Avidity of α-fucose on human milk oligosaccharides and blood group–unrelated oligo/polyfucoses is essential for potent norovirus-binding targets

There is agreement with respect to norovirus infection routes in humans regarding binding of the pathogen to gastrointestinal epithelia via recognition of blood group–active mucin-typeO-glycans as the initiating and essential event. Among food additives playing a potential role in applications to protect newborns, human milk oligosaccharides (HMOs) as competitors are of major importance. By focusing on fractions of high-molecular mass HMOs with high fucose contents, we attempted to identify the structural elements required for norovirus GII.4 (Sydney 2012, JX459908) capsid binding in neoglycolipid-based arrays. We provide evidence that HMO fractions with the strongest binding capacities contained hepta- to decasaccharides expressing branches with terminal blood group H1 or Lewis-b antigen. H2 antigen, as recognized by UEA-I lectin, is apparently not expressed in high-mass HMOs. Beyond affinity, sterical and valency effects contribute more to virus-like particle binding, as revealed for oligovalent fucose conjugates of α-cyclodextrin and oligofucoses from fucoidan. Accordingly, high-mass HMOs with oligovalent fucose can exhibit stronger binding capacities compared with monovalent fucose HMOs. The above features were revealed for the most clinically relevant and prevalent GII.4 strain and are distinct from other strains, like GII.10 (Vietnam 026, AF504671), which showed a preference for blood group Lewis-a positive glycans.

Norovirus infections belong to the most common causes of human gastroenteritis worldwide, and epidemic outbreaks are responsible for hundreds of thousands deaths annually, mainly among children under the age of 3 years. Data from the NoroNet network indicate that increased activities of the virus during November and December of 2012 were related to the development of a novel, highly virulent genotype II.4 variant, first reported in Australia (Sydney 2012, JX459908) (1) and currently the most prevalent strain.
Noroviruses, a genus in the Caliciviridae family, are small nonenveloped viruses that contain a single-stranded RNA genome surrounded by a capsid protein. The capsid is formed by assembly of 180 copies of the major capsid protein, VP1, and a low copy number of the minor capsid protein, VP2 (2). Based on variations within the VP1 gene, noroviruses are classified into seven genogroups, termed GI-VII (3). The GII norovirus strains are responsible for most infections in humans (4). To establish an infection, the virus needs to bind to the gastrointestinal epithelia, and this binding is at least partly mediated by specific interactions of a lectin-like activity in the norovirus capsid and blood group-active mucin-type O-glycans on membrane glycoproteins or mucins (5,6). Evidence is accumulating that fucose, a structural part of histo-blood group antigens (HBGAs) 2 or Lewis-like antigens, plays a major role in recognition of the protruding domain of the VP1 capsid protein of GII norovirus strains, including the highly infectious GII.4 and the recently emerged GII.17 noroviruses. ABH and Lewis fucoses of blood group antigens are introduced by the FUT2 and FUT3 fucosyltransferases, and homozygously recessive individuals lacking these enzymes were less susceptible or even resistant to an infection by certain strains in human challenge and outbreak studies (7,8). Certainly, similar to the recently identified cellular protein receptor for murine noroviruses, there might be other host factors required for attachment and entry of human noroviruses (9).
One of the strategies to combat the virus by preventing its binding to epithelial surfaces is based on food additives, like human milk oligosaccharides (HMOs), that mimic the structure of blood group-active mucin-type O-glycans. Human milk oligosaccharides represent an ideal source of potential competitors of viral glycan receptors, and although many of their struc-tural details were elucidated, knowledge regarding their precise structure of high-mass HMOs is limited. One of the smaller but strongly expressed HMOs in the milk of secretors, the trisaccharide 2-fucosyllactose (2ЈFL), has gained attention, as it is able to block norovirus binding quite efficiently (10,11). Although 2ЈFL has reached market approval as a safe food additive, the question remains whether other more complex glycans in the fraction of high-mass HMOs might contain components with higher competitive activity in norovirus binding and whether these glycans might give clues to the fine structural requirements of norovirus binding to host epithelia.
Based on these considerations, we have developed a neoglycolipid-array strategy that allowed us to screen complex fractions for components with high-binding capacity. These HMO fractions were characterized after HPLC separation of components by mass spectrometric analysis of their structures. Independent of this approach, we followed a second route of considerations, as it became evident that some of the observed effects exerted by HMOs were induced by epitope valency and were unrelated to blood group structures. In this context, we generated a series of ␣-cyclodextrin dendrimers presenting ␣-fucosyl residues at varying degrees of scaffold substitution and revealed stereochemical/valency-dependent enhancement of competitive activity in norovirus-binding assays. These findings with synthetic compounds were finally supported by binding and inhibition studies with natural compounds, poly-and oligo-fucoses, corresponding to native and fragmented fucoidan.

