Molecular and structural basis for Lewis glycan recognition by a cancer-targeting antibody.

Immunotherapy has been successful in treating many tumour types. The development of additional tumour-antigen binding monoclonal antibodies (mAbs) will help expand the range of immunotherapeutic targets. Lewis histo-blood group and related glycans are overexpressed on many carcinomas, including those of the colon, lung, breast, prostate and ovary, and can therefore be selectively targeted by mAbs. Here we examine the molecular and structural basis for recognition of extended Lea and Lexcontaining glycans by a chimeric mAb. Both the murine (FG88.2) IgG3 and a chimeric (ch88.2) IgG1 mAb variants showed reactivity to colorectal cancer cells leading to significantly reduced cell viability. We determined the X-ray structure of the unliganded ch88.2 fragment antigen-binding (Fab) containing two Fabs in the unit cell. A combination of molecular docking, glycan grafting and molecular dynamics simulations predicts two distinct subsites for recognition of Lea and Lex trisaccharides. While light chain residues were exclusively used for Lea binding, recognition of Lex involved both light and heavy chain residues. An extended groove is predicted to accommodate the Lea-Lex hexasaccharide with adjoining subsites for each trisaccharide. The molecular and structural details of the ch88.2 mAb presented here provide insight into its cross-reactivity for various Lea and Lexcontaining glycans. Furthermore, the predicted interactions with extended epitopes likely explains the selectivity of this antibody for targeting Lewis-positive tumours.


Introduction
shallow binding pocket [18,19]. Although each of these anti-Lewis antibodies have been well characterised, with some tested in clinical trials [12], none are currently approved as therapeutic mAbs.
Recently, a mAb known as FG88.2, raised against plasma membrane extracts from colorectal cancer cells, was shown to target Lewis glycans found on a wide range of colorectal, pancreatic, gastric, nonsmall cell lung and ovarian tumours. When screened against glycan arrays, the FG88.2 mAb bound to Le a -containing glycans and extended epitopes including di-Le a , Le a -Le x , and Le c Le x . Importantly the FG88.2 mAb displayed minimal binding to many other mammalian glycans or normal tissues indicating its capacity for Lewis-positive tumour selectivity. In addition, the antibody demonstrated antibodydependent cellular cytotoxicity (ADCC), complement-dependent cytotoxicity (CDC) and direct (caspaseindependent) tumour cell killing. It was also shown to internalise, colocalise with lysosomes and deliver saporin that killed cells with subnanomolar potency. In vivo studies revealed potent anti-tumour efficacy in a metastatic colorectal xenograft tumour model in mice, leading to significant long-term survival. This work indicates the potential of FG88 as a therapeutic mAb for the treatment of multiple solid tumours [20].
Here we characterise the potential molecular basis for recognition of Lewis glycans (Le a and Le x ) by the chimeric antibody ch88.2. Both mouse (FG88.2) and chimeric (ch88.2) variants showed reactivity to colorectal cancer cells, with mAb binding significantly reducing cell viability. We determined the X-ray structure of ch88.2 Fab, with both Fabs in the unit cell displaying similar structures with some differences within the CDR loops, particularly in the H chain. A mixture of molecular docking, glycan grafting and molecular dynamics simulations predicted a binding region with distinct sites for Le a and Le x glycan epitopes. The Le a -Le x hexasaccharide, which was previously identified by glycan array as the top binding motif [20], was found to interact via both its Le a and Le x components. Taken together, the structural and computational analysis suggest that the FG88.2/ch88.2 mAbs are capable of binding multiple extended Lewis glycans for effective targeting of Lewis-positive tumours.
Downloaded from https://portlandpress.com/biochemj/article-pdf/doi/10.1042/BCJ20200454/890280/bcj-2020-0454.pdf by guest on 17 August 2020 Biochemical Journal. This is an Accepted Manuscript. You are encouraged to use the Version of Record that, when published, will replace this version. The most up-to-date-version is available at https://doi.org/10.1042/BCJ20200454 Ethidium homodimer (EthD-1) and 2 μM Calcein-AM in PBS for 45 minutes at room temperature.
Coverslips were mounted onto slides and examined by fluorescence microscopy for red (dead) and green (live) fluorescence. Samples were examined using a Leica DM2500 epifluorescence microscope with a DFC310 digital camera, and images were captured using LAS software (V4.1; Leica Microsystems).
