The variable domains from the original J22.9 mouse IgG had been cloned onto a human IgG1 Fc domain for recombinant production, generating the chimaeric J22.9-xi. Although the constant regions of the four IgG subclasses show very high sequence similarities, the efficient activation of complement together with the high affinity binding to Fc receptors led to the choice of the IgG1 Fc with a view towards clinical application; the observed efficiencies of J22.9-xi in CDC (complement-dependent cytotoxicity) and ADCC (antibody-dependent cellular cytotoxicity) experiments subsequently justified this choice [18]. For the full humanization, identification of the human germline sequences corresponding to the J22.9-xi VH and VL chains was performed using IgBLAST [25] and subsequent sequence alignments indicated a total of 41 residue changes needed to produce fully humanized sequences, 16 in VH and 25 in VL (Fig. 1 A, B). Each modification was modelled into the J22.9-xi:BCMA complex structure to identify any potential clashes or unfavorable bond angles due to steric constraints. Nearly all substitutions (mostly surface exposed) appeared likely to be accommodated by the antibody without obvious impact on the binding site.

Humanizing mutations in J22.9-xi. A Protein sequences of the mouse variable region heavy (mVH) and light (mVL) chains aligned with the corresponding human germline sequences (hVH and hVL). CDRs 1, 2, and 3 from each chain are underlined. Humanizing mutations are highlighted in magenta, proposed CDR positions for stabilizing PTM mutations are highlighted in red and the complex disrupting A46 indicated in green. Vertical dashed lines indicate the corresponding boundaries of the cloning cassettes. B Space-filling model of the J22.9-xi variable domain showing the positions of the proposed mutations on the structure (PDB ID: 4ZFO). As in A, the VH is depicted in blue, the VL in yellow, humanizing mutation positions in magenta, stabilizing CDR PTM positions in red, and A46 in green. The right panel view depicts a 180° rotation along the vertical axis of the left panel view

The single exception concerned alanine 46 (A46) of the mouse VL, that in the human VL, is a leucine (variation = A46L). In the J22.9-xi:BCMA structure, A46 is buried in a hydrophobic pocket bounded by L99, D108, and W110 from VH and Y36, L47, and F55 from VL (Fig. 2). These residues, together with the β-carbon from the side chain of D108 from VH, form a small cavity closely packed around the alanine β-carbon atom that is directly adjacent to the BCMA binding site. The sidechain faces of both F55 and L99 opposite A46 are in direct contact with L17 in BCMA, the residue at the center of the DxL loop and the primary recognition feature for J22.9-xi and both native BCMA ligands. Additionally, F49 from VL is packed tightly against F55 from VL and L18 of BCMA. The additional volume required upon substitution with the much larger leucine sidechain was expected to induce substantial rearrangement of the interactions around the A46 pocket and therefore interfere with BCMA binding.

The A46 hydrophobic pocket of J22.9-xi. Two views of the A46 binding environment with corresponding residues in space-filling representation and labeled. VH is in blue, VL in yellow, and BCMA in violet. A46 (VL) is seen in the center in stick representation with color-coding of individual atoms (nitrogen blue, carbon green, and oxygen red) and the beta carbon atom in the pocket represented as a green sphere. The right panel view depicts a 180° rotation along the vertical axis of the left panel view. The close packing around the A46 sidechain is clear from both panels as are the interactions with L17 and L18 of BCMA

We divided the positions to be mutated between cloning cassettes that could be easily combined to test the maximum number of substitutions with as few new constructs as possible (Fig. 1A). It was reasoned that by stepwise combination of cassettes encoding the fully humanized sequences with those of the original mouse sequences, detrimental mutations (as qualitatively assessed by binding to cells displaying BCMA in FACS, described in [18]) could be rapidly identified. Mutated positions on cassettes whose presence disrupted binding could then be individually replaced by PCR to narrow down the problematic change(s). Several constructs were cloned, and the antibodies were produced in a small scale as described for J22.9-xi. It was quickly determined that only antibodies harboring the A46L substitution were negative for binding by FACS (not shown). Reversing the change back to alanine restored binding in the absence of any other changes, confirming the original predictions from the modelling. The final humanized variant comprising all substitutions in both the VH and VL, with the exception of A46 in the VL, is hereafter referred to as J22.9-H.

