin and lethal toxin. Each is composed of a common binding component, protective antigen (PA), 1 together with an enzymatic component, edema factor (EF), in the case of edema toxin and lethal factor (LF) in the case of lethal toxin (1)(2)(3). The current model for toxin entry into the cell illustrates the centrality of PA for toxin action. PA binds to cellular receptors, recently identified as splice variants of either tumor endothelial marker 8 (TEM8) (4 -6) or the closely related capillary morphogenesis protein 2 (CMG2) (7). Furin cleaves PA, releasing a 20-kDa fragment and leaving behind a 63-kDa portion (PA 63 ) capable of forming a heptamer, which has a newly exposed surface that binds EF and LF (8 -12). Heptamer complexes enter the endocytic pathway by receptor-mediated endocytosis (13), and upon acidification of the vesicle, the PA 63 heptamer undergoes a conforma-tional change to form a pore through which EF and LF translocate into the cytoplasm (10, 11, 14 -16). Once in the cytoplasm, EF and LF exert their toxic effects.
The PA protein can be divided into four domains based on its crystal structure, and functions can be attributed to the different domains based on mutational and biochemical analyses (16). Domain 1 (residues 1-258) contains the furin cleavage site as well as the hydrophobic portion of PA, which is exposed upon furin cleavage to allow EF and LF to bind (16,17). Several lines of evidence indicate that domain 2 (residues 259 -487) is involved in oligomerization and contains the loop that inserts into the membrane to form the channel through which the LF and EF enter the cytosol (16, 18 -20). Various amino acids in domain 3 (residues 488 -595) are necessary for oligomerization, and this has been the only function attributed to domain 3 to date (21,22). Domain 4 (residues 596 -735) is essential for binding to cellular receptor as indicated by several lines of evidence. A CNBr fragment of PA containing residues 663-735 successfully competed with PA in cell binding assays, and mutations in domain 4 prevented PA binding to cells (23)(24)(25).
Monoclonal antibody and mutational analysis studies further localized the residues necessary for binding to receptor. Monoclonal antibody 14B7 neutralized anthrax toxin by inhibiting PA binding to receptor and recognized a region of PA between residues 671 and 721 (17). Additionally, multisubstituted PA variants in and near the small loop of domain 4 (between ␤ strands 4␤ 8 and 4␤ 9 ) suggested that residues essential for binding were in that region (25). The alanine substitutions described in this work cover the small loop (amino acids 679 -693) of domain 4. We also included another residue, Asn 657 , located nearby in the crystal structure and implicated by our earlier multisubstituted variants.
The development of new reagents to prevent and treat anthrax can be aided by our understanding of the molecular interactions between the toxin components and cellular targets as well as by interactions between the toxin components and potential reagents. One of the molecular reactions it would be beneficial to understand in molecular detail would be that between anthrax toxin and its cellular receptor. As noted above, much progress has been made in the study of PA and recently also with its cellular receptors, TEM8 and capillary morphogenesis protein 2, but studies of the interaction between the toxin with receptor have to date been limited to deletions or multiple substitutions in the proteins (6,25). One goal of this study was to determine which individual amino acids in the previously identified region of PA were important for interaction with its receptor.
For development of improved vaccines and antibody-based therapeutics, it would be helpful to identify the residues in PA necessary for recognition and neutralization of PA. 14B7 is one of a group of monoclonal antibodies that neutralize anthrax toxin by inhibiting the binding of PA to cells (26). Each antibody in this class recognizes an antigenic region between amino acids 671 and 721 of PA (17,27). 14B7 and affinity-enhanced derivatives of it have been proposed as potential therapeutic reagents for anthrax (28). The second goal of this work was to identify amino acids in the small loop region of PA domain 4 necessary for interaction with the neutralizing monoclonal antibody 14B7. Additionally, we sought to elucidate any overlap and differences between the amino acid residues necessary for PA binding to receptor and to 14B7.

EXPERIMENTAL PROCEDURES
Materials-The enzymes and reagents for DNA manipulations were from New England Biolabs (Beverly, MA), Panvera (Madison, WI), and Stratagene (La Jolla, CA). Mutagenic primers were purchased from the Center for Biologics Evaluation and Research (CBER) at the FDA (Bethesda, MD). Phenyl-Sepharose Fast Flow and Sephacryl S-100 were purchased from Amersham Biosciences. Wild type PA, LF, and FP59 were purified in our laboratory as previously described (29,30). Rabbit anti-PA antibody (number 5308) was made in our laboratory, and goat anti-rabbit IgG-horseradish peroxidase (sc2054) was purchased from Santa Cruz Biotechnology, Inc. (Santa Cruz, CA). Monoclonal antibody 14B7 was purified from mouse ascites fluid.
