The selection and function of cell type-specific enhancers

Preface

The human body contains several hundred cell types, all with the same genome. In metazoans, much of the regulatory code that drives cell type-specific gene expression resides in distal elements called enhancers. Enhancers are activated by proteins called transcription factors that bind specific DNA motifs and recruit co-regulators to ultimately activate transcription. While the human genome contains millions of potential enhancers, only a small subset of them is active in a given cell type. Densely spaced clusters of active enhancers, referred to as super-enhancers, are associated with the expression of genes that specify cell identity and function. On a genomic scale, the function of enhancers is influenced by, and in turn affects higher-order chromatin structure and nuclear organization.

Introduction

The molecular mechanisms that enable and mediate cell-specific transcriptional responses to intra- and extra-cellular cues remain poorly understood. Early experiments indicated that sequences far away from gene promoters are often required to regulate cell type-specific gene transcription¹. Such genetic elements are termed enhancers, and were initially functionally defined as DNA sequences that have the potential to enhance basal transcription levels from gene promoters and transcriptional start sites (TSS)¹, at distances ranging from hundreds of base pairs to megabases². Recent genome-wide transcription factor-binding studies indicated that the majority of transcription factor binding sites are found in distal locations that frequently exhibit enhancer function^3–9. This is consistent with the profound role that enhancers play in shaping signal-dependent transcriptional responses^10–12.

When cell signaling induces an increase in the nuclear concentration and DNA binding of transcription factors, as occurs following the activation of steroid hormone receptors and NF-κB, the great majority of binding events typically occurs at genomic locations that already exhibit binding of other transcription factors and enhancer-like histone modifications^{5, 6}. Because the complement of active cis-regulatory elements is different across cell types, these findings introduced the notion that pre-existing sets of enhancers are largely responsible for cell type-specific gene expression and responses to external stimuli^13–15. The annotation of epigenetic features associated with enhancers in many different cell lines, primary cells and tissues by the ENCODE consortium provided evidence for the utilization of several hundreds of thousands of such elements in the human genome¹⁶, greatly exceeding the number of genes that encode mRNAs or long intergenic non-coding RNAs (lincRNAs). This raised the question of how the correct subsets of enhancers are selected from the large repertoire of potential enhancers in each particular cell type. Here, we review recent findings on the selection and function of enhancers that specify cell identity and underlie their distinctive responses to intra- and extracellular signals. We discuss the collaborative and hierarchical binding of transcription factors to DNA in the context of chromatin, which orchestrates enhancer selection and priming, and the transformation of chromatin from a silent, primed or poised state to one that actively supports transcription. We conclude with a discussion of the three-dimensional organization of enhancers in the nucleus and its importance for their function.

Enhancer characteristics

Genomic regions that function as transcriptional enhancers are enriched for closely spaced recognition motifs for sequence-specific transcription factors. Enhancer activation begins with the binding of transcription factors and local nucleosome remodeling. Recent genome-wide studies of nucleosome remodeling during differentiation of embryonic stem cells and induced pluripotent stem cells indicated that the majority of remodeling affects a single nucleosome, and that alterations in nucleosome occupancy are enriched at enhancers associated with pluripotency and differentiation¹⁷. Transcription factor binding leads to, and in some cases is facilitated by, the recruitment of co-regulators such as the histone acetyltransferase p300¹⁸, followed by the recruitment of RNA polymerase II (Pol II) and the transcription of enhancer-associated RNAs (eRNAs)^{19, 20}. Co-regulator recruitment and transcription are accompanied by the covalent modification (methylation and acetylation, among others) of histone tails in enhancer-associated nucleosomes. In organisms whose DNA exhibits methylation in the context of CG dinucleotides (CpG methylation), these enhancers become demethylated upon their activation, concomitant with transcription factor binding²¹. Thus, epigenetic modification patterns can be used to distinguish between different enhancer activation states²² and have been used extensively to annotate putative enhancers in different cell types¹⁶.

Enhancer states can broadly be classified as inactive, primed or poised, and active²². An inactive enhancer is essentially buried in compact chromatin and is devoid of transcription factor binding and histone modifications. Primed enhancers are characterized by closely bound, sequence-specific transcription factors that establish a DNase I-hypersensitive¹⁵, nucleosome-free²³ region of open chromatin. However, they may require additional cues to accomplish their function, which may include signal-dependent activation, the recruitment of additional transcription factors, and the eventual recruitment of co-activators that lead to enhancer activation. Poised enhancers can be defined as primed enhancers that also contain repressive epigenetic chromatin marks (see below), a state that is most commonly found in embryonic stem cells. The characteristic features of poised and active enhancers are depicted in Figure 1.

Figure 1.

