The binding mechanism of the virulence factor Streptococcus suis adhesin P subtype to globotetraosylceramide is associated with systemic disease

Streptococcus suis is part of the pig commensal microbiome but strains can also be pathogenic, causing pneumonia and meningitis in pigs as well as zoonotic meningitis. According to genomic analysis, S. suis is divided into asymptomatic carriage, respiratory and systemic strains with distinct genomic signatures. Since the strategies to target pathogenic S. suis are limited, new therapeutic approaches are needed. The virulence factor S. suis adhesin P (SadP) recognizes the galabiose Gal a 1–4Gal-oligosaccharide. Based on its oligosaccharide fine specificity, SadP can be divided into subtypes P N and P O . We show here that subtype P N is distributed in the systemic strains causing meningitis, whereas type P O is found in asymptomatic carriage and respiratory strains. Both types of SadP are shown to predominantly bind to pig lung globotriaosylceramide (Gb3). However, SadP adhesin from systemic subtype P N strains also binds to globotetraosylceramide (Gb4). Mutagenesis studies of the galabiose-binding domain of type P N SadP adhesin showed that the amino acid asparagine-285, which is replaced by an aspartate residue in type P O SadP, was required for binding to Gb4 and, strikingly, was also required for interaction with the glycomimetic inhibitor phenylurea-galabiose.


Introduction
Bacterial adhesion to host cell surfaces is a prerequisite for infectious disease. Host cell surfaces are heavily covered by surface carbohydrates that form the glycocalyx layer (1,2). Pathogenic bacteria and their toxins can treacherously exploit carbohydrates to attach and invade cells. One example is a group of endogenous glycosphingolipids, known as globo series membrane glycosphingolipids, which are grouped based on the Gala1-4Gal (galabiose) structure. Globo series glycolipids are cellular receptors for toxic action of Verotoxins and for the attachment of bacteria, such as uropathogenic Escherichia coli, Streptococcus suis and Pseudomonas aeruginosa (3)(4)(5)(6). Since antibiotics are increasingly losing their power, alternative antimicrobial compounds need to be invented and developed (7,8). Transformation of bacteria to antibiotic-resistant forms can be overcome by developing compounds that block in vivo colonization and virulence factors without killing the pathogens. Cell surface carbohydrate glycomimetics are the most tempting pipeline for the development of new generation of antimicrobials (9,10). Streptococcus suis is found as a commensal bacterium in all pigs and it is an important pig pathogen causing septicemia, pneumonia and meningitis. It also causes severe zoonotic meningitis. Based on clinical outcome, the S. suis strains can be classified as asymptomatic carriage, respiratory and systemic strains. Recent data show that S. suis produces an alarmingly vast array of factors causing high antibiotic resistance (11). Genomic analysis shows that the asymptomatic carriage and systemic strains have clear differences in their core and accessory genomes, whereas respiratory strains have more overlapping features (12). This could suggest that environmental factors such as overcrowding and synergistic activities with other keystone pathogens (13), can alter pigs to develop severe symptoms and pneumonia. Instead, the virulence genes are enriched in systemic strains and their accessory genomes contribute to the invasive S. suis infections and meningitis (12). There is multiple and redundant colonization and virulence factors required for S. suis disease. The distribution of virulence factors is also different depending on the geographical regions impacting the accessory genomes, which hampers the development of effective vaccines (14). In order to develop new therapeutics and vaccines the molecular mechanisms of accessory virulence factors need to be resolved in atomic level. S. suis strains expressing SadP adhesin bind specifically to Gala1-4Gal-containing oligosaccharides (15). Interestingly, galabiosebinding strains can be divided to subtypes PN and PO based on how they recognize galabiosecontaining carbohydrates in hemagglutination assays (16). Type PN strains recognize galactose, N-Acetylgalactosamine, and the ceramide-linked oligosaccharides Gb3 (Gala1-4Galb1-4Glc) and Gb4 (GalNAcb1-3Gala1-4Galb1-4Glc). Subtype PO recognizes only galactose and preferentially binds to Gb3. Recently, S. suis galabiose-binding adhesin SadP was identified as an LPNTGanchored cell-wall protein (6). The carbohydrate interaction of SadP has been shown to mediate S. suis binding to pharyngeal epithelium and intestinal epithelial cells (16,17). In an S. suis murine infection model, the mice deficient in globotriaosyl ceramide (Gb3) expression (knockout in alpha-1,4galactosyltransferase A4GALT) developed less severe brain inflammation and injury (18). Recently, low picomolar concentrations of glycomimetic 3-phenylurea-galabiose-containing dendrimers were shown to block S. suis SadP (type PN) adhesin binding activity (19). However, the molecular basis of differences in galabiose-binding mechanisms of SadP adhesins have remained elusive. In the present study, we have characterized the distribution of SadP subtypes in S. suis strains and the molecular binding mechanism of type PN and PO SadP to globosyl oligosaccharides. The results show that S. suis SadP type PN is distributed in the systemic strains and its specific N285mediated binding mechanisms to globotetraosyl ceramide distinguishes it from type PO sadP.

