The Illumina Stranded mRNA protocol is not strongly stranded for mRNA with low U content

The Illumina TruSeq Stranded and Illumina Stranded mRNA protocols are commonly used for strand-specific bulk RNA-seq and they typically yield >99% antisense reads. We show that these protocols can generate sense-oriented reads for transcripts with extremely low U content (<3%). Indeed, such regions can bypass the dUTP-based blockade of cDNA second strand amplification. A small number of genes are affected by this issue (three in Drosophila melanogaster, including the glue gene Sgs3, and 46 in Mus musculus). To prevent overestimation of expression levels, we recommend excluding sense reads for all genes.