Integrative approaches to study enhancer–promoter communication

Abstract

The spatiotemporal control of gene expression in complex multicellular organisms relies on noncoding regulatory sequences such as enhancers, which activate transcription of target genes often over large genomic distances. Despite the advances in the identification and characterization of enhancers, the principles and mechanisms by which enhancers select and control their target genes remain largely unknown. Here, we review recent interdisciplinary and quantitative approaches based on emerging techniques that aim to address open questions in the field, notably how regulatory information is encoded in the DNA sequence, how this information is transferred from enhancers to promoters, and how these processes are regulated in time.

Current Opinion in Genetics & Development 2023, 80:102052

This review comes from a themed issue on Genome Architecture and Expression

Edited by Giacomo Cavalli and Job Dekker

For complete overview of the section, please refer to the article collection, “Genome Architecture and Expression (2023)”

https://doi.org/10.1016/j.gde.2023.102052

0959–437X/© 2023 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Introduction

In metazoans, the spatiotemporal control of gene expression relies not only on promoters but also, and crucially so, on enhancers. These noncoding regulatory sequences are located outside promoter regions and activate target genes often over large genomic distances, which in mammals can extend over the megabase scale [1]. Despite the intervening distance, the regulatory role of enhancers is essential for accurate tissue- and developmental time-specific gene expression patterns. Genetic variation within these noncoding regions results in quantitative changes in gene expression and is a major driver of evolution, but also causal to developmental disorders and strongly associated with numerous human diseases [2]. However, despite our increasing ability to identify putative enhancer sequences on the basis of functional genomic annotations and genome-wide association studies, to date, it remains impossible to predict and causally explain distal genetic associations between enhancers and their target promoters from first principles. How enhancers control a promoter’s transcriptional level, and how this depends on their sequence, number, and genomic distance, remains to be understood, as well as how these dependencies relate to binding and interactions of transcription factors (TFs) and cofactors, as well as on modulations of enhancer–promoter (E–P) physical interactions (Figure 1).

Figure 1.

Enhancer-promoter (E-P) communication and transcriptional outputs depend on a number of variables, including enhancer and promoter sequence, number, and mutual genomic distance, in a way that depends on binding of TFs and interactions with cofactors, the physical proximity between regulatory sequences and their dynamics.

Several excellent reviews have covered recent progress in enhancer biology 1, 3, 4. Here, we focus specifically on emerging evidence from highly diverse approaches, which strongly suggest that a truly quantitative understanding of E–P communication requires understanding the spatiotemporal dynamics of the underlying molecular processes, which are highly stochastic and dynamic in individual living cells. We notably review technical and methodological developments bridging across biochemistry, molecular biology, and biophysics that have recently enabled exciting insights, and outline future challenges.

How are regulatory information and specificity encoded in enhancer and promoter sequences?

One key question is how an enhancer’s regulatory activity and its specificity for selected promoters are encoded in its DNA sequence. While it is well established that the activity of an enhancer depends on its ability to recruit combinations of TFs [5], it remains unclear how enhancer activity is modulated by TF binding site (TFBS) number, affinity, and arrangement (i.e. enhancer grammar) [6], and how it depends on covalent DNA or histone modifications as well as TF interactions with cofactors. The last few years have seen major advances in methods for the identification of putative regulatory sequences genome-wide [7] (some of which can be deployed at single-cell resolution [8]), the synthesis and/or mutagenesis of endogenous or ectopic sequences [9], and the massively parallel detection of their effects on transcriptional reporters [10]. Coupled with machine learning algorithms, these methods have given considerable insights into TFBS usage within enhancer regions and at least in some contexts have enabled the design of artificial enhancers with tissue-specific activity 11••, 12••.

While such approaches have unlocked the potential to reveal quantitative sequence determinants of intrinsic enhancer activity, it remains unclear if an enhancer can activate any promoter, or rather only a compatible subset of promoters. Answering this fundamental question requires measuring the transcriptional output of large numbers of E–P combinations in parallel. Initial experiments based on transiently expressed combinatorial libraries of enhancer-reporter pairs have suggested that housekeeping and developmentally regulated promoters are activated by distinct sets of enhancers in D. melanogaster cells [13]. However, in mammalian cells, a rather broad compatibility was observed, with quantitative rather than qualitative differences in E–P communication 14••, 15••, 16••. This is in line with the earlier finding that once randomly inserted into the mouse genome, ectopic promoter sequences are activated by the (unrelated) surrounding endogenous-regulatory landscapes 17, 18. To what extent these results can be extrapolated into general principles of E–P communication in a chromosomal context is unclear. Moving the field forward will critically depend on the development of large-scale reporter assays allowing to study E–P communication in an endogenous chromatin environment, possibly taking advantage of experimental setups where genomic sequence can be engineered bottom-up 19, 20, 21 and chromatin states and/or recruitment of TFs and cofactors can be artificially modulated 22, 23, 24, 25, 26, 27•, 28. Coupled to increasingly more refined ‘explainable’ artificial intelligence approaches [29], such experiments will allow disentangling regulatory layers and unravel quantitative contributions of enhancer/promoter sequence, TFBS usage, and chromatin-mediated effects to E–P communication.