Flow cytometry for antibody binding and antibody-mediated killing. To assess antibody binding, cells were seeded at 1 x 10 5 cells/well and treated with primary antibody followed by secondary antibody (as per immunofluorescent detection method above). Monoclonal human IgG1 (F598) [22] and mouse IgG3 (49-31.1) [23] antibodies against irrelevant targets were available in house and were used as isotype controls. Samples were fixed in 2% paraformaldehyde before being analysed in the FACS Canto (10,000 cells per sample). To assess antibody killing, cells were seeded at 1 x 10 5 cells/well and treated with 30μg/mL of antibody in complete RPMI media for 24 hours at 37°C with 5% CO 2 . As a control for total cell lysis (positive control), cancer cells were treated with 0.1% Triton X-100 in PBS for 15 minutes. Cells were imaged in a 1 μg/mL propidium iodide solution on the FACS Canto (10,000 cells per sample). The resultant data was analysed by FlowJo software where cell populations were gated by cell size and complexity, following which AF488 or propidium iodide positive populations (FITC or PE emission >10 3 ) were selected. One-way ANOVA tests were used to assess statistical significance in cell death (nonviability) compared to the negative controls. Individual data points are shown along with the mean (n = 2 for ch88 antibody binding from 1 experiment or n = 4 for all other samples from 2 experiments), and standard deviation was determined for antibody killing (n = 4).

Scanning Electron Microscopy (SEM).
Silicon wafer surfaces were coated with poly-D-lysine at 1 mg/mL and cells were seeded overnight in complete RPMI media before being treated with 30 μg/mL of antibody for 24 hours at 37°C with 5% CO 2 . For cellular imaging, samples were affixed using 3% glutaraldehyde, dehydrated and coated with a thin film of gold. Scanning electron micrographs were obtained using a FEI Verios 460L field-emission scanning electron microscope (FE-SEM) (FEI Company, Oregon, USA) operated with an accelerating voltage of 2-5 kV. The resultant images were analysed using the Gwyddion and Image J software suites.
Fab Production. A Pierce Fab Preparation Kit (Thermo Fisher Scientific) was used to produce Fab from IgG. Briefly, 0.5 mL of an 8 mg/mL IgG sample was digested with Papain-agarose for 6 hours at 37°C. Fc and residual intact IgG was separated from Fab using protein A affinity chromatography. Coomassie stained SDS-PAGE (precast 4-15% Bis-Tris Mini Gels and MES running buffer, BioRad) was used to examine the purity of the Fab compared to the Fc and intact IgG samples under non-reducing and reducing (β-mercaptoethanol) conditions. Fab was quantitated by absorbance at 280 nm (Nanodrop) assuming a mass extinction (E, derived concentrations in mg/mL) value of 1.0 (1.37 for intact IgG). DLS Downloaded from https://portlandpress.com/biochemj/article-pdf/doi/10.1042/BCJ20200454/890280/bcj-2020-0454.pdf by guest on 17 August 2020 was used to determine protein size and polydispersity using a Zetasizer Nano ZS instrument (Malvern Instruments). Cuvettes containing 100 µL of protein sample were measured (five replicates) at 25°C, and time-correlated light scattering data were analysed as size-distributions by scattered intensities (histograms of particle diameter (nm) versus % intensity). The Z-average hydrodynamic diameters (D H , in nm) and overall polydispersity of samples were estimated by the cumulants method.  Data collection and structure determination. Diffraction data were collected at the Australian Synchrotron using the MX2 beamline, by the ultrafine φ-slicing data-collection method using an EIGER X 16M detector and the qeGUI graphical user interface (oscillation range 180°, Δφ = 0.1°). Diffraction data were auto processed on the MX2 beamline using automated indexing with xdsme (using XDS and Pointless) and AIMLESS [24,25]. Data processing and hkl file conversions were implemented in the XDS and the CCP4 program package [26,27]. All modelling and crystallographic refinements were performed using Phenix, COOT and REFMAC, and the CCP4 program package [28][29][30]. Figures were generated using Discovery Studio (Dassault Systèmes BIOVIA, USA). X-ray data was processed in the P1 space group and the structure was determined by molecular replacement using Protein Data Bank (PDB) code 4X80.

