The Generation of Promoter-Mediated Transcriptional Noise in Bacteria

Noise in the expression of a gene produces fluctuations in the concentration of the gene product. These fluctuations can interfere with optimal function or can be exploited to generate beneficial diversity between cells; gene expression noise is therefore expected to be subject to evolutionary pressure. Shifts between modes of high and low rates of transcription initiation at a promoter appear to contribute to this noise both in eukaryotes and prokaryotes. However, models invoked for eukaryotic promoter noise such as stable activation scaffolds or persistent nucleosome alterations seem unlikely to apply to prokaryotic promoters. We consider the relative importance of the steps required for transcription initiation. The 3-step transcription initiation model of McClure is extended into a mathematical model that can be used to predict consequences of additional promoter properties. We show in principle that the transcriptional bursting observed at an E. coli promoter by Golding et al. (2005) can be explained by stimulation of initiation by the negative supercoiling behind a transcribing RNA polymerase (RNAP) or by the formation of moribund or dead-end RNAP-promoter complexes. Both mechanisms are tunable by the alteration of promoter kinetics and therefore allow the optimization of promoter mediated noise.

💡 Research Summary

The paper addresses the long‑standing observation that transcription in bacteria can occur in irregular bursts, as first reported by Golding et al. (2005) for the lac/ara promoter in E. coli. In those experiments, periods of rapid successive transcription (“on” periods) lasting only a few minutes were interspersed with long silent intervals (“off” periods) of tens of minutes. Such behavior cannot be explained by a simple Poisson process, nor by the classic three‑step McClure model of transcription initiation (closed‑complex binding, isomerisation to an open complex, and escape into elongation).

The authors first formalise the McClure model mathematically, assuming fast equilibrium for RNAP binding/unbinding and treating isomerisation (rate O) and escape (rate E) as the two kinetically significant steps. They show that when one step dominates (Class I) the inter‑event times follow a single exponential distribution, and when the two steps are comparable (Class II) the distribution acquires a peak at a non‑zero interval. Neither case reproduces the heavy‑tailed “off” periods observed experimentally; both produce far less variability than required.

To capture the bursty dynamics, the authors propose two novel branched‑path mechanisms.

1. Supercoiling‑mediated recruitment – An elongating RNAP generates negative supercoiling behind it, which can dramatically accelerate open‑complex formation at downstream promoters. The model introduces a probability q that the supercoiling left by a previous transcription event persists long enough to make the isomerisation step effectively instantaneous. When the promoter is in the supercoiled state, initiation proceeds at the fast escape rate E; when not, initiation is limited by the slow rate O. This yields a mixed exponential distribution (Eq. 3) that, with parameters τ_O ≈ 37 min, τ_E ≈ 29 min and q ≈ 0.55, reproduces the experimentally measured distributions of inter‑transcript intervals, burst sizes (average ≈ 2.2 transcripts per burst), and on/off period lengths. Gillespie simulations confirm the analytical predictions.

2. Dead‑end complex formation – Occasionally RNAP may become trapped in a non‑productive, long‑lived complex (e.g., a back‑tracked or arrested intermediate). Such a dead‑end state blocks the promoter for an extended time, creating the long “off” periods. The model treats entry into the dead‑end state and escape from it as stochastic events with their own rates, allowing the system to switch between an active (bursting) state and an inactive state. By tuning these rates, the same statistical features observed by Golding et al. can be reproduced.

Both mechanisms are biologically plausible: negative supercoiling is known to increase promoter activity by >10‑fold in vitro, and dead‑end complexes have been documented in biochemical studies. Importantly, the models show that promoter‑mediated noise can be tuned by altering kinetic parameters (e.g., the probability q or the lifetime of the dead‑end complex), providing a potential evolutionary lever for cells to optimise phenotypic variability.

The paper concludes that transcription initiation in bacteria is not a simple Poisson process but a dynamic multi‑state system influenced by DNA topology and RNAP‑promoter interactions. By incorporating these additional layers of regulation, the authors reconcile theoretical models with single‑cell experimental data and open new avenues for synthetic biology designs that deliberately harness or suppress transcriptional noise.