Use of a neutralizing antibody helps identify structural features critical for binding of Clostridium difficile toxin TcdA to the host cell surface

Clostridium difficile is a clinically significant pathogen that causes mild-to-severe (and often recurrent) colon infections. Disease symptoms stem from the activities of two large, multidomain toxins known as TcdA and TcdB. The toxins can bind, enter, and perturb host cell function through a multistep mechanism of receptor binding, endocytosis, pore formation, autoproteolysis, and glucosyltransferase-mediated modification of host substrates. Monoclonal antibodies that neutralize toxin activity provide a survival benefit in preclinical animal models and prevent recurrent infections in human clinical trials. However, the molecular mechanisms involved in these neutralizing activities are unclear. To this end, we performed structural studies on a neutralizing monoclonal antibody, PA50, a humanized mAb with both potent and broad-spectrum neutralizing activity, in complex with TcdA. Electron microscopy imaging and multiangle light-scattering analysis revealed that PA50 binds multiple sites on the TcdA C-terminal combined repetitive oligopeptides (CROPs) domain. A crystal structure of two PA50 Fabs bound to a segment of the TcdA CROPs helped define a conserved epitope that is distinct from previously identified carbohydrate-binding sites. Binding of TcdA to the host cell surface was directly blocked by either PA50 mAb or Fab and suggested that receptor blockade is the mechanism by which PA50 neutralizes TcdA. These findings highlight the importance of the CROPs C terminus in cell-surface binding and a role for neutralizing antibodies in defining structural features critical to a pathogen's mechanism of action. We conclude that PA50 protects host cells by blocking the binding of TcdA to cell surfaces.

