Density transitions in the regulation of transcription

SUMMARY

Transcription of the right genes at the right time is crucial for physiology and often dysregulated in disease. For a process that requires precise outcomes, the underlying molecular interactions regulating transcription can be surprisingly dynamic and short-lived. Various sources of multivalency have been proposed to stabilize short-lived interactions at specific genomic loci. Using multivalent contacts, components of the transcription machinery form complexes and above threshold concentrations can form condensates by undergoing reversible density transitions (i.e., phase separation). Importantly, coupling the density transition to networking generates a condensate-spanning network with emergent properties that soluble complexes do not possess, including enhanced dwell times of relevant components, distinct chemical environments, capillary forces, and biochemically active interfaces. Here we will review the current evidence for and against the role of density transitions in regulating transcription, discuss best practices for studying the functional role of these phenomena, and consider how emergent properties may regulate transcription.

eTOC blurb:

Navalkar, Eppert et al. review the role condensates may play in transcription. The review clarifies differences between molecular machines, soluble higher-order complexes, and condensates formed via density transitions. It highlights emergent properties of condensates and outlines approaches for dissecting the functional relevance of density transitions in transcription across biological contexts.

INTRODUCTION

Gene activation is a multi-step process that requires the coordination of dozens of factors in space and time. This multi-step process involves several rate-limiting intermediates starting with the interaction of transcription factors (TFs) with DNA regulatory elements and culminating in the assembly of an elongation-competent RNA Polymerase II ^1,2. While stable, stoichiometric complexes (referred to here as molecular machines; see Box 1) play important roles in transcription ^1,2, many of the crucial interactions that coordinate the multiple steps (Box 2) are surprisingly dynamic and short-lived ^3–7 as evidenced by short residence times on chromatin ⁸. Despite the low probability of many short-lived interactions occurring simultaneously, transcription in cells occurs in sustained periods of activity described as bursts ^9–11. In cells, the degree of transcription is regulated not by tuning a consistent rate of transcription but by increasing or decreasing how frequently genes can enter active states (i.e., the burst frequency) and how many elongation-competent RNA Polymerase II molecules are assembled during an active state (i.e., the burst size). While significant attention has been focused on how dynamic interactions influence target search and the specificity of DNA-binding TFs ^5,7,12–14, many more components must be locally concentrated together to activate gene transcription. The relevant question becomes: how do many dynamic and short-lived interactions lead to bursts of transcription in cells? Phrased another way: what is the physical basis for the switch between active and inactive states that is mediated by many dynamic and short-lived interactions?

Box 1: Glossary.

Molecular machines are macromolecular complexes with specific stoichiometry. The subunits typically assemble with high affinities, which means that this class of macromolecular complexes can be purified and visualized by structural biology approaches. We use this term to emphasize the differences between this class of complexes and the higher-order complexes that form through more dynamic and shorter-lived interactions. The term “machine” is used in the most general sense of an entity with multiple parts that performs a specific function irrespective of energy consumption or enzymatic activity. Examples of molecular machines include the Mediator complex and the RNA polymerase II holoenzyme.

Higher-order, soluble complexes:

Molecules that are multivalent for each other form higher-order complexes of varying stoichiometries with broad size distributions that are determined by the constituent concentrations, valencies and affinities ^15,16. This means that the complexes do not grow over time. They have also been referred to as pre-percolation clusters ¹⁵, given the fact that they have not crossed the percolation concentration. The lack of system-spanning percolation means that the effective off-rates of individual molecules can be high and are higher than off-rates from condensates of the same constituents. Pre-percolation clusters can exist below the saturation concentration and coexist with condensates. Examples include transcription factor–coactivator complexes that have been characterized in detail, including the Gcn4/Med15 complex ⁴⁵.

Condensates:

Condensates are nonstoichiometric assemblies that form via phase separation coupled to percolation, a process that is driven by multivalent interactions and poor solubility ¹⁹. When the solubility limit of biomolecules is exceeded, they cannot be further enriched in the dilute phase and are instead sequestered into the more concentrated dense phase. The dilute and dense phases co-exist in equilibrium and are thus equally thermodynamically favorable. Hence, phase separation is the consequence of a solubility gap, and the solubility of the biomolecules in the dilute and in the dense phase is equally high. In case multiple macromolecules contribute to the formation of the core percolated network, threshold concentrations depend on the input concentrations and molar ratios of the constituents. The condensate-internal percolated network generates long lifetimes, i.e., condensates do not typically dissociate and reform as soluble complexes do, even though individual molecules leave condensates and others join. Even on a per-molecule basis, the contact time is longer than in pre-percolation clusters, and this was shown to mediate selective activation downstream of phase-separated membrane receptors. Condensates can grow over time via fusion and Ostwald ripening, although processes such as dynamical arrest, liquid-to-solid transitions and attachment to chromatin may hinder this. Examples include stress granules ^128,129,147 and condensates formed by transcriptional machinery, including those promoted by Mediator and RNA Polymerase II.

Box 2: The multicomponent and dynamic process of gene activation.

The molecular details of the multi-component and multi-step process of gene activation have been reviewed by others ^1,2,148. Here, we provide a brief and simplified overview of the multi-step process with a focus on the roles of IDR-mediated interactions. Briefly, short-lived TF-DNA complexes recruit a series of chromatin-modifying coactivators through dynamic, multivalent contacts between TF IDRs and (often multiple ¹⁴⁹) domains of coactivators. Chromatin modifying coactivators enhance chromatin accessibility and provide binding sites for TFs, which recruit the Mediator complex and pre-initiation complex members through IDR-domain interactions to position RNA Polymerase II at the promoter. After promoter escape, RNA Polymerase II pauses, which is overcome by pTEFb-mediated phosphorylation and elongation factor recruitment. Beyond this phospho-regulatory switch, how elongation factors are recruited to pause-released RNA Polymerase II is not well understood, but it is proposed that recruitment occurs via low-affinity interactions involving IDRs ^150–153. Interestingly, the sites on RNA Polymerase II bound by elongation factors are occluded by factors involved in pre-initiation, initiation or pausing ¹⁴⁸. While some pre-initiation components can be recycled for new rounds of re-initiation ¹⁴⁸, elongation factors travel with the elongating polymerase and must be newly recruited to each RNA Polymerase II molecule. Furthermore, the paused RNA Polymerase II prevents the initiation of a second RNA Polymerase II molecule ^154–156. Although soluble complexes are sufficient to assemble elongation-competent RNA Polymerase II ²⁸, how multiple rounds of re-initiation and elongation are achieved during transcriptional bursts remains unresolved ¹⁴⁸.

