Revisiting the role of histo-blood group antigens in rotavirus host-cell invasion

Histo-blood group antigens (HBGAs) have been proposed as rotavirus receptors. H type-1 and Lewisb antigens have been reported to bind VP8* from major human rotavirus genotypes P[4], P[6] and P[8], while VP8* from a rarer P[14] rotavirus recognizes A-type HBGAs. However, the role and significance of HBGA receptors in rotavirus pathogenesis remains uncertain. Here we report that P[14] rotavirus HAL1166 and the related P[9] human rotavirus K8 bind to A-type HBGAs, although neither virus engages the HBGA-specific α1,2-linked fucose moiety. Notably, human rotaviruses DS-1 (P[4]) and RV-3 (P[6]) also use A-type HBGAs for infection, with fucose involvement. However, human P[8] rotavirus Wa does not recognize A-type HBGAs. Furthermore, the common human rotaviruses that we have investigated do not use Lewisb and H type-1 antigens. Our results indicate that A-type HBGAs are receptors for human rotaviruses, although rotavirus strains vary in their ability to recognize these antigens. Histo-blood group antigens (HBGAs) have been proposed as essential cellular receptors for rotaviruses, which can cause severe diarrhoea in children. Here, the authors show that rotavirus strains display important variations in their ability to recognize different HBGAs.

R otaviruses, members of the Reoviridae family, are the most important cause of severe diarrhoeal disease and dehydration in young children with around 500,000 deaths each year 1 . The introduction of two vaccines licensed in 2006 and 2008 has been successful in reducing disease severity and death rates 2 . However, several recent studies have revealed a significant increase in intussusception after the administration of the first or second rotavirus vaccine dose 3,4 , and vaccine-associated hospitalizations for severe rotavirus gastroenteritis have been reported 5 . These risks must be weighed up against the benefits of preventing rotavirus-associated illness. Rotaviruses with the P [8] genotype of the spike protein VP4 are predominantly associated with severe disease globally, with the P [4] and P [6] genotypes being substantially less common 6 . Virus infectivity is activated by trypsin cleavage of VP4 to produce the VP5* and VP8* subunits. Although both subunits remain associated on the virion surface, initial cell attachment to host cell membrane glycoconjugates is mediated by the VP8* head of the spike protein 7,8 . Gangliosides are glycosphingolipids with one or more sialic acid residues that are key receptor molecules for rotavirus infection [9][10][11][12][13][14] . Rotaviruses have been classified as sialidase-sensitive or sialidase-insensitive, based on the effect on their infectivity of host cell pre-treatment with sialidase. However, commonly used bacterial and viral sialidases remove only terminal sialic acids from sialylglycoconjugates. Sialic acid recognition by sialidasesensitive animal rotaviruses via VP8* is structurally well characterized through a series of crystallographic studies 13,15-21 . Our recent studies using nuclear magnetic resonance (NMR) spectroscopy and cell-based assays identified GM1 as a ganglioside receptor for the human rotavirus strains Wa (P [8]) and RV-3 (P [6]), and revealed that the internal Nacetylneuraminic acid (sialic acid) residue of GM1 plays a key role in VP8* recognition 12,22 . We concluded that human sialidase-insensitive strains such as Wa and RV-3, and most probably other human strains, require the presence of sialic acidcontaining receptors for efficient infection 12,22 .
A new paradigm of human P [14] rotavirus binding to A-type HBGA has been recently suggested, related to decreased infectivity of HAL1166 in HT-29 cells following anti-A-type antibody treatment and increased infectivity in CHO cells expressing A-type HBGA (ref. 23). The X-ray crystal structure of HAL1166 VP8* revealed a typical galectin-like fold with the twisted b-sheets separated by a shallow cleft as observed in all available VP8* structures, including those of the sialidaseinsensitive human rotaviruses Wa and DS-1 (P [4]), and the sialidase-sensitive animal strains RRV and CRW-8 VP8* (refs 16,23). Notably, the width of the binding site in HAL1166 VP8* appears narrower than in Wa and DS-1, resembling the binding site of RRV and CRW-8. Interestingly, sialic acids bind the VP8* of such animal rotaviruses at the same location within the narrow cleft as HBGA A-type trisaccharide (A-tri) and tetrasaccharide (A-tetra) binds to HAL1166 VP8*, although the key amino acids that interact with sialic acids and HBGA are different. Based on VP8* sequence alignments and rotavirus infectivity assays, it was predicted that the human P [9] rotavirus K8 could also bind to A-type HBGA, although no structural evidence has been reported. The VP8* of P [9], P [14] and P [25] genotype rotaviruses within P genogroup III bind A-type HBGAs in haemagglutination inhibition and synthetic oligosaccharidebased assays 24 . In addition, binding assays with P genogroup II rotaviruses of P genotypes [4], [6] and [8] showed interaction of H type-1 and Lewis b antigens with the VP8* of P [4] and P [8] human rotaviruses, while P [6] rotavirus VP8* bound only to H type-1 antigen 25,26 . In the current study, we sought to determine the importance of HBGA receptors for several major human rotavirus genotypes.