Sequence analysis of SadP subtypes PN and PO, structural modelling and distribution in S. suis systemic, respiratory and non-clinical strains
SadP adhesins are 80 kDa LPNTG-anchored cellwall proteins, which contain the signal sequence for secretion, the N-terminal galabiose binding domain, tandem repeat domains and C-terminal LPNTG cell-wall anchor domain (6). The galabiose-binding domains have no significant sequence homologs to other bacterial proteins or to other carbohydrate binding domains. The SadP Nterminal galabiose-binding domains (aa 31-328) of hemagglutinating type PN and type PO strains were cloned, sequenced and compared with multiple alignment (Fig. 1A). The structure consists of three a-helixes and ten b-strands (b1-b10) that form a bsandwich core domain. The sequences of the Nterminal galabiose-binding domain of type PN SadP adhesins are 100 % identical and are found in virulent serotype 2 strains. There is more sequence variation in the galabiose-binding site of type PO SadP. The carbohydrate binding sites of types PN and PO are highly conserved (Fig. 1B). Interestingly there is a conservative change at position 285 from asparagine to aspartate, which could have an effect for the interaction of type PN SadP with GalNAcb1-3Gala1-4Gal-oligosaccharide. The distribution of sadP genes in clinical and nonclinical strains analyzed with core genome phylogeny analysis is shown in Figure 1C. The gene of sadP type PN was found in 91.9 % of systemic S. suis isolates (88.1 % of strains isolated from brain), whereas only less than 1 % of strains had the gene encoding sadP type PO. Based on the maximum-likelihood core genome phylogeny analysis the type PN SadP gene is found in clonal virulent S. suis strains, which mostly belong to serotype 2. Interestingly, also serotype 14 strains isolated from human meningitis cases have type PN gene. Type PO sadP was found more frequently in respiratory and non-clinical strains. For the studies comparing the glycosphingolipid binding specificity of wild-type SadP type PN and PO adhesins, the N-terminal region (aa 31-328) containing the galabiose-binding domain were cloned to pET46 vector for production of recombinant proteins as described in the experimental procedures. For the site-specific mutagenesis, crystal structure and STD-NMR studies, only the galabiose binding domain (aa 125-329) of type PN SadP was used.
characterized. Moreover, whether SadP binds to Gb3 or Gb4 isolated from the pig tissues has not yet been directly shown. The glycolipids were isolated from porcine lung and erythrocytes, in order to control that the glycolipids isolated from the lungs were not contaminants derived from the erythrocytes remaining in the lungs. The total nonacid glycosphingolipid fractions were hydrolyzed with endoglycoceramidase II of Rhodococcus sp., and the oligosaccharides thereby obtained were analyzed by LC-ESI/MS using graphitized carbon columns. This method gives resolution of isomeric oligosaccharides, and the MS 2 analyses gives complete sequence information and allows differentiation of linkage positions by diagnostic cross-ring 0,2 A-type fragment ions (20). The base peak chromatograms from LC-ESI/MS of the oligosaccharides obtained by hydrolysis of the total non-acid glycosphingolipid fractions of porcine erythrocytes and lung were determined ( Fig. 2A). By comparison of the retention times and MS 2 spectra of oligosaccharides obtained from reference glycosphingolipids, the major oligosaccharides obtained from porcine lung were tentatively identified as globotriaosylceramide (Gb3, detected as a [M-H + ]ion at m/z 503) and the blood group H type 2 pentasaccharide (H5-2, detected as a [M-H + ]ion at m/z 852) ( Fig. 2A). The major oligosaccharides from porcine erythrocytes were globotriaosylceramide (Gb3, detected as a [M-H + ]ion at m/z 503), globotetraosylceramide (Gb4, detected as a [M-H + ]ion at m/z 706), and the blood group A type 2 hexasaccharide (A6-2, detected as a [M-H + ]ion at m/z 1055) (Fig. 2B). Thus, globotriaosylceramide was found in both samples, but otherwise the two samples were significantly different. The total non-acid glycosphingolipid fraction from porcine lung was thereafter separated by Iatrobeads chromatography. Aliquots of the fractions obtained were analyzed by thin-layer chromatography, and fractions that were coloured green by anisaldehyde were tested for binding of SadP using the chromatogram binding assay. The fractions were thereafter pooled according to the mobility on thin-layer chromatograms and their SadP binding activity.
The binding of SadP to neutral glycolipids isolated from porcine lung using TLC overlay assay The porcine lung non-acid glycosphingolipid fractions were analysed with type PN SadP . The separations gave two SadP binding fractions migrating in the triglycosylceramide region. One fraction (0.2 mg) migrating as a single band was denoted fraction PL-1 (Fig. 2B, lane 1), and one fraction (0.3 mg) migrating as a double band was denoted fraction PL-2 (

Endoglycoceramidase digestion and LC-ESI /MS of SadP binding glycosphingolipids from porcine lung
The base peak chromatograms from LC-ESI/MS of the oligosaccharides obtained by endoglycoceramidase digestion of fractions PL-1 and PL-2 were very similar (Fig. S3AB). Both had a [M-H + ]ion at m/z 503 which eluted at the same retention time (17.4-18.4 min) as the saccharide obtained from reference globotriaosylceramide (Fig. S3C), while the saccharide from reference isoglobotriaosylceramide eluted at 20.0-20.4 min (Fig. S3D). MS 2 of the ion at m/z 503 of fractions PL-1 and PL-2 gave in both cases two C-type fragment ions (C1 at m/z 179 and C2 at m/z 341) identifying a Hex-Hex-Hex sequence (Fig. S3EF). A 4-substitution of the internal Hex was demonstrated by the 0,2 A2 fragment ion at m/z 281 (20)(21)(22) . Taken together with the similarity to the MS 2 spectrum of reference globotriaosyl saccharide (Fig. S3G), this allowed identification of the saccharides of fractions PL-1 and PL-2 as globotriaosyl saccharides (Gala1-4Galb1-4Glc). LC-ESI/MS of the oligosaccharides obtained by hydrolysis of fraction PL-4 with Rhodococcus endoglycoceramidase II allowed a tentative identification of a globotetraosyl saccharide (GalNAcb1-3Gala1-4Galb1-4Glc).
This conclusion was based on the following spectral features: First, the base peak chromatogram of fraction PL-4 had a [M-H + ]ion at m/z 706 (Fig.  S4A), and MS 2 of this ion (Fig. S4B) gave a C-type fragment ion series (C1 at m/z 220, C2 at m/z 382, and C3 at m/z 544), demonstrating a HexNAc-Hex-Hex-Hex sequence. The 0,2 A3 fragment ion at m/z 484 demonstrated a 4-substituted Hex, while the 0,2 A4 ion at m/z 646, and the 0,2 A4-H2O ion at m/z 628, were derived from cross-ring cleavage of the 4-substituted Glc of the lactose unit at the reducing end. The features of this MS 2 spectrum were very similar to the MS 2 spectrum of the reference globotetraosyl saccharide (

