Robust production of monovalent bispecific IgG antibodies through novel electrostatic steering mutations at the C_H1-C_λ interface

ABSTRACT

Bispecific antibodies represent an increasingly large fraction of biologics in therapeutic development due to their expanded scope in functional capabilities. Asymmetric monovalent bispecific IgGs (bsIgGs) have the additional advantage of maintaining a native antibody-like structure, which can provide favorable pharmacology and pharmacokinetic profiles. The production of correctly assembled asymmetric monovalent bsIgGs, however, is a complex engineering endeavor due to the propensity for non-cognate heavy and light chains to mis-pair. Previously, we introduced the DuetMab platform as a general solution for the production of bsIgGs, which utilizes an engineered interchain disulfide bond in one of the C_H1-C_L domains to promote orthogonal chain pairing between heavy and light chains. While highly effective in promoting cognate heavy and light chain pairing, residual chain mispairing could be detected for specific combinations of Fv pairs. Here, we present enhancements to the DuetMab design that improve chain pairing and production through the introduction of novel electrostatic steering mutations at the C_H1-C_L interface with lambda light chains (C_H1-C_λ). These mutations work together with previously established charge-pair mutations at the C_H1-C_L interface with kappa light chains (C_H1-C_κ) and Fab disulfide engineering to promote cognate heavy and light chain pairing and enable the reliable production of bsIgGs. Importantly, these enhanced DuetMabs do not require engineering of the variable domains and are robust when applied to a panel of bsIgGs with diverse Fv sequences. We present a comprehensive biochemical, biophysical, and functional characterization of the resulting DuetMabs to demonstrate compatibility with industrial production benchmarks. Overall, this enhanced DuetMab platform substantially streamlines process development of these disruptive biotherapeutics.

KEYWORDS:

• Antibody
• bispecific
• chain
• charge
• electrostatic
• engineering
• DuetMab
• lambda
• pairing
• steering

Introduction

Over the past decade, bispecific antibodies have emerged as a leading class of biological therapeutics with extensive therapeutic potential. Unlike conventional monoclonal antibodies which bind a single target, bispecific antibodies are capable of engaging two distinct epitopes. This ostensibly simple change in binding capabilities has transformed the therapeutic landscape of antibodies due to the expanded functional capabilities that are accessible with bispecific antibodies. Simultaneous targeting of two independent disease markers offers improved therapeutic efficacy and greater resilience to escape mechanisms than mono-targeted therapies.^1,2 Additionally, dual-antigen avidity binding can enhance target selectivity on double-positive target cells over single-positive normal tissue, leading to reduced toxicity and improved therapeutic index.³ In addition to enhanced targeting capabilities, bispecific antibodies also provide access to an expanded set of mechanisms of action, including T-cell engagers, half-life extension, trans-membrane delivery, and targeted protein degradation.^4,5 These disruptive modalities are rapidly progressing and show exciting promise in many therapeutic areas, including cancer, autoimmunity, and metabolic diseases.⁶

Asymmetric monovalent bispecific IgG (bsIgG) therapeutics have the additional advantage of maintaining a native antibody-like structure, which provides favorable bioavailability and pharmacokinetic (PK) profiles and leverages the production expertise associated with conventional antibodies. The synthesis of bsIgGs, however, often requires the production of asymmetric molecules, which adds substantial complexity to the manufacturing process. A monovalent bispecific IgG, for example, is composed of two heavy chains (HCs) and two light chains (LCs), which can fold into a total of 10 different molecule combinations, only one of which contains the desired assembly of HCs and LCs.⁷ Several techniques have been developed to drive heterodimerization of the heavy chains during production, including the knobs-into-holes approach,⁸ electrostatic steering,⁹ and disulfide engineering.¹⁰ Alternative approaches include the use of Fab-arm exchange to produce the bispecific after production,¹¹ or bypassing the problem through the use of a common HC.¹² An additional challenge in asymmetric bsIgG production is the correct pairing of HCs and LCs. Although initially addressed through the use of a common LC,¹³ several protein engineering solutions have been developed to promote orthogonal chain pairing, including domain cross-over between C_H1 and C_L,¹⁴ engineering of the antigen-binding fragment (Fab) interface,^15,16 disulfide bond engineering,^17,18 and the introduction of charged residues in the HC-LC interface.^19,20

We have previously introduced the DuetMab platform for the production of monovalent bispecific IgG antibodies.^17,21,22 This approach combined knobs-into-holes Fc mutations with the introduction of an engineered disulfide at the C_H1-C_L interface to promote correct chain associations. Compatibility with both kappa (κ) and lambda (λ) LCs was demonstrated, and several of the presented DuetMabs contained Fabs with a combination of both LCs.

A persistent challenge in the development and application of bsIgGs is the significant influence that V_H and V_L sequences have on HC-LC pairing, which can overwhelm engineering methods that promote specific chain associations.^15,16,23 There thus remains a need for improvements in strategies for bsIgG production that build toward robust platforms that accommodate diverse Fv sequences. Additionally, the development of bispecifics that maintain the structural integrity and manufacturability of the native IgG remains a rate-limiting step for advancement of these biotherapeutics. Here, we present enhancements to the DuetMab platform that improve bsIgG pairing and production through the combination of disulfide engineering and electrostatic steering mutations to drive correct HC-LC pairing. Although electrostatic steering was explored and reported in the literature previously, no solution has been provided for antibodies with a lambda LC. Here, we expand the bispecific antibody toolbox by disclosing for the first time electrostatic steering mutations that promote cognate chain pairing for lambda LCs. Additionally, we also present a comprehensive biochemical, biophysical, and functional characterization of the resulting DuetMabs to demonstrate compatibility with industrial production benchmarks.

