The Bordetella Adenylate Cyclase Repeat-in-Toxin (RTX) Domain Is Immunodominant and Elicits Neutralizing Antibodies*

Background: The protective antigen adenylate cyclase toxin (ACT) has not been included in current pertussis vaccines partly due to incomplete understanding of its protective epitopes. Results: The repeat-in-toxin (RTX) domain is immunodominant in mice and contains neutralizing epitopes. Conclusion: The RTX domain induces similar neutralizing antibody responses as ACT. Significance: The RTX domain may be an alternative to ACT for inclusion in future vaccines. The adenylate cyclase toxin (ACT) is a multifunctional virulence factor secreted by Bordetella species. Upon interaction of its C-terminal hemolysin moiety with the cell surface receptor αMβ2 integrin, the N-terminal cyclase domain translocates into the host cell cytosol where it rapidly generates supraphysiological cAMP concentrations, which inhibit host cell anti-bacterial activities. Although ACT has been shown to induce protective immunity in mice, it is not included in any current acellular pertussis vaccines due to protein stability issues and a poor understanding of its role as a protective antigen. Here, we aimed to determine whether any single domain could recapitulate the antibody responses induced by the holo-toxin and to characterize the dominant neutralizing antibody response. We first immunized mice with ACT and screened antibody phage display libraries for binding to purified ACT. The vast majority of unique antibodies identified bound the C-terminal repeat-in-toxin (RTX) domain. Representative antibodies binding two nonoverlapping, neutralizing epitopes in the RTX domain prevented ACT association with J774A.1 macrophages and soluble αMβ2 integrin, suggesting that these antibodies inhibit the ACT-receptor interaction. Sera from mice immunized with the RTX domain showed similar neutralizing activity as ACT-immunized mice, indicating that this domain induced an antibody response similar to that induced by ACT. These data demonstrate that RTX can elicit neutralizing antibodies and suggest it may present an alternative to ACT.

Whooping cough is a highly infectious disease caused primarily by the bacteria Bordetella pertussis. While disease incidence has dropped dramatically due to the initiation of widespread vaccination programs using killed bacteria in the 1940s, in recent years rates have rebounded, reaching a 60-year high in the United States in 2012 (1)(2)(3). This trend is especially trou-bling for unimmunized infants, who are most susceptible to the disease and exhibit the highest rates of morbidity and mortality. Modified vaccination strategies, including booster immunization of adolescents, adults, and pregnant women, have been implemented to reduce transmission to neonates.
This increase in disease incidence coincides with the switch from whole cell to acellular vaccines in the 1990s, and has been attributed to several factors, including increased awareness, mismatch between vaccine and circulating strains, a Th1/Th2 immune response instead of the more effective Th1 response, and a shorter duration of protection conferred by acellular vaccines (4). Recently, Warfel et al. (5) demonstrated that acellular vaccines protect against disease symptoms but not subclinical infection or transmission in a novel non-human primate model. Taken together, these data provide a compelling argument for modification of the current vaccine.
Currently licensed acellular vaccines contain chemically detoxified pertussis toxin and up to four surface adhesins, including filamentous hemagglutinin, pertactin, and fimbriae 2/3. Exciting approaches in development to enhance vaccinemediated protective immunity include a genetically attenuated B. pertussis for intranasal delivery (6), nanoparticle formulations, including purified antigens and novel adjuvant formulations (7), as well as inclusion of additional highly conserved antigens in the current vaccine (8,9). A strong candidate for inclusion in any of these is the adenylate cyclase toxin (ACT), 2 which aids in immune evasion and is produced by three closely related Bordetella species, including B. pertussis, Bordetella parapertussis, and Bordetella bronchiseptica (10,11).
ACT-deficient Bordetella strains have shown significantly compromised colonization and persistence in various mouse models (12)(13)(14), whereas some hypervirulent strains express higher ACT levels (15). Moreover, active or passive immunization with polyclonal anti-ACT antibodies protected mice against lethal respiratory challenges by B. pertussis and B. parapertussis (15) and shortened the period of bacterial colonization in the respiratory tract (16). Finally, natural infection of humans results in a strong anti-ACT antibody response (17).
ACT is a large ϳ177-kDa protein consisting of two functionally discrete regions as follows: the catalytic domain (residues 1-385) and a pore-forming or hemolysin region that is part of the larger repeat-in-toxin (RTX) family, represented in Ͼ250 bacterial strains (Fig. 1A). After translocation into the cytosol, the catalytic domain binds eukaryotic calmodulin with low nanomolar affinity (18) and rapidly converts available ATP to cAMP via its adenylate cyclase activity (19). The resulting supraphysiological cAMP levels disrupt signaling and bactericidal activities in phagocytic cells (20 -22). The C-terminal ϳ1300 residues exhibit homology to the Escherichia coli ␣-hemolysin. This region consists of a hydrophobic domain capable of forming a cation-selective transmembrane channel (residues 525-715) (23), a modification region bearing two acylation sites at residues Lys-860 and Lys-983 (24), the RTX domain (residues 1006 -1600), consisting of ϳ40 calcium binding sites formed by glycine-and aspartate-rich nonapeptide repeats, and finally, a C-terminal secretion signal (residues 1600 -1706). The RTX region also harbors the receptor-binding site, with specificity for the ␣ M ␤ 2 integrin (also called CR3, Mac-1, and CD11b/ CD18) present on phagocytic leukocytes (25,26). Both posttranslational acylation by the co-expressed enzyme CyaC and calcium ion-mediated structural changes are essential for receptor binding, cAMP intoxication, and pore forming activities (24,27).
Despite evidence indicating ACT is a protective antigen, few neutralizing antibodies have been described, and the location of neutralizing epitopes remains unclear. Moreover, ACT is prone to aggregation and degradation when produced by Bordetella or recombinantly by E. coli, precluding its inclusion in current acellular vaccine formulations (28). Therefore, we aimed to identify neutralizing antibodies and their domain specificity and to determine whether any single domain, possessing desirable expression and protein stability characteristics, can recapitulate the antibody responses induced by the holo-toxin.
To enhance folding and solubility, the hydrophobic domain, encompassing the region between the catalytic and RTX domains (residues 399 -1096), was cloned into pMalc-5x vector (New England Biolabs) between NdeI and BamHI sites, downstream of the maltose-binding protein (MBP). The primers for PCR were 5Ј-gggcgcaCATATGcgccaggattccggct-3Ј and 5Ј-atcggcGGATCCttaatggtgatgatggtgatgggcgctggcctcggaaggctggtgcac-3Ј; with the boldface nucleotides encoding a Cterminal His 6 -tag. The cyaC gene was inserted downstream of the hydrophobic domain between the BamHI and HindIII sites, with an upstream ribosome binding to allow for co-expression.
ACT and Domain Expression and Purification-Full-length ACT was expressed from the plasmid pT7CACT3 with co-expression of the palmitoylating enzyme CyaC in E. coli strain XL-1 Blue (29). The holo-toxin was purified using a single-step calmodulin-agarose affinity chromatography as described by Sebo and co-workers (29). Purified ACT was stored in 50 mM Tris, 8 M urea, 2 mM EDTA, pH 8.0, at 4°C for short term or Ϫ80°C for long term storage. The protein concentration was determined by absorbance at 280 nm using a molecular extinction coefficient of 143,590 M Ϫ1 cm Ϫ1 as calculated from its amino acid sequence (30). ACT from BEI Resources was used as a reference for purity and toxicity.
