Biochemical Journal

Research article

Botulinum neurotoxin serotype D attacks neurons via two carbohydrate-binding sites in a ganglioside-dependent manner

Jasmin Strotmeier, Kwangkook Lee, Anne K. Völker, Stefan Mahrhold, Yinong Zong, Johannes Zeiser, Jie Zhou, Andreas Pich, Hans Bigalke, Thomas Binz, Andreas Rummel, Rongsheng Jin


The extraordinarily high toxicity of botulinum neurotoxins primarily results from their specific binding and uptake into neurons. At motor neurons, the seven BoNT (botulinum neurotoxin) serotypes A–G inhibit acetylcholine release leading to flaccid paralysis. Uptake of BoNT/A, B, E, F and G requires a dual interaction with gangliosides and the synaptic vesicle proteins synaptotagmin or SV2 (synaptic vesicle glycoprotein 2), whereas little is known about the cell entry mechanisms of the serotypes C and D, which display the lowest amino acid sequence identity compared with the other five serotypes. In the present study we demonstrate that the neurotoxicity of BoNT/D depends on the presence of gangliosides by employing phrenic nerve hemidiaphragm preparations derived from mice expressing the gangliosides GM3, GM2, GM1 and GD1a, or only GM3 [a description of our use of ganglioside nomenclature is given in Svennerholm (1994) Prog. Brain Res. 101, XI–XIV]. High-resolution crystal structures of the 50 kDa cell-binding domain of BoNT/D alone and in complex with sialic acid, as well as biological analyses of single-site BoNT/D mutants identified two carbohydrate-binding sites. One site is located at a position previously identified in BoNT/A, B, E, F and G, but is lacking the conserved SXWY motif. The other site, co-ordinating one molecule of sialic acid, resembles the second ganglioside-binding pocket (the sialic-acid-binding site) of TeNT (tetanus neurotoxin).

  • botulinum neurotoxin D (BoNT/D)
  • crystal structure
  • ganglioside-binding site
  • HC fragment
  • sialic acid complex


BoNT/D (botulinum neurotoxin serotype D), produced by Clostridium botulinum, belongs to the family of CNTs (clostridial neurotoxins). CNTs comprise the seven BoNT serotypes A–G and TeNT (tetanus neurotoxin) that cause the diseases botulism and tetanus respectively. Their neurospecific binding to unmyelinated areas of nerve terminals [2] is the main cause of toxicity and they have a median lethal dose (LD50) below 1 ng per kg of body weight, electing them to the most toxic agents known [3]. At motor neurons, BoNTs inhibit acetylcholine release, leading to flaccid paralysis, whereas TeNT is subjected to retrograde transport into inhibitory neurons and blocks release of glycine and γ-aminobutyric acid, which results in spastic paralysis.

The crystal structures of the 150 kDa BoNT/A, B and E holotoxins [46] revealed that all CNTs are composed of four functionally independent domains, which perform individual tasks in the multi-step intoxication process [7]. The 100 kDa HC (heavy chain) contains three of the four domains. The C-terminal 25 kDa HCC (C-terminal HC) domain mediates the neurospecific binding and uptake of BoNT/A, B, E, F and G, as the first step to attack neurons (for a review, see [8]) [9]. The role of the N-terminally neighbouring 25 kDa HCN (N-terminal HC) domain in the intoxication mechanism is still unclear, although it is suggested that BoNT/A HCN weakly interacts with phosphatidylinositol phosphates [10]. Together the HCC and HCN domains form the 50 kDa HC fragment that displays intact neurospecific binding; the HC fragment can be isolated from full-length BoNT by proteolytic digestion or can be autonomously expressed in Escherichia coli. So far, the crystal structures of isolated HC fragments alone or in complex with carbohydrates or receptor peptides have been determined for BoNT/A, B, F, G and TeNT [1117], whereas the structures of BoNT/C and D are still unknown. The N-terminal half of the HC, the 50 kDa HN translocation domain, forms a channel and delivers the partially unfolded disulfide-bond-linked 50 kDa LC (light chain) into the cytosol [18,19]. Within the cytosol, the released LC, a Zn2+-dependent endopeptidase, hydrolyses members of the three neuronal SNAREs (soluble N-ethylmaleimide-sensitive fusion protein-attachment protein receptors), thereby blocking Ca2+-triggered exocytosis [7].

Widely distributed gangliosides represent low-affinity neuronal receptors responsible for accumulating CNTs on neuronal surfaces. This interaction has been studied extensively using in vitro assays for all serotypes (for reviews, see [20,21]). Their physiological role has been demonstrated for all CNTs using mice or mouse hippocampal neurons displaying different expression patterns of polysialogangliosides, as a result of specific gene KOs (knockouts) [8,9]. Surprisingly, only BoNT/D does not seem to require gangliosides for neuronal uptake, but uses phosphatidylethanolamine instead [22]; however, it was previously demonstrated that the biological activity of BoNT/D is inhibited in mice and hemidiaphragm preparations by pre-incubation with GT1b in a dose-dependent manner [23].