A reversible local environment in which rate-limiting components are concentrated could enable such sustained transcription. Indeed, high local concentrations of gene activation machinery have been observed and associated with positive regulation of transcription ^{5,22,23,55–64,157–160}. Despite these observations, the physical basis for the generation of these high local concentrations remains debated.

Recent evidence supports models wherein multivalent interactions involving intrinsically disordered regions (IDRs) and repeats of binding domains of protein components mediate the formation of higher-order but soluble complexes ^15,16 or phase-separated networks (i.e., condensates ^17–19; see Box 1), which locally concentrate relevant factors and increase the frequency of gene activation by stabilizing crucial short-lived interactions at specific genomic loci. While higher-order complexes can mediate some of the functions of condensates, condensates have emergent properties that smaller complexes lack. Condensates formed via reversible density transitions ^17–19 may act as switches by locally concentrating the components required for gene activation and excluding unrelated proteins or negative regulators²⁰. By creating on-demand local proteomes ^20–23 through the selective stabilization of specific short-lived interactions, condensates may be able to increase the frequency of gene activation. While individual transcription factor and RNA polymerase II binding events are transient (typically lasting seconds to minutes), condensates often persist for substantially longer durations. The extended lifetimes of condensates may generate an environment that supports repeated cycles of transient binding, release, and rebinding. This dynamic but continuous exchange may enable sustained transcriptional activity and can underlie the temporal patterns characteristic of transcriptional bursting. Single-molecule and live-cell imaging studies reveal that transcription factors display dynamic binding behavior, with transient search events on sub-second to second timescales and more stable, sequence-specific binding lasting a few seconds in yeast cells ²⁴. Similar kinetics have been observed across developmental contexts, where TF dwell times last for around tens of seconds during zygotic genome activation in zebrafish ²⁵. Single-molecule imaging in Drosophila embryos showed typical TF residence times in the second range ²². Further, in mammalian cells, Mediator and Pol II clusters exhibit dynamic exchange (within seconds) and condensate persistence on the minute scale ²⁶. Several studies also establish that RNA polymerase II exhibits promoter-proximal pausing over tens of seconds to minutes ²⁷. These findings collectively highlight that transcriptional assemblies function across a broad spectrum of dynamic timescales in a context-dependent manner as captured by state-of-the-art single-molecule approaches. Based on the above findings, condensate formation of transcriptional proteins may be relevant for facilitating interactions on multiple timescales affecting the efficiency, specificity, and context-dependent regulation of transcription.

While there is evidence to support the role of density transitions in transcription regulation, an alternative model is that soluble complexes form large, higher-order assemblies without the need to invoke a density transition. Notably, the biochemical process of polymerizing ribonucleotides into RNA, i.e., the most basic definition of the term transcription, does not require a density transition. Soluble complexes are sufficient for the assembly of elongation-competent RNA Polymerase II ²⁸ and for basal transcription in vitro ^29,30. However, it remains unclear whether soluble complexes alone are sufficient to sustain transcriptional bursting within cells, or if periods of high-frequency transcription instead depend on protein networks formed through density transitions—namely, condensates that concentrate multiple copies of key molecular machines and cofactors.

In this review, we will define terms related to molecular machines, higher-order but soluble complexes, phase separation, and condensates (see Box 1). We will review the evidence for the role of density transitions in transcription and methods to evaluate this role, and we will discuss emergent properties of condensates that distinguish them from soluble complexes and may play roles in transcriptional processes.

Molecular machines, higher-order complexes and condensates: how are they different?

Progress in molecular understanding of transcription has been achieved via reconstitutions of biochemical activities of soluble molecular machines (e.g. TF-DNA complexes, nucleosomes, RNA Polymerase II, Mediator, etc.) and via the determination of their structures. These structures provide important clues regarding how these molecular machines assemble and function, but they do not visualize conformations and interactions of the long intrinsically disordered regions (IDRs) that many of the subunits contain. IDRs elude characterization using traditional structural biology techniques ³¹ because they do not adopt unique three-dimensional structures. Several key examples of functionally important IDRs in transcription include the C-terminal domain of RNA Polymerase II ^32–34, transactivation domains of TFs ^35,36, and histone tails ^37,38. Beyond these key examples, there are many more poorly characterized IDRs. As a group, proteins involved in the regulation of chromatin and transcription are among the most overrepresented for IDRs ^39,40. Although these IDRs do not fold into defined structures, they encode a variety of conformational and dynamical preferences via their amino acid composition and patterning. The molecular grammars underlying these preferences can be grouped into distinct classes of chemical diversity associated with distinct cellular functions ^41–43. In addition to the high-affinity interactions that mediate the formation of molecular machines, dynamic interactions—often mediated by IDRs and involving aspects of multivalency—are necessary for the function of transcription regulatory proteins within cells ^{5,39,40,44,45}. The IDRs often carry out their functions through interactions with multiple partners, often with low affinities relative to the interactions holding the molecular machine itself together. However, the progress through structural biology approaches is mainly constrained to the folded parts of molecular machines.

Despite over thirty years of knowledge of the functional importance of IDRs in gene activation ⁴⁶ and of their conformational heterogeneity, dynamics and interactions ⁴⁷, the lack of fixed structure has generated challenges for understanding their function. While some IDRs fold upon binding via defined binding motifs, other IDRs interact with their binding partners in a highly dynamic fashion and form dynamic or fuzzy complexes ^45,48,49. If interactions are multivalent, the result can be soluble higher-order complexes (see Box 1) that adopt wide size distributions that shift to somewhat larger sizes at increasing concentrations ¹⁵. If they are poorly soluble, they may eventually form condensates above threshold concentrations as the result of a density transition. A well-studied example of such a system is the yeast TF Gcn4 and its coactivator subunit Med15, which can form higher-order complexes ⁴⁵ and condensates ⁵⁰ depending on the concentrations of DNA, TF, and coactivator ⁵¹. The complexes and condensates can be modeled in vitro with Gcn4 and the Med15 subunit of the Mediator complex, but the condensates that may be relevant in cells are unlikely to be composed only of individual subunits. In cells, multi-component condensates will involve full Mediator, which has many subunits with long IDRs, and potentially other components. The resulting condensates will thus have internal networks formed by collections of molecular machines, not individual subunits or IDRs. The formation of some of the more labile complexes relevant for transcription may even be promoted and stabilized within condensates.