The catalytic and RTX domains of ACT were expressed in E. coli strain BL21(DE3). Briefly, 250 ml of TB media were inoculated from starter cultures to an absorbance at 600 nm (A 600 ) of 0.05, grown at 37°C until A 600 ϭ 0.3-0.6, at which time 0.4 mM isopropyl ␤-D-thiogalactopyranoside (IPTG) was added to induce expression. After 4 h of growth at room temperature, the cells were harvested, resuspended in Buffer A (50 mM Hepes, 250 mM NaCl, 2 mM CaCl 2 , 40 mM imidazole, pH 8.0), and lysed with a French press (Thermo Scientific). After a 20-min centrifugation step at 20,000 rpm (JA-20 rotor), the supernatant was applied to a HisTrap column on an ÅKTA FPLC (GE Healthcare), followed by elution with a linear gradient of Buffer B (Buffer A ϩ 500 mM imidazole). The hydrophobic domain was expressed in E. coli strain BL21 as above, and purification included immobilized metal affinity chromatography (IMAC) resin followed by an MBPTrap affinity column (GE Healthcare) and elution with 10 mM maltose.
Biophysical Characterization-To assess the oligomeric status of ACT and its domains, fractions eluted from HisTrap or IMAC were supplemented with either 100 l of 100 mM EGTA or 100 mM HBSC (HBS ϩ 2 mM CaCl 2 ), incubated on ice for 1 h, and loaded onto Superdex S200 column (except Superdex75 for CAT 400 ) equilibrated with HBS (50 mM Hepes, 150 mM NaCl, pH 7.8) or HBSC, respectively. Running buffers were HBS or HBSC to ensure the absence or presence of calcium ion. To assess secondary structure characteristics, purified ACT domains were dialyzed into 10 mM Tris-H 2 SO 4 , pH 8.0, and the concentrations were adjusted to 100 g/ml. To observe the effect of calcium ions on protein conformation, CaCl 2 were added to a final concentration of 2 mM and incubated at room temperature for 1 h. CD spectra between 180 and 260 nm were collected on a J-815 CD spectrometer (Jasco) at 25°C, at 1-nm intervals using a 1-mm rectangular cell. Each spectrum repre-sents the average of three scans subtracted with the spectrum of buffers (with or without CaCl 2 ). Data were fitted with the CDSSTR program on the DichroWeb server to estimate the percentage of secondary structures (31).
Murine Immunization-All protocols were approved by the University of Texas at Austin IACUC (protocol number 2012-00068), and all mice were handled in accordance with IACUC guidelines. As a source for antibody libraries, two 6-week-old BALB/c mice were primed intraperitoneally with 17 g of ACT (dialyzed against PBS to remove urea) in complete Freud's adjuvant. Four weeks later, the mice were bled through a tail vein and boosted subcutaneously with the same amount of PBSdialyzed ACT in incomplete Freund's adjuvant. Two weeks later, the mice were sacrificed, and blood was collected by cardiac puncture. Spleens were removed sterilely, sliced into pieces, and immediately immersed in 1 ml of cold RNAlater solution. After soaking overnight at 4°C, the solution was removed, and the spleens were stored at Ϫ80°C.
To assess the immunogenicity of individual ACT domains, 4 -6 BALB/c mice per group were immunized subcutaneously with equal moles of ACT and individual domains (10 g for ACT, 2.6 g for CAT 400 , 6.7 g for HP 1096 *, and 4.4 g for RTX 985 ) in complete Freund's adjuvant. Four weeks later, the mice were boosted subcutaneously with the same amount of antigen in incomplete Freud's adjuvant, a process that was repeated at 6 and 8 weeks. Blood was collected before immunization, 4 weeks after the first injection, and 2 weeks after each boost. Anti-ACT antibody titers were determined by ELISA (described below), with titer defined as the 50% effective concentration (EC 50 ) from a four-parameter logistic fitting to the ELISA data. Neutralization of ACT-induced cAMP intoxication of J774A.1 macrophage cells (ATCC number TIB-67) was tested with a 1:400 sera dilution.
Phage Display Antibody Library Construction-Total RNA was extracted from frozen spleens with TRIzol (Invitrogen) and the RNeasy mini kit (Qiagen) or PureLink RNA kit (Invitrogen) according to the manufacturers' instructions. The quality and concentration of total RNA were assessed by agarose gel electrophoresis and A 230:260:280 ratio (ϳ1:2:1 for pure RNA) measured by NanoDrop 2000 (Thermo Scientific). For first-strand cDNA synthesis, 5 g of total RNA was used. To maximize diversity, two separate reactions were performed using combinations of Superscript II ϩ d(T) 23 VN primer or Superscript III (Invitrogen)ϩ random hexamer (Thermo Scientific), following the manufacturers' instructions. The two sets of cDNA were pooled as template for amplification of the V L and V H repertoires using the primer sets and PCR conditions described by Krebber et al. (32). The PCR products were gel-purified, with 10 ng each of V L and V H used as template in an overlap PCR to generate V L -linker-V H fragments (scFv). This product was gelpurified and digested overnight with SfiI prior to directional ligation with similarly SfiI-digested pMopac24 vector (33). Ten individual electroporations were performed to transform XL1-Blue cells. The transformants were pooled, and an aliquot was 10-fold serially diluted and plated to count library size; the rest were plated on eight 150-mm 2ϫYT agar plates (10 g/ml tetracycline, 200 g/ml ampicillin, and 2% glucose). After incuba-tion overnight at 37°C, the bacterial lawns were scraped off in 2ϫYT medium and pooled to form the master library.
Phage Production, Purification, and Panning-Aliquots of the master library were used to inoculate 250 ml of 2ϫYT medium with 10 g/ml tetracycline, 200 g/ml ampicillin, and 2% glucose in 1-liter flasks to an A 600 of ϳ0.1. The cultures were grown at 37°C for 2-3 h until the A 600 reached ϳ0.6, induced, and rescued by adding 1 mM IPTG and M13KO7 helper phage (multiplicity of infection of ϳ20), incubated for 30 min without shaking at 37°C, and then returned to a shaking incubator at room temperature. Three hours after adding helper phage, the culture was supplemented with 50 g/ml kanamycin prior to overnight incubation with shaking. Phage were then purified by double precipitation with 0.2 volume of precipitation solution (2.5 M NaCl, 20% PEG-8000). The concentration of viable phage was assessed as colony-forming units (cfu), with serially diluted phage added to log-phase XL1-Blue cells, followed by plating on 2ϫYT agar plate with 200 g/ml ampicillin, and enumeration of colonies after overnight incubation.
Two rounds of panning were performed using ACT as bait. Eight ELISA plate wells (Costar) were coated with 50 l of 2 and 1 g/ml ACT in PBS at 4°C overnight for the first and second rounds, respectively. Input phage (100 l) were diluted into 900 l of 5% nonfat milk in PBST (PBS, 0.05% Tween 20) and incubated for 1 h before transferring 100 l to each of the 8 wells. After a 1-h incubation at room temperature, followed by five (or 10 for round 2) washes with PBST, bound phage were eluted with 100 l per well of 0.1 N HCl for 10 min at room temperature. The eluted phage was pooled and immediately neutralized with 48 l of 2 M Tris base. Half of the output phages was added into 5 ml of log-phase XL1-Blue culture grown with 10 g/ml tetracycline at 37°C to retain the F plasmid, incubated for 30 min without shaking and 1 h with shaking at 225 rpm in 37°C, spun down, and then plated on six 150-mm 2YT agar plates (200 g/ml ampicillin, 10 g/ml tetracycline, and 2% glucose). After overnight incubation at 37°C, colonies or lawn were scraped, pooled, and mixed thoroughly, and aliquots were used for second round of panning as described above.
Input and output phage titers (colony-forming units) were determined by infecting and plating E. coli as described above. Sequence diversity was monitored throughout all steps by performing colony PCR of random colonies on the phage titration plates, followed by BstNI fingerprinting and agarose gel electrophoresis. Clones with unique fingerprints were confirmed by DNA sequencing.