Our results demonstrate that A-type HBGAs are important receptors for human rotaviruses, although major variations between virus strains were detected. Using a multidisciplinary approach, we show that P [14] rotavirus HAL1166 and the related P [9] human rotavirus K8 bind A-type HBGAs without engaging the HBGA-specific a1,2-linked fucose moiety. Importantly, human rotaviruses DS-1 (P [4]) and RV-3 (P [6]) also use A-type HBGAs for infection, with fucose involvement. In contrast, the human P [8] rotavirus Wa does not recognize A-type HBGAs and no rotavirus strain we tested uses H-type-1 and Lewis b antigens.

Results
Binding of HBGAs to rotavirus VP8*. The evidence supporting the notion that HBGAs are common cell receptors for human rotaviruses remains incomplete and unclear. To address this issue, we report here an NMR solution study of the interactions between A-tri and A-tetra (Fig. 1a) 27 .
We have used Saturation Transfer Difference (STD) NMR spectroscopy to study the interaction of rotavirus VP8* protein and A-type HBGAs. This NMR method is based on the transfer of saturation from the protein to bound ligands that are in fast exchange with a binding site, and therefore allows the discrimination of binding and non-binding ligands. The degree of saturation a ligand proton receives from the target protein can be directly translated into a ligand-binding epitope. Ligand protons that are in close contact to the protein surface receive larger saturation and will therefore produce larger STD NMR effects compared to protons that are more solvent exposed 28,29 .
We obtained the 1 H NMR spectrum of A-tri ( Fig. 2a) and the control STD NMR spectra of A-tri with glutathione-S-transferase (GST) (Fig. 2b) and A-tri in complex with recombinantly expressed GST-tagged VP8* from human rotavirus strains Wa, K8 and HAL1166. The STD NMR spectrum of GST-Wa VP8* in complex with A-tri showed no recognition of A-tri (Fig. 2c). However, the STD NMR spectra of GST-K8 VP8* (Fig. 2d) and GST-HAL1166 VP8* (Fig. 2e) in complex with A-tri unequivocally show that both proteins bind to A-tri. The control STD NMR spectrum of the GST fusion tag alone with A-tri shows no significant signal intensities, confirming that A-tri does not interact with GST. It is worth noting that this important control STD NMR spectrum demonstrates that all observed signals in the spectra of HAL1166 and K8 VP8* originate from binding of A-tri to VP8*. The strongest interaction of A-tri was observed for the methyl protons of the GalNAc N-acetamido group (NHAc, d ¼ 1.73 p.p.m.) for both Hal1166 and K8 VP8* proteins. No STD NMR signal was observed for the H-disaccharide (Fuca1,2Galb), emphasizing the significance of the GalNAc moiety for A-tri binding to HAL1166 (Supplementary Fig. 1). It is most striking that STD NMR signals were also observed for the GalNAc H2 proton of two A-tri species: A-tri pa and A-tri pb . In addition, the H4 proton of the galactose moiety also received saturation for A-tri pa and A-tri pb . However, the X-ray crystal structure of A-tri with HAL1166 VP8* revealed that only the pyranose a-anomer, a biologically irrelevant configuration, binds to the protein 23 . In HBGA precursors, the glycosidic linkage between Gal and GlcNAc is always a b-configuration. Our solution binding study demonstrates that A-tri pa and A-tri pb are able to bind to K8 and HAL1166 VP8*, and that the binding epitope of A-tri for both proteins is identical, suggesting the same binding mode of A-tri, irrespective of the reducing end ( Supplementary Fig. 2a,b). A homology model of K8 VP8* was generated using the HAL1166 VP8* X-ray crystal structure (pdb accession code: 4DRV 23 ) ( Supplementary Fig. 3) with a sequence identity of 83.9% and sequence similarity of 90.7%. The binding-site residues show minor variations supporting the identical binding epitope of A-tri determined by STD NMR experiments.