hydrate-binding sites. Binding of TcdA to the host cell surface was directly blocked by either PA50 mAb or Fab and suggested that receptor blockade is the mechanism by which PA50 neutralizes TcdA. These findings highlight the importance of the CROPs C terminus in cell-surface binding and a role for neutralizing antibodies in defining structural features critical to a pathogen's mechanism of action. We conclude that PA50 protects host cells by blocking the binding of TcdA to cell surfaces.
Clostridium difficile is a Gram-positive, anaerobic bacterium that can colonize humans and other animals to cause mild-tosevere diarrhea and, in some cases, fulminant colitis and death (1). Infection is typically associated with antibiotic use and the resulting dysbiosis in the colonic microbiota that facilitates C. difficile growth. In 2011, C. difficile infection (CDI) 4 was thought to have caused 500,000 infections and 29,000 deaths in the United States (2). The cost of CDI to the United States healthcare system has been steeply increasing since the early 2000s (2,3), but the therapeutic approaches for treatment have remained limited. Strong antibiotics such as metronidazole, vancomycin, or fidaxomicin are used to combat the active infection, but recurrence is a significant problem (1). Roughly 30% of people who experience CDI once will suffer from recurrence, in part because the antibiotics used in treatment prolong the dysbiosis in the microbial communities that restrict C. difficile growth (4).
The development of additional or complementary therapeutic strategies for the treatment of CDI has become a significant priority (5). Targeting multiple processes that influence infection, such as bacterial colonization or host microbiota recovery, is likely to be more successful long-term at combating CDI than antibiotic treatment alone. For example, fecal microbiota transplantation has gained acceptance as a viable treatment for recurrent infection, with reported success rates between 83 and 100% (6). Nevertheless, practical considerations about the administration of fecal microbiota transplantation remain and include the potential for secondary infections and risks from the procedure itself (7). Production of encapsulated, orally administered fecal samples (8) as well as optimized mixtures of beneficial gut microbes has also led to successful results (7). Both approaches suffer at present from a lack of knowledge about the microbiota species that are required for effective treatment of a generalized patient population. Identification of the most critical steps to target along the infection pathway has also been problematic due to limited understanding of the pathways that bacteria use to cause persistent infection and disease. An alternate approach has been to focus treatments toward the root cause of disease symptoms and cellular damage in CDI, the TcdA and TcdB toxins produced by C. difficile.
TcdA and TcdB are large, homologous, multidomain proteins that are secreted during CDI and that target and disrupt the function of the colonic epithelium (9). Host cell intoxication is initiated by interaction of the toxins with receptors on the cell surface and by the subsequent endocytosis of the toxins into the cell (10,11). Acidification of the endosome by the host cell triggers a pH-sensitive conformational change within the toxin that allows for pore formation and translocation of the toxin's glucosyltransferase domain (GTD) across the endosomal membrane (12). The GTD is then released into the cytoplasm through an autoproteolytic cleavage event (13) and can modify and inactivate Rho family GTPases within the host cell (14). Downstream effects of GTPase inactivation can include loss of actin cytoskeletal structure, cytokine production, and apoptotic cell death (15).
A number of studies have demonstrated significant levels of protection by toxin-specific antibodies in animal models of CDI (16 -19). In human patients, persistent levels of anti-toxin antibodies, particularly those against TcdA, correlate with reduced recurrence (20 -23). Naturally circulating antibodies from human subjects have been isolated through B-cell cloning to generate specific neutralizing monoclonal lines with some clinical efficacy (24). The most complete clinical investigation of neutralizing antibodies against TcdA and TcdB has been with the monoclonal antibodies (mAbs) actoxumab (anti-TcdA) and bezlotoxumab (anti-TcdB) (5). The humanized mAbs were raised in mice immunized with either full toxoids or toxin fragments (25) and bind epitopes within the C-terminal combined repetitive oligopeptides (CROPs) domain of each toxin (26,27). In phase II clinical studies, these antibodies have shown efficacy in preventing CDI recurrence in human subjects (28). This outcome has been recently reproduced in Phase III clinical trials (5), although the combination of actoxumab and bezlotoxumab did not provide a protective advantage in prevention of recurrence over treatment with bezlotoxumab alone.
In considering why actoxumab did not provide more of a clinical benefit in the context of the Phase III study, one must consider the breadth and potency of neutralization across the full spectrum of sequences present in actively circulating C. difficile strains. A previous study shows that a different anti-TcdA CROPS antibody, PA50, could more effectively neutralize TcdA from multiple C. difficile strains (19) and that it recognized unique, although undefined, epitope(s) in the CROPs. The results raise the possibility that PA50 may provide clinical benefit in situations where actoxumab does not. The evolutionary conservation of TcdA in diverse clinical isolates demonstrates its importance in success of the pathogen and validates its viability as a therapeutic target with other agents or in other contexts than the trials that evaluated actoxumab. This clinical potential motivated an effort to better understand the mechanism of action of PA50.
The CROPs domain of TcdA is thought to contribute to the receptor-binding properties of the toxin (29). Although no single receptor has been identified, the TcdA CROPs is known to bind a series of carbohydrate structures that are present on the surface of colonic cells as glycoproteins. It is conceivable that antibodies that bind the TcdA CROPs can neutralize the toxin by blocking receptor-binding and entry. Alternatively, it is possible that mAb binding induces aggregation and/or conformational changes in the toxin that prevent receptor engagement. Here, we present structural and mechanistic characterization of the interactions between the neutralizing monoclonal antibody PA50 and C. difficile TcdA. We have used X-ray crystallography and electron microscopy (EM) to fully define the epitopes within the TcdA CROPs and cell-binding assays to reveal the specific mechanism of neutralization.

Identification of multiple PA50 epitopes on TcdA holotoxin
A goal for this study was to define the molecular interaction between TcdA and the PA50 antibody. Using negative-stain EM, a comparison of TcdA holotoxin (Fig. 1A) with TcdA in complex with PA50 mAb (Fig. 1B) revealed minor aggregation in the presence of the full monoclonal. This is likely the result of