The affinity of J22.9-H for human BCMA (hBCMA) was determined by SPR to be (1.5 ± 0.3) × 10−8 M—a nearly 200-fold loss compared to J22.9-xi at (2.8 ± 0.7) × 10−10 M (Table 1 and Fig. 3). In an attempt to mitigate this loss in affinity and impart greater long-term stability to the variable domains, we modelled additional mutations into the CDRs of both chains that were intended to prevent detrimental post-translational modifications (PTMs). The sulfur containing methionine is subject to oxidation by reactive oxygen species, particularly at low pH, and this has been shown to reduce the conformational and thermal stability of antibodies [26, 27]. In their ionized form, Asp sidechains can perform a peptide chain breaking reaction in which the sidechain carboxylate attacks the peptide bond carbonyl carbon [28] and Asp residues adjacent to those having hydroxyl containing sidechains, like serine, have been shown to be more reactive [29]. Three proposed PTM sites were identified in hVH, one in each of the three CDRs (Fig. 1A): methionine 34 in CDR1 to be substituted with isoleucine and phenylalanine (M34I/F); aspartic acid 54 in CDR2 to be substituted with serine (D54S); and methionine 107 in CDR3 to be substituted with tyrosine (M107Y). Only one mutation in hVL, aspartic acid 30 to glutamic acid (D30E) in CDR1, was proposed. Since both J22.9-H aspartic acid residues, D54 and D30, occur adjacent to a serine, both were targeted for replacement. Although conformational changes to the CDRs induced by the proposed substitutions could not be ruled out, an inspection of the J22.9-xi:BCMA complex structure made clear that none of these residues made either direct or indirect (over water molecules) contacts to BCMA [30]. The final constructs incorporating all of the humanizing mutations (except A46L) and the stabilizing CDR PTM substitutions, 44 residue changes in total, are referred to as J22.9-FSY and J22.9-ISY, based on the identity of the hVH M34 substitution (phenylalanine, F, and isoleucine, I, respectively).

Measurement of BCMA binding affinities of J22.9 variants. SPR sensorgrams of J22.9 variants (labeled) binding to hBCMA and cynoBCMA. The colored traces correspond to increasing concentrations of BCMA in the mobile phase binding to immobilized IgG. The leftmost vertical dashed line indicates the start of ligand flow over the immobilized antibodies, during which association is measured, and the rightmost indicates the switch to BCMA-free buffer, initiating the dissociation phase

By SPR, the set of stabilizing CDR substitutions did indeed improve the binding affinity for BCMA, fortuitously by more than tenfold over J22.9-H, putting them in the low nanomolar range (Table 1). Since toxicity studies carried out in preparation for clinical trials often use macaques (cynomolgus monkeys), the affinity of each variant for cynomolgus BCMA (cynoBCMA) was also determined. cynoBCMA differs from the human homolog at three positions—in cynoBCMA, alanine 20 is an aspartic acid and isoleucine 22 is a lysine—a potentially destabilizing change for the interaction with the antibodies, since it introduces a charged side chain directly into the hBCMA binding epitope; additionally, asparagine 31 in hBCMA is deleted in cynoBCMA. However, despite the presence of lysine 22, all variants were able to bind cynoBCMA, and in all cases, the affinity by SPR was approximately tenfold lower for cynoBCMA than for hBCMA.

All variants were subjected to crystallization trials and diffracting crystals were obtained for J22.9-H and J22.9-ISY. We noted that a D54N substitution in the hVH CDR2 would produce a eukaryotic N-linked glycosylation signal sequence (consensus sequence N–X–S/T, where X is any amino acid except proline; in J22.9-H, the D54N substitution produces N-S–S) and, since the CDR2 loop does not extensively contact BCMA, we introduced the D54N substitution into J22.9-FSY. With the resulting variant, referred to as J22.9-FNY, we also obtained diffracting crystals. We solved all three structures by molecular replacement: J22.9-H and J22.9-ISY to 3.1 Å and J22.9-FNY to 2.7 Å (Table 2), thereby allowing comparison of the binding interaction of the fully humanized and optimized IgGs with that of the original chimaeric and biologically active J22.9-xi.