We have also analysed interactions of A-tri with VP8* proteins from P[6] (RV-3) and P [4] (DS-1) human rotaviruses. A-tri binding affinity to RV-3 and DS-1 was identified (Fig. 3b,c, respectively), with a strong STD NMR signal observable for the N-acetamido group of the GalNAc residue. Strikingly, the methyl group of the fucose moiety (CH 3 Fuc, B1 p.p.m.) shows the most predominant STD NMR signal intensity, leading to the conclusion that the fucose moiety of A-tri does interact with the VP8* protein of DS-1 and RV-3. Our NMR study provides the first detection of a direct interaction of the HBGA fucose moiety with any rotavirus VP8* protein. However, absolute STD NMR signal intensities were substantially lower compared with the spectra acquired for A-tri in complex with HAL1166 and K8. To validate protein activity, an equimolar concentration of GM1a glycan, a confirmed receptor for the DS-1-like P [4] rotavirus strain RV-5 and for RV-3 (ref. 22) was added to the A-tri complex of VP8* protein from DS-1 or RV-3.
Quantitative analysis of the HAL1166 VP8*-A-tri STD NMR spectra revealed that the fucose residue of A-tri does not receive any saturation via the protein. This solution-based observation is in excellent agreement with the reported X-ray crystal structure 23 31,32 . Furthermore, a recent study has led to the proposal that HBGA-like substances associated with human enteric bacteria bind human norovirus and may have an impact on transmission and infection of this virus 33 . Interestingly, crystallography studies revealed that norovirus predominantly interacts with the fucose moiety of the blood group antigens [34][35][36][37] . In fact, the a1,3and a1,4-fucose residues of Lewis HBGAs are claimed to be essential for binding to GII.4 noroviruses 36,37 . STD NMR studies on GII.4 virus-like particles derived from norovirus  Our study also demonstrates a lack of A-tri recognition by VP8* from the clinically relevant human P[8, II] rotavirus Wa. We identified that GM1a glycan bound as expected to the A-tri complex of Wa VP8* and the Wa VP8* competed with Wa virus for infectivity, confirming protein function. Moreover, using STD NMR spectroscopy no recognition of H type-1 and Lewis b antigens by the P[8] Wa VP8* was observed ( Supplementary  Fig. 4). In contrast, a recent study based on assays for VP8* binding to human saliva and synthetic oligosaccharides led to the conclusion that H type-1 and Lewis b antigens can interact with Wa VP8* (ref. 25). Notably, the Wa strain represents an important current and past lineage of the most common human rotavirus serotype worldwide 40 but shows only 43% VP8* sequence identity with HAL1166, and most of the amino acids in the glycan-binding site differ substantially ( Supplementary Fig. 5). Importantly, our present study demonstrates for the first time that VP8* of DS-1 and RV-3 rotaviruses recognize A-tri. After P[8,II], these viruses represent the next most common genotypes P [  and show relatively low VP8* sequence identity to HAL1166 (44 and 47%, respectively). Significantly, the A-tri engagement with VP8* from both DS-1 and RV-3 involves the fucose moiety. This fucose recognition is consistent with a recent study in children from Burkina Faso, which showed an association of P [6] rotavirus disease predominantly with a Lewis-negative phenotype, independent of secretor status 41 .
A-tetra and A-tri have an identical binding epitope for K8 and HAL1166 VP8*. We have also used STD NMR spectroscopy to investigate the VP8* interaction of the biologically relevant A type-2 HBGAs tetrasaccharide (A-tetra), which has an additional GlcNAc residue ( Fig. 4a-g). The initial 1 H NMR spectrum of A-tetra shows no distinct forms as in the case of A-tri (Fig. 4a). A control STD NMR experiment of the GST fusion tag in the presence of A-tetra shows no significant STD NMR signals (Fig. 4b), confirming that all STD NMR signals for A-tetra in complex with GST-K8 VP8* (Fig. 4c,f) and GST-HAL1166 VP8* (Fig 4d,g) are genuine and are derived from the binding of A-tetra to VP8* and not GST. Similar to A-tri recognition, the N-acetamido group's methyl protons of the crucial A-tetra Gal-NAc moiety (NHAc, d ¼ 1.72 p.p.m.) received the highest saturation, implying a very close contact with the protein surface. However, our STD NMR studies demonstrated that the additional GlcNAc residue N-acetamido group methyl protons do not engage with either HAL1166 VP8* (Fig. 4g) or K8 (Fig. 4f). Furthermore, the GalNAc, and Gal H2 and H3 protons also show strong STD NMR signals, whereas the ring protons of the GlcNAc residue do not produce significant STD NMR signals. The fucose residue is entirely solvent exposed and does not interact with the protein. Therefore, we conclude that the binding epitope of A-tetra is identical to that of A-tri when complexed with HAL1166 VP8*. Finally, from this study we have discovered that the binding epitopes of A-tri and A-tetra are also identical when bound to K8 VP8*, with only very small differences observed in STD NMR signal intensities ( Supplementary  Fig. 2).
