Nanoscale nuclear environments, fine-scale 3D genome organization and transcription regulation

Abstract

Decades of in vitro biochemical reconstitution, genetics and structural biology studies have established a vast knowledge base on the molecular mechanisms of chromatin regulation and transcription. A remaining challenge is to understand how these intricate biochemical systems operate in the context of the 3D genome organization and in the crowded and compartmentalized nuclear milieu. Here we review recent progress in this area based on high-resolution imaging approaches.

Nuclear heterogeneity: Focal accumulation of Pol II and Regulatory factors

Early microscopy studies had revealed that transcription takes places in discrete nuclear foci [1]. The distribution of RNA Polymerase II and regulatory factors (RFs) also showed significant spatial inhomogeneity [2–4]. As many of these early studies relied on chemical fixation and antibody labeling, it was not easy to address the extent of potential disruption and the degree of labeling of the native structures that form in live cells. We note that classical nuclear “bodies” associated with transcription of gene arrays and clusters also exhibit focal accumulation of certain protein factors, e.g. the histone locus in Drosophila, or the miR-430 locus in zebrafish embryos. These much larger, several μm-sized structures are outside of the scope of this manuscript, which focuses on finer-scale organization at single-copy genes.

More recently, fluorescent tags compatible with live-cell imaging and genome editing tools to create endogenous tagged protein factors, in conjunction with more sensitive and higher resolution imaging modalities, stirred new interest in the organization of transcription. We now know that many important transcription factors (TFs) and co-factors, as well as Pol II, are distributed in mesoscopic (~100’s of nanometers) clusters throughout the nucleus. These include developmental and cell-identity TFs [5–7], downstream effectors of signaling pathways [8–10], co-activators [11, 12], chromatin adapters [11], elongation factors [13, 14], as well as pathological molecules such as oncogenic transcription factor fusions [15, 16].

Although the description of clusters of Pol II and RFs throughout the nucleus has been an important and significant advance, addressing the functional roles of such assemblies and their relationship to transcription has remained a challenge. New insights have now begun to emerge from techniques that can directly visualize Pol II and RFs at specific gene loci in live cells. A particularly challenging task has been reliably detecting the small number of molecules that might be engaged with loci of interest at a given time, and discriminating these target-engaged molecules from background molecules diffusing in-and-out of the detection volume, or binding nearby. Technological developments in single-molecule nanoscopy, in particular effective background suppression schemes [17*], recently enabled single-molecule detection sensitivity, at addressable 3D locations inside the crowded intracellular milieu.

The new techniques revealed that both Pol II and RFs (Sox2, Mediator, Brd4, Cdk9) form nanoscopic (~100 nm) clusters in the vicinity of pluripotency genes like Pou5f1, Nanog and Sox2, in live mouse embryonic stem cells (mESCs). These clusters, comprising just 5–20 molecules of each protein factor, could only be imaged, tracked and quantified by highly sensitive techniques. Pol II molecules cluster within ~100nm of the nascent RNA, consistent with the majority of Pol II being engaged in elongation. Enhancer-associated RFs accumulate slightly further, at ~200 nm from the nascent RNA. Interestingly, distinct from the fine clusters that form in the vicinity of active pluripotency genes, a different study described much larger Pol II and Mediator clusters (~300 nm in size and comprising hundreds of molecules) observed at undefined nuclear locations in mESCs [18], several μm away from pluripotency genes. There is thus a variety of Pol II and RF clusters that can form throughout the nucleus, whose functional significance remains to be elucidated.

How does RF clustering mechanistically relate to transcription activity? Evidence for at least three different mechanisms has been reported (Figure 1): acceleration of rate-limiting steps through mass action, modulation of target search and specificity, and gene co-regulation.

Figure 1.

Possible functional roles of focal RF accumulation at transcription sites. (A) RF clustering creates regions of high local concentration and by mass-action accelerates rate-limiting steps in the transcription cycle [17*]. (B) RF clustering modulates the target search processes of transcription factors and chromatin regulators, possibly by transient trapping in local zones [23**]. (C) Nano-scale environments of clustered RFs activate multiple embedded promoters [35**], facilitating gene co-regulation.

