Enhancer function: mechanistic and genome-wide insights come together

Summary

Enhancers establish spatial or temporal patterns of gene expression that are critical for development, yet our understanding of how these DNA cis-regulatory elements function from a distance to increase transcription of their target genes and shape the cellular transcriptome has been gleaned primarily from studies of individual genes or gene families. High-throughput sequencing studies place enhancer-gene interactions within the 3D context of chromosome folding inviting a new look at enhancer function and stimulating provocative new questions. Here we integrate these whole genome studies with recent mechanistic studies to illuminate how enhancers physically interact with target genes, how enhancer activity is regulated during development, and the role of non-coding RNAs transcribed from enhancers in their function.

Introduction

Enhancers are cis-acting DNA regulatory elements that increase the transcriptional output of target genes to influence the destiny of cells during development and differentiation. Enhancers may reside far from their in vivo targets, raising key mechanistic questions about how they communicate with promoters. Models such as enhancer looping to distant targets, linking by large protein complexes or tracking along intervening chromatin were animatedly discussed but failed to be definitely resolved (Bulger and Groudine, 2011). The development of chromosome conformation capture (3C) technology (Dekker et al., 2002) allowed physical interaction frequencies between specific enhancers and target genes to be determined (Tolhuis et al., 2002). This significant advance affirmed that enhancers establish proximity with the genes they activate although the original models are not mutually exclusive and may yet be found to contribute to enhancer function.

Enhancer DNA sequences are replete with clusters of binding sites for transcription factors whose occupancy confers upon them tissue specificity. Thus, it had long been proposed that factors binding to enhancers and genes could stabilize chromatin loops between them through homotypic or heterotypic interaction. Indeed, tissue specific proteins that are critical for looping have been identified in a limited number of mammalian model systems such as the β-globin locus and the human IFNγ and T_H2 cytokine loci, among others (Krivega and Dean, 2012). In addition, looping interactions of chromosome organizing proteins such as the insulator binding protein CTCF and its frequent partner cohesin have direct and indirect roles in facilitating enhancer-gene chromatin contacts (Ong and Corces, 2014). Furthermore, lineage specific activators and CTCF/cohesin engage in interactions with components of the RNA polymerase II (Pol II) machinery and with the Mediator transcriptional co-activator complex to tie enhancer loops directly to the transcription apparatus (Maston et al., 2012).

Whole genome studies now place enhancer interactions in a 3D nuclear context. These advances have depended on refinements in 3C-related approaches including 5C, Hi-C and ChIA-PET, capable of detecting chromatin loops at multiple levels (de Laat and Dekker, 2012). Studies across developmental stages detail the dynamic regulation of enhancers and enhancer looping during development (de Laat and Duboule, 2013). Moreover, recent results suggest a function for enhancer transcription into eRNAs as part of the mechanism of gene activation and possibly looping (Orom and Shiekhattar, 2013). The new data, coming in a deluge of publications over a very recent period paint a picture of multiple levels of long range genome interactions of which enhancer-gene contacts are a part. Descriptive by nature, the studies nevertheless allow mechanistic insights and raise new questions such as how chromosome folding occurs, what drives the changes in enhancers during development and how enhancer looping and transcription activation are integrated physically and spatially in the nucleus to achieve a unique gene expression.

Excellent recent reviews have covered genome-wide discovery of new enhancers and gene targets, activator and co-activator recruitment to enhancers and specific features of enhancers such as an open chromatin structure, the H3K4me1 histone mark and histone acetyltransferase p300 occupancy (Spitz and Furlong, 2012; Maston et al., 2012; Calo and Wysocka, 2013). Here, we specifically focus on mechanistic insights into enhancer function from individual gene and very recent genome-wide studies. We begin by considering the basis for enhancer-gene loops in individual loci and across chromosomes. We then discuss how enhancer activity is regulated during development and its potential functional significance. We end with an account of our current understanding of enhancers RNAs and their roles in enhancer activity and gene regulation.

Enhancers engage in long range interactions with target genes through lineage specific factors and ubiquitous architectural proteins

Enhancers are typically occupied by clusters of transcription factors that exclude nucleosomes and contribute to their DNAse I hypersensitive character. At select loci, 3C and RNAi studies have shown that specific enhancer binding proteins are required for enhancer-gene looping (Figure 1A). For example, a complex including GATA1 and co-factor FOG1 along with TAL1, LMO2 and LDB1 is required for β-globin locus control region (LCR) looping to globin genes and for transcription activation in mature erythroid cells (Vakoc et al., 2005; Song et al., 2007; Yun et al., 2014). The dimerization domain of LDB1 underlies the enhancer-gene proximity (Krivega et al., 2014). In fact, the dimerization domain alone, when linked to the β-globin promoter via a specially designed DNA binding zinc finger protein is capable of driving loop formation and partially activating transcription in immature erythroid cells where this does not normally occur, arguing for the causality of the loop in the activation (Deng et al., 2012). A large cohort of erythroid genes is regulated by the LDB1 complex, suggesting it can function broadly, likely through chromatin looping, to affect lineage commitment (Li et al., 2013a).