Crystallisation
Several rounds of fitting of the atomic model to electron density and crystallographic refinement were conducted. Data collection and refinement statistics are reported in Table S1.  Glycan Grafting. The Gly-Spec tool was used to assess the agreement between glycan microarray data and docked poses [33,34]. Gly-Spec grafts complete glycan structures onto a minimal binding determinant/receptor complex and can predict whether the grafted structure will bind experimentally based on steric clashes between the grafted ligand and the protein receptor. Gly-Spec can then compare to a set of known binders from a CFG glycan microarray and report the overall agreement between the grafted structures and the array data. To assess the agreement of docked poses with CFG array data, we reduced each docked pose to its core Le a determinant and used this structure as input to Gly-Spec.
Positive binders were determined from CFG glycan array data (http://www.functionalglycomics.org, last accessed 01/07/20, ID = primscreen_5844) using a method based on Median Absolute Deviation (MAD) [35] with a threshold z-score of z > 3.5 for assigning positive binders ( Figure S5). Poses that had high agreement with CFG array data (as determined with Gly-Spec) were used as starting points for MD simulations.
Molecular dynamics simulations. MD simulations were run using GROMACS 2018.2 [36,37]. Proteins were parameterised using the AMBER99SB-ILDN forcefield [38]. Carbohydrate topologies were generated using the Glycam06j forcefield via GLYCAM-Web [http://glycam.org], converted to GROMACS format using ACPYPE [39,40] and combined with the protein topology to form the complete proteincarbohydrate system. Initial coordinates for the protein-carbohydrate system were based on molecular dynamics simulations using the ch88.2 Fab structure (Fv portion only). The protein-carbohydrate system was placed in a rhombic dodecahedral box with ≥10 Å distance between the molecule and the edge of the box. The system was solvated using the TIP3P water model, then ionised and neutralised with Na + and Clto a concentration of 0.15 M. Energy minimization was performed using the steepest descent algorithm (5000 steps), follow by equilibration at constant volume and temperature (NVT ensemble) for 100 ps, with each replicate initialised with random velocities sampled from a Maxwell distribution at 300 K. This was followed by equilibration with constant pressure and temperature (NPT ensemble) for 300 ps. Pressure coupling was achieved via the Parrinello-Rahman barometer with a time constant of 2 ps and a reference pressure of 1 bar. Temperature coupling was achieved using the modified Berendsen thermostat, with separate temperature coupling groups for ligand/protein and water/ions respectively.
A time constant of 0.1 ps was used for temperature coupling, with a reference temperature of 300 K.
Equations of motion were integrated using the leap-frog integrator with a time-step of 2 fs. Hydrogen bonds were constrained during all steps using the LINCS constraint algorithm [41]. Long range Downloaded from https://portlandpress.com/biochemj/article-pdf/doi/10.1042/BCJ20200454/890280/bcj-2020-0454.pdf by guest on 17 August 2020 electrostatics were calculated using the Particle Mesh Ewald (PME) method with cubic interpolation.
Neighbour searching was performed using the Verlet cut-off scheme with a distance cut-off of 1.0 nm for both Coulomb and van der Waals interactions. Extended simulations were run using the Spartan high-performance computer system (University of Melbourne).
Analysis of MD simulations was performed using a combination of GROMACS 2018.2 and the MDTraj Python package (v1.9.1). Ligand root-mean-square deviation (RMSD) was calculated relative to the protein backbone carbon atoms (i.e. protein backbone carbon atoms were used to remove rotational and translational movement across frames, and ligand RMSD calculated for these transformed trajectories). Hydrogen bonds were identified using MDTraj, with the Baker-Hubbard criteria used to identify hydrogen bonds (θ > 120° and distance from H to acceptor atom < 2.5 Å) [42]. Hydrogen bonding matrices were generated using frames in which ligand RMSD < 0.5 nm. Side-chain dihedral angles were calculated with MDTraj using frames in which ligand RMSD < 0.5 nm. Plots were constructed using the Matplotlib 1.5 and Seaborn 0.9.0 Python libraries. Visualisation of docked poses was performed using USCF ChimeraX (v0.9). LigPlot+ version 2.2 was used to generate the ligand interaction diagram [43].