Alignment of J22.9-H and the fully humanized and CDR optimized J22.9-ISY and J22.9-FNY structures with that of J22.9-xi provides verification of the retention of the BCMA binding mode after incorporation of the necessary 44 substitutions (Fig. 4). The backbone root mean square deviation (RMSD) of the variable domains between J22.9-xi and J22.9-H is 0.86 Å, indicating no substantial changes in overall architecture. The VH and VL domain conformations of both J22.9-FNY and J22.9-ISY are, within experimental error, identical to those of the corresponding J22.9-H domains, demonstrating the minimal structural impact of the PTM modifications. There are only minor alterations in the backbone conformations of CDRs 1 and 2 from VL between J22.9-xi and the humanized variants, while differences between the VH chains are also minimal, confined to small shifts without changes in their topologies.

Structural alignments of J22.9-xi with humanized and optimized variants. A View into the binding pocket of all four J22.9 aligned variants with BCMA removed. Only the backbone atoms are shown as lines, with the CDRs depicted as sticks. The VH CDRs of J22.9-xi are colored blue, and the VL CDRs are colored yellow with J22.9-H shown in magenta, J22.9-ISY in light blue, and J22.9-FNY in light green. The positions of all 4 PTM modifications are indicated. B Separate views of the VL (left panel) and VH (right panel) domains with CDR depictions and coloring as in A. The conformations of all VL CDR loops are nearly identical, as are those of the VH CDRs of the humanized variants that all show a slight shift from the corresponding positions in J22.9-xi. C Overlay of BCMA from all J22.9 variants generated by alignment of all 4 complete complex structures. BCMA bound to J22.9-xi is shown in violet and BCMA molecules from the humanized structures are colored as per their corresponding VL and VH chains in A. The DxL loop residues are indicated

Assessment of the elbow angle (the angle between the pseudo-twofold axes relating VL to VH and CL to CH1 [30]) and the variable domain packing angles (the torsion angle between respective centroids determined based on the positions of conserved framework residues in VH and VL [31]) of each antibody showed the major change to be between the chimaeric J22.9-xi and the group of resulting humanized variants. At 173.8°, the elbow angle of the J22.9-xi falls within the range determined for the majority of VL-Fabs. The fully humanized J22.9-H showed a 13.3° difference to J22.9-xi, with an elbow angle of 160.5°; the CDR optimized variants displayed no further change in elbow angle, with J22.9-FNY at 159.4° and J22.9-ISY at 160.4°. The VH/VL packing angle differences followed the same pattern, with J22.9-xi at − 46.6° and all three humanized variants at − 41.9°. This calculated rotation is consistent with the minor shift in the CDR positions visible from the alignments in Fig. 4A and B.

Within experimental error, the antigen binding site is identical between the original and fully modified IgGs. In all four structures, the quality of the electron density for BCMA is highest in the region of the binding site, particularly the segment containing the DxL loop. Aside from minor variations in sidechain conformation, this loop is positioned identically in the binding cavities of all antibody variants (Fig. 4C), with a nearly perfect superposition of L17, the critical residue recognized by all known BCMA binders, when the complete complex structures are aligned.

The J22.9-FNY structure also shows retention of the important binding interactions between BCMA and IgG and only minor shifts in the backbone positions of the CDR loops despite the large NAG modification of VH CDR2 (Fig. 5). Here again, the positions of the DxL loop and L17 are nearly identical to those seen in the J22.9-xi structure. That structure showed the minimal interaction of CDR2 with BCMA, allowing the possibility that substantial changes to this loop could be tolerated without eliminating ligand binding. Challenging this supposition with the glycosylation confirmed that productive changes in the CDR2 loop were possible, further emphasizing the value of the structure-based approach.

J22.9-H and J22.9-FNY binding site comparison. Structural alignment of J22.9-H:BCMA and J22.9-FNY:BCMA showing the interaction of BCMA with CDR2 and the position of the NAG modification. The J22.9-H VH is colored blue, and the J22.9-FNY VH is colored light blue. The N-acetylglucosamine (NAG) modification on J22.9-FNY is colored orange. BCMA bound to J22.9-H is shown in violet and BCMA bound to J22.9-FNY in magenta. The NAG modification in J22.9-FNY induces no substantial changes in the CDR2 loop. The most important features of both binding partners are labeled, including the DxL loop of BCMA, showing that the central L17 residue occupies a nearly identical position in both structures relative to the antibody. The single most extensive interaction in the complex, between W33 in VH CDR1 and H19 in BCMA, is also explicitly shown and is unchanged between the two structures. The electron density (contoured at 1.0σ) for the NAG modification of J22.9-FNY is depicted as a grey mesh.