Results

Design of C_H1-C_λ interface variants for orthogonal Fab pairing

Previous engineering efforts by Dillon et al. to drive orthogonal HC-LC pairing identified the C_H1 S183K and C_L V133E mutations, which provide significant improvements to bispecific antibody production.¹⁹ However, electrostatic steering mutations have not yet been identified that promote correct chain pairing in the context of a C_H1-C_L interface with a lambda constant domain (C_λ). At the C_H1-C_λ interface, a tyrosine (Y178) is predicted to clash with the S183K-V133E interaction, whereas the corresponding threonine residue in kappa LCs is able to accommodate these mutations due to its smaller size (Figure 1). We sought to identify charge pair mutations that would maintain compatibility with lambda LCs. In an effort to identify engineering solutions that could serve as a robust platform that functions independently of Fv sequence, we restricted our search to pairs of amino acids at the C_H1-C_λ interface. To increase the likelihood that selected pairs have a direct interaction, only amino acids in which the Cβ atom were oriented toward the interface were selected (in the case of glycine residues, we required the carbonyl to be oriented away from interface). Additionally, we required that selected residues have sufficient space beyond the Cβ atom to introduce amino acids with longer side chains (i.e., arginine and glutamic acid). The last criteria for selection involved substituting amino acids at candidate positions and searching for possible steric clashes. Substituted residues included Arg, Lys, Asp, Glu, Ser, and Thr to introduce electrostatic interactions that may promote orthogonal chain pairing. Substitutions of these amino acids were explored in a pair-wise fashion. Candidate positions where substitutions could be made without steric clash for at least one combination of permitted Chi angles were selected for experimental evaluation. In total, these criteria identified the amino acid pairs and substitutions shown in Table 1.

Figure 1. Structural analysis of the C_H1-C_L interface with kappa and lambda LCs. Structural alignment of the engineered C_H1-C_κ interface (S183K C_H1, V133E C_κ; PDB ID 5TDN; gray) with the wild-type C_H1-C_λ complex (PDB ID 4LLD; C_H1, green; C_λ, red). In this alignment, the C_λ residue Y178 exhibits significant steric clash with the engineered C_H1 residue S183K.

Table 1. Proposed electrostatic steering mutations at the C_H1-C_λ interface.

Set	Cλ	CH1
#1	V134 (R,K)	L128 (D,E,S,T)
#2	V134 (R,K)	L145 (D,E,S,T)
#3	L136 (R,K)	V185 (D,E,S,T)
#4	T117 (R,K)	V185 (D,E,S,T)
#5	T117 (R,K)	A141 (D,E,S,T)
#6	F119 (R,K)	L128 (D,E,S,T)
#7	Y178 (R,K)	V173 (D,E,S,T)
#8	V134 (K)	S183 (D)

Charge pairs in the C_H1-C_λ interface promote correct heavy chain and light chain pairing

Paired mutations were evaluated in a model DuetMab bsIgG carrying the variable heavy (V_H) and variable light (V_L) domains of in-house anti-epidermal growth factor receptor (αEGFR) and anti-receptor tyrosine-protein kinase 2 (αHER2) mAbs. Specifically, the V_H and V_L genes of αEGFR were inserted into a human constant gamma-1 heavy chain (C_H1–C_H3) carrying the ‘hole’ mutations and a constant Kappa LC (C_κ), whereas the V_H and V_L genes of αHER2 were inserted into a human constant gamma-1 heavy chain carrying the ‘knob’ mutation and a constant Lambda LC (C_λ). The previously reported engineered V12 disulfide was incorporated into the ‘knob’ arm by mutation of F126C and S121C in the HC and LC, respectively, and mutation of the native disulfide cysteine residues to valine.¹⁷ For initial screening efforts, production of bsIgG was carried out in a high-throughput fashion at a scale of 3 mL. In this production, the “wild-type” version of this bsIgG (control 1) exhibited a correct HC-LC pairing of 70% (Figure 2, Table 2). Incorporation of C_H1 S183K and C_κ V133E mutations into the hole arm (control 2) improved chain pairing to 92%. We then assessed the chain pairing of bsIgGs containing both S183K/V133E in ‘hole’ arm and one of the 58 candidate mutation pairs in the ‘knob’ arm (Figure 2, Table 2). In this cohort, five charge-pair combinations exhibited improved chain pairing to values between 93% and 96%: variants 33, 34, 35, 36, and 41.

Figure 2. Screening C_H1-C_λ interface mutants for improved chain pairing properties. Bispecific αEGFR/αHER2 antibodies containing electrostatic steering mutations (Table 2) were generated via a small-scale, high-throughput production and purification platform. Relative abundances of kappa and lambda LCs in each antibody were measured by capillary electrophoresis, and the resulting stoichiometric ratios were used to calculate chain pairing percentages.

Table 2. Screening of candidate variants for improved chain pairing properties.

