Structural insights into antigen recognition of an anti-β-(1,6)-β-(1,3)-D-glucan antibody

Schizophyllan (SCH) is a high molecular weight homopolysaccharide composed of a β-(1,3)-D-glucan main chain with branching β-(1,6)-bound D-glucose residues. It forms triple helices that are highly stable towards heat and extreme pH, which provides SCH with interesting properties for industrial and medical applications. The recombinant anti-SCH antibody JoJ48C11 recognizes SCH and related β-(1,6)-branched β-(1,3)-D-glucans, but details governing its specificity are not known. Here, we fill this gap by determining crystal structures of the antigen binding fragment (Fab) of JoJ48C11 in the apo form and in complex with the unbranched β-(1,3)-D-glucose hexamer laminarihexaose 3.0 and 2.4 Å resolution, respectively. Together with docking studies, this allowed construction of a JoJ48C11/triple-helical SCH complex, leading to the identification of eight amino acid residues of JoJ48C11 (Tyr27H, His35H, Trp47H, Trp100H, Asp105H; Asp49L, Lys52L, Trp90L) that contribute to the recognition of glucose units from all three chains of the SCH triple helix. The importance of these amino acids was confirmed by mutagenesis and ELISA-based analysis. Our work provides an explanation for the specific recognition of triple-helical β-(1,6)-branched β-(1,3)-D-glucans by JoJ48C11 and provides another structure example for anti-carbohydrate antibodies.

Here, we have determined the crystal structure of the antigen binding fragment (Fab) of anti-SCH antibody JoJ48C11, both in the apo form and in complex with the oligosaccharide laminarihexaose. These structures allow modelling of an antibody/SCH complex which suggests that all three chains of the helical glucan engage in interactions with the paratope of JoJ48C11. Importantly, these interactions involve several protruding β-(1,6)-bound glucose residues of SCH, explaining the specificity of the antibody. The model was validated by introducing point mutations into JoJ48C11, corroborating its correctness.

Structure analysis of JoJ48C11 Fab and the JoJ48C11 Fab/laminarihexaose complex. In order
to obtain detailed information about the binding of SCH and other β-(1,3)-branched β-(1,6)-D-glucans by JoJ48C11, we resorted to an X-ray crystallographic structure analysis. For this, we produced recombinant Fab fragments of JoJ48C11 by adding human constant C H 1 and C kappa domains to the respective variable regions of the initial JoJ48C11 scFv fragment and crystallized the purified protein to collect diffraction data.
Initial phasing of the JoJ48C11 Fab fragment was performed by molecular replacement with the structure of a chimeric antibody Fab fragment against CD25 (PDB entry 1MIM) as a search model 27 . The resulting structure was refined to 3.0 Å resolution. The JoJ48C11 Fab fragment adopts the expected heterodimeric structure of a heavy chain fragment and a light chain both consisting of a constant (C) and a variable (V) immunoglobulin (Ig) domain. The Ig domains show the typical immunoglobulin fold characterized by two anti-parallel β-sheets packed against each other (Fig. 1a) 28 . Not surprisingly, backbone atoms can easily be superposed onto other Fab structures, and even several structures with almost identical C L , C H and V L domains including their CDRs can be found (Table 1). However, with respect to previously published Fab fragments, the V H domain of JoJ48C11 Fab is slightly different, especially in the CDR3 region (Fig. 1b,c).
For understanding of the molecular determinants that contribute to the recognition of SCH by JoJ48C11, we aimed at obtaining the crystal structure of a complex with SCH but failed despite extensive attempts. Therefore, we resorted to structure determination of a complex with the less complex oligosaccharide laminarihexaose, which is a β-(1,3)-linked unbranched glucose hexamer. Electron density for two glucose moieties was observed when crystals were soaked with high concentration of the ligand (50 mM laminarihexaose) for one week. Interestingly, laminarihexaose does not show binding to JoJ48C11 in an in vitro affinity assay 23 , indicating that this is a good example for the sensitivity of protein crystallography to detect lower affinity 29 . The JoJ48C11 Fab/ laminarihexaose structure was refined to 2.4 Å resolution. In the complex, Asp105 H of the heavy chain CDR3 region binds the O4-and O6-atoms of the non-reducing end of laminarihexaose in a bidentate fashion ( Fig. 2a and Supplementary Fig. S1). To confirm the observed interaction of Asp105 H , we generated an alanine mutant and produced it as scFv-Fc. Binding of JoJ48C11-Asp105Ala H scFv-Fc to SCH was completely abrogated in a titration ELISA, which corroborates the importance of Asp105 H and the significance of the JoJ48C11 Fab/laminarihexaose complex crystal structure.
