Structure-guided design of a potent Clostridioides difficile toxin A inhibitor

Crystal structures of camelid heavy-chain antibody variable domains (VHHs) bound to fragments of the combined repetitive oligopeptides domain of Clostridioides difficile toxin A (TcdA) reveal that the C-terminus of VHH A20 was located 30 Å away from the N-terminus of VHH A26. Based on this observation, we generated a biparatopic fusion protein with A20 at the N-terminus, followed by a (GS)6 linker and A26 at the C-terminus. This A20-A26 fusion protein shows an improvement in binding affinity and a dramatic increase in TcdA neutralization potency (>330-fold [IC50]; ≥2,700-fold [IC99]) when compared to the unfused A20 and A26 VHHs. A20-A26 also shows much higher binding affinity and neutralization potency when compared to a series of control antibody constructs that include fusions of two A20 VHHs, fusions of two A26 VHHs, a biparatopic fusion with A26 at the N-terminus and A20 at the C-terminus (A26-A20), and actoxumab. In particular, A20-A26 displays a 310-fold (IC50) to 29,000-fold (IC99) higher neutralization potency than A26-A20. Size-exclusion chromatography-multiangle light scattering (SEC-MALS) analyses further reveal that A20-A26 binds to TcdA with 1:1 stoichiometry and simultaneous engagement of both A20 and A26 epitopes as expected based on the biparatopic design inspired by the crystal structures of TcdA bound to A20 and A26. In contrast, the control constructs show varied and heterogeneous binding modes. These results highlight the importance of molecular geometric constraints in generating highly potent antibody-based reagents capable of exploiting the simultaneous binding of more than one paratope to an antigen.


Introduction
Clostridioides difficile is a spore-forming Gram-positive bacterium capable of infecting humans and causing symptoms ranging from mild diarrhea to pseudomembranous colitis (Hussack and Tanha, 2016;Kordus et al., 2022). C. difficile infection is one of the most prevalent hospital-acquired bacterial infections, costing health-care systems ~$5 billion per year and claiming up to 30,000 lives annually in the United States alone (Fu et al., 2021). The healthy balance of microbiota normally present in the gastrointestinal tract of individuals is thought to suppress C. difficile infection (Vasilescu et al., 2021). Changes to this balance, often via the introduction of broad-spectrum antibiotics, can increase individuals' susceptibility to C. difficile colonization and infection. C. difficile remains an Urgent Threat pathogen according to the Centers for Disease Control and Prevention, despite numerous advances in new therapeutic agents to treat C. difficile infection. 1 C. difficile infection is currently treated with antibiotics that include metronidazole, vancomycin and fidaxomicin; however, significant incidences of disease relapse have made the search for alternative treatments, including vaccines (Henderson et al., 2017), fecal transplantation (Shogbesan et al., 2018) and antibody-based immunotherapy (Hussack and Tanha, 2016), a top public health priority. The primary virulence factors of C. difficile are two high molecular weight toxins (Kordus et al., 2022), toxin A (TcdA) and toxin B (TcdB), which are largely responsible for the physical damage in the colon of infected individuals and are the foremost targets of vaccine-and immunotherapy-based approaches (Hussack and Tanha, 2016;Henderson et al., 2017). Recently, a monoclonal antibody (mAb) targeting TcdB was shown to significantly reduce C. difficile relapse and was approved by the Food and Drug Administration. The trial data showed that reduced incidences of relapse were driven solely by the anti-TcdB mAb (bezlotoxumab) and not enhanced by the inclusion of an anti-TcdA mAb (actoxumab) in combination with bezlotoxumab (Wilcox et al., 2017). Interestingly, neither the TcdB antibody alone nor both mAbs in combination had a significant impact on the duration or severity of C. difficile infection (Wilcox et al., 2017), possibly due to the relatively poor efficacy of actoxumab. Several studies have illustrated that circulating anti-TcdA and TcdB antibodies in human patients correlate with reduced recurrence (Kyne et al., 2000(Kyne et al., , 2001Leav et al., 2010), suggesting the development of TcdA inhibitors with considerably greater potency than actoxumab are warranted for use in combination with bezlotoxumab or other TcdB inhibitors. Efforts toward developing anti-toxin Ab-based therapeutics, including therapeutic approaches targeting both TcdA and TcdB, have been reviewed earlier (Hussack and Tanha, 2016).