How should one think about the differences between molecular machines, higher-order but soluble complexes and condensates, given that they all represent large protein (or protein–nucleic acid) assemblies (see Figure 1)? Molecular machines are stable macromolecular complexes with specific stoichiometries. Molecular machines can interact with each other, and if they possess inherent multivalence, e.g., encoded in multiple interaction domains or in IDRs, they will form higher-order complexes. Given the relatively weak interactions between these molecular machines, these complexes will have relatively short lifetimes, and the largest species occur rarely. While short lifetimes are likely sufficient for stimulating basal transcription in vitro ³⁰, they may not sustain bursts of transcription. Condensates with their internally percolated networks promote rebinding after individual contacts are lost and thereby generate effectively longer-lived interactions. These types of extended lifetimes are known to be exploited for favoring rare processes. A well-studied example is found in signaling pathways downstream of the Nephrin transmembrane receptor and the scaffold protein linker for activation of T cells (LAT), wherein multivalent interactions between these membrane receptors and downstream adaptor molecules mediate clustering and induce co-condensation of receptors and adaptors at the membrane ^52,53. Signaling effectors such as “Son of Sevenless” (SOS) are recruited to these condensates, and integration into the condensate network dramatically increases their residence time, or dwell time, relative to recruitment to individual, activated receptors. Given that activation of effectors such as SOS is typically a slow, rate-limiting step in the signaling cascade, the long dwell times in condensates increase the likelihood of activation. By contrast, short dwell times of effectors recruited to unclustered receptors are much less likely to result in activation. This well-documented example shows that the effective long-lived interactions within condensates are used to potentiate and regulate biological processes. While large higher-order complexes can also favor rebinding and thereby enhance dwell times, such large complexes are rare (Figure 1). By contrast, density transitions enable the formation of a considerable fraction of very large assemblies (Figure 1). Analogous condensate-mediated mechanisms may also operate in transcriptional processes, where condensates of transcriptional machinery may potentiate the otherwise rare event of a transcriptional burst.

Figure 1: Features of molecular machines, higher-order, soluble complexes, and condensates.

(A) Molecular machines are stable macromolecular complexes with specific stoichiometries and are typically amenable to structure determination. The structures of the RNA Polymerase II complex (PDB ID 7B7U) and the Mediator complex (PDB ID 8TQW) are shown as examples. The RNA Polymerase II (represented in yellow) encloses the DNA template and positions it for transcription, featuring a central cleft and a mobile clamp domain. The intrinsically disordered C-terminal domain (CTD) acts as a scaffold for the assembly of various co-transcriptional regulators, including the Mediator complex and cofactors. The Mediator complex (represented in blue), a modular coactivator, shown here with head, middle, tail, and kinase modules, forms a dynamic bridge between gene-specific transcription factors and RNA Polymerase II.

(B) Higher-order complexes are soluble, non-stoichiometric assemblies of multivalent proteins. The multivalence can be encoded by tandem repeat interaction domains or within repetitive IDRs (see Box 1). If molecular machines encode multivalent interactions, they can assemble into higher-order complexes. The size distribution of higher-order complexes (here shown on a log scale for visualization purposes) is broad and heavy-tailed ¹⁵. Increasing constituent concentrations generate somewhat higher fractions of large assemblies and increase the maximal accessible size, but the size distribution is always dominated by smaller species (yellow, orange, and red lines).

(C) Above threshold concentrations, higher-order complexes with poor solubility can form condensates via a density transition coupled to percolation. Under these conditions, the assemblies can become much larger (blue line), and a larger fraction of the constituents can be incorporated into the condensate (blue peak). The condensate-internal network promotes rebinding and effective lifetimes of biomolecules in the assembly, favoring transient binding events.

(D) The underlying multivalent interactions that form higher-order soluble complexes and condensates are often the same and include homotypic and heterotypic interactions among IDRs and globular domains of proteins. Various combinations of homo- and hetero-typic interactions between globular domains and IDRs are illustrated here. While RNA plays a role, it is not shown here for simplicity.

(E) An example of two multivalent proteins that interact, one containing repeats of SH3 domains and the other containing repeats of proline-rich motifs (PRMs), which bind to the SH3 domains. These can form higher order complexes and above threshold concentrations condensates driven by a density transition coupled to networking of these molecules. Above threshold concentrations, higher-order complexes and condensates coexist. Adapted from (Harmon et al., 2017)¹⁴⁶.

Components of the transcription machinery are frequently observed to form regions of high local concentrations in cells, often coinciding with sites of nascent transcription. This phenomenon has been observed across diverse biological systems, ranging from human and mouse cell culture to live organisms. Several studies have reported colocalization of high local concentrations of DNA-binding TFs, coactivators, RNA Polymerase II, and other regulatory factors with nascent transcription at a wide range of gene classes ^{22,23,54–64}. This association between high densities of regulatory proteins and bursts of nascent transcription—that is consistently observed across a wide range of techniques, methods, organisms, and genes—suggests a role in regulating transcription but does not demonstrate causality. Moreover, despite the expectation that large, higher-order complexes should be rare (Figure 1B), observations of high local densities cannot unequivocally determine whether they arise from a density transition or represent a rare soluble, higher-order complex. However, the appearance of a large number of high densities as concentrations increase clearly points to a density transition. Therefore, the consistent correlation between high local concentration and transcriptional bursting suggests the possibility that condensate-mediated mechanisms—similar to those described above for transmembrane receptors—may contribute to transcriptional bursting.

This possibility raises two critical questions: Are the multivalent soluble complexes formed without density transition sufficient? And do condensates make additional contributions to the regulation of transcription? Multivalent interactions among soluble complexes may well be sufficient to activate genes in the absence of a density transition. However, we propose that condensates have emergent properties that may favorably contribute to the regulation of transcription. Specifically, we propose that long-lived interaction networks within condensates can effectively stabilize crucial short-lived interactions, thereby increasing the frequency of gene activation. Distinguishing between these two models is experimentally challenging for technical and conceptual reasons, and the role of density transitions in transcription, therefore, continues to be debated ^65,66. First, establishing that a density transition has taken place in cells is technically challenging. Second, even if condensates are present, soluble complexes will coexist with them, and the source of the activity may be challenging to identify. Ideally, one could identify mutations that only influence density transitions without impacting complex formation and vice versa. While partial separation-of-function mutations should be possible ^19,51,67, the overlap of the interactions driving the formation of complexes and the condensate-interior networks (Figure 1E) makes this task challenging. Given the different concentration dependence of both types of assemblies, titrations of protein levels are likely necessary to evaluate the functional roles of the assemblies, with the caveat that dominant-negative effects from predominantly homotypic interactions at super-physiological concentrations need to be avoided, as we lay out below in more detail.