Screening of HMO neoglycolipid arrays reveals preference of norovirus GII.4 binding to blood group H1 on high-mass glycans
Fractions of HMOs, each containing about 5-20 different high-mass glycans with different degrees of fucosylation, were included in an in-house prepared neoglycolipid array. In addi-tion to these HMO fractions, a series of commercially available standard HMOs sizing up to hexasaccharides were included in the array. Strong binding activity of norovirus GII.4 VLPs was revealed for three HMO fractions (fractions 2, 7, and 46), whereas seven other fractions showed intermediate or weak VLP binding, and 32 HMO fractions were negative (Fig.  1A). Among the standard HMO compounds, LNFP-I and LNneoFP-I with monovalent blood group H1-and H2-trisaccharide at the nonreducing termini were strongly active (Fig. 1B). Smaller HMOs in the trisaccharide and tetrasaccharide range were inactive (2ЈFL, 3FL, difuco-lactotetraose, data not shown), probably due to sterical effects preventing access of the bulky VLPs to the lipid-embedded neoglycolipids. Also, all tested sialylated neoglycolipids covering the tri-to pentasaccharides 3-sialyllactose, 3-sialyl-N-acetyllactosamine, and sialyllacto-N-tetraose c were inactive (data not shown). In conclusion, GII.4 (Sydney, 2012, JX459908) VLPs bind with high efficiency to blood group H-positive HMOs, without discriminating between H1 and H2 antigens on LNFP-I and LNneoFP-I, respectively.

Comparative profiling of HMO fractions for GII.4 (Sydney, 2012, JX459908) VLP binding versus blood group H2 and Lewis-b activity
To reveal the relative proportions of blood group H2 epitopes in HMO fractions, we tested the neoglycolipid array and standard HMOs with the H2-specific lectin UEA-I for binding ( Fig. 2A). Although LNneoFP-I (H2-active) stained strongly positive with the lectin, neither LNFP-I (H1-active) nor any of the VLP binding-positive or -negative HMO fractions were stained by the lectin ( Fig. 2A and Fig. S2). Accordingly, the VLP-binding components in high-mass HMO fractions apparently express blood group H1.
Similarly, a panel of VLP-binding HMO fractions and standard HMOs was tested for their relative activities of the blood group H1-related Lewis-b antigen (Fig. 2B). Fractions 2, 7, and  Table S1. Two micrograms of neoglycolipid were immobilized per well in a duplicate assay on polystyrene plates. B, neoglycolipids derived from standard HMOs (LNFP-I, lacto-N-fucopentaose-I; LNneoFP-I, lacto-N-neofucopentaose-I; LNFP-II, lacto-Nfucopentaose-II) and selected HMO fractions were tested in triplicate for comparative norovirus GII.4 VLP binding.

Specificity and valency of norovirus binding targets
46 with high VLP-binding capacities were found to express high levels of Lewis-b antigen, which is in accord with the assumption that in particular type 1-based blood groups H1 and Lewis-b are responsible for the high binding capacities in highmass HMO fractions. However, also a series of negative VLPbinding fractions was tested positive for Lewis-b epitopes. A potential explanation could be the partial cross-reactivity of the antibody with Lewis-a, as revealed for the neoglycolipid of LNFP-II (Fig. 2B).

