De novo gene evolution: How do we transition from non-coding to coding?

Recent years have witnessed the discovery of protein–coding genes which appear to have evolved de novo from previously non-coding sequences. This has changed the long-standing view that coding sequences can only evolve from other coding sequences. However, there are still many open questions regarding how new protein-coding sequences can arise from non-genic DNA.

Two prerequisites for the birth of a new functional protein-coding gene are that the corresponding DNA fragment is transcribed and that it is also translated. Transcription is known to be pervasive in the genome, producing a large number of transcripts that do not correspond to conserved protein-coding genes, and which are usually annotated as long non-coding RNAs (lncRNA). Recently, sequencing of ribosome protected fragments (Ribo-Seq) has provided evidence that many of these transcripts actually translate small proteins. We have used mouse non-synonymous and synonymous variation data to estimate the strength of purifying selection acting on the translated open reading frames (ORFs). Whereas a subset of the lncRNAs are likely to actually be true protein-coding genes (and thus previously misclassified), the bulk of lncRNAs code for proteins which show variation patterns consistent with neutral evolution. We also show that the ORFs that have a more favorable, coding-like, sequence composition are more likely to be translated than other ORFs in lncRNAs. This study provides strong evidence that there is a large and ever-changing reservoir of lowly abundant proteins; some of these peptides may become useful and act as seeds for de novo gene evolution.