Crystal structure of botulinum neurotoxin subtype A3 cell binding domain in complex with GD1a co‐receptor ganglioside

Botulinum neurotoxins (BoNTs) are one of the most toxic proteins known to humans. Their molecular structure is comprised of three essential domains—a cell binding domain (HC), translocation domain and catalytic domain (light chain) . The HC domain facilitates the highly specific binding of BoNTs to the neuronal membrane via a dual‐receptor complex involving a protein receptor and a ganglioside. Variation in activity/toxicity across subtypes of serotype A has been attributed to changes in protein and ganglioside interactions, and their implications are important in the design of novel BoNT‐based therapeutics. Here, we present the structure of BoNT/A3 cell binding domain (HC/A3) in complex with the ganglioside GD1a at 1.75 Å resolution. The structure revealed that six residues interact with the three outermost monosaccharides of GD1a through several key hydrogen bonding interactions. A detailed comparison of structures of HC/A3 with HC/A1 revealed subtle conformational differences at the ganglioside binding site upon carbohydrate binding.

Botulinum neurotoxins (BoNTs) are one of the most toxic proteins known to humans. Their molecular structure is comprised of three essential domains-a cell binding domain (H C ), translocation domain and catalytic domain (light chain) . The H C domain facilitates the highly specific binding of BoNTs to the neuronal membrane via a dual-receptor complex involving a protein receptor and a ganglioside. Variation in activity/toxicity across subtypes of serotype A has been attributed to changes in protein and ganglioside interactions, and their implications are important in the design of novel BoNT-based therapeutics. Here, we present the structure of BoNT/ A3 cell binding domain (H C /A3) in complex with the ganglioside GD1a at 1.75 A resolution. The structure revealed that six residues interact with the three outermost monosaccharides of GD1a through several key hydrogen bonding interactions. A detailed comparison of structures of H C /A3 with H C /A1 revealed subtle conformational differences at the ganglioside binding site upon carbohydrate binding.
Botulinum neurotoxin (BoNT) causes the disease botulism by specifically targeting cells of the neuromuscular junction and cleaving a soluble N-ethylmaleimide-sensitive factor attachment protein receptor (SNARE) protein(s). Botulism is characterised by a descending flaccid paralysis that can be fatal without medical intervention. Considering that there are only a low number of incidences of botulism reported each year [1], there has not been a need for mass vaccination; consequently, it has been possible to use BoNT as a therapeutic for the treatment of hyperactive neuromuscular disorders. BoNTs are generally produced by Clostridium botulinum; however, bont gene clusters have recently been identified in different bacterial species [2,3]. There are currently seven distinct BoNT serotypes produced by C. botulinum, /A-/G. Serotypes /A, /B, /E and /F are associated with human botulism making them potential candidates for the development of BoNT-based therapeutics. These serotypes are further divided into subtypes (e.g., /A1-/A8) based on minor amino acid variations that may affect toxicity [4][5][6]. BoNTs are expressed as a single polypeptide chain (150 kDa) that is activated by post-translational cleavage into a di-chain consisting of a 50 kDa light chain (LC) linked to a 100 kDa heavy chain (HC) by a disulphide bond. The LC possesses zinc-endopeptidase activity, whereas the HC comprises two domains -an N-terminal translocation domain (H N ) and a C-terminal cell binding domain (H C ).
Gangliosides constitute 10-20% of neuronal cell membranes [7] with both GD1a and GT1b present at the neuromuscular junction [8]. They are amphiphilic molecules with a lipophilic ceramide tail that is inserted into the neuronal membrane, conjugated to a hydrophilic oligosaccharide moiety that is displayed extracellularly [8]. GT1b and GD1a differ by only one monosaccharide, with the latter lacking the third sialic acid (Sia) (Fig. 1B). All but one BoNT serotype (BoNT/D) bind to a ganglioside receptor and a protein receptor (dual receptors) via the H C domain, with the former occurring at a conserved ganglioside binding site (GBS) [9]. Crystallographic studies of H C /A1 alone and in complex with GT1b and GD1a revealed that a majority of the interacting amino acids did not alter conformation [10,11]. Upon binding to the target cell, the BoNT is internalised into an endosome and a drop in pH triggers conformational changes in the H N domain. One significant change involves a switch of buried a-helical regions into a b-hairpin structure that facilitates the embedding of the H N into the endosomal membrane [12]. The LC domain is then translocated into the cytosol of neurons at the neuromuscular junction where it catalyses the cleavage of its target SNARE protein [13]. Previously we had reported the crystal structure of H C /A3 at 1. 6 A resolution [14]; here, we report the structure of H C /A3 in complex with GD1a to 1.75 A resolution and highlight the key structural changes that occur upon ganglioside binding.

