High-throughput spheroid-based assay for functional breast cancer precision medicine facilitated by deep learning

Despite rapid advances in treatment, breast cancer remains the leading cause of cancer mortality in women, with triple negative breast cancers having a particularly poor prognosis. Some tumors have (epi)genetic alterations causing homologous recombination deficiency, providing opportunities for targeted therapeutics including poly (ADP-ribose) polymerase inhibitors. However, the effects of targeted treatments are variable; therefore, functional assays are needed to predict the best personalized treatment options.

We developed a high-throughput spheroid-based assay using patient-derived breast cancer xenograft models sensitive and resistant to cisplatin. Methods were developed for automatic spheroid segmentation using deep learning to measure response of spheroids to treatment with cisplatin, olaparib and radiotherapy. We developed a method to distinguish between sensitive and resistant tumors based on predicting the percentage of responding and non-responding spheroids.

Here we show that differences in treatment response between cisplatin-sensitive and resistant tumors faithfully correspond with the expected in vivo responses. The assay is able to discriminate between olaparib-sensitive and resistant tumors based on predicting the percentage of responding and non-responding spheroids.

We demonstrate that this assay, guided by automatic spheroid segmentation using deep learning, may report on the tumor's sensitivity to therapies with the potential to be applied to functional precision oncology for breast cancer.