TcdA neutralization by PA50
one antibody being able to cross-link two different toxins, as the aggregation was not present upon incubating TcdA with PA50 Fab (Fig. 1C). To better visualize the TcdA-PA50 Fab interaction, a dataset of 7891 particles in negative stain was collected using non-saturating conditions (excess toxin) and subjected to reference-free two-dimensional (2D) averaging. Classes containing TcdA holotoxin with no bound Fab could be isolated (Fig. 2B) and are consistent with previous structural studies showing that TcdA forms a globular head with two tails, short and long (30,31). The longer tail corresponds to the CROPS structure that contains multiple sequence repeats (32), which in TcdA from C. difficile strain VPI10463 is organized into seven long repeats (LR) with 31 intervening short repeat (SR) segments. The numbering scheme defines the CROPs as a series of seven larger repeat structures (R1-R7), each composed of one LR followed by two SRs and referred to by the numerical order of the LR ( Fig. 2A).
Within the subsaturating dataset, several classes revealed a single PA50 Fab bound to the extreme C terminus of the TcdA CROPs (Fig. 2C, supplemental Fig. S1). This structure is consistent with Fab binding to the last full repeat sequence in the CROPs, CROPs repeat seven (CROPs-R7). Examination of individual particles indicated that a second Fab may bind the CROPs immediately adjacent to the first at repeat six (CROPs-R6) (Fig. 2E), but the site was not highly occupied at this Fab:toxin ratio. A subset of particles that contained more than one Fab was selected (529 total), and subsequent 2D alignment and classification revealed complexes with two Fabs bound at CROPs-R6-R7 (Fig. 2D, supplemental Fig.  S2). Importantly, there were no particles found with only CROPs-R6 occupied, indicating that R7 is bound preferentially over R6.
A second dataset (18,991 particles total) was then collected using a 3-fold molar excess of Fab to TcdA. Particles were aligned, subjected to 2D averaging, and sorted into 94 classes (supplemental Fig. S3). All Fab-bound classes show CROPs-R6-R7 bound by PA50 Fab, as in the subsaturating dataset, with an additional Fab detected at repeat five (CROPs-R5). Three representative classes are shown in Fig. 2F, illustrating the three bound Fabs at CROPs-R5-R6-R7. The TcdA CROPs has a coiled, ␤-solenoid structure that could influence how the toxin sits on the grid and the level of detail seen at adjacent binding

TcdA neutralization by PA50
sites. Visualizing the Fab at CROPs-R5 appears to be sensitive to the overall orientation of the CROPs. This site may be angled further away from the surface of the grid compared with the sites at R6 and R7 when the complex is bound. Importantly, no classes were observed with Fabs bound at only R5 or R6, indicating a hierarchy of occupation starting with the most C-terminal repeat R7. In some complex averages (Fig. 2F, center), there was indication of a low occupancy fourth binding site that could not be clearly resolved in the full dataset analysis. Based on the model of the TcdA CROPs, the fourth site appeared to be located at CROPs repeat two (CROPs-R2). Manual selection of a subset of particles for averaging (515 total) allowed visualization of this additional binding site ( The stoichiometry of PA50 Fab binding to TcdA holotoxin was also assessed using size exclusion chromatography with multiangle light scattering (SEC-MALS). Incubation of TcdA with increasing amounts of PA50 Fab produced larger molecular weight complexes that could be separated by size exclusion chromatography ( Fig. 3A). At both 1:5 and 1:10 molar excess of Fab to toxin, multiangle light-scattering analysis indicated a complex with a molar mass of 425 kDa, corresponding in size to TcdA (300 kDa) with a total of three Fab fragments bound (47 kDa each) (Fig. 3B). This result is consistent with the predominant complex identified in the EM analysis, TcdA with Fabs at CROPs-R5-R6-R7. The complex with four bound Fabs was not detected by SEC-MALS, a result that could reflect the rarity of this complex and, possibly, the inability of a low-affinity fourth binding site at CROPS-R2 to be maintained over the course of chromatographic separation.

X-ray crystal structure of TcdA 2460 -2710 in complex with PA50 Fab
The portion of the CROPs containing the last two full repeats (R6 and R7) through the C terminus (residues 2460 -2710) was crystallized in the presence of excess PA50 Fab. Crystals diffracted to 3.23 Å resolution, and a structure was determined by molecular replacement using a structure of CROPS fragment 2456 -2710 (PDB code 2G7C) as a search model ( Table 1). The crystal structure revealed two Fabs bound to R6 and R7 of the CROPs (Fig. 4A), although density for the constant regions of the Fab at CROPs-R7 could not be completely resolved. Fab residues modeled at R6 include HC(2-214) and LC , and residues at R7 include only the variable regions HC(2-118) and LC(2-102). The CROPs R6-Fab and CROPs R7-Fab structures align with an r.m.s.d. of 0.572 Å over 254 C-␣ atoms, indicating little change in the overall structure between the two binding sites (Fig. 4B). In light of this similarity, the analysis will focus on the details of the CROPs-R6 site. A simulated annealing omit map from this interface is included (supplemental Movie S1).
The Fab-binding site of the R6 (and R7) CROPs has elements from the LR, SR1, and SR2 (Fig. 4C). Among the complementarity determining regions (CDRs) of the Fab, designated H1-H2-H3 for the heavy chain and L1-L2-L3 for the light chain, all CDRs except CDR-L2 have residues that make contact with the CROPs (Fig. 4D, Table 2). PA50 covers a relatively large surface area (1177 Å 2 ) with multiple structural features contributing to the binding interaction.
The molecular elements that contribute to binding can be divided into three smaller interfaces (Fig. 5A). The first is centered on the interaction of CDR-L3 with the CROPs LR. Most notably, Ser 91 of CDR-L3 appears to be positioned within hydrogen-bonding distance of four CROPs residues (Asp-2540,
The EM analysis at different Fab concentrations revealed an ordered preference for Fab binding, with initial occupation of CROPs-R7, followed by R6, then R5, and with binding at R2 observed only rarely (Fig. 2). Alignment of all the LR-SR1-SR2 sequences in the TcdA CROPs highlights the degree of sequence divergence among the repeats and points to residues that define PA50 specificity (Fig. 6A). The Asp-2540, Ala-2541, and Asn-2542 residues found in the Trp-103 pocket (Fig. 5C) are conserved through R5-R6-R7 (and R3) (Fig. 6A, blue asterisks). The change from Ala-2541 (R6 numbering here and below) to Asn in R2 would alter the interaction with Trp-103. In addition, the presence of Tyr in place of Asp-2540 in R2 should   Residues in hydrogen bonds are in bold, with the rest of residues involved in nonbonded contacts. Residues not located in a CDR are in parentheses.