Protein expression and purification
The binding domain of BoNT/A3 (residues 866-1292; 'H C / A3') was cloned into the pJ401 vector as previously described [14]. The construct was transformed into BL21 E. coli cells and grown at 37°C in 0.5 L TB. Cultures were induced with 1 mM IPTG upon reaching an OD 600 of 0.6 followed by incubation at 16°C for 16 h. Cells were lysed in 50 mM Tris pH 7.4, 0.5 M NaCl. Target protein was captured on a GE HisTrap column and further purified by size-exclusion chromatography using a GE Superdex 200 column and 50 mM Tris pH 7.4, 150 mM NaCl.

Protein crystallisation
Protein crystallisation was carried out using the sitting drop vapour diffusion method at 16°C in 96-3 well crystallisation intelli-plates. H C /A3 (5 mgÁmL À1 ) was added to 1.5 mM GD1a ganglioside sugar (Elicityl OligoTech) and incubated for 30 min at room temperature prior to setting up crystallisation trials with the following screens from Molecular Dimensions: PACT Premier, Morpheus I, Morpeus II, BCS, SGI and MIDAS+. Several crystal clusters formed in the BCS screen, with the best crystals observed in condition A10 (0.1 M sodium acetate, 22 % v/v PEG smear broad). These were optimised using 1 : 1, 2 : 1 and 1 : 2 protein: reservoir ratios. Crystals were mounted onto a cryo-loop without cryo-protection and flash-frozen for storage in liquid nitrogen.

X-ray diffraction data collection and structure determination
Crystals were kept at 100 K using a liquid nitrogen jet while a total of 7200 X-ray diffraction images were collected at 0.1°oscillations with exposures of 0.02 s using the I04 protein crystallography beamline at Diamond Light Source (Didcot, UK). Data processing was carried out using DIALS [15], and the structure was solved by molecular replacement with PHASER [16] using the structure of H C /A3 (PDB code: 6F0O) [14] as the search model. The model was refined with REFMAC5 [17] and manually fitted in COOT [18] as part of the CCP4 program suite [17]. Structure validation was performed using MolProbity [19], and figures were produced using the CCP4mg moleculargraphics software. Crystallographic data collection and refinement statistics are summarised in Table 1.

Results and Discussion
Crystal structure of H C /A3 in complex with GD1a oligosaccharide The crystal structure of H C /A3-GD1a was solved by molecular replacement in space group P2 1 2 1 2 1 to a

resolution of 1.75
A, with one molecule in the asymmetric unit (Table 1). An initial round of refinement revealed large, positive electron density within the GBS that indicated the presence of GD1a. Monosaccharides were modelled in the observed electron density (Fig. 1A) and subsequent rounds of refinement improved the map significantly. The quality of the electron density map was very good throughout the structure, with only two small loop regions (residues 1222-1228 and 1267-1271) that were not observable. The overall fold of the protein is very similar to H C / A3 and other BoNT binding domain structures [10,14,17] where the N-terminal half contains a 14 bstrand 'jelly-roll fold' and the C-terminal half folds into a 'b-trefoil' with a b-hairpin that contains the conserved GBS (H..SxWY..G) ( Fig. 2A). With regard to the GD1a oligosaccharide, Sia 5 -Gal 2 were clearly defined by the electron density and modelled with lower average B-factors for monosaccharides interacting with the protein. Sia 5 has a B-factor of 45.4 A 2 , Gal 4 29.2 A 2 , GalNAC 3 43.5 A 2 and Gal 2 57.7 A 2 respectively. Glu 1 is partially accounted for by the electron density with an average B-factor of 73.3 A 2 , whereas there was insufficient positive electron density to model Sia 6 .
Six residues of H C /A3 formed seven hydrogen bonds with GD1a (Table 2), with a conserved water molecule involved in a bridging interaction between GD1a and Leu 1250. Leu 1250 interacted with both Sia 5 at O4 and its glycosidic bond with Gal 4 (Fig. 2B). The hydroxyl group of Tyr 1263 formed a hydrogen bond with the carboxylic acid of Sia 5 (2.7 A) and the mainchain peptide of Gly 1275 formed an additional hydrogen bond with this monosaccharide (2.9 A) (Fig. 2B). Phe 1248, Ser 1260 and His 1249 all formed hydrogen bond interactions with Gal 4 (2.5, 2.7, and 3.1 A, respectively) ( Fig. 2C) and glucose (Glu) 1199 formed hydrogen bonds with both Gal 4 and N-acetylgalactosamine (GalNAc) 3 (2.7 and 2.5 A, respectively) ( Fig. 2C,D). Apart from these strong interactions, ring stacking interactions between Trp 1262 and Gal 4 and GalNAc 3 were also observed ( Fig. 2A).

