Detection of a sequence feature for recursive splicing

-acting sequences associated with recursive splicing have not been identified. Using probabilistic mixture models, we found that recursive splicing occurs more frequently in first introns, which are typically longer, and exhibit a distinct CG-rich sequence feature in the sequences flanking the upstream 5'SS, and depletion of CGs in the downstream polypyrimidine tract. Remarkably, recursive splicing is also more frequent in downstream introns of genes containing first introns with these properties. Mechanistically, these data suggest that early events in RNA synthesis and processing influence the prevalence of recursive splicing for the rest of the transcript. Finally, we developed a sequence-dependent classifier for recursive splicing, which we tested with a novel medium-throughput primer extension assay. In summary, the usage of recursive splicing sites is established at the beginning of RNA synthesis through newly-identified sequence motifs flanking both ends of the first intron.