Support for a critical role of chromatin effects in E–P communication is also supported by recent studies that measured TF and nucleosome occupancy with single-molecule resolution using DNA methyltransferase footprinting followed by high-throughput sequencing. These studies revealed that nucleosome eviction following TF binding to enhancer DNA often results in indirect binding cooperativity (e.g. not mediated by direct protein–protein interactions) 30, 31. The single-molecule resolution of such assays was also key to demonstrate that CpG methylation within enhancer regions directly prevents binding of at least some TFs [32]. Coupled to long-read DNA sequencing, single-molecule footprinting methods hold great promise for revealing correlations between TFBS occupancy at distal promoter and enhancer sites [33] and downstream transcription events [34], and even inferring their temporal order [35]. It will be exciting to see how such assays can be combined in the future with atomic models of molecular interactions provided by cryo-electron microscopy (cryo-EM) [36] as well as fluorescent microscopy-based measurements of the rates and stoichiometries of such interactions in single living cells 37, 38 to gain insight into the mechanistic bases of sequence determinants of E–P communication. Ultimately, studying E–P pairs in a genomic context will also be fundamental to reveal how their communication is influenced by the intervening genomic distance and their numbers [39]. This will lead to an understanding of how E–P interactions are modulated by additional regulatory elements and chromosome structure (see below).

How is regulatory information conveyed from enhancers to promoters?

The exchange of regulatory information between an enhancer and a promoter is thought to require at least some degree of physical proximity in the three-dimensional (3D) space of the cell nucleus. E–P communication is indeed generally restricted within topologically associating domains (TAD) 40, 41, which are established as the loop-extrusion activity of the cohesin complex is arrested by CTCF molecules bound to DNA in a specific orientation [42]. Compartment-like interactions as well as cohesin/CTCF-independent tethering elements also contribute to establishing physical associations between active regulatory sequences 43, 44, 45. The notion that proximity is required for conveying regulatory information is also supported by the fact that E–P interaction probabilities measured with cross-linking-based chromosome conformation capture (3C) correlate quantitatively with the degree of promoter activity 46, 47, 48 and that cohesin-mediated CTCF loops can often modulate E–P communication 46, 49, 50, although the effect of TAD boundaries and CTCF sites might depend on the actual intrinsic activity of the enhancer 46, 51 or compatibility with the promoter.

However, what level of ‘proximity’ is required and what molecular processes actually mediate regulatory exchange remains unclear. Spatial proximity is usually defined as the ability to form ligation products in 3C methods, but it is unclear at what spatial distance two genomic loci can be cross-linked and ligated, and whether this distance depends on local chromatin composition (e.g. active or repressive chromatin) [52]. Measurements of proximity-induced DNA methylation have shown that at least a fraction of cross-linking events detected by 3C methods occur in a range of a few nanometers [53]. Absolute quantification of physical distances between regulatory sequences may be enabled by optical methods, but is complicated by experimental uncertainty on distance measurements in fluorescence microscopy (approx. 50 nm in single-dye sequential FISH 54, 55, 56), which currently exceeds the scale of molecular interactions [57]. Indeed, although measurements of physical proximity based on 3C and optical methods often correlate well, discrepancies have been observed [58]. Recent developments in super-resolution microscopy start to bridge this gap achieving a localization precision in the range of 1 nm [59] and promise to yield more precise estimations of molecular proximity in the near future. However, since the mechanisms and factors that mediate E–P communication remain unclear, so do the physical scales that are the most relevant to study. If activation is mediated by direct TF-cofactor–RNA polymerase II (PolII) interactions, these are expected to take place in the few tens-of-nanometer range (as a reference, a single human Mediator/pre-initiation complex spans ∼25×10×10 nm based on recent cryo-EM structures [60]). If instead communication occurs through macromolecular complexes, phase-separated droplets nucleated by low-complexity protein domains 61, 62, or local diffusion of (post-translationally modified) transcriptional coactivators [63], it is possible that distances involved are much larger (in the range of hundreds of nm) (Figure 2). Both mechanisms, irrespective of whether they co-occur or are mutually exclusive, must result in transmission of information to the promoter that ultimately leads to PolII licensing into productive transcription [5].

Figure 2.