SadP subtype PN and PO binding to purified glycolipids
TLC overlay assay is a powerful method to analyze glycolipid receptor function. The carbohydratebinding specificity of SadP to globo series isoreceptors was determined using purified glycolipids (all TLC binding results are summarized in Table 1). Type PN SadP  bound preferentially to Gb3 and P1 glycolipid, and also weakly to Gb4 and the Forssman glycolipid (Fig. 2C). Type PO SadP(31-328) also bound strongly to Gb3 and P1, i.e. to glycolipids with terminal Gala1-4Gal, whereas there was no binding to Gb4 and the Forssman glycolipid (Fig.  2D). There was no binding to the non-acid glycolipid fraction from the pig intestine, containing the Globo H. The type PN 76 kDa full length SadP and SadP  binding to dilutions of globo series glycolipids was analyzed with TLC overlay assay to the detection limit for the various glycolipids (Fig.  S5A). The full length adhesin bound mainly to P1 glycolipid, with a detection limit at 0.4 µg, whereas the detection limit for the Gb3 glycolipid was 0.8 µg. The detection limit of SadP(31-328) N-terminal domain was 0.2 µg for both Gb3 and P1. Next, the type PN SadP(31-328) binding to 250 ng of glycolipids immobilized in plastic microtiter wells was quantitatively evaluated by solid phase binding assay (Fig. S5B). No binding to the Gb2 glycolipid, galabiosylceramide was detected, whereas there was weak binding to Gala1-4Galb1-O-bis-(SO2-C16H33)2. SadP was found to bind stronger to Gb3 with non-hydroxy ceramide than to Gb3 with hydroxy ceramide. There was also binding to Gb4 and iGb3 isoglobotriaosylceramide. Binding to P1 pentaosylceramide was at the same level as binding to Gb3 with non-hydroxy ceramide. The difference in the binding to iGb3 (Gala1-3Gal-containing glycolipid) in the microtiter well assay compared to TLC overlay assay could be due to the difference in the presentation of the binding epitope and the clustering of the glycolipid moieties.
Gala1-4Gal-dependent endothelial cell binding activity of S. suis SadP type PN S. suis type PN WT strain D282 and the corresponding mutant strain D282DsadP, containing an insertion mutation into SadP gene, were compared for their binding activity to cultured cell line EA.hy926, which is a primary human umbilical vein cell line fused with a thioguanineresistant clone of A549 (human lung adenocarcinoma cell line). In addition, binding of purified fluorescently labelled SadP was analysed for the binding to EA.hy926 cells. The cells were grown into glass coverslips for 48 h and the bacteria grown into early log phase were added to cells and were let to adhere for 1 h. After washing and staining with DiffQuick the bacteria were enumerated. The average adherence of wild-type bacteria was calculated to be 8.7 ± 0.5 bacteria/field, whereas the mean adhesion of 1.3 ± 0.5 of the insertion mutant DsadP was significantly less (unpaired t test > 0.0001) (Fig. 3A). In addition, the binding of the WT strain was inhibited with 10 µg/ml pigeon ovomucoid, which contains N-linked glycans with terminal Gala1-4Gal (23). The inhibitor reduced the adherence of S. suis bacteria to the cells to 0.7 ± 0.1 bacteria/field. For flow cytometry assay, the cells were grown in wells, washed and tested for the binding of FITC labelled type PN SadP . After washing, the cells were harvested with scraping and analyzed with flow cytometer. The flow cytometry results (dot blot and the histogram) are presented in Figure  3BCD. The median SadP binding with and without inhibitor (pigeon ovomucoid, 10 µg/ml) were 56.2 and 14.9 per 10 000 total events (median of untreated cells was 12.8).

Site-specific mutation changes the specificity of type PN recombinant adhesin to type PO
The binding mechanism of type PN and PO SadP was further studied by site-directed mutagenesis. Type PN adhesin SadP-D282(125-329) was mutated and the mutants were analyzed for the binding to galabiose oligosaccharides with AlphaScreen competitive inhibition assay and ITC. Multiple alignment of SadP homologues allowed us to predict amino acids from the galabiosebinding region that differed between type PN and PO (Fig. 1A). Based on that, the site-specific amino acid change N285D and E292Q, deletion ∆244-246 and N285D&∆244-246 mutants were constructed into the type PN SadP(125-328) cloned into the plasmid pET28a. In addition, the 28 amino acid long region H216-D246 of the type PN was replaced by the corresponding region of type PO. The WT SadP and mutants were expressed and purified as described in the methods and analysed with SDS-PAGE (Fig. S6). The galabiose-binding strengths of the above mutant constructs, full length SadP, SadP N-terminal constructs SadP(31-328), SadP(125-328) and as a negative control sitespecific mutant SadP(31-328)W258A, were compared by AlphaScreen asssay (Fig. S7A). The binding of full-length SadP to biotinylated pigeon ovomucoid was weaker compared to the N-terminal domain constructs, which could be due to a larger size of the full length SadP. The size of the protein could increase the distance of the acceptor and donor beads, and hence reduce the transmission of the singlet oxygen during the assay (19). As a negative control for the assay, there was no binding of site-specific mutant W258A to pigeon ovomucoid. Of the mutants in the type PN SadP(125-328) background, mutants N285D and E292Q bound well whereas the binding of mutants ∆244-246 and H216-D246delinsDELFNRFPNKVDSTNNGDGAPFRF FNKE was reduced. The binding of N285D&∆244-246 mutant was abolished. Next the specificities of mutants were analyzed with AlphaScreen assay using phenylureaderivative and Gala1-4Gal ( Fig. S7B-F). All mutants were equally inhibited by the Gala1-4Gal oligosaccharide, but only with the site-specific mutant N285D the inhibitory power of Phenylurea-Gala1-4Gal was weaker compared to Gala1-4Gal, suggesting a specific switch from type PN to PO specificity. Next, the Alphascreen inhibition assay was used to compare the effect of site-specific mutation in N285D to the specificity between Gala1-4Gal (terminal Gb3 structure) and GalNAcb1-3Gala1-4Gal (terminal Gb4 structure) TMSEt glycosides. SadP(125-329) type PN and site-specific mutant N285D were similarly inhibited with Gala1-4Gal, whereas N285D was not inhibited with even higher concentrations of GalNAcb1-3Gala1-4Gal (Fig.  5AB).

Structural comparison of recombinant SadP type PN and its N285D mutant
The site-specific mutant SadP(125-329)N285D was crystallized with five molecules in the crystallographic asymmetric unit (Fig. 7A, Table  S1). All molecules are similar with subtle differences as suggested by the low root mean square deviation (r.m.s.d) between them (0.25-0.31 Å). The structure of the N285D mutant is similar to that of the native SadP (PDB id 5BOB) with r.m.s.d. of 0.26 Å for 199 aligned residues. The structure consists of three a-helixes and ten bstrands (b1-b10) that form a b-sandwich core domain (Fig. 7B). The first b-sheet of the bsandwich is formed by antiparallel b1-b10-b9-b4-b7-b6 strands and the second b-sheet by b2-b3-b8-b5 in antiparallel fashion. Structural superposition with the Gala1-4Gal -Fhc structure revealed small differences (Fig. 7C). The side chain of D285 was found slightly rotated compared to the N285 side chain in the wild type.