The anatomies of poised and active enhancers. The characteristic features of poised and active enhancers are shown, including the binding of lineage-determining transcription factors (LDTFs) and collaborating transcription factors (CTFs) to closely spaced recognition motifs (blue and green sites, respectively) on the DNA. The binding of these factors in concert with nucleosome remodeling complexes (NRCs) initiates nucleosome displacement to form narrow nucleosome free regions at poised enhancers (top). At poised enhancers, the redundant histone methyltransferases (HMTs) myeloid/lymphoid or mixed-lineage leukemia protein 3 (MLL3) and MLL4 deposit the active H3K4me1 and H3K4me2 marks, whereas EZH2 (a component of the polycomb complex) deposits repressive H3K27me3 and histone deacetylase (HDAC)-containing complexes maintain histones in a repressed deacetylated state. Pol II is either absent or low at poised enhancers. In response to various cues, signal-dependent transcription factors (SDTFs) associate with recognition motifs in close association with LDTFs, which results in additional nucleosome displacement (bottom), as observed by widening of the DNase I-hypersensitive sites. SDTFs recruit co-activator complexes containing histone demethylase (HDM) complexes that remove H3K27me3 marks, histone acetyltransferase (HAT) that deposit H2K27ac, and the mediator complex. The transformation to elongating Pol II results in bidirectional transcription — a hallmark of active enhancers — and the generation of enhancer RNAs (eRNA), which is closely coupled to enhancer activity.

An insight important for the identification of potential enhancers was the understanding that specific histone methylation signatures mark enhancer-like regions. In particular, enhancers display enrichment for histone H3 Lys 4 mono- or dimethylation (H3K4me1 and H3K4me2, respectively) and depletion of H3K4me3, compared to promoters³. Whereas genomic regions exhibiting these features are not necessarily functional enhancers, it appears that the vast majority of regions that do function as enhancers exhibit these characteristics^{3, 7, 24}. Specifically, primed enhancer-like regions are marked with H3K4me1 and H3K4me2 and lack histone acetylation, and enhancers marked by the trimethylation of histone H3 at lysine 27 (H3K27me3, a repressive mark) are considered to be poised^24–26 (for a review, see²⁷)(Fig. 1). Features associated with active enhancers include histone H3 Lys 27 acetylation (H3K27ac)²⁵ and the presence of actively transcribing Pol II¹⁹. Examples of these features in the vicinity of the TAL1 locus in the genomes of 7 human cell lines, evaluated by the ENCODE consortium, are illustrated in Figure 2. Several developmental enhancers have been characterized for this locus: the −3.8 kb (upstream) and +19 kb (downstream) enhancers drive TAL1 transcription in endothelial cells (HUVEC) and hematopoietic stem and progenitor cells^{28, 29}, whereas the +51 kb enhancer is required for TAL1 expression in erythroid cells (K562)³⁰. On the whole, DNase I hypersensitivity at this locus corresponds with overall transcription factor binding, and the presence of the “active” epigenetic marks H3K4me2 and H3K27ac is correlated with cell type-specific enhancer activity. Conversely, in TAL1 non-expressing cells such hESC and NHEK, the +19 kb enhancer, promoter and gene body are devoid of DNase I-hypersensitive sites, and the −3.8 kb region and the gene body exhibit the repressive mark H3K27me3.

Figure 2.

Cell type-specific enhancers are marked by specific epigenomic features and chromatin accessibility. Genomic features of a ~60 kb region of human chromosome 1 centered around the TAL1 gene ENCODE consortium data of DNase-I hypersensitive (DNase HS) regions and ChIP-Seq for the marks H3K4me2, H3K27me3 and H3K27ac in 7 cell lines. Enhancers known to be responsible for TAL1 transcription in endothelial cells (the −3.8 kb and +19 kb enhancers, relative to the TAL1 promoter, in HUVEC cells) and erythroid cells (the +51 kb enhancer in K562 cells) exhibit cell type-specific DNase HS, H3K4me2 and H3K27ac signals. In cell types where TAL1 is not expressed, the promoter and gene body are devoid of DNase HS and histone modifications indicative of enhancer activation (H3K4me2, H3K27ac), and exhibit variable levels of the repressive H3K27me3 mark. Shaded boxes indicate cell-restricted or cell-specific enhancers regions.

Enhancer selection

The vast number of potential cis-regulatory elements in the genome and the cell-type selectivity with which they are utilized raises the question as to the series of events whereby unique enhancer repertoires are selected. Many lines of evidence indicate that enhancer selection is initially driven by so-called pioneer factors, exemplified by FOXA1, that are able to bind to their recognition motifs within the context of compacted chromatin³¹. By opening the conformation of the chromatin and initiating the process of enhancer selection, such pioneering factors can function as key cell lineage-determining transcription factors (LDTFs) to drive lineage-specific transcription programs. However, most sequence-specific transcription factors, including those that function as pioneer factors, recognize relatively short DNA sequences (of about 6 to 12 base pairs), and their typical DNA recognition motifs exhibit varying levels of degeneracy. This means that most sequence-specific transcription factors have millions of potential binding sites in the mammalian genome. Yet, chromatin immunoprecipitation followed by sequencing (ChIP-Seq) experiments have indicated that they bind only a small subset of all potential sites, and that a large fraction of the observed binding is associated with cell type-specific enhancers³². Cell type-specific binding sites often harbor motifs for additional pioneer factors, and experimental data strongly suggest that pioneer factors act in concert to jointly displace nucleosomes^{33, 34}. Here, we review evidence supporting a model in which pioneer factors, or LDTFs, prime cell type-specific enhancers through collaborative interactions^{7, 23, 35–40,35–40}.

