Paratope alanine scanning
Towards our aim of understanding how Burosumab interacts with FGF23, we first investigated the binding interface of Burosumab to select functional residues that may contribute to the binding. From amino sequence analysis, Burosumab possesses a hydrophilic CDR surface and a negatively charged HCDR3 with the net charge of -3 (Supplementary Fig. S2). The structure of Burosumab or Burosumab-FGFG23 complex has not been solved. In order to determine the critical paratope residues (hotspots) responsible for binding, we built a homology model of Burosumab Fv using Rosetta Antibody 3 module from ROSIE server in which Burosumab amino acid sequence was sourced from Drug Bank database (Supplementary Table S1). From the model, the fifty-six CDR residues that formed the paratope structure were mutated to alanine using site-directed mutagenesis, expressed as Burosumab alanine mutants in HEK293T cells, and purified using Protein A chromatography.
We next evaluated the effect of each Burosumab mutant on the FGF23-Burosumab interaction using ELISA (Supplementary Table S2 and S3). On the heavy chain, we found two alanine mutations: Y33A (HCDR1) and D95A (HCDR3) that caused significant reduction in binding (> 100-fold decrease in Kd), whereas another mutation D98A caused a mild reduction in binding (2.62-fold). Alanine mutations from HCDR2 did not significantly affect the binding. On the light chain, D50A (LCDR2) knocked out the binding (> 100-fold decrease in Kd), while F96A (LCDR3) decreased the binding by 2.32-fold (Fig. 1a,b). Collectively, alanine scanning identified five paratope residues critical to the binding of which three residues: VH: Y33, VH: D95, and VL: D50 were considered as hotspots. We also observed that three out of five residues are negatively charged residues, suggesting that negatively charged residues in CDRs may play a dominant role in the binding and imply a likelihood for interacting with positively charged residues on the binding interface of FGF23. Of note, these critical residues were clustered at the VH – VL interface, suggesting that the concave surface formed by VH and VL domains may also play a key role in the recognition of FGF23 (Fig. 1a).
Epitope alanine scanning
In order to determine the hotspots on the epitope, we first calculated the solvent accessibility of all FGF23 residues on the N-terminal domain to search for highly exposed FGF23 residues. The selection of highly exposed residues was predicated on the assumption that residues with high solvent accessibility were likely to participate to the binding interface with Burosumab. We used a crystal structure of FGF23 (PDB: 5W21) reported by Chen et al. (PDB: 5W21)14 for the calculation of solvent accessibility using BIOVIA Discovery Studio version 2017 R2. Residues with relative solvent accessibility (RSA) more than 20% of maximum RSA within the structure were designated as solvent exposed residues, consistent with published studies12,15. In total, we selected fifteen FGF23 residues that were highly solvent-exposed and covered every sub-domain of FGF23 for alanine scanning. Specifically, the FGF23 residues Y43, R48, H52, H66, Y70, R76, F82, R91, R99, Y107, F108, H117, E121, Y124, and H133 were mutated to alanine using site-directed mutagenesis. The N-terminal domain of FGF23 mutants was attached to the His-SUMO tag (denoted as FGF23 in subsequent parts), expressed in BL21(DE3), purified using Ni–NTA resin, and calculated for their purity using SDS-PAGE analysis (Supplementary Fig. S3). All FGF23 mutants were coated on ELISA plate in proportion to their purity (Methods) and evaluated for binding to Burosumab (Fig. 1c).
Results revealed five hotspot residues that significantly impacted the binding upon modification. Notably, alanine substitution at H52, R76, F108 residues of FGF23 reduced binding ranging between 7 to 10-fold. Y124A mutation reduced binding by almost 50-fold as compared with the wild-type FGF23 (Fig. 1d). Of note, H117A mutation alone completely knocked the binding (> 100-fold). We also noticed that 3/5 of FGF23 hotspot residues were positively charged as opposed to the negatively charged hotspot residues on the paratope interface, suggesting that FGF23-Burosumab interaction may be driven by electrostatically charge-charge complementary. Unexpectedly, we found that these epitope hotspots were located on two different regions, although they lined close to the central cavity of FGF23 overlapping with FGFR binding site (Fig. 1c). The first region included H52 residue on β2 strand and Y124 residue on β9 strand, which seemed to be in close contact with the D2 domain of FGF23 receptor (FGFR) and the second region included R76 residue on β4 strand, F108 residue on β7-β8 hairpin loop, and H117 residue on β8 strand, which were in proximity to the D3 domain of FGFR (Supplementary Fig. S1). Taken together, epitope alanine scanning identified five potential epitope hotspots: H52, R76, F108, H117, and Y124 overlapping with FGFR binding interface of FGF23.