Resolution of the interaction mechanisms of SadP to globo series glycolipids and C3'-Phenylurea-Gala1-4Gal
The key amino acids of both types PN and PO are conserved. Y198, Q255, R238 and G233 forms hydrogen bonds with the HO-3', HO-4' and HO-6' of the a'-Gal and the K316 with HO-2 and HO-interaction with the receptor saccharide Gala1-4Galb1-4Glc in solution, STD-NMR was applied using type PN SadP(125-329) and the corresponding mutant N285D (Fig. 7D). STD-NMR results suggest that both the type PN SadP and the mutant N285D interact mainly with the two terminal galactose units of Gb3. With the mutant N285D the interactions of H-4ʹ and H-4ʹʹ are, however, somewhat weaker compared to the interaction of H-6ʹʹ. Additionally, the mutant N285D does not seem to have a clear interaction with the β-glucoside unit at the reducing end of Gb3, whereas the type PN SadP appears to have interactions with H-2, H-3 and H-4. In order to find a plausible hypothesis explaining the binding preferences of the galabiose, GalNAcb1-3Gala1-4Gal and phenylureagalabioside for WT and the N285D mutant, molecular dynamic simulations of their respective complexes were performed. Briefly, the ligand structures were built without the TMSEt aglycon to simplify the simulations and were then placed in the binding site of SadP (pdb id 5BOA) with the Gala1-4Gal disaccharide oriented as in the crystal structure. The terminal GalNAcb1 residue was oriented with the dihedral glycosidic bond angles minimized with the H1 parallel to H4' of Gala1-4Gal. The three complexes with WT were mutated by exchanging N285 for D285. The complexes were unrestrainedly subjected to 100 ns molecular dynamics simulations and all starting conformations converged towards similar stable complex geometries (Fig. 8A-E), except for the phenylurea derivative in complex with the N285D mutant for which light constraints on the backbone atoms of protein secondary structures were applied (Fig. 8F). The a-face of the b-galactoside residue of all three ligands stacked to W258 forming CH-π interactions and the galabiose disaccharide moiety in all simulations retained the interactions and hydrogen bonds observed in the X-ray structure with the Gb3 trisaccharide (PDB id 5boa). The HO-4' of galabiose did not interact directly with N285, but instead via N285 sidechain NH hydrogen bond donation to a network of hydrogen bonds involving 3-5 water molecules and the G233 carbonyl oxygen (Fig. 8A and Movie S1). Mutation of N285 to D285 influenced the directionalities of the hydrogen bonding partners in this network ( Fig.  8D and Movie S2), which may explain the different STD-NMR data with WT and the D285 mutant.
The simulations with the GalNAcb1-3Gala1-4Gal and the phenylurea-galabioside in complex with the WT protein converged to geometries that revealed a highly populated interaction between the ligands and N285. The ring oxygen of the GalNAc residue of the trisaccharide and the urea carbonyl oxygen of the phenylurea derivative accepted hydrogen bonds from the N285 side chain NH2 (Fig. 8B-C and Movie S3 -S4). The GalNAc HO-4 of the trisaccharide formed transient hydrogen bonds to side chain amide groups of E286 and E234, while the phenyl ring of the phenylurea ligand did not form any persistent interactions with any protein residue. The highly populated hydrogen bonds of the trisaccharide GalNAc ring oxygen and the phenylurea carbonyl oxygen to N285 could not form in the N285D mutant ( Fig. 8E-F and Movie S5-S6). Instead, it was disrupted by the formation of a water network between the D285 side chain carboxylate and the GalNAc ring oxygen of the trisaccharide (Fig. 8E), while the phenylurea moiety of the phenylurea galabioside rotated away from the D285 side chain carboxylate (Fig. 8F). This lack of direct hydrogen bond interactions of the GalNAc ring oxygen and urea carbonyl oxygen may at least partly explain the less efficient binding of the trisaccharide and phenyl urea derivative to the N285D mutant.