The role of lineage-determining transcription factors

Experiments modulating the expression of LDTFs have demonstrated their ability to initiate the transition of enhancer elements from closed chromatin to a ‘primed’ or ’poised’ state, where transcription factors have gained access to the DNA and established nucleosome-free regions^{7, 10} (Figure 1). An example is provided by the ETS domain factor PU.1, a LDTF required for the development of macrophages and B cells. Importantly, PU.1 influences the establishment of distinct gene expression programs in each cell type⁴¹. The vast majority of PU.1 binding sites are located >500 bp from promoters and largely occupy different genomic locations in macrophages and B cells⁷. Macrophage-specific binding of PU.1 was observed at genomic locations that contained PU.1 binding sites in close proximity to binding sites of other macrophage LDTFs, such as the C/EBPs and AP-1 factors. Conversely, B cell-specific binding of PU.1 was observed in close proximity to other B cell LDTFs, including motifs recognized by EBF, E2A and Oct factors. The corresponding motifs were generally situated less than 100 bp from the PU.1 motif, but mostly not at a close (5–20 bp), invariable distance that would be indicative of direct ternary protein-protein-DNA interactions⁴². Notably, macrophage-specific PU.1-bound regions were depleted of B-cell LDTF motifs and vice versa, relative to neighbouring genomic regions. This, together with the finding that in a given cell type, non-bound PU.1 motifs that lie in transcriptionally inactive genomic regions are generally depleted of motifs of the LDTFs expressed in the cell³², suggests that (LDTF) motif composition may be one of the contributing factors to forming transcriptionally inactive and active genomic compartments (see below). Gain- and loss-of-function experiments revealed an interdependence of PU.1 with other LDTFs for effective DNA binding, suggesting that their collaborative interactions are necessary to compete with nucleosome for binding to DNA. By considering natural genetic variation between inbred strains of mice as mutations, LDTF binding site mutations not only impaired binding of the respective LDTF but also that of closely bound LDTFs^{43, 44}, consistent with a model in which enhancer selection is a collaborative effort of multiple DNA binding factors. Other examples of LDTF co-operativity in establishing specific LDTF binding patterns have been observed in developmental systems such as zebrafish hematopoiesis⁴⁵ and Drosophila melanogaster embryogenesis⁴⁶.

The use of computational methods to identify binding motifs that are enriched in genomic regions marked by H3K4me1 resulted in the identification of LDTF motifs of the corresponding cell type. For example, binding sites for transcription factors that are capable of reprogramming fibroblasts into induced pluripotent stem cells are highly enriched in the H3K4me1-marked regions of the genome in embryonic stem (ES) cells⁷. Conversely, ChIP-Seq studies of LDTFs commonly revealed that they occupy large fractions of the enhancers within the cells in which they exert lineage-determining functions^{7, 10, 16, 36, 47, 48}. Thus, whereas most cells express hundreds of transcription factors, the selection of a large fraction of cell type-specific regulatory elements may be driven by relatively simple combinations of LDTFs that interact with each other and with other factors. Collectively, these findings may facilitate computational efforts to predict the selection of cell type-specific enhancer elements based on the local organization of binding motifs and the combinations of expressed transcription factors.

The role of signal-dependent transcription factors

Whereas LDTF binding may be sufficient for activating some enhancers, additional signals will be required for other enhancers to be fully activated. Many of the cellular responses to internal and external signals are dependent on the function of widely expressed, signal-dependent transcription factors (SDTFs). Examples of SDTFs are members of the nuclear receptor and NF-κB families. These factors frequently activate common sets of genes in different cell types, but can also regulate gene expression in a cell type-specific manner. ChIP-Seq studies of SDTFs in different cell types found both common and cell type-specific binding sites^{37, 38, 48}. Two types of mechanisms were suggested to account for cell type-specific binding of SDTFs. In one, such SDTF binding occurs at genomic locations that exhibit features of pre-selected enhancers^{7, 36, 45} (Fig. 3A). In these cases, there is a hierarchical relationship between SDTFs and LDTFs, which are the pioneer factors responsible for the initial enhancer selection through interactions with additional, collaborating transcription factors (CTFs). In many cases loss of function of the LDTF results in a failure of both the LDTF and the SDTF to bind the enhancer, but not vice versa^{7, 48–51}. Alternatively, SDTFs could contribute directly to latent or de novo enhancer selection^{37, 50, 52} (Fig. 3B). This has been shown to involve collaborative interactions with LDTFs, which, owing to their restricted cell type-specific expression patterns, imposes cell type-specific enhancer selection at genomic locations that have the appropriate combination of motifs. Although mechanisms underlying collaborative DNA binding by transcription factors remain poorly understood, studies of the glucocorticoid receptor suggest that transcription factor binding can be highly dynamic and that even two factors that interact with the same recognition motif in the same cell can facilitate each other’s binding through a proposed assisted loading mechanism⁵³.

Figure 3.

Cell type-specific enhancer selection and activation. A. Collaborative interactions between lineage-determining transcription factors (LDTFs) and collaborating transcription factors (CTFs) select enhancers for binding and activation by signal-dependent transcription factors (SDTFs). Prior to signal-dependent activation, such regions may be ‘poised’ enhancers or exhibit basal enhancer activity (‘pre-existing’ enhancers) that is further induced by the binding of a SDTF. The resulting transcription is cell type-specific because the enhancers are selected by the cell type-specific LDTFs. B. SDTFs can direct the selection of latent or de novo enhancers. In these cases, the SDTF functions as an essential collaborative transcription factor to LDTFs to enable concurrent binding of all factors involved. The transcriptional output is cell type-specific because of the requirement for cell type-specific LDTFs for enhancer priming.