Transcription activation by high local RF concentration

In the simplest scenario, the high local RF concentration accelerates rate-limiting steps in the transcription cycle, through mass action. Simultaneous single-gene imaging of RF clustering and transcription activity supports this model [17*]. Fine-tuning Cdk9 cluster size at the Pou5f1 locus results in tunable amplitude of nascent transcription, a result consistent with the control of promoter-proximal pause release by Cdk9-mediated phosphorylation of the Pol II CTD. Similarly, fine-tuning the Brd4 cluster size at Nanog results in tunable burst initiation rate, as suggested by quantitative imaging of Nanog bursting kinetics and associated Brd4 clusters upon treatment with the Bromodomain and Extra-Terminal (BET) domain protein family inhibitor JQ1.

Both studies of Cdk9 and Brd4 clusters at Pou5f1 and Nanog relied on competitive inhibition of Brd4 binding to acetylated chromatin using small molecules. The development of a variety of chemical probes and targeted protein degradation tools [19] provides a further toolkit for attempting to fine-tune clustering of a variety of other RFs. One important aspect that has not yet been characterized is the effective local RF concentration seen by the Pol II machinery at the promoter. Zooming into the promoter region with higher spatial resolution, down to the ~10’s of nm-sized macromolecular complexes that form, could enable measuring directly how individual RF molecules interact with components of the Pol II pre-initiation complex and/or with promoter-proximally paused Pol II. Such measurements will be key to elucidate physical mechanisms of e.g. promoter-enhancer communication inside these nanoscale nuclear environments [20] and understand in detail the regulatory impact to transcription outputs.

Modulation of TF nuclear exploration and target search

Beyond regulating rate-limiting steps in the Pol II cycle through mass-action, RF clustering could also provide a means for more efficient target search and increased binding specificity, e.g. through mechanisms akin to “antenna effects” described for facilitated diffusion processes [21, 22]. This model presently remains untested experimentally at the single-gene level. However observations of TF movements throughout the nucleus using single-particle tracking suggest that the local environment might alter the character of 3D diffusion. The transcription factor CTCF exhibits anisotropic diffusion, consistent with a model postulating interactions within ~200 nm trapping zones [23**]. The authors further showed that trapping involves an RNA-interacting domain of CTCF and the trapping zones correspond to clusters of CTCF. Interestingly, disruption of the CTCF RNA-interacting domain results in changes in nuclear clustering, 3D genome organization, and gene expression, suggesting links between CTCF clustering, diffusion, and target search mechanism [24].

The glucocotricoid receptor (GR) also exhibits heterogeneous nuclear behavior. Machine-learning classification of single GR trajectories showed two populations [25**]: an apparent sub-diffusive population that might correspond to cognate DNA site binding (“chromatin-bound”), and a second less mobile population that might correspond to local “confinement”. Interestingly, this second population was dependent of an intrinsically disordered region (IDR) of the GR protein, while another nuclear receptor, PPARα, which contains a shorter IDR only exhibited the “chromatin-bound” but not the “confinement” population. Notably, although GR forms nuclear clusters upon hormone induction, this study did not address the mobility of GR specifically inside nuclear clusters.

It will be very interesting to see if CTCF and GR clusters form at specific genomic loci and if this type of nuclear exploration modulates TF target search in general. Importantly, if confinement accelerates target search, one would expect that locally trapped TFs would eventually bind to a cognate site. However, the single-particle tracking studies so far did not report transitions between the different states. Emerging techniques that can provide longer single-particle trajectories and increased spatio-temporal resolution [26–28] might enable direct tests and help provide a more detailed biophysical grounding of these ideas.

Gene co-regulation

Compartmentalization of the nucleus into local environments of clustered RFs could also provide a mechanism for the little-understood phenomenon of gene co-regulation [29–31]. To explain the simultaneous activation of two promoters by the same enhancers [32], as well as rationalize the ~200 nm apparent distance separating such co-regulated genes [33], it was hypothesized that a common pool of clustered Pol II or RF molecules might be shared by the two promoters [34]. Although such models remain speculative, an imaging study of Nanog and Sox2 showed that transcriptional bursting of two identical promoters on sister chromatids is coordinated [35**], and during such coordinated bursts a joint pool of clustered Brd4 molecules preferentially span the space between the two transcription sites. Whether this picture can be extended to different co-regulated genes, both linked in cis [32, 36*] and in trans [37–39], will require further studies of multiplex RNA tagging and simultaneous observation of RF clusters.

