Humanization of rabbit mAb 4A11 by grafting CDRs to the closest human frameworks
Rabbit monoclonal antibody 4A11 (rbt4A11) was humanized in a two-step strategy. First, the human germline frameworks that are closest to the variable light chain (VL) and variable heavy chain (VH) of rbt4A11 were identified based on the selection criteria of the highest amino acid sequence identity in variable region, frequent usage in antibody repertoire, and functional genes in different germline family. In total, we found four human light chain germline genes (IGKV1D-39*01, IGKV4-1*01, IGKV3-20*01, IGKV2-24*01) and two human heavy chain germline genes (IGHV3-48*01, IGHV4-59*06) as the closest acceptor frameworks, which were 62%, 61%, 54% and 54% identical to the VL, and 57% and 52% identical to the VH of rbt4A11, respectively. Subsequently, the rbt4A11 CDRs, which covered the definition of Kabat and Chothia6, and fifteen rabbit framework residues at the Vernier zone (position 2, 4, 36, 43 and 58 of VL, and position 2, 24, 37, 48, 49, 67, 71, 78, 91 and 105 of VH) were grafted onto each acceptor framework to generate four humanized VL and two humanized VHgene segments for molecular cloning (Figure 1). To assess the best humanized variants, a panel of eight humanized 4A11 variants (h4A11.v1 - h4A11.v8) was generated through mixing and matching the cloned humanized gene segments and expressing each variant as human IgG1 for further characterization (Table 1). The expression and purification profiles of all humanized variants were good (>99% monomer) with no noticeable differences between germline usage. The variable domains of these eight humanized 4A11 variants comprised an identity of 77% to 81% in the human immunoglobulin germline sequence, which were further improved by framework Vernier permutation from original rabbit residues to the corresponding human germline residues to meet World Health Organization standards as a humanized antibody (data not shown).
Functional characterization of humanized 4A11 variants
Each humanized 4A11 variant was purified and characterized by measuring its binding affinity and blocking activity against all three mature TGFβ isoforms. To minimize the avidity effect on binding, the affinity of each variant was measured by real-time Biacore SPR using label-free TGFβ in solution with low level of antibody bound to the biosensor surface, as previously described6. Further, the blocking activity of the variant IgGs to inhibit mature TGFβ induced TGFβ receptor dependent signaling was determined using reporter cell lines (HEK-Blue™ TGFβ)1.
After initial characterization, we observed that all eight primary humanized 4A11 variants preserved the same selectivity profiles as parental antibody (rbt4A11) by binding to human mature TGFβ2 and TGFβ3, but not TGFβ1 (Table 1). Although the humanized variants differed slightly in binding affinity compared to rbt4A11, they retained picomolar binding affinity. However, dramatic differences of blocking activity in the cell-based assay for these humanized variants were observed, which was unexpected because the variants contained almost identical CDRs and closely matched germline frameworks. After detailed analysis, we concluded that humanized variants (v3, 4, 7, and 8) using the heavy chain germline (IGHV4-59*06) as an acceptor framework maintained similar potency compared to rbt4A11, whereas the humanized variants (v1, 2, 5 and 6) using the other heavy chain germline (IGHV3-48*01) significantly lost their blocking activity and also turned into incomplete blockers without reaching maximum inhibition at the highest antibody concentration (<70%). The usage of different light chain germlines had minimal impact on the blocking activity. Among all eight primary humanized variants, v7 had the most desirable affinity (TGFβ2_KD: 6.3pM; TGFβ3_KD: 9.7pM) with the best blocking activity (TGFβ2_IC50: 0.9nM; TGFβ3_IC50: 0.2nM), and was pursued further as the lead candidate1. This result pointed out that the usage of two different yet closely related human heavy chain germlines in the humanization led to the dramatic difference in their blocking activities.
Crystal structures of TGFβ2 in complex with antigen-binding fragments (Fabs) of h4A11.v2 and h4A11.v7
To further investigate how the heavy chain germline acceptor framework differences in the humanized 4A11 variants resulted in significant functional differences in blocking TGFβ2 and TGFβ3, we selected one of the incomplete blockers (h4A11.v2) and the top complete blocker (h4A11.v7) to obtain crystal structures of Fab fragments in complex with TGFβ2. We reported the crystal structure of the h4A11.v7 Fab in complex with TGFβ2 recently1. To investigate any structural differences in the TGFβ2:Fab complexes for both v2 and v7 variants, we crystallized and solved the x-ray crystal structure of TGFβ2 in a complex with Fab from h4A11.v2 to 2.9 Å resolution (Table 2). The overall structures of both v7 and v2 complexes are very similar, with two Fab fragments binding to two TGFβ2 molecules and the epitope of each Fab containing contributions from both TGFβ2 monomers in the dimer for the h4A11.v2-Fab, as previously reported for the h4A11.v7-Fab (Figure 2). The overall RMSD for all atoms in TGFβ2 and the variable regions (Fv) of the complexes is 1.04 Å. The constant regions of the Fabs were excluded from these calculations because the wide variation in elbow angle of Fabs distorts the overall alignments in an irrelevant way7.