Methods to investigate roles for density transitions in transcription

While there is no single method that will definitively distinguish between the contribution of soluble complexes versus condensates, a number of approaches are available to compare transcription of genes under conditions where the protein of interest is soluble versus in condensates. These approaches include imaging the temporal relationship between condensates and transcription, investigating cells with protein concentrations below and above threshold concentrations for condensate formation, inducing synthetic multimerization of proteins to artificially drive density transitions, manipulating protein sequence to disfavor or promote density transitions, and strategies to solubilize proteins. Each of these strategies needs to be carefully implemented, and alternative models and potential caveats should be considered for the interpretation of the results. In this section, we will discuss the various approaches, their benefits and caveats (Figure 2), and evidence for and against roles density transitions may play in transcription using these approaches.

Figure 2: Methods to investigate the role of density transitions in transcription.

Each method involves the observation of whether condensates are present and how they relate to transcriptional activity. As discussed in the main text, each method has its benefits and caveats that need to be acknowledged in the interpretation of results.

Temporal relationship between condensates and transcription

Insights into the relationship between high local concentrations of transcriptional machinery and transcriptional output has come from live cell imaging approaches where the spatial relationship and relative dynamics of protein accumulation and transcription can be monitored over time. These live-cell approaches demonstrate that areas of high local concentration of the transcription machinery appear, grow, and dissipate on the same timescale of seconds to minutes as individual bursts of transcription. Investigation of Actb transcription in mouse embryonic fibroblasts found that clusters of RNA Polymerase II form at the Actb gene locus prior to an observed burst of transcription ⁵⁴. Furthermore, this study found that the lifetime of the RNA Polymerase II cluster was directly correlated with the number of transcripts produced per burst. In Drosophila blastoderm nuclei, lattice light-sheet tracking revealed that the pioneer factor Zelda assembles “multifactor hubs”, i.e., areas of high local concentration wherein the morphogen Bicoid is preferentially captured and promotes interactions with DNA. High local concentrations of Bicoid were spatially associated with transcription of its target gene hb²². These high local concentrations containing Bicoid also seem to correlate with increased transcriptional kinetics in a gene-specific manner⁶⁸. In developing Drosophila embryos, imaging of the burst kinetics of a reporter gene driven by a synthetic TF showed the rapid accumulation of high local concentrations of TFs preceding a transcriptional burst. These high local concentrations remain proximal throughout the burst period and dissipate upon burst termination. Manipulating the sequence of the synthetic TF to increase high local concentrations increased both burst size and frequency ⁶². Active-target-locking 3D nanoscopy at the Nanog gene locus in mouse embryonic stem cells (mESCs) showed a high local concentration of RNA Polymerase II, Mediator, BRD4 and other cofactors, and the cluster size correlated with transcriptional burst size. Pharmacological shortening of the cluster size curtailed RNA output ⁵⁷. By simultaneously imaging either RNA Polymerase II or Mediator and transcription at the Sox2 gene locus in mESCs, Du et al. observed a positive correlation between the proximity of high local protein concentrations to the gene and the burst size and frequency, demonstrating a direct temporal and spatial relationship ⁶⁹.

While in most of these examples only a single protein was labeled, other observations of transcription suggest that RNA Polymerase II and the dozens of factors required for gene activation are dynamically concentrated. Indeed, Li et al. tagged multiple components of the transcription machinery, showing that they were all locally concentrated ⁵⁷. Collectively, these studies converge on a simple chronology: the formation of a high local concentration of many components (henceforth termed ‘clusters’, which can refer to condensates or large, higher-order complexes) precedes promoter firing, the lifetime or size of the cluster scales with burst output, and cluster dissolution or movement away from the locus marks the return to inactivity. The kinetics of RNA production that tightly matches the sharp formation and dissolution of clusters may suggest a role for reversible density transitions that could create on-demand high local concentrations of the various short-lived complexes. However, these data do not explicitly distinguish whether function arises through condensates or soluble complexes. The benefits of these observational approaches are that they observe the process as it is occurring with minimal perturbation. Nevertheless, to test whether high local concentrations and the effect on transcriptional bursting are mediated through density transitions or soluble complex formation, careful and thoughtful perturbation followed by investigation of outcomes is required.

Measuring activity across a range of concentrations

Density transitions are discontinuous transitions, meaning they exhibit a strongly cooperative concentration dependence. Determining whether the activity of the protein(s) in question also exhibits a discontinuous transition over the same concentration range represents a valuable test of the functional role of density transitions. Given that soluble complexes coexist with condensates, the concentration dependence of complex formation will also have to be evaluated. Another possibility is inspecting whether the product is produced in the condensate or in the soluble phase. The earliest biochemical reconstitution of condensates in vitro of the nephrin-NCK-N-WASP actin polymerization reaction showed that rates of actin polymerization sharply increased above concentration thresholds for condensate formation and found actin filaments associated with condensates ¹⁶, and emanating from condensates ⁵². Similar experiments were performed with transcriptional machinery in cell-based assays, where changes in transcriptional coactivator activity were observed across a range of protein concentrations ⁶³.

Changes in the concentration of cell-type specific transcription regulators drive the regulation of cell fate decisions ^70–73. Many TFs and coactivators are only present at detectable levels during cellular differentiation and are often self-regulating in a positive feedback loop to reinforce their continued expression ^74,75. By investigating the physiological change in concentration of Myocardin (MYOCD), a smooth muscle specific coactivator, Gan et al. identified its threshold concentration for the formation of foci and found smooth muscle-specific transcripts emanating from high local concentrations of MYOCD foci in iPSC-derived smooth muscle cells ⁶³. To test whether the transcriptional activity was dependent on condensate formation, the physiological concentration range of MYOCD was modeled in orthogonal cell lines by ectopic expression within the physiologic range. Cells with MYOCD at concentrations below the threshold for condensate formation did not express relevant protein products from a reporter, but cells above the threshold concentration for condensate formation expressed reporter protein products. Follow-up work to investigate the real-time production of nascent RNA is still needed to connect the concentration-dependent condensate directly to transcription. These types of experiments require single-cell resolution approaches given that bulk measurements of transcription report on ensembles of cells with different concentrations above and below condensate thresholds.

The approach described above relies on the heterogeneous expression of the protein in question across cells and selection of cells that are within physiological concentration ranges. A benefit of this approach is that it uses a naturally occurring variable and does not require the introduction of artificial manipulations if the tested concentration range is physiological. Given that concentration is an important variable for driving transcription, supraphysiological overexpression is often challenging to interpret. E.g., supraphysiological overexpression of transcriptional regulators may lead to large, non-functional condensates formed by aberrant homotypic phase separation ⁷⁶ or sequestration of other transcriptional factors in non-functional, ectopic condensates. Such technical artifacts of supraphysiological expression were observed at the highest concentrations when transcription factor functions were observed over large concentration ranges ^77,78. The dominance of homotypic interactions leading to dysfunction was shown to be a mechanism of disease for alanine repeat expansions of HOXD13 in polydactyly ⁷⁹. These studies highlight the importance of conducting experiments within biologically relevant concentration regimes or with approaches that enable tracking of dominant-negative effects.