To produce monoclonal phage clones from panning outputs, single colonies from output plates were inoculated into sterile 96-well plates containing 100 l of 2YT medium (2% glucose, 200 g/ml ampicillin, and 10 g/ml tetracycline) and grown at 37°C overnight with shaking. The next morning, 10 l of the overnight culture was inoculated into another plate with 90 l per well of fresh medium containing 0.25% glucose and antibiotics, and grown at 37°C for 3 h, then 50 l of 2YT (200 g/ml ampicillin, 3 mM IPTG, and M13KO7 helper phage) was added and then shaken at room temperature for 3 h before adding 50 l of 2YT (200 g/ml ampicillin, 1 mM IPTG, 200 g/ml kanamycin). The plate was then shaken at room temperature overnight.
Antibody Expression and Purification-To convert phagedisplayed scFvs to soluble single chain antibody fragments (scAbs), consisting of a variable light chain domain (V L ) connected by a flexible (Gly 4 Ser) 2 to a variable heavy chain domain (V H ) and followed by a human constant domain to enhance expression and solubility, the scFv region was removed from pMopac24 phagemid vector by SfiI digestion and directionally ligated into SfiI-digested pMopac54 plasmid (34). For scAb production, 100 ml of TB supplemented with 200 g/ml ampicillin and 1% glucose were inoculated at A 600 ϭ 0.02 and grown overnight at room temperature. The next morning, cells were pelleted at 5000 ϫ g for 10 min at room temperature, resuspended in 100 ml of TB medium with ampicillin but no glucose, and grown at room temperature for 1 h before induction with 1 mM IPTG. After another 4 h, cells were harvested by centrifugation at 5000 ϫ g for 10 min at 4°C. Osmotic shock was performed as described (34). scAbs in the dialyzed shockates were purified by IMAC resin followed by size exclusion chromatography with a Superdex 200 column on FPLC (GE Healthcare). Protein concentrations were measured by BCA assays (Pierce) using a BSA standard with purity assessed by SDS-PAGE.
To convert scAbs into full-length IgG with enhanced stabilities and in vivo half-lives, the V L and V H genes were subcloned onto Ig-Abvec and IgG-Abvec vectors as described by Smith et al. (35). For IgG production, paired Ig-Abvec and IgG-Abvec plasmids were transiently transfected into CHO-K1 cells using Lipofectamine 2000 (Invitrogen) according to the manufacturer's instructions. Culture media were collected at 1-2-day intervals, neutralized with 1 Tris, pH 8.0, and pooled, and IgG was purified by ammonium sulfate precipitation followed by HiTrap protein A column. The purity and presence of aggregates were assessed by SDS-PAGE and size exclusion chromatography using a Superdex S200 column. The concentration was determined by A 280 using extinction coefficients calculated from deduced amino acid sequences (30).
Analysis of Antibody Binding by ELISA-For monoclonal phage screening, the 96-well phage production plates described above were spun at 3000 ϫ g for 20 min with 40 l of supernatant transferred to a coated (2 g/ml ACT in PBS) and blocked (5% nonfat milk in PBST, M-PBST) ELISA plate, containing 60 l of M-PBST per well, and incubated at room temperature for 1 h. After four washes with PBST, 50 l of 1:2000 HRP-conjugated anti-M13 antibody (GE Healthcare) was added and incubated for1 h at room temperature. The plate was washed four times, and 50 l/well 3,3Ј,5,5Ј-tetramethylbenzidine substrate was added and incubated at room temperature. The reaction was quenched by adding 1 M HCl to the 50 l/well, and the absorbance at 450 nm recorded with a SpectraMax M5 (Molecular Devices). Wells with absorbance higher than 2-fold of background were identified for further characterization.
Binding assays to assess domain specificity and relative affinity of soluble scAb or IgG proteins were performed in a similar manner. Purified ACT (1 g/ml) or domains (equimolar with ACT) were coated on ELISA plates, followed by blocking and serial dilutions of purified antibodies, and finally detection with goat anti-mouse IgG HRP-conjugated antibody (to detect murine antibodies), goat anti-human chain HRP-conjugated antibody (to detect scAbs), or goat anti-human IgG (Fc-spe-cific) HRP-conjugated antibody (to detect recombinant IgG). For competition ELISA, the antibody of interest was used at a fixed concentration determined to yield 70 -80% of the maximal signal, and mixed with an equal volume of serially diluted competitor antibody, followed by detection as above.
To assess the reactivity of sera from humans exposed to B. pertussis, purified ACT (1 g/ml) or domains (equimolar concentrations as ACT) were coated on ELISA plates. Plates were blocked as above; human sera were serially diluted in M-PBST, and bound antibodies were detected with goat anti-human IgG (Fc-specific) HRP-conjugated antibody.
Nine randomly selected samples were tested in duplicate. The absorbance value for each sample binding to a domain was normalized to that sample's signal on an ACT-coated well at a 100-fold dilution (in the linear dose-response range) as follows: (A 450 (domain coated well) Ϫ A 450(uncoated well) )/ (A 450 (ACT coated well) Ϫ A 450(uncoated well) ). Human sera were obtained from Vanderbilt University Medical Center under a protocol approved by the local institutional review board (IRB 061262, 070258, and 090806). Use of the samples was approved by the University of Texas at Austin (2009-05-0096). The study was conducted in accordance with the Declaration of Helsinki, with written informed consent obtained from each participant prior to study entry. The original consent forms allowed for sample use in subsequent studies. CaCl 2 levels in all ELISAs were maintained at Ͼ2 mM unless indicated, and each assay was performed at least twice.
In Vitro cAMP Intoxication and Neutralization Assay-J774A.1 cells were grown in DMEM (Sigma) supplemented with 10% fetal bovine serum, 1 mM sodium pyruvate, and penicillin/streptomycin. To measure the cAMP intoxication of J774A.1 cells by ACT or antibody neutralization, J774A.1 cells were seeded at 4 ϫ 10 4 /cm 2 in 24-well plates (Costar) 1 day before the assay. ACT alone or with antibodies was diluted in DMEM without supplements in a final volume of 1 ml; base DMEM contains 1.8 mM CaCl 2 . ACT was used at 125 ng/ml in all assays unless otherwise specified; antibodies were present at 160-fold molar excess. ACT and antibody mixtures were incubated at room temperature for 30 min, and cells were washed twice with plain DMEM prior to addition of 480 l of antibody/ ACT solution to duplicate wells. The plate was incubated at 37°C for 30 min in a CO 2 incubator, followed by two washes with cold PBS. Lysis solution (500 l; 0.1 N HCl, 0.1% Triton X-100) was added into each well, and the plate was rocked on ice for 10 min. The lysates were transferred into 1.5-ml tubes and boiled for 10 min to inactivate ACT and cAMP-hydrolyzing enzymes. To assess cAMP intoxication in CHO cells, ACT was used at 250 ng/ml, with all other assay conditions held constant.
The resulting cellular lysates were clarified by centrifugation at 13,000 ϫ g for 5 min. The supernatant was diluted 6-fold with 200 mM Hepes, 150 mM NaCl, 0.05% Tween, pH 8.0, prior to cAMP measurement using a competition ELISA as described by Karimova and Ladant (36). All assays were performed at least in duplicate, with cAMP concentrations normalized to total protein concentration in the lysates as measured by BCA assay (Pierce). FEBRUARY 6, 2015 • VOLUME 290 • NUMBER 6

JOURNAL OF BIOLOGICAL CHEMISTRY 3579
To evaluate ACT neutralization in the context of the whole bacterium, B. pertussis Tohama I was grown on a Bordet Gengou agar plate supplemented with 15% defibrinated sheep blood (BD Biosciences) at 37°C for 4 to 5 days. Bacteria were then inoculated into modified synthetic Stainer-Scholte medium and grown at 37°C with shaking at 225 rpm for 20 -24 h to an A 600 of 0.7-1.0. Bacteria were pelleted by centrifugation at 5000 ϫ g for 10 min, resuspended in PBS, and diluted to an A 600 of 0.4 in DMEM ϩ 10% heat-inactivated FBS, and then mixed with an equal volume of serially diluted neutralizing or control IgGs in DMEM ϩ 10% heat-inactivated FBS, and incubated at room temperature for 20 min before addition to adherent J774A.1 cells and incubation of 1 h at 37°C. Intracellular cAMP level was then measured as above.