Toward the goal of developing more effective therapeutics targeting TcdA, we previously isolated a panel of camelid single-domain antibodies (V H Hs or nanobodies) targeting C. difficile TcdA, including several V H Hs with the ability to neutralize TcdA cytotoxicity (Hussack et al., 2011). To better understand the determinants of molecular recognition and neutralization activity, we solved the crystal structures of two of these V H Hs targeting non-overlapping TcdA epitopes, namely A20.1 (A20) and A26.8 (A26), in complex with fragments from the combined repetitive oligopeptides (CROPs) domain of TcdA (Murase et al., 2014). These high-resolution crystal structures identified the specific locations of unique epitopes targeted by these antibodies and explained the synergistic TcdA neutralizing effects observed with these V H Hs (Hussack et al., 2011). While monomeric V H Hs and combinations of monomeric V H Hs showed modest TcdA inhibition in vitro, we hypothesized that assembling the V H Hs as multimers and biparatopic dimers may improve TcdA neutralizing potency. Combining V H Hs in tandem, targeting distinct and/or common epitopes, has been successfully demonstrated for a number of anti-toxin V H H systems, including: botulinum toxins (Mukherjee et al., 2012;Huang et al., 2017;Lam et al., 2020;Tremblay et al., 2020), anthrax toxin (Moayeri et al., 2015;Vrentas et al., 2016), ricin (Vance et al., 2013;Herrera et al., 2015Herrera et al., , 2016, scorpion toxins (Hmila et al., 2008(Hmila et al., , 2010, Shiga toxins (Tremblay et al., 2013;Mejías et al., 2016), tetanus toxin (Rossotti et al., 2015), α-neurotoxin (Wade et al., 2022) and C. difficile TcdA and TcdB (Yang et al., 2014;Schmidt et al., 2016). In many of these examples, V H H orientations were largely chosen at random, and the spacer lengths separating the two (or more) V H H domains were typically not extensively explored.
The crystal structure of A20 and A26 bound to the C-terminal 255 residues of TcdA (amino acid residues 2,556-2,710 of TcdA10463; referred to as TcdA-A2 fragment; PDB 4NC1) revealed that the C-terminus of A20 was located 30 Å away from the N-terminus of A26 (Murase et al., 2014; Figure 1A). Molecular modeling led us to hypothesize that connecting A20 to A26 by a 12 residue (Gly-Ser) 6 linker would generate a biparatopic fusion protein in which the two tethered V H H domains could simultaneously bind to a single molecule of TcdA, potentially generating a TcdA inhibitor with substantially higher binding affinity and inhibitory potency. To test this hypothesis, we have produced this biparatopic fusion protein and used in vitro binding and cytotoxicity inhibition measurements to confirm its high potency and specificity, capable of neutralizing TcdA at picomolar concentrations. The much weaker neutralization activities observed for a series of control constructs highlight the advantages of a structure-based approach in designing highly specific neutralization reagents and define some of the geometric constraints that are most critical for successfully combining V H Hs into biparatopic reagents. This study provides a dramatic example of how high-resolution crystal structures can be used to tailor the design of a multivalent toxin-inhibiting protein to match the molecular geometry of a toxin to generate novel and potent therapeutics.

