High-throughput interrogation of programmed ribosomal frameshifting in human cells

Martin Mikl; Yitzhak Pilpel; Eran Segal

Prediction of PRF events has up to now been largely limited to identifying sequences matching the 531 canonical pattern for slippery sites, followed by a downstream secondary structure (Belew et al., 2008; 532 Theis et al., 2008) . To the best of our knowledge, no attempts at quantitatively predicting the effect of 533 sequence variation exist to date. We used our measurements of frameshifting signal of thousands of 534 variants of known PRF sites to build a computational model based on known and novel sequence 535 features affecting frameshifting, leading to accurate prediction of frameshifting potential (up to ROC 536 AUC=0.93) and frameshifting rates (up to Pearson r=0.81). In many cases sequence properties other 537 than slippery site identity or downstream secondary structure rigidity yielded high prediction scores, 538 emphasizing the importance to include additional features such as amino acid properties and tRNA 539 availability in the investigation of PRF regulation. 540