J774A.1 Cell Lysis Assay-Cell lysis was monitored by enzymatic activity of lactate dehydrogenase released into the medium upon cell lysis. J774A.1 cells were seeded 10 5 cells/well in 96-well round-bottom plates 1 day before the assay. 100 l each of ACT alone (0.25 g/ml) or ACT preincubated with scAbs (10 g/ml, 160-fold molar excess) in plain DMEM were added to triplicate wells and incubated for 2 h at 37°C. The plate was then centrifuged at 250 ϫ g for 5 min, and lactate dehydrogenase activity in the supernatants was measured by a colorimetric assay with the CytoTox 96 kit (Promega, WI). After subtracting the background signal from control wells without ACT, the sample absorbance at 490 nm was normalized to ACT-only controls as follows: Analysis of ACT-Integrin Binding by ELISA-Recombinant murine ␣ M ␤ 2 integrin (R&D Systems) was coated at 1 g/ml in PBS at 4°C overnight and blocked with M-PBST. ACT or purified domains were serially diluted and incubated for 1 h at room temperature, followed by detection with polyclonal rabbit anti-ACT antibody and HRP-conjugated goat anti-rabbit antibody. To assess the effect of antibodies on ACT binding to integrin, ACT (1 g/ml) was mixed with an equal volume of serially diluted antibody (5-, 1.6-, 0.5-, and 0-fold molar excess) and incubated for 1 h at room temperature before transferring to a blocked ␣ M ␤ 2 ELISA plate. The bound ACT was detected as described above.
Flow Cytometry Analysis of ACT Binding to Cells-ACT was dialyzed against HBSC (50 mM Hepes, 150 mM NaCl, 2 mM CaCl 2 , pH 8.0) to remove urea and then biotinylated with 100fold molar excess of EZ-Link Sulfo-NHS-LC-Biotin (Thermo Scientific) at room temperature for 2-3 h before quenching with 1 M Tris, pH 8.0, and dialysis against HBSC overnight in 4°C. Biotinylated ACT (210 l of 0.8 g/ml) was incubated with an equal volume containing 120 g/ml purified scAb (M2B10, M1H5, M1F11, and M1C12) in DMEM ϩ 1% BSA at room temperature for 30 min. Then 200 l of the incubated mixtures was added to 4 ϫ 10 5 washed J774A.1 cells in duplicate and incubated on ice for 30 min to allow ACT binding but not internalization. After two washes with FACS buffer (HBSC ϩ 2% FBS), 200 l of 1:500 phycoerythrin (PE)-conjugated streptavidin (BioLegend) was used to detect cell-associated biotinylated ACT. After 20 min of incubation on ice and three washes, the cells were finally resuspended in 600 l of FACS buffer and analyzed on an LSR Fortessa II. Data analysis was performed with Flowjo software (Version 10).

RESULTS
ACT Is Prone to Aggregation and Proteolysis-The native ACT holotoxin secreted from B. pertussis readily aggregates at the bacterial surface and is prone to proteolytic degradation (28), while recombinant toxin expressed in the E. coli cytoplasm forms inclusion bodies (37)(38)(39). To increase the yield of purified protein, 8 M urea was used to extract the aggregated protein from bacterial cell pellets; even so, the protein is highly susceptible to proteolysis with early reports observing enzymatic activity in 43-and 45-kDa fragments (40). Efforts to remove urea, such as dialysis and dilution, result in significant aggregation and fragmentation (41).
As a result, standard purification protocols include solubilization of the cell pellet with urea, followed by calmodulin affinity or sequential anionic and hydrophobic interaction chromatographic steps followed by storage in 8 M urea. Assays using the toxin call for dialysis or dilution of urea-solubilized ACT into assay media immediately before experimentation, most likely resulting in an ensemble of fully and partially folded ACT molecules and uncertainty regarding the exact concentration of active toxin molecules. ACT purified in our laboratory exhibited similar in vitro activity when incubated with J774A.1 cells as described by Eby et al. (42): 125 ng/ml produced 10,000-ϳ40,000 pmol of cAMP per mg of total cellular protein in 30 min at 37°C.
However, dialysis or dilution of ACT into buffer without urea led to protein that was poorly behaved and retained by the SEC column (Ͻ5% of protein applied to the column was eluted). In contrast, overnight dialysis into HBSC containing 1 M urea resulted in high molecular weight aggregates that eluted off the SEC column at a size corresponding to ϳ600 kDa. An alternative refolding approach, rapid 10-fold dilution into HBSC (final urea concentration 0.8 M) followed immediately by SEC, yielded a broader aggregate peak and a smaller peak corresponding to the expected size for monomer (Fig. 1B). The different monomer yields may reflect a time-dependent aggregation process or the presence of folding intermediates with different aggregation propensities in the two refolding procedures. A recent report reiterated the challenges of refolding ACT (43).
Individual ACT Domains Are Biophysically Superior to ACT-To identify which, if any, ACT domains are predominantly recognized by polyclonal antibody responses, we expressed individual domains in E. coli with affinity tags to facilitate purification (Table 1). Based on prior reports (18, 29, 44 -46), the N-terminal catalytic domains (residues 1-373 (CAT 373 ), 1-385 (CAT 385 ), and 1-400 (CAT 400 )) and C-terminal RTX domains (residues 751-1706 (RTX 751 ) and 985-1706 (RTX 985 )) were cloned into the pET28a vector for cytoplasmic expression with N-terminal His 6 tags to facilitate purification. In our hands, RTX(482-1706) was poorly soluble and purified inefficiently; instead we selected RTX 985 as the largest fragment to exclude both acylation sites but retaining the N terminus before the first Gly-Asp rich repeat. To enhance solubility, the hydrophobic domain (residues 399 -1096 (HP 1096 ), encompassing the region between the catalytic and RTX domains) was fused down-stream of MBP, with a C-terminal His 6 tag and di-cistronic expression of the specific acylating enzyme CyaC (indicated by *). After cytoplasmic expression of each construct and cell lysis, a one-step affinity chromatography with a HisTrap column yielded ϳ5-80 mg of protein per liter of culture with Ͼ90% purity as determined by SDS-PAGE (Fig. 1C). For HP-MBP fusion proteins, a second chromatographic step with an MBP-Trap column was required to reach a similar level of purity, although purity, proteolysis, and solubility issues persisted.
To determine whether the purified domains exhibited native-like structure and expected calcium-dependent structural changes, SEC was used to assess the oligomeric state, and circular dichroism (CD) spectroscopy was used to assess the secondary structure content. The catalytic domains eluted as a single peak of expected size (40 kDa) with the estimated composition of secondary structures (56% helix, 14% strands, 13% turns, and 17% unordered) similar to that determined by x-ray crystallography ( Fig. 2A) (18). Although the catalytic domain formally encompasses residues 1-373, the construct encompassing residues 1-400 was selected for further use to include the neutralizing epitope recognized by antibody 3D1 (47).