Sample #	αEGFR Fab CH1	αEGFR Fab Cκ	αHER2 Fab CH1*	αHER2 Fab Cλ*	Day 7 Titer (µg/mL)	Pairing (%)	Sample #	αEGFR Fab CH1	αEGFR Fab Cκ	αHER2 Fab CH1*	αHER2 Fab Cλ*	Day 7 Titer (µg/mL)	Pairing (%)
1	WT	WT	WT	WT	184	69.8	33	S183K	V133E	A141D	T117R	242	96.0
2	S183K	V133E	WT	WT	122	91.8	34	S183K	V133E	A141E	T117R	257	95.6
3	S183K	V133E	L128D	V134R	94	38.4	35	S183K	V133E	A141S	T117R	214	93.1
4	S183K	V133E	L128E	V134R	125	26.7	36	S183K	V133E	A141T	T117R	132	95.0
5	S183K	V133E	L128S	V134R	130	30.1	37	S183K	V133E	V185D	T117K	89	46.2
6	S183K	V133E	L128T	V134R	105	25.8	38	S183K	V133E	V185E	T117K	108	57.0
7	S183K	V133E	L145D	V134R	166	48.8	39	S183K	V133E	V185S	T117K	152	76.4
8	S183K	V133E	L145E	V134R	158	72.6	40	S183K	V133E	V185T	T117K	132	74.1
9	S183K	V133E	L145S	V134R	163	47.8	41	S183K	V133E	A141D	T117K	168	93.5
10	S183K	V133E	L145T	V134R	129	37.5	42	S183K	V133E	A141E	T117K	177	90.8
11	S183K	V133E	L128D	V134K	128	37.2	43	S183K	V133E	A141S	T117K	167	80.4
12	S183K	V133E	L128E	V134K	108	27.9	44	S183K	V133E	A141T	T117K	120	82.2
13	S183K	V133E	L128S	V134K	138	30.6	45	S183K	V133E	L128D	F119R	79	56.4
14	S183K	V133E	L128T	V134K	149	35.8	46	S183K	V133E	L128E	F119R	100	18.5
15	S183K	V133E	L145D	V134K	158	33.3	47	S183K	V133E	L128S	F119R	100	28.0
16	S183K	V133E	L145E	V134K	156	45.6	48	S183K	V133E	L128T	F119R	87	28.8
17	S183K	V133E	L145S	V134K	194	21.6	49	S183K	V133E	L128D	F119K	48	56.7
18	S183K	V133E	L145T	V134K	131	21.9	50	S183K	V133E	L128E	F119K	106	43.0
19	S183K	V133E	WT	V134K	159	12.2	51	S183K	V133E	L128S	F119K	92	41.4
20	S183K	V133E	S183D	V134K	131	25.8	52	S183K	V133E	L128T	F119K	67	39.5
21	S183K	V133E	V185D	L136R	169	21.0	53	S183K	V133E	V173D	Y178R	95	46.7
22	S183K	V133E	V185E	L136R	167	8.6	54	S183K	V133E	V173E	Y178R	90	16.2
23	S183K	V133E	V185S	L136R	138	9.7	55	S183K	V133E	V173S	Y178R	106	21.5
24	S183K	V133E	V185T	L136R	149	11.9	56	S183K	V133E	V173T	Y178R	95	13.5
25	S183K	V133E	V185D	L136K	129	38.5	57	S183K	V133E	V173D	Y178K	139	10.8
26	S183K	V133E	V185E	L136K	129	32.7	58	S183K	V133E	V173E	Y178K	94	74.3
27	S183K	V133E	V185S	L136K	151	28.2	59	S183K	V133E	V173S	Y178K	130	12.9
28	S183K	V133E	V185T	L136K	144	27.4	60	S183K	V133E	V173T	Y178K	105	12.1
29	S183K	V133E	V185D	T117R	130	53.2
30	S183K	V133E	V185E	T117R	137	77.0
31	S183K	V133E	V185S	T117R	151	80.8
32	S183K	V133E	V185T	T117R	144	79.5

*αHER2 Fabs included the engineered V12 disulfide.

Liquid chromatography-mass spectrometry (LCMS) analysis of the intact, non-reduced antibodies identified the prominent mis-paired species in variant 1 as a heterodimer with two EGFR LCs (Supplementary Figure S1). Notably, this results in antibodies in which the EGFR LC containing the native C220 residue mis-pairs with the HER2 HC containing the engineered F126C residue. This pairing does not result in disulfide bond formation between the HC and LC, as evidenced by dissociation of the EGFR LC during mass analysis.

To evaluate potential effects on production, the selected charge-pair variants with improved pairing and control molecules were expressed at 100 mL scale. All DuetMabs exhibited similar titers with the exception of variant #41, which was reduced 2.5-fold (Table 3). Chain pairing at the 100 mL scale was consistent with the 3 mL scale productions and showed an average difference of 3 ± 2%. Overall, C_H1-C_λ charge-pair combinations improved HC-LC pairing compared to S183K/V133E alone, with variants 33, 34, 36, and 41 exhibiting >96% correct pairing after protein A purification. Secondary affinity purification and preparative size-exclusion chromatography (SEC) were used to remove mis-paired bsIgGs and aggregates in preparation for subsequent biochemical characterization. Bispecific IgG complexes were confirmed by subunit mass spectrometry analysis of the Fab-digested fragments, which identified masses consistent with the correctly paired bispecific antibodies (Supplementary Figure S2, Supplementary Table S1).

