What a living tumor model can tell us
Functional testing measures a response in a model. The next question is how well that model represents the patient. DNA, RNA, and protein suggest mechanisms. A living model lets researchers perturb those mechanisms with a drug.
Before you start: Review the molecular layers. Function adds an experiment rather than another parts list.
Choose a model for the question
| Model | What researchers do | What it preserves | Important limitation |
|---|---|---|---|
| Dissociated cells | Separate cells and expose them to drugs | Some original tumor-cell properties | Tissue organization and interactions change |
| Organoids or spheroids | Grow cells in three-dimensional culture | Some architecture and heterogeneity | Culture selects cells; standard epithelial cultures lack a complete immune compartment |
| Tissue fragments | Test living pieces without extensive expansion | Some local stromal and immune interactions | Short survival and uneven penetration |
| Patient-derived xenograft (PDX) | Grow human tumor in a mouse and treat the animal | Many tumor features and whole-animal exposure | Human stroma is largely replaced by mouse stroma; conventional mice lack an intact human immune system |
PDX is an in vivo model. A mouse-derived organoid tested in a dish is an ex vivo or in vitro experiment. Neither is universally best. Establishment time and success depend on specimen, method, and disease. TNBC often engrafts more readily than estrogen-receptor-positive breast cancer; there is no single success rate or guaranteed turnaround for a new specimen. Breast tumor graft study; TNBC establishment study.
Immune effects need the relevant immune cells
A tumor-only organoid cannot reproduce a checkpoint inhibitor's full immune mechanism. Researchers can add immune cells, use air–liquid-interface cultures, or build other co-cultures. “Organoid” does not tell you whether an assay contains functional T cells. Primary organoid immune-microenvironment study.
Ask which cells are present, whether tumor and immune cells come from the same person, what is measured, and which controls establish tumor-specific killing. A change in mass, growth, or viability is a readout; it does not automatically identify the mechanism.
A worked example: two antibodies, different scores
Imagine two anti-PD-1 antibodies scoring 80 and 30 in one well-based assay. The numbers are invented.
The difference could reflect exposure, viability, cell composition, sampling, or measurement variation. Check reproducibility and controls first. A same-class difference in a single unvalidated assay does not prove one antibody would work better clinically. It cannot justify replacing an evidence-based regimen or adding a drug with a different toxicity profile.
What can go wrong
- The model grows a subset of the original clones.
- Laboratory concentrations differ from achievable patient exposure.
- A drug reduces growth without killing cells.
- The model omits cells required for the mechanism.
- The experiment measures tumor response but not normal-tissue toxicity.
- Reported model–patient agreement excludes samples that failed to grow.
Interpretation needs dose ranges, controls, replicates, a stated endpoint, and validation in the relevant population. A model result does not establish an adjuvant indication. Retrospective agreement does not prove that model-guided treatment improves survival.
Try it
A tumor-only organoid does not respond to pembrolizumab. Has it shown that the patient's immune system will not respond?
Answer: no. The model may omit the required immune cells. Even a suitable co-culture requires clinical validation.
Explain it back
Read “the model responded” as an experimental finding, then ask what connects it to patient benefit.
Takeaway
Read “the model responded” as an experimental finding, then ask what connects it to patient benefit.
Next: Blood testing over time. Provider capabilities and specimen operations belong in Functional drug testing and the tracker.
Sources and scope
Source check: October 8, 2026. The studies above establish model characteristics, not a prescription for model-guided care. For controls and combination effects, see Reading a functional experiment.