Novel CH1:CL interfaces that enhance correct light chain pairing in heterodimeric bispecific antibodies

Abstract

Targeting two unique antigens with a single bispecific antibody is an attractive approach with potential broad therapeutic applicability. However, the production of heterodimeric bispecific antibodies (bsAbs) presents a challenge, requiring the co-expression and accurate pairing of two distinct heavy and light chain units. Several undesirable by-products can be formed in the production process, including heavy chain homodimers and non-cognate light chain pairings. Although additional downstream purification methods exist, they are often time consuming and restrict practical large-scale production. In this study, we identify and validate novel Fab interface mutations that increase cognate light chain pairing efficiencies within heterodimeric bsAbs. Importantly, the variable domains remain unaltered as interface mutations were restricted to the CH1 and CL domains. We performed several biochemical assays to demonstrate that the novel engineered interfaces do not adversely impact bispecific antibody expression, stability, affinity and biological function. The designs reported here can easily be applied in a generic manner to use existing antibodies as building blocks for bsAbs which will help to accelerate the identification and production of next generation bispecific antibody therapeutics.

Keywords: Fab interface design; bispecific antibodies; heterodimeric IgG; light chain pairing problem; orthogonal Fab engineering.

Fig. 1

Chain pairing variants in bsAbs. A bsAb combines heavy and light chains of two different parent antibodies. The co-expression of all four chains in a single host cell potentially leads to various misassembled by-products. If heavy chain heterodimerisation is successful, three possible mispaired variants can still occur where one or two light chains bind to a non-cognate heavy chain.

Fig. 2

Structural alignments using the align function of PyMOL. (A) Superimposition of C_H3:C_H3 domains (PDB ID:1oqo) and the C_H1:C_L domains (PDB ID: 3eo9). (B) Superimposition of C_H3 and C_L as isolated objects. Interface residues in C_H3 previously chosen to induce C_H3:C_H3 repulsion (Ying et al., 2012) and their homologs in C_L are indicated.

Fig. 3

Identification of candidate mutants in C_L. (A) Schematic drawing of the model antibody 3D6^Q44E. The mutation Q44E was introduced in V_H and V_L to enhance the repulsive effect generated by mutations in the C_H1:C_L interface. (B) Western blots to assess the impact of mutations in C_L on soluble antibody expression in HEK293-6E. Left, 3D6^Q44E-HQ10 (S85.1E:S86K:T88A), 3D6^Q44E-HQ13 (F7S:V22R:L24H:V82K:S85.1E:S86K:T88A) and 3D6^Q44E-HQ24 (F7S:V82K). Right, effect on antibody expression of 3D6^Q44E-HQ24 compared to single mutation F7S. (C) Analytical SEC and (D) circular dichroism of C_L domains with or without mutated interface residues. (E) Differential scanning calorimetry of wildtype C_L and C_L-F7S.

Fig. 4

(A–C) Effect of mutations in C_L and C_H1 on antibody expression. The antibody concentration in culture supernatants of HEK293-6E was determined 5 days post transfection using ELISA. 3D6^Q44E was expressed with one mutation in either C_L only (white), C_H1 only (black) or in both C_L and C_H1 (grey). The interface designs with mutations in both C_L and C_H1 were named as indicated in the graphs. The expression is given relative to the parental antibody 3D6^Q44E.

Fig. 5

Interface designs MaB5, MaB21, MaB40, and MaB45 compared to the respective sections of the wildtype C_H1:C_L interface (PDB ID: 3eo9). The structures of interface designs were modelled using SWISS-MODEL. Residues chosen for mutation in the wildtype structure and mutated residues in the structure models are underscored. Neighbouring residues within a radius of 4 Å are shown.

Fig. 6

Analysis of pairing of light to heavy chain in the bsAb B10v5 × hu225M by LC–ESI–MS, after protein A purification and deglycosylation. The bsAbs contained either no mutation in the Fab interface (wildtype) or contained mutations in the C_H1:C_L interface (see Table II for a complete list). The V_H:V_L interface was left unmutated in all cases. The antibodies were produced by transient transfection of (A) Expi293, (B) HEK293-6E or (C) ExpiCHO. The prevalence of correctly assembled bsAb is given in percent of all detected heterodimeric IgGs and represents a mean of two (A) or three (B and C) independent transfections (Table I).

Fig. 7

Biophysical and functional characterisation of B10v5 × hu225M with or without mutations in the C_H1:C_L interface. Analytical SEC after protein A purification of bsAbs produced in (A) HEK293-6E or (B) bsAbs produced in ExpiCHO after preparative SEC. The area under the curve in percent of all detected peak areas is given in the graph. (C) Simultaneous binding of both antigens cMET and EGFR using biolayer interferometry. BsAbs (wt, MaB40, MaB5/40, MaB21/40, MaB45/40) or antibodies consisting of only one Fab (oa for ‘one-armed’) were allowed to bind to cMET coated biosensors. Subsequently, biosensors were incubated with EGFR. (D) Differential scanning calorimetry of wildtype, interface mutants and ‘one-armed’ constructs (Table III). (E) Western blot analysis of the inhibition of EGFR and cMET phosphorylation in A549 cells by bsAbs with our without interface designs.

Fig. 8

Analysis of pairing of light to heavy chain in the bsAb B10v5 × huOKT3 by LC–ESI–MS, after protein A purification and deglycosylation. The bsAbs contained either no mutation in the Fab interface (wildtype) or contained mutations in the C_H1:C_L interface (see Table I for a complete list). The V_H:V_L interface was left unmutated in all cases. The antibodies were produced by transient transfection of HEK293-6E. The prevalence of correctly assembled bsAb is given in percent of all detected heterodimeric IgGs and represents a mean of three independent transfections (Table IV).