An analysis of TGFβ2:h4A11.v2-Fab and TGFβ2:h4A11.v7-Fab crystal structures revealed that the epitopes of both h4A11.v2 and h4A11.v7 antibodies on the TGFβ2 dimer are identical (Figure 3). This, along with identical CDRs between the two variants, suggested that the functional differences observed between the two variants cannot be explained by a potentially altered epitope. Further analysis of the two structures revealed an intriguing interaction of the 2 heavy chain molecules in the TGFβ2:h4A11.v2-Fab complex with each other, which was not present in the TGFβ2:h4A11.v7-Fab complex. In the TGFβ2:h4A11.v2-Fab complex structure, Y79, R19 from one heavy chain molecule form a ternary π-stacking interaction with Q81 from the other heavy chain molecule and vice versa, resulting in a pseudo-symmetric interaction (Figure 4A), which likely orients the relative positions of the two heavy chains and the conformation of the Fabs relative to TGFβ2 and alters the dynamics of the h4A11.v2 Fab when bound to TGFβ2 in solution, compared to the h4A11.v7 Fab. Supporting this hypothesis, in the TGFβ2:h4A11.v7-Fab complex structure, the corresponding residues in the heavy chain framework region are S19, S79 and K81, which do not allow for the specific interactions observed in the TGFβ2:h4A11.v2-Fab complex (Figure 4B). Although the distances between the Cα atoms of the respective residues at each position (19-19, 79-79, 81-81) in the heavy chains of both the h4A11.v2 and h4A11.v7 Fabs are almost identical in the two structures, we hypothesized that in solution, the interactions between the side chains at 19, 79 and 81 among the 2 heavy chain molecules and subtle effects on the conformations of the Fabs may be the difference between a functional ‘complete blocker’ and an ‘incomplete blocker’, where the slight differences induced in the dynamics of the TGFβ2 dimer itself due to the presence or absence of the subtle Fab-Fab interaction could play a role. For such an interaction, we assume that two Fabs from the same antibody molecule would bind to a TGFβ2 dimer in solution since the rbt4A11 antibodies have monovalent interactions with dimeric TGFβ2 and TGFβ3. To test this hypothesis, we performed step by step mutagenesis to try and convert a v2-like function (incomplete blocking) to a v7-like function (complete blocking) with an increasing number of mutations listed in Table 3. We hypothesized that a simple R19S substitution in h4A11.v2 would disrupt the π-stacking interaction, and so would Q81N or Q81A. In addition, to convert a v7-like function to a v2-like function, we hypothesized that we would need to introduce a greater number of mutations at positions 19, 79, 81 to re-build the stabilizing interactions of a v2-like molecule. In this process, we also included an aromatic sidechain W at position 79, as it could result in a π-stacking interaction similar to Y in the TGFβ2:h4A11.v2-Fab structure (Figure 4).
Generation and functional characterization of structure-guided h4A11.v2 and h4A11.v7 heavy chain framework variants
To test the hypothesis in accordance with our structural analysis, a new panel of variants was generated with single or multiple mutations at position 19 (HC-FWR1), 79 and 81 (HC-FWR3) of h4A11.v2 and h4A11.v7. Following expression and purification, all variants were fully characterized and compared with parental rbt4A11, h4A11.v2 and h4A11.v7. As predicted from our structural analysis, all h4A11.v2-based variants (h4A11.v2.1 - h4A11.v2.6) demonstrated binding affinity improvement as well as regained complete blocking activity in the cell-based assay against TGFβ2 and TGFβ3 (Table 3). All three single mutation variants (v2.1_R19S; v2.2_Q81N; v2.3_Q81A) were sufficient to restore complete blocking function similar to rbt4A11 (v2.1 in Figure 5). Furthermore, by adding other mutations to R19S (v2.4_ R19S, Y79S; v2.5_ R19S, Q81S; v2.6_ R19S, Y79S, Q81S), did not further change the effect of the individual mutations, suggesting the critical role of positions 19, 79 and 81 in the observed h4A11.v2 heavy-chain interactions upon TGFβ2 and TGFβ3 binding. In contrast, all h4A11.v7-based variants (h4A11.v7.1 - h4A11.v7.6) with double mutations (v7.1_S19R, S79Y; v7.2_S19R, S79W) and triple mutations (v7.3_S19R, S79Y, K81Q; v7.4_S19R, S79Y, K81R; v7.5_S19R, S79W, K81Q; v7.6_S19R, S79W, K81R) appeared to have a broad range of affinity drop (4-34 fold) compared to rbt4A11 and turned into incomplete blockers against TGFβ2 and TGFβ3 (v7.1 in Figure 5). These results supported the important functional role of short sidechain amino acids at human framework positions 19 and 79 in the active h4A11.v7 molecule.