The extent to which SDTFs operate on poised enhancers or participate in de novo enhancer selection appears to vary depending on the factor, cell type and signal in question. FoxP3, a SDTF required for the acquisition of the Th2 phenotype of CD4-positive T cells, was found to bind almost exclusively to a poised enhancers upon their activation³⁶. In contrast, the receptor for the steroid hormone ecdysone, a member of the nuclear receptor family mediating transcriptional responses to ecdysone in insects, primarily binds newly selected enhancer elements in combination with cell type-specific transcription factors³⁷. Both mechanisms of enhancer selection (Fig. 3) can occur simultaneously in the same cell type. For example, following macrophage activation by lipopolysaccharides, approximately 90% of the binding of the p65 subunit of NF-κB occurs at already primed enhancers, whereas the remainder is associated with the de novo selection of latent enhancers in collaboration with LDTFs such as PU.1 and C/EBPα^{43, 50}. Recent studies of macrophage populations from different tissues demonstrated that local environmental signals regulate the expression and activities of TFs that can function to selectively activate enhancers that are primed in common between macrophage subsets or drive the priming and activation of enhancers that are specific to macrophage subsets or^{54, 55}. Of note, the histone methylation signature of latent enhancers persists after the cessation of cell stimulation and is associated with more rapid and diverse transcriptional responses to subsequent stimulation⁵². These observations provide evidence that the writing of the H3K4me1 signature in enhancers provides a molecular memory of prior activation.

Enhancer activation

While transcription factor binding is a requirement enhancer activity, not all promoter-distal transcription factor binding sites appear to function as enhancers, as judged by H3K4me1 and H3K4me2, and not all regions of the genome enriched with H3K4me1 and H3K4me2 exhibit marks of active enhancers, such as H3K27ac. This raises the question of what determines whether transcription factor binding will result in an active enhancer. Many different enhancer states can be defined based on particular combinations of histone posttranslational modifications²² (Figure 1), which are deposited by transcription co-regulators that are recruited to enhancers and promoters by transcription factors. Transcription co-regulators include histone methyltransferases (HMT), including the MLL proteins⁵⁶, histone acetyltransferases (HAT) such as p300 and CBP⁵⁷, histone deacetylases (HDAC) which are components of co-repressors such as NCoR and SMART⁵, as well as chromatin remodelers such as the BRG1- or hBRM-associated factor (BAF) complexes (also known as the mammalian SWI/SNF complexes)^{58, 59} and the mediator complex⁶⁰. Recruitment of co-regulators to enhancers is the more frequent the more transcription factors are co-bound to a given enhancer^{7, 61}. Co-regulators are large proteins with multiple distinct interaction sites for transcription factors^{18, 62, 63}, and likely serve as both facilitators and integrators of transcription factor binding and intracellular signals at enhancers, similar to their known roles at promoters⁶⁴.

Enhancer transcription

The epigenetic marks deposited by co-regulator complexes serve as binding sites for chromatin readers such as TFIID⁶⁵ and Brd4—P-TEFb⁶⁶, which function in transcription pre-initiation complex assembly and in transcription elongation, respectively.

The presence of transcription pre-initiation complex and elongation factors at enhancers^{67, 68} is in line with the finding that Pol II is found at enhancers. More than 20 years ago Pol II was observed to generate non-coding RNAs at locus control regions⁶⁹, but it was only recently appreciated that mammalian enhancers are broadly transcribed and generate enhancer RNAs (eRNAs)^{19, 20, 70–72}. Pol II recruitment to enhancers and signaling-dependent changes in eRNA expression are highly correlated with changes in the expression of nearby genes, suggesting a functional link between eRNA and gene expression^{73, 50, 74, 75}. The distinguishing features of eRNAs are that most are short (< 1 kb), are not subjected to polyadenylation or splicing^{19, 20}, and are rapidly degraded by the exosome⁷¹. Similarly to what has been shown for short promoter antisense transcripts⁷⁶, these characteristics are likely caused by the lack of a 5’ splice donor proximal to eRNA TSS^{71, 72}, which is a prerequisite for splicing and promotes elongation⁷⁷, packaging into messenger ribonucleoprotein particles (mRNP), polyadenylation and nuclear export⁷⁸, all of which contribute to the stability of transcripts. As a side note, the fact that enhancers resemble promoters in almost every aspect, except for lacking proximal splice donors⁷¹ and H3K4me3 marks⁷⁹, suggests that stable mRNAs or lincRNAs could be created by simply introducing a splice donor downstream of an eRNA TSS⁷². This would be in line with the ability of intronic enhancers to serve as alternative promoters⁸⁰, and the fact that 98% of all lincRNAs have only two exons (that is, a single splice donor downstream of a promoter), compared with merely 6% of coding genes⁸¹.

