Manufacturable by Design: De-Risking Cell Line Development for Bispecifics, Multispecifics & Difficult-to-Express Proteins

Speed to IND has become the metric every biologics program is measured against. But bispecific and multispecific antibodies don’t follow a typical monoclonal antibody (mAb) development process, and neither do other difficult-to-express (DTE) proteins. Their structural complexity slows down exactly the step that speed depends on most: cell line development (CLD).

Understanding and responding to the pressure to move fast and working with molecules that are structurally harder to express reliably is the starting point for any CLD strategy. Dealing with it means building a strategy that will identify a molecule’s specific liabilities long before they cost a program months, rather than relying on a process designed for a simpler format.

Abzena Featured Author & Resident CLD Expert:

• John Gill, VP & Scientific Leader of Cell Line Development at Abzena

The pressure behind the timeline

The rationale for urgency is reflected in the markets. One market analysis estimates the next-generation antibody market at $5.7 billion in 2025, projected to reach $11.3 billion by 2032 (1), it’s a projection that will vary by source and by how “next-generation” is defined, but the direction is consistent across analyses: a growing share of the pipeline is bispecific, multispecific, or otherwise structurally complex. Investors and patients are both pushing for a faster path to IND.

But that pressure runs up against a precedent problem. The reduced timelines biopharma has come to expect were established with mAbs, i.e., molecules with a single binding domain and a single heavy and light chain pair, refined over decades of optimization. Bispecifics and multispecifics don’t inherit that history, since two (or more) heavy and light chain combinations have to assemble correctly inside the same cell, and the resulting heterogeneity is first a production problem.

Why these molecules are difficult to express

A mAb has a dominant correct assembly outcome, with comparatively limited mispairing. Half-antibodies and aggregates can occur at low frequency even within a single heavy and light chain pair, but there’s no second light chain species for it to mispair with. A bispecific multiplies the possible outcomes and introduces exactly that risk. When two distinct heavy chains and light chains are co-expressed, they can pair in several combinations (e.g., correct heterodimer, homodimers, various mispaired species), and conventional Chinese hamster ovary (CHO) expression systems have no specific mechanism to enforce the correct heterodimer pairing. Endoplasmic reticulum quality control (ERQC), folding kinetics, and secretion bias all influence which species make it through, but none is specific enough to reliably favor the intended pairing over the alternatives.

The knobs-into-holes (KiH) approach addresses part of this at the design stage, where complementary mutations engineer a “knob” on one heavy chain, and a “hole” on the other (2), so the two are structurally biased toward pairing with each other rather than with themselves. KiH resolves heavy chain heterodimerization but does nothing to stop the light chains pairing with the wrong heavy chain, which is why it is frequently combined with a common light chain or a CrossMab-style domain swap to control that separately (3). Even with both strategies in place, mispairing is reduced rather than eliminated, and heterodimer formation still has to be confirmed empirically in every cell line and process developed around a given construct. The KiH approach also does not address expression level challenges that can result in low overall process yield.

This is also why there’s no single template for bispecific CLD. Every format (KiH, common light chain, appended domains, or other heterodimerization approaches) presents its own combination of chain-pairing risk and expression imbalance between chains, plus specific structural liabilities. A process optimized for one bispecific format will not transfer cleanly to the next.

The same expression bottleneck extends to other DTE proteins, including fusion proteins and multi-domain constructs. For these molecules, low titer is frequently more than a transcription problem, as mRNA stability and translation efficiency are common limiting steps, and nuclear export can also contribute in specific engineered contexts, which is why vector engineering aimed only at boosting transcription often underperforms in these formats specifically.

Designing the lead candidate with the endpoint in mind

The highest-leverage point in the entire process is at lead candidate selection. In silico sequence risk assessment, applied to the primary amino acid sequence before a cell line is ever built, can flag liabilities that may affect stability, aggregation propensity, or chain-pairing risk. Addressing these at the sequence level (engineering out a liability or consciously choosing to carry it forward with a mitigation plan already in place) is far cheaper than discovering it during scale-up.

This is where the case for starting CLD early is strongest. Running CLD in parallel with lead optimization, rather than sequentially after a single lead is locked, allows several candidates to be evaluated as fast stable pools simultaneously. Bulk stable pools can be generated in a matter of weeks, ahead of full clonal cell line development, and used to generate early material for developability screening along with preliminary formulation work and analytical method development. The trade-off here, where transiently or pool-derived material comes with a higher risk of altered product quality attributes than a qualified clonal line, is acceptable for comparative lead selection in exchange for the time saved.

Engineering the vector and host for the harder cases

Where a molecule’s expression ceiling is set by translation and stability rather than transcription alone, vector design has to work on multiple fronts at once. Our AbZelectPRO^TM platform illustrates the scope of what that involves in practice: dual-promoter architecture to increase transcription, elements supporting mRNA transport out of the nucleus, and additional elements that enhance translation initiation and mRNA stability, combined with an evolved CHO host cell line selected through repeated bioreactor passaging for higher viable cell density at faster doubling times.

In a case study comparing standard expression vectors against this vector technology across multiple CHO host backgrounds (including CHOk1SV, HD-BIOP3, CHOZN, CHO DG44, CHO-S, and CHO-K1), cell lines built with the enhanced vector produced 2–6-fold higher titers across a panel that included a mAb, a bispecific antibody, and an Fc-fusion protein (4, 5). In CHO DG44 cells specifically, a 3–4-fold increase in protein production was observed relative to a conventional vector, with further gains achieved through methotrexate amplification and sub-cloning. In the CHO DG44 line, scale-up to 300 L bioreactors was carried out as part of further process optimization. Separately, in CHO-S and CHO-K1 lines, additional media and scale-up optimization by the client achieved production levels exceeding 6 g/L. These figures are specific to the constructs and conditions tested; expression gains for any individual bispecific or DTE candidate still need to be confirmed empirically, since chain balance and folding behavior vary by format.

