The authors begin from a striking statistic: only about one percent of the 3.2 billion base pairs of the human genome codes for functional proteins, yet roughly 80.4 percent of the genome participates in non-protein-coding regulation of gene expression in at least one cell type. This enormous regulatory investment underscores how vital epigenetic gene regulation is at every stage of single- and multicellular life. Epigenetic modifications, chemical changes to DNA and the histone proteins around which it is wrapped, fold the genome into permissive and restrictive chromatin structures and nuclear arrangements. When these systems go awry, they contribute to the initiation and maintenance of diseased states. Yet while whole-genome mapping has revealed the landscape of these marks, the dynamics and variability of their interactions at the level of individual cells remain far less understood.

That variability is what biologists call transcriptional noise, the cell-to-cell variance in messenger RNA content, often quantified by the coefficient of variation, a mean-normalised standard deviation. Noise comes in two flavours. Extrinsic noise arises from subtle differences in the cellular microenvironment that affect processes such as growth rate and cell volume, and it tends to influence all gene expression within a given cell uniformly. Intrinsic noise, by contrast, stems from the randomness of biochemical reactions involving small molecule numbers, such as transcription factors binding to their cognate motifs, and it affects each gene uniquely. The distinction was famously established in bacteria by the Elowitz laboratory, which used two reporter genes expressing yellow and cyan fluorescent proteins under the same promoter: fluctuations in the combined signal reflected extrinsic noise, while divergence between the two colours within a single cell revealed intrinsic noise. The greatest source of intrinsic noise, the review emphasises, is the stochastic switching of gene promoters between active and inactive states.

The consequence of this switching is that messenger RNA is produced in short discrete bursts interspersed with longer periods of relative inactivity, a process occurring on timescales ranging from minutes to hours. Cell-to-cell variability results from the burst size, the number of RNA molecules produced during the active state; the burst frequency, how often transitions to the active state occur; and the degradation rate of the resulting transcripts. Remarkably, recent work has shown that fluctuations in gene expression can persist for several cell divisions, in line with the idea that bursting may drive phenotypic divergence of otherwise identical cells. On shorter timescales, bursting provides a way for cells to respond rapidly to stimuli. Understanding its regulation matters greatly for fields such as oncology, where heterogeneity in gene expression is associated with poor prognosis and drug resistance.

Here the review makes its most provocative conceptual move. The traditional two-state model, in which promoters flip between a single ON and a single OFF state, is increasingly challenged by evidence for more than two promoter states. The authors propose, in agreement with others, that bursting should be understood as a multi-scale phenomenon: during a macroscopic ON state, multiple small microbursts occur, corresponding to convoys of RNA polymerase II traversing the gene, separated by short and shallow OFF states lasting roughly 1.5 minutes, while deep OFF states extend far longer, around 9 to 34 minutes. The accumulation of these microbursts within each ON state produces a larger macroburst. This distinction matters because different experimental techniques operate at different scales. Single-cell transcriptomics generally captures the macroscale, lacking time resolution and fitting pre-determined models to distribution data, whereas live-cell imaging using MS2 or PP7 bacteriophage coat protein reporter systems offers very high temporal resolution and therefore probes the microscale. Even gene length matters: imaging a long gene captures macroburst dynamics as multiple polymerase convoys progress simultaneously, while a short gene reveals individual microbursts.

Armed with this framework, the authors turn to the epigenetic machinery. DNA methylation, long viewed as a stable, repressive mark, turns out to be far more dynamic than generally appreciated. It responds rapidly to extracellular stimuli, as seen in immune cells following bacterial infection, primarily at distal enhancers, and in post-mitotic neurons, where 1.4 percent of assessed CpG sites rapidly underwent demethylation or de novo methylation after stimulation. Intriguingly, 87 percent of differentially methylated CpG sites showed decreased methylation after an associated increase in transcription factor binding, suggesting that changes in DNA methylation are often a consequence rather than a cause of altered gene expression. Multiple transcription factors can recruit the methyltransferases DNMT3A or DNMT3B to trigger promoter hypermethylation, and methyl-CpG-binding proteins such as MeCP2 recruit repressive complexes, although MeCP2 has also been shown to activate genes and induce extensive chromatin unfolding in models of Rett syndrome.

