Crystal structures reveal that Lewis-x and fucose bind to secondary cholera toxin binding site – in contrast to fucosyl-GM1

Cholera is a life-threatening diarrhoeal disease caused by the human pathogen Vibrio cholerae. Infection occurs after ingestion of the bacteria, which colonize the human small intestine and secrete their major virulence factor - the cholera toxin (CT). Recent studies suggest that the GM1 receptor may not be the only target of the CT, and that fucosylated receptors such as Lewisx (Lex) and histo-blood group antigens may also be important for cellular uptake and toxicity. However, where and how Lex binds to the CT remains unclear. Here we report the high-resolution crystal structure (1.5 Å) of the receptor-binding B-subunit of the CT bound to the Lex trisaccharide, and present matching SPR data for CT holotoxins. Lex, and also L-fucose alone (at 500-fold molar excess), bind to the secondary binding site of the toxin, distinct from the GM1 binding site. In contrast, fucosyl-GM1 mainly binds to the primary binding site due to high-affinity interactions of its GM1 core. The two binding sites are likely connected by allosteric cross-talk, potentially affecting toxin uptake. We also discuss why secretors are protected from severe cholera. Author summary Cholera is a severe diarrhoeal disease that is still a major killer in many parts of the world, especially in regions struck by natural disasters and wars. However, some individuals experience milder cholera symptoms. These so-called ‘secretors’, who have blood group antigens also in their bodily fluids like their saliva and the slimy mucus layer covering their stomach and intestines, appear to be somewhat protected. Here we present detailed atomic structures of cholera toxin and quantitative binding data that give clues of the protective mechanisms. Interactions of the protein toxin with sugar molecules are of crucial importance both for toxicity and protection. In addition, we identify a new tool for biochemical studies, and lay the groundwork for the design of cholera drugs and vaccines that may save countless human lives.

contrast, fucosyl-GM1 mainly binds to the primary binding site due to high-affinity interactions 23 of its GM1 core. The two binding sites are likely connected by allosteric cross-talk, potentially 24 affecting toxin uptake. We also discuss why secretors are protected from severe cholera. 25 26 Author summary 27 Cholera is a severe diarrhoeal disease that is still a major killer in many parts of the world, 28 especially in regions struck by natural disasters and wars. However, some individuals experience 29 milder cholera symptoms. These so-called 'secretors', who have blood group antigens also in their 30 bodily fluids like their saliva and the slimy mucus layer covering their stomach and intestines, 31 appear to be somewhat protected. Here we present detailed atomic structures of cholera toxin and 32 quantitative binding data that give clues of the protective mechanisms. Interactions of the protein 33 Introduction to the GM1 binding site, the BGA binding site or both 32 . Likewise, it is unknown where and how 80 L-fucose binds to the CT. 81 Here we investigated the binding of CTB to the Le x trisaccharide, L-fucose and the fucosyl-GM1 82 oligosaccharide (os) (Figure 1) by X-ray crystallography and surface plasmon resonance (SPR) 83 spectroscopy. We hypothesized that Le x and related sugars interact solely with the secondary 84 binding site of CTB based on their similarity to known toxin-bound BGAs and human milk 85 oligosaccharides (HMOs) 19 . To test this hypothesis, we crystallized CTB with the Le x trisaccharide 86 or L-fucose. Indeed we found both sugars in the secondary, but not in the primary binding sites. In 87 contrast, fucosyl-GM1os occupied the primary GM1 receptor site (and some of the secondary 88 binding sites, via its fucose), as expected due to the similarity of the two compounds. It binds the 89 CT with the same strong affinity as GM1 and with a two-fingered grip, but the fucose residue 90 contributes little to their interaction, serving more as decoration. 91 92

