Bispecific Antibody Manufacturing: Expression, Assembly, Purification, and Analytical Challenges

What Are Bispecific Antibodies?

Bispecific antibody manufacturing is the production of engineered antibodies that bind two different epitopes or antigens simultaneously, enabling therapeutic mechanisms impossible with conventional monoclonal antibodies. The defining challenge is assembling two distinct binding arms into a single molecule with high fidelity while keeping misassembled species below regulatory limits.

A bispecific antibody (BsAb) is a protein therapeutic containing two different antigen-binding sites within one molecule. Unlike standard IgG monoclonal antibodies, which bind a single target with two identical Fab arms, bispecific antibodies engage two targets at once. This dual specificity enables T-cell redirection (engaging CD3 on T cells while binding a tumour antigen), receptor co-engagement, and bridging of two signalling pathways.

As of mid-2026, 15 bispecific antibodies have received FDA approval. The pace of approvals has accelerated sharply: 9 of those 15 were approved between 2022 and 2025, reflecting both improved manufacturing platforms and expanding clinical utility. Approved molecules span T-cell engagers (blinatumomab, teclistamab, mosunetuzumab, glofitamab, epcoritamab, talquetamab, linvoseltamab), factor-bridging (emicizumab for haemophilia A), and receptor co-engagement (amivantamab for NSCLC). For the platform-level comparison of where bispecifics sit against mAbs and ADCs, see the mAb vs bispecific vs ADC manufacturing comparison.

The manufacturing complexity of bispecific antibodies arises from the need to express, correctly pair, and purify a molecule assembled from three or four distinct polypeptide chains, compared to the two identical chains of a standard IgG. This article covers the dominant IgG-like formats, their expression and assembly strategies, the purification trains required to remove format-specific impurities, and the orthogonal analytical panels needed for characterisation and release.

Bispecific Antibody Formats and Assembly Strategies

IgG-like bispecific antibodies use Fc engineering to force two different heavy chains into a heterodimer, and additional strategies to ensure each light chain pairs with its cognate heavy chain. Five formats dominate commercial and clinical manufacturing.

Knobs-into-holes (KIH) is the most widely adopted heavy chain heterodimerization strategy. Introduced by Ridgway, Presta, and Carter in 1996, it introduces a bulky tryptophan residue (T366W, the "knob") in the CH3 domain of one heavy chain and complementary small residues (T366S, L368A, Y407V, the "hole") on the other. This steric complementarity drives heavy chain heterodimerization to 92-95% while disfavouring both knob-knob and hole-hole homodimers. KIH is used in emicizumab (Hemlibra) and amivantamab (Rybrevant).

CrossMab, developed by Roche/Genentech, combines KIH with a domain crossover in one Fab arm. By swapping the CH1 and CL domains of one half-antibody, each light chain becomes structurally incompatible with the wrong heavy chain, reducing light chain mispairing from 5-15% to below 1%. Emicizumab uses the CrossMab_CH1-CL variant.

Common light chain formats eliminate light chain mispairing entirely by engineering both Fab arms to use the same light chain with different heavy chain variable domains. Teclistamab (Tecvayli) and other Janssen bispecifics use this approach. The trade-off is that the shared light chain must retain adequate affinity for both targets, which constrains the antibody discovery campaign.

DuoBody (controlled Fab-arm exchange), developed by Genmab, takes a fundamentally different approach: two standard homodimeric IgG1 antibodies are produced separately at full mAb titers (5-10 g/L each), then combined under mild reducing conditions (2-mercaptoethylamine at 31°C for 5 hours) to exchange half-molecules. CH3 mutations (F405L on one parent, K409R on the other) ensure that only heterodimeric recombination is thermodynamically stable, driving exchange to greater than 95% heterodimer.

Expression Systems and Chain Pairing

CHO cells are the dominant expression host for IgG-like bispecific antibodies, using the same fed-batch platform as conventional mAb production but with modifications for multi-chain expression. The core challenge is achieving balanced co-expression of three (common LC) or four (two-LC formats) distinct polypeptide chains from a single cell.

Three expression strategies are used in practice:

• Single-cell, multi-gene expression: All chains are encoded on one or two expression vectors within a single CHO cell line. Promoter strength, gene copy number, and codon usage are tuned to balance chain ratios. This is the most common approach for KIH and CrossMab formats. Typical titers range from 0.5 to 2 g/L for four-chain formats and 1 to 3 g/L for common light chain (three-chain) formats.
• Two-cell co-culture: Each half-antibody is expressed in a separate cell line and secreted as a half-molecule. The two half-molecules are combined in vitro under mild reducing conditions (as in DuoBody/cFAE). This avoids intracellular mispairing entirely and allows each cell line to be optimised independently.
• Separate production + in vitro assembly: Two parental homodimeric antibodies are produced in independent manufacturing campaigns, then combined for Fab-arm exchange. Each parent achieves standard mAb titers (5-10 g/L). The two-step process doubles the upstream campaigns but eliminates all chain-pairing issues.

