BIONEWS
Weill Cornell tested a PARP inhibitor, a class of targeted cancer drugs, on lab grown tumors from patients who wouldn't normally get one. 58% responded anyway.
When Andrea Sboner's team at Weill Cornell Medicine tested the PARP inhibitor talazoparib on lab-grown "mini-tumors" derived from patients who, under current clinical rules, would never be prescribed a PARP inhibitor, 58% of those supposedly ineligible organoids responded to the drug anyway. The finding, [published this month in Science Advances](https://www.science.org/doi/10.1126/sciadv.adz3351), points to a constructive problem in precision oncology: the molecular gatekeepers that decide who gets PARP inhibitors may be filtering out patients whose tumors would actually respond.
PARP inhibitors are a class of targeted cancer drugs that work especially well in tumors with DNA-repair weaknesses, most famously in breast and ovarian cancers carrying BRCA mutations. Clinicians use a set of clinical criteria, including BRCA mutation status and broader measures of homologous recombination deficiency (HRD), to decide which patients get them. Those criteria are conservative by design: the drugs are expensive, the side effects are real, and physicians do not want to prescribe them to patients who will not benefit. The Sboner team's data, drawn from their pan-cancer patient-derived organoid platform, suggest the criteria may be conservative in the wrong way.
A patient-derived organoid, or PDO, is a tiny three-dimensional tumor grown from a patient's own cancer cells, kept alive in a dish and used as a stand-in for testing drugs. The Weill Cornell / Englander Institute for Precision Medicine (EIPM) team built 220 such organoids from 191 patients, spanning 15 cancer types. In validation work, the organoids matched their parent tumors: 93% concordance on histopathology (the visual appearance of the tumor under a microscope), 80% median concordance on driver mutations, and a 0.85 median correlation on gene expression. Expression profiles stayed largely stable across 10 passages, and 85% of dominant tumor clones were preserved. The dishes were still acting like the patients' tumors.
That fidelity is what made the talazoparib arm interpretable. The team ran the drug on organoids from patients who, on paper, would not be eligible for any PARP inhibitor under current rules, then asked which molecular features predicted response. The 58% sensitivity rate among the "ineligible" cohort is the headline. The susceptibility signatures the team fingerprinted are the working argument: if those signatures hold up in prospective trials, they would give oncologists something better than BRCA or HRD status alone to filter patients by.
Sboner, quoted in the Weill Cornell newsroom, frames the platform as a practical tool: organoids "appear to be very good preclinical models... practical because they can be used long-term." The implicit critique is that current eligibility rules, built before patient-specific testing at this scale was possible, are under-inclusive rather than over-inclusive. That is a precision-oncology problem worth naming out loud.
The caveats are real and the source sets them. Organoid response is not patient response. The 58% number is preclinical, retrospective within a defined cohort, and exploratory; the team itself treats it as a hypothesis generator, not a treatment recommendation. The molecular signatures need prospective validation before they can replace, or even formally supplement, current criteria. The platform covers 15 cancer types but not all of them, and drug-screening coverage beyond talazoparib is still limited.
What to watch: whether the susceptibility signatures move into a prospective trial design, and whether the broader PDO cohort gets used to interrogate the same question for other targeted-therapy classes. The EIPM Director's Memo for July 2026 flags the platform as a long-running EIPM asset rather than a one-off study, which means the data and the patient-derived reagents will keep generating hypotheses after this paper.
The trade press coverage and a parallel write-up on news-medical.net have led with the platform itself: 220 mini-tumors, 15 cancer types, 93%, 80%, 0.85. Those numbers are real and they are the engine. But the question a cancer patient, an oncologist, or a payer actually wants answered is who qualifies, and on the evidence Sboner's team has put on the table, the answer may be wider than the current rules suggest.