Screening for the right clone, not just a productive one

Titer alone does not qualify a clone for GMP use. For bispecifics in particular, a high-producing pool can still be dominated by mispaired species, so screening has to select on productivity and correct assembly together, and do it fast enough not to lose the time gained upstream.

Microfluidic single-cell screening platforms, such as Cyto-Mine®, address the throughput side of this by encapsulating individual cells in picodroplets and measuring productivity and growth kinetics per cell across thousands of clones simultaneously, while generating strong, well-documented evidence of monoclonal origin through imaging and deposition records. Regulators accept monoclonality on a probabilistic basis rather than requiring absolute proof, but the strength of that documented evidence is still what a GMP submission is judged on, which is why the record generated at single-cell screening matters as much as the productivity data it produces. This is a meaningfully different proposition from limiting dilution or FACS-based sorting, both of which trade throughput for the depth of per-clone data available at the point of selection.

What GMP-ready actually requires

Speed at the DNA-to-research cell bank (RCB) stage is not the same thing as speed to a GMP-compliant cell bank. An RCB can be established in as little as 10 weeks under an optimized process. Progressing from there to a fully characterized, GMP-ready master and working cell bank (meeting ICH Q5D expectations for genetic stability, sterility, mycoplasma and broader contaminant testing, and documented traceability) typically takes 6–12 months of further development and qualification.

For bispecifics and other DTE formats, this stage carries its own format-specific risk: genetic or phenotypic drift in a cell line expressing a structurally complex, multi-chain construct can show up as a shift in chain stoichiometry or an increase in mispaired species, not just a decline in yield. Stability testing across the intended number of generations in production, not just at the point of banking, is what catches this before it reaches a GMP batch.

Where speed and GMP-readiness meet

The apparent conflict between speed and rigor dissolves once the CLD strategy is built around the molecule’s actual liabilities rather than a generic reduced-timeline template. The main risk sits in treating a DTE construct’s early data as equivalent to a monoclonal’s, and discovering the gap only once a Phase 1 cell bank fails a stability check or a monoclonality verification that a standard mAb process was never designed to catch.

That gap is rarely caught by CLD working in isolation. The chain-pairing ratios, clone stability data, and sequence liability flags generated during CLD are the same data a downstream release testing and cGMP process depend on. When CLD sits apart from analytical characterization and process development, that data has to be regenerated once the molecule changes hands, at exactly the point in the timeline where a delay is hardest to recover. Abzena runs CLD alongside analytical characterization, process development, and cGMP manufacturing, so the chain-pairing and stability data established at clone selection carries forward into process scale-up and release testing rather than being rebuilt from scratch.

For bispecifics, multispecifics, and other difficult-to-express formats, that continuity is where the time gained earlier in CLD either holds all the way to a GMP batch or gets spent again downstream.

FAQs

Why do bispecific antibodies take longer to develop cell lines for than monoclonal antibodies?
Bispecifics require two distinct heavy and light chains to assemble correctly inside the same cell. This creates multiple possible mispaired species alongside the intended product, a structural problem mAb expression systems were never built to resolve, and it has to be screened for empirically in every cell line generated.

What makes a protein difficult to express?
DTE proteins, including many bispecifics and fusion proteins, often have expression bottlenecks beyond transcription, such as limited mRNA stability or poor translation efficiency, with nuclear export sometimes contributing in specific engineered contexts. Vector strategies that only increase transcription frequently underperform for these constructs specifically.

How much faster can cell line development be with an optimized platform?
In one case study across multiple CHO host backgrounds, an enhanced vector and host system produced 2–6 fold higher titers than a conventional vector, including for a bispecific antibody and an Fc-fusion protein, with high-producing clones remaining stable through scale-up to 300 L. Results for a specific molecule still depend on its individual expression and chain-pairing behavior.

Is a fast RCB the same as a GMP-ready cell bank?
No. An RCB can be generated in around 10 weeks under an optimized process, but progressing to a fully qualified GMP master and working cell bank that meets ICH Q5D expectations for stability, sterility, and traceability, typically takes 6 months of additional characterization and qualification.

If you are looking to develop a Bispecific and need to find out more you can visit Bispecific & Multispecific Antibody Development, or contact our experts today.

References

1. Next-Generation Antibody Therapeutics Industry, Persistence Market research. https://www.persistencemarketresearch.com/market-research/next-generation-antibody-therapeutics-market.asp.
2. Y. Xu, J. Lee, C. Tran, T. H. Heibeck, W. D. Wang, J. Yang, R. L. Stafford, A. R. Steiner, A. K. Sato, T. J. Hallam, G. Yin, Production of bispecific antibodies in “knobs-into-holes” using a cell-free expression system. mAbs 7, 231–242 (2015).
3. D. T. Abdeldaim, K. Schindowski, Fc-Engineered Therapeutic Antibodies: Recent Advances and Future Directions. Pharmaceutics 15, 2402 (2023).
4. Enhancing protein expression in CHO cell lines: Unleasing the potential of 2G UNic technology, Proteonic (2026). https://abzena.com/wp-content/uploads/2023/12/ProteoNics-CHO-Application-Note_Published.pdf.
5. CHO Cell Line Performance with 2G UNic® Technology Case Study, Abzena. https://abzena.com/resources/case-studies/cho-cell-line-performance-with-2g-unic-technology-case-study/.

The post Manufacturable by Design: De-Risking Cell Line Development for Bispecifics, Multispecifics & Difficult-to-Express Proteins appeared first on Abzena.