Discussion
SadP carbohydrate specificity to neutral glycolipids from lungs was studied by TLC overlay assay. The carbohydrate receptors for S. suis adhesion in respiratory tract and lungs have remained elusive. We studied here how SadP type PN and PO recognize glycolipids isolated from pig lung. The major pig lung neutral glycolipids were identified as Gb3, Gb4, the H type 2 pentaosylceramide, the Galili pentaosylceramide and Le b and Le y hexaosylceramides (Fig. 2). In TLC overlay assay, type PN SadP bound to Gb3 and Gb4, whereas type PO SadP bound only to Gb3. Both SadP types showed no binding to Gala1-3Gal-containing Galili pentaosylceramide. (Fig. 2AB, Table 1). Therefore, our findings suggest that globo series glycolipids are the major carbohydrate receptors for S. suis in pig lung. Although fucose-binding lectins have been described both in bacterial and fungal lung pathogens, no fucose-specific binding activity was found in SadP. The globo series glycolipids have been shown to be widely expressed in many tissues of pigs (26); hence, they can potentially serve as SadP receptors for other host organs and brain capillary veins critical for S. suis invasive infection. We show that the binding of type PN S. suis serotype 2 sadP-mutant to endothelial cell line was significantly reduced and binding of both the WT S. suis bacteria and recombinant SadP adhesin were inhibited with the galabiose-containing pigeon ovomucoid. The oligosaccharide inhibition assays performed with both whole bacteria and recombinant adhesin suggest that the reducing end b-Glc of Gb3 is required for optimal binding (16,19). The TLC overlay assay using purified glycolipids showed that SadP bound strongest to the P1 glycolipid ( Fig.  2BC and Fig. S5A). The preferential binding to the P1 glycolipid might also be due to a more optimal presentation of the epitope on the longer core chain, whereas hindered presentation of binding epitope in galabiosyl ceramide might explain, why SadP did not bind to Gb2 in TLC overlay assay. The presence of P1 structure is not well known in pig, however it was found in a1,3GalT knockout pig kidney, since this glycosyltransferase is a direct competitor of a1,4GalT (27). This might be important for surveillance of pathogens in xenotransplantation. The fine specificity of SadP subtypes was further analyzed with ITC and competitive inhibition assay. The ratio of KD values (Gala1-4Gal / GalNAcb1-3Gala1-4Gal) for type PN SadP was 0.4 whereas for type PO it was 0.004, which is in accordance to the corresponding hemagglutination inhibition MIC values observed with bacterial hemagglutination assays (16). Moreover, the IC50 values for Gala1-4Gal and 3'-phenylurea-Gala1-4Gal are in accordance to previous hemagglutination inhibition MIC concentrations for type PN and PO S. suis (25). The mechanism of how the two SadP subtypes differ in their recognition of Gb3, Gb4 and glycomimetic 3-phenylurea-galabiose has not been studied before. The multiple-sequence alignment of SadP showed that specific amino acids G233, E249 and N285 of type PN in galabiose-binding domain are different compared to type PO SadP (Fig. 1A). Of these, only substitution from asparagine to aspartate at position 285 is found in all hemagglutinating type PO adhesins. To analyze the mechanisms how SadP type PN can bind to Gb4 terminal trisaccharide (GalNAcb1-3Gala1-4Gal) and 3'-phenylurea-galabiose, we designed mutations to type PN SadP and analyzed their binding to galabiose. The SadP type PN region from H216 to D246 was swapped to the corresponding region of type PO (to contain substitution from amino acid G to N at position 233). The binding of this mutant to galabiose was reduced compared to the wild type adhesin (Fig. S6A), but was still sufficient to further analyze the oligosaccharide specificity. Deletion ∆244-246 and site-specific mutations N285D and E292Q bound also to galabiose at sufficient level (Fig. S6A) allowing testing with 3'-phenylurea-derivative to compare for type-specificity. Of these mutants, only N285D lost the ability to be efficiently inhibited by 3'phenylurea-derivative ( Fig. S6B-F), suggesting that N285 has an important role in interacting with the phenylurea-group. The amino acid region 216-246 including G233N substitution have possibly no significant role alone in the interaction with HO-3' substituted phenylurea-galabiose. However, G233 could still synergistically interact with phenylurea. Therefore, the N285D mutant was chosen for further studies for the type PN binding mechanism. The WT and N285D mutant were compared for their binding to the soluble Gb4 terminal trisaccharide GalNAcb1-3Gala1-4Gal and to globo series glycolipids Gb3 and Gb4 (Fig. 5 and  6). The results show consistently that N285 plays a specific role in Gb4 binding as demonstrated by competitive inhibition assay, ITC and liposome binding assay. Molecular dynamics calculations were performed with the structures representing type PN SadP(125-329) and its N285D site-specific mutant. Both proteins were analyzed with Gala1-4Gal, GalNAcb1-3Gala1-4Gal and 3'-phenylurea-Gala1-4Gal. In accordance to the co-crystal structure (PDB id. 5boa), the galabiose HO-4', HO6', HO-2 and HO-3 were hydrogen bonded to Y198, G233, R238, Q255 and K316. Hydrophobic interaction with the a-face of the b-galactoside stacked to W258 forming CH-π interactions. The above interactions were conserved in WT and N285 mutant. WT SadP N285 formed hydrogen bonds to GalNAc O-5 ring oxygen of GalNAcb1-3Gala1-4Gal and to urea's carbonyl group (Fig. 8B). The amino acid D285 of site-pecific mutant formed hydrogen bonds with solvent water molecules when simulated with GalNAcb1-3Gala1-4Gal and 3'-phenylurea-Gala1-4Gal. These results together with the AlphaScreen and ITC data give us a plausible explanation for the molecular mechanism of how type PN SadP interact with Gb4 terminal GalNAc and with phenylurea-galabiose derivative. In addition, WT SadP amino acid G233 interacts with the terminal GalNAc of Gb4 via hydrogen bond to N-Acetyl-group or to urea's -NH group (Fig. 8C). SadP does not have any sequence homologs in other galabiose-recognizing proteins, however structurally E. coli PapG and SadP galabiosebinding sites share some similarities. They both contain a b-sheet composed of antiparallel bstrands, which are adjacent to the galabiose-binding domain (PDB; E. coli 1j8r, S. suis 5boa). In SadP, the binding site is formed by the alpha-helix and a non-helix loop, whereas in PapG the galabiose binding site is located in the middle of two nonhelix loops. Other galabiose binding proteins such as P. aeruginosa LecA (4yw6), E. coli verotoxin (1bos) and Lyophyllum decastes fungal lectin (4ndv) show different combining site architectures. As previously suggested, the galabiose-binding adhesin SadP from gram-positive and E. coli P fimbrial adhesins from gram-negative bacteria are an example of convergent evolution towards binding to the galabiose oligosaccharide. The structural similarities thus confirm this hypothesis. Hemagglutination positive strains expressing type PN SadP from pig and zoonotic meningitis strains belong to serotype 2 and clonal group CC1 (6, 17). S. suis hemagglutination type PO strains were mostly found from serotype 4 and were isolated from lungs of strains causing pneumonia (28). The majority of strains that have the gene encoding type PN SadP belongs to a distinct population consisting of systemic isolates, which were identified in a large BRaTP1T consortium functional genomic study (12). S. suis strains that cause human meningitis comprise a separate clade and are thought to have evolved when the pig farming was intensified in 1920s (12). Zoonotic S. suis strains as a specific population have been thought to exponentially spread to different geographical areas due to the selection of pigs that are optimal in terms of productivity. The BRaTP1T consortium study of S. suis genomic signatures for pig and human infections has suggested that the systemic infections are genetically determined and the genome size of highly virulent strains is reduced and enriched with virulence genes. Since the type PN SadP adhesin gene has not been lost in strains causing systemic infections, it is tempting to speculate that the recognition of Gb4 may play a specific role in invasive infections. In a previous study type PN SadP (designated SadP1) present in CC1 clonal group of S. suis strains (17) was found to specifically adhere better to human intestinal cells (Caco-2) than to pig intestinal cells. This could indicate that type PN SadP has specific binding preferences to human intestine, thus promoting zoonosis. In summary, the results of the present study could have implications toward the therapeutic applications targeting SadP and S. suis systemic infections (Fig. 9). The gene encoding SadP type PN has a downstream homolog, which is a pseudogene (has a stop codon). It could be speculated, that the gene encoding SadP type PN is a putatively duplicated gene, which has evolved to bind Gb4 (Fig. 9A). This study shows the molecular mechanism how N285 contributes to the binding of terminal GalNAcb1-saccharide of Gb4 by a specific interaction with GalNAc hexose ring oxygen (Fig. 9BC). The analogous N285-mediated binding mechanisms to Gb4 and glycomimetic phenylurea-galabiose could be utilized in the design of therapeutics against the invasive systemic infections without disturbing the commensal S. suis bacteria colonizing pig upper respiratory tract.

Bacterial strains
The chromosomal DNA of type PN and PO S. suis strains (6) was purified using GelElute bacterial genomic DNA kit according to the manufacturer's instructions. Streptococci were grown in Todd Hewitt broth supplemented with 0.5 % yeast extract or in Columbia agar plates supplemented with 5 % sheep blood in 5 % CO2 at 37°C and E. coli strains were grown in LB medium. Antibiotics used were 30 µg/ml kanamycin (pET28) and 100 µg/ml ampicillin (pET46EkLIC) in E. coli and 500 µg/ml kanamycin for S. suis D282DSadP mutant (6). Escherichia coli NovaBlue (Novagen) was used for the cloning of SadP constructs and strain BL21(DE3) for expression of the adhesin.