VLP-binding activity and fucose contents of HMOs
Valency of active glycotopes and hence the stereochemistry of their presentation to VLPs are as expected important determinants for the strength of binding. Accordingly, among highmass glycans with branched structures, we can expect species with higher numbers of fucose and higher VLP-binding activity.
Mass spectrometric profiling of selected HMO fractions (Table  1) revealed 24 distinct species based on compositions of monosaccharides. Some of these compositional species defined by their M ϩ Na ϩ masses were found in only one HMO fraction (see trifucosylated species in fraction 49), although others were present in several fractions. Of particular interest are those HMO fractions containing species with high numbers of fucose residues (fractions 7, 46, and 49), as these may support the hypothetical relation between VLP-binding activity and degree of fucosylation. Fraction 49, containing glycans with up to three fucose residues, exhibited low VLP-binding activity, most likely because the structure of the dominant undecasaccharide (refer to MS/MS fragmentation data given in Fig. S3) revealed no H-type but Lewis-a fucosylation. Fractions 7 and 46 contain high proportions of species with up to five and four fucose residues, respectively, and showed strong VLP-binding

Specificity and valency of norovirus binding targets
activity. Also fraction 2 exhibiting strong VLP-binding activity contains glycans with up to two fucose residues. In summary, these findings may support a relation of fucose content in HMO species and their VLP-binding capacities. However, valency effects cannot be clearly discriminated from contributions by structural specificity.
In support of the hypothesis that valency effects might be important in norovirus capsid binding, 2ЈFL and LNFP-I, which both carry monovalent fucose, exhibited equally low activity in VLP-binding inhibition assays (Fig. 3). The IC 50 values of both HMOs were above 30 mM. Accordingly, a blood group H1-active HMO, the assumed specific target of the viral lectin, appears to have a similar low affinity in binding competition on gastric mucins as the reference HMO 2ЈFL.

Structural analysis of HMO fractions with high VLP-binding activity
Oligosaccharides in HMO fraction 46 were separated by reversed-phase chromatography on graphitized carbon, and HPLC subfractions 1-12 were tested after conversion into neoglycolipids for GII.4 VLP-binding capacities (Fig. 4A). Subfraction HPLC-7, as the fraction with the highest VLP-binding activity, was chosen for further analysis of glycan components. Four dominant glycan species were revealed by MALDI-MS profiling of oligosaccharides as the major components (Fig. 4B). The molecular ions registered at m/z 1241, 1387, 1533, and 1679 indicated the presence of mono-to tetrafucosylated hexasaccharide cores of the composition F1H4N2, F2H4N2, F3H4N2, and F4H4N2, respectively (Fig. 4B). The corresponding permethylated glycans registered at m/z 1550, 1724, 1898, and 2072, respectively, were analyzed in the post-source decay mode to obtain fragmentation spectra and sequence information ( Fig. 4B). According to these MALDI-MS/MS spectra, the trifucosylated nonasaccharide at m/z 1898 exhibits two nonreducing termini branching at the galactose of the lactose core, as indicated by the sodium adducts of primary fragment ions B 3 (m/z 660.3), B 4 (m/z 834.4), Y 2␤ (m/z 1086.5), Y 2␣ (m/z 1260.6), Z 4␤ (m/z 1487.7), and Z 5␣ (m/z 1691.8).
These structural features of components in subfraction HPLC-7 are reflected in the patterns of partially methylated alditol acetates identified by GC-MS in HMO-46 (Fig. 5). The glycans in this fraction contain a 3,6-disubstituted galactose (3,6-Gal) and 3,4-disubstituted N-acetylglucosamine as the major hexose and hexosamine derivatives indicating the presence of a backbone branch at the lactosyl core and Lewis-type structures in the two branches. A minor proportion of the branches terminates with a galactose, and less than about 30% of the branches yields 4-monosubstituted GlcNAc, which indicates the presence of H-type structures.

