Multiple approved therapies now target the same validated myeloma antigen. They do not perform the same. Response rates differ. Durability differs. Toxicity profiles differ. The biology of the target has not changed. What changed is how each molecule was designed to engage it.
This is not an anomaly. It is what the field has consistently observed across every major T-cell engager class: two molecules targeting the same antigen, developed with comparable scientific intent, separating substantially in the clinic. The explanation is rarely the target. It is almost always the design — specifically, what was optimised during discovery and what was measured to evaluate candidates before they entered patients.
What binding affinity does and does not predict
Most T-cell engager discovery selects candidates by binding affinity. It is measurable, scalable, and technically tractable. Lead molecules are selected because they bind their antigen with high affinity and specificity. This is a reasonable starting point. It is a poor finishing criterion.
Binding affinity does not predict immune synapse geometry. It does not determine whether the epitope engaged produces a productive killing conformation between T cell and tumour cell, or an abortive one. It does not measure the ratio of activating to exhaustion signals generated across a heterogeneous T-cell population. It does not tell you whether the cytokine environment produced by your molecule resolves after target clearance or whether it escalates into uncontrolled release.
These are the variables that separate clinical profiles. And they can only be observed in functional assays — not binding measurements, not computational prediction from structure alone, and not cell lines engineered for assay convenience. They are only legible in primary human immune cells, in conditions that approximate the disease biology the molecule must eventually navigate.
The measurement gap
The gap between what discovery programmes measure and what determines clinical performance is not a secret. The field has known for years that binding-selected candidates routinely fail to translate — or translate at significantly lower efficacy than preclinical data suggested. The gap persists because solving it is hard. Functional screening in primary human immune cells at meaningful scale requires infrastructure that most drug discovery operations were not built to run. It is slow, expensive, and produces data that is difficult to work with computationally.
The result is a consistent pattern across the CAR-T and T-cell engager literature: molecules targeting identical biology produce different outcomes because they were evaluated against different proxies for the biology that matters. Better binding. Cleaner structural models. Higher selectivity ratios in simplified cell assays. These are not irrelevant properties. They are insufficient ones.
Design as the differentiating variable
When two molecules targeting the same antigen diverge in the clinic, the explanation is almost always traced to decisions made during discovery — epitope selection, linker geometry, format choice, affinity tuning — and whether those decisions were guided by measurements that predicted functional immune behaviour or by measurements that were simply available.
This is the core argument for functional-first design. Not that binding data is useless — it remains a necessary filter — but that the functional readouts measured in living immune cells are the signal that predicts clinical behaviour, and that selecting for those readouts from the start produces molecules with fundamentally better-characterised therapeutic profiles before a single patient is dosed.
The goal is not faster discovery. The goal is different discovery — one in which the candidates selected for clinical development were evaluated against the biology they must eventually engage, not against tractable proxies chosen for assay convenience.
What this means for multispecific design
The complexity compounds in multispecific formats. A bispecific or trispecific molecule must not only engage two or three antigens — it must engage them simultaneously, in the right spatial configuration, producing a defined immune response profile across the full range of antigen expression states present in patient tumours. The variables governing this behaviour — cooperative binding geometry, conditional activation logic, the balance of signals generated when antigen expression is heterogeneous — are not measurable in binding assays. They emerge from functional behaviour in relevant disease conditions.
Molecules that appear equivalent by binding metrics can generate radically different immune activation profiles when evaluated in primary human cells with patient-relevant antigen expression patterns. This is the design space where multispecific T-cell engagers are differentiated — and the space in which functional-first discovery provides the clearest advantage.
The compounding effect
There is a second-order consequence of functional-first discovery that the clinical data has begun to surface: the candidates that survive functional pre-selection not only perform differently at their lead epitope — they tend to perform more consistently across the antigen expression variance present in real patient populations. Selection in primary human cells, measured across multiple donor backgrounds, generates molecules that have been evaluated against biological variability before development begins.
Same target. Different design. Different outcome. The clinical record is consistent on this point. The variable that explains the variance is not the antigen — it is what was measured and selected for during the years before the first patient was dosed.