Choose a target with a usable safety window
A target must support enough tumor recognition while limiting harmful recognition of healthy tissue. A disease name is not a target specification. This is step two of the engineered-cell guide.
Before you start: A chimeric antigen receptor (CAR) usually recognizes an accessible surface feature. A conventional T-cell receptor (TCR) recognizes a peptide–human leukocyte antigen complex.
Where this step sits
Choose the recognition requirement before tuning the receptor or delivery. A perfect delivery system cannot turn an unsafe antigen into a safe target.
The intuition: look beyond a positive label
“Target-positive” compresses a distribution into a label. Ask which cells carry the target, how much is accessible, and where else it occurs. Expression can differ within one lesion, between lesions and after treatment.
An epitope is the particular feature the recognition element binds. Receptor geometry and epitope accessibility can matter even when the underlying protein is present. RNA abundance does not directly measure accessible surface density.
A target worksheet
| Question | Useful evidence | Tempting overread |
|---|---|---|
| Which cancer cells display it? | Appropriate protein and spatial measurements across relevant samples | One positive region represents every lesion |
| How much is accessible? | Product-relevant binding and functional assays | Any staining exceeds the CAR’s activation threshold |
| Where does healthy tissue express it? | Relevant normal-tissue distribution and functional testing | “Tumor-associated” means tumor-exclusive |
| Can coverage change? | Serial samples and resistant-population analysis | The baseline sample stays representative forever |
| Does recognition create a usable window? | Dose, function and safety evidence for the complete product | Strong binding alone proves a better therapy |
Targets studied in solid tumors include many normal human proteins. A named target is a research category, not evidence of eligibility or a recommendation. Dated programs and target menus belong in the therapeutic landscape.
Worked example: coverage and harm
Invented numbers for learning: a tumor has 100 cells; 70 strongly display target X and 30 do not. A model CAR-X kills the 70 positive cells. The mass shrinks, but the negative cells remain.
Now add a healthy tissue with low X expression. Increasing sensitivity might improve recognition of some target-low cancer cells while also increasing healthy-tissue injury. The comparison needs both benefit and harm measurements. “More sensitive” is not a complete design goal.
On-target and off-target are different
On-target, off-tumor injury occurs when the receptor correctly recognizes its intended antigen on healthy cells. Off-target recognition involves an unintended molecular target. A normal-tissue problem can exist even with a highly specific binder.
A second target or a logic gate may broaden coverage or restrict activation. It also introduces a new design whose expression patterns, timing and failure behavior require testing. A diagram of two conditions does not prove a reliable safety system.
What can go wrong at this step
Sampling misses negative populations. Assays measure intracellular rather than accessible protein. A target changes after therapy. Healthy tissue under inflammation may differ from the baseline comparison. These uncertainties belong beside the target-positive result.
Try it
A CAR kills every target-positive cell in a dish. What remains unresolved before calling the target suitable?
Answer: Coverage across relevant tumor cells, healthy-tissue recognition, access, product behavior and clinical safety. The assay establishes activity under its tested conditions.
Explain it back
“A usable target needs ___ across cancer cells and ___ in healthy tissue.” One answer: “sufficient accessible coverage; an acceptable safety window.”
Takeaway
Assess tumor coverage and healthy-tissue recognition together for the actual receptor product.
Next: Build the receptor.
Sources and scope
Source check: October 8, 2026; expert and learner review pending. The example is fictional; no threshold or target is recommended.
- FDA CAR T-cell product-development guidance.
- NCI solid-tumor barriers.
- Primary report of ERBB2-targeted off-tumor toxicity — one specific clinical product and dose, not a universal risk estimate.