Structure-guided design of V H H fusion proteins
The ternary complex of the TcdA-A2 fragment bound to A20 and A26 (PDB 4NC1) indicated that the N-terminus of A26 was located 30 Å away from the C-terminus of A20. Models of linkers containing a repeating Gly-Ser dipeptide motif with backbone geometry similar to that seen in antiparallel β-strands were generated using PyMOL (Version 2.0, Schrödinger LLC). A linker with six (GS) dipeptides appeared to be sufficient to bridge the gap observed in the 4NC1 crystal structure, but the presence of disordered residues at the N-and C-termini of the V H H proteins introduced a modest degree of uncertainty into the design. As a result, a series of constructs with a small range of linker lengths were generated.

Size-exclusion chromatography
Purified proteins (400 μg in PBS) were loaded onto a Superdex™ 75 Increase 10/30 GL (Cytiva Life Sciences) column, at a flow rate of 0.5 mL/min, controlled by an AKTÄ™ FPLC (Cytiva Life Sciences). To determine the size of each construct, a standard curve was generated using the Bio-Rad™ Size-Exclusion Chromatography Standard (BioRad, Hercules, CA) and apparent molecular masses (M app s) of antibodies were calculated by interpolation using their elution volumes (V e s). Sizeexclusion chromatography (SEC) chromatograms were normalized as described (Kim et al., 2012).

Enzyme-linked immunosorbent assay
Nunc-immuno microtitre plate wells with Maxisorp surface (Thermo Fisher) were coated in triplicates with 30 ng/well of full-length A B

FIGURE 1
Model of the A20-A26 fusion protein bound to the CROPs domain of Clostridioides difficile TcdA. (A) The structure of full-length C. difficile TcdA was generated by superimposing the crystal structure of the N-terminal 1832 amino acid residues of TcdA (4R04) (Chumbler et al., 2016) onto the crystal structure of TcdB (6OQ5) (Chen et al., 2019). To model the CROP domain missing from the TcdA structure, we made the assumption that the orientation of the CROP domain relative to the N-terminal domains is similar in the two toxins. By superimposing the N-terminal short repeat of a model of the TcdA CROP domain (Ho et al., 2005;Pruitt et al., 2010;Pruitt and Lacy, 2012) onto the N-terminal short repeat in the CROP domain of TcdB (6OQ5), a model of full-length TcdA was generated. The resulting model shows similar features to the low resolution cryo-EM envelopes of full-length TcdA (Pruitt et al., 2010;Pruitt and Lacy, 2012). The structure of the complex formed between A20-A26 and TcdA was generated by modeling the structure of a (GS) 6 linker in an extended antiparallel β-strand conformation using PyMOL and positioning this linker between the C-terminus of the A20 V H H and the N-terminus of the A26 V H H observed in the crystal structure of the complex formed between A20, A26 and the TcdA-A2 fragment (4NC1) (Murase et al., 2014). TcdA is shown in cartoon ribbon and semi-transparent surface representations. A20, A26 and the A20-A26 V H H domains are drawn in cartoon representation. Each polypeptide chain is colored according to the colors of the rainbow, starting from blue at the N-terminus to red at the C-terminus. RBD, receptor binding domain; GT, glucosyl transferase. (B) Schematic representation of the V H H constructs generated in this work.

In vitro TcdA neutralization assay
Vero cells (CCL-81) were obtained from ATCC (Manassas, VA) and cultured according to ATCC's instructions in 96 well microtiter plates (Nunc) at 2 × 10 4 cells/well in MEM media (Gibco) supplemented with 10% (v/v) heat-inactivated fetal bovine serum (Gibco) at 37°C in 5% CO 2 . The neutralization activity of antibodies was determined by co-incubation with 80 ng/mL (260 pM) of full-length TcdA, with antibody and toxin added simultaneously to the cells. After 72 h of incubation, the cell viability was quantified with the Cell Proliferation Assay Reagent WST-1 (Roche Diagnostics, Laval, Canada) according to the manufacturer's instructions. Briefly, the media was replaced with 100 μL of MEM (without fetal bovine serum) containing 10% of WST-1, incubated for 40 min at 37°C in 5% CO 2 and finally the absorbance read at 450 nm. The neutralizing activity was calculated as % inhibition: A450 test is the absorbance of cells incubated with TcdA and varying concentrations of antibodies; A450 low is the absorbance of cells incubated only with TcdA (0% inhibition); and A450 high is the absorbance of cells incubated only with media (100% inhibition).