Cell Lines and Culture Media-All culture media and cell culture solutions were purchased from Invitrogen. RAW264.7, a mouse macrophage cell line, was grown in Dulbecco's modified Eagle's medium with 10% fetal bovine serum, 10 mM HEPES, 2 mM Glutamax I, and 50 g/ml gentamycin. CHO cell clone 6 (CHO-CL6) and a furin-deficient cell line, CHO FD11, were previously described (6,31) and were grown in ␣-minimal essential medium supplemented with 10% fetal bovine serum, 10 mM HEPES, 2 mM GlutaMax I, and 50 g/ml gentamycin.
Production and Purification of Small Loop Mutants-PA variants were constructed from parental plasmid pYS5 (32). For each variant, mutagenic primers encoding the desired amino acid substitution were made for each strand. The region between the change and the unique BamHI site downstream of the PA coding region on pYS5 was amplified with one mutagenic primer and a primer matching the BamHI site. The other mutagenic primer and a primer complementary to the unique PstI site in the PA coding region were used to amplify the region between the desired mutation and the PstI site. The products from the two reactions were then used to prime off each other in a third PCR whose product gave a fragment containing the desired change, having PstI-and BamHI-cut sites on the ends. The PCR products and pYS5 were each digested with PstI and BamHI prior to ligation. The construct for each variant was verified by sequencing.
PA variants were produced and purified in a modification of previously described methods (25,33). Briefly, PA was expressed and secreted by B. anthracis strain BH445. Culture supernatants containing the desired protein were adjusted to 5 mM EDTA and filter-sterilized prior to the addition of ammonium sulfate to 2 M. Phenyl-Sepharose Fast Flow resin was then added in batch. After binding for 1 h with gentle mixing at 4°C, resin was collected in a plastic Buchner funnel and washed with 600 ml of cold buffer (1.5 M ammonium sulfate, 10 mM HEPES, pH 7.4, 1 mM EDTA). PA variants were eluted from the resin with 0.3 M ammonium sulfate, 10 mM HEPES, pH 7.4, 1 mM EDTA. The resultant proteins were further purified by precipitation with ammonium sulfate (47 g/100 ml of eluted protein) followed by gel filtration chromatography on a Sephacryl S-100 column. Fractions were eluted from the column with 10 mM Tris, pH 8.0, 50 mM NaCl, 0.5 mM EDTA. Pooled fractions containing the PA variants were ϳ90% pure, as judged on gels stained with Coomassie Brilliant Blue and were stored frozen at Ϫ80°C.
Cytotoxicity and Neutralization Assays-Cytotoxicity assays were performed with both RAW264.7 (mouse macrophage) and CHO-CL6 cell lines. For RAW264.7 assays, cells were plated in 96-well plates at a density of 30,000 -35,000 cells/well and incubated overnight. The next morning growth medium was removed by aspiration and replaced with 125 l of new medium containing 200 ng/ml LF and a dilution series of PA or PA variants ranging from 0 to 1000 ng/ml, where each tested dilution was made in triplicate wells. Cells were incubated at 37°C for 4 h. Cell viability was determined by the addition of 3-[4,5-dimethylthiazol-2-yl]-2,5-diphenyltetrazolium bromide (MTT). Viable cells converted the MTT to an insoluble blue pigment, which was measured as previously described (34). Results were plotted and analyzed with Prism software (GraphPad Software Inc., San Diego) as the percentage viability of control wells containing LF without PA. EC 50 values were determined by nonlinear regression sigmoidal dose-response analysis with variable slopes. Each assay was performed three times, and the graphs from a representative assay are shown in Fig. 1. Variant to wild type ratios of EC 50 values were calculated.
CHO-CL6 cytotoxicity assays were performed similarly, with two exceptions: 1) the cells were plated at a density of ϳ15,000 cells/well, because after the addition of the toxins (50 ng/ml FP59 with a dilution series of PA or PA variants from 0 to 2000 ng/ml), cells were incubated 48 h prior to MTT addition; 2) FP59, a recombinant toxin consisting of the N-terminal PA-binding portion of LF fused to the ADP-ribosylation domain of Pseudomonas exotoxin A, was used in place of LF, because CHO cells are not lysed by lethal toxin. EC 50 values were determined for three independent assays as above and are reported with S.E.
