Polysome fractionation analysis reveals features important for human nonsense-mediated mRNA decay

Nonsense-mediated mRNA decay (NMD) is a translation-dependent mRNA surveillance pathway that eliminates transcripts with premature termination codons. Several studies have tried defined the features governing which transcripts are targeted to NMD. However, these approaches often rely on inhibiting core NMD factors, which often have roles in non-NMD processes within the cell. Based on reports that NMD-targeted transcripts are often bound by a single ribosome, we analyzed RNA-Seq data from a polysome fractionation experiment (TrIP-Seq) to characterize the features of NMD-targeted transcripts in human cells. This approach alleviates the need to inhibit the NMD pathway. We found that the exon junction complex (EJC) model, wherein an exon-exon junction located ≥50 nucleotides downstream of a stop codon is predicted to elicit NMD, was a powerful predictor of transcripts with high abundance in the monosome fraction (bound by a single ribosome). This was also true for the presence of an upstream open reading frame. In contrast, as 39 UTR lengths increase, the proportion of transcripts that are most abundant in the monosome fraction does not increase. This suggests that either longer 39 UTRs do not consistently act as potent triggers of NMD or that the degradation of these transcripts is mechanistically different to other NMD-targeted transcripts. Of the ribosome-associated transcripts annotated as "non-coding", we find that a majority are bound by a single ribosome. Many of these transcripts increase in response to NMD inhibition, including the oncogenic SHNG15, suggesting many might be NMD targets. Finally, we found that retained intron transcripts without a premature termination codon are over-represented in the monosome fraction, suggesting an alternative mechanism is responsible for the low level of translation of these transcripts. In summary, our analysis finds that the EJC model is a powerful predictor of NMD-targeted transcripts, while the presence of a long 39 UTR is not.