A-tri adopts a different conformation when bound to K8 VP8*.
The conformation of glycans when bound to proteins (bioactive ARTICLE conformation) can significantly vary in solution compared with a single conformation obtained with X-ray crystallography. To determine the bioactive conformation of A-tri when bound to VP8* in solution, we acquired Nuclear Overhauser Enhancement spectroscopy (NOE) NMR spectra of free A-tri in solution and transferred NOE (trNOE) NMR experiments of A-tri in complex with K8 VP8* following a similar approach to that previously reported 11 . The NOE NMR spectrum shows that A-tri has positive NOEs (diagonal and cross-signals of opposite sign) at a temperature of 305 K (Fig. 5a). On the contrary, the trNOE experiment of a VP8* K8:A-tri-complex (Fig. 5b) at a 1:10 (protein:ligand) molar ratio clearly indicate that the NOEs are negative, leading to the assumption that A-tri has adopted the rotational correlation time (t c ) of the VP8* due to association with the protein. Conversely, the trNOE experiment of A-tri in complex with Wa VP8* does not result in a change of the sign of NOE signals, indicating that A-tri has no binding affinity for Wa VP8* (Fig. 5c). The observed reduced intensity of the NOE signal ( Fig. 5c) can be explained by a substantially lower ligand concentration in the trNOE compared with the NOE experiment of the free ligand in solution. A more detailed comparison between the NOE and trNOE spectra indicates significant differences in the NOE pattern. Most striking is the disappearance of the interglycosidic NOE between the fucose methyl group, the CH 3 (Fuc) residue and the H1 (Gal) proton of bound A-tri. This particular NOE shows a strong signal in the uncomplexed NOE spectrum (Fig. 5a), indicating that the bioactive conformation of A-tri must be different to the predominant solution conformation adopted in the absence of the protein. The disappearance of a crucial NOE can be explained by a conformational rearrangement of the ligand on binding to the protein, most probably around the a1,2-glycosidic linkage between the fucose and galactose residue. This is in excellent agreement with the binding epitope obtained from STD NMR  experiments, revealing that the fucose is not engaged with the protein and remains solvent exposed, resulting in a higher degree of flexibility.
Molecular dynamics simulation reveals flexibility of the fucose residue. To further confirm the flexibility of the fucose residue, molecular dynamics (MD) simulation of A-tri pa in complex with HAL1166 ( Supplementary Fig. 6) was conducted based on the X-ray crystal structure (pdb 4DRV) 23 . Our MD simulation reveals that the a1,2-glycosidic linkage between the fucose and galactose residue is highly flexible. Furthermore, our simulation indicates that this linkage can easily adopt a conformation where the intermolecular distance between the H1 (Gal) and CH 3  Interestingly, the observed severe line broadening for the H1 (GalNAc) signals for both pyranose aand b-anomers (A-tri pa and A-tri pb ) suggests close protein interaction and is in excellent agreement with the determined A-tri glycointeractome by STD NMR spectroscopy (see Fig. 2).
Specificity and affinity of A-type HBGA binding to VP8*. To quantify the affinity of A-tri and A-tetra when bound to rotavirus VP8* protein, we have determined K D -values using 1 H-15 N heteronuclear single quantum correlation (HSQC) chemical shift perturbation NMR experiments 43,44 . Upon addition of the ligand to uniformly 15 N-labelled VP8* protein, changes in 1 H and 15 N chemical shifts can only be detected for those backbone amides that are in close contact with the ligand. A K D -value of 1.2 mM has been previously reported using the same approach for the sialidase-sensitive rhesus rotavirus (RRV) VP8* protein and 3 0sialyllactose as ligand 14 . In the current study, we have determined the K D -values for A-tri and A-tetra following the glycan induced chemical shift perturbations of the backbone amides of K8 and HAL1166 VP8* for ten residues that show the strongest changes (Fig. 6). A complete 1 H-15 N HSQC spectral analysis of these perturbations was undertaken ( Supplementary Figs 7-9). We found that the affinity of A-tri is significantly stronger for HAL1166 VP8* (K D 1.5 mM) compared to K8 VP8* (K D 3.4 mM).