CTB in complex with Le X 95
To study the interaction of CT with Le x , we crystallized purified classical CTB (cCTB) in complex 96 with Le x triaose. Crystals were obtained in two different crystallization conditions. They belonged 97 to space group P212121 and contained two CTB pentamers in the asymmetric unit. Therefore, we 98 can observe 20 crystallographically distinct primary binding and secondary binding sites. The 99 protein structures show the typical "doughnut-shaped" CTB structure of five symmetrically 100 arranged B subunits, each consisting of two three-stranded antiparallel β-beta-sheets with α-helices 101 on both sides 13 . The structures were determined to 2.0 Å and 1.5 Å resolution, respectively. In both 102 crystal forms, Le x is only observed in the secondary binding sites of the CT. The binding mode 103 and interactions of Le x described here are based on the 1.5 Å resolution structure, which was 104 refined to a high quality model (Figure 2 a, Table 1; Rfree = 22.4%). Generally, the structure is well 105 defined except for the flexible loop comprising residues 50-61 and the C-terminal asparagine 106 residue, which exhibits some disorder. Inspection of the electron density maps revealed the 107 presence of Le x triaose in eight of ten secondary binding (Figure 2 b). One additional binding site 108 contained electron density in sufficient quality to place the terminal L-fucose, and the last 109 secondary binding site was blocked by crystal contacts. No ligand density was observed in any of 110 the primary binding sites, even at low sigma cut-offs.  Table S1) 19 . Its reducing end (N-acetylglucosamine; GlcNAc) points towards residue 47. The 114 blood-group antigen binding site of CT targets Lewis antigens mainly through the α1,3 fucose 115 (although Le y itself can bind the CT in two orientations, where either Fucα2 or Fucα3 can serve as 116 main interaction points). As observed for cCTB bound to Le y and A-Le y (5ELB, 5ELD 19 ), the 117 fucose forms hydrogen bonds to Gln3# from the adjacent B-subunit and to the backbones of 118 residues 47 and 94 (Figure 2  TALON-purified cCTB was co-crystallized with L-fucose in a molar ratio of 1:500 (B-subunit to 135 ligand). The crystals belong to space group C2 and contained two CTB pentamers in the 136 asymmetric unit. The structure was refined to 1.95 Å and an Rfree value of 23.7% (Table 1). The 137 electron density of the loop regions and the C-terminal Asn103 were more disordered compared 138 to the cCTB-Le x structure, however, overall the structure is well defined. Electron density for β-139 L-fucose is observed in nine of ten secondary binding sites (Figure 3 a,b, Supplementary Table  140 S2). We did not observe electron density corresponding to L-fucose in the GM1 binding site, 141 confirming that fucose only binds to the secondary binding site. Additionally, L-fucose was found 142 sandwiched between two B-pentamers and covalently attached to some of the N-terminal residues. 143 The latter is likely due to non-enzymatic glycosylation 40 . These interactions are unlikely to be 144 biologically relevant and probably caused by the high molar excess of L-fucose added during 145 crystallization. 146

CTB in complex with fucosyl-GM1os 147
Our data strongly suggest that the toxin secondary binding site is the sole binding site for Le x and 148 similar sugars. There is, however, one fucosylated oligosaccharide that would be expected to bind 149 to the primary binding site, namely fucosyl-GM1os. Fucosyl-GM1 binds CT almost as strongly as 150 GM1 41 . We crystallized TALON-purified cCTB with fucosyl-GM1os in a 1:10 molar ratio (B-151 subunit to ligand), yielding crystals that diffracted to 1.6 Å resolution. The structure was refined 152 to an Rfree value of 21.4%. Inspection of the electron density revealed that fucosyl-GM1os indeed 153 binds to the primary binding site, similarly to GM1os, with the additional fucose residue facing 154 outwards toward the solvent (Figure 3 c). The electron density for the fucose residue is less well 155 defined compared to the other sugar residues (Figure 3 d), suggesting that it binds more weakly. It 156 also has higher B-factors (average B-factors for sugar ring atoms in chain A: Fuc>GalNAc>Gal, 157 28.6>20.8>17.1 Å 2 ). This explains the small difference in CT affinity between fucosyl-GM1 and 158 GM1, as the fucose residue is not a major contributor to binding. In addition to the primary binding 159 sites, we observe fucose binding in some of the secondary binding sites of cCTB. In these sites, 160 fucosyl-GM1os is not well resolved, however, electron density extending from the fucose is 161 compatible with a larger oligosaccharide like fucosyl-GM1os. 162

