High-throughput interrogation of programmed ribosomal frameshifting in human cells

Martin Mikl; Yitzhak Pilpel; Eran Segal

The features of the different subsets driving the prediction provide additional information about the 465 The copyright holder for this preprint (which was not peer-reviewed) is the author/funder. It . https://doi.org/10.1101/469692 doi: bioRxiv preprint effect of specific sequence properties on the readout (Fig S11A) , e.g. recapitulating earlier 466 observations that XXXCCCZ ("Y=C") and XXXGGGZ ("Y=G") slippery sites are disfavored (Fig S11A 467 upper left, cf. Fig 2D) . For a model trained on the MFE of different lengths of downstream region, 468 stronger secondary structure not always led to prediction of higher PRF efficiency (Fig S11A lower 469 left). When taking into account regions of e.g. 60 nt after the slippery site ("dg60"), a lower feature 470 value favored prediction of PRF (Fig S11B) .). While the positive effect of strong secondary structure 471 (low MFE) in the first 60 nt on PRF prediction was strongly enhanced if also the downstream region 472 as a whole showed low MFE (Fig S11B) , in general we find strong secondary structure of the entire 473 downstream region (120 nt) to be associated with prediction of no PRF (Fig S11A) . This could either 474 be explained through an effect on frameshifting efficiency or on RNA stability and overall expression 475 levels. The latter option is unlikely to explain the effect in full as we do not observe a correlation 476 between MFE of the downstream region and steady-state RNA levels ( Fig S11C) . the frameshifting site, but not in the slippery site itself (Fig S12A) . Frameshifting rates of the variants 486 are distributed around the rate observed for the lab strain HXB2 (Fig 6A) . This constitutes actual 487 differences in frameshifting rates and not only experimental variability, as isolated clones show 488 remarkably good correlation given the small differences in PRF signal we are measuring (Fig 6B) . 489