The apo structure of JoJ48C11 Fab and the complex with laminarihexaose were used as template for modeling of JoJ48C11 Fab in complex with the triple helical SCH as described below.
Modelling the interaction of JoJ48C11 with triple helical SCH. To further refine insight into the JoJ48C11 Fab/SCH interaction, we used SwissDock to generate models of a complex with the repeating unit (RU) of SCH and the apo structure of JoJ48C11 Fab. 256 models in 44 clusters were generated and then inspected to select the most likely models for further analysis. Two similar docking poses positioning the RU inside the typical antigen binding area of antibodies were identified. They proposed that Trp33 H , His35 H , Trp47 H , Tyr50 H , Glu59 H , Trp100 H , Trp90 L and Asn93 L were interacting residues. Subsequent mutagenesis studies proved His35 H , Trp47 H , Trp100 H and Trp90 L to be necessary for binding SCH, since their respective alanine mutants showed significant decrease of binding strength with EC50 values of 1.1 nM to 10.9 nM, which are higher than 0.6 nM (3 times the EC50 value of wild type JoJ48C11 scFv-Fc). It is important to note that the main interaction sites predicted by docking are different from the position identified in the crystal structure with laminarihexaose, suggesting that JoJ48C11 may contain two separated recognition motives for β-(1-6)-linked glucose moieties (Fig. 2b).
With the information from the crystal complex and the docking poses with the RU of SCH, we aimed at generating a model of JoJ48C11 in complex with triple helical SCH. Interestingly, the binding sites for the terminal glucose moiety of laminarihexaose observed in the crystal structure and the second binding site identified by docking calculation are approx. 18 Å apart, which corresponds very well to the helix pitch distance of the SCH chains in the triple-helical form (Figs 2b and 3a) 30 . We therefore created a model of JoJ48C11 Fab in complex with triple-helical SCH by superimposing a SCH triple helix structure from the literature onto the β-(1,6)-linked glucose unit of the docked SCH RU ( Supplementary Fig. S2) 31 . To corroborate this model, we examined additional residues that are located close to the predicted binding sites of the triple-helical SCH, namely Tyr27 H , Ser31 H , Tyr57 H , Asp49 L , Lys52 L and Ser55 L , by generating respective alanine exchange mutants in the scFv-Fc format. This revealed Tyr27 H , Asp49 L and Lys52 L to be significantly involved in binding of SCH, validating our structural model of the JoJ48C11 Fab/triple-helical SCH interaction ( Fig. 3b and Table 2).
The model suggests that five glucose units are involved in binding to JoJ48C11: two β-(1,6)-linked moieties of SCH chain 1 fit into the two binding cavities found in the crystal structure with laminarihexaose and in the docking studies with the RU of SCH, one β-(1,6)-linked and one β-(1,3)-linked glucose group of SCH chain 2 interact with V H and V L in the area between the two binding cavities, and one β-(1-3)-linked moiety of SCH chain 3 interacts with V L near the docking site of the RU of SCH. The distribution of torsion angles values at the β-(1,3)-glycosidic bonds of triple-helical SCH in the modeled complex with JoJ48C11 after energy minimization is very similar to values observed in crystal structures, indicating the high geometrical quality of the model ( Supplementary Fig. S3). According to our binding studies, the antibody itself contributes eight amino acid residues (Tyr27 H , His35 H , Trp47 H , Trp100 H , Asp105 H ; Asp49 L , Lys52 L , Trp90 L ) to this interaction, acting mainly SCIentIfIC RepoRtS | (2018) 8:13652 | DOI:10.1038/s41598-018-31961-x through hydrogen bonds or π-effects (Fig. 3c,d). Taken together, this provides an explanation for JoJ48C11's specificity towards SCH triple helices.