Chromatin topology and gene regulation: communication between enhancers and promoters

Eukaryotic genes are often controlled by enhancers, cis-elements that can activate target promoters over large genomic distances. Uncoupling a promoter and its regulatory elements enables more complex regulatory scenarios for regulating a gene under different conditions, and also puts the 3D configuration and folding of the chromatin fiber in center stage. Distal enhancers and target promoters must somehow communicate. Most models for promoter-enhancer communication posit transmission of signals either through direct physical interactions or propagations of signals. Exactly how communication takes place is not well understood.

Enhancer-promoter physical distances

Cross-linking and proximity ligation assays often show preferential association between enhancers and promoters, but the physical distances cannot be accurately defined by these indirect measurements. Imaging studies that directly measure physical distances have shown increased enhancer-promoter proximity accompanying transcription activation. In Drosophila embryos, physical distances between a distal synthetic reporter and endogenous even skipped (eve) enhancers reduce from ~700 nm to ~200–300 nm, when this distal reporter was activated. [40]. Sustained enhancer-promoter proximity was necessary for productive transcription. In MCF7 breast cancer cells, induction of estrogen responsive genes, e.g. TFF1, was accompanied with increased proximity (defined as <600 nm separation) of multiple distal enhancers [6].

The increased proximity paradigm does not apply generally, even for the same gene under different conditions. Shh, encoding a morphogen important for tissue growth and patterning during development, is controlled by tissue-specific enhancers located within a ~1 Mb domain. In Shh-expressing tissue in the developing limb-bud, a ~1Mb distal enhancer (ZRS) is brought into spatial proximity to Shh, suggesting a preformed chromatin topology [41, 42]. Contrary, in neural progenitor cells and in the Shh-expressing cells in the embryonic neural tube, enhancers >100 kb upstream of Shh remain spatially separated from Shh [43*]. Similar puzzling results, showing weak or no correlation between proximity and transcription, have been reported for a variety of systems, including the bithorax Hox gene complex in Drosophila [44] and the Sox2 gene in mESCs [45*].

Many imaging studies to date have caveats that need to be taken into consideration. Studies using Fluorescence In-situ Hybridization (FISH) have limited resolution, often using ~10 kb to >100 kb FISH probes, while they rely on chemical fixation and denaturation. Live cells studies have often used exogenous probes such as nascent RNA tags as proxies for the genomic DNA, or large arrays with hundreds of tandem sites for DNA binding proteins. The native structures could be perturbed on the nanometer scale in FISH assays; the exogenous tags used in live-cell imaging might also not accurately represent the positions of the endogenous genomic elements. Due to the various localization errors, the measured distances between fluorescent tags can thus be quite a bit longer than the true distances between e.g. a gene promoter and its distal enhancers [46]. How often and how close promoters and enhancers come together remains an open question.

Refinements of the original chromosome conformation capture (3C) method have enabled measurements of genome architecture at increasing resolution: ~10 kb with Hi-C [47, 48], ~100–1,000 bp with Micro-C [49, 50] and ~1–100bp with Micro-capture-C [51]. Interestingly, genomic contacts appear progressively more well-defined well looked at higher resolution. More sensitive 4D microscopy methods [27] as well as insights based on physical models of the chromatin polymer [52], together with approaches that better preserve structures in fixed cells [53–55], or enable nanometer distance measurements in live cells [35**, 56] pave the way for optically imaging interactions between promoters and enhancers at higher resolution. By optically imaging and tracking a single DNA bound molecule, the resolution would be pushed down to that molecule’s DNA footprint - a few 10’s of bp.

Enhancer-promoter communication and enhancer-associated RF clusters

The ~100’s of nanometers apparent distances between enhancers and promoters can be interpreted as an indication that approximate enhancer-promoter proximity is sufficient for activation, while molecular-scale contacts are not strictly needed. Some models further postulate that although promoters and enhancer remain separated, clusters of RF molecules can effectively span the intervening distance, bringing some RF molecules to the vicinity of the transcription machinery at the promoter. Zooming into the nano-environment of active transcription sites and visualizing with <1kb resolution the juxtaposition of individual promoters and enhancer elements, as well as the relative movements of single molecules of RFs and the Pol II machinery will enable testing these ideas in more detail.