Several studies have demonstrated an accelerated uptake of BoNT/A–G and TeNT into the phrenic nerve prepared together with the adjacent hemidiaphragm and into isolated neuronal cultures, upon electrical and chemical stimulation respectively, as neuronal stimulation causes increased rates of exo- and endo-cytosis of SVs (synaptic vesicles) (for a review, see [8]) [9]. The N-terminal luminal domains of the SV membrane proteins Syt (synaptotagmin)-I and Syt-II were identified as the protein receptors for BoNT/B [24,25]. Co-crystallization of Syt-II with BoNT/B refined the binding segment to an α-helical 17-mer peptide adjacent to the transmembrane domain of Syt-II [12,26]. BoNT/G interacts with identical segments of Syt-I and Syt-II, but none of the other CNTs bind to either Syt-I or Syt-II [27]. Thereafter, the three isoforms of SV2 (SV glycoprotein 2), a 12-transmembrane-domain protein, were identified as protein receptors for BoNT/A [28,29], BoNT/E [30] and BoNT/F [9,13]. It is still under debate whether BoNT/C and D engage protein receptors for their neuronal uptake [22].

In more recent years, a conserved ganglioside-binding site in the HCC domain, containing an ‘E(D)…H…SXWY…G’ motif that interacts with the terminal NAcGalβ3–1Galβ (where NAcGal is N-acetylgalactosamine and Gal is galactose) moiety of GT1b and GD1a, has been described for BoNT/A [11,31], BoNT/B [5,31], BoNT/E [9], BoNT/F [9,13], BoNT/G [15,32] and TeNT [17,33]. A description of our use of ganglioside nomenclature is given in [1]. In addition, the binding site for the protein receptor of BoNT/B and G has been allocated to their HCC domains [12,26,32,34]. Interestingly, the TeNT HCC domain contains a second ganglioside-binding site (called the sialic-acid-binding site) at the homologous location with the protein-receptor-binding site in BoNT/B and G, which interacts either with the disialic acid branch of GD1b and GT1b [16,17,33,35,36], or, eventually, with a glycosylated protein receptor [37]. Only binding to both receptors allows CNTs to show neurospecific uptake, thereby confirming the double receptor mechanism [38]. However, analysis of an amino acid sequence alignment of all BoNT HCC domains does not reveal such a conserved ganglioside binding motif in BoNT/D [31] and its potential binding to gangliosides is currently ambiguous.

In the present study we employed MPN (mice phrenic nerve) hemidiaphragm preparations derived from mice only expressing either GM3 or the a-series gangliosides (GM3, GM2, GM1 and GD1a), and demonstrate that the biological activity of BoNT/D depends on the expression of complex polysialogangliosides. Binding of GT1b to HCD (BoNT/D HC fragment) was revealed by MALDI (matrix-assisted laser-desorption ionization)–TOF (time-of-flight)-MS. To further understand the mechanism underlying BoNT/D–ganglioside recognition, we have determined the crystal structures of HCD alone and in complex with sialic acid at 1.7 Å (1 Å = 0.1 nm) and 2.0 Å respectively. A carbohydrate-binding site in BoNT/D was identified that locates similarly to the protein-receptor-binding site in BoNT/B and G, as well as the sialic-acid-binding site in TeNT. Structure-based site-directed mutagenesis in BoNT/D was performed, focusing on the newly identified sialic-acid-binding site, as well as the equivalent ‘conserved’ ganglioside-binding pocket as observed in BoNT/A, B, E, F and G. Mutants in either site showed drastically decreased binding to neuronal membranes, as well as severely reduced biological activity in the MPN hemidiaphragm assay. Hence BoNT/D binding to cell-surface gangliosides is required in order to exert its neurotoxicity and it employs two receptor pockets within its HCC domain for the carbohydrate interaction.


Statistical analyses were performed using GraphPad Prism 4.03 software.

Plasmid construction

Plasmids encoding the HC fragments (pHCDS for recombinant expression in E. coli, pSP72-HCD for in vitro transcription/translation) and the full-length BoNT/D (pBoNTDS) of C. botulinum strain BVD/-3 (EMBL accession number X54254) have been described previously [39,40]. Mutations in HCD were generated using the GeneTailor™ method (Invitrogen) using suitable primers and pHCDS and pSP72-HCD as template DNA. Mutated expression plasmids for full-length BoNT/D were generated by swapping DNA fragments between pBoNTDS and mutated pHCDS and pSP72-HCD plasmids respectively. Nucleotide sequences of all mutants were verified by DNA sequencing.

Production of recombinant proteins

Recombinant full-length neurotoxins were expressed under biosafety level 2 containment (project number GAA A/Z 40654/3/123) in the E. coli strain M15[pREP4] (Qiagen), following 16 h of induction at 22 °C. Proteins were purified on Strep-Tactin® Superflow® columns (IBA) according to the manufacturer's instructions and were stored in 100 mM Tris/HCl, pH 8.0. All recombinant proteins were shock-frozen in liquid nitrogen, and kept at −70 °C.