Structural differences between H C /A3 bound and unbound to GD1a
The structures of H C /A3-GD1a (present structure) and H C /A3 (PDB code: 6F0O) are conformationally very similar, with an RMSD of 1.0 A for 403 C a atoms. There are, however, some noticeable differences in and around the GBS (Fig. 3A,B). Residues 1195-1196 and 1273-1277 are now clearly visible in the electron density for the H C /A3-GD1a complex. The latter is located in a loop near the GBS that interacts with GD1a, which would be consistent with increased order to a flexible loop. The formation of a hydrogen bond between Gly 1275 and Sia 5 is accompanied by flipping of positions for Phe 1274 and Thr 1273 (Fig. 3C) and loss of a water molecule.
Beyond the loop, there are additional differences in the H C /A3-GD1a structure compared to the unbound H C /A3 structure. For example, Trp 1262 is positioned some 4 A away from Gal 4 ; Tyr 1263 has moved 1.1 A to within hydrogen bonding distance of Sia 5 , displaced a water molecule and formed a further interaction with the backbone amine of Phe 1248; and the side chain of Asn 1264 has rotated~180°to form a hydrogen bond with the backbone amine of Phe 1245 (Fig. 3D). Elsewhere in the complex structure, His 1249 appears closer to the GBS and forms two hydrogen bonds with Gal 4 , and several hydrophobic residues (Phe 1113, Val 1198 Glu 1199, Tyr 1251 and Trp 1262) come together to form a shallow groove occupied by Sia 5 ? Gal 4 . This is further contributed by the C c atom of Glu 1199 that is rotated by~110°a bout C a -C b bond adapting a different rotamer (Fig. 3B), and the carboxylate of this residue also forms a hydrogen bond with GalNAc 3 .  Table 2. Hydrogen bonding distances observed for ganglioside binding in H C /A3-GD1a, H C /A1-GT1b and H C /A1-GD1a structures. Watermediated interactions are indicated in italics by a '-H 2 O molecule (n 1 , n 2 )' where n 1 is the distance between the amino acid residue and the water, and n 2 is the distance between the water and monosaccharide. D Indicates they are the equivalent water molecule for each structure. Comparison to H C /A1 and H C /A3 structures in complex with GT1b or GD1a