Bioinformatics
To compare the distribution of SadP alleles in S. suis from different infections e.g. invasive, respiratory and non-clinical samples, the presence of either PN or PO allele was detected from 374 previously published S. suis isolates (12) using SRST2 software version 0.2.0 with default settings (29). The SadP genes from S. suis P1/7 (AM946016.1) and 6407 (NZ_CP008921.1) were used as reference sequences for PN and PO, respectively. The distribution of PN and PO alleles in the S. suis phylogeny was investigated by creating a core-genome maximum-likelihood tree using Roary version 3.12.0 (30) and RAxML version 8.2.8 (31) from the 374 isolates.

Expression and purification of SadP proteins
The bacteria were grown at 30°C, 250 rpm, to an OD600 of 0.5, and the protein expression was induced with 0.2 mM IPTG for 3.5 hours. Bacteria were harvested by centrifugation with 3000 x g, at +4°C and stored at -84°C. The recombinant protein was purified with Ni-NTA affinity chromatography. Briefly, bacteria were lysed with 0.4 mg/ml hen egg lysozyme (Sigma) in 50 mM sodium phosphate buffer pH 8.0, containing 0.5 M NaCl, EDTA-free protease inhibitor cocktail (Pierce), 20 mM imidazole, 20 µg/ml deoxyribonuclease and 1 mM MgCl2 on ice for 30 mins. The lysate was sonicated to further homogenize the cell debris and was centrifuged at 20000 x g, +4°C for 30 min. The filtered lysate was purified with Ni-NTA affinity chromatography using HiPrep FF column connected to Äktaprime plus, GE Healthcare at +25°C. Further purification was done with a gel filtration HiLoad 16/60 Superdex 200 column using Tris-HCl, pH 7.5, 0.15 M NaCl as running buffer. The purity of the recombinant proteins was analyzed with SDS-PAGE chromatography.

Crystallization
SadP N285D mutant (15 mg/ml) was crystallized with the hanging-drop vapor-diffusion method. The well condition contained 1.3-1.5 M sodium citrate tribasic, 0.1 M sodium cacodylate, pH 6.5. Equal volumes of well solution and protein (2 µl+2 µl) were mixed and the crystals grew at 16 °C, initially as haystacks of very thin needles. Single crystals suitable for data collection were subsequently grown by using the seeding technique. A small part of the haystack was crushed with a metallic needle in the crystallization drop and 0.2 µL were diluted in 20 µl of mother liquor to create a seeding stock. Pre-equilibrated (1-d old) crystallization drops were seeded with 0.1 µl of the seeding stock. Single crystals started to grow after ~3 days.

Data collection and crystal structure determination and refinement
Diffraction data were collected on the BioMAX beamline at MAX IV (Lund, Sweden) from a single crystal under cryogenic temperatures (100K). The crystal diffracted to 1.85 Å but the resolution was later truncated during processing to 2.05 Å owing to completeness. The crystal was found to belong to the C2 space group. Data processing was carried out with automated procedures in EDNA (32) and scaling was done with AIMLESS (33). Molecular replacement with PHASER (34) was carried out to obtain initial phases using the native structure as a template. Five molecules in the crystallographic asymmetric unit were located (Matthews coefficient 2.67 Å 3 /Da corresponding to ~53.9% solvent content). Initial building of the structure was carried out using the automated building procedure in BUCCANEER (35) using the ccp4i2 interface (36). After the initial building that produced an almost complete structure, the refinement continued in PHENIX v.1.17.1-3660 (37) using simulated annealing at 1000K with maximum likelihood as target function. The first round of simulated annealing resulted in Rwork/Rfree of 0.227/0.269. Inclusion of waters and rounds of manual rebuilding using COOT (38) alternated with PHENIX refinement resulted in the final structure with Rwork/Rfree of 0.174/0.221 (Table S1). A glycerol molecule was found bound in each of the A, C, and D subunits. Two glycerol molecules were located in the binding site of subunit B.

Isolation of SadP binding glycosphingolipids from porcine lung
Non-acid glycosphingolipids were isolated from porcine lung as described (39). Briefly, the lung tissue was lyophilized, and then extracted in two steps in a Soxhlet apparatus with chloroform and methanol (2:1 and 1:9, by volume, respectively). The material obtained was subjected to mild alkaline hydrolysis and dialysis, followed by separation on a silicic acid column. Acid and nonacid glycosphingolipid fractions were obtained by chromatography on a DEAE-cellulose column. In order to separate the non-acid glycolipids from alkali-stable phospholipids, this fraction was acetylated and separated on a second silicic acid column, followed by deacetylation and dialysis. Final purifications were done by chromatographies on DEAE-cellulose and silicic acid columns. The total non-acid glycosphingolipid fraction (9 mg) from porcine lung was first separated on an Iatrobeads (latrobeads 6RS-8060; Iatron Laboratories, Tokyo) column (1.0 g) and eluted with increasing volumes of methanol in chloroform. Aliquots of the fractions obtained were analyzed by thin-layer chromatography. Fractions that were coloured green by anisaldehyde were tested for binding of SadP using the chromatogram binding assay. SadP bound to a fraction containing glycosphingolipids migrating in the tri-and tetraglycosylceramide regions on thin-layer chromatograms. This fraction (1.0 mg) was further separated on an Iatrobeads column (1.0 g), eluted with chloroform/methanol/water 65:25:4 (by volume), 30 x 0.5 ml, followed by chloroform/methanol/water 65:25:4, 10 mi. The SadP binding triglycosylceramide was eluted in fractions 12-17, and these fractions were pooled into one fraction migrating as a single band (0.2 mg, denoted fraction PL-1), and fraction migrating as a double band (0.3 mg, denoted fraction PL-2) on thin-layer chromatograms. In addition, a SadP tetraglycosylceramide was present in fractions 19-31, and pooling of these fractions gave 0.4 mg (denoted fraction PL-4).

Reference glycosphingolipids
Total acid and non-acid glycosphingolipid fractions were isolated as described (39), and the individual glycosphingolipids were obtained by repeated chromatography on silicic acid columns, and by HPLC, and identified by mass spectrometry (20,40) and 1H NMR spectroscopy (41).

Thin-layer chromatography
Aluminum-or glass-backed silica gel 60 high performance thin-layer chromatography plates (Merck, Darmstadt, Germany) were used for thin-layer chromatography, and chromatographed with chloroform/methanol/water (60:35:8 by volume) as solvent system. The different glycosphingolipids were applied to the plates in quantities of 4 µg of pure glycosphingolipids, and 20-40 µg of glycosphingolipid mixtures. Chemical detection was done with anisaldehyde (42).

Radiolabeling
Aliquots of 100 µg of the different SadP protein preparations were labeled with 125 I by the Iodogen method according to the manufacturer's instructions (Pierce/Thermo Scientific), giving approximately 2000 cpm/µg protein.