CTB and Le x triaose 164
To determine the binding affinity of cCTB to Le x , we performed SPR experiments, in which the 165 toxin B-pentamer was immobilized on the SPR chip and the sugar was injected as the analyte over 166 a range of different concentrations. Le x triaose was found to have a lower binding affinity to cCTB 167 than Le y tetraose or A-Le y pentaose, which both feature a second fucose residue (Kd=10 ± 3 mM 168 (Le x triaose), compared to 1.05 ± 0.04 mM (Le y ) and 2.2 ± 0.1 mM (A-Le y ) 19 Table 2). Native folding of these variants was confirmed by 176 circular dichroism (CD) spectroscopy (Supplementary Figure S2). Toxin variants or wild-type CT 177 were immobilized on SPR chips, before injecting oligosaccharides at different concentrations. 178 First, we showed that substitution of W88A in the primary binding site prevented GM1os binding 179 ( Figure 4). Since all other toxin variants bound GM1os, this confirmed an intact primary binding 180 site. We then compared binding of Le x oligosaccharides to wild-type CT and CT variants. 181  Figure S1). This result was expected, since there is no major structural difference 186 between free CTB and the B-pentamer in the holotoxin 13,42 . In our crystal structures, Le x and 187 related sugars pre-dominantly bind with the α-anomer of the reducing end GlcNAc, whereas 188 relevant glycoconjugates on the cell surface are β-glycosidically linked to proteins or lipids. Le x 189 tetraose, with a fixed β-anomer, and Le x triaose bound equally well to cCT (Kd=8.1 ± 3.2 mM for 190 triaose, 7.7 ± 0.5 mM for tetraose), suggesting that the triaose core is mainly responsible for 191 binding and that the linkage does not significantly affect binding affinities. 192 Next, we determined binding affinities for all toxin variants using Le x tetraose ( Figure 4, Table 2). 193 The primary binding site variant W88K was found to bind Le x tetraose with an affinity comparable 194 to wild-type CT (Kd=8.6 ± 0.1 mM for W88K, 7.7 ± 0.5 mM for cCT), whereas H18A, which 195 features a mutation in the secondary binding site, exhibited significantly reduced binding to Le x 196 tetraose (Kd>50 mM for H18A, 7.7 ± 0.5 mM for cCT), in agreement with our structural data. This 197 reduced binding affinity is not due to disruption of the primary binding site, since GM1os binding 198 was found to be even stronger than for the wild-type protein (Kd=39.5 ± 0.9 nM for H18A, 57.7 ± 199 0.3 nM for cCT; Figure 4). The double mutant H18AH94A, which was designed to disrupt the 200 water network in the binding site and additionally precludes H-bond formation to the His94 side 201 chain, only bound Le x tetraose at the highest analyte concentration applied (40 mM), but bound to 202 GM1os almost as strongly as wild-type CT (Kd=80.0 ± 0.39 nM for H18AH94A, 57.7 ± 0.3 nM 203 for cCT). Our results confirm Le x binding to the secondary binding site of CT and suggest cross-204 talk between the primary and secondary binding sites, since mutations in the latter affect GM1 205 binding. 206