Lewis-b and H-type 1 active glycans on human gastric mucins are preferred ligands for GII.4 VLP binding
Mucins isolated from human gastric juice express preferentially O-linked glycans with blood group ABH and Lewis-type structures (12). Testing GII.4 VLPs for their binding to mucins from blood group Lewis-defined individuals revealed strong activity on Lewis-b and Lewis-(Ϫ)/H1 mucins, but only weak binding to Lewis-a mucins (Fig. 6A). This binding pattern contrasts with that observed for GII.10 VLPs, which did not discriminate between mucins from Lewis-a and Lewis-b individuals (Fig. 6B). The finding is in agreement with data from neoglycolipid arrays, where GII.10 VLPs bound most strongly to Lewis-a active, branched HMOs, 3 whereas GII.4 VLPs did not bind to LNFP-II exposing the Lewis-a trisaccharide (Fig. 1B).
Arrays of neoglycoproteins revealed binding of GII.4 VLPs to blood group A and B trisaccharides, weaker binding to H-disaccharide, and no binding to sialylated Lewis epitopes (sLe-a and 3 F. G. Hanisch, T. Feizi, and G. S. Hansman, unpublished observations. Table 1 Human milk oligosaccharide fractions positive with GII.4 (Sydney, 2012, JX459908) VLPs MALDI mass spectrometry of native oligosaccharides in the positive-ion mode is shown. S is NeuAc; F is fucose; H is hexose; and N is N-acetylhexosamine. Molecular masses given in boldface represent compositional species with high fucosylation. HMO fractions in boldface contain mostly highly fucosylated species. * indicates that these components in fraction 2 were detectable in trace amounts. ** indicates HMO species in fraction 49 with more than two fucosyl residues were structurally characterized to expose terminal Lewis-a trisaccharides but no H type 1 or Lewis-b.

Specificity and valency of norovirus binding targets
sLe-x) or to LSTc pentasaccharide. Again, part of these data needs to be critically evaluated under the assumption that short glycans in the di-to trisaccharide range might reveal artificial results, similar to 2ЈFL, which is inactive as a neoglycolipid but active as a free trisaccharide in solution.

Valency/cluster effects demonstrated with fucoidan and oligo-␣-fucose-␣-cyclodextrin dendrimers
Among natural compounds, the polyfucoses (fucans) with structural features that are unrelated to blood group Lewis and H antigens were tested for their competitive activities in

Specificity and valency of norovirus binding targets
norovirus-to-mucin-binding assays (Fig. 7A). Not only the native fucoidan from Fucus vesiculosus but also a solvolytically desulfated and partially fragmented fucoidan were strongly inhibitory in GII.4 VLP-binding assays on human gastric mucins. These findings strongly indicate that the observed competitive effect is independent of sulfation of the polyfucose. Fucus species have their central chains composed of repeating (133)-and (134)-linked ␣-L-fucopyranose residues, and some minor amounts of ␣-L-fucopyranose are (1-2)-linked to fucoses of the core chains (Fig. 7B) (13). The fragment sizes ranged up to dp 20 -25 (Fig. 7C). The structural aspects of fucoidan are unrelated to the blood group H antigen, because ␣-L-fucose is linked (1-2) to L-configurated 6-deoxygalactose (L-fucose) instead of its D-stereoisomer.
Based on observations with fucoidan, we asked the question whether a mere clustering of ␣-fucosyl residues on sugar-type scaffolds, like ␣-cyclodextrin, could increase inhibitory activity of free ␣-fucose, which exhibits IC 50 values in the range of 60 -80 mM. A further reference of a monovalent sugar with norovirus binding inhibitory capacity, 2ЈFL, blocks GII.4 VLPs with IC 50 values more than 30 mM (see above). We tested dendrimers of ␣-fucose substituted at varying degrees to ␣-cyclodextrin: FCD1, FCD2, and FCD3 (Fig. 8). Whereas the lowsubstituted FCD1 with less than one fucose per molecule was of weak activity in VLP-binding competition, the FCD2 preparation with two fucosyl residues per molecule, on the average, showed increased activity with IC values in the range of 20 -30 mM. Strongest effects were seen for the FCD3 with three fucose residues per molecule, and IC 50 values of about 4 mM. The structural features of the fucose dendrimers were revealed by MALDI-MS (profile of substitution isomers, Fig. 9A), 1 H NMR (anomery of fucose, Fig. S1), and GC-MS linkage analysis VLPs to human gastric mucins from blood group Lewis-defined individuals (triplicate assay); the sample derived from individual Lewis-a positive gastric juice corresponds to sample HGM-Le-a (bar 1) in B. B, binding of norovirus GII.10 VLPs to human gastric mucins from blood group Lewis-defined individuals (triplicate assay); gastric juice from three independent individuals of blood group Lewis-a were tested (HGM- Le-a (1-3)). (Fig. 9B). The major portion of the L-fucose was ␣-linked, and more than 98% was linked 1-6 to glucose in the cyclodextrin scaffold.