UPLC-SEC-multiangle light scattering analysis
Monomeric and dimeric V H Hs were mixed with TcdA-A2 at 1:1 molar ratios in PBS at final concentrations of 1.91 mg/mL and incubated at 4°C overnight. Control experiments which included monomeric V H Hs alone (0.91 mg/mL), dimeric V H Hs alone (0.91 mg/mL) and TcdA-A2 alone (1 mg/mL) were performed under the same conditions as the V H H-TcdA-A2 mixtures. UPLC-SEC analysis of protein samples was performed on a BEH200 SEC column (4.6 × 150 mm, 1.7 μm particle column, Waters, Milford, MA) using a Waters H-Class Acquity UPLC system equipped with a diode array detector, a Wyatt MiniDawn multiangle light scattering (MALS) detector, and a Wyatt Optilab T-rEX refractive index detector (Wyatt Technology, Santa Barbara, CA). The mobile phase was PBS (HyClone SH30028.02, Cytiva Life Sciences) containing 0.02% (v/v) polysorbate 20 at a flow rate of 0.4 mL/min. The column temperature was 30°C. Molar masses were determined using Astra software version 6.1.7.17 (Wyatt Technology) using either absorbance at 280 nm (A 280 ) or refractive index (RI) as the concentration measure. Similar results were obtained using either A 280 or RI, but band broadening reduced resolution in the RI signal so masses determined using the A 280 signal are reported.

Structure-guided design of V H H fusion proteins
We previously solved the X-ray structures of two C. difficile TcdAbinding V H Hs, A20 (K D = 2 nM) and A26 (K D = 12 nM), in complex with fragments from the CROPs domain (formerly referred to as the RBD, receptor binding domain) of TcdA (Hussack et al., 2011;Murase et al., 2014). Because in the ternary complex of the TcdA-A2 fragment bound to a single molecule of A20 and a single molecule of A26 (PDB 4NC1) the N-terminus of A26 was observed to be located 30 Å away from the C-terminus of A20, we hypothesized that a short linker peptide could be used to fuse the two V H H domains together in a recombinant fusion protein (Figure 1). Models of linkers containing a repeating Gly-Ser dipeptide motif were generated with backbone geometry similar to that seen in antiparallel β-strands. A linker with six Gly-Ser dipeptides (GS) 6 Frontiers in Microbiology 05 frontiersin.org appeared to be sufficient to bridge the gap observed in the 4NC1 crystal structure, but the presence of disordered residues at the N-and C-termini of the V H H proteins indicate that the ideal lengths and geometric details of linkers is challenging to model precisely. As a result, in addition to A20-A26 with a (GS) 6 linker, a series of constructs with a range of linker lengths were generated ( Figure 1B). These included A20-A26 (G 4 S), A20-A26 (G 4 S) 2 , and A20-A26 (G 4 S) 3 , which possessed linkers of shorter To test this prediction, the A26-A20 fusion protein was generated with the reverse polarity (i.e., A26 at the N-terminus and A20 at the C-terminus). To evaluate whether the tethering of two V H H domains generates a non-specific multivalency effect, two other fusion proteins containing two A20 V H H domains (A20-A20) and two A26 V H H domains (A26-A26), each tethered by the (GS) 6 linker, were also generated.