14B7 neutralization assays were also performed in 96-well plates in a similar manner as described above with RAW264.7 cells. In the 14B7 neutralization assays, PA or toxic PA variant and LF (200 ng/ml each) were preincubated for 30 min at 37°C with 14B7 in a serial dilution from 0 to 10 g/ml. Neutralization by rabbit polyclonal anti-PA antibody 5308 was tested by adding serum at a 1:800 dilution for preincubation with PA or toxic PA variant and LF (200 ng/ml each). For both types of assays (14B7 or polyclonal 5308), after the preincubation, the toxin/antibody solutions were added to cells from which medium had been aspirated and were incubated at 37°C for 4 h. The MTT addition and analysis were identical to those in the cytotoxicity assays.
Assay of PA Binding to Cells-CHO FD11 cells were grown in 24-well plates to confluence. Cells were chilled to 4°C in 0.5 ml of binding medium (minimal essential medium with Hanks' salts, 1% fetal bovine serum, 10 mM Hepes). PA proteins were added to a concentration of 0.5 g/ml, and incubation continued for 2 h. Cell monolayers were then washed five times with Hanks' balanced salt solution (Biofluids, Rockville, MD) containing 1% fetal bovine serum. The cells were lysed in 100 l of modified radioimmune precipitation lysis buffer (50 mM Tris-HCl, pH 7.4, 1% Nonidet P-40, 0.25% sodium deoxycholate, 150 mM NaCl, 1 mM EDTA, 2 mM MgCl 2 , 10 units/ml benzonase DNase (Novagen), and 1 tablet of Complete protease inhibitor (Roche Applied Science) per 20 ml. The cell lysates were subjected to SDS-PAGE using 4 -20% Trisglycine gradient gels (Novex, San Diego, CA). Prior to loading, the cell lysates were boiled for 5 min in 1ϫ SDS sample buffer (50 mM Tris-HCl, pH 6.8, 2% SDS, 100 mM dithiothreitol, 0.01% bromphenol blue, 6% glycerol). The proteins were then transferred to nitrocellulose membranes, followed by Western blotting. PA was visualized by chemiluminescence using the West Pico Kit (Pierce).
Surface Plasmon Resonance (SPR) Analysis-Biosensor experiments were conducted with a Biacore X instrument. The antibody was covalently attached to a carboxymethyl dextran surface of an F1 chip using standard amine coupling (35), with the antibody dissolved at 0.08 mg/ml in 10 mM sodium acetate, pH 5.5. The reaction was stopped when 300 -400 resonance units were cross-linked. Binding experiments were performed at 25°C, in 10 mM HEPES, pH 7.4, 150 mM NaCl, 3 mM EDTA, 0.005% Tween 20, at a flow rate of 5 l/min, and the surface was regenerated with 10 mM glycine, pH 2.0. For wild type PA and each variant, a series of association and dissociation sensorgrams were collected at different protein concentrations, starting at the stock concentration between 1 and 10 mg/ml and proceeding in serial dilutions by a factor 2 or 3 until no significant binding was detected. Each association sensorgram was observed until steady-state was reached and was corrected for buffer refractive index offsets by subtraction of the signal from a nonfunctionalized surface. After each concentration series, the stability of the surface was verified by injection of 1 M (83 g/ml) wild type PA.
Mathematical data analysis was based on a model with a heterogeneous population of surface sites with continuous binding constant distributions (36). In brief, the measured equilibrium surface binding signal s(c) was described as a superposition of binding signals s eq (K A , c) from different populations of sites with different affinity, each following a Langmuir isotherm (37).
The differential affinity distribution P(K A ) was integrated over the peak of the high affinity sites to give a weight average estimate of the binding constant (36).
Analytical Centrifugation-Sedimentation equilibrium experiments were conducted in a Beckman Optima XL-I/A analytical ultracentrifuge (Beckman Coulter, Fullerton, CA). Antibody, wild type PA or variants, and mixtures were dissolved in 150 l of phosphate-buffered saline at loading concentrations of 0.1-0.3 mg/ml, and equilibrium absorbance profiles were acquired at rotor speeds of 9,000, 13,000, and 17,000 rpm; at a rotor temperature of 4°C; and at wavelengths of 230, 250, and 280 nm. Protein extinction coefficients at 280 nm were estimated using the software SEDNTERP, kindly provided by Dr. J. Philo. Buoyant molar mass values for the proteins were obtained experimentally but were consistent with estimated values derived from amino acid sequence. Binding constants were calculated by global nonlinear regression with the software SEDPHAT, based on equations for the radial concentration distribution of ideally sedimenting species in mechanical and chemical equilibrium as described elsewhere (38,39). The model for the analysis of the radial total protein distribution included four exponential terms, accounting for free antibody, free PA, and 1:1 and 1:2 complexes, linked at each radius by mass action law and the statistical factors for independent binding to two identical sites.