Activity and membrane effects of mAb 88.2 against colorectal cancer cells. Previous work by Chua et
al. demonstrated the ability of FG88.2 to target colorectal cancer cells [20]. In order to confirm activity of the chimeric mAb ch88.2, binding and killing experiments were conducted using colorectal cancer cell line COLO205. Both ch88.2 and FG88.2 showed strong binding to COLO205 cells by immunofluorescence staining ( Figure 1A) and flow cytometry, with mean binding determined to be 99.9% for ch88.2 and 89.9% for FG88.2 and minimal binding in both the negative (0.02% and 0.05%) and isotype (0.09% and 2.3%) controls ( Figure 1C). Cell viability was determined after 24 hours of antibody treatment, with mAb treatment shown to cause cell death as measured by Live/Dead staining. The average non-viability increased from 0% in the negative control to 21% after ch88.2 treatment and 44% after FG88.2 treatment, compared to 100% in the positive control ( Figure 1B). Cell viability was also measured by flow cytometry through propidium iodide uptake, with non-viability significantly increasing from 4.1 ± 1.0% in the negative control (and 4.9 ± 0.4% in the isotype control) to 16.6 ± 0.9% after ch88. To characterise the structural basis for recognition of Lewis glycans by ch88.2, crystallisation of the Fab:glycan complex was attempted. However, electron density was not observed for the Le x (described here) or the Le a trisaccharides, and crystallisation with the Le a -Le x hexasaccharide was not attempted.
Instead, the crystal structure of ch88.2 Fab in the free form was determined to 2.3 Å (R work /R free = 0.213/0.293) from a triclinic P1 crystal, with two Fabs in the asymmetric unit (designated here as Fab1 and Fab2). X-ray diffraction data collection and crystallographic refinement statistics are presented in Table S1. The electron density maps allowed fitting of the L and H polypeptide chains for each Fab, except for CH1 residues 141-146 (sequential numbering), which are distant from the antigen-binding site and can be disordered in Fab crystal structures [44,45]. There is clear electron density corresponding to most binding site residues, although the maps are not as well defined around  Figure S3), were shown to be similar with most of the minor variations occurring in the CDR loops surrounding the binding site ( Figure 3C). End-on surface views of each Fab illustrate that the main difference between Fab1 ( Figure 3A) and Fab2 ( Figure 3B) occurs in CDR H3. This is further evident in the CDR view of each Fab. In the overlay, there is little to no variation in L1, L2, L3 and H1 CDR loops, but some variation in H2 and larger differences in H3, which can be seen in the mainchain as well as some individual side-chains ( Figure 3D-F). While these differences may be due to crystal packing, with Fab1 packed against Fab2, they also demonstrate the overall plasticity of the ch88.2 Fab CDR regions.

Molecular docking of Le a -Le x reveals a range of potential binding poses.
To elucidate the structural basis for recognition of Le a and Le x by the ch88.2 mAb, molecular dynamics simulations and molecular docking were used, combined with a re-analysis of previously published glycan microarray data for FG88.2 [20]. FG88.2 and ch88.2 contain the same variable chain sequences and were shown to bind to the same cell lines and tissues [20]. Thus, recognition of top glycan binders, identified with FG88.2 mAb, were explored using the structure of the chimeric variant. Microarray data revealed that FG88.2 is capable of binding a large number of Lewis-containing sugars. The top binders were a mixture of Le aand/or Le x -containing glycans, with Le a being found in most of the strongest binders, while the top binder was a Le a -Le x hexasaccharide ( Figure 4A-C). Initial attempts to dock small Le a or Le x trisaccharides Downloaded from https://portlandpress.com/biochemj/article-pdf/doi/10.1042/BCJ20200454/890280/bcj-2020-0454.pdf by guest on 17 August 2020 proved difficult, with docked poses tending to bury the reducing end of the carbohydrate ligand. This is problematic, as the reducing end is typically attached to either another monosaccharide unit or a glycolipid/glycoprotein on the cell surface.
Therefore, to determine the likely binding mode for Le a , we performed molecular docking on the highest binding Le a -Le x hexasaccharide. We used this extended structure for docking to enforce some of the structural/steric constraints that would not be captured by docking Le a alone. Vina-Carb was used to dock the Le a -Le x ligand to 3 different structures. Namely, the two Fv regions from the crystallographically determined Fabs, and a third Fv structure obtained by molecular dynamics simulations to remove potential crystal packing artefacts. As the crystal structures were unliganded, docking was performed with flexible side-chains for residues around the putative binding region. The best 9 docked poses were extracted for each structure, giving a total of 27 potential docked poses. We observed very little agreement between docked poses, with a wide range of potential binding modes identified ( Figure S4).

