A multispecific can produce an impressive tumor-killing curve and still leave a team with the wrong architecture. The added arm may contribute little outside a favorable assay, selectivity may weaken as concentration rises, or the intended molecule may be difficult to assemble and recover. Those limitations need to shape the design decision while there are still alternatives worth pursuing.

The central question is what the added complexity buys the program. A third specificity, another binding domain, or a different geometry should address a defined biological limitation and remain useful in a molecule that can be produced consistently. That requires functional selectivity and product quality to enter the same conversation early, before potency alone settles the ranking.

Give the additional arm a specific job

A costimulatory arm can strengthen the response where initial T-cell activation is insufficient. But cytotoxicity, proliferation, and cytokine production do not necessarily move together. Changes to CD3 and CD28 affinities, epitopes, or domain placement can improve one response while altering another, including activation when tumor cells are absent.

The comparison with a corresponding CD3 bispecific should therefore start with the intended gain. If the limitation is inadequate proliferation, a better killing curve does not resolve the question on its own. If the goal is stronger cytotoxicity, that improvement needs to be weighed alongside cytokine release and unwanted T-cell engagement. Keeping these responses separate makes the tradeoff visible.

Tumor-absent conditions belong in that comparison from the outset, alongside relevant target-expression levels and cellular contexts. A third arm that improves activity only in one favorable setting has earned another experiment, but the case for committing to its architecture remains incomplete. The next round should establish where the benefit holds and where the added stimulation becomes a liability.

Test selectivity where it is most likely to weaken

Avidity-based selectivity depends on the difference between multivalent engagement on densely expressing tumor cells and binding to lower-density cells. A strongly expressing tumor line can demonstrate potency while revealing very little about that discrimination. The useful experiment spans the antigen-density range over which the design is expected to separate desired from unwanted activity.

That response depends on the whole format. Affinity, epitope, linker design, topology, and CD3 positioning can interact in nonlinear ways. Adding tumor-binding valency may improve discrimination in one arrangement and increase engagement more broadly in another. More binding sites do not establish a selective mechanism without the functional evidence to support it.

AND-gated designs need equally demanding comparisons. Dual-positive, single-positive, and negative cells should be assessed at relevant expression levels and across concentration. Recognition of two antigens does not establish that both are required for activity, and single-positive cells may become susceptible as exposure increases. The concentration range over which the proposed gate holds is part of the design’s value.

A focused matrix of affinities, domain positions, and valencies can expose these relationships without making the screen unmanageable. Changing one feature at a time may miss a useful combination when its effect depends on orientation. Changing everything at once makes a successful result difficult to interpret. The panel should be built around the interactions most likely to determine the intended selectivity, with expression and biophysical quality tracked alongside function.

Know which molecule produced the result

Functional rankings become harder to trust when preparations contain different proportions of intended product and closely related species. Total expression gives an incomplete account if much of the output is mispaired or partially assembled. Before comparing architectures, teams need to establish what was tested and how consistently that material can be made.

Chain balance is one practical place to look. Heavy- and light-chain plasmid ratios can influence yield, product quality, and mispairing during transient expression. Modestly increasing selected chains may improve assembly in some formats, making a limited ratio assessment worthwhile before an otherwise promising architecture is rejected.

The breadth of the workable conditions matters too. A construct that assembles reliably across the tested range presents a different development proposition from one that performs well only at a narrowly tuned ratio. Transient expression can reveal this sensitivity early, although it cannot establish how the molecule will behave in a later manufacturing process.

Correctly assembled product and recovered usable material should therefore sit alongside total yield in the comparison. That keeps biological performance tied to the molecule the team intends to develop and helps separate an architecture problem from an expression condition that can reasonably be improved.

Let purification influence the design decision

The species an architecture produces also determine how difficult it will be to isolate the intended molecule. Closely related unwanted species may share enough properties with the product to make separation challenging. Predicted charge differences can guide ion-exchange development, but the useful process still depends on pH, resin chemistry, buffer composition, and the material actually being produced.

Purity and recovery need to be reviewed together. High purity obtained by sacrificing most of the product presents a different problem from high purity achieved with workable recovery. Likewise, an expression improvement adds limited value if it mainly increases an unwanted species that follows the product through purification.

These results can justify revisiting chain pairing, domain arrangement, or scaffold choice before lead nomination. An architecture that reduces the separation burden may be worth a modest functional tradeoff if it still meets the required biological objective. That decision is easier to make while alternatives remain in the panel, rather than after the team has committed to a format and asked process development to make it work.

Reassess developability in the complete format

Individual-arm measurements are useful for removing clear liabilities, but they cannot fully predict the behavior of the assembled multispecific. Format conversion creates new interfaces and surface arrangements that can change self-interaction, hydrophobicity, and stability. Those changes can improve developability as well as worsen it.

This makes the parent antibody a starting point for assessment. A favorable isolated domain does not guarantee a favorable complete molecule, while a property that looks limiting in one format may behave differently in another. Teams need enough architectural variation to learn which liabilities persist and which depend on the arrangement.

A practical workflow screens domains for clear concerns, builds a representative architecture panel, and compares function, assembly, purification behavior, and biophysical properties on the complete constructs. This keeps early screening manageable while avoiding a premature commitment based on measurements that may not survive format conversion.

Choose the tradeoffs the program can support

An architecture review should explain what the added function contributes, where unwanted activity remains, and how reliably the intended molecule can be produced. Together, those answers give meaning to a potency gain. They also identify whether the next engineering cycle has a tractable problem to solve or whether the team should reconsider the format.

The strongest candidate may still need affinity tuning, a different chain balance, or purification optimization. The important point is to connect that work to a defined limitation and a measurable improvement. Functional assays can support a selectivity claim within the tested conditions, but they do not establish clinical safety; production results likewise need to be interpreted within the process used.

Multispecific design offers considerable freedom to combine functions. The most convincing architectures use that freedom with purpose, preserving the biological gain through assembly, purification, and format-specific assessment. That is a stronger basis for development than the most potent construct in an isolated assay.

Continue the conversation at PEGS AI

Carry the conversations from PEGS Europe into PEGS AI in January. Join the PEGS community to explore how AI can inform protein engineering and the experimental decisions behind more effective, developable biologics.

Register for PEGS AI