Aberrant transcript usage is associated with homologous recombination deficiency and predicts therapeutic response

Abstract

Genomic scars, mutational signature3 which were accumulated by homologous recombination deficiency (HRD), and BRCA1/2 mutations were used as prediction markers for DNA damage inducing drug sensitivity. However, predicting sensitivity of DNA damage inducing drugs is suffering considerable difficulties by various mechanism of acquiring resistance including restoration of homologous recombination (HR). Especially, identifying resistant patients has been becoming highly important issue, because genomic scars and mutational signature 3 cannot represent a functionality of homologous recombination. In this study, we used RNA sequencing data as a highly dynamic data sources to overcome limitations of static data sources (e.g. DNA data). Consequently, we found significantly different splicing patterns in DNA repair pathway genes between HRD group and HR proficient (HRP) group. Because these splicing patterns do not represent their functional difference, we compared relative transcript expression to confirm their functionality. Strikingly, we found that significantly increased aberrant transcript usage (aTU) in HRD group. Because these patterns represent a reduced functionality of genes, we suggest transcriptional HRD (tHRD) machine learning model to predict DNA damage inducing drug sensitivity and HRD functionality based on aberrant transcript usage of DNA repair genes. Finally, the tHRD model was validated by showing better performance than genomic scars and mutational signature3 for classifying responder group of DNA damage inducing drug in breast and ovarian cancer patients and cell lines. These results suggest that importance of RNA splicing and isoform usage in expanding the application of the DNA damage inducing drug and establishing patient specific treatment strategies.