Le x and similar fucosylated sugars bind to the secondary binding site of CT 208
We set out to explore the molecular interaction of CT with fucosylated sugars, in particular Le x . 209 Already twenty years ago, CTB was shown to bind fucose 43 . More recently, evidence was 210 presented that BGAs may act as functional toxin receptors for the related LTB 44,45 , and fucosylated 211 glycan structures were shown to be functionally active and cause cellular uptake of CT 30,32,33 . For 212 example, CTB binding to jejunal epithelial cells can be blocked by the Le x trisaccharide or a 213 monoclonal antibody against Le x 32 . Crosslinking and immunoprecipitation of CTB-bound cellular 214 proteins revealed that CTB binds to glycoproteins modified with Le x 32 . It was, however, unclear 215 how Le x binds to CT. Also G33D, a CTB variant with greatly reduced affinity to GM1 15,46 , showed 216 weaker binding to Le x 32 . Since Gly33 is located in the GM1 binding site, it was suggested that 217 also Le x might bind to the primary binding site 32 . To explore this possibility, Cervin et al. 218 performed competition experiments with plate-bound GM1 or Le x and CTB pre-incubated with 219 the free sugars 32 . Competition experiments were also performed with human granulocytes, T84 220 and Colo205 cells. Despite the fact that these cell lines only express low levels of GM1, pre-221 incubation of CTB with GM1os had an inhibitory effect. Taken together, these results were 222 interpreted to indicate that GM1 competitively blocks fucosylated sugars from binding to the CT 223 primary binding site 32 . However, it should be noted that even at high concentrations, GM1os could 224 not completely block CTB binding and Le x could not block binding to plate-bound GM1. 225 Here we show that Le x binds solely to the secondary binding site of cCTB, and even L-fucose alone 226 (when co-crystallized in 500-fold molar excess) did not bind to the primary binding site. Moreover, 227 SPR experiments with toxin variants confirmed that disruption of the secondary binding site, but 228 not the primary binding site, lowered the affinity to Le x ( Figure 4, Table 2). We also found that 229 Le x bound more weakly to CTB than Le y or A-Le y , which is in good agreement with inhibitor 230 studies identifying Le y as the most potent small-molecule inhibitor of CTB 33 . The binding mode 231 of Le x is highly similar to that of Le y tetraose and A-Le y pentaose 19 , which both feature an 232 additional fucose residue (Figure 2 c). The partial competition of GM1 and Le x observed by Cervin 233 et al. 32 is most likely due to allosteric cross-talk, in line with results from a recent NMR study of 234 the homologous LTB 47 . Similarly, we found that the CT secondary binding site variant H18A 235 exhibits increased GM1os binding. Furthermore, even a small number of receptors in the cell-236 based assays could allow CT uptake, due to the high affinity of GM1 to CTB, which could explain 237 the findings in the recent study 32 . However, even though Le x and related fucosylated glycans and 238 glycoconjugates do not bind to the primary CT receptor site, clearly additional fucose residues 239 allow for additional attachment points that could interfere with toxin uptake, explaining the strong 240 potency of a fucose-based polymer as CT inhibitor 33 . 241 addition of a hydroxyl group to C6 was tolerated, but the removal of C6 resulted in a complete 250 loss of CTB binding suggesting the importance of a hydrophobic patch within the binding pocket. 251

Comparison of ligand structures to structure-activity relationship data
Addition of a methyl group to OH1, locking the anomeric hydroxyl group in the β-configuration, 252 resulted in less efficient CTB binding compared to L-fucose or α-locked fucose. Finally, the SAR 253 study showed increased binding of α1,2 linked fucose compared to α1,3 linked fucose 33 . 254 Our current ligand structures and previous structures of cCTB in complex with Le y or A-Le y (5ELB 255 and 5ELD 19 ) show that fucose OH3 can form one or two hydrogen bonds to His94 and to a 256 conserved water molecule (not shown in Figure 2), explaining the observed reduced CTB binding 257 upon its removal in the SAR study. OH4 was reported to contribute the strongest to CTB binding 33 , 258 which is in good agreement with the fact that OH4 forms two hydrogen bonds to backbone atoms 259 of residues 47 and 94 (Figure 2 d). The crystal structures also identify a match for the proposed 260 key hydrophobic residues interacting with C6 in Phe48, possibly together with Ala46 (Figure 2 d). 261 cCTB binds L-fucose with its free anomeric hydroxyl group in the β-configuration (Figure 3 b), 262 thus free fucose is not limited to bind in the α-configuration. However, locking the fucose in the 263 β-configuration by the addition of a methyl group would cause steric clashes with residues Thr47 264 and Gly45, explaining why α-linked fucose bound stronger to CTB than β-linked fucose 33  suggest that CTB can bind sugars at both binding sites at the same time, specifically Le y tetraose 292 and galactose 19 . However, it has not yet been shown if this also holds true for GM1os. A recently 293 published hetero-multivalent binding model suggests that CTB first binds to a high-affinity ligand 294 such as GM1. Subsequent binding occurs much more readily (up to 10,000-fold faster) also for 295 low-affinity ligands like GM2, since binding is then confined to the 2D membrane surface 55 . In 296 fact, cooperativity was found to be enhanced for a heterogeneous mixture of ligands 55 . In a 297 biological context, it seems more likely that the toxins first bind to the broadly available 298 fucosylated structures on the cell surface or mucus layer until they can bind to the few available 299 GM1 receptors, preventing the flow in the small intestine to remove unbound toxins from the 300 intoxication site 19  Allosteric cross-talk between the two binding sites may facilitate toxin binding and uptake. GM1 306 binding to CTB is known to be positively cooperative 57-59 , and the small structural changes 307 associated with cooperative binding to the primary site could easily be transmitted to the secondary 308 binding site, for example via the helix connecting the two sites or by stabilization of the loop region 309 comprising residues 55-60 60 . Recent NMR data for the homologous LTB from E. coli (ETEC) are 310 consistent with such a cross-talk between binding sites 47 . Likewise, CT variant H18A showed 311 increased GM1os affinity compared to wild-type CT, even though the mutation is in the secondary 312 binding site. Cross-talk between the two sites could also explain the partial competition of GM1 313 and fucosylated receptors and the reduced binding of G33D to fucosylated structures observed by 314 Cervin et al. 32 . 315