Discussion
Basic characteristics for the paratope structure of carbohydrate binding antibodies were early established by studies of anti-dextran binding antibodies. They predicted that the respective antibodies consist of a cavity-shaped paratope for accommodation of terminal saccharide residues or of a groove structure for up to seven internal saccharide residues of a linear polysaccharide chain 32,33 . To date, the structures of a few anti-carbohydrate antibodies have been determined by X-Ray crystallography. They indeed show the predicted cavity or groove structures of their paratope and some even consist of a combined structure of both types 34 . To name one interesting example, the monoclonal antibody F22-4 that is directed against the O-Antigen (lipopolysaccharide) from serotype 2a Shigella flexneri consist of a groove-shaped paratope with two small cavities integrated. Co-crystallization with a synthetic fragment (decasaccharide) of the O-antigen clearly showed the binding-orientation of the helically bowed fragment along the groove with two branches of single glucose residues fitting into the small cavities 35 . For JoJ48C11 we stated that the combination of the triple helical structure with the protruding β-(1,6)-linked D-glucose is a crucial factor for binding 23 . Therefore, we expected a groove-shaped paratope for interchain recognition of SCH with at least one cavity-shaped binding site to accommodate a β-(1,6)-linked D-glucose.
Here, we discovered via protein crystallography and in silico docking that the paratope structure of JoJ48C11 consists of two cavity-shaped binding sites that are connected by a shallow groove. These cavities can accommodate single glucose residues and are approx. 18 Å apart, a distance that corresponds strikingly well to the helical pitch in structural models of triple-helical SCH 30 . This enabled us to generate a model of the JoJ48C11 Fab/SCH complex, which is defined by the protruding β-(1,6)-linked D-glucose residues positioned in the two cavities. Experimental validation by mutagenesis indicated that eight amino acids of JoJ48C11 predicted to be involved in binding of SCH, namely Tyr27 H , His35 H , Trp47 H , Trp100 H , Asp105 H , Asp49 L , Lys52 L and Trp90 L , are indeed required, confirming the correctness of the model.
The complete abrogation of binding by mutation of Asp105 H reveals the critical function of this residue. As part of one of the two binding cavities, it was identified to engage in two hydrogen bonds with the glucose unit at the non-reducing end of laminarihexaose in the complex crystal structure. Such bidentate interactions between acidic residues and saccharide moieties are an important motif in protein/carbohydrate complexes 36 and the model with SCH suggests that Asp105 H contributes a similar interaction to a β-(1,6)-linked moiety from SCH chain 1.
In silico docking suggested that His35 H , Trp47 H , Trp100 H and Trp90 L establish walls around a second cavity for the binding of a second β-(1,6)-linked glucose moiety of SCH chain 1. Histidine and tryptophan are often found in carbohydrate binding sites, where they seem to interact through aromatic contacts 36 . Such interactions are known to determine the specificity of carbohydrate binding proteins, but their contribution to the stability of such complexes is rather modest 37,38 , which is in line with the observation that mutation of these residues led to a smaller affinity decrease than in the case of the Asp105Ala H substitution. However, the function of Trp47 H and  Interestingly, in order to avoid clashes with SCH, Trp100 H has to adopt a different sidechain rotamer than is observed the apo structure. The required conformation is found in two out of four copies of the laminarihexaose complex, however, which may be indicative of crosstalk between the two glucose binding sites ( Supplementary Fig. S4). The potential involvement of Tyr27 H , Asp49 L and Lys52 L in SCH recognition was only discovered from the JoJ48C11/triple-helical SCH complex model. Tyr27 L seems to interact with a β-(1,6)-bound glucose of SCH chain 2, whereas Asp49 L and Lys52 L engage in hydrogen bonds with one β-(1,3)-bound glucose unit of the same SCH chain. Of these residues, Asp49 L is especially important as can be concluded from the large decrease in binding affinity observed after its mutation to alanine (Fig. 3b).  In summary, we have determined the structure of the anti-SCH antibody JoJ48C11 and obtained further insight into SCH binding by combining crystal structure analysis of a complex with a molecule that does not show binding in ELISA experiments, in silico docking of a SCH RU and modeling of the JoJ48C11 Fab/triple-helical SCH complex by hand. The model is consistent with experimental data and validates our previous hypothesis about the specificity of JoJ48C11 towards SCH and other similar β-(1,6)-branched β-(1,3)-D-glucans, namely that the β-(1,6)-bound D-glucose residues in conjunction with the triple-helical structure constitute the binding epitope for the antibody JoJ48C11 23 . Our work provides first insight into the structural principles that govern the binding of triple-helical SCH to a SCH-specific antibody and provides another example of an anti-carbohydrate antibody structure in complex with its antigen.