The occurrence of global enhancer transcription in mammalian cells raises the question of its functional significance. Recent studies provide evidence that eRNAs contribute to local enhancer activity, potentially by facilitating enhancer--promoter interactions through chromatin looping, recruitment of co-factors such as the mediator complex (Fig. 4A; reviewed in ref. 66) and release of negative elongation factor⁸². As of yet there is limited evidence for specific sequence features of eRNAs that could be necessary for their function, and not all eRNAs appear to contribute to enhancer function. To date, little attention has been directed at the possibility that the process of enhancer transcription itself (independent of the eRNA product) could influence enhancer activity. Pol II is a powerful nucleosome remodeling machine⁸³, and transcription initiated from an enhancer sequence may contribute to maintaining an open chromatin configuration that enables access of sequence-specific transcription factors. In addition, enhancer transcription may play an important role in contributing to the deposition of H3K4me1 and H3K4me2 marks at enhancers (Fig. 4B). Genetic studies indicate that the D. melanogaster H3K4 methyltransferase trithorax-related (Trr) and its mammalian homologues MLL3 and MLL4 play important roles in the writing of these marks^{84, 85}, but the mechanisms that recruit these enzymes and determine the overall distribution of histone methylation remain poorly understood. Studies of newly selected or de novo enhancers in activated macrophages provided evidence that the methylation of H3K4, but not the acetylation of H3K27, required enhancer transcription and the presence of MLL3 and MLL4⁵⁰.

Figure 4.

Enhancer activation and function. A. Interactions between enhancers and promoters involve structural connections (orange oval) that include cohesin and the mediator complex to promote pre-initiation complex formation, initiate transcription and/or overcome Pol II pausing. A potential role of enhancer RNAs (eRNAs) could be to promote transcription by facilitating chromatin looping, possibly by mediating interactions with cohesin. Another potential role could be to mediate interactions with protein complexes required for transcriptional elongation, such as the mediator complex. LDTFs, lineage-determining transcription factors; CTFs, collaborating transcription factors; SDTFs, signal-dependent transcription factors. B. Potential roles of enhancer transcription. In activated macrophages the NF-κB proteins p50 and p65 are signal-dependent transcription factors and PU.1 is a lineage-determining transcription factor that collaboratively select de novo enhancers. The subsequent recruitment of histone acetyltransferases (HAT) results in histone acetylation, which is bound by the Brd4 component of the P-TEFb complex, allowing its Cdk9 component to phosphorylate the C-terminal domain (CTD) of Pol II. Phosphorylated CTD acts as docking sites for the MLL3 and MLL4 histone H3K4 methyltransferases. MLL3 and MLL4 are proposed to deposit H3K4me1 and H3K4me2 during successive rounds of Pol II elongation.

A model of enhancer activation based on time-resolved studies of transcription factor binding, eRNA transcription, H3K4 methylation and H3K27 acetylation at de novo enhancers, and on results of gain- and loss-of-function experiments⁵⁰, is illustrated in Fig. 4B: Signal-dependent activation of NF-κB (p50 and p65) results in its collaborative binding with PU.1 and the recruitment of co-activator complexes that contain histone acetyltransferases (HAT). These events result in nucleosome remodeling, histone acetylation and the recruitment of Pol II. The conversion of Pol II from a paused to an elongating form involves P-TEFb, which is recruited to at least some sites of transcription initiation by interactions between Brd4 and acetylated histone H4. Cyclin-dependent kinase 9 (CDK9), a component of P-TEFb, phosphorylates the C-terminal domain (CTD) of Pol II, providing docking sites for the histone methyltransferases complexes myeloid/lymphoid or mixed-lineage leukemia protein 3 (MLL3) and MLL4. MLL 3 and MLL4 progressively methylate H3K4 during successive rounds of transcription elongation. Consistent with this model is the distribution of H3K4me1 and H3K4me2, which was found to correlate with the extent of enhancer transcription, and to be dependent on transcription elongation⁵⁰. The generality of this model with respect to the mechanisms by which H3K4 methylation marks are established at other classes of enhancers, such as those that are selected during cellular differentiation, remains to be determined. For example, in contrast to the activation of de novo enhancers in the context of extracellular signaling responses, studies of the distribution of H3K4me1 and H3K4me2 at cell type-specific enhancers selected during muscle and adipocyte differentiation suggested that MLL complexes can interact directly with LDTFs such as C/EBPβ and MyoD at cell type-specific enhancers, where MLL3 and MLL4 are also required for acetylation of H3K27, mediator and Pol II recruitment⁸⁵.

The function of H3K4me1 and H3K4me2 marks remains an open question. As they are known to recruit histone remodeling complexes⁸⁶, they could conceivably contribute to keeping previously bound and modified enhancers open and accessible, which would help explain the observation that previously activated latent enhancers are more rapidly re-activated by subsequent stimuli⁵².

Enhancer function

Promoter activation requires that many components of the transcriptional machinery come together in order to assemble the pre-initiation complex, initiate transcription, overcome Pol II pausing, and eventually lead to productive transcription elongation. Through looping of the intervening DNA, enhancers get into close proximity of promoters, and are thought to affect any or all of the aforementioned processes by increasing the local concentration of the factors that carry them out⁸⁷ (Fig. 4A). These factors include co-activator complexes such as the mediator complex, which increases the loading of transcription factors on promoters and enhancers⁶⁰; scaffold proteins such as cohesin, that mediate stable, often cell type-specific promoter-enhancer interactions^{60, 88}; and factors involved in releasing paused Pol II and in the initiation of elongation, such as Brd4⁸⁹. A major challenge in deciphering cell-specific enhancer functions is connecting active enhancers to their target genes in vivo.