Several studies used experiments wherein the concentration of TFs was varied to test the relevance of TF density transitions for function and concluded that they were not needed or even detrimental. However, some of this evidence is inconclusive. It has been argued that the absence of a discrete saturation concentration demonstrates the absence of a density transition ⁶⁵, but multicomponent phase separation results in saturation concentrations that depend on the concentrations of all components and are therefore variable ^16,80,81. In another example, overexpression of the disordered prion-like domains of FET family fusion oncoproteins promoted the formation of condensates but dampened activity. This was taken as evidence that phase separation does not drive transcriptional activity, but the overexpressed domains effectively functioned as dominant-negative effectors because of their missing DNA-binding domains ⁷⁷, meaning that the results do not provide general insights into the function of TF condensates. The study provides a useful example of how manipulation of condensate composition can influence transcriptional activity. In this case, dominantly homotypic condensates or partitioning of transcriptional activators into the nucleolus perturbed transcription. Given that the experiments are analogous to classic studies with GAL4 in which overexpression of the activation domain had an inhibitory effect on transcription, referred to as “squelching” ^82,83, this study ⁷⁷ provides a potential mechanism underlying the long-observed phenomenon of squelching. Another study varied the expression levels of synthetic TFs, to test whether crossing a threshold for condensate formation has an effect on transcription of a reporter gene⁷⁸. The utility of the elegant experimental design was limited in its ability to examine the impact of a density transition due to technical challenges of assigning the threshold concentration for condensate formation at the reporter gene locus. Instead, the authors defined the threshold concentration as that for the formation of extra-chromosomal condensates away from the reporter locus and then investigated the impact of crossing that threshold on transcription of the locus. Given that high expression levels can promote the formation of non-functional, extra-chromosomal condensates ⁷⁶, the chosen threshold limits the interpretation of the reported results. Previous studies using the same reporter cell line mapped a threshold for a density transition at the reporter locus ⁵⁶, which would have been the appropriate threshold to test. Despite this technical issue, the study provides compelling evidence that increasing multivalent interactions of a TF or increasing TF dwell time increases transcription.

Synthetic multimerization: inducible homotypic oligomerization

To investigate the impact of density transitions on transcription, multiple strategies are available that induce homotypic oligomerization to promote condensate formation in cells. Optogenetic and chemogenetic tools are powerful methods for correlating function and density transitions, and a particular advantage is that they enable probing of the system with and without phase separation at identical concentrations ^84–86.

Wei et al. fused optogenetic Cry2 domains to the N-terminal IDRs of TAF15 and FUS, and condensate formation induced via blue light illumination led to an increase in nascent RNA synthesis at sites of condensate formation ⁸⁶. Schneider et al. performed similar experiments with Cry2 fused to the N-terminal IDR of FUS, but included the DNA-binding domain of TetR, which allowed them to investigate expression of a reporter gene with a TetO array upstream of a minimal CMV promoter⁸⁷. Light-induced condensate formation led to expression of the reporter gene in cell culture and in animal models. Similarly, the LAMPS (Light Activated Macromolecular Phase Separation) system utilized dCas9–Cry2 fused to either the N-terminal IDR of FUS or the C-terminal IDR of BRD4 and targeted the native HS2 enhancer of the β-globin locus ⁸⁸. The LAMPS constructs were equivalently recruited to the HS2 locus in the presence or absence of light, but light-induced condensate formation resulted in enhanced expression of β-globin locus genes, increased accessibility of regulatory DNA elements, and increased enhancer–target gene contacts. Locus-specific proteomics revealed that condensate formation led to a shift in the local proteome at the HS2 locus with a depletion of negative regulators and enrichment in RNA Polymerase II and multiple transcriptional coactivators, further supporting a direct, causal relationship between condensate formation and enhanced transcriptional activity.

Chemogenetic approaches that use chemically induced oligomerization offer complementary evidence. Wu et al. employed rapamycin-induced dimerization of FKBP-tagged activators with multivalent FRB scaffolds, rapidly forming transcription factor clusters at a reporter gene ⁸⁹. Live two-color imaging of a single-copy 7×TetO reporter carrying PP7 stem-loops showed that cluster formation markedly increased the fraction of time each locus spent in the ON state, reduced the lag to the first transcriptional burst, and synchronized activation across the ~11 integrated reporter alleles. DNA-FISH confirmed physical docking of clusters at the reporter locus, and rapamycin titrations revealed that stronger clustering raised effective DNA-binding affinity and yielded switch-like expression profiles at high oligomerization propensities.

However, inducible oligomerization experiments have also illustrated limitations to the generality of condensate-mediated regulation. Yang et al. appended a tetramerizing coiled-coil tag to N-Myc, enabling rapid condensation upon chemical induction, followed by genome-wide RNA-seq analysis ⁸⁴. The forced condensation altered transcriptional regulation of only ~3% of N-Myc target genes, suggesting that most targets were efficiently activated by soluble complexes alone without the need for the presence of visible condensates. A major strength of these studies was their investigation of genome-wide transcription as opposed to single targeted gene loci or reporters.

Together, these inducible oligomerization studies offer evidence that switching transcriptional regulators between diffuse and condensed states can directly modulate transcriptional activity at specific loci. Nevertheless, the variability in the magnitude and directionality of condensate-specific effects underscores that while density transitions can strongly enhance transcription in certain contexts, they may not be universally required across all transcriptional regulatory scenarios. Inducible oligomerization approaches often rely on large tags with genetically encoded oligomerization domains on the protein of interest, which may change organization in complexes and condensates, alter the network structure and composition of condensates and thereby the functional readout. This drawback may explain why the use of synthetic homo-oligomerization methods to study transcription have led to opposing conclusions as to the role of density transitions in transcription regulation, with different studies reporting positive, negative, neutral, or mixed gene-specific effects on transcription ^{78,84–88,90}. It is likely that the nature of the homo-oligomerization and the impact on the ability of the condensates to communicate with chromatin and other critical components of transcription machinery will determine whether it is a faithful model of natural condensates. It is likely that favoring homotypic interactions between TFs and concomitantly disfavoring the heterotypic network of interactions required for TF activity is a suboptimal approach. Future iterations that preserve endogenous heterotypic interactions will be essential for generalizing the conclusions.