Materials and Methods
Preparation of JoJ48C11 as Fab for crystallography. JoJ48C11 was produced as a Fab in HEK239-6E cells (National Research Council, Biotechnological Research Institute). Towards this, the DNA encoding V H and V L of JoJ48C11 scFv were cloned into the respective vectors pCSE2.5-Fab-k-His-XP and pCSE2.5-Fab-h-His-XP 23 . A mixture containing 12.5 µg of both vectors was used for PEI-mediated transfection of 25 mL culture (about 1.5 × 10 6 cells mL −1 ). Transfection and production were carried out as described previously 23,41 . After production, the Fab was isolated by His-tag affinity purification. For optimization of the crystallization, macroseeding was carried out with drops consisting of 500 nL protein solution, 600 nL precipitant (0.1 M HEPES pH 7.5, 0.1 M sodium acetate, 0.2 M lithium sulfate, 25% w/v PEG 4000) and 100 nL macroseed crystal from 0.1 M Tris pH 8.5, 0.2 M lithium sulfate and 30% w/v PEG 4000, set up with an OryxNano crystallization robot (Douglas Instruments Ltd). Crystals were observed after 2 months and were soaked by adding 50 mM laminarihexaose to the drop for 1 week and subsequent flash freezing in liquid nitrogen without additional cryoprotectant. Diffraction data were collected from 3600 0.05°-oscillation images to a maximum resolution of 2.4 Å. The complex crystal belonged to space group P2 1 with cell parameters of a = 81.73 Å, b = 131.50 Å, c = 91.03 Å and β = 91.51°. The asymmetric unit contained 4 copies of the Fab fragment, with a calculated Matthews parameter (V M ) of 2.39 Å 3 /Da (solvent content 48.5%). Diffraction data were processed and scaled using autoPROC with XDS, pointless and aimless [42][43][44][45][46] . Statistics for the data collection are described in Table 3.
Structure determination and refinement. The structure of the anti-SCH JoJ48C11 Fab was solved by the automated molecular replacement pipeline BALBES (https://www.ccp4.ac.uk/ccp4online/), using the structure of a Fab fragment of a chimeric antibody against CD25 (PDB entry 1MIM) as a search model 27 . The complex structure with laminarihexaose was solved by molecular replacement with PHASER as included in the PHENIX software suite using the apo structure of JoJ48C11 Fab as a search model 47,48 . An initial atomic model was manually improved using COOT and refined with PHENIX and REFMAC by standard protocols 49,50 . Refinement statistics for the JoJ48C11 Fab fragment structures are summarized in Table 3. Coordinates and structure factors have been deposited in the Protein Data Bank (PDB) with accession codes 6EV1 (JoJ48C11 Fab) and 6EV2 (JoJ48C11 Fab:laminarihexaose complex) 51 . Figures were prepared using PYMOL 52 . Molecular modeling. Molecular docking calculations were performed using the SwissDock web service (http://swissdock.vital-it.ch/) with default parameters and settings 53,54 . The variable region of the apo structure of JoJ48C11 Fab (residue 1 to 116 of chain A (V H ) and 1 to 106 of chain B (V L )) was selected as the target. Water molecules were removed from the structure. For the ligand, a three-dimensional model of the SCH RU was obtained from the PubChem Compound Database (CID:24777) and was converted into MOL2 file format using YASARA View 55,56 . Results were visualized with the UCSF Chimera package and unique docking poses were selected by visual inspection 57 .