The hydrophobic fusion proteins eluted as broad aggregate peaks when acylated or nonacylated. In the absence of acylation, multiple smaller peaks were observed, suggesting that acylation may stabilize folding of this domain and protect against proteolysis. In both cases, the CD spectra were not characteristic of unfolded or aggregated proteins (Fig. 2B). The solubility and CD spectra of these constructs may be dominated by the MBP fusion partner, but we did not attempt to remove it, as a shorter hydrophobic domain was reported to further aggregate under these circumstances (48).
The RTX domain includes ϳ40 calcium binding Gly-Asp repeats, grouped into five blocks, and separated by non-RTX flanking regions (Fig. 1A). Structural data for RTX-containing proteins suggest the repeats fold into parallel ␤-helix structures. In the presence of calcium ions, the protein converts from an intrinsically disordered domain into a compact ␤-roll structure with an altered CD spectrum and a reduced hydrodynamic radius that appears to be further stabilized by acylation (43,49,50). RTX 985 exhibited a shift from largely monomer in the presence of calcium (78 kDa) to a mixture of oligomers (ϳ380 kDa) upon the addition of EGTA to chelate calcium ions. These structural changes are captured by CD, which shows a more ordered state in the presence of calcium ions (Fig. 2C), consistent with that observed with a similar construct also lacking the acylation sites (residues 1006 -1706) (51). SDS-PAGE indicates the two peaks observed with calcium have the same molecular weight suggesting that RTX 985 forms two stable states with different hydrodynamic radii (Fig. 2C).
Theorizing that these two forms are a consequence of the missing acylation sites, we generated a larger construct to include both sites. RTX 751 expressed without CyaC eluted as a single peak of expected size (ϳ110 kDa) in the presence or absence of calcium. When co-expressed with CyaC, presumably resulting in acylation at residues Lys-860 and Lys-983,

TABLE 1 Biophysical analysis of ACT constructs
ACT and various derivatives were purified and subjected to size exclusion chromatography in the presence of 2 mM calcium or an excess of EGTA to chelate free calcium ions. The expected molecular mass for each construct is noted, as is the observed size and thermal melting temperature of the major peak. Constructs noted with * were co-expressed with CyaC to acylate residues Lys-860 and Lys-983. ND means not determined. RTX 751 * exhibited a calcium-dependent conversion from a compact monomer (ϳ90 kDa) to a soluble higher molecular weight aggregate (ϳ600 kDa) after depletion of calcium (Fig.  2D). The catalytic and RTX domains retain much of the expected structural behavior, with RTX 751 * appearing to better stabilize the monomeric form than RTX 985 . Similar to previous reports on ACT behavior (26, 27, 51-53), the presence of calcium and acylation appears to stabilize the RTX monomers. Anecdotally, the RTX domains were stable for at least 6 months at 4°C with minimal aggregation or degradation (RTX 751 and RTX 751 * were slightly more stable than RTX 985 ), as measured by SDS-PAGE and SEC to assess the monomeric fraction. Although the CAT 400 domain also remains monomeric, it starts to degrade after 3 months under the same conditions, as measured by SDS-PAGE. The compact ␤-roll structure RTX domains adopt in the FIGURE 2. ACT domain oligomeric state and secondary structure. Purified domains were separated by size exclusion chromatography (Superdex200 column, except Superdex75 for CAT 400 ), with far UV circular dichroism spectra (Jasco J-815) used to assess secondary structure in the presence of 2 mM CaCl 2 or the absence of calcium ions. A, catalytic domain, spanning residues 1-400 (CAT 400 ), eluted as a single peak of expected size. The secondary structure is similar to that observed in the CAT 373 crystal structure (56% helix, 14% strands, 13% turns, and 17% unordered). B, HP domain, spanning residues 399 -1096, with an N-terminal MBP fusion protein eluted off SEC as high molecular weight aggregates whether acylated (*) or nonacylated. C, RTX 985 formed high molecular weight aggregates in the absence of calcium ions but eluted as two peaks, one corresponding to the expected molecular mass of 78 kDa in the presence of calcium. Circular dichroism revealed significant conformational change upon addition of 2 mM CaCl 2 corresponding to an increase in ␤-strand content. D, RTX 751 domain exhibits a similar calcium-dependent delay in elution volume, which is more pronounced when the protein is acylated (*), and under these conditions it yields a single monomer peak. Inset, SDS-polyacrylamide gels show proteins present in the indicated peaks, with arrows indicating the expected monomer size.

Domain
presence of calcium may contribute to their overall higher melting temperature and resistance to proteolysis than CAT 400 (Table 1).
ACT Domains Are Biochemically Similar to ACT-To determine whether our domain constructs retain structural elements present in ACT, we screened a panel of nine previously characterized monoclonal antibodies for binding to ACT and individual domains by ELISA (47). All nine antibodies tested recognized only the expected domain and did not distinguish between acylated and nonacylated domains (Table 2), supporting the notion that the domains are properly folded. One exception is 2B12, whose epitope includes residues 888 -1006, but did not recognize HP 1096 . This may be due to incomplete folding of HP 1096 or the binding site may require additional residues distal to residue 1006 not present in this construct.
As the RTX domain harbors the receptor-binding site between residues 1166 and 1281 (26), we assessed the ability of our RTX constructs to bind purified murine extracellular ␣ M ␤ 2 receptor, a known ACT cell surface receptor (25). Although ACT and RTX 751 * both bound the murine receptor when acylated and in the presence of calcium (apparent EC 50 ϳ20 nM; Fig. 3), ACT exhibited considerable nonspecific binding to wells without the receptor. This is similar to the sticky behavior observed when ACT without urea was applied to and retained by the SEC column and likely reflects the presence of solventexposed hydrophobic patches in misfolded ACT molecules. Monomeric RTX 985 did not bind the ␣ M ␤ 2 receptor, consistent with prior studies showing that post-translational acylation is essential for receptor binding (29,54). In summary, the individual domains were readily purified, with yields of CAT 400 at ϳ80 mg/liter culture, nonacylated RTX and HP domains at ϳ5 mg/liter culture, and the acylated domains at Ͻ2 mg/liter culture. The CAT and RTX domains share many structural features with ACT, although the HP domain is mostly aggregated.
Anti-ACT Antibodies Primarily Recognize RTX-To determine whether a single ACT domain dominates the immune response, we used phage display to analyze murine antibody repertoires after ACT immunization. Two mice were immunized intraperitoneally with 17 g of ACT in complete Freund's adjuvant and were boosted subcutaneously once with incomplete Freund's adjuvant. The resulting sera neutralized the toxic activities of ACT at a 1:400 dilution in an in vitro cAMP intoxication assay (data not shown). The spleens were harvested, and each was used to construct an antibody phage display library, each containing ϳ10 7 total transformants. After two rounds of panning against ACT, 90 individual clones from each library were grown in 96-well plates and assessed for ACT binding by phage ELISA. Of these, 57 and 60 clones from the two libraries respectively yielded signals 2-fold above background, with 29 and 21 expressing unique sequences, as determined by BstNI fingerprinting and DNA sequencing.
To determine the domain specificities of unique antibodies, monoclonal phage were assessed for binding to individual domains in ELISA (Fig. 4A). Few antibodies bound the catalytic domain; none bound the hydrophobic domain, whereas the rest (27 of 29 and 20 of 21, respectively) bound the RTX 985 domain. This observed high frequency of RTX 985 -specific antibodies concurs with previous reports by Lee et al. (47) that the majority of antibodies discovered using hybridomas recognized the RTX domain and by Betsou et al. (29) that this domain may be immunodominant.
Anti-RTX Antibodies Can Neutralize ACT-Next, we screened unique antibodies identified from phage libraries for the ability to neutralize ACT activities using an in vitro cAMP assay. There are several steps during ACT intoxication of cells that are susceptible to antibody-mediated neutralization, including receptor binding, membrane insertion, and translocation, and interference with any of these will be reflected in decreased intracellular cAMP accumulation in or reduced lysis of target cells. For this assay, we employed the murine macrophage cell line J774A.1 bearing the ␣ M ␤ 2 integrin.
