Caution needs to be taken when assigning transcription start sites to ends of protein-coding genes: a rebuttal.

Naturally occurring stress-induced transcriptional readthrough is a recently discovered phenomenon, in which stress conditions lead to dramatic induction of long transcripts as a result of transcription termination failure. In 2015, we reported the induction of such downstream of gene (DoG) containing transcripts upon osmotic stress in human cells, while others observed similar transcripts in virus-infected and cancer cells. Using the rigorous methodology Cap-Seq, we demonstrated that DoGs result from transcriptional readthrough, not de novo initiation. More recently, we presented a genome-wide comparison of NIH3T3 mouse cells subjected to osmotic, heat, and oxidative stress and concluded that massive induction of transcriptional readthrough is a hallmark of the mammalian stress response. In their recent letter, Huang and Liu in contrast claim that DoG transcripts result from novel transcription initiation near the ends of genes. Their conclusions rest on analyses of a publicly available transcription start site (TSS-Seq) dataset from unstressed NIH3T3 cells. Here, we present evidence that this dataset identifies not only true transcription start sites, TSSs, but also 5'-ends of numerous snoRNAs, which are generally processed from introns in mammalian cells. We show that failure to recognize these erroneous assignments in the TSS-Seq dataset, as well as ignoring published Cap-Seq data on TSS mapping during osmotic stress, have led to misinterpretation by Huang and Liu. We conclude that, contrary to the claims made by Huang and Liu, TSS-Seq reads near gene ends cannot explain the existence of DoGs, nor their stress-mediated induction. Rather it is, as we originally demonstrated, transcriptional readthrough that leads to the formation of DoGs.