Expression, purification and biophysical characterization of V H H fusion proteins
While monomeric V H Hs were expressed in the periplasm of E. coli, the dimeric V H Hs were expressed in mammalian HEK293-6E cells to obtain higher expression yields. Nonetheless, in contrast to A20-A26 (G 4 S) and A20-A26 (G 4 S) 2 , which could be expressed in high yields similar to the dimers with the (GS) 6 linker (Table 1), the A20-A26 (G 4 S) 3 construct was surprisingly only expressed at very low yields despite repeated attempts, excluding its full analysis in the current study. The proteins were purified by immobilized metal-ion affinity chromatography and ran as single bands of the expected molecular masses on SDS-PAGE under reducing and non-reducing conditions ( Figure 2A). Control proteins (V H H B39, V H H-V H H B39-B39, mAbs CDA1 and MDX1388) were expressed in E. coli or HEK293-6E cells and purified by immobilized metal-ion affinity chromatography or protein A affinity chromatography (data not shown). Purified yields of A20 and A26 were in the range of 15-20 mg/L, and yields of the four dimeric V H Hs ranged from 80-114 mg/L ( Table 1). The SEC profiles of all V H Hs and V H H dimers produced single, monodispersed peaks devoid of aggregates (100% "monomeric") with V e s consistent for a V H H monomer to dimer transition and relative to protein standards ( Figure 2B; Table 1).
Binding of V H Hs and V H H-V H Hs to immobilized TcdA was demonstrated by ELISA ( Figure 2C; Table 1). The four dimeric constructs, A20-A20, A26-A26, A20-A26 and A26-A20, showed the strongest binding to TcdA with similar apparent EC 50 s of 32-47 pM. In contrast, monomeric A20 and A26 demonstrated significantly lower TcdA binding, with EC 50 s of 194 pM (A20) and 14.6 nM (A26). The off-rates (k d s, s −1 ) of all constructs were determined by SPR by flowing V H Hs, V H H-V H Hs or control CDA1 mAb over amine coupled TcdA for long dissociation times at 37°C ( Figure 2D; Table 1). Consistent with the ELISA results, A20 and A26 dissociated rapidly with k d s of 2.5 × 10 −2 s −1 and 7.5 × 10 −3 s −1 , respectively. All dimeric constructs clearly showed avid binding and demonstrated very slow dissociation rates as expected; however, subtle differences were evident. Of the dimeric proteins, A26-A26 dissociated the fastest followed by A20-A20 and A26-A20, which had similar k d s to the CDA1 IgG benchmark. The dissociation of A20-A26 was very slow (6.9 × 10 −6 s −1 ), which is at the instrument limit of detection, achieving only 4% dissociation after an hour and corresponding to an estimated half-life of 28.1 h (Table 1).

UPLC-SEC-MALS analysis of V H H-TcdA complexes
To gain further insight into the mechanism of action underlying the high neutralization potencies observed with A20-A26 dimers, several monomeric and dimeric V H Hs were incubated with TcdA-A2 in solution at 1:1 molar ratios for formation of antibody-TcdA-A2 complexes. The toxin/antibody mixtures were then subjected to SEC-MALS analysis to obtain observed molecular masses (M obs s) and retention times (T r s) of free (TcdA-A2 [Ag], antibody [Ab]), and complexed (Ag.Ab) species. M obs and T r values, complemented with neutralization data, were used to determine the types of binding complexes ("species" and "modes") formed ( Figure 4; Table 3).
