Approximate Bayesian computation of transcriptional pausing mechanisms

Abstract

At a transcriptional pause site, RNA polymerase (RNAP) takes significantly longer than average to transcribe the nucleotide before moving on to the next position. At the single-molecule level this process is stochastic, while at the ensemble level it plays a variety of important roles in biological systems. The pause signal is complex and invokes interplay between a range of mechanisms. Among these factors are: non-canonical transcription events - such as backtracking and hypertranslocation; the catalytically inactive intermediate state hypothesised to act as a precursor to backtracking; the energetic configuration of basepairing within the DNA/RNA hybrid and of those flanking the transcription bubble; and the structure of the nascent mRNA. There are a variety of plausible models and hypotheses but it is unclear which explanations are better.\n\nWe performed a systematic comparison of 128 kinetic models of transcription using approximate Bayesian computation. Under this Bayesian framework, models and their parameters were assessed by their ability to predict the locations of pause sites in the E. coli genome.\n\nThese results suggest that the structural parameters governing the transcription bubble, and the dynamics of the transcription bubble during translocation, play significant roles in pausing. This is consistent with a model where the relative Gibbs energies between the pre and posttranslocated positions, and the rate of translocation between the two, is the primary factor behind invoking transcriptional pausing. Whereas, hypertranslocation, backtracking, and the intermediate state are not required to predict the locations of transcriptional pause sites. Finally, we compared the predictive power of these kinetic models to that of a non-explanatory statistical model. The latter approach has significantly greater predictive power (AUC = 0.89 cf. 0.73), suggesting that, while current models of transcription contain a moderate degree of predictive power, a much greater quantitative understanding of transcriptional pausing is required to rival that of a sequence motif.\n\nAuthor summaryTranscription involves the copying of a DNA template into messenger RNA (mRNA). This reaction is implemented by RNA polymerase (RNAP) successively incorporating nucleotides onto the mRNA. At a transcriptional pause site, RNAP takes significantly longer than average to incorporate the nucleotide. A model which can not only predict the locations of pause sites in a DNA template, but also explain how or why they are pause sites, is sought after.\n\nTranscriptional pausing emerges from cooperation between several mechanisms. These mechanisms include non-canonical RNAP reactions; and the thermodynamic properties of DNA and mRNA. There are many hypotheses and kinetic models of transcription but it is unclear which hypotheses and models are required to predict and explain transcriptional pausing.\n\nWe have developed a rigorous statistical framework for inferring model parameters and comparing hypotheses. By applying this framework to published pause-site data, we compared 128 kinetic models of transcription with the aim of finding the best models for predicting the locations of pause sites. This analysis offered insights into mechanisms of transcriptional pausing. However, the predictive power of these models lacks compared with non-explanatory statistical models - suggesting the data contains more information than can be satisfied by current quantitative understandings of transcriptional pausing.