Crystallization and diffraction data collection

The isolated wild-type HCD was further purified with an ÄKTA Purifier using a Superdex-200 16/60 column (GE Healthcare) in buffer containing 30 mM Tris/HCl, pH 7.6, and 500 mM NaCl and were then concentrated to ~7 mg/ml for crystallization. Initial crystallization screens were carried out using a Phoenix crystallization robot (Art Robbins Instruments) and high-throughput crystallization screen kits from Hampton Research, Qiagen or Emerald BioSystems. The best crystals were grown at 18 °C by the hanging-drop vapour-diffusion method in an 1:1 (v/v) ratio of protein and reservoir solution containing 0.1 M Hepes, pH 8.0, and 7% (v/v) PEG [poly(ethylene glycol)] 10K. The crystals were cryoprotected in the same mother liquor, supplemented with 25% (v/v) glycerol, and then flash-frozen in liquid nitrogen. The crystals of the sialic acid complex HCD–NAcNeu (N-acetylneuraminic acid or salic acid) were obtained by soaking the HCD crystals in a solution containing 10 mM Hepes, pH 8.0, 150 mM NAcNeu, 14% (v/v) PEG 10K and 25% (v/v) glycerol for 8 h at 18 °C. These crystals were then directly frozen in liquid nitrogen. The X-ray diffraction data of the apo-HCD and the HCD–NAcNeu complex were collected at 100K at beam line 9–2 of the SSRL (Stanford Synchrotron Radiation Laboratory), using a Mar-325 CCD (charge-coupled-device) detector. All data sets were processed and scaled by using HKL2000 [41]. Data collection statistics are summarized in Table 1. The crystals of the apo-HCD or the HCD–NAcNeu complex belong to the P 21 21 21 space group with unit cell dimensions of a = 60.7 Å, b = 89.9 Å, c = 94.1 Å, or a = 60.2 Å, b = 90.0 Å, c = 93.5 Å respectively.

View this table:
Table 1 Data collection and refinement statistics

Values in parentheses represent the highest resolution shell. Rmerge = ΣhΣj|Ihj−<Ih>|/ΣhΣj|Ih,j|; Rwork = (Σ∥Fo|−|Fc∥)/Σ|Fo|; 5% of the reflections were set aside for calculation of Rfree.

Structure determination and refinement

The structures of apo-HCD and the HCD–NAcNeu complex were determined by molecular replacement using Phaser [42] and the crystal structure of HCA (BoNT/A HC fragment; PDB code 2VU9) was used as the search model. The structures were subsequently refined with REFMAC5.5 and re-built with COOT in an iterative way [43,44]. Refinement progress was monitored with the Rfree value using a 5% randomly selected test set [45]. The apo and the complex structures were refined to 1.72 Å with Rwork/Rfree = 0.18/0.21 and to 2.0 Å with Rwork/Rfree = 0.17/0.22 respectively. Both structures, with excellent stereochemistry, were validated through the Molprobity web server [46]. Structural refinement statistics are listed in Table 1. Figures were prepared with PyMol (

Synaptosome-binding assay

Binding of 35S-labelled in vitro transcribed/translated HCD mutants to freshly prepared rat brain synaptosomes was conducted as described previously [31].

MPN hemidiaphragm assay

The MPN hemidiaphragm assay was performed in Krebs–Ringer solution (1.19 mM KH2PO4, 1.2 mM MgSO4, 25 mM NaHCO3, 118 mM NaCl, 4.75 mM KCl, 2.54 mM CaCl2 and 11 mM glucose, pH 7.4) under an 95% oxygen/5% carbon dioxide atmosphere, as described previously [47]. The phrenic nerve was continuously stimulated at 5–25 mA with a frequency of 1 Hz, or, in the indicated cases, at 0.016 Hz (with an 0.1 ms pulse duration). Isometric contractions were transformed using a force transducer and recorded with VitroDat Online software (FMI). The time required to decrease the amplitude to 50% of the starting value (paralytic half-time) was determined. To determine the altered neurotoxicity of the BoNT mutants, or wild-type BoNT/D with ganglioside-deficient MPN hemidiaphragm preparations, a concentration–response curve, consisting of five data points determined in triplicate was compiled to which a power function could be ascribed: y(5, 15, 30, 50 or 100 pM wild-type BoNT/D) = 203.41x−0.3438, R2 = 0.9825. Resulting paralytic half-times were converted into corresponding concentrations of the respective wild-type BoNT/D, using the equation given above, and neurotoxicity was finally expressed as the percentage of the wild-type BoNT/D neurotoxicity.

Complex-polysialoganglioside-deficient mice

Animal use was according to section 4.3 of German animal protection law; animals were killed by trained personnel before dissection of organs, with their number reported to the animal welfare officer of the Central Animal Laboratory and to the local authority Veterinäramt Hannover.

Phrenic nerves were derived from 20 g of wild-type NMRI mice or tissue from C57BL/6 mice lacking the genes B4galnt1 encoding NAcGalT (β-1,4-N-acetylgalactosamine transferase; EC and/or St8sia1 encoding GD3S [GD3 synthetase (CMP-sialic acid:GM3 α-2,8-sialyltransferase); EC]. Whereas neurons of NMRI mice contain the full set of complex polysialogangliosides, the nerve cells of complex-ganglioside-deficient mice contain only GM3 or predominantly GM1 and GD1a [48]. No difference in the paralytic half-time was observed when phrenic nerve hemidiaphragm preparations from NMRI and wild-type C57BL/6 mice were compared.

MALDI–TOF-MS analysis

Samples were analysed in an AB Sciex 5800 MALDI–TOF mass spectrometer controlled by TOF/TOF Series Explorer 4.0.0. For each sample eight subspectra were generated with 200 shots and a laser intensity of 6297 instrument specific units. Masses were detected from 20000 m/z to 100000 m/z. Samples were co-crystallized with 13 mg/ml sinnapinic acid solution in 50% acetonitril, 0.1% trifluoroacetic acid, mixed at a ratio of 1:1 (0.5 ul of each) with sample directly on a stainless steel MALDI target plate. BSA (66430 Da), trypsinogen (23980 Da) and Protein A (44620 Da) were used as standard proteins for external calibration.


