The method in the madness: Transcriptional control from stochastic action at the single-molecule scale

Abstract

Cell states result from the ordered activation of gene expression by transcription factors. Transcription factors face opposing design constraints: they need to be dynamic to trigger rapid cell state transitions, but also stable enough to maintain terminal cell identities indefinitely. Recent progress in live-cell single-molecule microscopy has helped define the biophysical principles underlying this paradox. Beyond transcription factor activity, single-molecule experiments have revealed that at nearly every level of transcription regulation, control emerges from multiple short-lived stochastic interactions, rather than deterministic, stable interactions typical of other biochemical pathways. This architecture generates consistent outcomes that can be rapidly choreographed. Here, we highlight recent results that demonstrate how order in transcription regulation emerges from the apparent molecular-scale chaos and discuss remaining conceptual challenges.

Figure 1:. Visualization of transcription at the single molecule level in living cells.

(a) In live-cell nascent mRNA imaging, the MS2 Coat protein (MCP) fused to a fluorescent protein, binds to a target RNA containing genetically encoded MS2 Binding Sites (MBS) (left, center). The fluorescence intensity of transcription sites over time provides access to bursting kinetics (right). Multiple orthogonal stem loop sequences that bind unique RNPs, such as PP7-PCP, enable the imaging of different RNA species simultaneously. Extended review on these technologies can be found here: [83]. (b) In SMT, transcription regulators are labeled with fluorescent dyes (left). Sparse labeling via fluorophore photoactivation (or limited dye conjugation) allows the sequential tracking of individual molecules free of background (middle). Single-molecule trajectories provide access to the fraction of bound vs. freely diffusing proteins, and their residence times on chromatin (right). (c) Multiple binding sites for fluorescently-labeled proteins inserted near regulatory elements (left) enables measuring looping dynamics (middle, right). Cartoons (right) represent typical metrics.

Figure 2:. Myths vs. reality at the single molecule scale.

(a) Gene expression was originally conceived as uniform across a cell population (top), but is highly heterogeneous (bottom). (b) Active genes do not produce mRNAs continuously (top), but in intermittent “bursts” (bottom). While bursting in yeast is adequately described by simple on-off models, metazoan bursting is likely driven by more regulatory states [84]. (c) TF binding to enhancers is not stable (top), but rather turns over within seconds (bottom). (d) Cofactor recruitment often does not occur via a conventional “lock and key” protein-protein interaction (top), but rather is the product of weak, multivalent interactions encoded by disordered domains (bottom). (e) Enhancer-promoter loops do not represent long-lived molecular scale contact (< 30 nm) in live imaging experiments (top); chromatin configurations consist of a stochastic dynamic ensemble (bottom).

Figure 3:. The length scales of transcription control. (a) Representation of various players at scale.

Data sources: C-terminal Domain of Pol II radius of gyration, 6 nm [85]; Mediator Pre-Initiation Complex, 27 nm [86]; Single-molecule tracking positioning accuracy ~50 nm; human enhancer-promoter median separation ~ 200 nm based on the trend measured by DNAseqFISH+ [87] for a typical separation of ~21 kbp [67] (see c); permissive radius, 400 nm, from [65,68]. (b) Enhancer-Promoter physical distances of various loci measured in their inactive (light gray circles) vs. active (colored circles) states, overlaid with the median physical distances measured for 36,599 loci pairs in mESCs (dark gray curve) [87]. Data sources: Sox2¹ [51], Sox2² [60], Sox9 [88], Nanog [89], SHH [54], eve [90], Hox [91], synthetic enhancers [58]. + denotes experiments measuring enhancer-promoter distances separately on active and inactive alleles within the same cell type; * denotes experiments measuring enhancer-promoter distances in two cell types, one in which the gene is inactive, and one in which the gene is active (i.e. without sorting individual alleles based on their activity state); o denotes an experiment measuring the enhancer-promoter distance in different cells where the gene is either inactive (gray), active and regulated by the enhancer (color), or active due to the action of a separate enhancer than the one which distance is plotted (gray). (c) Cumulative probability of the physical distances separating pairs of loci separated by 25 kbp in mESCs. From a full dataset of 1119 loci pairs, 51 loci pairs sampling the 1–99 percentiles of median physical distances are plotted [87].

Figure 4:. Models of enhancer function.

(a-b) Transcription activity exhibits a hypersensitive response to enhancer distance if the promoter continuously cycles through multiple states, a fraction of which are transcriptionally active (marked by increasing coactivator molecules or activating marks, a), and the enhancer deposits co-activators or activating marks upon brief contacts (b) [73]. (c) In the hub model, brief TF binding events nucleate local concentration of activating factors around the enhancer and promoter via a network of weak protein-protein interactions [26]. (d) In the Transcription Factor Activity Gradient model [76], TFs are activated (e.g. via acetylation) during brief binding events at enhancers occupied by coactivators; upon diffusing away, activated TFs are captured by a nearby promoter where they trigger transcription.