Figure 1. Enhancers and promoters communicate by chromatin looping.

(A) Left: Two lineage specific genes and an enhancer are depicted along unfolded chromatin with neither gene being transcribed. Right: Lineage specific transcription factors mediate long range interaction between the enhancer and one of the genes through homotypic and/or heterotypic protein interaction. The gene in contact with the enhancer is activated; the other gene (inactive) is looped away from the elements that are in proximity. (B) Left: A CTCF binding site and an enhancer are depicted with an inactive gene along unfolded chromatin. Right: The gene is activated by lineage specific activators that coopt CTCF into long range interaction with the gene. (C) A non-interacting enhancer and gene. Right: The enhancer is bridged to the gene promoter by Mediator and cohesin with participation of lineage specific factors, activating the gene. (D) Left: A locus containing a gene and enhancer reside in an unfolded and inactive state. Center, Right: Enhancer-gene looping is depicted as being mediated by lineage specific activators before accumulation of Pol II and the appearance of a transcription factory, and transcription.

Other focused investigations in the IFNγ and MYB loci indicate that architectural proteins CTCF/cohesin can participate directly in enhancer-gene looping (Figure 1B) (Sekimata et al., 2009; Stadhouders et al., 2012; Hadjur et al., 2009). For example, enhancers that bind the lineage specific factor T-bet in T_H1 cells are interspersed with CTCF/cohesin binding elements in the IFNγ locus (Sekimata et al., 2009). As naïve T cells differentiate, CTCF promotes a T_H1 cell specific IFNγ locus looped conformation joining both kinds of sites and activating IFNγ transcription. CTCF locus occupancy and looping are dependent on enhancer binding by T-bet but whether these proteins interact is unknown. These examples illustrate how lineage specific transcription activators mediate enhancer gene looping but also how they can cooperate with a ubiquitous looping factor (CTCF) to drive a cell type specific regional architecture conducive to transcription.

Enhancer-gene interactions are dismantled during cell division. Interestingly, recent studies indicate that key lineage factors FoxA1 in hepatoma cells and GATA1 in erythroid cells remain associated with select sites on mitotic chromosomes, some of which have enhancer markings (Caravaca et al., 2013; Kadauke et al., 2012). Moreover cohesin remains on chromatin during mitosis in colon cancer cells after eviction of clustered transcription factors with which it was associated at enhancer-like sites in interphase (Yan et al., 2013). We speculate that enhancers may be appropriate repositories for mitotic bookmarks to re-establish these long range interactions pattern after cell division.

Enhancer loops are tied to the RNA Polymerase II transcription complex and transcription activation

The enhancer’s main job is to increase transcriptional output. This activity could be manifest at different stages including transcript initiation, elongation or termination. Enhancer-promoter interactions can involve components of the basal transcriptional machinery (Ren et al., 2011; Liu et al., 2011; Koch et al., 2011). Moreover, Mediator occupies enhancers of many ES cell genes together with pluripotency factors OCT4, SOX2 and NANOG (Kagey et al., 2010) and bridges the enhancers to Pol II at target promoters by direct interaction with cohesin (Figure 1C). These studies intimately link enhancer loops to transcription initiation. Other studies suggest a role for enhancers in elongation (Sawado et al., 2003; Deng et al., 2012). New work now shows that certain enhancers specifically function to release Pol II pausing and allow elongation (Liu et al., 2013). These anti-pause enhancers loop to target promoters and permit activation of the P-TEFb complex, which is required for release of Pol II into elongation. Enhancers, thus, have diverse mechanistic functions in transcriptional regulation.

Further insight emerges from a ChIA-PET study that determined interactions centered on Pol II across the genome in human cell lines (Li et al., 2012). Networks of interacting promoters and promoters/enhancers were revealed. Enhancer-promoter interactions were significantly enriched for cell type-specific genes. Most genes in multi-gene complexes were highly transcribed and gene families regulated by a common transcription factor were significantly over-represented. A link between these multi-gene complexes and Pol II foci was elegantly drawn by incorporating 3D DNA-FISH and immuno-staining at select loci. New work in murine cells of progressive lineage commitment revealed that the gene clustering patterns are different, linking the looping interactions, coordinated transcription and overall nuclear organization (Zhang et al., 2013). Spatial chromatin connectivity based on Pol II linking of select co-regulated genes revealed that several pluripotency genes in ES cells are connected within one major hub suggestive of a transcription factory.