Ganglioside deficiency clearly reduces the neurotoxicity of BoNT/D in the MPN assay

Almost all BoNT serotypes require complex polysialogangliosides as receptors to exert their neurotoxicity [8,9]. However, surprisingly BoNT/D does not lose its activity when applied to GM3S (GM3 synthetase)-deficient mice which should theoretically express only lactose-ceramide (Figure 1a) [22]. To clarify this discrepancy between BoNT/D and the other BoNT serotypes, we employed phrenic nerve hemidiaphragm preparations derived from St8sia1-KO mice or from a combined St8sia1/B4galnt1-double-KO mice to check the potency of BoNT/D in the MPN assay. The St8sia1-KO prevents synthesis of GD3S, resulting in expression of only GM3, GM2, GM1 and GD1a (Figure 1a). The double-KO mice lack GD3S and NAcGalT and thus express only GM3 [48]. We found that the neurotoxicity of BoNT/D is clearly reduced to 14% and 6% in GD3S-KO and GD3S/NAcGalT-KO MPN hemidiaphragms respectively (Figure 1b). In GD3S-KO hemidiaphragm preparations, the clear loss of potency of BoNT/D is in contrast with the moderate decrease in BoNT/A potency, which is due to the binding of BoNT/A to the terminal NAcGalβ3–1Galβ moiety present in the GM1 and GD1a gangliosides that are still expressed. Obviously, BoNT/D requires the 2–8NAcNeuα present in the ganglioside b-series, such as GD1b and GT1b, to exert its biological activity. This explanation is supported by the fact that the neurotoxicity of BoNT/D is only moderately further diminished in hemidiaphragm preparations where only GM3 is expressed, especially when compared with BoNT/A and C, which display one order of magnitude lower potency in GM3-only neurons (Figure 1b) [9]. Hence the absence of the disialic acid moiety, such as is found in GD1b and GT1b, clearly interferes with BoNT/D activity and indicates that sialic acid is the direct binding partner. This finding is supported by the results of a carbohydrate screen showing that HCD displays highest affinity to GD2, which also contains a disialic acid branch (; primscreen_1682). Furthermore, similar to earlier reports [31,33], we detected a direct binding of at least one GT1b molecule to HCD when employing MALDI–TOF-MS (Figure 2). Incubation of the wild-type HCD with GT1b in a molar ratio of 1:4 yielded an additional peak (asterisk in Figure 2b) with a mass of 51921 Da. This mass increase corresponds to approximately one molecule of GT1b. An earlier study reported the inactivation of BoNT/D by GT1b, which represents another piece of evidence for the interaction of BoNT/D with gangliosides [23]. We presume that the alternative expression of GM1b and GD1c in the GM3S-deficient mice [22] can interact with BoNT/D and facilitate its uptake. In contrast, the GM3-only double-KO mice used in the present study clearly express no complex polysialogangliosides [48] thereby demonstrating the ganglioside dependence of BoNT/D.

Figure 1 Ganglioside deficiency reduces neurotoxicity of BoNT/D

(a) Schematic representation of the biosynthetic pathway for complex polysialogangliosides. The scheme shows the enzymes GM3S, GD3S and NAcGalT, which are involved in ganglioside biosynthesis, and the encoding genes (St3gal5, St8sia1 and B4galnt1), which are deleted in the respective KO mice leading to an altered ganglioside expression pattern. (b) BoNT/A, BoNT/C [9] and BoNT/D show reduced neurotoxicity when applied to phrenic nerve hemidiaphragm preparations obtained from single GD3S-KO or combined GD3S/NAcGalT double-KO mice. These mice only express GM3, GM2, GM1 and GD1a, or GM3 respectively. Results are means±S.D. (n = 3–9).

Figure 2 Binding of GT1b to the HC fragment of BoNT/D (HCD) detected by MALDI–TOF-MS

(a) The molecular mass of the wild- type HCD (theoretical molecular mass of 50271 Da) was determined as 50246 Da by MALDI–TOF-MS. The signal at m/z 49675 belongs to an N-terminally truncated protein, due to a methionine at amino acid position six. (b) The wild-type HCD (25 μM) was incubated with GT1b (molecular mass of 2145 Da) in a molar ratio of 1:4 for 30 min at 37 °C and subsequently analysed by MALDI–TOF-MS. A significant additional peak (*) was visible at m/z 51921, indicating the binding of one molecule of GT1b.

High-resolution crystal structure of the HC fragment of BoNT/D

Since BoNT/D clearly shows polysialoganglioside-dependent neurotoxicity, it is important to understand the mechanism underlying the BoNT/D–ganglioside recognition. To this end, the crystal structure of HCD was determined. The construct of HCD (residues Ser863–Glu1276, GenBank® accession number CAA38175, with an N-terminally added MRGSAMA and a C-terminally fused PPTPGWSHPQFEK Strep-tag) was designed based on homology modelling with the crystal structures of HCA and HCB (BoNT/B HC fragment) and secondary-structure prediction [11,12]. Isolated HCD retains its full binding ability as it inhibits the biological activity of full-length BoNT/D in the MPN hemidiaphragm assay [9]. The structure of HCD was determined to 1.72 Å resolution using X-ray crystallography (Figure 3a and Table 1). Electron density was clearly visible for all residues except for two residues (Gly1180-Gly1181) that located in a flexible surface loop with minimal electron density and were not modelled.

Figure 3 Crystal structure of the HC fragment of BoNT/D (HCD)

(a) Ribbon diagram of HCD (PDB code 3OBT). Sialic acid and the one glycerol molecule (GOL) that binds at the conserved ganglioside-binding site are shown in a space-filling model. Regions that show large conformational differences in comparison with the structures of HCA and HCB are coloured blue. (b) An overall Cα atom alignment of the structures of HCD (salmon), HCA–GT1b complex (PDB code 2VU9; green) [11] and HCB–Syt-II complex (PDB code 2NM1; light blue) [12]. The carbohydrate portion of GT1b is shown as sticks (green) and Syt-II is shown as a ribbon (blue). The orientation of HCD is the same as that in (a).