Manipulating protein sequence

IDRs of TFs and coactivators do not only play roles in forming condensates, they also recruit factors through protein-protein interactions and facilitate DNA target search ^12,45,91,92. Loss-of-function approaches that delete or mutate IDRs not only disrupt the density transition, but also protein-protein interactions mediated by the domains (see Box 3). These manipulations, therefore, are not sufficient to provide direct insight into the functions of density transitions, and other interpretations should be considered.

Box 3: Common Pitfalls in Assessing Functional Roles for Density Transitions.

The ability of a protein to phase separate does not inherently indicate that this mediates function. The central question is whether phase separation occurs under physiological conditions in cells and whether it regulates the activity of the protein.

Although mutations that affect both phase separation and function may appear compelling, they typically do not offer definitive proof, given that they can also alter higher-order complex formation. Distinguishing condensate-driven activity from other assembly-based mechanisms remains challenging. Bremer et al. directly compared condensate- and complex-based models and were able to generate partial separation-of-function mutants ⁵¹. Of note, most mutations that alter the density transition also alter the drive to form soluble complexes.

Loss-of-function studies that involve deleting or strongly mutating intrinsically disordered regions (IDRs) can obscure interpretation when researchers do not compare outcomes to models based on soluble complexes. Without such comparisons, observed phenotypes may reflect broader disruptions in protein assembly or function, not specific loss of condensate formation. Similarly, replacing native IDRs with orthogonal IDRs that also form higher-order complexes does not provide a definitive test of condensate-dependent function, as these substituted regions may support gene activity through other assembly mechanisms.

Several widely used approaches to attempt to assess phase separation include fluorescence recovery after photobleaching (FRAP), treatment with aliphatic alcohols, and in vitro reconstitution. While these methods can provide important insights, they require careful interpretation because they do not prove that condensates are present in cells and contribute to function. The individual limitations of these techniques (reviewed elsewhere ^161–163) highlight the need to use complementary approaches and interpret evidence for phase separation with caution.

Overexpression of proteins can drive phase transitions that may not occur under endogenous conditions. To avoid misleading conclusions, it is essential to complement overexpression studies with experiments that preserve endogenous expression levels. Moreover, extreme overexpression often leads to the formation of large, off-chromatin condensates that are likely driven by homotypic interactions. Such condensates may not recapitulate the diverse, multivalent interactions required for transcriptional activation.

Use of fluorescent tags or large oligomerization domains may perturb the system and generate behaviors that may not occur with endogenous proteins, including solubilization, which suppresses phase separation, or oligomerization, which can drive phase separation ^164,165. Alternative methods using small epitope tags—ideally through knock-in—or minimally perturbative tagging strategies are preferable.

To ensure robust conclusions, the quantification of effects across cell populations is advantageous. Drawing inferences from single images or rare events risks misinterpretation due to variability in cell state, imaging artifacts, or stochastic fluctuations. Statistical rigor and replication are essential to determine whether condensate formation plays a specific and physiologically meaningful role in regulating protein function.

In conclusion, a rigorous, multifaceted approach—grounded in physiological context, physical principles, comparative modeling, and quantitative analysis—is essential to accurately define when and how condensate formation contributes to protein function in cells.

Making more subtle mutations to the amino acid sequence of TF IDRs can also lead to unwarranted conclusions, even if the IDRs are well annotated for the presence of binding motifs. Recent work has highlighted that the same interactions that mediate networking in condensates also lead to the formation of soluble complexes below the saturation concentration and has therefore pitted two models, i.e., the formation of condensates versus soluble complexes, against each other ⁵¹. Bremer et al. determined in vivo transcriptional activities of nearly 30 variants of the yeast transcription factor Gcn4, where some variants carried mutations in annotated binding motifs, others outside of these. The transcriptional activities were compared with the abilities of the variants to form complexes with Med15, which is its cognate Mediator subunit, versus the abilities to recruit Med15 into condensates ⁵¹. The affinities and phase separation abilities were highly correlated and explained the activities similarly well, highlighting the difficulty of separating the functions of soluble complexes and condensates by mutations. However, a small number of strongly interacting variants had activities that were better explained by phase separation rather than the formation of soluble complexes, and these suggested that internal networking of TF condensates could dampen activities ⁵¹.

Mutations with validated biophysical effects on phase separation can be useful. Naderi et al ⁶⁷ leveraged biophysical insights made by Martin et al ⁹³ on the role of aromatic spacing in determining threshold concentrations for the low complexity domain of hnRNPA1 to manipulate the spacing of aromatic residues in the activation domains of TFs. Many sequence manipulations to various TFs were tested for their change in threshold concentrations in vitro and their activity in cells using reporter assays and cellular reprogramming assays. The rearrangement of aromatic amino acids in a TF activation domain to generate uniform spacing promoted density transitions at lower concentrations, as predicted by previous work. Strikingly, the perfectly spaced aromatic mutant led to enhanced transcriptional activity in reporter assays and made a number of TFs more potent in cellular reprogramming assays. The study also generated several loss-of-function mutations that diminished TF activity. Such loss- and gain-of-function mutations may be good candidates for partial separation-of-function mutations if they preserve the ability for networking while tuning solubility.

In conclusion, mutations that change phase behavior often impact the ability to form soluble complexes with the transcriptional machinery ^19,51,84. Therefore, the use of deletions, mutations, or substitutions may not be sufficient in isolation to support a role for density transitions in transcription (and other biological processes) and should be tested head-to-head with orthologous experiments for their performance in competing models.

Methods to solubilize condensates

To investigate the impact of density transitions on transcription, several strategies have been implemented that disrupt condensate formation by increasing the solubility of the condensate components. Some of the non-specific approaches have limitations and confounding effects and should therefore be used and interpreted with caution (see Box 3), while specific approaches are valuable tools in the repertoire.

The aliphatic alcohol 1,6-hexanediol dissolves some condensates and is frequently used in attempts to evaluate the functions of condensates. However, the dissolution of assemblies—or lack thereof—upon hexanediol treatment does not confirm or refute their identity as condensates and therefore should not be used as evidence for the role of density transitions in transcription. Additional caveats are that hexanediol can have many unintended side effects, such as inhibiting enzymes, altering chromatin 3D architecture, and altering cell state and fate ^94–96. Thus, conclusions about the state of protein assemblies and the role of condensates in transcription must be drawn with caution when relying on hexanediol-based perturbations.