Chain ratio imbalance is the primary upstream failure mode. If one heavy chain is over-expressed relative to the other, the excess forms homodimer rather than waiting for its heterodimeric partner. The CHO cell does not "know" which chains should pair; it assembles whatever is available in the ER. Promoter engineering (e.g., weaker promoter for the over-expressed chain), internal ribosome entry sites (IRES), and 2A self-cleaving peptides are used to fine-tune ratios. Most development programmes screen 200-500 clones to find lines with balanced expression and acceptable titer.

How to Solve Light Chain Mispairing in Bispecific Manufacturing

Light chain mispairing is the single largest source of product-related impurities in bispecific antibodies that use two different light chains, accounting for 5-15% of the Protein A eluate in KIH formats without a dedicated solution. The mispaired species has the correct heavy chain heterodimer but swapped light chains: LC-A paired with HC-B and LC-B paired with HC-A. Because the mispaired molecule has the same molecular weight as the correctly assembled bispecific, it cannot be separated by SEC and is difficult to detect without mass spectrometry.

Five strategies address light chain mispairing, each with distinct manufacturing trade-offs:

1. CrossMab domain crossover: Swapping CH1 and CL domains in one Fab makes it structurally incompatible with the other light chain. Reduces mispairing to below 1%. Used in emicizumab. Adds no purification burden but requires re-engineering of one Fab arm.
2. Common light chain: Both arms share one light chain, eliminating the problem. Used in teclistamab, talquetamab, and other Janssen DuoBody-platform molecules. Constrains the discovery campaign because the shared LC must bind both targets adequately.
3. Charge-pair mutations: Introducing complementary charged residues at the CH1-CL interface (e.g., K-D or R-E pairs) creates electrostatic steering that favours correct LC-HC pairing. Multiple academic and commercial implementations exist, each reducing mispairing to 2-5%.
4. Orthogonal Fab interfaces: Redesigning the VH-VL or CH1-CL interface of one arm so that only the cognate light chain fits. Similar principle to charge-pair but with more extensive interface engineering.
5. Separate production (cFAE): Each parent mAb is produced with its own light chain in a dedicated cell line. Light chain mispairing is structurally impossible because only one light chain is present per cell.

Purification: Removing Homodimer, Half-Antibody, and Aggregates

Bispecific antibody purification requires a multi-step chromatography platform that goes beyond the standard mAb Protein A capture, viral inactivation, and ion exchange polish. The additional complexity arises from four product-related impurity classes that are structurally similar to the target heterodimer and absent from conventional mAb processes.

The typical three- to four-step purification train for a KIH bispecific is:

1. Protein A capture: All Fc-containing species bind (heterodimer, homodimers, half-antibodies, aggregates). Removes HCP, DNA, and media components. Step yield 85-95%. This step does NOT resolve bispecific-specific impurities.
2. Low-pH viral inactivation: Standard hold at pH 3.5-3.7 for 60 minutes. Aggregation risk is higher for asymmetric bispecifics than for mAbs; optimise pH and hold time carefully.
3. Mixed-mode or HIC polishing: The critical step for homodimer removal. Capto MMC ImpRes (mixed-mode) or butyl/phenyl HIC exploits the subtle charge/hydrophobicity difference between heterodimer and homodimers. Homodimer is reduced from 5-15% to below 2%. Step yield 70-85%.
4. AEX flowthrough or CEX bind-elute: Removes aggregates, DNA, and endotoxin. Standard mAb polish step, adapted for the bispecific's isoelectric point. Step yield 85-95%.

The choice between mixed-mode chromatography and HIC for homodimer polishing depends on the physicochemical properties of the specific bispecific. If the heterodimer and homodimers differ in hydrophobicity (common when the two Fab arms have different CDR surface character), HIC with a descending ammonium sulfate gradient achieves baseline separation. If the difference is primarily charge-based, CEX with a shallow NaCl gradient or mixed-mode chromatography (which combines both charge and hydrophobic interactions) is preferred.

Half-antibody removal is often accomplished at the Protein A step itself by using an intermediate-pH wash (pH 4.5-5.0) that selectively elutes half-molecules while the intact bispecific remains bound. Alternatively, hydrophobic interaction chromatography resolves half-antibodies from intact molecules based on their lower molecular weight and different surface hydrophobicity.

