Discovery and functional characterization of recurrent gene fusions from 7,470 primary tumor transcriptomes across 28 human cancers. C. Bandlamudi1, P. Lin1, J. Tian2, R. Grossman1, K. White1 1) University of Chicago, Chicago, IL; 2) Duke University, NC.
Gene fusions are consequences of somatic rearrangements in cancer genomes. Many oncogenic fusions have been discovered in different tumor types that currently serve as diagnostic, prognostic and therapeutic markers. However, an emerging theme from recent sequencing studies is that many tumorigenic fusions appear at frequencies 2% or less within the respective tumor types, suggesting that many functional fusions with clinical significance remain to be discovered using large sample sizes and sensitive detection approaches. We have developed a novel algorithm, Minimum Overlap Junction Optimizer (MOJO), that uses a transcriptome guided approach to detect fusions. Using 20 tumor transcriptomes with experimentally validated fusions, we show that MOJO demonstrates the highest sensitivity and specificity compared to nine other published methods. We performed fusion discovery using MOJO on 7,470 transcriptomes in the Cancer Genome Atlas (TCGA). Using 1,800 normal tissue transcriptomes from Genotype Tissue Expression (GTEx) consortium, we developed filters to model and account for technical and biological noise in fusion discovery through RNAseq. We demonstrate our sensitivity by recovering all fusions that have so far been validated in these samples. We nominated 16,114 high confidence fusion calls including 430 known fusions and 1,039 events involving known cancer genes in COSMICs Cancer Gene Census. We find that the frequency spectrum of fusion events ranges from a median of 8.3 events/tumor in Ovarian to 0.3 events/tumor in Thyroid. Our integrated analysis of copy number and fusion events in a subset of tumor types suggest that the rate of fusion events is correlated with the overall degree of genomic instability. Our analysis identified 201 fusion genes found in 5 or more samples across multiple tumor types. Using additional filtering criteria, we selected 29 in-frame fusion genes for gain-of-function validations. We generated stable MCF10A cell lines expressing these in-frame fusions, and so far, we have assayed 18 of them and found that 50% showed a statistically significant increase (p-value < 0.01) in proliferation, as compared to the GFP-only control. Intriguingly, we find that a majority of the validated fusions are identified in three or more tumor types, albeit, with low frequencies.
You may contact the first author (during and after the meeting) at