Interestingly, the blood group-A tetrasaccharide (A-tetra) showed  (Fig. 6). This can be explained by the fact that A-tetra does not adopt a,b-anomeric mixtures of both the pyranose and furanose forms in solution as does A-tri where only the aand b-pyranose forms of Gal showed binding affinity to VP8* and not the aand b-furanose forms. Nevertheless, A-tetra also shows a higher affinity for HAL1166 VP8* over K8 VP8*. No changes in the total number of chemical shift perturbations were identified for protein complexes with A-tetra compared to A-tri. This confirms the STD NMR outcomes that the additional GlcNAc residue does not interact with the protein and that the binding epitopes of A-tri and A-tetra are identical ( Supplementary Figs 8 & 9). No significant chemical shift perturbations were observed for human rotavirus Wa VP8* in the presence of A-tri (58 mM), providing further evidence that A type HBGAs do not bind to Wa VP8* ( Supplementary Fig. 10). Interestingly, addition of Lewis b even at a very high concentration (19.7 mM), to 15 N-labelled Wa VP8* also did not result in any chemical shift perturbations. These results are in agreement with our STD NMR results and indicate that Lewis b does not bind to VP8*.
A-Tri blocks cell binding by VP8* of HAL1166 but not Wa.
A-tri inhibits infection by K8, HAL1166, RV-3, ST-3 and DS-1 but not Wa. The effect of A-tri treatment on MA104 cell infection by a range of human rotaviruses was determined. All of these  viruses were passaged to a similar low extent. Infectivity measurements were more sensitive than VP8*-cell binding assays for the detection of A-tri use, as a 37% decrease in K8 infectivity following exposure to 5 mM A-tri was observed (P ¼ 0.0009; Fig. 7b). As expected from the increased HAL1166 infectivity in cells genetically modified to express A-tri (ref. 23), the infectivity of HAL1166 in MA104 cells was inhibited by 42% after A-tri exposure (t-test, Po0.0001). In contrast, the presence of A-tri did not alter Wa infectivity (t-test, P ¼ 0.09; Fig. 7b), in excellent agreement with our NMR results. Overall, we have observed that K8 rotavirus is less dependent on A-type antigen for VP8*-cell binding than HAL1166, and shows a similar dependence on this antigen to HAL1166 for infectivity. In contrast, Wa rotavirus does not recognize this antigen. Consistent with the A-tri recognition of the VP8* of DS-1 and RV-3 detected by STD NMR (Fig. 3), A-tri at 5 mM inhibited the infectivity of DS-1 and RV-3 rotaviruses, by 36% (t-test, P ¼ 0.001) and 87% (t-test, Po0.0001), respectively (Fig. 7c). Infectivity blockade was still evident at 2 mM A-tri for RV-3 (88%; t-test, Po0.0001) and another P [6] human rotavirus ST-3 (79%; t-test, Po0.0001), but not DS-1 (Fig. 7c).

H type-1 and Lewis b HBGA do not inhibit P[II]/[III] viruses.
We could not detect any blockade of HAL1166, K8 or Wa rotavirus infectivity by H type-1 and Lewis b antigen at a 10 mM concentration ( Supplementary Fig. 12). This is supported by both STD NMR and 1 H-15 N HSQC chemical shift perturbation NMR experiments, which revealed no H type-1 and Lewis b antigen recognition by HAL1166, K8 or Wa rotavirus VP8* ( Supplementary Figs 1, 4 and 10). Similarly, 10 mM H type-1 or Lewis b antigen did not affect the infectivity of RV-3, ST-3 or DS-1 ( Supplementary Fig. 12).  [4]) and ST-3 (P [6]), whose VP8* were previously reported to bind one or both of these antigens 25 . We have discovered that A-type HBGA interacts with the VP8* of human P [4] and P [6] rotaviruses. In these complexes, we determined, for the first time, an involvement of the fucose moiety with the protein surface. This engagement may help explain why A-tri reduces infection by P [6] rotavirus RV-3 more substantially than the P [14] HAL1166 and P [9] K8 strains. However, as the DS-1 infectivity reduction by A-tri was similar to that of HAL1166 and K8, other factors are also likely to be involved, including the difference in usage of other receptors such as integrins between DS-1 and RV-3 (ref. 45).

