Shrinking Breast Cancer Into a Petri Dish: Patient-Derived Organoids Reveal Drug-Resistance Vulnerabilities

One of the most difficult challenges in breast cancer treatment is not only finding a drug that can attack a tumor, but knowing in advance which tumor will resist it. A team at the University of California, San Francisco attempted to grow patients’ cancer tissue into three-dimensional “mini tumors,” then classify these organoids using molecular data accumulated through clinical trials, allowing treatment responses to be tested in a petri dish.

The study collected and analyzed 44 patient-derived breast cancer organoids, selecting gene-expression and protein biomarkers that could still be measured in the organoids. The researchers then combined these with transcriptomic data and treatment outcomes from the I-SPY2 breast cancer trial to build response models for treatments including immunotherapy, PARP inhibitors combined with platinum chemotherapy, and HER2-targeted drugs. The central aim of this design was to allow predictions derived from clinical data to undergo functional testing in three-dimensional models that retain some characteristics of patients’ tumors.

Using triple-negative breast cancer as an example, the team validated a response model for a combination of the PARP inhibitor veliparib and a platinum drug. The organoids’ drug sensitivity or resistance in laboratory experiments matched predictions based on data from I-SPY2 patients, suggesting that clinical biomarkers and in vitro functional testing may complement each other, without relying solely on a single mutation to determine the direction of treatment.

The organoids were also used to investigate a more practical follow-up question: when a standard combination fails, what other vulnerability can be targeted? The researchers screened 386 compounds in a triple-negative breast cancer organoid resistant to veliparib and platinum treatment, identifying candidate regimens that could be paired with cisplatin to suppress tumor growth again, including strategies that promote cancer cell apoptosis. These results provide leads for further drug development, but cannot yet be regarded as proven new treatments for patients.

The key limitation of this work also lies in the word “prediction.” The study primarily tested whether a model built from data from a group of I-SPY2 participants could reproduce corresponding drug responses in organoids; it did not prospectively select drugs for each organoid donor and then demonstrate that the patients achieved better clinical outcomes as a result. Culture success rates, production time, intratumoral heterogeneity, and the absence of immune cells and a complete microenvironment may also affect how closely organoid responses align with responses in the human body.

The currently available version of the study is hosted on a biology preprint platform, whose database clearly states that it has not yet undergone journal peer review. Advancing this approach into clinical practice will require prospective validation in independent and larger patient populations, demonstrating that testing can be completed reliably within treatment-decision timeframes and that it improves drug selection and patient outcomes more accurately than existing biomarkers.