Circular RNAs composed of exonic sequence have been described in a small number of genes. Thought to result from splicing errors, circular RNA species possess no known function. To delineate the universe of endogenous circular RNAs, we performed high-throughput sequencing (RNA-seq) of libraries prepared from ribosome-depleted RNA with or without digestion with the RNA exonuclease, RNase R. We identified >25,000 distinct RNA species in human fibroblasts that contained non-colinear exons (a "backsplice") and were reproducibly enriched by exonuclease degradation of linear RNA. These RNAs were validated as circular RNA (ecircRNA), rather than linear RNA, and were more stable than associated linear mRNAs in vivo. In some cases, the abundance of circular molecules exceeded that of associated linear mRNA by >10-fold. By conservative estimate, we identified ecircRNAs from 14.4% of actively transcribed genes in human fibroblasts. Application of this method to murine testis RNA identified 69 ecircRNAs in precisely orthologous locations to human circular RNAs. Of note, paralogous kinases HIPK2 and HIPK3 produce abundant ecircRNA from their second exon in both humans and mice. Though HIPK3 circular RNAs contain an AUG translation start, it and other ecircRNAs were not bound to ribosomes. Circular RNAs could be degraded by siRNAs and, therefore, may act as competing endogenous RNAs. Bioinformatic analysis revealed shared features of circularized exons, including long bordering introns that contained complementary ALU repeats. These data show that ecircRNAs are abundant, stable, conserved and nonrandom products of RNA splicing that could be involved in control of gene expression.

Abstract Personalized cancer medicine, the matching of therapies to a given patient's somatic alterations, depends on highly accurate and complete identification of patients’ somatic alterations, or their mutome. Advances in sequencing technologies (exome sequencing, RNAseq, and whole genome sequencing) have provided a means to examine large portions of the genetic content of patients’ cancers. Computational tools have arisen that make somatic mutation predictions utilizing particular sequencing assays; however, each sequencing assay has limitations and existing mutation detection tools exhibit less than ideal agreement when analyzing the same data. The task of identifying all somatic mutations in one patient's cancer remains a challenge to personalized cancer medicine. Typically, somatic mutation detection is performed utilizing DNA sequencing. Because RNA sequencing is often a component of genome characterization projects along with DNA sequencing, we sought to evaluate the possible added value of RNA sequencing in somatic mutation detection. We have developed an original computational method, UNCeqR, that makes patient-specific somatic mutation predictions utilizing RNA sequencing combined with DNA sequencing. DNA mutations and RNA mutations are statistically modeled separately and results are combined in a meta-analytic fashion, resulting in up to three predictions for a locus: DNA-only, RNA-only, and DNA+RNA. In addition to de novo genomewide mutation predictions, UNCeqR can query specific a priori mutations. UNCeqR was applied to The Cancer Genome Atlas (TCGA) lung squamous cell carcinoma sequencing data, consisting of Ilumina RNAseq and Illumina exome sequencing. Of annotated exons, 20% had very low to zero coverage in RNA and 5% had very low to zero coverage in DNA, indicating that both sequencing assays add new genomic territory for mutation detection. Limiting to regions with both DNA and RNA coverage, 56% of mutations detected from DNA were also predicted by RNA, providing an independent validation of these mutations. To evaluate if mutation detection using DNA+RNA is superior to detection using DNA-only, cancer specimen DNA and RNA reads were randomly split into subsamples. UNCeqR was executed on each of the subsamples and mutation agreement was compared among pairs of subsamples within regions of DNA and RNA coverage. Compared with the DNA-only method, DNA+RNA mutation detection exhibited a 42% relative increase in percent agreement across subsamples and a 230% relative increase in the number of mutations detected. Therefore, RNA sequencing adds positive value to somatic mutation detection via UNCeqR. Citation Format: {Authors}. {Abstract title} [abstract]. In: Proceedings of the 103rd Annual Meeting of the American Association for Cancer Research; 2012 Mar 31-Apr 4; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2012;72(8 Suppl):Abstract nr 3975. doi:1538-7445.AM2012-3975