Discussion
Taken together, our data revealed that HBGAs are used as rotavirus receptors, consistent with their abundance on the surface of mucosal epithelia of the gastrointestinal tract. However, it is notable that major rotavirus strain variations in HBGA recognition were detected. The relevant human clinical strain Wa did not recognize HBGAs, which is consistent with the observed lack of influence of genetically controlled expression of different HBGAs on pediatric rotavirus gastroenteritis 46,47 . Interestingly, two recent studies suggested that non-secretor individuals (null homozygote fucosyltransferase, FUT2) are not susceptible to infection with the most common rotaviruses 48,49 . However, this finding appears inconsistent with the well-established fact that by 2-3 years of age, essentially all children excrete rotavirus at least once and/or seroconvert to rotavirus regardless of their secretor status 50,51 . Although only a small number of children were analysed, a further recent study 41 may shed light on this discrepancy with its finding that non-secretors may be preferentially infected by P[6] rather than P [8] rotaviruses. This also indicates that all children are susceptible to rotavirus infection, independent of their blood group. Consequently, the interdependence of rotavirus infection and disease on secretor status remains questionable. The possibility that rotaviruses, such as noroviruses, may bind HBGA-like substances on intestinal bacteria also should be considered 33,52 , especially in the light of recent evidence that antibiotic treatment reduces rotavirus infection 53 . The complex regulation of fucosylation by FUT2 on intestinal epithelium, proposed to be controlled by innate lymphoid cells and involve IL-22 induction by commensal bacteria 54,55 also may contribute to the apparent inconsistencies between studies of the roles of HBGA in rotavirus receptor usage and disease. Our study indicates that HBGA play a role in host cell invasion by several human rotavirus genotypes. However, the clear differences we found between rotavirus strains in their ability to use HBGA show that HBGA may not be essential in this process. In this context, it is important to note that FUT2 activity is not restricted to the biosynthesis of H type-1 precursors (Galb1,3GlcNAc). Glycolipids in the epithelial tissues of the gastrointestinal tract are also fucosylated 56 by FUT2. Clearly, secretor status of the individual and FUT2 enzyme activity are not correlated with blood group specificity. More extensive structural, virological and epidemiological studies are required to shed light on the exact role played by HBGAs in rotavirus infection and disease.
Carbohydrates and reagents. Deuterium oxide was purchased from Sigma Aldrich (Australia) and all glycans, that is A antigen trisaccharide (A-tri), A antigen tetrasaccharide (A-tetra), Lewis b and H type-1 antigens, were purchased from Elicityl (France).
VP8* protein production. Plasmids expressing the gene encoding the VP8* core proteins of HAL1166 (VP8* 64-224 ), K8 (VP8* 64-224 ), Wa (VP8* 64-223 ), RV-3 (VP8*  ) and DS-1 (VP8*  ) for NMR and rotavirus-binding assays were synthesized (HAL1166) or cloned from virus (Wa, K8, RV-3 and DS-1) as described previously for Wa (ref. 12). In brief, complementary DNA produced from viral dsRNA by reverse transcription was used as template for PCR with specific oligonucleotide primers (Supplementary Table 1) to produce the VP8* gene fragment, which was cloned into the pGEX-4T-1 vector (GE Healthcare). DNA sequencing indicated that the predicted amino acid sequence of pGEX-Wa-VP8* was identical to the published Wa VP4 sequence 59 (GenBank accession code L34161), except for leucine substituting a phenylalanine as position 163; pGEX-K8-VP8* amino acid sequence was identical to the published sequence (GenBank accession code Q01641), except for a change from tyrosine to aspartic acid at amino acid position 149; and the sequence of pGEX-RV-3-VP8* 22 was identical to the most recent database entry for RV-3 VP4 (GenBank accession code FJ998273). Our sequence and FJ998273 both differ from the original published sequence (GenBank accession code U16299) in showing serine rather than proline at aa position 71 (refs 60,61). The pGEX-DS-1-VP8* amino acid sequence was identical to that published (GenBank accession code HQ650119). The VP8* gene fragment of HAL1166 was synthesized (GenScript) with 5 0 -BamH1 and 3 0 -EcoR1 restriction sites and ligated into the pUC57-Kan cloning vector through a single blunt-end EcoRV restriction site. Following digestion with BamH1 and EcoR1 of the pUC57-HAL1166-VP8* plasmid, the VP8* gene fragment was ligated into pGEX-4t-1, yielding pGEX-4t-1-HAL1166-VP8*. The final expression plasmid encoded for amino acids 64 to 224 of the VP4 protein and was fused to an N-terminal GST protein. For STD NMR, trNOE and cell-binding assays, the GST portion was not cleaved from the VP8* proteins. Protein expression was performed as described previously 19 . To prevent the formation of insoluble