The structure of HCD can be divided into two domains: the N-terminal domain HCN, composed of residues Ser863–Arg1082, and the C-terminal domain HCC, composed of residues Asn1083–Glu1276. HCND is formed from two anti-parallel β-sheets organized into a jelly-roll barrel motif (β-barrel domain) whose function is still unknown. HCCD adopts a β-trefoil fold similar to the other known BoNT HCC domains that are directly involved in dual receptor recognition [8].

Despite different receptor-binding specificities, the overall structure of HCD is similar to the HC fragment of the other BoNT serotypes, as reflected by high DALI Z-scores when comparing protein structures in three-dimensions [49]. The Z-scores are 41.6, 39.5, 40.6, 41.2, 40.2 and 40.3 for BoNT/A, B, E, F, G and TeNT respectively (it should be noted that pairs with Z-scores of 2.0 are already considered to be structurally similar). The pair-wise structure comparison between HCD and the HC fragments of BoNT/A (PDB code 2VUA), B (PDB code 2NM1), E (PDB code 3FFZ), F (PDB code 3FUQ), G (PDB code 2VXR), and TeNT (PDB code 3HMY) shows RMSD (root mean square deviation) values of 1.8 Å, 1.9 Å, 1.8 Å, 1.7 Å, 1.9 Å and 2.0 Å over the 366, 365, 356, 356, 364 and 369 best-aligned amino acids respectively [6,1113,15,50]. We selected HCA and HCB for further structural analyses because the mechanism by which they interact with their dual receptors has been well characterized, as revealed by biochemical data and the crystal structures of HCA–GT1b and HCB–Syt-II complexes [11,12,31,32].

The structural differences among HCD, HCA and HCB mostly reside within the solvent-exposed loops of the HCC domain. The loops with the largest differences in HCD comprise residues Glu1114–Val1117, Tyr1123–Leu1129, Ser1141–Tyr1146, Thr1172–Cys1187, Ser1195–Ile1202, Asn1209–Cys1217, Ser1223–Asn1227, Tyr1235–Tyr1246 and Val1251-Ser1262 respectively (coloured blue in Figure 3a). Interestingly, the Thr1172–Cys1187 loop of HCD is much longer and adopts an extended conformation when compared with its counterpart in HCA (loop Asn1196–Val1202) and HCB (loop Tyr1183–Glu1189). The conformation of this loop is expected to be flexible as evident by the relatively weak electron density in this area. The function of this loop is unclear. The Tyr1235– Tyr1246 loop of HCD corresponds to a loop in HCB that separates the binding sites for the ganglioside and the protein receptor. Its hydrophobic segment Phe1242–Tyr1246 is likely to be involved in membrane association. In HCD, this loop moves closer to the previously defined protein-receptor-binding site, as revealed by the HCB–Syt-II complex. Structural comparison between HCD and HCB suggests that the loop of HCD will physically clash with the C-terminus of Syt-II and prevent its binding in a mode observed in the HCB–Syt-II complex. Lastly, the Val1251–Ser1262 loop of HCD adopts a significantly different conformation in comparison with other BoNT serotypes. The equivalent region in HCA (loop Ser1264–Gly1279) has five residues, Ser1264, Trp1266, Tyr1267, Ser1275 and Arg1276, which are directly involved in GT1b binding [11,31].

Crystal structure of BoNT/D HC fragment in complex with sialic acid

Since our data clearly demonstrated that BoNT/D requires gangliosides as a receptor and sialic acid as a direct binding partner, the next goal was to explore the structural basis of the BoNT/D–ganglioside interactions. To this end, we determined the structure of the HCD–NAcNeu complex at 2.0 Å. One sialic acid molecule was clearly observed to bind at the tip of the HCC domain via extensive interactions with HCD (Figures 4a and 4b). The carboxy group of sialic acid forms four pairs of hydrogen bonds or salt bridges with residues Thr1172 (main chain nitrogen), Asp1173 (main chain nitrogen), Lys1192 (NZ atom) and Arg1239 (NH2 atom). In parallel, the O-2 of sialic acid contributes two more hydrogen bonds with the NH2 and NE atoms of Arg1239; the O-6 and O-7 atoms of sialic acid form hydrogen bonds with the NZ atom of Lys1192. Besides the direct interaction between sialic acid and HCD, several well-defined water molecules were identified in the sialic-acid-binding pocket which bridge the O-2, O-4, O-7 and O-10 atoms of sialic acid to HCD (Supplementary Table S1 at Surprisingly, the sialic-acid-binding site in HCD partially overlaps with the protein-receptor-binding site identified previously in BoNT/B and BoNT/G [32]. The novel sialic-acid-binding site in BoNT/D is reminiscent of the two carbohydrate-binding sites present in TeNT (a lactose-binding site and a sialic-acid-binding site [17,33]). The sialic-acid-binding site of HCD is indeed similar to the sialic-acid-binding site of TeNT (Figure 4c); Arg1239 of HCD adopts the functional role of Arg1226, the key residue involved in sialic acid binding in TeNT [33], whereas Lys1192 of HCD aligns structurally with Arg1226 in TeNT. The similarity of these sialic-acid-binding sites is reflected by biochemical data showing that HCD can inhibit the potency of TeNT and HCT that of active full-length BoNT/D in the MPN hemidiaphragm assay [9]. When comparing the apo-HCD and HCD-NAcNeu structures, one sees that the side chain of Lys1192 moves to form hydrogen bonds with sialic acid, whereas Arg1239 is unaltered. Also, the side chain of Asp1171 moves away to accommodate the carboxy group of sialic acid.

