Design of orthogonal constant domain interfaces to aid proper heavy/light chain pairing of bispecific antibodies

Abstract

The correct pairing of cognate heavy and light chains is critical to the efficient manufacturing of IgG-like bispecific antibodies (bsAbs) from a single host cell. We present a general solution for the elimination of heavy chain (HC):light chain (LC) mispairs in bsAbs with $\kappa$ LCs via the use of two orthogonal constant domain (C_H1:C$\kappa$) interfaces comprising computationally designed amino acid substitutions. Substitutions were designed by Rosetta to introduce novel hydrogen bond (H-bond) networks at the C_H1:C$\kappa$ interface, followed by Rosetta energy calculations to identify designs with enhanced pairing specificity and interface stability. Our final design, featuring a total of 11 amino acid substitutions across two Fab constant regions, was tested on a set of six IgG-like bsAbs featuring a diverse set of unmodified human antibody variable domains. Purity assessments showed near-complete elimination of LC mispairs, including in cases with high baseline mispairing with wild-type constant domains. The engineered bsAbs broadly recapitulated the antigen-binding and biophysical developability properties of their monospecific counterparts and no adverse immunogenicity signal was identified by an in vitro assay. Fab crystal structures containing engineered constant domain interfaces revealed no major perturbations relative to the wild-type coordinates and validated the presence of the designed hydrogen bond interactions. Our work enables the facile assembly of independently discovered IgG-like bispecific antibodies in a single-cell host and demonstrates a streamlined and generalizable computational and experimental workflow for redesigning conserved protein:protein interfaces.

Keywords: Bispecific antibodies; antibody engineering; computational protein design; developability; heavy chain/light chain pairing; in vitro developability assay.

Figure 1.

Schematic summarizing the design and experimental process followed to obtain novel heavy chain/light chain constant domain pairing mutation sets. (a) Rosetta flex ddG scores (after 18,000 backrub steps) of the top five best single Fab interface designs (SIDs), representing the initial in silico scoring step. Larger predicted energetic gaps between the positive design state (red circle) and negative design states (blue circles) indicate a prediction of stronger ability to enforce correct pairing. The right side of the schematic depicts the experimental workflow used to quantify pairing. Panitumumab and ustekinumab variable domains are shaded brown and blue, respectively. IgG-like bispecifics that incorporate the top-scoring Rosetta designs in one of the Fab arms, and C_H3 knob-into-hole mutations for heavy-chain heterodimerization, were expressed in HEK-293 cells. Correct pairing was then assayed via two experimental methods: cation exchange chromatography on intact IgGs, and LC-MS quantification on IgGs digested into Fabs. (b) Rosetta flex ddG scores for double interface design’s (DIDs) that contain mutation sets installed in both Fab arms of the IgG-like bispecific.

Figure 2.

(a) Correct pairing in multiple variable region contexts, as measured by LC-MS quantification of Fab material digested from HEK-293-produced IgGs. Variable domains are assigned colors (shown in legend). Wild-type bispecific controls, with no C_H1:C$\mathit{\kappa }$ mutations, are depicted as squares. Data for DID bispecifics produced with the 1443 and 1993 designs are depicted as circles. The different split colorings for the circles represent alternate combinations of variable regions and constant domain mutation sets. (b) CEX main peak quantification of intact IgG-like bispecific containing variable domains from panitumumab and ustekinumab. Monospecific control samples arising from 2-chain transfections, with the same 1443 or 1993 set of mutations on both Fab arms, are depicted as triangles. (c) Representative CEX traces of panitumumab/ustekinumab bispecific IgG sample produced with corresponding constant region mutation sets 1443/1993, and Hole/Knob C_H3 mutations, along with an antibody schematic illustrating the domain sequences in the intended, correctly paired bispecific sample. Main peak quantification of this trace (96%) is represented as the circular point with the highest main peak quantification in panel (b). (d) CEX traces of the alternative variable domain/constant domain pairing shown in panel (c), with a C_H1:C$\mathit{\kappa }$ constant domain containing mutation set 1443 linked with an ustekinumab variable domain, and mutation set 1993 with panitumumab. Main peak quantification of this trace (88%) is also depicted in panel (b). (e) Reference CEX trace of a control produced with wild-type C_H1:C$\mathit{\kappa }$ domain sequences and wild-type C_H3 domains, with a main peak quantification of 49%. The antibody schematic represents the corresponding bispecific IgG with correctly paired heavy and light chains.

Figure 3.

Summary of high-throughput biophysical and developability characterization data. Variable domains are assigned colors (shown in legend). Monospecific IgGs containing one set of variable domain sequence are shown as solid colored circles. Bispecifics, consisting of unique variable domains, are shown as split circles, with each half of the circle indicating one of two paired variable domains present in the molecule. Most bispecific combinations of variable domains are present twice, indicating that they were produced in two separate orientations with 1443/1993 (e.g., 1443 paired with variable domain a in one point, or 1443 paired with variable domain B in the other). A schematic of example constructs is shown below panel C. (a) Size exclusion chromatography (SEC) purity, as measured by main peak percentage. (b) Hydrophobic interaction chromatography (HIC) column retention time. (c) Affinity-capture self-interaction nanoparticle spectroscopy (AC-SINS) (d) Fab T_m, as measured via differential scanning fluorimetry (DSF) assay. No boxplot is shown for monospecific T_m values for design 1443 due to lack of available data for sifalimumab.

Figure 4.

1443/1993 antibody pairing mutation sets do not confer additional risks for immunogenicity. a) Frequency of the top 20 most prevalent HLA-DR alleles among North American populations, averaged across gold-standard datasets on allelefrequencies.Net, and among the PBMC donor cohort selected for in vitro T cell stimulation assessments. b) Calculated stimulation index (SI) values for enrichment of CD134 and CD137 (CD134 ∪ CD137) or ICOS expression in CD4+ T cell populations following 2 or 7 days of co-culture with KLH, as assessed by flow cytometry. Statistical comparisons calculated by two-sided Friedman’s tests with Dunn’s correction for multiple comparisons. *, p < 0.05; **, p < 0.01; ***, p < 0.001. c,d) Calculated CD134 ∪ CD137 SI values among CD4+ T cells following 7 days of co-culture with either monospecific control IgGs with known clinical anti-drug antibody (ADA) response rates (c) or monospecific and bispecific IgGs based on palivizumab and bevacizumab bearing 1443 or 1993 antibody pairing mutation sets, alongside wild-type IgG controls (d). e,f) Calculated ICOS SI values among CD4+ T cells following 7 days of co-culture with either monospecific control IgGs with known clinical ADA response rates (e) or monospecific and bispecific IgGs based on palivizumab and bevacizumab bearing 1443 or 1993 antibody pairing mutation sets, alongside wild-type IgG controls (f).

Figure 5.

(a) Crystal structure of design 1443. (b) Crystal structure of network 1993. Hydrogen bonds (bonds and hydrogen atom placement determined in Rosetta via H-bond feature analyzer) are shown in both panels. Mutated side chains, or residues with side chains involved in hydrogen bonds with mutated residue side chains, are shown as sticks.