A biotech startup's lead oncology bispecific had stalled at <0.3 g/L expression with >60% mispairing contamination — unusable for IND studies. AntibodyLLM's AI platform redesigned the heavy chain interface, implemented a KiH/electrostatic pairing strategy, and engineered a stable CHO cell line that achieved 1.8 g/L with >95% correctly paired bispecific and passed a 6-month accelerated stability panel.
The client was developing a T-cell engager bispecific antibody targeting a tumor-associated antigen on one arm and CD3 on the other — a well-validated mechanism in oncology but one of the most technically demanding formats in biologics development. Their lead molecule had been designed in-house using a conventional IgG1 scaffold, and initial transient expression runs in HEK293 cells produced a complex mixture of species: correctly paired bispecific (~35%), homodimers from each arm (~45%), and half-antibody fragments (~20%).
The mispairing problem was not just a purity issue — it was a safety and efficacy issue. CD3-arm homodimers can cause systemic T-cell activation independent of tumor targeting, a known on-target off-tumor toxicity risk. The regulatory path for an IND submission required demonstrating <5% mispairing species in the drug substance.
With a 12-month runway and an IND filing target, the team needed a complete redesign of the molecular format, a new stable CHO cell line, and a purification process that could reliably deliver >95% correctly paired bispecific — at commercially viable yields.
A conventional IgG has two identical heavy chains and two identical light chains — they always pair correctly. A bispecific needs two different heavy chains and two different light chains to assemble with absolute precision. Without engineered pairing control, the cell produces a statistical mixture of all possible combinations.
When two different heavy chains are co-expressed, they preferentially form homodimers with their identical partner rather than assembling into the desired heterodimer. Without engineering to break homodimer thermodynamic preference, mispairing dominates.
Even with correct heavy chain heterodimerization, each light chain can associate with either heavy chain. A bispecific with two different light chains generates up to four distinct heavy-light pairings — three of which are wrong. Light chain selectivity engineering is required for reliable correct assembly.
Even with correct design, unequal expression of the two heavy chain genes in a single CHO cell can skew the stoichiometry, increasing the fraction of unwanted species. Stable cell line engineering must ensure balanced co-expression throughout passaging and production runs.
The solution required engineering at three levels simultaneously: the molecular format (heavy chain interface), the light chain strategy, and the cell line itself. Each level addressed a distinct root cause of the mispairing problem.
AntibodyLLM applied a dual-layer heavy chain engineering strategy. The first layer used a Knobs-into-Holes (KiH) modification at the CH3 interface: a large residue (T366W, the "knob") was introduced into one heavy chain, paired with a complementary cavity (T366S/L368A/Y407V, the "hole") in the other. KiH creates strong steric preference for heterodimer over homodimer formation — favoring the desired pairing by approximately 95% thermodynamically.
The second layer used AI-designed electrostatic complementarity: AntibodyLLM's structural modeling identified additional CH3 interface positions where placing opposite charges (K→E on one chain, E→K on the other) could further destabilize homodimers without affecting heterodimer stability. The AI screened 120 candidate electrostatic pairs computationally before selecting the combination with the highest predicted ΔΔG penalty for homodimer formation.
The combined KiH + electrostatic design pushed heavy chain heterodimerization to >98% in transient expression — resolving the primary source of mispairing contamination before any cell line selection or purification step.
| Strategy | HC Heterodimer % | LC Selectivity | Expression Impact |
|---|---|---|---|
| Unmodified IgG1 (client original) | ~35% | None | — |
| KiH only | ~90% | Partial | Slight reduction |
| Common light chain | ~85% | Full | Activity may be compromised |
| KiH + Electrostatic (AntibodyLLM) | >98% | Full (CrossMAb LC) | 1.8 g/L maintained |
With heavy chain heterodimerization solved, the light chain misassignment problem remained. For a bispecific with two structurally distinct light chains, there is no simple way to force each light chain to associate only with its correct heavy chain — unless the heavy chain is made structurally incompatible with the wrong light chain.
AntibodyLLM implemented a CrossMAb light chain strategy: the CH1 domain of one heavy chain was swapped with a CL (constant light) domain, and the corresponding light chain was given a CH1 domain instead of CL. This domain exchange creates a domain mismatch that enforces correct pairing through steric and electrostatic incompatibility — the "wrong" light chain simply cannot assemble stably with the swapped heavy chain.
Combined with the KiH + electrostatic heavy chain design, overall bispecific purity in the harvested cell culture fluid reached >93% before any purification — significantly reducing the burden on the downstream process and enabling a simpler, higher-yielding purification train.
A correctly designed bispecific molecule that mispairs at low frequency in transient expression can still fail in stable cell lines if the two heavy chain genes are expressed at different levels. AntibodyLLM engineered the stable CHO cell line using a bicistronic vector with an IRES element and a 2A self-cleaving peptide to drive co-expression of the two heavy chains from a single promoter — ensuring equimolar production throughout the production run.
Cell line selection prioritized clones with both high specific productivity and a demonstrated stable bispecific ratio over 60 days in culture. Out of 48 clones screened, 6 met both criteria; the top clone was selected for production scale-up based on yield, growth profile, and stability panel results.
The final cell line produced 1.8 g/L in a 14-day fed-batch process with the bispecific purity remaining above 95% across three independent production runs — confirming process reproducibility for the IND submission CMC package.
IND CMC package completed — full analytical characterization, three production lots, and 6-month accelerated stability data included.
Purification cannot fix a fundamentally mispaired product — the mispairing species are often too similar to the target molecule to remove by chromatography. Engineering high-purity correct assembly at the molecular level is the only scalable solution.
KiH alone addresses heavy chains but not light chains. The CrossMAb strategy addresses light chains. Neither is sufficient alone. Using both in combination is what achieves the >95% purity threshold required for IND-enabling studies.
Bispecifics that pass initial characterization can still fail stability studies if the engineered interfaces are thermodynamically marginal. Incorporating DSC screening and accelerated stability testing early — not as a final IND box-check — prevents costly late-stage failures.
Chain mispairing, low expression, and stability failures are solvable problems — if the molecular engineering is right from the start. Our AI-guided design platform has the structural modeling depth to get the format right before you commit to a cell line program.
Discuss Your Bispecific