TcdA neutralization by PA50
prevent formation of the binding pocket through steric hindrance. Consistent with this, the introduction of these mutations into R7 (R7 D2631Y/A2632N ) reduced the PA50-binding affinity from 812 pM to 153 nM (Fig. 6B).
The non-binding repeats also differ in the residues that would otherwise interact with CDR-L3. Most notably, there is a switch from Arg-2550 in R6-R7 (Fig. 6A, first red asterisk) to Leu or Val in the other repeats, including R2 and R5. Unexpectedly, the introduction of the Arg-to-Leu mutation in R7 (R7 R2641L ) resulted in a slight increase in binding affinity (Fig.  6B). Whereas we did not test the effect of mutating this to Val, this residue may contribute to binding through hydrophobic rather than charge interactions.
The CROPs residues found in the Met-65 pocket are strictly conserved through R6 and R7 (with only a conservative change from Tyr-2557 in R6 to His in R7), but there is divergence in both R2 and R5 (Fig. 6A, black asterisks). Although the residues corresponding to Phe-2555 and Tyr-2583 remain the same, substitution of a small, polar Thr for Tyr-2557 and positively charged Lys for Ile-2562 disrupts the hydrophobicity of the site. Overall, the non-epitope CROPs repeats such as R1 and R4 display high divergence from the sequences in repeats containing PA50 epitopes. Whereas R3 is closer in sequence to the epitope of R5, the substitution of a negatively charged Glu for what is otherwise a positively charged residue in all other repeats (Arg-2554 in R6) is the difference that likely

TcdA neutralization by PA50
accounts for its lack of PA50 binding to R3 (Fig. 6A). Arg-2554 has hydrogen bonds with Ser-91 of CDR-L3. Mutating the comparable residue in R7 to Ala (R7 R2645A ) reduces the PA50-binding affinity by 377-fold (from 812 pM to 306 nM) (Fig. 6B). Lastly, there are two sequence differences between the epitope residues in R6 and R7, a Tyr to His change at R6 residue 2557 and an Asp to Asn change at R6 residue 2579 (Fig. 6A). These differences could contribute to the preferential binding at R7.

PA50 blocks binding of TcdA holotoxin to human colonic epithelial cells
Previously, PA50 was shown to neutralize the toxicity of TcdA on T84 cells in vitro, with no loss of efficiency after humanization of the monoclonal (19). To examine whether PA50 can directly block toxin binding to cells, cell-binding assays were performed using Caco-2, a human colonic epithelial cell line. Although a 2-fold molar excess of mAb had a minor impact on TcdA-binding, the amount of TcdA bound to the surface of the cells was significantly reduced in the presence of a 4-fold molar excess of PA50 mAb (Fig. 7, A and B). An even greater reduction in binding was seen in the presence of the PA50 Fab (Fig. 7, C and D). In this case, the binding of TcdA to cells was nearly eliminated using either a 4-fold (4:1 Fab:TcdA) or 8-fold (8:1 Fab:TcdA) molar excess of Fab, exhibiting 3% of the binding seen with TcdA alone. These data imply that the Fab is more efficient at preventing cell-surface binding, as almost no toxin was bound with even a 4-fold molar excess of Fab (Fig. 7D), which provides only half the antibody-binding sites of the mAb experiment.