Table 3. Production and chain pairing profiles of engineered DuetMab variants.

Sample #	αEGFR Fab CH1	αEGFR Fab Cκ	αHER2 Fab CH1	αHER2 Fab Cλ	Titer day 14 (μg/mL)	Post-ProA pairing (%)	Post-ProA monomer (%)	Final pairing (%)	Final monomer (%)
1	WT	WT	WT	WT	361	63	92	91	>99
2	S183K	V133E	WT	WT	317	92	92	94	>99
33	S183K	V133E	A141D	T117R	291	99	94	>99	>99
34	S183K	V133E	A141E	T117R	343	97	92	>99	>99
35	S183K	V133E	A141S	T117R	383	90	94	95	>99
36	S183K	V133E	A141T	T117R	389	100	93	94	>99
41	S183K	V133E	A141D	T117K	104*	96	93	99	>99

*Sample 41 was generated in a separate production and exhibited a 2.5-fold reduction in titer compared to Sample 1.

DuetMabs with charge pair variants have similar physicochemical properties

A crucial feature for the general application of electrostatic steering mutations in bsIgG production is that they have minimal impacts on the native physicochemical and biological properties of the antibody. To determine whether the C_H1-C_λ charge pair mutations altered antibody function or structure, we performed a series of biochemical and biophysical assessments on the selected charge pair variants.

We assessed the effects of C_H1-C_λ mutations on the binding affinity of the αHER2 arm by biolayer interferometry. DuetMabs without charge mutations in the αHER2 Fab (controls 1 and 2) had K_D values of 0.35 and 0.51 nM, respectively, whereas variants 33, 34, 35, 36, and 41 had similar kinetic values from 0.24 to 0.49 nM (Supplementary Figure S3 and Supplementary Table S2). Differential scanning calorimetry (DSC) was used to measure the thermal stability of the DuetMab antibodies. Thermograms of control 2 and the selected variants exhibited highly similar profiles (Figure 3, Table 4). Furthermore, the T_onset value of control 2 was 48.4°C, whereas selected variants had values of 48.3–52.5°C. Differential scanning fluorimetry (DSF) further identified no significant difference in T_onset or T_m values between electrostatic steering mutants and control molecules (Supplementary Table S3). Aggregation and fragmentation propensity of the variants was further assessed by an accelerated stability heat stress study wherein SEC was used to monitor the levels of aggregates and fragments before and after incubation at 45°C for 14 days. No significant increase in aggregation was observed for any of the DuetMab variants, and only variant 35 exhibited a low-level increase in fragmentation (Supplementary Table S3). Additionally, all variants exhibited SEC retention times consistent with that of an IgG1 control. Collectively, these results provide evidence that the incorporation of the selected electrostatic steering mutations into the C_H1-C_λ interface had minimal effects on the physicochemical properties of the antibody.

Figure 3. Differential scanning calorimetry on engineered DuetMabs. Thermograms show no difference between control 2 and the engineered variants. Differences between control 1 and control 2 are likely due to the S183K and V133E mutations in the C_H1 and C_κ domains, respectively.

Table 4. Differential scanning calorimetry on engineered DuetMab variants.

Sample #	αEGFR Fab (CH1/Cκ)	αHER2 Fab (CH1/Cλ)	TM1 (°C)	TM2 (°C)	TM3 (°C)	TM4 (°C)	Tonset (°C ± sd)
1	WT	WT	67.6	68.5	72.1	79.8	49.5 ± 0.7
2	S183K/V133E	WT	64.8	68.7	NA	79.3	48.4 ± 0.3
33	S183K/V133E	A141D/T117R	64.5	68.8	NA	79.2	49.3 ± 1.4
34	S183K/V133E	A141E/T117R	63.7	68.7	NA	79.4	50.3 ± 1.1
35	S183K/V133E	A141S/T117R	64.0	68.7	NA	79.4	48.3 ± 0.5
36	S183K/V133E	A141T/T117R	64.5	68.7	NA	79.4	52.5 ± 0.3
41	S183K/V133E	A141D/T117K	64.5	68.7	NA	79.3	50.8 ± 0.3

DuetMabs with charge pair variants have similar biological and pharmacokinetic properties

Anti-EGFR antibodies have been shown to inhibit EGFR ligand binding, EGFR/HER2 heterodimerization, downstream signaling, and cell proliferation.²⁴ The cytotoxic activity of the engineered DuetMab variants was assessed using a viability assay with a non-small cell lung cancer cell line, NCI H358, which expresses both EGFR and HER2 on the cell surface. Half-maximal effective concentration (EC50) values of control 1 and 2 for inducing cell cytotoxicity were 0.57 and 1.08 nM, respectively, and EC50 values of the selected variants all fell within a similar range from 0.40 to 0.81 nM (Figure 4). Maximum killing values of the controls and variants were also comparable and ranged between 45% and 48%.

Figure 4. Biological activity of engineered DuetMabs. Controls and select engineered variants were assayed for cytotoxic activity on NCI H358 cells. Each point represents the mean values of triplicate wells and error bars represent ± the standard error of the mean (SEM). R347 is an isotype control antibody.