As expected, the individual unmixed TcdA-A2 and V H H samples treated under the same conditions as the Ag-Ab mixes had M obs s very close to the theoretical M r values, indicating the lack of significant homotypic interactions for each individual protein. The M obs s for A20:TcdA-A2 and A26:TcdA-A2 complexes were consistent with a 1:1 TcdA-A2-V H H complex type (Ag.Ab) and the previous SEC data obtained for these complexes (Murase et al., 2014). The extent of complex formation was higher for A20 (80%) than for A26 (69%), consistent with its higher affinity for TcdA-A2 (Hussack et al., 2011). Homodimeric A20-A20 gave a major complex (92%) of (Ag) 2 .Ab type and a minor complex (8%) of Ag.Ab type. For the A26-A26 homodimer, however, it was not clear from the M obs s, whether the complex was of Ag.Ab or (Ag) 2 .Ab type for the major SEC species (76%) or of Ag.Ab or Ag type for the minor SEC species (24%) due to intermediary M obs values relative to expected M r s. Nonetheless, all possible complexes point to a lack of binding avidity. Thus, as in the case of A20 vs A26 V H H, a higher % of complex formation for A20-A20 may be due to a higher intrinsic affinity of A20 for TcdA-A2. Moreover, the lack of binding Frontiers in Microbiology 07 frontiersin.org avidity explains why A26 and A26-A26 have similar potencies, and in the case of A20-A20, it suggests the acquired neutralization capability of A20 upon homodimerization may be due to increased steric hindrance. A20-A26 formed predominantly an Ag.Ab type complex (91%), and along with its ultra-potent neutralization capability, indicates that both V H H domains in A20-A26 simultaneously engage with a single molecule of TcdA-A2 in a biparatopic fashion with 1:1 stoichiometry (Mode 1), as predicted from the original design based on the crystal structure ( Figure 4). Conversely, A26-A20 did not appear to form any biparatopic complexes, a result that is also consistent with the crystal structure predictions. It formed two major complexes at 55% and 43% with significantly different T r s but similar M obs s, indicative of (Ab) 2 .(Ag) 2 tetrameric (Modes 2, 3, and 4) and (Ab) 2 .Ag and/or Ab.(Ag) 2 trimeric complexes (Modes 5 and 6), respectively. Of the three tetrameric possibilities, one can adopt a cross-biparatopic binding arrangement (Mode 2), while the other two arrange in such a manner that excludes binding avidity (Modes 3 and 4, Figure 4). The far weaker neutralization potency of the A26-A20 construct relative to the A20-A26 biparatopic construct, on the one hand, and its similar neutralization capability to monomeric and homodimeric A20/A26 constructs, on the other, indicate that only one V H H domain in A26-A20 is able to bind to a single epitope in TcdA-A2 within any one particular molecular complex, thus excluding the possibility of Mode 2 binding. Interestingly, A20-A26 (G 4 S) and A20-A26 (G 4 S) 2 which had the same V H H-V H H orientations as A20-A26, formed complexes with M obs s consistent with Mode 1 and Mode 2 biparatopic Ab.Ag engagement. Molecular models based on the crystal structures indicate that the much shorter linker found in A20-A26 (G 4 S) when compared with A20-A26 would likely prevent the formation of complexes where both of the A20 and A26 V H H domains are able to bind to a single molecule of TcdA-A2 simultaneously. This likely explains why only 53% of the complexes detected for A20-A26 (G 4 S) show 1:1 stoichiometry (Ab.Ag; Mode 1), whereas 47% of the complexes show 2:2 stoichiometry ((Ab) 2 .(Ag) 2 ; Mode 2) based on the M obs and T r parameters. In comparison, 91% of the complexes formed by A20-A26 and A20-A26 (G 4 S) 2 , which both have linkers with similar lengths that are expected to allow both V H H domains to bind simultaneously to a single molecule of TcdA-A2, show 1:1 stoichiometry (Ab.Ag; Mode 1). Only up to 9% of the complexes have 2:2 stoichiometry ((Ab) 2 .(Ag) 2 ; Mode 2). A20-A26 (G 4 S) 3 was not analyzed by SEC-MALS, Comparison of neutralizing potency of the structure-guided A20-A26 construct containing a 12-amino acid linker, to A20-A26 constructs containing 5, 10, and 15 amino acid linkers. In all assays the TcdA concentration was 260 pM and antibody concentrations as described. Error bars indicate standard deviation (SD) of three technical replicates.
Frontiers in Microbiology 08 frontiersin.org because it could not be expressed in sufficient quantities. However, it is expected to possess a similar binding mode to that observed for A20-A26 and A20-A26 (G 4 S) 2 , because the longer linker length would be expected to allow both V H H domains to bind to a single molecule of TcdA-A2 simultaneously.