Discussion
Numerous studies have addressed the blood group dependence of norovirus binding to human gastrointestinal epithelia as an infectivity factor (5). X-ray crystallography studies of the GII noroviruses revealed that there is a major contribution of terminal ␣-fucose to a functional epitope (14), whereas other parts of the glycan are much less involved. These additional interactions can be quite complex, and evidence from X-ray crystallography suggests that adaptive movements of both the capsid loop and the HBGA glycan can account for these. Whereas fucose in a blood group H2 trisaccharide is held by six direct hydrogen bonds, the central galactose forms only two and the subterminal GlcNAc just one direct hydrogen bond with the capsid protein. There is also evidence that the H2-and H1-trisaccharides in the P2 pocket are oriented similarly and exhibit similar numbers of direct interaction with the protein.
Only a few other reports have dealt with these aspects of fine specificity referring to the contributions of subterminal sugars (see below). On the other side, epitope valency and epitope presentation in clusters have come into the research focus, in particular with respect to low-or medium-affinity interactions of glycans with lectin-like proteins, as revealed for norovirus P2 domains. The latter represents an important aspect when orientational constraints imposed on membrane glycoproteins or glycolipids are considered. For these reasons, we chose a screening strategy based on neoglycolipids that are immobilized by embedding their nonpolar tails in a lipid layer on poly-

Specificity and valency of norovirus binding targets
styrene surfaces. In this way, the polar glycans are presented at high densities to the liquid aqueous phase and can exert the desired clustering effects.
In contrast to X-ray crystallography studies, recently published Saturation-Transfer Difference NMR in-solution experiments showed strong contributions also of the sugars at the reducing end (15). It was demonstrated that linkage of the subterminal galactose to GlcNAc is an important structural determinant for effective binding, and different from type 1 blood group chains, type 2 structures yield very poor saturation transfer (15). Another study referred to glycosphingolipids derived from small intestinal epithelium of humans (type 1 chains) and dogs (type 2 chains) and revealed strong GII.4 capsid binding to all type 1 chains, but low or no binding to type 2 chains (16). Similarly, Shirato et al. (17) found more tight binding of GII.4 (1997, Narita104, AB078336) to type 1 HBGAs than to type 2 HBGAs. However, all the above studies have used GII.4 strains that are less prevalent (16,17), but in this study we examined a GII.4 (Sydney 2012) that is still a major strain and has been responsible for the most outbreaks worldwide since around 2012 (1).
The GII.4 variants have dominated over the past decade and have caused four pandemics. Although the ABH-fucosebinding amino acid residues in the P2 subdomain are typically conserved among GII.4 variants, the residues surrounding the pocket are more diverse. These mutations generate virus variants that might escape the host immune response. At the same time, these variants may alter the interactions with subterminal sugar units and create specific preferences to a certain glycan conformation. In agreement with the previous studies (15,16), GII.4 (Sydney 2012) VLPs displayed strong binding to H-type 1 and Le b structures in high-mass HMOs, however, in contrast to the previous studies without discriminating between blood . F refers to the mass increment of deoxyhexose (dHex). Up to six dHex residues were added to the scaffold by acid reversion with an average of about three residues. B, GC-MS analysis of partially methylated alditol acetates derived from FCD3. Three major signals were detected at ion traces m/z 175 (terminal Fuc at 8.8 min) and m/z 118 (4-Glc and 4,6-Glc at 11.2 and 12.3 min, respectively). The ratio of 4,6-Glc and 4-Glc in the ion trace at m/z 118 indicates that at least one-third of the available sites on the cyclodextrin scaffold is substituted with fucose. Minor or trace components were registered in the m/z 190 trace at 9.9 min (3-Fuc) and 11.9 min (2,4-Glc), which correspond to less than 2% of the base-peak area (4-Glc) in the TIC chromatogram. a.u., arbitrary unit.