In contrast to hexanediol treatment, targeted solubilization of condensates by direct recruitment of a solubilizing protein^85,97,98 or by the removal of synthetic homo-oligomerization ⁸⁵ offers a more specific means to probe the functional relevance of condensates in transcriptional regulation without perturbing other protein assemblies. Condensates were dissolved by the recruitment of MBP ⁹⁷, HSP70 ⁹⁸, mCherry ⁸⁵, or through the removal of synthetic oligomerization ⁸⁵. Chung et al. used chemogenetic dissolution of YAP-MAML2 oncogenic condensates and found that 95% of genes activated by expression of the fusion oncoprotein persisted in the absence of condensates ⁸⁵. Only a small subset of genes strictly required the presence of condensates. Brumbaugh-Reed et al. used optogenetic dissolution of FUS-CHOP oncogenic condensates and observed a partial reversal of the transcription program elicited by FUS-CHOP ⁹⁷. The results from these studies suggest that density transitions for transcriptional activation can be highly context-dependent and locus-specific.

Expanding the Condensate Toolkit

As described above, a major challenge for studying density transitions is the lack of tools to selectively manipulate this feature of proteins. Several approaches have been proposed in the literature, but each has its own set of potentially confounding effects that need to be controlled for and acknowledged when interpreting results. Tool and method development are needed to faithfully model and selectively perturb condensates. While approaches to drive synthetic homo-oligomerization are powerful, they likely do not produce faithful mimics of natural condensates. Physiologically relevant features that can be tuned to regulate condensate formation remain to be fully identified. Changes in concentration within physiological ranges—above or below threshold concentrations for condensate formation—particularly for cell type–specific factors, could be a useful strategy. Tools to tightly control protein concentration across a population would be valuable; degron-based approaches might enable this type of control ⁹⁹. Beyond manipulating concentration, selective manipulation of condensate composition and dynamics by nanobody-mediated targeting of HSP70 ⁹⁸ or highly associative peptide sequences ¹⁰⁰ have recently been proposed as novel approaches to perturb condensate properties in cells.

In addition to tools to selectively manipulate condensates in cells, more sophisticated in vitro methods are needed that couple in vitro reconstitution of transcription with state-of-the-art quantification of phase separation. Reconstitution systems are powerful but fail to recapitulate key features of cellular transcription, as evidenced by the inability to reproduce transcriptional bursting ¹⁰¹. This suggests that essential components are missing or are improperly regulated. Recent advances towards observing bursting in vitro are promising ²⁹. Future iterations that build on the current reconstitutions may want to attempt to incorporate the effects of physiologically relevant condensates to test whether this enables recapitulation of bursting in vitro.

Following a large number of different players in real time in and around local densities that mediate transcription is currently difficult due to the dynamic nature of the process and technical limitations of fluorescence microscopy. Nevertheless, remarkable progress has been made in labeling DNA, protein, and RNA for live cell imaging studies. The combination and further development of these methods to simultaneously image genomic loci of interest, key transcription regulators, and the RNA product will generate additional spatial and temporal insights and enable robust interpretations. Protein labeling systems such as SunTag ^102,103 and covalent tags like HaloTag ^56,104,105 and SNAP-tag ^69,106, enable sensitive, multiplexed imaging of protein dynamics in live cells. Similarly, RNA stem-loop tags like MS2 and PP7 allow single-molecule tracking of transcript synthesis ^69,107. Live cell DNA labeling has been achieved by two main strategies 1) insertion of an array of orthogonal binding sites ^108,109 and 2) use of fluorescently-labeled catalytically dead CRISPR/Cas9 with various strategies to amplify signal ^110–115. The simultaneous live-cell imaging of DNA, RNA, and protein has the potential to provide an integrated view of the interplay between condensate formation and transcription, but each of these techniques introduce non-trivial perturbations with unknown and often uncontrollable effects on the system under investigation. Together, these techniques complement existing single-molecule and biochemical assays and will ultimately strengthen conclusions.

Moving forward, the development of more refined tools is critical for elucidating the specific roles of condensates in transcription. Integrating these approaches with high-resolution imaging and locus-specific analyses promises to uncover when and where density transitions are functionally required. The complexity of transcriptional processes suggests that only the combination of several approaches will enable us to form a full picture of the role density transitions play in transcription. Caveats and alternative explanations for data from each technique should be acknowledged explicitly.

Additional emergent properties and their potential functional roles.

So far, we have focused on the ability of condensates to stabilize short-lived interactions in their networked interior, but condensates have additional emergent properties which may also influence transcriptional activity. For example, condensates promote the generation of protein conformations that differ from those in the dilute phase. Conformations can be further modified at interfaces, and interfaces can encode electric potentials due to differential ion partitioning. In addition, dense phases can generate environments with a different pH compared to the dilute phase, and with sequence-encoded interfacial tensions and viscoelastic properties. We will briefly describe these emergent properties and their potential impacts on transcriptional activities below (see Figure 3).

Figure 3: Emergent properties of condensates and their potential impact on regulation of transcription.

Large assemblies such as condensates have properties that smaller assemblies do not possess called emergent properties. These include the promotion of longer effective lifetimes of interactions, preferred conformational properties of biomolecules in the bulk of the condensate or at their interfaces, preferential partitioning of biomolecules or ions into condensates or their exclusion, pH gradients across interfaces, viscoelasticity and surface tension-mediated capillary forces. Emergent properties of condensates may promote transcriptional bursts, selective partitioning of positive regulators into transcriptional condensates and exclusion of negative regulators, chromatin remodeling and looping.

Interfacial tensions of condensates and the resulting capillary forces

Given that interfaces are unfavorable locations for biomolecules, interfaces are minimized through condensate fusion and Ostwald ripening. The driving force underlying this behavior is known as interfacial tension, and this, together with the viscoelastic properties of condensates ^116–118, generates capillary forces ¹¹⁹. Through capillary forces, transcriptional condensates can likely distort chromatin and even bring unrelated regions of chromatin together. In vitro studies show that the mechanical properties of condensates formed by pioneering factors can exert forces that distort chromatin structure ¹²⁰. Recent work has demonstrated that condensates exert mechanical forces sufficient to reposition intracellular components, including chromatin ^121–125. Changes to the 3D genome organization have the potential to alter gene activation by bringing cis-regulatory DNA elements (i.e., enhancers) into the proximity of previously inactive genes, resulting in their activation. The condensates formed by NUP98-HOXA9 fusion oncoproteins in cells alter DNA looping and are proposed to activate oncogenic genes via this mechanism ¹²⁶.