products, inclusion bodies and aggregates, the expression protocol was optimized for RV-3 VP8* by adding 20 ml l À 1 of a 50 Â stock solution of Augmedium (Athena Environmental Sciences, Inc., Baltimore, MD, USA), to precondition the cells and induce expression of chaperone proteins, followed by 30 min incubation. Isopropylthiogalactoside was added to reach a final concentration of 1 mM, to induce protein overexpression. Frozen cell pellets were thawed in PBS (137 mM NaCl, 2.7 mM KCl, 10 mM Na 2 HPO 4 , 2 mM KH 2 PO 4 pH 7.3) supplemented with 1 mM phenylmethylsulfonyl fluoride (Roche Diagnostics). The cells were lysed with 1 mg ml À 1 lysozyme supplemented with 1% (w/v) Triton X-100 and 20 mg ml À 1 DNaseI. Cell debris was removed by centrifugation at 20,000g for 30 min and the supernatant was passed over a glutathione-Sepharose column (Amersham-Pharmacia Biotech). The column was washed with PBS-binding buffer and bound GST fusion protein was eluted with 50 mM Tris-HCl pH 7.4 supplemented with 10 mM glutathione. 2) and incubated for 1 H at 25°C before expression was induced by the addition of 0.1 mM isopropylthiogalactoside. Cells were harvested after 4 h. Purification including removal of the GST fusion protein was accomplished as previously described 19 .
Standard NMR experiments. Chemical shift assignment of A-Tri and A-Tetra in 20 mM deuterated phosphate buffer and 70 mM NaCl, pH 7.1, was achieved by COSY, TOCSY, HSQC, 1D NOESY and 1 H NMR experiments at 280 and 305 K. Chemical shift assignments were in agreement with published values 27 .
STD NMR experiments. All STD NMR spectra were acquired in Shigemi Tubes (Shigemi, USA) with a Bruker 600 MHz Avance spectrometer at 280 K using a conventional 1 H/ 13 C/ 15 N gradient cryoprobe system. NMR experiments of A-tri in complex with VP8*-GST protein and GST as a control experiment were prepared using a protein concentration of 24 or 36 mM and a ligand concentration of 2.4 mM, resulting in a total protein-ligand ratio of 1:100 and 1:67, respectively. The final volume was 200 ml containing 20 mM deuterated phosphate buffer and 70 mM NaCl, pH 7.1. The protein was saturated with a cascade of 40 Gaussian-shaped pulses with duration of 50 ms each at À 0.1 p.p.m., resulting in a total saturation time of 2 s. The off-resonance was set to 33 p.p.m. and 1,024 experiments were acquired. A WATERGATE sequence was used to suppress the residual HDO signal. A spinlock filter with a strength of 5 kHz and duration of 10 ms was applied to suppress the protein background.
NOE and trNOE NMR experiments. NOE NMR experiments of free A-tri were acquired by diluting 1 mg of A-tri in 50 ml of 20 mM deuterated phosphate buffer and 70 mM NaCl, pH 7.1, at 305 K on a Bruker 600 MHz Avance spectrometer. The mixing time (t m ) was set to 800 ms and the relaxation delay to 2 s. trNOE NMR experiments were carried out using 1 mg GST-Wa or GST-K8, with a total molecular weight of 44 kDa and 145 mg of A-Tri that equals a protein:ligand ratio of 1:12 in 20 mM deuterated phosphate and 70 mM NaCl, pH 7.1, at 305 K and 600 MHz. The mixing time (t m ) for VP8*-GST protein was set to 150 ms to avoid false-positive NOEs due to spin diffusion effects. A trNOE NMR experiment of GST-K8 protein using the identical experimental setup but without ligand was acquired to subtract it from the trNOE with ligand to remove water background. In a first analysis step, the amide shifts were calculated for the highest ligand concentration weighing proton shifts over nitrogen by a factor of five d( 1 H, 15 Supplementary Fig. 7). A threshold value of Dd( 1 H, 15 N) ¼ 0.04 p.p.m. is considered as binding of the corresponding amino acid. Alternatively, this threshold is set to the s.d. of all chemical shifts. However, the full analysis shows that both alternative threshold values are very similar for the binding of A-tri and A-tetra to the VP8* proteins of the RV strains K8 and HAL1166 (s.d. for HAL1166 A-tri ¼ 0.036, HAL1166 A-tetra ¼ 0.042, K8 A-tri ¼ 0.037, K8 A-tetra ¼ 0.040; Supplementary Figs 8 and 9). Notably, the s.d. for the binding of A-tetra is slightly higher compared with A-tri. Binding affinities were calculated on the basis of the ten strongest backbone amide chemical shifts, which all showed a characteristic pattern of shift movements confirming the specificity of those changes ( Supplementary Figs 8 and 9). The chemical shifts were quantified by the chemical shift index for those strongly affected backbone amides according to the following formula: oI4 ¼ S{[(N 0 À N)/0.5] 2 þ [(H 0 À H)/0.1] 2 } 1/2 , where N 0 and N are the amide 15 N chemical shifts in the absence and presence of ligand, respectively; H 0 and H are the backbone amide hydrogen chemical shifts in the absence and presence of ligand, respectively 11 . Calculations were performed using GraphPad Prism 6.0, by using nonlinear regression following a one-site total binding model according to oI4 ¼ B max *[A-tri]/(K d þ [A-tri]) þ NS*[Atri] þ Background, with B max the maximum specificity binding and NS the slope of nonspecific binding.
