Choosing the right drug for breast cancer can still involve an uncomfortable amount of trial and error. Current treatment selection relies largely on biomarkers and molecular profiling as directly testing multiple drugs against a patient’s own tumor has generally been too slow for routine clinical use. Researchers in France are now developing a way to test the drugs directly against a patient’s living tumor cells that could shrink that functional testing window from weeks or months to just four days.
In a study published in Cell Reports Medicine, researchers at Institut Curie developed a patient-derived “tumor-on-a-chip” platform designed to rapidly determine whether breast cancer cells are sensitive or resistant to different treatments. The approach correctly identified 88% of drug-sensitive cases and 91% of resistant cases when compared with matched patient-derived xenograft (PDX) models.
The “tumor-on-a-chip” technology tackles a persistent limitation of precision oncology: knowing what mutations a tumor carries does not necessarily reveal which drug will kill it. Response rates to commonly used oncology treatments, from chemotherapies to targeted therapies like antibody-drug conjugates (ADCs), can vary widely, and robust biomarkers for selecting among them are lacking.
Functional precision medicine takes a different approach. Instead of predicting response solely from molecular information, researchers expose a patient’s actual cancer cells to potential therapies and observe the results. PDX models, which implant human tumors into mice, or lab-grown patient-derived organoids allow scientists to do this. But PDX testing can take months, while organoids often require weeks to establish and cannot be successfully generated from every tumor. The tumor-on-chip shrinks that process dramatically.
The Institut Curie researchers isolated breast cancer cells, embedded them in a three-dimensional collagen matrix inside a microfluidic chip, and exposed them to drugs while monitoring cell viability. Across six PDX models representing different breast cancer subtypes, the system distinguished responses to a panel of standard chemotherapies—including carboplatin, paclitaxel, and 5-fluorouracil—within four days.
Importantly, those responses largely mirrored what happened when the corresponding tumors were treated in mice. At clinically relevant concentrations, the researchers reported no false-positive drug responses. Overall, the platform delivered results at least five to 10 times faster than the in vivo models.
The researchers also tested ADCs, such as trastuzumab deruxtecan (T-DXd), finding that the chip reproduced sensitivity and resistance seen in corresponding PDX models. That suggests the technology could potentially evaluate newer targeted treatments in addition to conventional chemotherapy.
Perhaps most important for clinical translation, the team designed a smaller chip requiring only about 10,000 tumor cells, making it possible to generate models from the limited material available in biopsies. In seven patient samples, the chips detected markedly different drug-response profiles. One fresh biopsy came from a patient with highly treatment-resistant triple-negative breast cancer; none of the tested drugs substantially reduced tumor-cell viability on the chip, matching the patient’s clinical history of resistance.
Led by co-first authors Clara Helal, Lea Pinon, Juliette Jin, and Niccolo Schintu, the researchers contend that the technology remains proof-of-concept rather than a clinical decision-making test. The models simplify the tumor microenvironment, some patient samples could not be reliably analyzed, and larger prospective trials will need to determine whether on-chip responses actually predict how patients fare on treatments selected using the assay.
But if those studies succeed, tumor-on-chip technology could add a functional layer to precision oncology: rather than only asking what molecular alterations a patient’s cancer contains, clinicians could rapidly test which drugs that individual tumor actually responds to.