Specificity and valency of norovirus binding targets
groups H1 and H2 on the standard lacto-N-fucopentaoses LNFP-I and LNneoFP-I. Lewis-a-positive glycans on gastric mucins are only weakly recognized by GII.4 (Sydney 2012), whereas the rarely detected GII.10 Vietnam026 showed similar binding to both Le a and Le b mucins.

Implications of the specificity toward H-type 1 and Le b for the infectivity
Interestingly, this pattern is not only limited to noroviruses. Other enteropathogens that utilize HBGAs as attachment factors exhibit similar specificity. For instance, the dominant rotavirus strains P [4] and P [8] also preferentially recognize H-type 1 and Le b structures, whereas the less prevalent P [19] rotavirus interacts with A, B, and H-type 1 HBGAs but not those containing the Lewis epitope (18). Similarly, the bacterial pathogen Helicobacter pylori recognizes H-type 1 and Le b HBGAs (19 -21).
H-type 1 and Le b HBGAs as found in gastrointestinal epithelia and secretions are products of the Secretor (FUT2) and Lewis (FUT3) genes. These genes are active in 80 and 90% of the general population, respectively. Accordingly, type 1 carbohydrates are highly abundant as glycan epitopes in the gastrointestinal tract of most individuals. Thus, high availabilities of the H-type 1 and Le b in the gastric epithelium of the general population might serve as a selection factor and define infectivity of many gastrointestinal microbes.

Does the length and branching of the carbohydrate chain matter for the binding of free HMOs?
The norovirus binding to HBGAs is likely dependent on the proper orientation of ABH and Lewis fucose. However, it is not clear whether this can play a role when HBGA/HMOs are present as free oligosaccharides. Because of the limited availabilities of HMOs, most studies so far have used HMOs conjugated to lipid or protein carriers. Only a few have tested HMOs as free oligosaccharides (2ЈFL and 3FL, or LNFP-I in this study). Based on surface plasmon resonance studies, there is not much difference between type 1 and type 2 core elements, when looking at binding of GII.17 capsids. 4 Most likely, the degree of fucosylation of HMOs or other antivirals and their proper stereochemical presentation with respect to occupancy of multiple HBGA-binding sites per virion are more important for effective inhibition.