For the neutralization assay, antibodies were expressed as recombinant scAbs, composed of the variable light chain (V L ) joined to a flexible (Gly 4 Ser) 2 linker and the variable heavy chain (V H ), followed by a C-terminal human chain constant region to increase solubility and serve as a detection handle (9). Based on multiple sequence alignments, 31 antibodies with FIGURE 3. ACT and RTX domains bind purified ␣ M ␤ 2 receptor. Soluble murine ␣ M ␤ 2 receptor was coated onto ELISA plates and blocked, and ACT or domains were serially diluted in M-PBST. Bound protein was detected with polyclonal rabbit anti-ACT antibody and goat anti-rabbit HRP. To assess nonspecific binding, control wells were not coated with ␣ M ␤ 2 receptor but blocked with M-PBST only. A, RTX 751 *, and B, acylated ACT* showed receptordependent binding, although ACT also exhibited significant nonspecific binding. All other domains showed no specific or nonspecific binding.

TABLE 2 Biochemical analysis of ACT constructs
Binding of previously characterized anti-ACT murine antibodies (47) to ACT and various derivatives described in this work in ELISAs. ACT constructs were coated on ELISA plates, and thee antibodies were titrated and detected with anti-mouse IgG-HRP. unique complementary determining regions were selected for scAb expression (Fig. 4B). Ten scAbs either expressed poorly (Ͻ200 g/liter culture) or bound ACT weakly (concentration Ͼ238 nM required for saturation) and were not tested further.
To identify antibodies neutralizing ACT function, ACT and individual scAbs were incubated at a 1:160 molar ratio before addition to adherent J774A.1 cells and determination of intracellular cAMP levels by competition ELISA. Of the 21 scAbs tested, nine reduced the cAMP level by more than 90%, as compared with cells treated with ACT alone, which we consider highly neutralizing in this assay (Fig. 4C). We also determined the ability of these scAbs to rescue J774A.1 cells from lysis using a lactate dehydrogenase release assay, observing a strong correlation with cAMP neutralization (Fig. 4D). This is in agreement with findings by Basler et al. (55) that intracellular ATP depletion is sufficient to promote cell lysis. Notably, all neutralizing antibodies identified recognize the RTX domain.
Two Novel Neutralizing Epitopes in the RTX Domain-We next sought to classify neutralizing antibodies based on their recognition of unique or overlapping epitopes. Here, we used a competitive binding ELISA, in which a single phage-displayed antibody was mixed with buffer or a second antibody in the scAb format, added to an ELISA well coated with ACT, and followed by detection of bound phage remaining in the well. Reduced signal in the presence of a second antibody compared with phage antibody alone indicates competition between the two antibodies for the same or overlapping epitopes. Using this approach, the nine neutralizing antibodies were divided into two groups binding nonoverlapping epitopes, although nonneutralizing antibodies recognized four unique epitopes, for a total of six epitopes represented in this study (Fig. 4A).
One representative antibody binding each neutralizing epitope was selected based on sequence uniqueness, expression level, and binding affinity for conversion into a full-length chimeric immunoglobulin, with human IgG1 and constant domains (Fig. 5A). ELISAs with the two RTX constructs helped to further define the epitopes recognized by these two antibodies, named M1H5 and M2B10. Both antibodies bound RTX 751 * with almost the identical affinity as full-length ACT (Fig. 5B), whereas M1H5 bound the shorter RTX 985 domain weakly (Fig.  5C), suggesting that its epitope is not fully contained or properly presented in this construct.
To determine whether the antibodies identified here bind epitopes overlapping with those of previously defined murine monoclonal antibodies (47), we performed a second set of competition ELISAs (Fig. 5D and data not shown). None of the murine antibodies competed with M2B10 or M1H5 for ACT binding, demonstrating that these antibodies recognize previously undescribed neutralizing epitopes in the RTX region.
Together, we observed antibodies binding six nonoverlapping epitopes on RTX, two neutralizing and four non-neutralizing, and all but two required the presence of calcium (Fig. 4A). A representative antibody binding one epitope, M1C12, competes with antibody 1H6, one of the four non-neutralizing RTX binding antibodies discovered by Lee et al. (47). Their remaining three anti-RTX antibodies bind unique epitopes, suggesting a total of at least nine distinct epitopes in the RTX domain. Identification of new epitopes is not unexpected, because neither hybridoma nor phage display technology is exhaustive, and each has trade-offs; the former is low throughput and laborintensive, and the latter does not preserve the pairing between the light and heavy chains and preferentially selects for antibod- . ACT immunization induces a diverse antibody response. A, phylogenetic tree depicting antibody sequence relatedness was generated using the light and heavy variable region amino acid sequences. Neutralizing scAbs are colored gray, with unique shapes denoting recognition of distinct epitopes among this antibody group as determined by competition ELISA. Open symbols denote antibodies whose binding does not depend on the presence of calcium. Antibodies competing with previously characterized monoclonal antibodies are indicated; all antibodies bind RTX except M1C5, M1F11, and M2G5, which bind CAT 400 . B, representative SDS-PAGE of scAbs after purification by IMAC and Superdex S200. Arrow indicates expected size of ϳ40 kDa, 2 g each of M1F11, M1C12, M1H5, and M2B10 scAbs were loaded. C, 21 unique scAbs identified from the immune phage libraries were tested for the ability to neutralize ACT-mediated increases in intracellular cAMP concentration. ACT was incubated with a 160-fold molar excess of scAb protein before adding to J774A.1 cells. Data are reported as the percent relative cAMP, calculated from the total cAMP concentration in the cellular lysate as determined by cAMP ELISA, divided by the protein concentration of the lysate, and normalized to control cells treated only with ACT (open bar). Error bars indicate range of duplicate assays. D, 21 scAbs were evaluated for their ability to rescue J774A.1 macrophages from ACT-induced lysis, using a similar protocol as for cAMP neutralization. Cell lysis was measured via lactate dehydrogenase release using the Cytotox 96 kit (Promega), normalized to control cells treated only with ACT (empty bar), and reported as the percent relative lysis. Error bars indicate standard deviation of triplicate assays.
ies with high bacterial expression levels. Even recently described repertoire mining approaches based on high throughput sequencing of antibodies from individual B cells do not identify the same sequences as phage display (56).
Antibodies Binding Novel RTX-neutralizing Epitopes Disrupt ACT-␣ M ␤ 2 Integrin Binding-ACT primarily targets cells bearing the ␣ M ␤ 2 receptor under conditions in which the RTX domain assumes a receptor-binding competent conformation mediated by the presence of calcium ions and post-translational acylation (26). Combining this with our observation that the M2B10 and M1H5 antibodies showed a much weaker neutralizing effect at the same 160-fold molar excess when CHO-K1 cells lacking this receptor were used than when J774A.1 cells expressing the receptor were used (Fig. 6A), we hypothesized that these two antibodies act by blocking the interaction between ACT and the ␣ M ␤ 2 integrin.
To test this hypothesis, we used flow cytometry to monitor ACT bound to J774A.1 cells in the presence or absence of the M2B10 and M1H5 scAbs. Biotinylated ACT was incubated with a 300-fold molar excess of neutralizing or non-neutralizing scAbs, added to J774A.1 cells and detected with phycoerythrin-conjugated streptavidin by FACS. The M2B10 and M1H5 antibodies significantly reduced binding of ACT-biotin to J774A.1 cells, whereas two non-neutralizing control scAbs (M1F11 and M1C12) had no significant effect (Fig. 6B). This difference was not due to affinity, as all four scAbs have similar affinities (EC 50 ϭ 0.3-0.7 nM; data not shown).