Discussion
We have shown that the neutralizing monoclonal antibody PA50 is able to specifically recognize a repeating epitope within the CROPs domain of TcdA. EM imaging showed that PA50 binds to three sites near the C terminus of TcdA, with a fourth lower-occupancy binding site closer to the N-terminal end of the CROPs (Fig. 2). The imaging results are consistent with measurements made by SEC-MALS, which indicate a stoichiometry of three Fabs bound to each TcdA molecule (Fig. 3). The crystal structure of PA50 Fabs bound to TcdA 2460 -2710 (R6 and R7) provides an atomic view of the interaction and defines the CROPs epitope (Figs. 4 and 5). Both PA50 mAb and Fab can significantly reduce the amount of toxin bound to the surface of colonic epithelial cells, with the Fab exhibiting a higher efficiency of inhibition than mAb (Fig. 7). Although we initially speculated that the mechanism of neutralization could be one of TcdA aggregation in the presence of mAb (Fig. 1B), the capacity of PA50 Fab to effectively block toxin-binding strongly supports an alternative mechanism of receptor blockade.
In addition to an analysis of how epitopes differ across the 7 CROPs repeats (Fig. 6A), we also considered whether the preference of PA50 for certain CROPs repeats could be influenced by differences in epitope presentation or accessibility. Although we cannot exclude this possibility, the ␤-solenoid conformation of the CROPs has only been observed as a rigid elongated structure, and with the exception of R4, which could be occluded by the TcdA delivery domain ( Fig. 2A), we predict that all other epitopes are fully accessible in solution. The EM and crystallography data indicate that neighboring PA50 Fabs are oriented away from each other. It is, therefore, unlikely that individual Fabs could interfere with binding at other epitopes. Finally, we note that Fab binding does not induce conformational changes in the CROPS structure. The Fab-bound structure aligns to the unbound structure of TcdA 2456 -2710 with an r.m.s.d. of 0.882 Å over 170 C-␣ residues.
The ability to compare TcdA sequences that do and do not bind the PA50 Fab in the context of a crystal structure allows us to propose a subset of residues that define the PA50 epitope on TcdA (Fig. 6A, R6 and R7 residues with asterisks). Importantly, these residues are strictly conserved in the R6 and R7 TcdA sequences from multiple C. difficile strains (VPI10463, 630, CD196, and R20291). A previous study showed that PA50 was capable of neutralizing TcdA from a diverse panel of strains with EC 50 s in the range of 20 -127 pM (19). CDA1/actoxumab, by contrast, has proven to be especially ineffective at neutralizing TcdA from epidemic, 027 ribotype strains, with EC 50 values in the nanomolar range (19,33). Localization of highly conserved binding sites within the last repeats of the CROPs could, therefore, explain the broad neutralizing activity of PA50.

TcdA neutralization by PA50
(1,4)-␤-GlcNAc trisaccharide (Fig. 8A) (34) and with the recently reported structures of TcdA CROPs in complex with two neutralizing single-domain antibodies (sdAbs) (35). The sdAbs are the only other anti-TcdA antibodies that have crystal structures defining the epitope interactions. Because the sdAbs only contain a heavy chain, the surface area covered by the sdAbs is smaller than that of PA50 (730 -790 Å 2 versus 1177 Å 2 ). Notably, neither the sdAbs nor PA50 Fab occlude the trisaccharide-binding site (Fig. 8B), and in competition experiments, the sdAbs were not able to compete with the binding of trisaccharide molecules (36).
One of the two sdAbs, A20.1, is bound to a region of the CROPs that overlaps with the PA50 epitope (Fig. 8C). The A20.1 epitope is located even further away from the trisaccharide pocket than that of PA50 with the majority of the interactions mediated by CDR1 and CDR3 of the sdAb and LR and SR1 of the CROPs. The other sdAb, A26.8, binds a non-overlapping site in R7 at the extreme C terminus of the CROPs. The epitope is not predicted to be repeated elsewhere in the CROPS.
The importance of the most C-terminal part of the TcdA CROPs to cell binding and entry has been noted in two studies where CROPs truncation constructs that lack repeats 5, 6, and 7 were not able to bind and enter cells (10,37). Notably, a construct containing only R5, R6, and R7 bound cells with an apparent K D equal to that of full-length CROPS (37). These observations along with the data included herein suggest that the CROPs C-terminal repeats (R5, R6, and R7) play a critical role in cell-surface receptor recognition.
Although the TcdA-CROPs is considered the primary site for toxin binding to host cells, the nature of the required receptor(s) for TcdA intoxication remains unclear. Although the ␣-Gal-(1,3)-␤-Gal-(1,4)-␤-GlcNAc trisaccharide is a known ligand for the CROPs (38), this sugar is not present in humans. TcdA can bind other related types of carbohydrates that are present in the human intestine (39,40), but it is still unclear if these interactions are important for TcdA intoxication. Although we and others have assumed that these carbohydrates bind in the same binding sites as the ␣-Gal-(1,3)-␤-Gal-(1,4)-␤-GlcNAc trisaccharide, it is possible that the human sugars bind in distinct locations. It is also possible that there are carbohydrate-independent interactions that are important not only for binding but for endocytosis of the toxin.
Future research is needed to identify the physiologically relevant receptor(s) for TcdA intoxication. It is possible that the common CROPs region recognized by both PA50 and the A20.1 sdAb represents a binding site for such a receptor. This work highlights the importance of the CROPs C terminus in cell-surface binding and the role that neutralizing antibodies can play in identifying important structural features that are critical to a pathogen's mechanism of action. It also demonstrates that PA50, a humanized mAb with both potent and broad spectrum neutralization activity, protects the host cell by blocking the cell-surface binding of TcdA.