Super-enhancers

Based on their epigenetic features, and depending on the experimental methods used to define enhancers, ~10,000–50,000 putative enhancers can be identified in a given cell type^{13, 16, 90}, implying there are more enhancers than expressed genes. Along the linear DNA molecule, enhancers are located non-uniformly in respect to genes, such that some genes reside in enhancer-rich regions of the genome, whereas others have few or no enhancers in their vicinity. Although a single enhancer is sufficient to activate the expression of a nearby gene³⁷, high levels of cell type-specific and/or signal-dependent gene expression are most frequently observed for genes located in enhancer-rich regions of the genome, exemplified by the relationship of enhancer-rich locus control regions and the expression of the globin genes in erythroid cells⁶⁹. Such enhancer-dense regions have recently been termed “super-enhancers”^91–93.

Super-enhancers were initially defined as large (tens of kilobases-long) genomic loci with an unusually high density of enhancer-associated marks, such as binding of the mediator complex, relative to most other genomic loci^{91, 92}. These regions can also be defined by high density⁹¹ and/or extended (> 3 kb)⁹⁴ depositions of the histone mark H3K27ac. Using differences in the density of mediator complex-binding sites or of H3K27ac marks to distinguish super-enhancers from regular enhancers, most cell types are found to have between 300 and 500 super-enhancers⁹¹. A substantial fraction of super-enhancers and nearby genes are cell type-specific, and the gene sets that are associated with super-enhancers in a given cell type are highly enriched for the biological processes that define the identities of the cell types^91,94. For example, many of the genes encoding factors required for pluripotency and self-renewal of ES cells reside near ES cell-specific super-enhancers⁹¹. In keeping with their tissue-specificity, super-enhancers active in certain cell types are enriched for disease-associated alleles relevant to that cell type^91,94. Not surprisingly, the individual enhancers of cell type-specific super-enhancers are enriched for binding sites of the corresponding LDTFs⁹². Collectively, the specific set of super-enhancers within a particular cell type may provide a means of simplifying the problem of defining what are the quantitatively most important transcriptional programmes required for establishing cell identity, and to identifying disease-relevant, non-coding genetic variation.

Three-dimensional chromatin interactions

In the nucleus, the genome is organized and partitioned into functional compartments in the three-dimensional (3D) space⁹⁵, and considerable effort is being directed at understanding enhancer function in the context of 3D chromatin interactions. One strategy is to identify the long-range looping interactions involving enhancer elements using a variety of chromosome conformation capture (3C)-based techniques⁹⁶. Genome-wide applications of these techniques to define the chromatin interactomes of human and mouse cells confirmed that the genome is divided into active and inactive compartments⁹⁶. These are further organized into sub-megabase-sized topologically associated domains (TADs) that correlate with regions of the genome that constrain the spread of heterochromatin and are relatively conserved across cell types⁹⁷. Although the genome-wide resolution of such studies remains somewhat limited, the resulting chromatin connectivity maps suggest that only approximately 7% of the looping interactions are made between adjacent genes, indicating that linear genomic adjacency is not necessarily a good predictor for long-range interactions⁹⁸. In addition, promoters and distal enhancer elements are frequently engaged in multiple long-range interactions and form active chromatin hubs^{98, 99} (Fig. 5). Whereas super-enhancers are identified along the linear DNA sequence by virtue of their high density of typical epigenomic features, it is clear that the enhancers within a super-enhancer form 3D interactions that are a feature of the folded genome in the nucleus⁹³ (Fig 5). Interestingly, studies of tumor necrosis factor α (TNFα)-responsive enhancers in human fibroblasts indicated that they are already in contact with their target promoters before the activation of TNFα signaling¹⁰⁰, suggesting that enhancer-promoter interactomes are already set up during development. This is in line with data from D. melanogaster showing that only 6% of spatial genome interactions change during early embryonal development¹⁰¹. It is not yet known when higher-order chromatin interactions are established during development, but it likely coincides with the occurrence of gap phases following the mid-blastula transition, which is accompanied by establishment of a non-random nuclear chromatin conformation and the transcriptional activation of chromatin domains¹⁰².

Figure 5.

The linear and the three-dimensional organization of enhancers in the nucleus. The outer circle represents the linear coordinates of a region of human chromosome 1 surrounding the Atf3 (activating transcription factor 3) gene in C57BL/6J mouse macrophages. The locations of individual genes are indicated by gene names and purple bars. The three successive concentric inner circles depict ChIP-Seq data of, respectively, histone H3 Lys 27 acetylation (H3K27ac), the transcription factor PU.1, and the transcription repressor CCCTC-binding factor (CTCF), which is enriched at boundaries of topological domains. A region of high density of H3K27ac in the vicinity of the Atf3 gene is designated as a super-enhancer. Purple and black lines in the center of the circle indicate physical contacts involving promoters and other genomic regions, respectively, as determined by statistically significant genome-wide chromatin connectivity measurements determined by tethered conformation capture¹³⁰. This locus demonstrates the multitude of connections between the individual enhancers comprising the Atf3 super-enhancer, which essentially forms its own TAD, as well as the longer-range enhancer-enhancer and enhancer-promoter interactions outside of the TAD.