Cholera blood group dependence: Why are secretors protected? 316
Many diseases show blood group association 61 , and cholera is the textbook example. Based on 317 high-resolution structures of CTB in complex with BGAs, we recently elucidated the molecular 318 basis of this phenomenon, i.e., why individuals with blood group O experience the most severe 319 symptoms 19 . Le y , characteristic for blood group O, exhibits a dual-binding mode and binds more 320 strongly to CTB than the BGAs of blood groups A or B 19 , allowing for more efficient cellular 321 uptake and more severe intoxication 31 . The H-antigen exists in two types, type-1 and type-2, with different linkages, both of which contain 336 an α1,2-linked fucose. Addition of a second fucose residue (Fucα4 or Fucα3) then gives rise to 337 Le b/y determinants, respectively. If the same fucose is added to H-antigen precursors instead, this 338 yields Le a/x epitopes. So far, all evidence points towards only type-2 antigens being able to bind to 339 the CT 11,19,20 , focusing the attention on Le y and Le x . The lower affinity of Le x compared to the 340 other BGAs makes it unlikely that Le x binding is the cause for more efficient CT uptake in non-

Conclusions and perspective 354
The importance of fucosylated sugars for cholera intoxication has recently been in the 355 limelight 19,23-25,30-33 , but there has been limited structural information on their interaction with the 356 CT. Here we show that fucosylated receptors, except fucosyl-GM1, bind solely to the secondary 357 CT binding site, in contrast to recent suggestions 32 . We provide detailed information on how Le x 358 and L-fucose bind to CTB, and corroborate our findings with quantitative binding data of holotoxin 359 variants. In addition, we present the first CT variant -H18Awith reduced affinity to fucosylated 360 sugars, but increased affinity to GM1os. This not only lends further support to our hypothesis of a 361 possible cross-talk between the primary and secondary CT binding sites 68 , but may also be of great 362 practical value. Currently CT is widely used as a marker of GM1 and lipid rafts 69 . We suggest to 363 replace wild-type CTB with variant H18A as more specific GM1 marker, to limit the number of 364 false-positive results caused by the interaction with non-GM1 receptors. Furthermore, CT variants 365 H18A and H18AH94A, being devoid of a functional secondary binding site, are predicted to 366 facilitate the analysis of cellular uptake in cellular and organoid models 31,70 by allowing 367 discrimination of the effects caused by interaction with primary and secondary sites. Finally, this 368 study is expected to facilitate the design of more potent fucose-based CT inhibitors and CTB-369 containing vaccines, which may save countless human lives. 370

392
The gene for cCTB (Uniprot: Q57193) was heterologously expressed in E. coli BL21 (DE3) using 393 a cCTB-pET21b+ construct. For protein production, cells were grown at 37°C in LB medium 394 containing ampicillin until OD600 nm of 0.5 was reached. The temperature was reduced to 25°C and 395 IPTG was added to a final concentration of 0.5 mM to start cCTB production. 396 The genes for CT and CT variants (W88K, H18A, H18AH94A) were heterologously expressed in 397 OverExpress™ C43 (DE3) cells (Sigma) using pARCT5 or pARCT5 derivatives. For protein 398 production, cells were grown at 37°C in TB medium containing chloramphenicol until OD600 nm of 399 2.0 was reached. L-arabinose was added to a final concentration of 0.2% (w/v) to start holotoxin 400 production.      weighted Fo-Fc electron density map for fucosyl-GM1os (grey mesh, contoured at 3.0σ, generated 843 before placing the ligand) and Trp88 shown in stick representation. Carbohydrate residues are 844 labelled in italics. The terminal fucose and glucose residues show weaker electron density 845 compared to the four core residues of fucosyl-GM1os. 846 847