The PK profile of control 1 and variants 33, 34, and 41 was evaluated in human FcRn transgenic Tg32 mice. Mice were dosed intravenously at 5 mg/kg and the PK properties were assessed over 35 days. Calculated antibody half-lives were ~180 hours and estimates in human as determined by allometric scaling from mouse PK data were ~17 days (Table 5, Supplementary Figure S4). Variant 34 exhibited slightly faster clearance and a reduced half-life by ~1.3-fold. Altogether, these results demonstrate that electrostatic charge pair variants preserve the biological and PK properties of the parent antibodies.

Table 5. Pharmacokinetic properties of engineered DuetMab variants.

Sample #	Cmax (µg/mL)	CL (mL/h/kg)	Vss (mL/kg)	t1/2 (hr)	Human t1/2 (days)
1	296.44	0.22	50.03	184.82	17.03
33	234.28	0.23	55.26	182.64	16.83
34	269.02	0.30	53.05	137.62	12.68
41	218.14	0.21	46.46	180.21	16.61

Structural investigation of charge pair mutations in the C_H1-C_λ interface

To determine the molecular implications of electrostatic steering mutations at the C_H1-C_λ interface, X-ray crystallographic structures were determined for two Fab molecules containing A141D/T117R and A141E/T117R (variants 33 and 34), which showed excellent pairing and developability profiles. Crystals of these Fabs diffracted to resolutions of 2.1 Å and 2.0 Å, respectively (Supplementary Table S4). The resulting refined structures show close alignment with the wild-type C_H1-C_λ structure (RMSD < 0.7 Å) and determined that the mutated amino acids engage in a salt bridge interaction (Figure 5). These residues are in a buried position in the C_H1-C_λ interface and have calculated solvent-accessible surface areas of 15–29%. Any energetic penalty due to the desolvation of these charged residues appears to be compensated by the favorable interactions of the protein–protein interface.²⁵ The observed improvements in HC-LC pairing likely arise due to a combination of the favorable interactions of the engineered salt bridge, as well as unfavorable interactions that may form during chain mispairing where unsatisfied hydrogen bond donors and acceptors may be present.

Figure 5. Crystallographic analysis of engineered DuetMabs. (a) Crystal structure of the Fab fragment in engineered variant #33 (C_H1 A141D, C_λ T117R, PDB ID 8TJF) and (b) engineered variant #34 (C_H1 A141E, C_λ T117R, PDB ID 8TI4) in structural alignment with the wild-type IgG1 C_H1:C_λ complex (PDB ID 4LLD). RMSD values in these alignments were 0.669 Å and 0.577 Å, respectively. Dotted lines highlight hydrogen bonds present in engineered salt bridge interactions (A141D/T117R, 2.6 Å; A141E/T117R, 3.1 Å).

Engineered variants exhibit robust improvements in chain pairing across multiple antibody pair combinations

Chain-pairing in bsIgGs has been demonstrated to exhibit significant Fv dependence.^15,16,23 To determine whether the observed improvements in chain pairing from the selected variants extend to other antibody sequences, we designed four additional model DuetMab bsIgGs: αEGFR(clone A, C_κ)-αCD3(clone A, C_λ), αPD1(C_κ)-ISO(isotype control, C_λ), αSTEAP2(C_κ)-αCD3(clone B, C_λ), and αCMET(C_κ)-αEGFR(clone B, C_λ). Notably, although the charge pair S183K/V133E in C_H1/C_κ alone exhibited >95% correct chain pairing for αSTEAP2-αCD3, this value was reduced to 66–83% in the remaining three DuetMabs. In contrast, the charge variants significantly improved pairing ratios independent of Fv sequence (Figure 6, Supplementary Table S5). In particular, variants #33, 34, and 41 exhibited excellent chain pair ratios ranging from 90% to 100% for all five DuetMab designs. These data suggest that the engineered charge-pair variants show consistent improvement in chain pairing across several Fv sequences.

Figure 6. Chain pairing of engineered DuetMabs with diverse Fv sequences. Controls and selected charge pair variants (33, 34, 35, 36, and 41) were incorporated into a total of five DuetMabs with different Fv sequence combinations. Chain pairing is displayed after protein A as determined by kappa and lambda light chain stoichiometry.

Discussion

The efficient and reliable production of bispecific IgG antibodies presents an engineering challenge that requires delicate balance. On one hand, natural antibody sequences exhibit promiscuous binding when co-expressed with other antibody chains, and protein engineering strategies are needed to drive the correct assembly of complex bispecific antibodies. On the other hand, each departure from the native antibody sequence can be a liability due to potentially unfavorable effects on production, stability, or immunogenicity. Indeed, the therapeutic field is trending toward more native IgG-like structures that can leverage the efficient production techniques of conventional mAbs while limiting developability risks. A further complication is that HC-LC pairing properties exhibit significant Fv dependence, and solutions for one bsIgG may have mixed success when applied to other Fv sequences (Figure 6, Supplementary Table S5).^15,16,23 The ideal engineering solutions for bsIgG production, therefore, are those that can efficiently achieve cognate chain pairing across a variety of bsIgGs with minimal alteration to native sequences.