Sedimentation velocity experiments were performed at a rotor speed of 45,000 rpm and a temperature of 20°C. Antibody, wild type PA or variants, and mixtures were diluted to a concentration of 0.2 mg/ml. 220 l of PA solutions were filled in the sample sector of Epon double-sector centerpieces, and mixtures of PA with antibody were filled in the reference sector. The light intensity transmitted through both sectors at 280 nm was measured with the absorbance scanning system in time intervals of 400 s. Using the software SEDFIT, the intensity data were transformed into pseudoabsorbance data (40) and analyzed with the model of diffusion-corrected sedimentation coefficient distributions, c(s) (41,42). As a result, separate sedimentation coefficient distributions for the PA and the mixtures were obtained, which resolved free PA, antibody, and the complex. Interpretation of the sedimentation coefficient distributions directly allowed both the determination of the active concentration of PA variants and the semiquantitative assessment of the strength of the interaction between PA and the antibody from the sedimentation coefficient of the complex.

Several Residues in the Small Loop Region of PA Domain 4
Are Important for Toxicity-Previous studies in our laboratory with multiple substitutions in PA domain 4 indicated that the small loop region was critical for binding to the anthrax receptor (25). To determine the importance of individual residues, we made a set of alanine-scanning mutants covering the domain 4 small loop and including residue Asn 657 , which is outside the small loop but located nearby in the crystal structure. In a few cases, we made doubly or triply substituted proteins to ascertain if effects would be additive or cooperative. Initial assessments of binding ability were made utilizing the mouse macrophage RAW264.7 cell line, which rapidly lyses upon the addition of PA and LF. If the PA variants failed to bind, the cells would be unaffected by the addition of LF and would metabolize the MTT dye similarly to untreated controls, forming a blue pigment. Variants that successfully bound to the receptor would transport LF, causing cell lysis and the inability to convert MTT to a blue pigment. The assay was performed three times, with results from a representative assay shown in Fig. 1A. The RAW264.7 assay exhibited slightly variable EC 50 values from day to day, but the ratios of the substituted proteins to wild type remained constant. The range of ratios for PA variant/wild type EC 50 values calculated from the three independent assays is reported in Table I. PA variants D683A, N657A/N682A, D683A/L685E/Y688C, and D683A/L685E/Y688K were all nontoxic at 1 g/ml, the highest level tested in the assay (Fig. 1A). Several amino acid substitutions shifted the EC 50 value from that of wild type toxin, 13 ng/ml. The N682A variant exhibited such low toxicity in this assay that it was not possible to determine an EC 50 value ( Fig. 1A and Table I). Substituting alanine for residues Pro 686 and Tyr 681 also had large effects on toxicity, making the variants 9-and 11-fold less toxic than wild type PA, respectively (Table I). Variants N657A, L687A, and I689A were about 5-fold less toxic than wild type. A smaller effect on toxicity was seen for the K680A variant, which was ϳ2-fold less toxic than PA. The remainder of the substitutions did not appear to affect toxicity and had variant/wild type EC 50 ratios of 1.
Due to the variability and limited sensitivity of the cytotoxicity assays with RAW264.7 cells, the panel of variants was also assayed in a CHO cell line, CHO-CL6. Although CHO cells do not undergo lysis upon the addition of PA and LF, they do express ϳ10,000 PA receptors/cell (43). To circumvent the insensitivity of CHO cells to LF, in its place we used FP59, a fusion toxin between the N-terminal portion of LF (residues 1-254) and the ADP-ribosylation domain of Pseudomonas exotoxin A. Such fusions are extremely toxic to CHO cells when applied with PA (44). EC 50 values for wild type PA were lower for CHO cells than for the RAW264.7 cell line (Fig. 1, compare  A and B), making the CHO-CL6 cytotoxicity assay more sensitive than the RAW cell assay. EC 50 values for the wild type protein and the panel of variants are reported in Table I.
