When a targeted therapy stops working because tumour cells have shed the antigen it was designed to engage, the field calls it antigen escape. The name implies a biological event — something the tumour does, something unpredictable, something outside the designer's control. That framing is wrong. Antigen escape is evolutionary selection, and evolutionary selection responds to engineering decisions made years before the first patient is dosed.
The pattern is well-documented across targeted oncology. A therapy engages its target with precision. Response rates are strong. Then, over months, tumour cells that have downregulated or lost the targeted antigen — cells that were present in the original tumour mass but invisible to the therapy — survive while sensitive cells are cleared. They repopulate. The patient relapses. The antigen expression profile of the relapsed tumour looks different from the tumour at baseline. The therapy is no longer engaging what is there.
What escape actually is
Antigen loss under therapeutic pressure is not a random mutation event. It is the predictable outcome of applying selective pressure to a heterogeneous cell population. In any tumour, antigen expression is not uniform across every cell. Some cells express the targeted antigen at high levels. Some express it at intermediate levels. Some have already reduced or lost expression through natural variation in that tumour's evolutionary history. The therapy kills the high-expressers first. Intermediate expressers survive longer. Low or no expressers survive longest and repopulate.
This is Darwinian selection operating at the speed of cancer cell division. The therapy does not cause the escape — it reveals which cells were already positioned to survive it. From the tumour's perspective, losing a surface antigen is a straightforward fitness advantage when that antigen is the molecule a therapy is using to kill it.
The structural vulnerability of single-target therapy
A therapy targeting one antigen provides one escape route. The evolutionary path to resistance requires a single step: reduce or eliminate expression of that antigen. Tumour cells with a pre-existing tendency toward lower expression of the target are already partway there. The selective pressure the therapy applies does the rest.
This is not a criticism of any particular therapy — the single-target bispecifics that have reached approval in multiple myeloma represent genuine clinical progress and have extended the lives of patients with limited options. The constraint is structural. A molecule designed around one target can only be as durable as that target's expression is stable under therapeutic pressure.
Why AND-gate dual-target designs are better — and still insufficient
First-generation dual-target designs address this by requiring two antigens to be present simultaneously before the molecule engages. The logic is sound: a tumour cell must lose both antigens to escape the therapy, and losing two independent antigens is a more difficult evolutionary step than losing one.
But AND-gate architecture introduces its own vulnerability. If the molecule requires Antigen A AND Antigen B to engage, then a tumour cell that loses either antigen alone has already escaped — the molecule simply will not bind in a productive killing configuration. One escape route remains: lose one antigen, not necessarily both. In a heterogeneous tumour where antigen expression varies by cell, the subpopulation that has already shed Antigen A or Antigen B is the subpopulation the AND-gate molecule cannot engage.
Dual-target AND-gate designs raise the escape barrier. They do not close it.
OR-gate architecture: closing the second escape route
The OR-gate approach encodes a different logic: each tumour-targeting arm drives killing independently. The molecule does not require both antigens to be present simultaneously. If Antigen A is present, that arm engages and the tumour cell dies. If Antigen B is present, that arm engages independently and the tumour cell dies. Both arms must fail simultaneously — both antigens must be lost — before the molecule loses its ability to engage.
The escape arithmetic changes materially. To survive an OR-gated therapy, a tumour cell must have lost both independent antigens — it must present neither, whether those losses occurred sequentially or arose from pre-existing heterogeneity. Each antigen loss carries its own evolutionary fitness cost — surface proteins serve biological functions in the tumour cell beyond serving as therapeutic targets, and losing them has consequences. Requiring two independent fitness-costly loss events is not an impossible evolutionary path, but it is a substantially more difficult one than requiring a single step.
The design logic runs as follows:
AND-gate: Antigen A AND Antigen B both required for engagement. Lose one antigen — the molecule no longer engages. One escape route. One evolutionary step.
OR-gate: Antigen A OR Antigen B sufficient for engagement. Each arm kills independently. To escape, the tumour cell must have lost both antigens. Two independent loss events required, each carrying its own fitness cost in principle.
An engineering decision made before the first patient
The critical point is timing. Antigen escape is not discovered after a therapy fails — it is a predictable consequence of the design choices made during molecular engineering. The escape architecture of a therapy is determined at the stage when the activation logic is encoded into the molecule. A single-target molecule carries single-target escape risk from day one. An AND-gated molecule carries its escape profile from day one. An OR-gated molecule with mechanistically complementary target selection carries a structurally different escape profile from day one.
This means the question of whether a therapy was designed for the reality of antigen escape has a single correct answer moment: during design. The biology a patient's tumour presents when the therapy eventually stops working is, in significant part, determined by decisions made years earlier at the engineering stage. Escape is not something that happens to a therapy — it is a structural property of the therapy, either accounted for in the design or not.
The problem is solvable. Not perfectly — biology is not software, and evolution does not stop. But encoding OR-gate activation logic into a molecule's architecture, with target pairs selected specifically because their expression is not co-regulated, shifts the escape probability substantially before a single patient is dosed. That is an engineering problem. It has an engineering answer.