Previous studies have shown that the incorporation of pairs of charged residues into the C_H1-C_L interface of bsIgGs promotes orthogonal chain pairing.^19,20 This work introduced substantial contributions to the antibody engineering toolkit for bsIgG production, but the efficacy of this strategy has only been established for the Kappa C_L domain (C_κ) and has not been demonstrated for Lambda C_L domains (C_λ). A possible explanation may be due to the presence of a tyrosine in the C_H1-C_λ interface that is predicted to clash with the C_H1 S183K mutation (Figure 1).

To address this unmet need, we sought to identify electrostatic steering mutations at the C_H1-C_λ interface that promote cognate chain pairing and exhibit compatibility with the DuetMab platform for bsIgG production. DuetMab antibodies consist of four independent chains and drive cognate HC-LC pairing by the substitution of the native C_H1-C_L disulfide bond on one Fab arm with an engineered V12 disulfide.^17,21,22 We used a structure-guided approach to identify 58 candidate pairs of electrostatic steering mutations at the C_H1-C_λ interface (Tables 1 and 2). In order to ensure compatibility with the V12 disulfide as well as previously established C_H1-C_κ charge pair mutations, we incorporated candidate mutation pairs into a model DuetMab targeting EGFR and HER2. The EGFR Fab arm contained the native HC-LC disulfide and the C_H1 S183K and C_κ V133E mutations, while the HER2 Fab arm contained the engineered V12 disulfide and the candidate C_H1-C_λ electrostatic steering mutations. Generation and subsequent screening of these candidates identified a select set of variants with highly specific chain pairing properties: C_H1/C_λ mutations A141D/T117R, A141E/T117R, A141S/T117R, A141T/T117R, and A141D/T117K (Figure 2 and Table 3). Subsequent analysis of these DuetMabs demonstrated that these mutations largely preserve the production, physicochemical, and biological properties of the parent antibodies.

Structural characterization of Fabs containing the C_H1/C_λ mutations A141D/T117R and A141E/T117R confirmed the presence of salt bridge interactions between the mutated interface residues (Figure 5). The energetic favorability of these interactions may directly promote orthogonal HC-LC pairing. Additionally, potential unfavorable interactions that may occur at the C_H1-C_λ interface in the context of mis-paired chains could also contribute to correct pairing. The extent to which the observed improvements in chain pairing are due to one or a combination of these phenomena is unknown.

In summary, we introduced novel electrostatic steering mutations at the C_H1-C_λ interface that, together with the engineered V12 disulfide, provide substantial improvements to chain pairing and preserve favorable antibody developability profiles. Importantly, this enhanced DuetMab design does not require any engineering to Fv sequences and is robust when applied to a variety of bsIgG combinations (Figure 6). We anticipate that this improved DuetMab design will serve as a platform for the reliable production of bsIgGs, as well as the expansion into more complex multispecific antibody designs with enhanced functional capabilities.

Materials and methods

Expression and purification of bispecific antibodies

Bispecific antibodies were produced using two bicistronic mammalian expression vectors: one plasmid containing two heavy chains (pDuet-Heavy) and a second containing two LCs (pDuet-Light) similar to as described previously.¹⁷ Briefly, the first operon of pDuet-Heavy contained a heavy chain with ‘hole’ (T366S, L368A, and Y407V), heterodimer disulfide (Y349C), and RF (H435R/Y436F) mutations in the C_H3 domain, while the second operon contained the ‘knob’ (T366W) and heterodimer disulfide (S354C) mutations in C_H3.^13,26 Additional V12 disulfide (F126C, C220V) and/or electrostatic steering mutations were incorporated into C_H1 domains for molecules as listed in Table 2.¹⁷ The first operon of the pDuet-Light vector contained LCs with a constant kappa (C_κ) domain, while the second operon contained LCs with a constant lambda (C_λ) domain and the V12 disulfide mutations (S121C, C212V). Additional electrostatic steering mutations were incorporated into C_κ and C_λ domains for molecules as listed in Table 2. Antibody amino acid sequences are available in Supplementary Table S6.

Antibodies were produced by transient co-transfection of Chinese hamster ovary (CHO) cells with pDuet-Heavy and pDuet-Light expression vectors using PEI-MAX (Polysciences) as a transfection reagent. CHO cells were maintained using a proprietary culture medium. Cell supernatants were collected 7 and 13 days after transfection and filtered through a 0.22 μm sterile filter in preparation for titer measurements. Antibody concentrations were determined using protein A sensors (Sartorius) on an Octet384 instrument using the manufacturer’s recommended protocols. Antibodies were purified by either protein A magnetic bead affinity purification (GenScript) or standard protein A affinity chromatography (Cytiva), followed by LC affinity chromatography if necessary, in accordance with the manufacturer’s protocol, and were subsequently buffer exchanged into phosphate-buffered saline (PBS; pH 7.2).

The purity and oligomeric state of purified molecules were assessed by microfluidics-based electrophoresis and analytical SEC. Microfluidics-based electrophoresis was performed using a 2100 Bioanalyzer system (Agilent) in accordance with the manufacturer’s protocols. In this analysis, we observed that kappa and lambda LCs, despite being similar molecular weight, are efficiently resolved, possibly due to differences in amino acid composition between the C_κ and C_λ domains which may contribute to different CE-SDS mobilities. Relative abundances of antibody kappa and lambda LCs were used to calculate the percentage of correctly paired antibody.