The pattern of toxicity levels for the variant panel was similar in CHO-CL6 cell assays to that in RAW264.7 cell assays (Table I), but the increased sensitivity of the CHO-CL6 assay provided further information. It was possible, for instance, to determine an EC 50 value for the N682A protein, which was not possible in the assay with RAW264.7 cells. This substitution decreased toxicity almost 150-fold. The D683A variant also exhibited weak toxicity in tests with CHO-CL6 cells, and in one of the three assays there were sufficient data points to determine an EC 50 value of 828. This protein seems to fall at the limit of EC 50 determination, because at protein levels higher than 2 g/ml (the highest level tested in this assay), nonspecific uptake of toxins becomes a concern, and toxicity in this case would not reflect receptor-PA interaction.
The N657A/N682A doubly substituted protein did not exhibit any toxicity at 2 g/ml in either cell culture system, whereas single alanine substitutions at these residues decreased toxicity by 7-and 140-fold, indicating that effects of the substitutions were synergistic. The two variants triply substituted at residues Asp 683 , Leu 685 , and Tyr 688 also remained nontoxic for CHO-CL6 cells. The triple variants were interesting because the singly substituted D683A protein did exhibit some toxicity at high concentrations in the CHO-CL6 cytotoxicity assay, and neither the L685A nor Y688A single substitutions greatly affected toxicity. It may be that the non-alanine amino acid substitutions for Leu 685 and Tyr 688 in the triple variants had greater effects on binding ability than the alanine substitutions. It is also possible that 1.5-fold loss of toxicity with the single amino acid substitution at position Tyr 688 was additive in the D683A background.
Variants Y681A and P686A were 32-and 17-fold less toxic than PA in the assay with CHO-CL6 cells, distinguishing between the similar values (9 and 11, respectively) obtained in the RAW264.7 assay (Table I). Single amino acid substitutions N657A, L687A, and I689A again were grouped closely together, exhibiting 6-, 5-, and 4-fold losses of toxicity, respectively. Results for the K680A variant were very similar to those in the assay with RAW264.7 cells, with a 2-fold reduction in toxicity levels.
Decreases in Toxicity Are Due to Decreases in Cell Binding-Based on previous studies in our laboratory, it was likely that the decreased cytotoxicity in the cell culture assays was due to the decreased ability of the variants to bind to the cellular receptor. To rule out the possibility that the less toxic variants bound cells similarly to the wild type protein and exhibited a loss in toxicity for other reasons, we assessed the ability of the proteins to bind to CHO FD11 cells. CHO FD11 cells are deficient in furin, so this cell line provided a background in which receptor binding by the variants was investigated without allowing further processing. The variants were allowed to bind to CHO FD11 cells for 2 h at 4°C. After washing to remove any unbound protein, cell lysates were prepared, subjected to SDS-PAGE, and transferred to nitrocellulose for Western blot analysis with polyclonal anti-PA serum. Fig. 2 shows that wild type PA bound strongly to cells. The results of the binding assay correlated very well with the toxicity assays. None of the less toxic variants bound to cells in amounts similar to the wild type toxin. The K680A and Y688A substitutions caused small decreases in toxicity and likewise bound more similarly to wild type PA than other substitutions that had more severe toxicity defects. N657A, L687A, and I689A bound more poorly in correlation with their greater defects in toxicity; binding by the nontoxic variants was barely detectable. As expected, the toxic variants retained the ability to bind to cells in levels similar to that of wild type protein.
Several Small Loop Residues Are Essential for 14B7 Neutralization-Monoclonal antibodies that block PA binding to cells have been isolated and shown to bind to domain 4 of PA (17,26). From our cell binding and toxicity assays, it was clear that several amino acids in the small loop region of domain 4 are required for PA binding to cells, and it was conceivable that the same region of PA could also be recognized by the monoclonal antibodies blocking PA binding to cells. We tested this idea with one such neutralizing antibody, 14B7. For these assays, only toxic variants could be tested. In the assay, 200 ng/ml each of LF and PA variant were preincubated with varying amounts 14B7 from 0 to 10 g/ml prior to the addition to RAW264.7 cells. The ability of 14B7 to protect the RAW cells from lysis was then assayed with MTT as in the previous cytotoxicity assays. Results from a representative assay are shown in Fig. 3.