Homology modelling of K8 VP8*. A homology model of K8 VP8* protein was built using the X-ray crystal structure HAL1166 X (pdb accession code: 4DS0) as template using YASARA Structure molecular modelling package (Ver. 13.9.8) 62 ( Supplementary Fig. 5). The HAL1166 structure (pdb code: 4DRV 23 ) was chosen as a suitable template, as it has a sequence identity of 83.9% and sequence similarity (BLOSUM62 score is 40) of 90.7%. The subsequent, fully unrestrained simulated annealing minimization, run for the entire model, resulted in an overall quality Zscore of À 0.081 and indicates a high-quality homology model. Both proteins share the amino acid sequence (Ser187-Tyr188-Leu189-Leu190-Thr191) that has been identified as containing key residues for HBGA binding in the case of HAL1166 VP8* 23 . After the side chains had been built, optimized and fine-tuned, all newly modelled parts were subjected to a combined steepest descent and simulated annealing minimization (that is, the backbone atoms of aligned residues were kept fixed to avoid potential damage. The hm_build.mcr macro of the YASARA package with default parameters was used. MD simulations. To analyse the flexibility of the a(1,2)-glycosidic linkage of A-tri, the three-dimensional structure HAL1166 VP8* protein in complex (pdb 4DRV) 23 a 2.8Ns MD calculation was performed using the YASARA Structure molecular modeling package (Ver. 13.9.8). The model was subjected to further refinement using the md_refine.mcr macro using the AMBER99 force field. Simulation parameters were kept at the values defined by the macro.
Flow cytometric analysis of VP8*-cell binding. Binding by GST-VP8* from HAL1166, K8 and Wa rotaviruses to MA104 cells was assayed in the same experiment at 4°C as previously described 22 . The optimal (saturating) concentration of each GST-VP8* was initially determined from its dose-dependent cell-binding curve. For A-Tri blockade studies, A-Tri (2 mM or 10 mM final conc.) or DMEM was reacted for 1 H at 4°C with the optimal level of GST-VP8* for HAL1166 (150 mg ml À 1 ), K8 (600 mg ml À 1 ) or Wa (300 mg ml À 1 ), then added to cells for 45 min. Stained single cells were analysed by flow cytometry as previously described 22 . The results provided are representative of those obtained in three independent experiments.
Assays for inhibition of rotavirus infectivity. A-tri (2, 5 or 10 mM), H type-1 antigen (10 mM), Lewis b antigen (10 mM), rotavirus-neutralizing monoclonal antibody (as mouse ascites fluid) or DMEM diluent were incubated with trypsinactivated rotavirus for 1 h at 37°C. The antibodies used were directed either to VP4 or to rotavirus outer capsid protein VP7, which confers glycoprotein (G) serotype specificity. Antibodies comprises the G8-specific B37:1 (ref. Following inoculum removal and washing as above, infected cells were incubated for 15 h at 37°C in 95% (vol/vol) air with 5% (vol/vol) CO 2 . Acetone-fixed cell monolayers were sequentially stained with rabbit antiserum to rotavirus (1 in 1,000) and fluorescein isothiocyanateconjugated goat anti-rabbit IgG (Invitrogen; 1 in 80), and virus titres determined by fluorescence microscopy as described previously 68 . For all assays, data represent the mean of triplicate samples and two independent experiments. Statistical analysis. Student's t-test was used, with significance set at the 95% level. Error bars on graphs indicate the s.d.