Two alternative (and not mutually exclusive) potential mechanisms of information transfer from enhancers to promoters. Information might be passed on to the promoter through direct interactions between TFs and cofactors, or through local enrichment of TFs and cofactors in the vicinity of the promoter. Both mechanisms must result in the transfer of information being used by the promoter to enhance PolII recruitment, post-transcriptional modifications, or events leading to transcriptional initiation or elongation (PTM=post-translational modifications).

Identifying cofactors involved in E–P communication (and in transcriptional regulation in general) has proven difficult because contrary to TFs, coactivators and corepressors do not bind DNA in a sequence-specific manner and no specific ‘interaction code’ with TFs has been identified so far [1]. The last few years have nonetheless seen exciting progress in the identification and characterization of transcriptional cofactors. Genetic screens and new methods for the multiplexed recruitment of cofactors 64, 65•• have revealed large numbers of protein domains with either activating or repressive effects on transcription. The large-scale, unbiased design of these assays has, for example, enabled the discovery of previously unknown amino acid compositional biases in protein fragments that show coactivator activity and a distinctive role of post-translational SUMOylation of transcriptional repressors [65]. Another step ahead in the study of cofactors is the recent development of inducible degradation methods and availability of small-molecule inhibitors, which complement gene knockout approaches by making it possible to study the acute effects of temporary depletion of essential cofactors [66]. Degradation of Mediator subunits was notably key to characterizing their cell-type-specific roles in transcriptional regulation [67] and a role in facilitating E–P interactions 68, 69. Interestingly, depletion and chemical inhibition have shown that chromatin remodeler cofactors, such as SWI/SNF (SWItch/Sucrose Non-Fermentable), not only regulate promoter expression by modulating TFBS accessibility in a TF-specific manner 70, 71, but can also regulate chromosome interactions by modulating accessibility of CTCF binding sites [70].

A better characterization of cofactors will likely enable a more functional, rather than purely physical understanding of E–P contacts. This would allow one to tag cofactors specifically involved in long-range communication with enzymes that promote proximity-mediated chemical modifications (e.g. through proximity-mediated biotinylation followed by mass spectrometry [72] or fluorescent complementation assays [73]). This could also account for the temporal aspect of such interactions. One interesting step in this direction was recently provided by proximity-assisted photoactivation [74], in which energy transfer between two rhodamine dyes allows the fluorescent detection of tagged molecules separated by a distance of few nanometers at single-molecule resolution. Finally, the exciting perspective to visualize physical proximity between genomic locations together with the factors that mediate functional communication might come from spatial sequencing techniques [75], especially combined with methods to fluorescently detect chromatin-binding proteins in situ [76]. Recent advances in correlative light and electron microscopy [77] also open up the possibility that functional interactions might become observable at nearly atomic resolution in a not-too-distant future.

How do stochastic molecular processes at enhancer–promoter interfaces mediate transcriptional regulation?

Irrespective of the exact nature of the biochemical reactions and molecular players mediating the transfer of regulatory information, many studies have shown that the physical distances of E–P pairs vary substantially from one cell to another within the same cell population or tissue 78, 79, 80. This suggests that if E–P communication requires physical proximity, it can generally only occur in a subset of cells at any given time. But it is unclear how often these windows of opportunity are created and how long they last in time within individual cells. This ultimately depends on how fast chromosome structure changes, especially within TADs where most E–P interactions occur in mammalian cells.

Since TADs are established by the loop-extrusion activity of cohesin, the dynamics of chromosome structure inside a TAD is expected to be linked to the kinetics of loop extrusion and of CTCF binding. A series of recent exciting studies based on quantitative microscopy of reconstituted protein complexes demonstrated that mammalian cohesin indeed extrudes loops of DNA and can be blocked by CTCF, and even measured rates of extrusion in vitro (∼1 kb/s) 81, 82, 83. Single-molecule tracking and analysis of protein mobility in vivo 84, 85 additionally suggested that cohesin resides on DNA for 5–20 min on average. Together, these studies suggested that the conformation of the chromatin fiber within a TAD might vary considerably over the timescale of minutes in living cells under the action of cohesin-mediated extrusion.

Analysis of the motion of genomic locations 86, 87, 88, 89 or chromatin-bound proteins [90] initially suggested that chromatin motion is subdiffusive and that promoters or enhancers explore the surrounding nuclear space in a way that might depend on their activity 86, 87. Only very recently, live-cell imaging studies provided insights into how CTCF and cohesin control the looping dynamics of chromosomal sequences located within or close to TAD boundaries. Analysis of the dynamics of pairs of genomic loci in mouse embryonic stem cells revealed that cohesin-mediated loops between pairs of convergent CTCF sites last ∼5–30 min on average 91••, 92••, which is a relatively small fraction of the cell cycle duration (∼12 hours in this cell type). One study estimated that in the absence of CTCF sites, two sequences within a TAD only spend around 5 min on average within ∼150 nm from each other (i.e. the upper range of estimated E–P communication distances, see above) [92]. Together with other recent studies of chromatin motion 93•, 94•, these new results suggest that windows of opportunities for regulatory exchange might be transient events occurring as enhancers and promoters explore the nuclear space in the context of a highly dynamic chromatin fiber.