In stark contrast to wild type PA, singly substituted PA variants K684A, L685A, L687A, and Y688A were not neutralized by 14B7, even when antibody was added in 25-fold molar excess (10 g/ml) over the variants (Fig. 3A). Of the four substitutions rendering the variants resistant to 14B7, K684A and L685A had no measurable effect on toxicity. L687A reduced toxicity 5-fold, and Y688A consistently exhibited a slightly reduced level of toxicity. Four other single substitutions in PA had decreased neutralization by 14B7; variants K679A, I689A, S690A, and N691A required greater concentrations of 14B7 for neutralization than did wild type PA (Fig. 3B). Of those substitutions affecting 14B7 neutralization, only L687A significantly affected toxicity. The remaining variants were neutralized by levels of 14B7 similar to the levels that neutralized wild type protein (Fig. 3C). Variants Y681A and P686A were already less toxic than wild type at the levels tested in this assay (200 ng/ml), so their neutralization may not reflect the same level of binding needed for neutralization of the other variants and wild type protein. It  exhibiting significant toxicity could be tested for neutralization, so this assay was not useful for interpreting the contributions of residues Asn 682 and Asp 683 to interaction with 14B7. Resistance to antibody neutralization was specific to 14B7; all toxic variants were completely neutralized by polyclonal anti-PA serum (data not shown).
Biophysical Measurements Reveal That Residue Asp 683 Is Not Essential for 14B7 Binding-Equilibrium constants of the antibody binding to the different variants were determined independently by SPR, sedimentation equilibrium, and sedimentation velocity. Sedimentation equilibrium experiments provide the most direct measure for the equilibrium constants and gave results for the wild type interaction that are consistent with the affinity previously reported (28) (Fig. 4, top panel). Unfortunately, the extended time necessary to attain sedimentation equilibrium at different rotor speeds, combined with the rigorous requirements for sample purity did not permit sedimentation equilibrium experiments with all variants. Sedimentation velocity provided a measure for the relative affinities through comparison of the populations of protein migrating at the sedimentation rate of free PA, free antibody, and with the complex, all of which can be resolved in the diffusioncorrected sedimentation coefficient distributions. Because the reversible binding reaction is taking place on the time scale of the sedimentation experiment, an indirect relative measure of the binding constant could be derived from a single sedimentation experiment by comparing the sedimentation rate of the complex (Fig. 4, lower panel). However, this measure is a combined measure of the rate and affinity constants of the interactions. In addition, the hydrodynamic separation of the protein from peptides and aggregates allowed the determination of the active PA concentration. This information was used for the following SPR analysis.
The analysis of the kinetics of PA binding to immobilized antibody revealed that the measured surface binding kinetics in the SPR biosensor was transport-limited for the wild type PA and most of the high affinity variants. An analysis of the binding progress in the presence of transport limitation can be very difficult, because it requires assumption about, and modeling of, the detailed reaction/diffusion process at the sensor surface that is governing the measured surface binding. Nevertheless, the empirically observed rate constants of the surface binding signal can be taken as lower limits for the chemical rate constants. (The true chemical rate constants may be higher and masked by the transport being the rate-limiting step, but they cannot be slower than the observed apparent rate constants.) For the wild type, the minimal estimates of the on-rate and off-rate constants were as follows: k off Ͼ 0.003/s,

FIG. 2. Losses and reductions in toxicity reflect losses and decreases in binding to cells.
Wild type and substituted PA proteins (0.5 g/ml) were allowed to bind to CHO FD11 cells for 2 h at 4°C. Cells were washed to remove unbound PA and lysed. Cell lysates were subjected to SDS-PAGE followed by Western blotting with anti-PA serum to detect bound PA. The first lane contains wild type PA, and the following lanes contain the variants, as labeled. Molecular weight standards (not shown) identified the single band shown as the 83-kDa PA. and k on Ͼ 5 ϫ 10 5 /ms (Fig. 5, top). Because of these difficulties in the analysis of the binding kinetics, the quantitative comparison of the binding properties of the variants was essentially restricted to the determination of the equilibrium binding constant from the interpretation of the equilibrium isotherms (Fig. 5, middle). When using a standard CM5 dextran-coated sensor chip, artifactual binding sites with ϳ100-fold lower affinity were observed that dominated the surface binding. The generation of lower affinity sites has been commonly observed when using the random immobilization chemistry, and similar effects have been attributed also to steric hindrance and/or influences of the local microenvironment in the immobilization matrix (45). With the short dextran layer of the F1 chip, the low affinity sites amounted to only ϳ50% of all sites, and they could be discriminated from the higher affinity PA binding sites that reflected the unimpeded interaction, as judged by consistency of those binding constants with sedimentation equilibrium results (Fig. 5, bottom).