Analytical Methods for Bispecific Antibody Characterization

No single analytical method can resolve all bispecific antibody impurities. Regulatory agencies expect an orthogonal panel of at least three complementary techniques that together cover size-based, charge-based, hydrophobicity-based, and mass-based characterisation. The challenge is that homodimers are nearly identical to the heterodimer in size (differing by only 0-5 kDa) and sometimes in charge, making conventional mAb SEC and CEX methods insufficient on their own.

Intact mass spectrometry is the definitive identity test for bispecific antibodies. Under denaturing conditions (reversed-phase LC-MS on a QTOF or Orbitrap), the heterodimer gives a single peak at the expected molecular weight (typically 145-150 kDa for an IgG1 bispecific), while homodimers appear at slightly different masses. Native MS under non-denaturing conditions can additionally resolve higher-order oligomers. The mass accuracy must be within 1-2 Da of the theoretical value to confirm correct chain assembly.

Dual-target binding assays (bridging ELISA, SPR, or BLI) are functionally the most important test: coat plate with Antigen-A, add sample, detect with labelled Antigen-B. Only correctly assembled heterodimer with both functional binding arms generates signal. This is the basis of the potency assay for release testing.

Worked Example: KIH Bispecific Purification Campaign

The 61% overall yield compares to 75-85% for a standard mAb purification train. The yield penalty comes almost entirely from the mixed-mode polishing step, where the tight cut needed to exclude homodimer sacrifices heterodimer in the overlapping region of the chromatogram. Optimising the gradient slope and pooling criteria at this step is the single highest-impact development activity.

Scale-Up Considerations

Bispecific antibody scale-up follows standard mAb principles for the upstream (constant P/V, matched kLa) and most downstream steps, but introduces three format-specific risks that require attention during process characterisation.

• Chain ratio sensitivity: The ratio of the two heavy chains (and two light chains, if applicable) must remain balanced at manufacturing scale. Nutrient gradients, dissolved CO₂ accumulation, and osmolality drift at large scale can differentially affect the expression of one chain over another. Monitor chain ratio at the Protein A step by analytical HIC or intact MS during engineering runs.
• Aggregation propensity: Asymmetric bispecifics are more aggregation-prone than symmetric mAbs, particularly during low-pH viral inactivation, freeze-thaw, and high-concentration formulation. The asymmetric Fc creates a hydrophobic patch that is absent in homodimeric IgG1. Hold-time studies at each process intermediate should be tighter than for a standard mAb platform.
• Mixed-mode chromatography scale-up: The gradient resolution required for homodimer removal is sensitive to column packing quality, flow rate, and bed height. HETP qualification criteria should be tighter than for standard ion exchange (target HETP < 0.04 cm, asymmetry 0.8-1.5). A shallow gradient (15-20 CV) is typical, which increases buffer consumption and processing time.

Frequently Asked Questions

Related Tools

• Chromatography Calculator — Column sizing, loading capacity, and buffer volume calculations for Protein A, CEX, AEX, HIC, and mixed-mode steps.
• Scale-Up Calculator — Bioreactor engineering parameters (P/V, kLa, tip speed, Re) for scaling CHO fed-batch from bench to manufacturing.
• Filtration Calculator — TFF membrane sizing and sterile filter area for UF/DF concentration and formulation steps.

References

1. Ridgway JBB, Presta LG, Carter P. 'Knobs-into-holes' engineering of antibody CH3 domains for heavy chain heterodimerization. Protein Eng Des Sel. 1996;9(7):617-621. doi:10.1093/protein/9.7.617
2. Schaefer W, Regula JT, Bähner M, et al. Immunoglobulin domain crossover as a generic approach for the production of bispecific IgG antibodies. Proc Natl Acad Sci USA. 2011;108(27):11187-11192. doi:10.1073/pnas.1019002108
3. Labrijn AF, Meesters JI, de Goeij BECG, et al. Efficient generation of stable bispecific IgG1 by controlled Fab-arm exchange. Proc Natl Acad Sci USA. 2013;110(13):5145-5150. doi:10.1073/pnas.1220145110
4. Surowka M, Klein C. A pivotal decade for bispecific antibodies? mAbs. 2024;16(1):2321635. doi:10.1080/19420862.2024.2321635
5. Wang Q, Chen Y, Pelzer M, et al. Non-affinity platform for processing knob-into-hole bispecific antibody. Bioresour Bioprocess. 2024;11:103. doi:10.1186/s40643-024-00827-8