Figure 4 Structure of the HCD–sialic acid complex

(a) Close-up view of the sialic-acid-binding pocket in HCD (PDB code 3OBT). HCD is in salmon and sialic acid (Sia) is in green. Key residues involved in complex interactions are shown in a stick representation. Hydrogen bonds and salt bridges are indicated by black dashed lines. (b) A σA weighted FoFc omit electron density map (contoured at 1.5 σ) around the bound sialic acid, overlaid with the final model. (c) The sialic-acid-binding site in HCD partially overlaps with the known sialic-acid-binding site in TeNT. The superimposed structures of HCD (salmon) and the TeNT HC-fragment (HCT) in complex with GT2 (PDB code 3HMY; blue) [50] are shown as a ribbon model. Sialic acid (binding to HCD) and GT2 (binding to HCT) are shown as stick models in green and blue respectively.

Several glycerol molecules were observed in the structure of the HCD–NAcNeu complex, which was used as the cryoprotectant for X-ray diffraction data collection at 100 K. Interestingly, one glycerol was found in a pocket (Figures 3a and 5) that has been described as the ‘conserved’ ganglioside-binding pocket from the structures of HCA–GT1b (PDB code 2VU9) [11] and BoNT/B–sialyllactose (PDB code 1F31) [5]. This glycerol molecule forms two hydrogen bonds with the main chain oxygen of Thr1252 and Asn1253 in the loop of Val1251–Ser1262. A similar glycerol molecule was also observed in the structure of apo-HCD. As discussed above, the equivalent loop in HCA is directly involved in GT1b binding. Therefore one might assume that BoNT/D accommodates a second receptor-binding site in an area where BoNT/A, B, E, F and G possess their singular conserved ganglioside-binding site. However, the Val1251–Ser1262 loop of BoNT/D misses the conserved ‘SXWY’ motif in which the tryptophan residue is responsible for the parallel packing of the hydrophobic side of the terminal galactose. It implies that BoNT/D might have a different carbohydrate-binding property within this site, which will be further characterized below.

Figure 5 Site-directed mutagenesis analysis of binding of HCD mutants to synaptosomes

In vitro translated 35S-labelled HCD mutants were bound to freshly prepared rat brain synaptosomes for 2 h at 0 °C and quantified by SDS/PAGE and subsequent autoradiography. Results are means±S.D. (n = 3–12). HCD mutants are grouped for mutations in the ‘sialic acid site’ (the structure is shown in the top left-hand panel; residues in green sticks), the ‘F1242–Y1246 loop’ (structure in the top middle panel; residues in blue sticks), and the ‘conserved binding site’ [structure in the top right-hand panel; residues in green sticks, with a GT1b molecule (cyan) that is modelled based on the structure of the HCA–GT1b complex].

Mutagenesis and membrane-binding studies reveal two carbohydrate-binding sites in the HC fragment of BoNT/D

We have now identified two potential ganglioside-binding pockets in BoNT/D: one is the newly identified sialic-acid-binding site and the other is the counterpart of the conserved ganglioside-binding pocket in BoNT/A, B, E, F and G (these are referred to as the sialic acid site and the conserved binding site respectively in the following discussion). To study further the physiological relevance of these two sites, we performed a series of structure-based mutagenesis studies on HCD, focusing on key residues that reside in these two sites. The HCD mutants were then examined for binding with synaptosomes that were freshly isolated from rat brain and display predominantly complex polysialogangliosides in their physiological environment on their outer membrane. In vitro transcribed/translated 35S-labelled HCD mutants were incubated with synaptosomes at 0 °C and the amount of bound radioactive HCD was quantified.

Mutation of Arg1239 in the sialic acid site into alanine and tyrosine caused a drastic decrease in binding by 80% and 65% respectively (Figure 5) because the important salt bridge between O-1 of sialic acid and the NH2 atom of Arg1239, as well as the hydrogen bond between O-2 of sialic acid and the NE atom of Arg1239, was destroyed (Figure 4a). The hydrogen bonding of sialic acid with Lys1192 seems to play a supporting role since the binding of HCD-K1192A was only reduced by 35%. Interestingly, mutation of Trp1238 and Phe1240 caused drastic reductions in binding although they are not directly involved in co-ordination of the sialic acid. Whereas the lack of the phenyl ring of Phe1240 reduces binding by 70%, a replacement with a tryptophan residue exceeds the binding of wild-type HCD. The indole ring of Trp1238 is crucial for the structural integrity of the sialic acid site because neither the aliphatic leucine residue nor the aromatic phenylalanine and tyrosine residues could rescue its loss. It is conceivable that both residues might interact with the adjacent galactose in the carbohydrate chain of gangliosides. As expected, removal of the carboxy group of Asp1173, whose main chain nitrogen forms a hydrogen bond with the carboxy group of sialic acid, did not alter the binding affinity. Combining the mutations K1192A and R1239A to remove important hydrogen bonding interactions with sialic acid markedly reduces the binding of HCD (by 92%; Figure 5). Similar behaviour has been demonstrated for TeNT by mutation of Arg1226 in the sialic acid site [33]. The residual binding of HCD with mutations in the sialic acid site is presumably mediated by a second binding site.