It is unclear how the 3D organization of the genome is determined, however what is known is that cohesin and the mediator complex⁶⁰, which are scaffold proteins of the replication machinery and the transcription machinery, respectively, are involved in the formation of high-order chromatin structures. Since cohesion appears to be recruited to enhancers through clusters of LDTFs¹⁰³, it is likely that protein-protein interactions and the genomic sequence together shape 3D genomic conformations, although this hypothesis still awaits experimental confirmation.

Given that the conformation of the genome appears to be mostly fixed across developmental stages¹⁰¹, individual cells¹⁰⁴, cell types⁹⁷ and signaling states¹⁰⁰, it is tempting to speculate how enhancers work in the 3D space: Promoters are known to also function as transcriptional enhancers with regard to the activation of promoters in their proximity¹⁰⁵, and enhancers have sequence features identical to those found in promoters^{71, 72}. Both are juxtaposed within TADs as part of linear super-enhancers⁹², as well being brought into proximity by higher-order structures⁷⁴, leading to the co-regulation of promoters and enhancers within a domain^74,106. Knocking out enhancers within a TAD shows that the loss of an enhancer often only leads to a graded reduction in expression^{107, 108,74} and in developmental dysregulation¹⁰⁹ of the associated gene, suggesting that at least in some cases enhancers work in an additive manner. The distribution of gene regulation among a multitude of enhancers, some of which reside linearly within or beyond neighboring genes (in their “shadow”, hence the term “shadow enhancers”¹¹⁰) but are close in 3D space, is thought to increase robustness of the regulatory system to mutations¹¹¹. Whereas the high-order chromatin structure of genomic regions >1 Mb is largely invariant, single loci can move between inert and active chromosomal compartments, depending on their activation status, leading respectively to stable repression or to a state poised for transcription¹¹². In contrast, within these large-scale compartments, inside TADs, the chromatin structure of regions smaller than 100 kb does differ in a cell type-specific fashion¹⁰⁰, implying that different regulatory regions within TADs can be dynamically juxtaposed in a stimulus-specific manner. This way, genome topology could contribute to cell type-specific transcription programs, meaning that mapping the genomic topology and elucidating the mechanisms that govern the 3D structure of the genome will be important steps toward understanding how the genome functions.

Conclusions and perspective

Although initially described more than 30 years ago, we still do not have a clear understanding of the mechanisms by which enhancers regulate gene expression. However, the development of a plethora of methods for genome-wide mapping of diverse enhancer features, their functional relationship with promoters, and their ultimate transcriptional outputs have resulted in a number of striking and unexpected discoveries, ranging from the identification of the great number of enhancers in metazoan genomes to the widespread production of enhancer-derived RNAs. The observation that more than 80% of disease-associated alleles identified by genome-wide association studies reside in non-coding regions of the genome¹¹³ implies they have yet unappreciated regulatory functions. Consistent with this, several studies have demonstrated an enrichment of disease-associated loci in cell type-specific regulatory regions, including in super-enhancers, of the corresponding disease-relevant cells types^{91, 114–117}, and a number of studies are beginning to document the direct effects of common variation in enhancer elements on enhancer states^118–120, gene expression^{117, 121, 122} and disease^123–127.

Beyond the simple annotation of regulatory regions in the genome, it is important to understand how cells select the full complement of enhancers that is required for maintaining their identities and functions. In essence, we would like to be able to ‘read’ the genomic template and predict from the combination of active transcription factors the enhancers that will be functional in a cell type-specific manner. The principle of collaborative transcription factor-interactions at closely-spaced DNA recognition motifs provides a starting point for predicting genome-wide patterns of transcription factor binding required for enhancer selection. These predictions can be validated by mutating binding sites or by taking advantage of naturally-occurring genetic variations. However, transcription factor-binding maps are insufficient to predict enhancer activity. The discovery that enhancer transcription is highly correlated with nearby gene expression is likely to be an important clue in understanding how enhancers function. The evidence that eRNAs contribute to activities of at least some enhancers provide impetus for determining their mechanisms of action. In addition, the relative importance of enhancer transcription itself in maintaining enhancer accessibility and contributing to enhancer-related H3K4 methylation requires further study, as the functional roles of H3K4 methylation beyond providing a memory of prior enhancer activation, remain obscure⁵².

Defining functional enhancer-promoter interactions remains an important goal. Despite being informative, chromatin connectivity maps do not directly relate chromatin interactions to the regulation of gene expression. Definitive evidence that a specific enhancer-like region exerts a transcriptional regulatory function requires mutating that region, and encouragingly, site-specific mutagenesis should be greatly facilitated by recently developed genome editing methods¹²⁸. Such tools will enable us to systematically delete enhancer elements and modify enhancer sequences to evaluate chromatin connectivity and gene expression. As a complementary approach, recent studies have demonstrated the utility of using natural genetic variation as a tool to study the relationships between transcription factor binding, enhancer selection and the regulation of gene expression⁴³. Improving the understanding of the mechanisms underlying the selection and function of enhancers will likely not only enable predicting the consequences of genetic variation on gene expression and phenotype, but may also provide approaches to directly alter enhancer function for therapeutic purposes.