Integration of glycan microarray data highlights a likely binding configuration.
To narrow down the set of likely poses, we used the Gly-Spec tool to compute the overall agreement between each pose and experimental FG88.2 glycan microarray data [33,34]. Gly-Spec grafts whole glycan structures onto a minimal binding determinant in a glycan-protein complex, and predicts binding based on steric clashes between grafted glycans and the protein structure. This predicted binding can then be compared to experimental glycan microarray data. As Le a is the proposed minimal binding determinant, we grafted glycans onto the Le a portion of docked poses.
After grafting onto the Le a portion of all 27 docked poses, only a single pose was deemed to have an acceptable level of agreement with CFG glycan array data, with an overall agreement of 87% to CFG array data ( Figure 4D & E). For this pose, one glycan (glycan ID=385) was predicted to be a non-binder by Gly-Spec, despite being within the positive binding set by glycan microarray. This glycan also contains a separate Le x moiety, suggesting that the Le a moiety may not be the minimal binding determinant for this one glycan structure. Indeed, for this structure, the Le a moiety has an alpha-2 linked fucose on the Gal residue, which is likely to impact binding via the Le a portion of the glycan ( Figure 4A, Table 1). The other conflicting result (glycan ID=533) was predicted to be a binder by Gly-Spec, but not by glycan microarray.
However, this glycan is just below the threshold for positive binders by microarray data, suggesting that this disagreement is the result of uncertainty in the threshold for calling a positive binder, rather than a true disagreement by Gly-Spec. Overall, this suggests a strong level of agreement between this docked pose and the glycan microarray data.  Table 1). In general, the sulfated glycans appear to have lower binding strength when compared to the top binding glycans, indicating that binding is permissive of the sulfate group. For example, the sulfate group on glycan ID=28 and glycan ID=492 is positioned away from the binding site so as not to interfere with the suggested binding mode. The ch88.2 mAb also appears capable of binding to structures like glycan ID=24 and glycan ID=291, which are closely related to Le x containing Galβ1-4(Fucα1-3)Glc instead of Galβ1-4(Fucα1-3)GlcNAc. Both of these glycans are also sulfated in one or two positions. In addition, ch88.2 is predicted to bind to glycan ID=430. This glycan contains Fucα1-3GlcNAc and Galβ1-4GlcNAc, which are both found in Le x but does not contain a complete Lewis glycan structure. This demonstrates broad recognition of ch88.2 for Lewis and Lewis-like glycans.

Stability of the proposed ch88.2 Le a -Le x complex.
Given that ligand docking was performed on a fixed structure, with only some sidechain flexibility surrounding the CDR regions, we performed a small number of initial MD runs to identify the most stable configuration for the ch88.2 Fv:Le a -Le x complex.
Using the single docked pose identified by Gly-Spec, we performed 3 x 100 ns MD runs. From these initial simulations, we identified the most stable configuration for the antibody-glycan complex, where Le a -Le x remained in position throughout the run. This structure was then run in 10 x 100 ns MD runs to assess overall stability.
The Le a -Le x hexasaccharide was stable in the binding site, remaining bound for 9 out of 10 runs ( Figure   5A). Some fluctuations in overall ligand RMSD were observed across runs, and this was typically due to movement in the Le a portion of the ligand ( Figure 5B), with the Le x end remaining relatively fixed ( Figure   5C). Interestingly, the reverse behaviour was observed when running MD simulations of just the Le a and Le x portions of the ligand; Le a was stable in 6 out of 10 runs, with minimal RMSD fluctuation ( Figure 5D), whereas Le x displayed greater RMSD variation and only remained bound in 3 out of 10 runs ( Figure 5E).