Analytical SEC-HPLC was performed on an Agilent 1260 Infinity HPLC system using a TSK-gel G3000SWxL column (Tosoh Biosciences). Preparative SEC-HPLC was carried out using a Superdex 200 column (Cytiva) to remove protein aggregates. Concentrations of purified antibodies were determined by absorbance measurements at 280 nm.

Biolayer interferometry

Binding measurements were made using an Octet RED384 instrument (Forte Bio). Briefly, streptavidin (SA) biosensors were loaded with biotinylated HER2 protein (ACRO Biosystems) in PBS (pH 7.2) with 1 mg/ml bovine serum albumin (BSA) and 0.05% (v/v) Tween. Loaded biosensors were washed in the same buffer before carrying out association and dissociation measurements with various antibodies at the indicated times. Kinetic parameters were determined using a non-linear fit of the data with the Octet software (v.12.2.1.24).

Mass spectrometry

Proteins were deglycosylated using PNGase F (Promega) before intact mass analysis or digested using IgdE (Genovis) before sub-unit mass analysis. The protein mass was acquired by a UPLC coupled to a Xevo G2-XS QTof mass spectrometer (Waters). A BEH C4 column (300 Å, 1.7 µm, 2.1 × 50 mm; Waters) was used for the reversed-phase separation. Mobile phases A and B were 0.02% TFA in water and in acetonitrile, respectively. Mass spectra were collected at a mass-to-charge ratio range of 800–4,500. Molecular mass was determined by deconvolution using the MaxEnt I algorithm within Waters MassLynx.

High performance-SEC

Antibody samples were analyzed using HP-SEC to determine levels of aggregate, monomer, and fragment. Samples (100 μg in PBS buffer) were injected on an Agilent 1200 series high-performance liquid chromatography (HPLC) instrument and separated using a TSKgel G3000SWxl size-exclusion column (Tosoh Bioscience #08541). The mobile phase was 100 mM sodium phosphate (pH 6.8), and sample flow rate was 1 mL/min. Ultraviolet (UV) detection was performed at 280 nm.

Heat stress accelerated stability

Samples were diluted to 1 mg/mL in PBS (pH 7.2) and incubated for 2 weeks at either 4°C or 45°C. Samples were analyzed by HP-SEC to determine monomer, aggregate, and fragment percentages based on curve integration using the HPLC ChemStation software (Agilent). The change in monomer, aggregate, and fragment content was calculated from the difference between each sample incubated at 45°C versus 4°C.

Differential scanning fluorimetry

DSF measurements were performed using a previously established method with minor modifications.²⁷ Samples were prepared by combining 20 µL of protein sample at 1 mg/mL in PBS (pH 7.2) with 5 µL of SYPRO Orange dye (Invitrogen S-6651) diluted to 40X in PBS (pH 7.2) in a 96-well PCR plate in duplicate. The plate was sealed and measurements performed in a QuantStudio 7 Flex Real-Time PCR System (Applied Biosystems). Samples were subjected to an initial equilibration step at 25°C for 2 min, followed by a temperature ramp to 99°C at 0.05°C/sec increments. The fluorescence emission was monitored using the FAM filter set. The T_m value for each sample was calculated in the Protein Thermal Shift software (Applied Biosystems) using the Boltzmann method, and T_onset values were approximated from these curves.

Cell viability assays

Cell viability was determined using the CellTiter-Glo™ Luminescent Cell Viability Assay (Promega), which measures ATP abundance as an indicator for metabolically active cells. In brief, NCI H358 cells were seeded in 96-well plates at a density of ~1 × 10⁴ cells/well in RPMI 1640 media supplemented with 0.1% BSA, and 0.2 ng/mL human recombinant EGF. Antibodies at various concentrations were added to samples in triplicate, and cells were incubated for 72 hr at 37°C and 5% CO₂ in a humidified incubator. After treatment, the cells were exposed to CellTiter-Glo® reagent (Promega) for ~15 min and OD409 was measured using an EnVision 2104 Multilabel plate reader (PerkinElmer). Cell viability was determined by comparing ATP signal relative to that of control cells with no antibody treatment.

Expression and purification of antibody Fabs

Fabs were transiently expressed in a suspension of human embryonic kidney (HEK) 293 cells, using 293fectin Transfection Reagent (Life Technologies) following the manufacturer’s protocols. Cells were grown in FreeStyle 293-F Expression Medium (Life Technologies) for 10 days and fed with a proprietary cell feed solution. Cells were harvested, and the supernatant was filtered through a 0.2 µM filter prior to purification. Fabs were purified using a 5 ml CaptureSelect CH1-XL column (Thermo Fisher Scientific), dialyzed against 25 mM Hepes (pH 7) and further purified using a 5 ml HiTrap SP HP cation exchange column (Cytiva) in a NaCl gradient.

Crystallization, harvesting, and X-ray diffraction

Fab fragments were purified using a Superdex 200 Increase 10/300 GL column (Cytiva) pre-equilibrated with 25 mM HEPES, pH 7.5, and 100 mM NaCl to ensure homogeneity of the samples. Initial crystallization trials were carried out by the sitting-drop vapor-diffusion method at 20°C. Crystallization drops were dispensed in 96-well crystallization plates (Intelli-Plate 102-0001-20, Art Robbins Instruments) using a Phoenix crystallization robot (Art Robbins Instruments) and commercially available crystallization screens. The drops were composed of equal volumes of protein and reservoir buffer.

