Abstract 4110: Pathway activation driven miRNA depletion highlights cancer signaling vulnerabilities and targeted therapy sensitivity

Identification of cancer vulnerabilities is pivotal for the success of precision oncology. Comprehensive molecular profiling technologies improved clinical outcome of targeted drugs. However, the complexity of molecular interactions still challenges appropriate data interpretation. Hence, new strategies are needed to support current treatment decisions. One approach would be to consider additional molecular information, for instance microRNA (miRNA) expression patterns. By analyzing various cancer datasets, we observed that inhibition of miRNA expression occurs in a directed manner. Strong cancer driving mutations, such as in KRAS, result in a block of miRNAs inhibiting the cancer driver gene itself, or its associated downstream pathways. As many cancer drivers can activate diverse downstream pathways, we examined whether groups of strongly repressed miRNAs can uncover essential pathways. Therefore, we developed a miRNA depletion based cancer gene dependency model and related drug prediction workflow using miRNA sequencing data from cancer cell lines. We defined a pathway ranking score (PRS) and focused on druggable target genes. To validate our drug prediction workflow, we performed miRNA sequencing in various patient derived 3D cancer models (PDCMs). We predicted top hit pathway dependencies and individual druggable target genes for each culture. Subsequently, samples were used for in vitro drug response assays. Here, samples were significantly more sensitive to predicted targeted drugs, compared to standard of care (SOC) chemotherapy, or non-predicted targeted drugs. Next, we applied our miRNA based drug prediction workflow to four patients and compared three sources for RNA sequencing for each patient: PDCM, primary tumor, and microdissected purified tumor from FFPE tissue. Of note, FFPE purified miRNAs extraction led to comparable results to PDCMs and primary tumors. Finally, we were interested whether our prediction workflow matches with a comprehensive genomic based drug prediction workflow from the MASTER precision oncology trial. Therefore, we sequenced miRNAs from 95 patients and performed subsequent drug prediction. Interestingly, a complete overlap was overserved in only 20% of patients. In 35% of the patients, MASTER drug recommendation was biologically related to our drug prediction, and 40% revealed no match. In 75 out of 91 patients (82%) with available RNA Seq data, at least one of our predicted target genes was noticeably overexpressed on