Enhancer cluster “super-clusters” and systems of multiple enhancers

Developmental and cell type-specific genes are often embedded inside complex regulatory landscapes, characterized by intricate non-binary configurations with multiple enhancers and promoters [57]. Characterizing multi-way genomic interactions by population-averaged chromosome conformation capture data has been challenging, recent studies using single-allele proximity ligation methods [58*, 59, 60*, 61] suggest that multiple promoters and enhancers simultaneously interact, in multi-way hubs. Highly connected enhancer hubs might be a characteristic of cell-type-specific transcription activity [36*, 62], and genes important for cell identity are proposed to be controlled by extended genomic regions with enhancer activity [63]. Cooperative interaction hubs formed in trans [64, 65], bringing together different chromosomes might also be important for gene regulation.

Although genomics assays elaborated on the highly-connected nature of such multi-way genomic interaction hubs, their physical manifestation has been hard to pinpoint. A number of imaging studies have now shed new light on enhancer hubs, and established connections between enhancer hubs and focal accumulation of Pol II RFs at nuclear foci.

In Drosophila embryos, the transcription factor Utrabithorax (Ubx) and its co-factor Homothorax (Hth) co-concentrate in regions of high local concentration [66]. Active transcription sites of the svb gene and svb enhancers colocalized in these areas of high local Ubx concentration, to within a few hundred nanometers [66]. The multiple enhancers controlling svb expression have overlapping expression patterns, a redundancy that ensures robust developmental phenotypes under stress conditions [67]. Deletions of individual svb enhancers results in lowered local Ubx concentration and reduced svb transcription activity. Interestingly, an ectopically re-inserted full svb regulatory region in a different chromosome was observed in close proximity to svb transcription sites and also rescued the deletion efficiencies. These results suggest that related enhancers from different chromosomes can share the same local microenvironment.

In mESCs, multiple distal enhancers from extended genomic loci reside in ~100–200 nm average proximity to the Pou5f1 and Sox2 transcription sites [35**], within the average distances of clusters of Sox2, Mediator and Brd4 proteins [17*, 35**]. The high frequency of pair-wise interactions further suggested that the extended locus frequently adopts configurations that juxtapose multiple distal enhancers and transcription site [35**]. These enhancer cluster “super-clusters” provided a framework for connecting genome-topology, enhancer-promoter communication and RF clustering at active transcription sites (Figure 2).

Figure 2.

Enhancer cluster “super-clusters” [35**]. Multiple distal enhancer clusters, from an extended genomic locus, are in the proximity (~100–200 nm) of a target gene. The multiple DNA and chromatin binding sites brought into close proximity for a scaffold for formation of RF clusters at active gene loci.

Underlying chromatin: scaffold, tether or seed for RF clusters?

Although the abundance of RF into clusters and their dynamics have been now well characterized, a biophysical basis for their formation is still not well established. A particularly important question relates to the role of the underlying chromatin in potentially scaffolding the formation of RF clusters, vs. the contributions of RF-RF interactions (Figure 3). In one end of the spectrum, RF clustering reflects occupation of multiple DNA and chromatin binding sites in close proximity. On the other end of the spectrum, RF clusters might be built-up through weak and multi-valent protein-protein interactions, with the underlying chromatin merely seeding cluster formation or tethering preformed clusters at specific sites. The prevalence of IDRs on Pol II RFs is hypothesized to drive formation of such macromolecular assemblies, in a manner akin to in vitro droplet formation through liquid-liquid phase separation.

Figure 3.

Models for RF cluster formation at single gene loci (top row) and their predictions for recruitment of factors (bottom row). (A) Specific molecular recognition model. Clustering of RFs reflects multiple RF molecules recognizing multiple specific DNA and/or chromatin binding sites. Binding sites might be clustered in 1D and/or in 3D. Bottom panel in (A): molecules with IDR deletions can still incorporate into RF clusters but mutations or deletions of specific binding domains diminish cluster incorporation ability. (B and C) Models that depend on IDR-IDR interactions. (B) Sub-saturation clustering model. The system is globally below the saturating concentration, so droplet formation is thermodynamically unfavorable. However, binding of multiple RF molecules to multiple specific sites at the enhancer increases the local concentration above saturation, leading to clustering of additional RF molecules due to IDR-IDR interactions [15, 80]. (C) Droplet formation model [7, 11]. The system is above saturation and droplet formation is thermodynamically favorable. A large number of RF molecules accumulate at the enhancer, giving rise to a structure that exhibits properties of macroscopic droplets – interfacial tension and clear separation between bulk phases (inside and outside the droplet). Bottom panels in (B and C): RF molecules with mutations or deletions of specific binding domains are readily incorporated into RF clusters.