Overall, the apparent stability of the ligands in MD simulations is in agreement with CFG glycan microarray data, in which Le a -Le x is the strongest binder, followed by Le a -containing glycans then Le xcontaining glycans ( Figure 4A). the proposed binding site, while the H chain CDRs are too far away to form any direct hydrogen bonds with the glycan. Key residues include Tyr-62H, whose backbone is involved in two highly populated hydrogen bonds to NAG1 (Le x ), one to the nitrogen on the N-acetyl group and the other to O1. Ser-94L forms one hydrogen bond to O2 on GAL3 (Le x ), while Glu-64H forms one bond to either O2, O3 or O4 on FUC2 (Le x ). Binding also involves the N-terminal residue (Asp-1L), which forms one hydrogen bond to either O5 or O6 on NAG4 (Le a ) as well as one bond to O4 of FUC2 (Le x ) ( Figure 6D, Figure 7A, Figure   S6). Notably, the majority of hydrogen bonds are to the Le x portion of the structure. However, the Le a portion appears to be supported by CH-π interactions between the alpha face of GAL6 and Trp-92L. In portion of the hexasaccharide. The most highly populated bonds are between Tyr-62H to the nitrogen on the N-acetyl group and to O1 on NAG1, Ser-94L to O2 on GAL3, Glu-64H to either O2 or O4 on FUC2, and Asp-1L to O4 of FUC2 or O4 of GAL3 ( Figure 7C & Figure 8). Recently, there has been an increasing trend towards targeted therapeutics for the treatment of cancer to increase specificity and reduce off-target effects. Lewis system carbohydrates (Le a , Le b , Le x , Le y ) are often aberrantly expressed on tumours derived from tissues that are normally negative or have minimal expression on restricted tissues [12]. This includes carcinomas of the colon, lung, breast, prostate and ovary that can therefore be selectively targeted by anti-Lewis mAbs. A number of Lewis-binding mAbs have been structurally characterised, including five co-crystal structures and three unbound structures of anti-Lewis antibodies [14]. Although each of these antibodies have been well characterised and some have progressed through clinical trials [12], none have been approved for therapeutic use. In addition, these structures describe anti-Le x and anti-Le y antibodies, but no crystal structures of anti-Le a antibodies have been determined. In this study we examine the nature of a chimeric-human IgG1 mAb that can target both Le a -and Le x -containing glycans.

Potential interactions between ch88.2 and Le
Previous work by Chua et al. demonstrated the ability of the murine antibody FG88.2 to target colorectal cancer cells [20]. Here we also conducted further additional studies to assess the killing ability of the chimeric variant ch88.2 IgG sample, which was used in our structural studies. FG88.2 was previously shown to have direct tumour cytotoxicity via a mechanism distinct from complementmediated or cellular-mediated effects [20]. Murine IgG3 mAbs have previously been shown to be particularly effective at agglutination [46]. Murine IgG3 are also able to self-associate via their constant regions, forming large oligomeric networks [47]. As such, we sought to determine if the chimeric ch88.2, which is built on a human IgG1 backbone, could also display direct killing activity against Lewisexpressing colorectal cancer cells. Treatment with ch88.2 IgG1 caused a reduction in cell viability, although not to the levels observed by the murine FG88.2 IgG3 (Figure 1). This was further reflected during SEM imaging, where a percentage of cells appeared damaged with membrane irregularities including pores and large blebs ( Figure 2). Interestingly, by DLS, intact ch88.2 (also used for structural studies) was shown to be highly aggregated in solution (around 90%) ( Figure S2). This level of aggregation may have led to increased clustering of Lewis epitopes, resulting in modest direct cell killing effects.
Several other mAbs have been developed against type II Lewis glycans, including BR96 (anti-Le y ), hu3S193 (anti-Le y ), 291-2G3-A (anti-Le x ) and 54-5C10-A (anti-Le x ). Co-crystal structures of both BR96 and hu3S193 mAbs with Le y show very similar mechanisms for carbohydrate recognition, with almost identical binding sites and carbohydrate orientation for these two structures. In these structures, Le y is accommodated in a large but relatively deep binding pocket, with antibody-glycan interactions dominated by the heavy chain [15][16][17]48]. The mAbs 291-2G3-A and 54-5C10-A were both developed Downloaded from https://portlandpress.com/biochemj/article-pdf/doi/10.1042/BCJ20200454/890280/bcj-2020-0454.pdf by guest on 17 August 2020 against Schistosoma parasites, although only 291-2G3-A has been resolved in complex with Le x . This structure involves a shallower binding pocket that is centrally located at the VL-VH interface [18,19].
