Clinical Biomarkers Guide Platinum-Resistance Screens in Breast Cancer Organoids

Using Patient Biomarkers to Select a Resistant Laboratory Model

A Cell Reports Medicine study led by the University of California, San Francisco connects clinical response biomarkers with functional drug testing in breast cancer organoids. The investigators asked whether gene-expression patterns associated with treatment response in patients could identify laboratory models of resistant disease. They then used the selected models to test treatment hypotheses experimentally.

The analysis combined an existing 26-organoid biobank with 14 newly established triple-negative breast cancer organoid cultures. The organoids preserved many tumor-related expression signatures, but immune and stromal features were incompletely represented. Around three quarters of the 100 evaluated biomarker signatures varied detectably in the existing biobank. Some hormone-receptor-positive cultures also showed evidence of reduced estrogen signaling, a relevant culture-associated change.

Clinical modeling used I-SPY2 data from 987 patient samples across 10 treatment arms. Of 23 qualifying clinical gene-expression biomarkers, nine were both evaluable in organoids and nonproprietary. A model using PARPi7 and LYMPHS_PCA signatures predicted pathological complete response to veliparib plus carboplatin in triple-negative disease, with an area under the receiver operating characteristic (ROC) curve of 0.912 in the clinical dataset. This was the model carried forward into organoid testing.

Testing a Different Platinum Agent in Organoids

The 14 triple-negative organoids spanned a range of model-predicted responses. The investigators used clinical-data-derived thresholds to classify seven as sensitive, five as intermediate and two as resistant. These scores were predictions to be compared with experiments, not observed organoid response categories.

The laboratory assay combined veliparib with cisplatin, whereas the clinical model was based on veliparib plus carboplatin. The authors chose cisplatin for its stability and potency in their experimental setting. A rapid assay used 10 μM cisplatin plus 20 μM veliparib, with cell-death staining and bright-field imaging. Detailed follow-up in four representative cultures found broadly concordant responses: TORG40 was resistant, TORG146 sensitive, and TORG139 intermediate; TORG58 also showed resistance despite its intermediate predicted score.

This comparison links biomarker-defined disease states with functional response while retaining the change in platinum agent. It also shows why the output is a proof of concept rather than validation of an identical clinical regimen in every organoid.

Combination Hits and an Exploratory Clinical Follow-Up

The researchers screened 386 compounds in the resistant TORG40 culture, each with and without a fixed 5 μM cisplatin backbone. Twenty compounds had at least a 40% reduction in dose-response area under the curve with cisplatin. Most did not produce an important improvement over monotherapy. Eight selected compounds were tested in a second imaging-based assay, which confirmed an additive effect at one or more tested concentrations.

A BCL-2/BCL-xL inhibitor, ABT-263, was then evaluated with cisplatin in three intermediate or resistant organoids. Initial single-dose effects prompted formal dose-matrix experiments. Mean Bliss synergy scores ranged from 11.02 to 31.57, with the strongest interactions at 5–10 μM cisplatin. These findings prioritize an experimental combination; they do not establish its effectiveness or tolerability in patients.

The screen also identified HSP90 inhibitors as potent single agents. Ganetespib activity was confirmed in additional organoids, motivating a retrospective analysis of the biomarker-defined HER2-negative, immune-negative, DNA-repair-deficiency-negative I-SPY2 subgroup. That clinical subgroup was broader than the triple-negative subset modeled in the organoid screen.

Among 224 patients in this subgroup, 38 received ganetespib-containing therapy and 186 received control chemotherapy. Pathological complete response was similarly uncommon: 13% versus 9%, respectively, with p = 0.55. Distant recurrence-free survival favored the ganetespib group in the exploratory analysis, with p = 0.008. The pathological-response comparison was not statistically significant, and the authors described the survival analysis as hypothesis-generating, given its retrospective design, small treatment subgroup, lack of covariate adjustment and receptor-subtype imbalance.

The study demonstrates a reverse-translational workflow: use clinical biomarkers to identify a resistant tumor state, locate comparable organoid cultures, and test treatment hypotheses. Establishment bias, missing immune/stromal components, bulk-expression measurements and modest cohort sizes constrain generalization. The resulting combinations and biomarker associations require independent validation before they can support treatment selection.