Expression and purification of recombinant TcdA constructs
Recombinant full-length TcdA holotoxin (representing a sequence from strain VPI10463; pBL282) was expressed in Bacillus megaterium as previously described (41) and purified by nickel affinity, anion exchange, and size exclusion chromatography. The regions of the TcdA gene corresponding to either the two full C-terminal repeats of the CROPs (TcdA 2460 -2710 ; TcdA-R6-R7; pBL840) or the last CROPs  3-6). Whole cell lysates were prepared for SDS-PAGE and Western blotting. The blot was probed with antibodies against TcdA and GAPDH. Cells that did not receive toxin and cells that received only toxin without any antibody were used as controls (lanes 1 and 2). Positions of relevant molecular weight markers are included. B, experiments shown in A were quantified by densitometry, and binding of TcdA to cells was determined by normalizing bound TcdA levels to that of GAPDH. Relative TcdA binding to cells was determined by normalizing the binding values to that of the toxin only controls (*, p Ͻ 0.05). C and D, similar experiments performed with TcdA-Fab complexes (*, p Ͻ 0.05). Results reflect the mean Ϯ S.D. of three independent experiments and were analyzed using one-way analysis of variance.

Construction, expression, and purification of PA50 mAb and PA50 Fab
DNA constructs for the light chain and heavy chain of PA50 were cloned into an in-house vector by standard molecular biology methods to produce both the PA50 mAb and Fab. For the Fab, the heavy chain consists of the variable region, the C H 1 domain, and a partial hinge region (including the EPKSC sequence). Expression and secretion of the proteins were driven by a cytomegalovirus promoter and the native immunoglobulin light chain signal peptide. The antibodies were transiently expressed in a proprietary Chinese hamster ovary (CHO)-derived cell line. Cell cultures were fed with an in-house developed media until expression was terminated 10 -13 days post transfection. Fabs were purified by light chain affinity purification (KappaSelect resin, GE, 17-5458-02), whereas mAb was isolated by protein A affinity (HiTrap Protein A HP, GE, 17-0403-01) according to the manufacturer's instructions.

Electron microscopy of TcdA holotoxin with PA50 Fab
Complexes of TcdA holotoxin with PA50 Fab were set up in 20 mM Tris, pH 8.0, 100 mM NaCl under conditions of either excess toxin (100 nM Fab:300 nM TcdA) or excess Fab (300 nM Fab:100 nM TcdA). Reactions were incubated for 30 min at room temperature before dilution in the same buffer to 6 ng/l toxin. Samples were immediately applied to glow discharged, carbon-coated copper grids for negative staining with 0.75% uranyl formate (42). Initial micrographs of TcdA in complex with either PA50 mAb or Fab were collected at 44,000 magnification on a Morgagni (100 keV; FEI) transmission electron microscope. Micrographs of TcdA in complex with PA50 Fab were collected at 62,000 magnification on a Tecnai F20 (200 keV; FEI) transmission electron microscope equipped with a Gatan 4K ϫ 4K CCD camera. Images were converted to mixed raster content (mrc) and binned by 2, giving a final value of 3.5 Å/pixel. Individual particles were manually selected in Boxer/ EMAN (43) (box size 150 ϫ 150 pixels), and image stacks were generated in SPIDER (44). Reference-free two-dimensional class averaging was performed in Scipion (45) using the Xmipp3-CL2D algorithm (46).

