Detection of alternative isoforms of gene fusions from long-read RNA-seq with FLAIR-fusion

Abstract

Gene fusions are important cancer drivers and drug targets, but are difficult to reliably identify with short-read RNA-sequencing. Long-read RNA sequencing data are more likely to span a fusion breakpoint and provide more sequence context around the breakpoint. This allows for more reliable identification of gene fusions and for detecting alternative splicing in gene fusions. Notably, alternative splicing of fusions has been shown to be a mechanism for drug resistance and altered levels of oncogenicity. Here, we present FLAIR-fusion, a computational tool to identify gene fusions and their isoforms from long-read RNA-sequencing data. FLAIR-fusion can detect fusions and their isoforms with high precision and recall, even with error-prone reads. We also investigated different library preparation methods and found that direct-cDNA has a higher incidence of artifactual chimeras than direct-RNA and PCR-cDNA methods. FLAIR-fusion is able to filter these technical artifacts from all of these library prep methods and consistently identify known fusions and their isoforms across cell lines. We ran FLAIR-fusion on amplicon sequencing from multiple tumor samples and cell lines and detected alternative splicing in the previously validated fusion GUCYA2-PIWIL4, which shows that long-read sequencing can detect novel splicing events from cancer gene panels. We also detect fusion isoforms from long-read sequencing in chronic lymphocytic leukemias with the splicing factor mutation SF3B1 K700E, and find that up to 10% of gene fusions had more than one unique isoform. We also compared long-read fusion detection tools with short-read fusion detection tools on the same samples and found greater consensus in the long-read tools. Our results demonstrate that gene fusion isoforms can be effectively detected from long-read RNA-sequencing and are important in the characterization of the full complexity of cancer transcriptomes.