Drug combinations are often required to improve clinical benefits for patients with advanced, relapsed or refractory malignancies. Due to the massive number of possible drug-dose combinations, an exhaustive exploration of the combinatorial space is experimentally unfeasible, especially considering the scarcity of patient cells.
There is a need for systematic, yet practical approaches, that can address both between-patient and intratumoral heterogeneity when designing combinatorial treatment regimens for cancer patients. These approaches should rationally combine drugs that preemptively inhibit multiple disease- or resistance-driving mechanisms in individual patients, tumors and clones.
In a recent article published in Nature Communications, researchers from FIMM, BRIC, HUS, and iCAN developed an experimental-computational approach that leverages large-scale public pharmacogenomics datasets to pre-train a machine learning model, named scTherapy. scTherapy accurately predicts clone-specific therapeutic options based solely on a single data modality, single-cell RNA-sequencing.
The article demonstrates how the approach enables identification of personalized multi-targeting treatment options in heterogeneous patient and cell populations, with case studies in hematological malignancies (relapsed and refractory acute myeloid leukemia) and solid tumors (metastatic ovarian carcinoma).
“We demonstrate that the predicted combinations do not only show synergistic effect in overall cancer cell killing, but also result in minimal toxic side effects in non-cancerous cells, thereby increasing the likelihood for clinical translation”, says the first author of the study, Doctoral Researcher Aleksandr Ianevski from the Institute for Molecular Medicine Finland (FIMM).
Overall, >95% of the predicted combinations were experimentally confirmed to show synergy, highlighting their potential for improved therapeutic efficacy. Importantly, >80% of the predictions resulted in low-toxicity to normal cells, as confirmed by single-cell drug response assays in the AML patients, and organoid drug assays in the ovarian cancer patients, each with different disease stages (diagnosis, refractory or metastatic disease).
“Since the approach uses only a limited number of patient primary cells, it is widely applicable to any patient samples that are amenable to scRNA-seq profiling. Selective combinations among approved drugs also provide straightforward repurposing opportunities for cancer treatment”, says FIMM Doctoral Researcher Kristen Nader, the joint first author of the study.
The study was done in collaboration with the researchers from Biotech Research and Innovation Centre (BRIC) of the University of Copenhagen, Helsinki University Hospital, University of Helsinki (HiLIFE and Faculty of Medicine), and iCAN Digital Precision Cancer Medicine Flagship project.
The open-source R-codes provide a systematic approach for identifying personalized combinatorial regimens for individual patient samples that selectively co-inhibit malignant cells, while avoiding inhibition of nonmalignant cells. Such approaches that can guarantee maximal cancer-selectivity are expected to significantly accelerate the future design and testing of combination therapies, as well as increase the likelihood of their success in clinical studies.
Original publication: Ianevski, A., Nader, K., Driva, K. et al. Single-cell transcriptomes identify patient-tailored therapies for selective co-inhibition of cancer clones. Nat Commun 15, 8579 (2024). https://doi.org/10.1038/s41467-024-52980-5