Crystals were harvested from sitting drop plates from the following crystallization solutions: for A141E/T117R, 0.1 M Bis-Tris at pH 6.5, 25% w/v PEG 3350, and a protein concentration of 18.4 mg/ml; for A141D/T117R, 200 mM sodium chloride, 0.1 M Bis-Tris at pH 5.5, 25% w/v PEG 3350, and a protein concentration of 9 mg/ml. Harvested crystals were flash-cooled in liquid nitrogen and diffraction experiments were performed on a beamline B14–1 at Stanford Synchrotron Radiation Lightsource at 100 K. Diffraction data were collected from a single crystal for each Fab and were processed, integrated, and scaled with the XDS software package.²⁸

Structures of both Fab molecules were determined using molecular replacement method with the MolRep program²⁹ from the CCP4³⁰ suite of crystallographic software. Model building was performed using Coot³¹, and Refmac5³² was used for refinement. Experimental structure factors and corresponding refined models for C_H1 A141D:C_λ T117R and A141E:C_λ T117R Fab structures were deposited with the RCSB PDB with accession codes 8TJF and 8TI4, respectively.³³

Pharmacokinetic analysis

Female Tg32 mice with human neonatal Fc receptor (FcRn) between 8 and 12 weeks-old (Jackson Laboratory), were injected with antibodies intravenously via tail vein at a dose of 5 mg/kg into 6 animals per group. Serial submandibular bleeds were obtained from mice at 1 hr, 6 hr, 24 hr, 2 d, 4 d, 7 d, 11 d, 14 d, 21 d and 28 d. Terminal bleeds were taken with cardiac puncture on day 36. After 30 min at room temperature, blood samples were centrifuged at 2500 rpm for 15 min and sera were collected for analyses. All PK studies were approved by the Institutional Animal Care and Use Committee at AstraZeneca.

The concentration of antibodies in mouse serum samples was determined using a universal ELISA assay. The method procedure was a stepwise format in which wash steps followed each incubation step. Microtiter plates were coated with a sheep anti-human IgG antibody. A standard curve of each antibody was prepared in 10% pooled naïve mouse serum. Three levels (high, medium, and low) of quality control (QC) samples were prepared in 100% pooled naïve mouse serum. Following a blocking step with Casein buffer, standard curve calibrators, QCs, and samples diluted to the method minimum required dilution (MRD) of 1:10 were added to the microtiter plates and incubated for 1 h with shaking at room temperature. After a wash, HRP-labeled goat anti-human IgG antibody was added to the plate and incubated for 1 hr with shaking. Reactions were stopped with acid, and the plate was read on a spectrophotometer at 450 nm. Data were analyzed with SoftMax® Pro (SMP), version 7.1. The standard curve was established using a 4-parameter logistical curve fit model without weighting. The quantitative range of this assay was 15.625 ng/mL (lower limit of quantitation [LLOQ]) to 2000 ng/mL (upper limit of quantitation [ULOQ]). PK parameters were estimated via non-compartmental analysis using the software Phoenix WinNonlin version 8.2 (Certera).

Abbreviations

Standard 3-letter and 1-letter amino acid abbreviations are used throughout this manuscript. Antibody amino acids are numbered according to the Eu convention.
bsIgG	=	Bispecific immunoglobulin G
CH1	=	Immunoglobulin constant domain 1
CH3	=	Immunoglobulin constant domain 3
CL	=	Constant light chain domain
Cλ	=	Lambda constant light chain domain
Cκ	=	Kappa constant light chain domain
CD3	=	Cluster of differentiation 3
CMET	=	Tyrosine-protein kinase Met
DSC	=	Differential scanning calorimetry
DSF	=	Differential scanning fluorimetry
EC50	=	Half-maximal effective concentration
EGFR	=	Epidermal growth factor receptor
Fab	=	Antigen-binding fragment
FcRn	=	Neonatal crystallizable fragment receptor
Fv	=	Variable fragment
HC	=	Heavy chain
HEK	=	Human embryonic kidney
HER2	=	Receptor tyrosine protein kinase 2
HPLC	=	High-performance liquid chromatography
ISO	=	Isotype control
LC	=	Light chain
LCMS	=	Liquid chromatography-mass spectrometry
LLOQ	=	Lower limit of quantitation
MRD	=	Minimum required dilution
PD1	=	Programmed cell death protein 1 receptor
PK	=	Pharmacokinetic
SA	=	Streptavidin
SEC	=	Size exclusion chromatography
SEM	=	Standard error of the mean
TFA	=	Trifluoroacetic acid
ULOQ	=	Upper limit of quantification
QC	=	Quality control

Disclosure statement

The authors were employed by AstraZeneca at the time this work was performed.

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/19420862.2023.2273449

Additional information

Funding

Use of the Stanford Synchrotron Radiation Lightsource, SLAC National Accelerator Laboratory, is supported by the U.S. Department of Energy, Office of Science, Office of Basic Energy Sciences under Contract No. DE-AC02-76SF00515. The SSRL Structural Molecular Biology Program is supported by the DOE Office of Biological and Environmental Research, and by the National Institutes of Health, National Institute of General Medical Sciences (including P41GM103393).