To further confirm that the diminished binding of ACT to J774A.1 cells was due to interference with a specific receptor, a competition ELISA with soluble ␣ M ␤ 2 integrin was performed. ACT (0.5 g/ml) was incubated with the M2B10, M1H5, M1F11, or 3D1 antibodies in molar ratios ranging from 5 to 0.5 and transferred to an ␣ M ␤ 2 integrin-coated plate. The result was consistent with the flow cytometry assay; M2B10 and M1H5 reduced ACT binding to immobilized ␣ M ␤ 2 integrin in a dose-dependent manner, Ͼ90% at a 2-fold molar excess (Fig. 6C). In contrast, 3D1, a neutralizing IgG that blocks translocation of the catalytic domain (57), and 2A12, a neutralizing antibody with unclear mode-ofaction, had no effect. Minimal nonspecific binding was observed under these assay conditions.
To determine whether ACT neutralization occurs in the context of the whole bacterium, we repeated this assay with live B. pertussis instead of purified ACT. According to Gray et al. (28), newly synthesized ACT is responsible for intoxication. Therefore, B. pertussis was washed in PBS to remove any secreted ACT. Bacteria (A 600 ϭ 0.2) added to J774A.1 cells resulted in cAMP levels similar to that induced by 125 ng/ml purified ACT. When the M2B10 and M1H5 but not the nonneutralizing M1F11 or 7C7 (47) antibodies were added with the bacteria, they resulted in dose-dependent reduction of cAMP levels (Fig. 6D). This suggests these antibodies may be able to neutralize ACT in the context of active infection. Microtiter plates were coated with ACT or ACT domains at equimolar concentrations, followed by serial dilution of antibody from 1 nM, and followed by detection with anti-human Fc antibody-HRP conjugate. D, competition ELISA determined that the M2B10 and M1H5 antibodies bind novel nonoverlapping epitopes. A 200-fold molar excess (20 nM) of previously described murine mAbs (3D1, 2A12, 10A1, 2B12, 9D4, 6E1, 7C7, and 1H6) (47) or scAb versions of M1H5 and M2B10 were mixed with M2B10 and M1H5 IgG (0.1 nM) and incubated on ACT-coated ELISA plate, with bound M2B10 or M1H5 detected as above. The absorbance was normalized to that of M2B10 or M1H5 with no competitor; absorbance significantly Ͻ1.0 indicates competition between the antibody pair. FEBRUARY 6, 2015 • VOLUME 290 • NUMBER 6

RTX Domain Is Immunodominant and Elicits Neutralizing
Antibodies-To determine whether any ACT domain dominates the immune response, we immunized mice with ACT and tested the resulting sera 4 weeks after primary immunization for binding to different ACT domains. Interestingly, strong responses were observed for ACT, RTX 985 , and RTX 751 , but no responses were observed for the CAT 400 or HP 1096 * domains (Fig. 7A).
To determine whether this was the result of RTX immunodominance or a lack of immunogenicity by the CAT and HP domains, we immunized additional groups of mice with CAT 400 , HP 1096 *, or RTX 985 . Here, RTX 985 was selected because it shares many features with RTX 751 *, including recognition by at least one neutralizing antibody, yet it lacks the acylation sites rendering it simpler to produce and less likely to engage the native receptor. The calcium concentration in extracellular fluid in the body is 2.2-2.7 mM (59), which is sufficient to support bacterial secretion and folding of ACT during an infection and is expected to support proper folding during immunization.
Mice immunized with ACT or RTX 985 showed high anti-ACT titers 4 weeks after the first injection, which increased after boosting (weeks 6 and 8, Fig. 7B). On the contrary, the catalytic domain was much less immunogenic, reaching a detectable anti-ACT titer of ϳ1500 only after two boosts (week 8; Fig. 7B). This weak response may reflect structural similari-ties between CAT and eukaryotic adenylate cyclases, resulting in immunological tolerance or an evolutionary mechanism to protect key toxin components from neutralizing antibodies (60,61). Only one of the four mice immunized with the hydrophobic domain reacted with ACT, supporting the SEC data that this construct is poorly folded (Fig. 7B). Antibody responses in the three remaining mice were directed toward the MBP fusion, as determined by ELISA (data not shown).
Next, we wanted to determine which domains induced sera best able to neutralize ACT cAMP intoxication activities in vitro with J774A.1 cells. Here, only immunization with RTX 985 elicited sera able to protect cells to a similar extent as sera elicited by full-length ACT (Fig. 7C). Although mechanisms other than receptor blockade may contribute to neutralization, the presence of M2B10-and M1H5-like antibodies in the sera of ACT-or RTX-immunized animals was confirmed by competition ELISA (Fig. 8). Because RTX 985 binds antibody M1H5 weakly, this domain may have sufficient conformational similarities as ACT to induce antibodies binding overlapping but nonidentical epitopes as M1H5.
Finally, to determine whether humans show a similar bias toward RTX recognition, we tested nine serum samples from humans exposed to B. pertussis, selected randomly from a larger collection (62). All nine sera recognized RTX 751 at a similar level as full-length ACT, whereas sera from only one individual bound CAT 400 (Fig. 9). Taken together, these data  Fig. 4. B, antibody blockade of ACT binding to J774A.1 cells assessed by FACS. Biotinylated ACT was incubated with a 300-fold molar excess of scAb and added to 4 ϫ 10 5 J774A.1 suspension cells on ice. After washing, bound biotinylated ACT was detected with streptavidin-PE and analyzed by FACS (mean fluorescence noted next to each peak). Controls include untreated cells (Cells only) and cells treated with nonbiotinylated ACT followed by streptavidin-PE (ACT). C, antibody blockade of ACT binding to soluble ␣ M ␤ 2 integrin by ELISA. ACT (0.5 g/ml) was incubated with serial dilutions of M2B10, M1H5, 3D1 and 2A12 antibodies at 5-, 1.6-, and 0.5-fold molar excess, before transfer to an ELISA plate coated with murine ␣ M ␤ 2 integrin. Bound ACT was detected with rabbit anti-ACT polyclonal antibody followed by HRP-conjugated goat anti-rabbit IgG antibody. D, antibody neutralization of ACT secreted by B. pertussis. Antibodies at 10, 2.5, 0.63, and 0.16 g/ml were incubated with live B. pertussis (A 600 ϭ 0.2) before adding to adherent J774A.1 cells. The resulting intracellular cAMP concentrations were measured, normalized to total protein concentration, and expressed as % relative cAMP.
provide proof-of-concept that RTX dominates the anti-ACT immune response and that the RTX domain can recapitulate the humoral immune responses induced by ACT.

DISCUSSION
The recent surge in pertussis cases, coupled with increasing recognition of the current acellular vaccine's shortcomings, has motivated design of third generation vaccines to prevent pertussis. In humans, even a single dose of whole cell vaccine significantly reduces the risk of illness (63), although in baboons, acellular immunization prevented the severe symptoms of disease but allowed bacterial persistence and transmission to naive animals (5). To design future vaccines that minimize subclinical disease and reduce transmission to susceptible infants, it is crucial to understand the roles played by various protective antigens. ACT has been shown to be protective in animal models and is immunogenic in humans (15,16,40,64,65). Because ACT activities hinder local anti-bacterial immune responses (66 -68), anti-ACT antibodies may protect these cells, indirectly facilitating bacterial elimination. Here, we demonstrated that the RTX domain is able to largely recapitulate the protective humoral immune response induced by ACT in mice and is better expressed and more stable than the intact ACT.