One molecule of glycerol binds to the analogue of the singular conserved ganglioside-binding site in BoNT/A, B, E, F and G, which is also the lactose-binding pocket in TeNT. Whereas the conserved pocket consists of an ‘E(D)…H…SXWY…G’ motif and is split up into a hydrophilic and a hydrophobic side in other BoNTs, this motif is neither found in the primary sequence nor in the three-dimensional structure of HCD. The functional role of the conserved binding site in BoNT/D was assessed by mutations N1186A, D1233A, Y1235A/W, V1251F, N1253A, K1257A and S1262F, which were selected by structural homology modelling based on the structure of the HCA–GT1b complex (Figure 5) [11]. Whereas the binding to synaptosomes could be blocked efficiently by the mutation V1251F, only a minor effect was observed in the case of S1262F. Mutation of the sterically neighbouring Asp1233 and Tyr1235 to an alanine residue showed a clear loss of binding by 80% and 45% respectively. Again, introduction of a tryptophan residue at position 1235 preserves wild-type levels of binding. Removing the side chains of Asn1253 and Lys1257 caused a moderate impact on binding, whereas the mutant N1186A did not alter the binding affinity of HCD. The clear effects upon mutation of Asp1233, Tyr1235 and Val1251 demonstrate the presence of a second ganglioside-binding site in BoNT/D that is located in the position of the conserved binding site, but does not display a conserved architecture in comparison with other BoNTs. Combining the most severe substitution in each pocket, D1233A in the conserved site and R1239A in the sialic acid site, one obtains a double-mutant that displays hardly any detectable binding to neuronal membranes, further supporting the presence of two receptor pockets (Figure 5).

Interestingly, the Tyr1235–Tyr1246 loop of HCD, which directly connects the sialic acid site and the bottom of the conserved site, contains three partially hydrophobic residues (Phe1242, Asn1244 and Tyr1246) that are surface-exposed but do not contribute to the crystal packing. Mutation of these three residues into alanine or serine residues reduced the neuronal binding of HCD by ~70%, whereas the mutant F1242W behaved like the wild-type protein. Similar motifs are present in BoNT/A (Phe1252, Gln1254 and Phe1255) and BoNT/B (Phe1250, Glu1251 and Tyr1253). We hypothesize that this amino acid stretch could substantially contribute to interactions of HC with the neuronal membrane surface.

In a previous study the mutants HCD E1114K and G1132K displayed an improved binding affinity to synaptosomes [51]. Neither residue participates in the formation of the sialic acid or the conserved site. Gly1132 is found far away, on the side opposite to the conserved site, whereas Glu1114 stabilizes the Tyr1235–Tyr1246 loop, which is important for sialic acid binding, through formation of a salt bridge to Lys1243. The reason for the improved binding affinity of these two mutants needs to be further studied.

Reduced ganglioside binding decreases the neurotoxicity of BoNT/D

To further examine the effects of mutants in BoNT/D on its physiological function, selected mutations were transferred to full-length BoNT/D and their impact on the neurotoxicity of recombinant BoNT/D was determined employing the MPN hemidiaphragm assay (Figure 6). The low synaptosomal binding affinity of mutants R1239A or R1239Y similarly caused a reduction of neurotoxicity by more than 95%. Removing all hydrogen-bonding side chains in the sialic acid site, as in the double mutant K1192A/R1239A, decreased the potency of BoNT/D to well below 2%. In addition, the lack of the indole ring at position 1238 resulted in proteins displaying less than 2% neurotoxicity; only introduction of another aromatic side chain could attenuate this loss. A similar result was observed for Phe1240. Mutations in the conserved site correspondingly reduced the neurotoxicity of full-length BoNT/D, as the mutants D1233A and V1251F were 70% and Y1235A was 60% less potent than the wild-type protein. However, the effects of mutations in the sialic acid site turned out to be much more drastic than those in the conserved site, indicating a higher receptor affinity towards the sialic acid site. It seems that the lack of the conserved amino acid architecture within the conserved site causes a lower binding affinity to carbohydrates in comparison with other BoNTs. Combining the key mutations of both sites, the BoNT/D-D1233A/R1239A double-mutant exerts less than 0.4% neurotoxicity, approx. one order of magnitude lower potency than the single-site BoNT/D-R1239A mutant, thereby supporting a double receptor interaction for BoNT/D. Furthermore, the removal of aromatic residues Phe1240, Phe1242 and Tyr1246 also impaired the neurotoxicity, indicating that the putative membrane interaction contributes to the uptake of BoNT/D. In summary, the selected BoNT/D mutants, with mutations in either binding site, show altered synaptosomal binding and similarly display changed neurotoxicity.

Figure 6 Reduced ganglioside binding of BoNT/D leads to a drastic decrease in neurotoxicity

Selected recombinant full-length BoNT/D mutants were analysed in the MPN hemidiaphragm assay and the remaining paralytic half-times were converted by a corresponding power function to a percentage of the neurotoxicity of the wild-type. Results are means±S.D. (n = 3–6). The biological activity of BoNT/D-R1239A mutant, which displays a deactivated sialic-acid-binding site, was also analysed via MPN hemidiaphragm preparations derived from GD3S-KO mice (inset; plotted on a logarithmic scale).