To derive a model for the interaction of JoJ48C11 with triple-helical SCH, the selected docking pose of the SCH RU was superposed onto chain A and B of the laminarihexaose complex. Residues Leu99 to Asp105 of chain A and Lys44 to Val57 as well as Gln89 to Asn93 of chain B were replaced with the corresponding residues from chain A and B of the apo structure due to the fact that Trp100 of chain A and Asp49 and Trp90 of chain B adopted different sidechain rotamers than in the apo structure that had been used for the docking calculations with SCH RU. The model of triple-helical SCH was generated using a structure of the triple-helical curdlan as a template (http://polysac3db.cermav.cnrs.fr/polysacdb/curdlan-II/curdlan2_3chains_expanded.pdb in PolySac3DB), which was modified into triple-helical SCH by manually adding β-(1,6)-bound glucose moieties homogenously at every third glucose moiety of each chain 58,59 . Subsequently, triple-helical SCH was fitted onto the JoJ48C11 Fab structure by visual superposition of two β-(1,6)-bound glucose with the two glucose moieties from the complex structure with laminarihexaose and the docking model with SCH RU. Subsequently, the two β-(1,6)-bound glucose of triple-helical SCH used for the superposition were replaced with the two glucose moieties from the laminarihexaose complex and the SCH RU docking pose. Finally, energy minimization was performed with the YASARA Energy Minimization Server (http://www.yasara.org/minimizationserver.htm) 60 . The complete docking and modeling procedure is summarized in Supplementary Fig. S1.
Generation and preparation of JoJ48C11 mutants as scFv-Fc. The    and thirteen bases in the 3′-direction. The mutated scFv DNA sequence was cloned into pCSE2.6-mIgG2c-Fc-XP vector and used for transfection of HEK cells (EXPI293-F ™ , Thermo Fisher Scientific) 41 . The cells were cultivated as described above. For transfection, 5 µg of the vector DNA was added to 250 µL culture medium and mixed with 250 µL medium containing 25 µg polyethylenimine. 5 mL of cells (ca. 1.5 × 10 6 cells mL −1 ) were transferred into 40 mL screw cap tubes (Sarstedt), the DNA-PEI mixture was added and the culture was incubated at 37 °C, 190 rpm as well as 5% CO 2 in the atmosphere (HERAcell ™ , Thermo Fisher Scientific and Celltron, Infors). 48 h after transfection the culture was fed with 5 mL HEK TF medium (Xell AG) supplemented by 8 mmol L −1 L-glutamine and 1 mL of HEK FS medium (Xell AG). The scFv-Fc mutants were isolated by protein A affinity purification. 9.5 ml of cell free supernatant was incubated with 500 µL of protein A purification resin (MabSelect ™ SuRe ™ , GE Healthcare). The loaded resin was washed with 20 volumes PBS and the scFv-Fc was eluted in 2 mL of 100 mM citrate puffer (pH 3). For pH neutralization, 220 µL of 1 M Tris was added immediately after elution. Afterwards the buffer was exchanged for PBS by centrifugation with Zeba ™ desalt spin columns (Thermo Fisher Scientific). The concentrations of the antibody preparations were calculated by their absorbance at 280 nm (NanoDrop 2000, Thermo Fisher Scientific).
Titration ELISA. Titration ELISA (enzyme-linked immunosorbent assay) was performed on Carbo-BIND ™ multiwell plates (Corning) loaded with proteinase K-treated SCH (SCH-PK) as described before 23 . Briefly, serial dilutions of the scFv-Fc antibodies ranging from 0.03 pM to 950 nM were prepared in blocking solution (2% w/v milk powder and 0.05% w/v Tween ® 20 in PBS) and added to the blocked multiwell plates (100 µL per well). After 1 hour incubation, the plates were washed 3 times with PBST (0.05% w/v Tween ® 20 in PBS). For the titration of the point mutants in the scFv-Fc format, the plates were incubated with peroxidase-conjugated goat anti-murine Fc antibody (A0168, Sigma-Aldrich, 1:40,000 in blocking solution, 100 µL per well). For Detection of JoJ48C11 in the Fab format, the plates were first incubated with a mouse anti-His-tag antibody (DIA-900, Dianova, 100 µL per well, 40 ng ml −1 ) before incubation with the peroxidase-conjugated antibody. The ELISA was developed by adding 100 µL TMB solution (20 volumes 30 mM potassium citrate, 0.5 M citric acid, pH 4.1 and 1 volume 10 mM 3,3′,5,5′-tetramethylbenzidine, 10% v/v acetone, 90% v/v ethanol, 0.3% v/v H 2 O 2 )) per well. The color reaction was stopped after 10 minutes by addition of 100 µL of 0.5 M H 2 SO 4 per well. The resulting absorption was measured at 450 nm (reference wavelength 620 nm). Each mutant was investigated in three independent experiments.
The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.