Promotion of favorable conformations

The condensate-spanning networks in condensate interiors promote the generation of condensate-specific conformations that differ from those in the dilute phase. This is caused by the multivalent attachments of proteins to the condensate-spanning network and can be the case for individual proteins and protein complexes. Experimentally characterized examples include the polar organizing protein Z (PopZ) in α-proteobacteria, G3BP1 in stress granules in human cells, and enzymes in synthetic condensates. A client recruitment domain of PopZ forms autoinhibitory interactions with the oligomerization domain of PopZ in the dilute phase, counteracting interactions with clients. Condensation requires crosslinking of the oligomerization domain with the condensate-spanning network and relieves the autoinhibitory interactions, generating different conformations in the condensate and thereby enabling client recruitment exclusively to condensates ¹²⁷. The stress granule hub G3BP1 forms intramolecular interactions between positively and negatively charged IDRs, thereby disfavoring interactions with RNA ^128,129. Cellular stress and subsequent translation termination liberate mRNA from polysomes and lead to co-condensation of G3BP1 with mRNA. This likely promotes open conformations of G3BP1 in the dense phase ^128,129. Post-translational modifications on the charged IDRs alter the propensity for intramolecular interactions and the threshold for stress granule formation in cells.

Condensates can also promote active conformations of enzymes, and this was demonstrated for enzymes that promote SUMOylation and were engineered to form synthetic condensates. In these condensates, the activity was higher than expected based on the concentrations of enzyme and substrate, and this and was attributed to favorable conformations adopted by the phase-separating scaffold that positions substrate and enzyme relative to each other ¹³⁰. Of note, condensate network structures do not exclusively generate active conformations. They can also decelerate processes by disfavoring specific conformations or the formation of complexes. This was recently observed for condensates formed by RNA-binding proteins or tau, whose interiors suppress amyloid fibril formation despite the high local concentrations of the proteins in question ^131,132. Whether condensates enhance or suppress biochemical processes will depend on whether their network structures favor or disfavor active conformations.

Active conformation at interfaces

While it is equally favorable for molecules to be in the dilute and in the dense phase at equilibrium, molecules at interfaces experience neither of these favorable environments. Due to their unique position at the interface, these molecules can neither form fully favorable interactions with the solvent nor with other molecules in the dense phase. Hence, their conformations ^133–135 and/or their orientations ^135,136 differ from those in the dilute and in the dense phase. These conformations and orientations can be viewed as unfavorable relative to those in the dilute phase and the interior of the dense phase and can thus act as activated conformations. As an example of interface-mediated activation processes, condensate interfaces can promote nucleation on pathway to fibril formation ^{132,137–139}.

Interphase electric potentials

Not only do macromolecules partition differentially between coexisting dilute and dense phases, but ions also partition differentially to equalize chemical potentials and osmotic pressures between coexisting phases ^140,141. This differential partitioning sets up ion gradients across the interface and generates interphase electric potentials that can directionally approach membrane potentials ¹⁴⁰. Condensates are therefore similar to capacitors in their ability to store charge. Interphase potentials give rise to electric double layers at condensate interfaces, explaining their recently observed electrochemical activities that can catalyze ATP hydrolysis ^142,143. Although additional studies are needed to characterize the phenomenon and its effects in more detail, such interphase electrical potentials may also contribute to gene regulation.

pH differences between condensate interiors and dilute phase

Condensate-forming proteins can generate pH differences between the condensate interiors and the dilute phase ^144,145. For example, the IDRs of nucleolar proteins are characterized by long D/E tracts (stretches of the charged amino acids Asp and Glu) that bind to protons and thereby set up pH differences between the nucleolus and the nucleoplasm ¹⁴⁴. Consequently, condensates and dilute phases can favor different biochemical processes with different pH optima. Condensate formation is a potential mechanism of regulating pH at a specific genomic locus. Whether and how this phenomenon contributes to transcriptional processes remains to be seen.

Potential regulatory opportunities conferred by emergent properties of condensates

Although it remains to be fully elucidated which emergent properties of biomolecular condensates regulate transcription, several compelling possibilities have emerged. Notably, condensates generate distinct microenvironments that may spatially and temporally organize transcriptional machinery, differing from soluble complexes by concentrating and coordinating specific molecular components. This localized enrichment can increase the frequency and effective duration of interactions and may thereby support sustained transcriptional bursting.

Condensates may also influence gene expression by promoting specific conformational states of transcriptional regulators. For example, they could promote rapid re-initiation and the assembly of multiple elongation-competent RNA Polymerase II complexes, potentially facilitating repeated transcriptional bursts—a hypothesis suggested by recent correlations between condensate formation and transcriptional dynamics ^54,62,68,69.

In addition, condensates can selectively enrich or deplete factors, such as coactivators and repressors, thereby stabilizing regulatory states. Depending on the composition of the compartments, they may either promote gene activation or facilitate silencing, particularly through the inclusion or exclusion of chromatin-modifying enzymes ²⁰.

Finally, physical features of condensates—including their interfaces and the generation of capillary forces—may alter chromatin architecture, adding an additional layer of transcriptional control. Together, these mechanisms highlight the diverse and dynamic roles that condensates may play in regulating gene expression.

Conclusions

We have highlighted recent advances in understanding how reversible density transitions may regulate transcription by locally concentrating and stabilizing highly dynamic, short-lived interactions, while also pointing out caveats in the interpretation of extant data. A growing body of evidence from live-cell imaging, concentration-dependent assays, synthetic inducible oligomerization, and sequence manipulation experiments support a model in which condensates formed through multivalent, heterotypic interactions facilitate sustained transcriptional bursts by creating distinct environments that stabilize transient interactions, prolong dwell times, and enable emergent properties absent in small, soluble complexes (Figure 4). Such condensates likely arise not from simple homotypic interactions of individual proteins but from extensive heterotypic interactions among TFs, coactivators, Mediator, RNA Polymerase II, and chromatin regulators. Conceptual frameworks should therefore emphasize the “co-condensation” of the transcriptional machinery as a collective, integrated network rather than the condensation of individual factors. However, dissecting the specific functional roles of condensates versus soluble higher-order complexes remains a key experimental challenge.

Figure 4: A model for the role of condensates in the regulation of transcription.

While basal transcription does not require condensates (left), we propose that multi-component condensates of relevant factors enable bursts of transcription by increasing the rates of both RNA Polymerase II reinitiation and assembly of elongation-competent Pol II complexes (right). For simplicity, many crucial components are not illustrated here including the general transcription factors, chromatin, and chromatin-associated factors. In this proposed model, condensates promote bursting by supplying a larger number of factors and positive regulators, promoting longer effective interactions, favoring elongation-competent RNA Pol II complexes by selective partitioning of positive regulators while excluding negative regulators, and bridging regulatory chromatin elements together through looping.

Future progress will depend on the development of (1) tools to precisely and selectively manipulate condensate formation under physiologically relevant conditions and (2) refined experimental designs capable of distinguishing density transitions from the formation of soluble complexes. Such approaches will illuminate the contexts and conditions under which condensate-mediated regulation is advantageous or essential for transcriptional control, ultimately enhancing our understanding of cell-type specificity, developmental programs, and transcriptional dysregulation in disease.