ACT was first discovered based on its ability to increase cAMP levels in neutrophils, inhibiting their anti-bacterial functions, including phagocytosis and respiratory burst, and promoting the early stages of disease establishment (66). At physiologically relevant concentrations (Ͻ50 ng/ml) (69), ACT results in cytotoxicity to the murine macrophage cell line J774A.1 after 2 h of exposure; ACT also causes chlorine efflux from polarized epithelial cells and hinders IL-2 secretion and proliferation of T cells (42). More recently, ACT has been shown to suppress development of an IL-17-mediated immune response that appears key for bacterial clearance (21). As a result, passively administered antibodies blocking ACT function may be able to enhance neutrophil-mediated phagocytosis of opsonized bacteria (67). Murine studies have shown that immunization with ACT alone or as a supplement to the acellular vaccine reduces bacterial colonization, an effect that correlated with increased immunoglobulin levels and a Th1/Th2 cytokine phenotype (16,54). Finally, ACT is a highly conserved antigen, able to induce protective immunity in mouse models against the three dominant Bordetella species (B. pertussis, B. parapertussis, and B. bronchiseptica) (70 -72). Although ACT is unlikely to be highly protective as an isolated antigen, it may be a valuable addition to vaccines.
The complex mechanism by which ACT directly translocates its catalytic domain into the host cell cytosol remains incompletely understood. The current working model consists of three steps (48,73,74). First, in the presence of millimolar levels of calcium and acylation, the RTX domain forms a ␤-barrel that binds the ␣ M ␤ 2 receptor on neutrophils or the ␤ 2 -containing integrin lymphocyte fusion-associated antigen 1 receptor on T cells through N-linked oligosaccharides. Second, this is followed by insertion of two loops (four predicted transmembrane ␣-helices between residues 502-522 and 565-591) into the host cell membrane, resulting in a translocation intermediate that permeabilizes the membrane to allow an influx of extracellular calcium ions, activating calpain-mediated cleavage of the integrin's talin tether. Third, the ACT-receptor complex is then free to diffuse to cholesterol-rich lipid rafts, which triggers complete translocation of the catalytic domain dependent on residues 375-485 (48). Interestingly, the catalytic domain allows insertion of peptides up to 206 amino acids long for intracellular delivery (75).
This structure-function information provides insight into epitopes required for cellular intoxication and those likely to induce protective responses. For instance, antibodies could block translocation steps, yet only one such antibody has been characterized. Immunization with ACT followed by analysis of the resulting polyclonal serum suggested that antibodies recognizing the RTX domain dominate the response, as 6 of 12 monoclonal antibodies recognized this domain (47,72). The 3D1 antibody binds a conformational epitope between residues FIGURE 7. RTX domain is immunodominant and elicits neutralizing antibodies. A, immunization with ACT yields sera preferentially recognizing the RTX domain. Purified domains were coated at equal moles on microtiter plates, with sera serially diluted starting at 1:200. The average EC 50 for individual domains is shown. Error bars are the standard deviations of the EC 50 value among four mice. B, immunogenicity of purified domains. Mice were immunized with intact ACT or individual domains, with the serum EC 50 for ACT measured by ELISA after the first boost (6 weeks) or second boost (8 weeks). C, sera from mice immunized with the ACT and RTX 985 domain neutralize cAMP intoxication similarly. Sera from each immunization group at a 1:400 dilution were incubated with 125 ng/ml ACT in DMEM before adding to J774A.1 cells. Intracellular cAMP levels were measured by cAMP ELISA, divided by the total protein concentrations, and normalized to cells treated with ACT alone, as in Fig. 4. pre indicates baseline sera collected prior to immunization. Statistical significance was determined by one-way analysis of variance with Tukey's test; **, p Յ 0.01; ***, p Յ 0.001. For all panels, * indicates an acylated domain. n.s. indicates not significant (p Ͼ 0.05).
373 and 399, adjacent to the catalytic domain, trapping a translocation intermediate and preventing complete delivery of the catalytic domain to the host cytosol (48,73), while the antihydrophobic region antibody 2A12 inhibits intoxication and, to a lesser extent, hemolysis, and the anti-RTX antibody 6E1 inhibits only hemolysis (47). Consistent with these prior reports, we observed that the majority of antibodies recovered from phage libraries bind the RTX domain, whereas sera from four mice immunized with the holo-toxin bind RTX only. To the best of our knowledge, no antibodies blocking the ACTreceptor binding, such as M1H5 or M2B10, have been previously described.
A barrier to development of additional technologies based on ACT has been the challenges of recovering monomeric protein from in vitro refolding processes. Standard protocols, including dilution and dialysis from denaturing buffers containing 8 M urea, recover high molecular weight species with variable activity levels. Recently, refolding on a size exclusion column was performed to prevent aggregation of partially folded species and resulted in purification of monomers with very high activity, which depended on the presence of calcium, acylation and molecular confinement (43). Although promising, the yields and scalability of this process are currently unclear. As shown here, the RTX domain retains many structural features, is more readily expressed, and exhibits greater stability than ACT. Furthermore, because RTX lacks the catalytic domain, it has no homology to endogenous proteins and thus poses no potential autoimmune concerns for human use. We evaluated two different RTX constructs, initially RTX 985 lacking the acylation sites and thus simpler to express, and then the larger RTX 751 * retaining the acylation sites. Although both exhibited expected calcium-dependent structural shifts and binding to previously described monoclonal antibodies, RTX 751 * appears superior to RTX 985 in terms of monomericity, stability, and recognition of soluble ␣ M ␤ 2 receptor in vitro.
Because RTX 985 is well behaved and binds at least one neutralizing antibody (M2B10), we used this construct for immunization before discovering that RTX 751 and RTX 751 * are better behaved and retain the ability to bind both neutralizing antibodies. Regardless, sera from mice immunized with RTX 985 neutralized cAMP intoxication in vitro as efficiently as sera from ACT-immunized mice (Fig. 7C). This may be because these nonoverlapping epitopes induce antibodies neutralizing toxin via similar receptor blocking mechanisms. Supporting this idea, no synergy was observed when the M2B10 and M1H5 antibodies were combined in vitro (data not shown). Thus, it may be possible to induce a strong neutralizing response when only one of the epitopes is structurally intact. We do not have data on RTX 751 * immunization, but based on the biochemical and biophysical data presented here, we expect it to perform as well or better than RTX 985 .
Structure-function analyses of antibody-antigen interactions can identify residues forming protective epitopes, key information to guide design of immunogens able to elicit neutralizing antibodies. This approach has been employed for complex antigens with high sequence variability and metastable protective epitopes, such as fHBP from Neisseria meningitidis (76), and the F-protein from respiratory syncytial virus (77). We have demonstrated proof-of-concept that constructs based on the RTX domain are better behaved than ACT, while retaining key epitopes and inducing neutralizing antibodies. Using the structures of homologous RTX toxins (58,78), RTX variants with enhanced stability, expression level, and reduced immunogenicity of non-neutralizing epitopes could be engineered to evaluate RTX as a vaccine antigen in murine challenge models.  RTX 985 (B). Sera at a 1:200 dilution were incubated on ELISA plates coated with ACT, before addition of 0.1 nM M2B10, M1H5, or M1F11 as a competitor. After incubation, immobilized monoclonal antibody was detected with anti-human Fc-HRP, with absorbance normalized to wells without sera. Lower absorbance indicates a higher concentration of epitope-specific murine antibodies. FIGURE 9. RTX dominates the human immune response to ACT. Nine serum samples from humans exposed to B. pertussis were tested for reactivity to the catalytic domain, RTX 751 , or intact ACT by ELISA. Absorbance values at a 100-fold dilution of the sera were normalized to that of ACT at the same dilution. A paired t test was used to determine the statistical significance between signals for CAT and RTX binding domains. ***, p Յ 0.001.