TcdA neutralization by PA50 Stoichiometry analysis by SEC-MALS
To determine binding complex stoichiometry of PA50 Fab to TcdA, the indicated molar ratios of PA50 Fab and TcdA in PBS were combined in a final volume of 136 l. Mixtures were incubated at room temperature for at least 60 min before analysis. Samples were analyzed by size exclusion chromatography using a WTC030S5 column with a 14-ml bed volume (Wyatt Technologies, Santa Barbara, CA) on an Agilent 1100 HPLC (Agilent, Santa Clara, CA) at room temperature. The samples were eluted isocratically in PBS at a flow rate of 1 ml/min for 20 min. Eluted proteins were detected using UV absorbance at a wavelength of 280 nm. Data analysis was done using the Agilent software ChemStation (version A.02.10). Column calibration was performed with a set of molecular weight standards ranging from 10 to 500 (Bio-Rad). In-line SEC-MALS was performed. Sample measurements were done on a DAWN HELEOS II MALS with an Optilab Rex refractometer (Wyatt Technologies). Molecular mass of each protein within a defined chromatographic peak was calculated using ASTRA, version 6.1 (Wyatt Technologies).

X-ray crystallography of TcdA 2460 -2710 with PA50 Fab
TcdA 2460 -2710 (18 M final) was incubated with a 2.1-fold molar excess of PA50 Fab for 30 min at room temperature. The resulting complex was isolated over a Sephadex 200 size exclusion column (1 ϫ 25 cm, GE Healthcare) in 20 mM Tris, pH 8.0, 100 mM NaCl, and the elution peak corresponding to the TcdA fragment with two bound Fab molecules was pooled and concentrated to 10 mg/ml. The complex was subjected to broad matrix crystallization screens to find conditions for crystal growth, and these initial hits were optimized to produce diffraction quality crystals. Final crystallization conditions were in 0.1 M Tris, pH 7.5, 0.6 M CaCl 2 , 19% polyethylene glycol 3350, and individual crystals were harvested and cryoprotected in mother liquor containing 20% ethylene glycol. X-ray diffraction data were collected at Advanced Photon Source, Argonne National Laboratory (Lemont, IL). Data were processed with Xia2 through the DIALS pipeline using programs within CCP4 (Pointless/Aimless) for indexing and scaling (47)(48)(49)(50). Molecular replacement was performed with Phaser (51) using the structures of the TcdA CROPs repeats (PDB code 2G7C) and a highly similar humanized antibody Fab fragment (PDB code 3L5X) as search models to phase the data. Refinement was performed in Phenix, and Coot was used for model building (52,53). All structure figures were made in UCSF Chimera (54) or PyMOL (The PyMOL Molecular Graphics System, Version 1.8, Schrödinger, LLC). Software used in this project was curated by SBGrid (55). The atomic coordinates for the crystal structure are available in RCSB PDB under PDB code 5UMI.

Epitope analysis of TcdA 2460 -2710 with PA50 Fab
Epitope surface area, non-bound contacts, and residues directly involved in hydrogen or electrostatic bonds in the final refined crystal structure were analyzed with ePISA (56) and PDBSum (57).

