Integrating computational and experimental approaches to optimize RNA fusion identification in BRCA associated breast cancer

Background: Germline testing has improved detection of high-risk breast cancer patients but risk reduction in germline BRCA carriers is limited to surgical intervention with no chemoprevention options. RNA fusions, somatic events from chimeric transcripts, generate neoantigens that can serve as targets for preventative vaccines but its presence in BRCA associated breast cancer is unknown. While these fusions can be predicted from RNA sequencing by tools like Arriba , the prediction pipelines need in-vitro validation to optimize target identification. We hypothesize RNA fusions can be identified in silico in BRCA associated cancers and validated at high frequency. Methodology: For identification of fusions in BRCA tumors, public repertoire of RNA sequencing of 34 BRCA associated breast cancers was utilized. Mammary tissue from 21 cancer free subjects were used as controls. For validation of in silico predictions, RNA sequencing was performed on mammary tissues or tumors from 7 normal-like BRCA c o/co Cre +\+ p53 +\+ controls and 23 BRCA c o/co Cre +\+ p53 +\- mice. RNA fusions in both human and mouse samples were detected using Arriba pipeline, followed by post-Arriba filters excluding fusions in normal controls, intra-genic, non-coding RNA and fusion without coding or splice-site breakpoints. Mouse fusion candidates were validated using PCR and Sanger Sequencing. Results: In human BRCA tumors 338 fusions were detected by Arriba ; with post- Arriba filters, 33 noncanonical, 251 noncoding genes & 3 fusions in CDS/CDS splice sites were removed with 51 patients specific fusions remaining. Fusions were detected in 50% of patient tumors. In 23 BRCA mouse samples, Arriba initially detected 352 fusions ; 157 noncanonical, 107 noncoding genes and 22 fusions present in normal control samples were excluded, yielding 69 fusions. Five recurrent mouse fusions- Thoc1-Usp14 (2 breakpoints), Tmcc-Raf1 , Usp14-Thoc1 , Krt5a-Krt6 (5 breakpoints), Runx2-Tmem191c with read counts of 4-21, 31-49, 2-4, 1-3 and 1 respectively were selected for validation. Four of these (except Runx2-Tmem191c) were validated by PCR followed by Sanger sequencing. In Krt5a-Krt6a fusions, three of five fusion junctions with >1 read count were confirmed. Notably, Usp14-Thoc1 showed fusions in 8 more samples, suggesting isoforms in common gene partners missed due to low sequencing depth. Overall validation rate of integrated Arriba -post Arriba pipeline was 80% reaching 100% with >1 supporting reads. Conclusion: RNA Fusions were successfully detected in both human and mouse samples using Arriba. Computational predictions using Arriba plus stringent post-filtering, reliably predicts RNA fusions when supported by >1 read count underscoring the importance of integrating computational prediction with in vitro validation to accurately characterize fusion landscapes in BRCA-associated breast cancer models.