Glossary

DNase I-hypersensitive site

Genomic DNA is packaged into chromatin, which makes it less accessible to DNase I, and nucleases in general. Binding of proteins to regulatory regions makes these genomic sites more accessible, and hypersensitive to DNase digestion.

Locus control region (LCR)

Confers tissue-specific expression to a linked transgene irrespective of the integration site of the transgene in the genome. Thus, LCRs display characteristics of both enhancers and insulators.

Exosome

A multiprotein complex that is involved in surveillance, degradation and maturation of RNA transcripts in both nucleus and cytoplasm. In the nucleus, the exosome is involved in 3’ processing of rRNA and snoRNA, degrades many types of aberrant RNA transcripts, including pre-mRNAs and pre-tRNAs, as well as non-coding transcripts such as eRNAs and promoter upstream transcripts (PROMPTs), and. In the cytoplasm, mRNAs are the primary target of the exosome, where it contributes to mRNA turnover, nonsense-mediated decay of mRNAs with premature stop codons, mRNAs without a stop codon, and 5’ ends of mRNAs that have stalled in translation or have been cleaved by the RNAi pathway¹²⁹.

Mediator complex

A multisubunit protein complex that influences nearly all stages of transcription and in conjunction with cohesion contributes to the 3D organization of the genome. The mediator complex is recruited to the DNA by binding to sequence-specific transcription factors via its individual subunits, and integrates a wide gamut of intracellular signals to affect pre-initiation complex formation, transcription initiation and elongation.

Chromosome conformation capture (3C)

A method to probe the higher order structure of genomic DNA in the nucleus. In living cells, DNA gets fixed to the chromatin in living cells by formaldehyde, digested with a restriction enzyme, and protein-DNA complexes get diluted and ligated. DNA that were spatially close in the nucleus are more likely to reside within the same protein-DNA complex. These are preferentially ligated together, and interaction frequencies of studied loci are quantified by quantitative PCR. This experimental strategy forms the basis for unbiased genome-wide methods such as Hi-C and tethered conformation capture (TCC), where the restriction sites are labeled with biotin prior to ligation, and either arm of the biotinylated, mixed ligation product is sequenced to reveal the genome-wide interaction frequencies of all genomic loci in a cell.

Chromatin hub

A spatial arrangement of regulatory DNA elements (LCRs, enhancers, promoters) and genes into a domain that leads to correct gene expression of the associated genes. The smallest unit of a hub could be a topologically associated domain (TAD), and the largest could comprise an entire nuclear compartment (see TAD).

Active enhancers

In addition to marks of poised enhancers, active enhancers are marked with acetylated H3 lysine 27 (H3K27ac), produce eRNAs, are bound by the Mediator complex, and exert regulatory function to increase the transcription of target genes.

Latent or de novo enhancers

An inactive enhancer locus whose selection requires the binding of a combination of transcription factors that includes SDTFs.

Primed enhancers

Enhancers that have been selected by LDTFs and CTFs and are marked with histone modifications characteristic of enhancers, such as mono- and dimethylation of histone H3 lysine 4 (H3K4me1 and H3K4me2), but do not produce eRNA.

Poised enhancers

Similar to primed enhancers, but distinguished by the presence of tri-methylation of H3 lysine 27, which must be removed for the transition to an active enhancer state.

Topologically Associated Domains (TADs)

Genomic, largely self-interacting domains of sub-megabase sizes, further organized into multi-megabase-sized structures called nuclear compartments. Genes within TADs are co-regulated and their expression patterns are highly correlated. TADs are determined by chromosome conformation capture experiments and are largely conserved across cell types and throughout development.

Biographies

Sven Heinz is the Director of the Next Generation Sequencing Core at the Salk Institute for Biological Studies in La Jolla, California, USA. After receiving his Ph.D. in Chemistry from the University of Regensburg, Germany, he did postdoctoral work with Christopher Glass at the University of California, San Diego, California, USA. His research focuses on the transcription regulatory mechanisms that underlie genome function in different cell types and individuals during development, homeostasis and disease.

Casey E. Romanoski completed her Ph.D. in 2010 in Human Genetics under supervision of Aldons J. Lusis at the University of California, Los Angeles. There, she studied the genetics of endothelial cell expression and responsiveness to oxidized phospholipids. Since, she has worked as a postdoctoral fellow with Christopher K. Glass where she is interested in the molecular mechanisms of gene regulation and the consequences of genetic variation on expression networks and disease traits.

Christopher Benner is the Director of the Integrative Genomics and Bioinformatics Core at the Salk Institute. Dr. Benner's research focuses on how genomic sequences are organized to encode complex regulatory mechanisms. He develops innovative software tools for the analysis of large-scale sequencing-based genomics experiments, including ChIP-Seq, GRO-Seq and conformation capture assays. Dr. Benner is the developer of the HOMER software suite that is widely used by the genomics community.

Christopher K. Glass is Professor of Cellular and Molecular Medicine and Professor of Medicine at the University of California San Diego School of Medicine. He has had a long-standing interest in understanding how sequence-specific transcription factors, co-activators and co-repressors regulate macrophage gene expression. His current work employs genome-wide approaches to investigate mechanisms underlying the selection of enhancers that determine macrophage identity and function.