Microtiter well assay
Binding of radiolabeled SadP to glycosphingolipids in microtiter wells was performed as described (43). In short, 250 µg of pure glycosphingolipids in methanol were applied to microtiter wells (Falcon 3911, Becton Dickinson Labware, Oxnard, CA). When the solvent had evaporated, the wells were blocked for 2 h at room temperature with 200 µl of BSA/PBS. Thereafter, the wells were incubated for 4 h at room temperature with 50 µl of 125 I-labeled SadP (2 x 10 3 cpm/µl) diluted in BSA/PBS. After washing 6 times with PBS, the wells were cut out and the radioactivity was counted in a gamma counter.

LC-ESI/MS of native glycosphingolipids
Native glycosphingolipids were analyzed by LC-ESI/MS as described (44). Glycosphingolipids were dissolved in methanol:acetonitrile in proportion 75:25 (by volume) and separated on a 200x0.150 mm column, packed in-house with 5 µM polyamine II particles (YMC Europe GmbH, Dinslaken, Germany). An autosampler, HTC-PAL (CTC Analytics AG, Zwingen, Switzerland) equipped with a cheminert valve (0.25 mm bore) and a 2 µl loop, was used for sample injection. An Agilent 1100 binary pump (Agilent technologies, Palo Alto, CA) delivered a flow of 250 µl/min, which was split down in an 1/16" microvolume-T (0.15 mm bore) (Vici AG International, Schenkon, Switzerland) by a 50 cm x 50 µm i.d. fused silica capillary before the injector of the autosampler, allowing approximately 2-3 µl/min through the column. Samples were eluted with an aqueous gradient (A:100% acetonitrile to B: 10 mM ammonium bicarbonate). The gradient (0-50% B) was eluted for 40 min, followed by a wash step with 100% B, and equilibration of the column for 20 min. The samples were analyzed in negative ion mode on a LTQ linear quadropole ion trap mass spectrometer (Thermo Electron, San José, CA), with an IonMax standard ESI source equipped with a stainless steel needle kept at -3.5 kV. Compressed air was used as nebulizer gas. The heated capillary was kept at 270 °C, and the capillary voltage was -50 kV. Full scan (m/z 500-1800, two microscans, maximum 100 ms, target value of 30,000) was performed, followed by data-dependent MS 2 scans (two microscans, maximun 100 ms, target value of 30.000) with normalized collision energy of 35%, isolation window of 2.5 units, activation q= 0.25 and activation time 30 ms). The threshold for MS 2 was set to 500 counts. Data acquisition and processing were conducted with Xcalibur software version 2.0.7 (Thermo Fisher Scientific). Manual assignment of glycosphingolipid sequences was done with the assistance of the Glycoworkbench tool (Version 2.1), and by comparison of retention times and MS 2 spectra of reference glycosphingolipids.
Briefly, 50 µg of glycosphingolipids were resuspended in 100 µl 0.05 M sodium acetate buffer, pH 5.0, containing 120 µg sodium cholate, and sonicated briefly. Thereafter, 1 mU of enzyme was added, and the mixture was incubated at 37°C for 48 h. The reaction was stopped by addition of chloroform/methanol/water to the final proportions 8:4:3 (by volume). The oligosaccharide-containing upper phase thus obtained was separated from detergent on a Sep-Pak QMA cartridge (Waters). The eluant containing the oligosaccharides was dried under nitrogen and under vacuum. The glycosphingolipid-derived oligosaccharides were resuspended in 50 µl of water and analyzed by LC-ESI/MS as described (20). The oligosaccharides were separated on a column (200 x 0.180 mm) packed in-house with 5 µm porous graphite particles (Hypercarb, Thermo-Hypersil, Runcorn, UK). An autosampler, HTC-PAL (CTC Analytics AG, Zwingen, Switzerland) equipped with a cheminert valve (0.25 mm bore) and a 2 µl loop, was used for sample injection. An Agilent 1100 binary pump (Agilent technologies, Palo Alto, CA) delivered a flow of 250 µl/min, which was split down in an 1/16" microvolume-T (0.15 mm bore) (Vici AG International, Schenkon, Switzerland) by a 50 cm x 50 µm i.d. fused silica capillary before the injector of the autosampler, allowing approximately 2-3 µl/min through the column. The oligosaccharides (3 µl) were injected on to the column and eluted with an acetonitrile gradient (A: 10 mM ammonium bicarbonate; B: 10 mM ammonium bicarbonate in 80% acetonitrile). The gradient (0-45% B) was eluted for 46 min, followed by a wash step with 100% B, and equilibration of the column for 24 min. A 30 cm x 50 µm i.d. fused silica capillary was used as transfer line to the ion source. The oligosaccharides were analyzed in negative ion mode on an LTQ linear quadrupole ion trap mass spectrometer (Thermo Electron, San José, CA). The IonMax standard ESI source on the LTQ mass spectrometer was equipped with a stainless-steel needle kept at -3.5 kV. Compressed air was used as nebulizer gas. The heated capillary was kept at 270 °C, and the capillary voltage was -50 kV. Full-scan (m/z 380-2 000, 2 microscans, maximum 100 ms, target value of 30 000) was performed, followed by data dependent MS 2 scans of the three most abundant ions in each scan (2 microscans, maximum 100 ms, target value of 10 000). The threshold for MS 2 was set to 500 counts. Normalized collision energy was 35%, and an isolation window of 3 u, an activation q = 0.25, and an activation time of 30 ms, was used. Data acquisition and processing were conducted with Xcalibur software (Version 2.0.7). Manual assignment of glycan sequences was done on the basis of knowledge of mammalian biosynthetic pathways, with the assistance of the Glycoworkbench tool (Version 2.1), and by comparison of retention times and MS 2 spectra of oligosaccharides from reference glycosphingolipids (20).