Testing whether phase separation occurs in cells remains technically very challenging [68, 69]. However it is worth noting that for some Pol II RFs (Sox2 and Brd4 in mESCs) previously shown to form IDR-driven droplets in vitro, and large nuclear clusters when artificially multimerized in cells, IDR-IDR interactions provide minimal contribution to recruitment into endogenous clusters and to active genes [35**]. These results suggest that interpreting such clusters as biomolecular condensates or droplets likely does not accurately describe the underlying biophysical and molecular clustering mechanisms.

Sox2 and Brd4 recruitment into clusters depends on specific molecular recognition of DNA and chromatin targets. Thus clustering of Sox2 and Brd4 throughout mESC nuclei and at specific transcription sites likely reflects multiple molecules that simultaneously occupy closely-spaced specific binding sites on DNA/chromatin. In another example, a point mutant Ubx protein incapable for specific DNA binding did not display nuclear clustering [66], suggesting that specific DNA binding might be more generally important for focal accumulation of TFs.

The oncogenic fusion protein EWS/FLI1 was shown to form clusters in A673 cancer cells, with an estimated median cluster size of ~24 molecules. Given that the median number of GGAA microsatellite repeats across the 1D genome is ~15 and a single bound EWS/FLI1 occupies two repeats, the authors concluded that IDR-IDR interactions mediate formation of super-stochoimetric EWS/FLI1 clusters [15]. It will be interesting to measure the sizes of EWS/FLI1 clusters forming at specific genomic loci with different GGAA repeats, as well as characterize in more detail the 3D arrangement of GGAA repeats in the nucleus and whether all EWS/FLI1 clusters form at GGAA repeats.

Putative roles of IDRs in RF clustering in cells still remain to be definitely proven, because conclusions in several studies were drawn from artificial systems involving exogenous protein overexpression. A recent study [70] used knock-in tags and gene editing of the endogenous Utx in mouse embryonic stem cells. Interestingly, deletion of a “core” IDR (cIDR) domain resulted in less granular pattern of GFP-tagged UTX protein compared to WT, indicating that the cIDR partially contributes to the focal enrichment of UTX in the nucleus.

Non-linearity, cooperativity and regulatory wiring of multi-enhancer systems

How individual enhancers collaborate to activate transcription is not well understood. Previous studies provided conflicting results on the relative importance of the individual elements within enhancer clusters [71]. In some experiments the individual enhancers contribute additively and independently from each other [72]. In agreement with these results, a linear “activity-by-contact” model, where the individual enhancer contributions depend on “activity” of the enhancer (judged by level of chromatin modifications or bound factors) and the contact frequency with the target promoter, seems to have the highest predictive power for enhancer-gene connections across cell types [73*]. Conversely, other experiments reveal a certain hierarchy of enhancers within a cluster, with some enhancers proposed to activate other elements within the cluster [74, 75]. Given this conundrum [76], the functions and interdependencies of constituent elements within enhancer clusters remain a critical knowledge gap in our understanding of such complex gene regulation schemes.

Many of the genes controlled by enhancer clusters, like the pluripotency genes in mESCs, are important for the specific cell state and are also part of auto-regulatory networks. Thus enhancer perturbations that disrupt gene expression might lead to changes in cell state and/or turn on feedback mechanisms that can confound interpretations. This can be problematic with genetic perturbations (e.g. enhancer deletions) that take place over multiple cell generations. Rapid perturbations by recruiting regulatory effector domains to specific genomic loci provide a powerful alternative for capturing primary effects that occur on the minutes to hours time-scales [77–79]. We expect that experiments that will combine imaging of nascent transcription activity, genomic interactions and RF clustering, together with targeted enhancer perturbations, will be very informative to tease out the regulatory logic of enhancer cluster “super-clusters”.

Conclusions and outlook

Recent advances in imaging technologies have led to many new insights about the connections between nuclear compartmentalization, 3D genome topology and transcription regulation. Pol II RFs form nano-scale clusters at active gene loci, on an underlying scaffold of enhancer cluster “super-clusters”. Many of the basic properties of these striking macromolecular assemblies have now been described. New technological developments in terms of spatial and temporal resolution will enable zooming into these nanoscale environments and following the movements and interactions of single RF molecules and individual constituent genomic elements. Together with chemical and epi-genetic tools that enable rapid targeted perturbations, these experiments will help establish a more detailed biophysical and molecular picture of these complex multi-component systems and their regulation.