To further prove that BoNT/D contains a second ganglioside-binding site, other than the sialic-acid-binding site, we applied the BoNT/D-R1239A mutant to GD3S-KO MPN preparations. Since the BoNT/D-R1239A mutant bears a deactivated sialic-acid-binding site, one would not expect a significant change in residual potency when applied to the wild-type or the KO mice preparation in the absence of a second site. However, we found that the potency of the mutant BoNT/D R1239A is ~90-fold lower when applied to GD3S-KO MPN preparations in comparison with the wild-type MPN preparations (Figure 6, inset). This drastic decrease strongly suggests the presence of a second binding site in BoNT/D that can partially compensate the deactivation of the sialic-acid-binding site as long as the receptor portfolio is intact.


We have clearly demonstrated that the biological activity of BoNT/D depends on the expression of complex polysialogangliosides. This proceeds from our findings in MPN hemidiaphragm preparations derived from mice only expressing either GM3 or the a-series gangliosides (GM3, GM2, GM1 and GD1a). This finding is endorsed by detection of a HCD–GT1b complex in MALDI–TOF-MS studies. Our results are also supported by earlier studies showing that BoNT/D activity is inhibited by GT1b [23] and that its HC fragment binds to the disialoganglioside GD2 (; primscreen_1682). Thus, in the first step of attacking motor neurons, all BoNTs accumulate on the cell surface via interaction with complex polysialogangliosides followed by receptor-mediated endocytosis. Despite the lowest amino acid sequence identity between BoNT/D and the human pathogenic BoNT/A, B and E serotypes (~25%), the high-resolution crystal structure of the BoNT/D HC fragment displays a similar overall architecture as the HC fragments of BoNT/A, B, E, F, G and TeNT. Furthermore, the co-crystal structure of HCD in complex with sialic acid, a key building block of gangliosides, exhibits a novel carbohydrate-binding site that locates similarly to the protein-receptor-binding site in BoNT/B and G, as well as the sialic-acid-binding site in TeNT. Site-directed mutagenesis within this sialic acid site yielded mutants that showed drastically decreased binding to neuronal membranes, as well as severely reduced biological activity in the MPN hemidiaphragm assay. In addition, mutation of residues within a pocket in the HCC domain of BoNT/D, which is the analogue of the conserved ganglioside-binding pocket identified previously in BoNT/A, B, E, F and G, also clearly reduced the binding affinity and potency of BoNT/D. However, this pocket of BoNT/D lacks the conserved ‘E(D)…H…SXWY…G’ motif found in other BoNTs; this might cause a reduced carbohydrate-binding affinity of BoNT/D in comparison to, e.g., TeNT, which also displays two ganglioside-binding sites. In conclusion, BoNT/D binds to the cell-surface gangliosides and employs two receptor pockets within its HCC domain to exert its neurotoxicity.


Jasmin Strotmeier and Anne Katrin Völker designed and performed the mutagenesis studies, recombinant protein expression, the synaptosomal binding assay and the MPN hemidiaphragm experiments, and acquired and analysed the data as part of their Ph.D. theses. Yinong Zong, Kwangkook Lee and Jie Zhou performed the protein characterization, crystallization, structure determination and analysis. Johannes Zeiser performed the MS experiments and Andreas Pich validated and interpreted the MS data. Stefan Mahrhold, Hans Bigalke and Thomas Binz contributed important intellectual content during the project and assisted in revision of the manuscript. Rongsheng Jin and Andreas Rummel supervized the project and wrote the manuscript, with all authors involved.


This work was supported by the Bundesministerium für Bildung und Forschung [grant number 01KI0742 (to H.B.)]; the Deutsche Forschungsgemeinschaft [grant numbers IIB2-Bi 660/2–3 (to T.B.), GSC108 (Exzellenzinitiative grant to A.K.V.)]; the Alfred P. Sloan Research Fellowship (to R.J.); and by the start-up research fund from the Sanford-Burnham Medical Research Institute (to R.J.). The SSRL Structural Molecular Biology Program is supported by the U.S. Department of Energy, Office of Biological and Environmental Research, and by the National Institutes of Health, National Center for Research Resources, Biomedical Technology Program and the National Institute of General Medical Sciences.


We thank Nadja Krez, Ulrike Fuhrmann, Beate Winter, Bianca Janzen and Dr Shenyan Gu for excellent technical assistance, and Dr Tino Karnath for support with FPLC. We also thank Dr Stuart Lipton and Dr Robert Liddington for support. Portions of this research were carried out at the Stanford Synchrotron Radiation Lightsource, a national user facility operated by Stanford University on behalf of the U.S. Department of Energy, Office of Basic Energy Sciences.


  • The co-ordinates and diffraction data for the apo-BoNT/D HC fragment and BoNT/D HC fragment–NAcNeu have been deposited in the PDB under codes 3OBR and 3OBT respectively.

Abbreviations: BoNT, botulinum neurotoxin; CNT, clostridial neurotoxin; GD3S, GD3 synthetase; GM3S, GM3 synthetase; HC, heavy chain; HCA, BoNT/A HC fragment; HCB, BoNT/B HC fragment; HCD, BoNT/D HC fragment; KO, knockout; LC, light chain; MALDI, matrix-assisted laser-desorption ionization; MPN, mice phrenic nerve; NAcGal, N-acetylgalactosamine; NAcGalT, β-1,4-N-acetylgalactosamine transferase; NAcNeu, N-acetylneuraminic acid (salic acid); PEG, poly(ethylene glycol); RMSD, root mean square deviation; SV, synaptic vesicle; Syt, synaptotagmin; TeNT, tetanus neurotoxin; TOF, time-of-flight


View Abstract