Stochastic pausing at latent HIV-1 promoters generates transcriptional bursting

Abstract

Promoter-proximal polymerase pausing is a key process regulating gene expression. In latent HIV-1 cells, it prevents viral transcription and is essential for latency maintenance, while in acutely infected cells the viral factor Tat releases paused polymerase to induce viral expression. Pausing is fundamental for HIV-1, but how it contributes to bursting and stochastic viral reactivation is unclear. Here, we performed single molecule imaging of HIV-1 transcription, and we developed a quantitative analysis method that manages multiple time scales from seconds to days, and that rapidly fits many models of promoter dynamics. We found that RNA polymerases enter a long-lived pause at latent HIV-1 promoters (>20 minutes), thereby effectively limiting viral transcription. Surprisingly and in contrast to current models, pausing appears stochastic and not obligatory, with only a small fraction of the polymerases undergoing long-lived pausing in absence of Tat. One consequence of stochastic pausing is that HIV-1 transcription occurs in bursts in latent cells, thereby facilitating latency exit and providing a rationale for the stochasticity of viral rebounds.

Quantitative imaging of transcription in livingDrosophilaembryos reveals the impact of core promoter motifs on promoter state dynamics

“…We then used a recently described novel machine-learning method to infer the transcriptional bursting mechanism 59 . This method involves three major steps: detection of successive initiation events for each nucleus, multi-exponential parametric regression of the distribution of waiting times between successive events, and identification of Markovian promoter state transition models.…”

“…In order to detect initiation events, we considered that each trace results from the convolution between i) the sequence of initiation events marked by a rise in GFP intensity plotted over time, and ii) the signal produced by a single polymerase ( Figure 2B, B') 33,59 . The deconvolution procedure uses a genetic algorithm to determine optimal Pol II positioning within the gene body ( Figure 2C) and thus the time between successive Pol II initiation events (Δt) ( Figure 2D).…”

“…The entire computational pipeline was tested using artificial data (see also 59 ). Artificial traces were generated by simulating the model using the Gillespie algorithm with parameter sets similar to those identified from data.…”

“…The KS test is based on differences between estimated and empirical probabilities, being thus sensible to errors in the larger probabilities of shorter waiting times, and less sensitive to rare, very long waiting times. Our fitting procedure combines linear and logarithmic scales to find a balance between short and long time scales (see 59 ). Although relatively small (in terms of the objective function), the resulting errors on short times are large enough to be considered significant by the KS test for all models.…”

“…This model was clearly incompatible with our data, as the fitting exceeded the bounds of the 95% confidence interval (Supplemental Figure 6L, Supplemental Figure 7I"). Recent work on the prototypic model of a highly paused promoter, the HIV-1 promoter, also found that modeling transcriptional dynamics with obligatory pausing was not in agreement with live imaging data in HeLa cells, and instead proposed an alternative model of pausing where only a fraction of polymerases are subject to pausing in a stochastic manner 59 . We asked if this alternative nonobligatory pausing model could be applicable to our findings ( Figure 6C'').…”