S. suis cell binding assay
EA.hy926 cells were maintained and grown in DMEM, 10 % FCS medium at 37°C, 5 % CO2. For bacterial binding assay the cells were detached with Trypsin-EDTA, counted and 15 000 cells/well added into round glass coverslips in 24 well-plates. Cells were grown for 24 -48 h to subconfluency. S. suis D282 WT type PN strain and D282-DsadP were grown in THY overnight at 37°C, 5 % CO2. Bacteria were diluted 1/20 into prewarmed 37°C THY and were grown to OD550 of 0.2 and were diluted 1/100 into prewarmed DMEM, 10 % FCS without antibiotics. 500 µl of bacterial dilution was pipetted into the wells (MOI 100:1) and the plate was centrifuged with 800 x g, 15 min at 20°C. The plate was incubated at 37°C for 1 h. The wells were washed 4 x 1000 µl of PBS. The cells were stained with DiffQuick kit, washed and the coverslips were mounted on glass slides with Permount. The bacterial binding was quantitated by microscopy with 100 x objective with immersion oil. The bound bacteria were enumerated by counting two wells for each sample. The results were expressed as average of bacteria / optical field. For flow cytometry, N-terminal domain of type PN SadP(31-328) was labelled with FITC (Sigma-Aldrich) in 0.2 M borate buffer, pH 8.0, 0.16 M NaCl (50:1 molar ratio of FITC/SadP, 83 nmol of SadP and 1.7 µmol of FITC) and the labelling was stopped with Tris-buffer. The labelled SadP was purified from the free FITC with PD-10 (GE Healthcare) desalting column. The EA.hy926 cells were seeded into 6 well culture dishes (1 x 10 5 cells/well) and were grown for 48 h. The cells were washed with DMEM and were incubated with 400 ng/ml of labelled SadP with or without 10 µg/ml pigeon ovomucoid. After binding, the wells were washed with phosphate Buffered Saline (PBS, 0.15 M NaCl, 2.7 mM KCl, 8.1 mM Na2HPO4, 1.5 mM KH2PO4). and were fixed with 2 % (w/v) paraformaldehyde in PBS for 15 min. The cells were washed twice with PBS and the cells were scraped of the wells and suspended into the PBS. 10 000 cells were analyzed for SadP binding with FACSCalibur.

Amplified luminescent proximity homogeneous assay (AlphaScreen)
The AlphaScreen assay was optimized as described before (19). Briefly, the assay was performed using AlphaScreen Streptavidin Donor beads and NiNTA Acceptor beads (PerkinElmer). The molar concentrations of galabiose-containing biotinylated ovomucoid (receptor) and adhesin were optimized by finding the hook point for the interaction by setting up a matrix of proteins in 96 well AlphaPlates (Perkin Elmer) in 10 mM Tris-HCl, pH 7.5, 0.15 M NaCl, 0.2% BSA, 0.05 % Tween 20 0.2% (TBST-0.2%BSA). For the optimized assay dilutions of 5 µl of the His-tagged adhesin and 5 µl of biotinylated ovomucoid were pipetted into the wells and the plates were centrifuged for 1 min, 1000 x g. The plates were incubated at +25°C for 1 hr. This was followed by addition of Ni-NTAbeads (20 µg/ml) and incubation for 1.5 hrs. Finally, streptavidin donor beads were added (20 µg/ml) and the mixtures were incubated for 30 mins and were then measured with the Ensight multimode reader (Perkin Elmer) using excitation wavelength of 680 nm with donor beads and measurement of emission wavelength of 615 nm from AlphaLISA anti-HIS acceptor beads. For inhibition assays the oligosaccharides, whose structures have been described before (25) or depicted in Figure 4 for TMSEt-glycosides of GalNAcb1-3Gala1-4Gal and 3Gala1-4Gal, were diluted into TBST-0.2%BSA buffer and were mixed with the His-tagged adhesin and biotinylated ovomucoid. The binding was measured as described above. The binding inhibition data was fitted using Prism with settings of log(inhibitor) vs. response slope (four parameters).

Preparation of liposomes and SadP binding assay
Vesicles were prepared as described before (45). Briefly, POPC (1-palmitoyl-2-oleoyl-glycero-3phosphocholine, Avanti Polar lipids), cholesterol and glycolipids were dissolved and mixed into the chloroform : methanol (2:1, vol/vol) to contain 30 % cholesterol and 2 mM glycolipids. The solvent was evaporated and the lipid film was suspended for 30 min at 60 °C in 10 mM Tris-Cl, 140 mM NaCl buffer. The lipid-buffer suspension was briefly vortexed followed by the extrusion procedure (Avanti mini extruder using 0.1 µm polycarbonate membranes filter, (Avanti Polar Lipids, Alabaster, AL, USA)) to form large unilamellar vesicles. PVDF membrane was wetted (wetting in 100 % MeOH, washing 4 x 5 min MilliQ water) and was overlaid onto wet Whatman filter paper. Liposomes were serially diluted 1/10 into Tris-buffered saline and 2 µl of liposomes containing 2 mM, 0.2 mM and 0.02 mM of glycolipid was pipetted onto the membrane and the membrane was not allowed to dry during the pipetting. The membrane was saturated with 2 % BSA (w/v) in TBS for 1 h at 20°C and was incubated with 10 µg/ml recombinant SadP for 2 h at 20°C. The membrane was washed 4 x 5 min with TBS and the bound SadP was detected with 1:10000 dilution of anti-His primary antibody (Sigma) and 1:20000 dilution of HRP-labelled secondary rabbit anti-mouse antibody (DakoCytomation). The membrane was incubated with ECL substrate (WesternBright Quantum, Labtech) and the SadP binding was imaged with Fuji LAS-4000.

STD-NMR
STD-NMR measurements were carried out on a Bruker AVANCE III HD spectrometer (Bruker BioSpin GmbH, Rheinstetten, Germany) operating at 600.13 MHz, equipped with a nitrogen-cooled triple-channel TCI inverse CryoProbe™. The pulse sequence used for the experiments was the stddiffesgp provided by Bruker. A 1x PBS buffer with a pH of 7.45 with 10% D2O, containing 50 µM protein and 2500 µM Gb3 trisaccharide was used for the experiments. The on-resonance pulse was set to the methyl groups at 0.9 ppm and the offresonance pulse was set at 20 ppm. A saturation time of 2 s was used and the experiments were carried out at 313 K to somewhat improve the weak signals, likely resulting from slow koff. A blank experiment with only Gb3 and no protein was done to rule out any self-STD.

Molecular dynamics simulations
Molecular dynamics simulations were performed with the OPLS3 force field in Desmond Maestro-Desmond Interoperability Tools, Schrödinger, New York, NY, 2017) using default settings except for the length of the simulation. Simulations were performed with periodic boundary conditions using orthorhombic simulation boxes with SPC water model. Counter ions were used to neutralize each other. Starting conformations of ligands in complex with SadP (pdb id 5BOA) were built by manually placing ligand galabiose residues in an identical position to that in the crystal structure (pdb id 5BOA). Complexes with the mutants were generated by exchanging N285 for D285. The complexes were then subjected to 100 ns molecular dynamics simulations. The complex of the phenylurea derivative with the N285D mutant was not stable during the 100 ns simulations. Therefore, light constraints (1 kcal·mol -1 ·Å -1 ) on the protein secondary structures were applied. Molecular images were generated using PyMOL v2.30 (Schrodinger LLC).