Molecular docking of FGF23 and Burosumab
To determine how Burosumab interacts with FGF23, we docked the structure of FGF23 (source: PDB: 5W21) with the Fv model of Burosumab generated using the HADDOCK2.4 server (https://wenmr.science.uu.nl/haddock2.4/). The identified paratope and epitope hotspot residues from alanine scanning were used as restraints during docking. HADDOCK2.4 has been validated on benchmark version 5.0 (BM5) datasets, which is comparable with other docking programs evaluated by CAPRI studies16,17. Since the epitope hotspots mapped to distinct regions of FGF23 placing them in close proximity to D2 or D3 subunits of FGFR, we performed docking with three different sets (scenarios) of epitope residue restraints: (1) H52 and Y124 residues (residues contacting D2-FGFR), (2) R76, H117, and F108 residues (residues contacting D3-FGFR), and (3) H52, Y124, R76, H117, and F108 residues (combined D2 and D3 scenarios), while the same set of paratope residue restraints: VH: Y33, VH: D95, VH:D98, VL:D50, and VL:F96 were used throughout the docking. HADDOCK2.4 takes input from experiments including site-directed mutagenesis to guide the docking process. The interface is selected based on a combination of traditional energetics and shape complementary metrics. The HADDOCK score is calculated based on interaction energy that includes van der Waal energy, electrostatic energy, empirical desolvation energy and ambiguous interaction restraints (AIR)18. We ran HADDOCK2.4 under a default setting, which generates 1,000 poses in total for each run. In order to minimize the risk of selecting a non-native pose, we employed knowledge-based constraints and experimental data to eliminate spurious models. Briefly, the top pose from each scenario was analyzed using seven criteria: (1) degree of stereo blockage of FGFR binding, (2) binding energy, (3) percent correlation with paratope alanine scanning data, (4) buried surface area at the interface, (5) percent contribution of framework residues (%FWR) at the interface, (6) percent contribution of CDR residues at the interface (%CDR), and (7) percent contribution of HCDR3 domain at the interface. Notably, structural analysis of antigen-antibody structural complexes from PDB have established upper bound on features 4–7, which help eliminate models that significantly deviate from near-native structures.
The top pose selected from the three scenarios were studied further (Fig. 2). Pose27 and pose180 exhibit a strong blockage of FGFR with similar buried surface areas (906.7 and 920.8 Å2, respectively) whereas pose39 possesses a slightly smaller buried surface area (859.3 Å2). To investigate residue contributions at the interface, each pose was submitted to PISA server (https://www.ebi.ac.uk/pdbe/pisa/). Interfacial data of pose27 showed that 77.77% of interface area was contributed by CDR residues of which HCDR3 (H3) and LCDR3 (L3) residues accounted for 22.21% and 21.27%, respectively while the remaining four CDRs account for 34.29%. This suggests that pose27 had a relatively high contribution of HCDR3 and LCDR3 to the binding, as commonly observed in other antibodies19–21. Unlike pose27, pose39 and pose180 showed relatively smaller HCDR3 contribution (pose27: 22.21%; pose39: 15.91%; pose180: 14.05%), even though they both had higher overall CDR contribution than pose27 (pose27: 77.77%; pose39: 86.53%; pose180: 86.04%). Interfacial data also showed that pose27 had the highest % FWR contribution (22.23%), which was accounted by residues flanking HCDR1 (Y33), HCDR2 (I50, T58, and S59), and LCDR2 (Y49). Most importantly, pose27 showed the highest % correlation with paratope alanine scanning, which was followed by pose39 and pose180.
