Detecting copy number alterations in RNA-Seq using SuperFreq

Calling copy number alterations (CNAs) from RNA-Seq is challenging, because differences in gene expression mean that read depth across genes varies by several orders of magnitude and there is a paucity of informative single nucleotide polymorphisms (SNPs). We previously developed SuperFreq to analyse exome data of tumours by combining variant calling and copy number estimation in an integrated pipeline. Here we have used the SuperFreq framework for the analysis of RNA sequencing (RNA-Seq) data, which allows for the detection of absolute and allele sensitive CNAs. SuperFreq uses an error-propagation framework to combine and maximise the information available in the read depth and B-allele frequencies of SNPs (BAFs) to make CNA calls on RNA-seq data. We used data from The Cancer Genome Atlas (TCGA) to evaluate the CNA called from RNA-Seq with those generated from SNP-arrays. When ploidy estimates were consistent, we found excellent agreement with CNAs called from DNA of over 98% of the genome for acute myeloid leukaemia (TCGA-AML, n=116) and 87% for colorectal cancer (TCGA-CRC, n=377), which has a much higher CNA burden. As expected, the sensitivity of CNA calling from RNA-Seq was dependent on gene density. Nonetheless, using RNA-Seq SuperFreq detected 78% of CNA calls covering 100 or more genes with a precision of 94%. Recall dropped markedly for focal events, but this also depended on the signal intensity. For example, in the CRC cohort SuperFreq identified 100% (7/7) of cases with high-level amplification of ERBB2, where the copy number was typically >20, but identified only 6% (1/17) of cases with moderate amplification of IGF2, typically 4 or 5 copies over a smaller region (median 5 flanking genes for IGF2, compared to 20 for ERBB2). We were able to reproduce the relationship between mutational load and CNA profile in CRC using RNA-Seq alone. SuperFreq offers an integrated platform for identification of CNAs and point mutations from RNA-seq in cancer transcriptomes. The software is implemented in R and is available through GitHub: https://github.com/ChristofferFlensburg/SuperFreq .