Monosaccharide
Cell-based assays have shown BoNT/A3 to have 107fold and 4-fold less activity compared to BoNT/A1 in iCell neurons and HIP neurons, respectively [20]. Considering that both LC and HC can effect potency separately [21], it is possible that the H C domain may be partly responsible for this difference in activity between these BoNT subtypes. For H C /A1, the structure in complex with GD1a or GT1b gives an RMSD of only 0.5 and 0.3 A (for C a atoms) compared with the uncomplexed molecule, respectively. GD1a and GT1b differ by just 1 monosaccharide (Fig. 1B) and both exhibit high affinity for the toxin [22]. In both ganglioside-bound structures, Sia 6 is stabilised by hydrogen bonding to Trp 1266 and Arg 1276 of H C /A1 (Table 2), whereas for H C /A3, Sia 6 could not be modelled, suggesting a lack of hydrogen bonding with the corresponding residues, Trp 1262 /A3 and Arg 1272 /A3 . Furthermore, upon binding ganglioside, there is an accompanying shift of Trp 1262 /A3 and Trp 1266 /A1 in opposite directions, and together with His 1249 /A3 and His 1253 /A1 , respectively, these residues form the opening of a groove where the ganglioside binds. For H C /A3, these residues have moved much farther (Fig. 4A) than those observed for the H C / A1 GT1b and GD1a bound structures, respectively (Fig. 4B,C). This is consistent with an induced fit mechanism for ganglioside binding where the tryptophan and histidine residues of H C /A3 translate~7 A. In addition to changes in relative positions of residues after ganglioside binding (Fig. 4G), there were noticeable differences in hydrogen bonding, especially to Sia 5 . As mentioned previously, Tyr 1263 /A3 , Gly 1275 /A3 and Leu 1250 /A3 form hydrogen bonds with the monosaccharide, the latter of which does so through a water molecule-bridged interaction. Although Leu 1250 /A3 is not conserved when compared to BoNT/A1 (the corresponding residue is Gln 1254 /A1 ), both ganglioside-bound structures do display a conserved water molecule. However, the orientation of the Sia 5 prevents a water-mediated interaction to Gln 1254 /A1 ; this difference in ring orientation is likely due to a nonconserved residue, Phe 1113 /A3 / Tyr 1117 /A1 (Fig. 4D-F). The other two residues (Tyr 1263 /A3 and Gly 1275 /A3 ) are conserved in BoNT/A1, Tyr 1267 /A1 and Gly 1279 /A1 , and both similarly form hydrogen bonds with the same monosaccharide in the GD1a-bound structure, but via a water molecule. There is an additional water-mediated and direct hydrogen bond interaction with Sia 5 via Arg 1276 /A1 and Tyr 1117 /A1 , respectively, which is different to that observed with H C /A3. For the GT1b-bound structure, however, the Sia 5 interactions are the same except that the water-mediated hydrogen bond with Tyr 1267 /A1 is replaced with a direct hydrogen bond with Ser 1275 / A1 .

Ganglioside binding affinity may affect BoNT potency
Both H C /A1 and H C /A3 bind to three carbohydrate moieties common to GD1a and GT1b-GalNAc 3 , Gal 4 , and Sia 5 . However, considering the distinct electron density maps, greater number of interacting residues, and low average B-factors, Gal 4 appears to be the most tightly bound monosaccharide. Four conserved residues form hydrogen bonds to Gal 4 with an additional water mediate interaction for Leu 1250 /A3 and the equivalent Gln 1254 /A1 but only for the GT1b-bound structure ( Table 2). An equivalent water molecule is present in the H C /A1-GD1a structure but the position of the backbone amide hinders its interaction with the ganglioside.
Overall, H C /A1 forms 10 hydrogen bonds with GT1b and 7 with GD1a, while H C /A3 forms seven with GD1a. Similarly, H C /A1 has a more extensive network of water-mediated interactions with ganglioside (four water molecules and five residues for GT1b, and two water molecules and four residues for GD1a) than H C /A3 (one water molecule and one residue for GD1a). This difference in combination of water-mediated interactions and hydrogen bond interactions would be consistent with the relative ganglioside binding affinities of BoNT/A1 [23][24][25] and suggests that BoNT/A3 binds GD1a with a lower affinity than BoNT/A1. Furthermore, the difference in degree of interaction between ganglioside and H C may partly explain the reported difference in potency between BoNT/A1 and BoNT/A3 and would be consistent with the observation that BoNT/A2 has a higher affinity for gangliosides than BoNT/A1 and also enters neuronal cells more efficiently [21,26].

Conclusion
The high-resolution crystal structure of H C /A3 in complex with the carbohydrate moiety of GD1a presented here reveals the interactions that are involved with ganglioside binding and the consequent change in conformation. A total of 6 residues form seven hydrogen bonds and one water-mediated interaction with Sia 5 , Gal 4 and GalNAc 3 . Although similar to H C /A1 binding to ganglioside, there are fewer interactions overall, mostly due to steric effects of Trp 1262 /A3 and Arg 1272 /A3 . This would indicate a lower ganglioside binding affinity for BoNT/A3 and may be a contributing factor to its reported lower toxicity compared to BoNT/A1.