Fucose multivalency as an approach for norovirus drug design
The conservation of the HBGA-binding site and its dependence on fucose recognition can serve as a starting point for development of anti-entry antivirals against noroviruses. Our recent work has identified two HMOs, 2ЈFL and 3FL, as possible broadly reactive antivirals. However, HMOs and other analogs of HBGA receptors bearing a single fucose unit are assumed to be less efficient in competing with multivalent virusmembrane glycan receptor interactions. Therefore, these single-fucose inhibitors have to be administered at higher doses. Alternatively, one can explore multivalent fucose analogs. For instance, Rademacher et al. (22) generated high-avidity norovi-rus binders by the attachment of a fucose-bearing compound to a polymer scaffold (polyacrylamide). In our approach, fucoses in fucosyl-cyclodextrins were attached to sugar-type scaffolds. Thus, they are potentially more safe and still of high potency. As an alternative to synthetic compounds, like fucosyl-cyclodextrins, natural products could be evaluated as antiviral food additives, like polyfucoses or fucans. Among the latter we explored the potential of native F-fucoidan from F. vesiculosus, desulfated F-fucoidan, and oligofucoses derived from the brown algal polysaccharide by partial acid hydrolysis or solvolysis. Although these oligo-L-␣-fucoses in the size range 2-20 dp do not exhibit any structural relationship to blood group glycans, they revealed strong inhibitory potentials in norovirus binding inhibition on human gastric mucins.

Production of norovirus VLPs
Two different strains of the genogroup GII were included in this study: GII.4 (Sydney, 2012, JX459908) and GII.10 (Vietnam 026, AF504671). VP1 protein was recombinantly expressed in insect Sf9 cells as described previously (23). Briefly, Sf9 cells were transfected with recombinant VP1 bacmids using Effectene. The cells were incubated at 26°C and harvested 6 days postinfection. The baculovirus harvest was clarified by lowspeed centrifugation and was used to infect Tn5 cells at 26°C. The cells were then harvested 6 days postinfection. The VLPs were concentrated by ultra-centrifugation and then purified by CsCl equilibrium gradient ultracentrifugation. The VLP band was collected from the side of the tube, and the VLP morphology was examined using EM.

Synthesis of ␣-fucosyl-␣-cyclodextrin dendrimers
A variable molar excess of L-fucose over ␣-cyclodextrin was solubilized in 0.12 weight % H 2 SO 4 and incubated for 1 h at 110°C to achieve variable substitution of the cyclic glucose scaffold with fucose by acid reversion chemistry. The products were separated from the bulk of fucose by carbograph solidphase extraction (150-mg cartridges, Extract Clean, Grace Davison Discovery Sciences, Lokeren, Belgium). Formation of preferential ␣-glycosidic fucose bonds to the C6 of glucose was confirmed by 1 H NMR and GC-MS linkage analysis, respectively. The degree of fucose substitution on ␣-cyclodextrin was analyzed by MALDI-MS. Three distinct preparations could be obtained with less than one (FCD1), two (FCD2), and three (FCD3) fucose residues per cyclodextrin on average. A method description for one-and two-dimensional 1 H NMR TOCSY spectrum of FCD3 is given in Fig. S1.

Preparation of F-fucoidan-derived oligo-fucoses
F-Fucoidan isolated from F. vesiculosus was purchased from Sigma (Munich, Germany). To achieve desulfation with partial fragmentation, the native sulfated F-fucoidan (30 mg) was protonated by treatment with Dowex 50WX8(Hϩ) cation-exchange resin and afterward neutralized with pyridine. The pyridinium salt was dried by vacuum rotation and incubated for further drying in the presence of phosphorous pentoxide overnight at 60°C in vacuo. The extensively dried pyridinium salt of the polysaccharide was then treated at 100°C with 5 ml of a mixture of DMSO/ pyridine (5:12, v/v) for 9 h. The solvolysis product was dialyzed with several changes against water and dried by vacuum rotation.

Preparation of neoglycolipid arrays
Neoglycolipids were prepared from reducing oligosaccharides as described previously (24) with minor modifications. In brief, HMOs (20 g) were taken up in 8 l of 40 mM imidazole buffer, pH 6.5, before 156 l of 1 mg of dipalmitoyl phosphatidylethanolamine/ml chloroform/ethanol (1:1, v/v) was added. The reaction mixture was kept at 50°C for 2 h, and after addition of 10 mg of sodium cyanogen borohydride/ml ethanol (8 l), it was incubated overnight under the same conditions. Dried by vacuum rotation, the neoglycolipids were solubilized in 500 l of methanol for application onto polystyrene plates by drying 50-l aliquots per well at 37°C.