Taken together, we considered pose27 as the top-ranked pose for use as model to demonstrate the FGF23-Burosumab interaction as it best agreed with (1) experimental data (highest % correlation with paratope alanine scanning), (2) physiochemical data (lock-and-key propensity implied by high %FWR correlated with a comparable size between FGF23 and Burosumab Fv), (3) biological data (strong blockage of FGF23), and (4) immunological data (highest contribution of HCDR3 which is a common of therapeutic antibodies with high specificity and high affinity; Burosumab is a high affinity antibody, given the Kd of 10− 11 M).
Pose27 showed that Burosumab recognized three main epitope regions on FGF23 including β1-β2 hairpin loop (epitope-1), β8-β9 hairpin loop (epitope-2), and β10 strand (epitope-3) by forming multiple hydrogen bond interactions, salt bridges, pi-anion interactions, and attractive charge interactions (Supplementary Table S4). Interfacial analysis showed that epitope-1 and epitope-3 of FGF23 together formed a planar binding site for Burosumab in which A47 and Y43 of the epitope-1 formed hydrogen bonding networks with LCDR1 and LCDR2, while R160 and R161 of the epitope-3 formed both salt bridge and attractive charge interactions with LCDR2 and HCDR3 (Fig. 3a). On the contrary, epitope-2 formed a conformational binding site (sharp turn) for protruding into a pocket region at the VH-VL interface of Burosumab, which was formed mainly by LCDR3, HCDR1, and HCDR2 (Fig. 3b). We speculated that H117, T119, D125, and S159 were four key FGF23 residues that maintain the structural shape of epitope-2 through intra-interaction networks (Fig. 3c), while R160A and R161A were two key residues that stabilized the complex of the interaction.
To validate pose27, six additional FGF23 alanine mutants: T119A, D125A, L158A, S159A, R160A, and R161A were generated and tested for binding using ELISA (Fig. 3d). T119A, D125A, and S159A mutations were selected to see the effect of the mutation to the intra-interaction networks, which determines the sharp turn conformation of β8-β9 hairpin loop or epitope-2 (Fig. 3c). While L158A, R160A, and R161A mutations on β10 strand or epitope-3 were selected to investigate the effect of the mutation to the hydrophobic interaction at the center of interface (by L158A) and to the electrostatic interaction at the peripheral interface (by R160A and R161A) (Supplementary Table S4).
Among these mutants, D125A mutation remarkably reduced the binding consistent with the structural model that showed the intra-electrostatic interaction between D125 and H117 (hotspots), which is necessary for the sharp turn of β8-β9 hairpin loop. This finding suggests a conformation-specific recognition of Burosumab. Intriguingly, the other single mutations caused no significant change to the binding. Notably, R160 and R161 which appeared to be involved in electrostatic interactions did not drop binding upon mutation to alanine. We speculated that the close proximity of these residues may render one to compensate for the other. To test this hypothesis, we created a double mutant – R160A-R161A – and tested its binding to Burosumab. The double alanine mutant dropped the binding by 7-fold, suggesting that each of these residues in fact play a compensatory role in the absence of the other (Fig. 3d).
Assessment of antibody cross-reactivity
Next, we sought to apply our model to explain the cross-species reactivity of Burosumab. As reported by the originator in the European Medicines Agency (EMA), Burosumab binds to FGF23 from human, monkey, and rabbit species (so-called binder species), but not from mouse and rat species (so-called non-binder species)2. We first performed amino acid sequence alignment of N-terminal domain of FGF23 from human, monkey, rabbit, rat, and mouse. The FGF23 from the different species exhibited varying degrees of sequence identity to human FGF23: rabbit: 87%; monkey: 88%; mouse: 78%; and rat: 78%. We identified thirty-three residues that were different between human and mouse FGF23 within the N-terminal domain. These residues include three residues: I40, N49, K57 on β1-β2 hairpin loop, two residues: N58 and A64 on β3 strand, four residues: R76, F82, V88, S90 on β4-β5 hairpin loop, five residues: Y93, R99, H106, Y107, D109 on β6-β7 hairpin loop, nine residues: R114, Q116, H117, Q118, H128, P130, Q131, Y132, F134 on β8-β9 hairpin loop, five residues: A144, L146, M149, Y154, S159 on β10 strand, and five residues: I164, I167, N170, P172, I173 on α-helix region (Fig. 4a). Seventeen out of thirty three (17/33) or 52% of these residues are located on β1-β2 hairpin loop, β8-β9 hairpin loop, and β10 strand, which make up the predicted epitope region.