Thus, each of the biophysical techniques applied exhibited specific limitations for the characterization of some PA variants. However, for the variants where the methods overlap, the binding constants from the high affinity surface binding site, from sedimentation equilibrium, and from the semiquantitative ranking by sedimentation coefficient distributions gave consistent results (with affinity constants by ultracentrifugation and SPR generally within a factor of 2-3 or better). The combination of the results from the three techniques provided a complete description of the affinity of the variants (Table I,  far right column). From these data, it was possible to distinguish a set of single amino acid substitutions that had a large effect on affinity to antibody, including L685A, which reduced the affinity to 14B7 at least 10,000-fold compared with wild type PA. Three individual substitutions reduced binding affinity ϳ70 -100-fold: N682A, L687A, and Y688A. Substitutions The sedimentation coefficient of the wild type PA-14B7 complex is highest, due to the highest affinity and stability. For L685A, virtually no complex is formed, and the peaks in the sedimentation pattern are virtually like those measured for the separate proteins. PA P686A shows a broader distribution with some complex and some free 14B7. The inset shows pseudoabsorbance versus radius at different times for the original velocity sedimentation data of the PA P686A/14B7 mixture, corrected for systematic noise components.
K684A and P686A reduced affinity 18-and 26-fold, respectively. Where measurable by neutralization assays, this group of six single substitutions prevented 14B7 neutralization, with the exception of P686A, which exhibited greatly reduced toxicity compared with wild type protein (Fig. 3, compare wild type and P686A traces at low levels of 14B7) and so could most likely be neutralized more easily than more toxic variants tested by neutralization. Other single amino substitutions had less effect on affinity. N693A exhibited a K D value similar to that of wild type. The remaining singly substituted variants had K D values between 6-and 14-fold higher than wild type. This class of variants was at least partially neutralized by 14B7 (for variants that were toxic enough to be tested in that assay). Interestingly, substitution D683A falls into this category with a K D value 9-fold higher than wild type PA. Whereas the D683A substitution prevented toxicity (and hence neutralization testing), the biophysical measurements indicated that it did not greatly affect 14B7 binding affinity. When the binding constant of the double mutant N657A/N682A was compared with those of the single mutants, within the error limits of the experiments, no significant thermodynamic cooperativity of these residues for 14B7 binding was found.

DISCUSSION
The binding of PA to its cellular receptor is essential for anthrax toxin activity. One way to prevent anthrax toxicity is by blocking this step. In order to do so, it would be helpful to understand at the molecular level the residues of PA that interact with the cellular receptors and with antibodies that block PA cellular binding. For this study, we created a set of alanine-scanning mutants covering the small loop of PA domain 4 to elucidate amino acid residues important for binding to cellular receptors and to a neutralizing antibody, 14B7. These studies revealed that the PA domain 4 small loop residues most important for toxicity overlapped with, but were distinct from, the set of residues most important for 14B7 binding. This finding is illustrated in Fig. 6, where the substitution of residues shown in red (Tyr 681 and Asp 683 ) or, to a lesser extent, pink (Asn 657 ), created large defects primarily in cell binding and toxicity; substitutions of residues shown in green (Lys 684 , Leu 685 , Leu 687 , and Tyr 688 ) greatly decreased or eliminated 14B7 binding with milder effects on cell binding and toxicity; and substitutions of residues shown in yellow (Asn 682 and Pro 686 ) eliminated or greatly reduced binding to both cells and 14B7. It should be noted that the color coding in Fig. 6 reflects only the most extreme defects. For instance, the L687A variant did exhibit a reduction in binding and toxicity, but the reductions were not as great as those seen for the substitutions in residues Asp 683 , Tyr 681 , and Asn 657 . Additionally, the Y689A variant was neither fully toxic nor fully neutralized, but again the defects were less pronounced than those for substitutions of residues highlighted in Fig. 6.