Interestingly, these experimental estimates for the timescales of CTCF looping and physical proximity represent an intermediate level between those of nucleosome remodeling and TF binding/unbinding, which typically occur on the second timescale, and those of promoter bursting, which in mammalian cells takes place over tens of minutes (Figure 3a) [95]. How these substantially disconnected timescales (at least in mammals; this might not be the case in budding yeast [96]) are integrated to elicit the correct transcription levels and cell-to-cell and temporal variability remains to be understood. Given the fast binding/unbinding kinetics of many TFs and the potentially longer timescales of E–P proximity, it is possible that time-averaged TFBS occupancies within enhancer and promoter regions are sufficient to enable the exchange of regulatory information throughout the duration of a contact. But even if this is the case, does a burst of transcription initiate every time that an enhancer is close enough to its target promoter that their transiently bound TFs can interact (Figure 3b, i)? Or does this happen with a delay, possibly due to the kinetics of intermediate regulatory steps (e.g. assembly or interactions with cofactors) (Figure 3b, ii)? Or rather does a burst of transcription only initiate after several contacts, for example, because of cooperativity in one or more sequential events occurring at the promoter (Figure 3b, iii)? Is every E–P pair actually expected to behave the same, or do alternative modes of signal integration [97] result in different behaviors at different E–P pairs? And how is this related to bursting kinetics at the promoter? Although the 1-contact-1-burst scenario has been observed in live Drosophila embryos, it is likely that it was favored by ectopically induced, unusually stable E–P interactions [98]. Live-cell imaging experiments in mammalian cells suggested a total absence of correlation between E–P and transcription bursts [99], but optical resolution limits discussed above urge caution on the interpretation of these results. Theoretical arguments that are compatible with scenarios ii) and iii) have been recently evoked to explain striking nonlinearities in how transcription levels depend on E–P contact probabilities 46, 100, although none has been formally proven. The rapid advance of super-resolution microscopy techniques that are compatible with live-cell approaches 101••, 102 will likely provide measurements of E–P proximity with increased accuracy together with single-molecule-resolved binding/unbinding of TFs and cofactors [96], thus allowing for a more precise assessment of the temporal order of events at E–P interfaces over timescales ranging from a fraction of a second to minutes and hours.

Figure 3.

Temporal dynamics of E–P communication. (a) Transcription initiation and the resulting transcriptional bursts are preceded by multiple processes, notably fast TF and cofactor binding (typically in the second to tens-of-seconds range) and looping of the chromatin fiber that establishes physical proximity, recently suggested to occur on the order of minutes. (b) While TF-binding dynamics are fast and occur frequently, possibly enabling continuous occupancy of TFBS at the enhancer and the promoter, E–P proximity is presumably longer and occurs less frequently, leading to potential models of E–P communication in time: i) one interaction might lead to one event of transcription initiation, ii) possibly with a delay, or alternatively more complicated scenarios iii) where one initiation event can only be achieved by multiple consecutive interaction events.

We would finally like to note that mathematical and physical models have been instrumental to push forward our understanding of how stochastic and dynamic regulatory processes in single cells convert regulatory inputs into transcriptional outputs. To cite just a few, models have provided mechanistic bases for the processes that shape chromosome organization and dynamics [42] as well as alternative hypotheses on how transcription bursts are generated [97] and connected to E–P interactions 46, 100. Further, they have allowed to quantify how regulatory information is processed by E–P modules into cell states [103]. We expect that the role of models will become even more prominent in the future as the data in the field become more and more quantitative.

Conclusions and outlook

Understanding how genomic sequence determines gene expression programs and their dynamics during development and homeostasis relies on our ability to understand E–P communication from a quantitative point of view, that is, to a level that allows us to predict how sequence mutations translate into changes of transcription levels and their cell-to-cell variability. Over the last few years, we have witnessed spectacular progress in this direction, thanks to the fast-paced rate of technological development and the advent of quantitative methods for the measurements of biological processes. It will be exciting to see how these methods can be coupled to another emerging family of approaches relying on the ‘bottom-up’ generation of regulatory landscapes with controllable levels of complexity 20, 21, 46, 47. By reducing the number of biological degrees of freedom to an amenable amount, such genome engineering approaches have the potential to disentangle regulatory layers and push the field forward toward the identification of mechanistic regulatory principles of E–P communication. We very much look forward to the next phase of experimental and theoretical studies, which holds great promise to open exciting insights into the molecular mechanisms of long-range transcriptional regulation.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data Availability

No data were used for the research described in the article.