Surface plasmon resonance (SPR) assays
Kinetic rate constants (k a and k d ) for binding of PA50 to purified wild-type or mutant TcdA-CROPs-R7 were measured by employing an IgG capture assay on a BIAcore T200 instrument. Mouse anti-huIgG-Fc was immobilized on a CM4 sensor chip with a final surface density of 1600 resonance units. A reference flow cell surface was also prepared on this sensor chip by use of an identical immobilization protocol. PA50 mAb was prepared at 5 nM in instrument buffer (HBS-EP buffer: 0.01 M HEPES, pH 7.4, 0.15 M NaCl, 3 mM EDTA, and 0.005% P-20) along with dilutions of R7 WT (0.0025 nM to 50 nM), R7 D2631Y/A2632N (0.62 nM to 450 nM), R7 R2641L (0.0025 nM to 50 nM), or R7 R2645A (0.62 nM to 1350 nM) in instrument buffer. A sequential approach was utilized for kinetic measurements. PA50-huIgG1 was first injected over the capture surface at a flow rate of 10 l/min. Once the binding of the captured PA50 stabilized, a single concentration of TcdA-CROPs-R7 was injected over both surfaces at a flow rate of 75 l/min. The resulting binding response curves yielded the association phase data. After the injection of TcdA-CROPs-R7, the flow was switched back to instrument buffer for 10 min to permit the collection of dissociation phase data, with a subsequent 2-min pulse of 3 M MgCl 2 to regenerate the IgG capture surface on the chip. Binding responses from duplicate injections of each concentration of TcdA-CROPs-R7 were recorded against PA50. In addition, several buffer injections were interspersed throughout the injection series. Select buffer injections were used along with the reference cell responses to correct the raw data sets for injection artifacts and/or nonspecific binding interactions, commonly referred to as "double referencing." Fully corrected binding data were then globally fit to a 1:1 binding model (Biacore T200 Evaluation 2.0 software; GE Healthcare) that included a term to correct for mass transport-limited binding, should it be detected. These analyses determined the kinetic rate constants k a and k d from which the apparent dissociation constant (K D ) was calculated as k d /k a .

Cell-binding assays
Purified TcdA (25 nM) was preincubated with 2-or 4-fold excess of PA50 mAb or an isotype control mAb (PA41) for 20 min at room temperature. For binding assays with Fab fragments, the antibody concentration was doubled to keep the available number of antigen-binding sites similar between the mAb and Fab experiments. TcdA-antibody complexes were then added to Caco-2 monolayers that were prechilled to 10°C. Cells that did not receive toxin and cells that received only toxin without any antibody were used as controls. Toxins were allowed to bind at 10°C for 1 h. Media containing unbound toxin inoculum were then removed, and cells were washed twice with ice-cold PBS. Cells were dislodged with a cell scraper, collected, and pelleted at 1000 ϫ g for 5 min. To pre-

TcdA neutralization by PA50
pare samples for Western blotting, cell pellets were suspended in 60 l of lysis buffer (10 mM Tris, pH 7.4, 250 mM sucrose, 3 mM imidazole) supplemented with protease inhibitor mixture (1:100, P8340; Sigma) and homogenized by passing 20ϫ through a 27-gauge needle fitted to a sterile 1-ml syringe. Nuclei and debris were pelleted and removed by spinning at 1500 ϫ g for 15 min. Samples were then diluted with Laemmli sample buffer containing 2-mercaptoethanol and heated at 95°C for 5 min. Equal volumes were loaded on a 4 -20% Mini-Protean gradient gel (Bio-Rad). Proteins were transferred in Tris-glycine buffer to PVDF membranes at 100 V for 1 h and blocked with 5% milk in PBS containing 0.1% Tween 20 (PBST) overnight. Primary antibodies against TcdA (1:1000, NB600-1066; Novus Biologicals) and GAPDH (1:3000, sc-25778; Santa Cruz) were diluted in 5% milk-PBST and incubated with the membranes for 2 h at room temperature. Membranes were washed four times with PBST and then incubated with anti-mouse (7076S; Cell Signaling) or anti-rabbit (7074S; Cell Signaling) HRP-linked secondary antibodies for 1 h at room temperature (1:2000 for TcdA and 1:5000 for GAPDH). Membranes were washed four times with PBST, and HRP was detected using ECL Western blotting Substrate (32106; Pierce).

Cell assay statistical analyses
Results reflect the mean and S.E. of three independent experiments and were analyzed using one-way analysis of variance. p values were generated using Dunnett's multiple comparisons test in GraphPad Prism. A p value Յ 0.05 was considered significant.
Author contributions-H. K. K., A. C. N., G. J. R., P. W., B. W. S., and D. B. L. designed the study. H. K. K. performed and analyzed the EM experiments, purified and crystallized the TcdA-PA50 Fab complex, and determined its X-ray structure. H. K. K. and R. C. designed the expression clones and purified the proteins. R. C. designed, performed, and analyzed the cell-binding experiments, and K. R. and R. W. performed the SEC-MALS and SPR analysis. X. J. expressed and purified PA50 mAb and Fab. M. D. O. and B. W. S. assisted with the analysis of the EM and X-ray data, respectively. H. K. K. and D. B. L. wrote the paper. All authors reviewed the results and approved the final version of the manuscript.