Abstract
Transcription in mammalian cells is a complex stochastic process that displays bursts, resulting in vastly different numbers of an mRNA species in isogenic cell populations. Several factors contributing to this phenomenon have been identified, usually classified as intrinsic, i.e., local effects at single genes, or extrinsic, relating to the macroscopic state of the cell, such as its size or position in the cell cycle. However, each factor only accounts partially for the observed “transcriptional noise”, and some possible contributors have not been explored yet. In this paper, we investigate the role of crosstalk between processes at the 5’ and 3’ ends of a gene, which has been suggested to facilitate polymerase recycling during transcription. Using a systems analysis approach based on flow cytometry and Bayesian model fitting, we were able to measure the transcriptional noise of β-globin and HIV-1-env transgenes with sufficient accuracy and precision to unveil significant patterns. We find that perturbation of polymerase recycling from the gene terminator back to the promoter typically reduces the average burst size and increases the burst frequency, thus limiting the noise. The observed noise patterns are reproduced by a generative model that encodes polymerase recycling. This suggests that construction of the mammalian transcriptional machinery is subject to trade-offs between noise and average expression level.