Notably, no crystal structures of mAbs in complex with type I Lewis glycans (Le a or Le b ) have been determined, so comparisons can only be made to type II glycan structures. The binding mechanism proposed here for the larger Le a -Le x epitope involves an extended binding groove, composed of two adjoining shallow binding pockets. The binding pocket for Le a is formed entirely by the light chain, whereas the pocket for Le x involves significant contributions from both heavy and light chain residues.
This extended binding groove with distinct sites for Le a and Le x may explain the high affinity for Le a -Le x as opposed to di-Le a or other Le a -containing glycans as observed by glycan microarray. Additionally, this may also explain the relatively high affinity for Le c -Le x , despite Le c and Le x being weaker binding determinants than Le a . The fucose residue of Le a (FUC5) is relatively unengaged in direct binding interactions, only making contact with Glu-27L in a small number of simulation frames. This suggests that FUC5 is not required for binding. As this residue is not present in the Le c -Le x pentasaccharide, it would likely bind in a similar way, in agreement with glycan array data.
An interesting feature that appears common to many anti-carbohydrate antibodies is the involvement of multiple aromatic residues in carbohydrate binding, in particular tyrosine (Tyr), tryptophan (Trp) and phenylalanine (Phe). Aromatic residues can participate in binding via direct hydrogen bonding with the glycan, stacking interactions between aromatic rings and the hydrophobic face of carbohydrate rings (involving CH-π interactions) or other hydrophobic interactions. These types of interactions have been noted in all previously determined co-crystal structures of anti-Lewis antibodies [15,16,18]. The proposed binding mode for ch88.2 is no exception, with interactions with aromatic residues playing a key role in binding to the Le a portion of the structure. The predicted high stability of Le a as compared to Le x , may be explained by potential CH-π interactions with aromatic residues in the L chain via GAL6 despite Le x being involved in a greater number of hydrogen bonding interactions. In general, CH-π interactions differ significantly between monosaccharides, with beta-D-Gal having a relatively high propensity for forming CH-π interactions via its alpha face [49]. For ch88.2, these interactions likely occur between the alpha face of GAL6 and Trp-92L. Additionally, other hydrophobic interactions may play a key role in ligand stability. It is thought that desolvation of aromatic residues and carbohydrate CH groups contributes significantly to overall antibody-glycan interactions [50]. In the binding mode predicted here, Trp-92L and Phe-32L are angled such that GAL6 is inserted end-on into a hydrophobic pocket formed by these two residues.
In summary, it is the balance of hydrogen bonding and hydrophobic interactions that engage the extended glycan structure. The combination of molecular docking, glycan grafting and molecular dynamics have provided crucial insights into the binding of Le a -and Le x -containing glycans to ch88.2, Downloaded from https://portlandpress.com/biochemj/article-pdf/doi/10.1042/BCJ20200454/890280/bcj-2020-0454.pdf by guest on 17 August 2020 highlighting the cooperative interplay between Le a and Le x in the extended Le a -Le x structure. The Le a portion is largely surrounded by antibody residues and appears to dominate the interaction with ch88.2.
However, the Le x portion is partially exposed to solvent indicating that a larger glycan may be accommodated within the binding site. On the surface of a cancer cell the glycan targets are likely larger than the hexasaccharide reported here and presented as glycolipids or glycoproteins. Thus, the mode of recognition of Lewis glycans identified here is compatible with binding of ch88.2 to larger antigenic targets on tumour cells. Detailed characterisation of the interactions identified here could assist in structure-guided manipulation of the antibody to enhance its selectivity for tumour-related Lewis antigens and drive the development of further therapeutic antibody candidates.    is identified by Glycan array ID (CFG array version 5.1). Antibody was tested at 50 μg/mL. Two Lewis motifs within the same glycan are indicated by darker shades of blue. RFU is shown on a linear scale. A full list of all binding glycans is provided in Table 1. B & C) The top binder in the ch88.2 glycan microarray is a Le a -Le x hexasaccharide shown as a cartoon (in SNFG nomenclature) and depicted as sticks (coloured by atom type). D & E) Agreement between docked poses and glycan microarray data for all 27 docked poses examined. Binder agreement is defined as the percentage of experimental binders which are also predicted to bind following glycan grafting, and conversely for non-binder agreement. Each pose is represented in grey, with the top pose shown in red.