The results from the 14B7 binding and neutralization assays were very interesting, because it was apparent from the biophysical measurements that with the exception of the N693A substitution, all of the singly substituted variants had at least a 5-fold weaker binding to 14B7 (Table I). However, only larger changes correlated with the loss of biological activity as seen by the neutralization assay. It may be that in addition to the large contributions made by the residues highlighted in Fig. 6, many residues in the small loop make small contributions to PA-14B7 interaction and that disrupting one such small contribution does not prevent productive binding. The additive nature of multiple changes in reducing binding affinities can be seen in the doubly substituted N657A/N682A variant. The singly substituted N657A variant exhibited an 8-fold reduction in binding affinity, and the singly substituted N682A variant showed a 75-fold reduction. The double N657A/N682A substitution caused a 200-fold loss in binding affinity (Table I). Although additive, however, these residues did not exhibit a significant cooperativity, suggesting that their free energy contribution to antibody binding is largely independent. The two triply substituted variants also had very low binding affinities for 14B7, but in this case it is difficult to determine whether any additional effect was seen over that of the single Leu 685 substitution, which alone completely disrupted binding.
It is apparent from our results that several small loop residues are critical for productive 14B7 binding and that other small loop residues also contribute to the binding affinity. It is possible that there are other residues involved in 14B7 binding that were not covered by our collection of PA variants. Previous studies localized the 14B7 epitope to a region between PA residues 671 and 721 (17), so our collection falls within this region but does not completely cover it. It is important to note that whereas some substitutions completely prevented neutralization by monoclonal antibody 14B7, all of the variants studied in this work were neutralized by polyclonal serum, indicating that the changes affected the ability of the 14B7 to block receptor binding but did not affect other PA epitopes contributing to protection by polyclonal serum.
The cell binding and toxicity assays implicated D683 as a critical residue in receptor binding. This finding is consistent with an interaction between the carboxyl group of Asp 683 and the metal coordination site of the anthrax receptors, TEM8 and CMG2. Recent studies in our laboratory have shown that the extracellular domain and a transmembrane domain or membrane anchor are necessary for PA activity, whereas the cytoplasmic tail is not essential (6). The extracellular portions of the receptors contain a von Willebrand factor type A domain with a metal ion-dependent adhesion site (MIDAS) motif. Thus, although the physiological functions of the receptors remain unknown, some things can be surmised about TEM8 ligand binding based on other proteins containing von Willebrand factor type A domains with MIDAS motifs. One such well studied class of proteins is the integrin family. The ␣ 2 subunit of ␣ 2 ␤ 1 integrin contains a domain (the I domain) homologous to von Willebrand factor type A domain and a MIDAS motif implicated in ligand binding (46). Crystallization and mutagenesis studies have indicated that the MIDAS motif is essential for binding to collagen (47,48). The crystal structure of a complex between the ␣ 2 I domain and a collagen peptide revealed that a collagen glutamate carboxylate oxygen formed a direct bond to the metal bound by the MIDAS motif (49). The authors suggested that such a metal bridge between the integrin and an acidic ligand residue would be a common feature of such interactions. A similar interaction has been reported as critical for another integrin-ligand interaction, that of the ␣L MIDAS motif and the carboxylate of a glutamate residue in its ligand, intercellular adhesion molecule-1 (50,51). Similarly, it is likely that the reason the D683A variant had the greatest defect in toxicity of all of the singly substituted variants was that a carboxylate oxygen of Asp 683 forms such a bridge in the TEM8-PA and CMG2-PA binding interactions.
Binding interactions between I domains and their ligands are affected by changes other than those in the MIDAS motif as well, and it is likely that other residues critical for toxicity could interact with other portions of TEM8 or CMG2. Additionally, some residues may act indirectly to orient the Asp 683 carboxylate into the MIDAS motif. Such an indirect affect may be the case for Asn 657 (Fig. 6, pink), which is partially buried in the PA structure and might not be very accessible for direct interaction with TEM8 or CMG2. As was the case for 14B7 interactions, the cell binding and toxicity effects of the N657A and N682A single substitutions were additive in the N657A/ N682A variant, which was completely nontoxic in our assays. The triple variants D683A/L685A/Y688C and D683A/L685E/ Y688C were also completely nontoxic in our assays. The single D683A substitution can account for the majority of the defect but not all of it (Fig. 1B). The Y688A substitution reproducibly caused a slight decrease in toxicity, and the additive effect of substituting both Asp 683 and Tyr 688 presumably caused the complete loss of toxicity.
Understanding the relevance of individual residues in PA to binding interactions with TEM8 or CMG2 and with neutralizing antibodies provides useful information for developing anthrax therapeutics. The residues identified in this work may provide targets for directed antitoxins. Additionally, the results of individual amino acid changes in PA disrupting neutralization by 14B7 imply that monoclonal antibody therapies should include a mixture of antibodies to different PA epitopes. Additionally, the PA variant panel used in this study could be useful in evaluating potential therapeutics directed against the binding domain of PA.