Norovirus capsid binding and binding inhibition assays
VLP binding and inhibition assays were performed either on immobilized human gastric mucins (HGM, 10 g/ml) from pooled gastric juice of blood group undefined subjects or on HGM from blood group Lewis-defined individuals. For HMO inhibition experiments, immobilized porcine stomach mucins (type III, Sigma, 10 g/ml) were used. Other series of experiments were based on the above-described neoglycolipid arrays. Polystyrene-immobilized HGM, porcine stomach mucins, or neoglycolipid arrays were blocked with 5% BSA/PBS for 1 h at 37°C before VLPs were applied at 10 g of protein per ml of 0.05% Tween 20/PBS in the absence or presence of inhibitor. After incubation for 1 h at 37°C and three plate-washing steps with 0.5% BSA/PBS, bound capsids were detected with polyclonal rabbit GII.4 capsid protein antiserum (diluted 1:3000 in 0.5% BSA/PBS). A monoclonal anti-GII.10 was used in combination with anti-mouse IgG-biotin. Anti-rabbit Ig-alkaline phosphatase, diluted 1:5000, or anti-rabbit Ig-horseradish peroxidase (HRP), diluted 1:2500 in 0.5% BSA/PBS (GII.4), or StrepTactin-alkaline phosphatase, diluted 1:1000 (GII.10), were used for development (1 h at 37°C), followed by three washing steps and addition of the substrate: 1) para-nitrophenyl phosphate (1 mg/ml) in 0.1 M diethanolamine buffer, pH 9.8, containing MgCl 2 (0.5 mM) or 2) ortho-phenylenediamine/hydrogen peroxide in phosphate citrate buffer. Plates were read after 30 min at 405 or 490 nm in a Tecan Sunrise (Remote Control) run with Magellan software.
In binding inhibition experiments, appropriate amounts of inhibitor in the millimolar range were dried from stock solutions and resolubilized in 0.5% Tween 20/PBS containing 10 g/ml VLP protein. In binding analyses with anti-blood group Lewis-b, the monoclonal mouse antibody 2-25LE (25) (Biozol, Eiching, Germany) was used at 1:200 dilution. For lectin-bind-ing studies, the agglutinin from Ulex europaeus isoform I was used in a biotinylated form (Sigma), diluted to 20 g/ml in 0.5% BSA/PBS, and applied in 50-l aliquots onto neoglycolipid arrays prepared and developed as described above.

Chromatographic separation of milk oligosaccharides
Between 100 and 300 ml of mature human milk from different donors were defatted, deproteinized, and separated by gelfiltration and anion-exchange chromatography as originally described by Egge et al. (26). Identification and characterization of HMOs, particularly of isomeric structures, were achieved by high performance anion-exchange chromatography with pulsed amperometric detection (HPAEC-PAD) in combination with MALDI-TOF as described previously (27,28).
Separation of reducing oligosaccharides from human milk was performed by reversed-phase chromatography on 100 ϫ 2.1-mm Hypercarb column (Thermoquest, 5) run at 0.5 ml/min in a gradient from 0 to 80% acetonitrile in water (duration 20 min). UV detection of the carbohydrates was performed at 192 nm.

Linkage analysis by GC-MS
Permethylation of glycans was performed as described previously by Albersheim et al. (29). The completeness of derivatization was controlled by MALDI-MS, and the oligosaccharides were hydrolyzed, reduced, and acetylated to obtain partially methylated alditol acetates (29). Separation and identification of partially methylated alditol acetates were achieved on a Fison MD800 mass detector (Fison/Thermo) coupled to a gas chromatograph (GC8000 series). A temperature gradient from 60 to 100°C (40°C/min), and from 100°C to 260°C (10°C/min) was run on a Restec Rxi-Sil MS (15 m, 0.25-mm inner diameter, 0.25 m).