The genomic location of methylation determines its effect on noise. Promoter methylation has been implicated with higher levels of transcriptional noise, apparently acting as a trigger that forces ON-to-OFF state transitions and increases the burstiness of genes, while unmethylated promoters can remain active and quieter. Gene body methylation, by contrast, is associated with reduced noise, a finding replicated in Arabidopsis thaliana and the sea anemone Exaiptasia pallida, and may work by suppressing transcription from cryptic intragenic promoters or by inhibiting deposition of the histone variant H2A.Z, which is itself linked to increased noise. In mouse embryonic stem cells, DNA methylation was identified as a key regulator of stochastic phenotypic state switching, although studies of Polycomb-targeted active genes found no such effect, underscoring the context dependency of these mechanisms.

Histone variants and nucleosome density add another layer. The H3.3 variant has a turnover time of approximately two hours, strikingly similar to the interval between transcription bursts for hormone-responsive genes, suggesting that H3.3 occupancy could act as a form of transcriptional memory; indeed, H3.3 and its methylatable fourth lysine residue are essential for retaining memory of gene activity states after nuclei transplantation. The chaperone HIRA deposits H3.3 at decondensed, transcriptionally active promoters and enhancers, while DAXX deposits it in condensed inactive chromatin, recruiting histone deacetylase II to maintain repression. The variant macroH2A, when paired with the transcription factor NRF-1, reduces noise by maintaining the ON state, whereas macroH2A nucleosomes lacking NRF-1 may promote stochastic ON-OFF switching. Independently of variant identity, a higher density of promoter-proximal nucleosomes was found to lower burst frequency and increase expression noise, slowing the rate of OFF-to-ON transitions.

Histone post-translational modifications show equally nuanced effects. Broadly, histone acetylation marks active genes, and evidence is accumulating that it mainly regulates burst size: the HDAC inhibitors trichostatin and SAHA increased the burst size of latent HIV production and, in a PP7-tagged cell line, predominantly affected burst size. Yet site-specific epigenetic rewriting using CRISPR-based dead Cas9 fusion proteins has revealed that targeted promoter acetylation can predominantly influence burst frequency, as at the circadian Bmal1 promoter, while enhancer acetylation increased burst duration of the FOS gene in neurons, mediated by the reader protein BRD4 and the release of paused RNA polymerase II. Histone methylation appears more tied to transcriptional memory: H3K4 methylation persists along gene bodies in yeast for over an hour after transcription ceases, transmits burst frequency between mother and daughter Dictyostelium cells, and in breast cancer cells, broad H3K4me3 peaks were associated with more homogeneous expression, while narrow peaks accompanied heterogeneous expression and drug resistance via the demethylase KDM5B.

Finally, the three-dimensional organisation of the genome and its enhancer networks constrain these dynamics. Macro-level features such as A and B compartments and topologically associated domains appear tied to stable, lineage-specific expression rather than acute bursting, although TAD disruption can cause developmental malformations by rewiring enhancer contacts. Enhancer-promoter loops, which can span up to three megabases, are far more dynamic: deleting specific CTCF sites in T helper 2 cells increased gene expression noise by destabilising loops, while forced looping and stronger enhancers in Drosophila increased burst frequency without changing burst size. Genome-wide analyses confirm that enhancers primarily modulate burst frequency, and the mediator complex drives the rapid succession of initiation events that produces a burst of polymerase activity. The authors conclude that resolving the true genome-wide effects of epigenetics on bursting will require high-throughput imaging of thousands of loci under epigenetic perturbation, but with such techniques on the horizon and single-cell transcriptomics exploding, the field stands on the verge of finally connecting the epigenetic code to the pulsating rhythm of gene activity.

Subject of Research: Epigenetic regulation of transcriptional bursting and gene expression noise in eukaryotic cells

Article Title: Transcription bursting and epigenetic plasticity: an updated view

Article References: Beckman, W. F., Jiménez, M. Á. L., & Verschure, P. J. (2021). Transcription bursting and epigenetic plasticity: an updated view. Epigenetics Communications, 1(1), Article 6. https://doi.org/10.1186/s43682-021-00007-1

Image Credits: AI Generated

DOI: 10.1186/s43682-021-00007-1

Keywords: transcription bursting, epigenetics, gene expression noise, DNA methylation, histone variants, histone modifications, chromatin architecture, enhancers, RNA polymerase II, topologically associated domains, transcriptional memory, single-cell imaging