Discussion
In this work we used the crystal structures of V H Hs A20 and A26 in complex with C. difficle TcdA that we had previously determined to guide the design of a novel and ultra-potent toxin inhibitor which we have named A20-A26. This fusion protein is comprised of the A20 V H H at the N-terminus, followed by a 12 residue (GS) 6 linker and then the A26 V H H at the C-terminus. A20-A26 expressed well in mammalian cells (>100 mg/L), and was monodispersed and free of aggregates. A20-A26 also showed high, avidity-driven apparent binding affinities for immobilized TcdA and TcdA in solution. Most importantly, A20-A26 was extremely potent at neutralizing TcdA in cytotoxicity assays with Vero cells, achieving a ≥ 2,700-fold improvement in potency  Frontiers in Microbiology 09 frontiersin.org compared to monomeric A26 and A20 based on IC 99 . While this dramatic increase in potency is likely driven by avidity, the steric disruption of receptor binding due to the larger footprint of A20-A26 may also be important.
In contrast, and as predicted from the crystal structures, the reverse orientation of this construct with A26 positioned N-terminal to A20 was much less effective (29,000-fold less potent) at neutralizing TcdA cytotoxicity. The bivalent monoparatopic constructs (A20-A20 and A26-A26), and the anti-TcdA mAb (CDA1 or actoxumab), also show much poorer neutralization activity than A20-A26. To further explore the effects of length and flexibility in the GS-based linkers, we also produced A20-A26 constructs separated by spacer lengths of 5-, 10-, and 15-amino acids. All of these designs were almost as potent as the original version of A20-A26 containing a 12-residue linker, in which the length was suggested by a crude and simplistic model for the linker adopting a fully extended conformation. Combinations of individual V H Hs demonstrated poorer efficacy than the top design, pointing to a need for simultaneous engagement of both binding arms by the same antibody molecule to achieve potent cytotoxicity inhibition improvements, an observation reminiscent of analogous multivalent antibody designs evaluated for ricin (Herrera et al., 2015(Herrera et al., , 2016.
It is important to note that the IC 50 and IC 99 of A20-A26 were essentially the same, 0.16 nM and 0.27 nM, respectively, highlighting the highly cooperative nature of its binding to TcdA. In contrast, for other constructs such as A26, A26-A26, and A26-A20, wide gaps between IC 50 and IC 99 values were observed. In particular, an IC 50 of ~50 nM and an Values are mean ± SD with from two technical replicates. a Total may not add to 100% in some cases, due to minor peaks not identified in the analysis. Relative proportion is only for the representative data plotted in Figure 4. b Peaks with <5% relative proportion were omitted from species assignment.
IC 99 of ~7,900 nM, an increase of ~160-fold, was observed for the A26-A20 construct. Based on IC 99 values, A20-A26 showed a 2,700-fold improvement in potency compared to A26 monomer, and a similar improvement with respect to A26-A26. However, it displayed a much higher potency improvement (29,000-fold) relative to A26-A20 due to the much higher IC 99 of A26-A20 (7,866 nM [A26-A20] vs 719 nM [A26] vs 1,189 nM [A26-A26]). These results suggest that since the A20 moiety binds TcdA more strongly than the A26 moiety of the A26-A20 construct, the A20 moiety may actually sequester the A26 moiety away from the neutralizing A26 epitope, leading to a much higher IC 99 for A26-A20 compared to A26 or A26-A26. While this does not explain why A20, A20-A20, and A26-A20 have similar IC 50 s, it does point to the unexpected structural and functional subtleties that arise as a consequence of antibody multimerization, even in the case of very simple multimerization designs such as the linear, flexible linkers used in this study. SEC-MALS analyses, when considered alongside the neutralization data, reveal that A20-A26 achieves ultra-high TcdA inhibition through a biparatopic interaction with 1:1 stoichiometry (Mode 1; Figure 4A). SEC-MALS also suggests that A20-A26 constructs with 10 [A20-A26 (G 4 S) 2 ] and 12 [A20-A26] residue linkers likely have sufficient length and