These findings prompted us to investigate the underlining mechanism of the cross-reactivity of Burosumab using pose27. We hypothesized that the physiochemical properties of the central cavity, which plays a central role for binding recognition of Burosumab, may also contribute to the species-specific binding behavior (Fig. 4b). To test this hypothesis, we created four back mutations in which polar and hydrophobic residues surrounding the central cavity of mouse FGF23: T49, T76, Q117, and A159 were replaced by corresponding charged and polar residues of human: N49, R76, H117, and S159, respectively. Three combined mutations: (1) combined Q117H and A159S mutations, (2) combined T49N, Q117H, A159S mutations, and (3) combined T49N, T76R, Q117H, A159S mutations were also created. These seven mouse FGF23 mutants were expressed and tested their binding to Burosumab using ELISA (Fig. 4c). Among those individual mutations, both Q117H and T76R mutations slightly improved the binding as compared to wild-type human FGF23 and mouse FGF23, while T49N mutation exhibits similar binding to wild-type mouse FGF23 and A159S mutation worsens the binding. Among the three combination mutations, Q117H and A159S mutations and T49N, Q117H, A159S mutations significantly rescued the binding nearly equivalent to wild-type human FGF23. These results suggest that the co-occurrence of H117 and S159 residues was necessary for the binding to mouse FGF23, whereas R76 does not play a significant role to the binding. It is also important to note that while the originator reported no binding with murine FGF23 based on competitive SPR, we found that Burosumab bound weakly to murine FGF23 based on ELISA. This disparity may arise from different formats of the binding assay. In particular, competitive SPR format has a higher stringency as compared with the ELISA format which permits avidity effects from the bivalent binding of Burosumab.
Affinity enhancement of Burosumab
The structural insights of FGF23-Burosumab interaction gained from computational docking prompted us to further design Burosumab variants with improved binding affinity. Using pose27 as the starting model, four mutations on VL domain – A32S, S52D, S67Y, T69D – were predicted to enhance the interatomic. VL:S52D and VL:T69D mutations were aimed to enhance electrostatic interactions with R160 and R48 of FGF23, respectively. On the other hand, VL:A32S and VL:S67Y mutations were designed to enhance hydrogen bond formation with Y124 and N49 of FGF23, respectively. The four single mutants were generated and tested for their binding to FGF23 as compared with wild-type Burosumab using ELISA. Binding results showed that only VL: A32S had noticeably enhanced affinity by 1.6-fold (Fig. 5a). To further enhance the affinity of Burosumab, we combined VL: A32S mutation with VH: V97A mutation. The latter mutation improved binding affinity (1.2-fold) in paratope alanine scanning (Fig. 5b). This combined Burosumab variant (combined VH: V97A and VL: A32S mutations) was generated and tested for binding using ELISA. Results showed that this combined variant improved binding affinity by 3.7-fold as compared with the wild-type (Fig. 5c,d).
It is important to note that although the positions of V97A and A32S mutations are far away from each other, they may contribute to each other to promote a greater favorable interaction between FGF23 and Burosumab. We believed that 3.7-fold enhancement may arise from entropy lost-enthalpy gain compensation; in that the entropy lost from the decrease of flexibility of V97A upon complex formation was compensated by the enthalpy gain of A32S due to more complex stabilization by inter-hydrogen bond formation of A32S with Y124 of FGF23 (Fig. 5e). These findings suggest that the gain-in-function with least overall structural/energetic disturbance may be the underlying strategy for affinity enhancement of therapeutic antibodies that possess rigid/densely packed CDRs.