flexibility to allow both V H H domains to simultaneously engage with the same molecule of TcdA to form a biparatopic Ab.Ag complex. In comparison, the A20-A26 construct with a shorter five-residue linker [A20-A26 (G 4 S)] does not appear to readily bind a single molecule of TcdA with high affinity, presumably because the shorter linker prevents both V H H domains from simultaneously engaging with the same molecule of TcdA. Somewhat surprisingly, this construct is still able to neutralize TcdA with a similar level of potency to A20-A26 and A20-A26 (G 4 S) 2 . SEC-MALS suggests that this is accomplished by two biparatopic V H H-V H H fusion protein molecules simultaneously engaging with two TcdA molecules in a second, biparatopic (Ab) 2 . (Ag) 2 interaction (Mode 2), which is presumably nearly as effective a neutralizing mechanism as the intramolecular biparatopic one (Mode 1). Both Mode 1 and Mode 2 binding arrangements involve avidity in solution, i.e., in the toxin neutralization assay setting. This explains why the overall impact of linker length on TcdA neutralization potency for all of these A20-A26 constructs was largely negligible at the TcdA concentrations used in the Vero cell neutralization assays. However, given that Mode 1 biparatopic complex structures are less disturbed by changes in concentrations than Mode 2 biparatopic complex structures, it is likely that A20-A26 would be a more potent inhibitor than A20-A26 (G 4 S) at lower, pathologically relevant, TcdA concentrations. The opposite orientation (A26-A20) and monoparatopic controls (A20-A20, A26-A26) are much less effective at neutralizing TcdA due to a lack of avidity when binding to the toxin, although they did display high apparent binding affinities from bivalent engagement in artificial settings (ELISA, SPR). Based on our previous crystal structures (Ho et al., 2005;Greco et al., 2006;Murase et al., 2014) and the work of others (Pruitt et al., 2010;Chumbler et al., 2016;Chen et al., 2022;Kordus et al., 2022), we hypothesize that the bidentate engagement of A20-A26 with TcdA may lead to ultra-potent toxin inhibition through three possible mechanisms: (i) A20-A26 may block enough of the carbohydrate binding pockets in TcdA to substantially disrupt binding interactions between TcdA and cell-surface carbohydrate receptors; (ii) A20-A26 may inhibit pH-induced TcdA conformational changes that are required for cellular toxicity (Pruitt et al., 2010;Chumbler et al., 2016;Chen et al., 2022); and (iii) we previously deduced the importance of the extreme C-terminus of TcdA playing a role in toxicity (Murase et al., 2014), based on the A26 V H H monomer binding at this region of the toxin and neutralizing TcdA relatively effectively (Hussack et al., 2011). It is also instructive to consider studies on anti-ricin antibodies, where Herrera et al. (2016) showed that separate V H H domains in bispecific anti-ricin antibody fusions neutralized the toxin through a mechanism that involved a single antibody fusion binding more than a single toxin molecule.
In conclusion, we have validated a structure-guided approach for designing a highly-potent C. difficle TcdA inhibitor. In the process we show the importance of V H H orientation and geometry in biparatopic constructs, the impact of linker length, and the mode of V H H:toxin interactions on achieving ultra-high neutralization potencies.

Data availability statement
The original contributions presented in the study are included in the article/supplementary material, further inquiries can be directed to the corresponding authors.

Author contributions
GH and JT conceived experiments, analyzed data, and wrote the manuscript. MR performed protein expression and purification, SDS-PAGE, SEC, ELISA, and neutralization assays, analyzed data, and edited the manuscript. HF performed SPR experiments and analyzed data. TM and LE performed SEC experiments and contributed reagents. JS designed and analyzed SEC-MALS experiments and edited the manuscript. KN conceived experiments, analyzed data, and edited the manuscript. All authors contributed to the article and approved the submitted version. Publisher's note All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.