These studies strongly advocate for enhancer looping as key and likely required for transcription activation. This view is supported by data showing formation of new enhancer loops precedes transcription activation in the β-globin locus (Krivega et al., 2014) (Figure 1D). The Pol II chromatin connectivity data (Zhang et al., 2013) are concordant with earlier data showing enhancer looping and co-localization of erythroid genes regulated by a common transcription factor in a shared transcription factory (Schoenfelder et al., 2010). Whether transcription drives this co-localization is unresolved. It seems reasonable to propose that long range enhancer-gene interactions collectively drive nuclear relocalization and clustering, underlying dynamic focal accumulation of Pol II and the appearance of transcription factories (Buckley and Lis, 2014). Interestingly, not all genes in such foci are active as examples exist where genes with opposite transcriptional responses to a stimulus or differentiation signal were found to co-localize (Hakim et al., 2011; Lin et al., 2012).

Enhancer loops within the global 3D organization of the nucleus

High-throughput studies now place enhancer-gene interactions within a genomic context. Thousands of enhancer-gene interactions associated with the unique transcriptional profile of cells have been observed in 5C studies (Sanyal et al., 2012; Thurman et al., 2012). Hi-C revealed that long range interaction between genomic loci in ES cells and certain differentiated cell types occur predominantly within circumscribed topological domains or topologically associating domains (TADs) (Dixon et al., 2012; Nora et al., 2012). Mega base-sized TADs (0.5–3Mb) are largely conserved across a range of cell types and during development. TAD borders are enriched for CTCF sites, although CTCF sites also occur within TADs. Long range interactions within the same TAD are much more frequent than between TADs and intra-domain loops are more variable among cell types, suggesting that the intra-TAD contacts are involved in cell specific transcription activation. TADs typically have a transcriptionally active or inactive character that can change during development or differentiation when gene activity changes (Denholtz et al., 2013; Lin et al., 2012).

TAD-type organizational structure has been widely observed across multiple mouse tissue and cell types (Shen et al., 2012). Clusters of co-regulated enhancers and promoters were revealed that were defined as enhancer-promoter units (EPUs). Hi-C confirmed looping between the identified promoters and enhancers preferentially within and not between EPUs that overlap with defined TADs (Dixon et al., 2012). TADs also encompass ‘super enhancers’ in ES cells (Whyte et al., 2013). These large domains of up to 0.5 Mb contain clusters of individual enhancer elements that are highly occupied by Mediator and pluripotency factors. Enhancers within these regions are consistently associated with genes that encode important regulators of cell identity and interact with these targets with high frequency.

Adding to the concept of TADs is the genome-wide description of sub-TADs that vary in a tissue specific manner (Phillips-Cremins et al., 2013). CTCF/cohesin and Mediator play roles at different levels of long range interactions that form TAD or sub-TAD organization although a complete description of how these borders differ from one another is still missing. Sub-TAD organization is exemplified by certain HoxA genes and their enhancers that occupy the same TAD but are grouped into distinct topological sub-domains in limb buds where the genes are active (Berlivet et al., 2013). 5C experiments suggest that the enhancers and genes interact with each other through contacts between sub-TADs possibly mediated by CTCF and cohesin, although this was not functionally tested. Supporting this idea, enhancer-gene long range interactions are strengthened by but do not depend on enhancer activity.

Are TADs functionally relevant for proper long range enhancer activation of genes? In support, deletion of a TAD border in the Xist locus resulted in new ectopic contacts and long range transcription mis-regulation (Nora et al., 2012). Moreover, CTCF (but not cohesin) reduction in HEK293T cells resulted in a gain of inter-TAD interactions, suggesting CTCF helps to maintain TAD borders, although changes in gene expression were modest (Zuin et al., 2014). Cohesin reduction in post-mitotic thymocytes had no effect on TAD organization (Seitan et al., 2013), similar to the results of Zuin et al, but in astrocytes relaxation of TAD structure and gene mis-regulation was observed upon cohesin reduction (Sofueva et al., 2013). Differing analysis methodologies may underlie this apparent discrepancy but, overall, the results highlight that much work remains to be done to define the contributions of CTCF, cohesin, Mediator and other factors to maintenance of TAD boundaries and to understand their function. TAD borders are enriched for housekeeping genes, transfer RNA genes and SINES (Dixon et al., 2012) so the field is wide open.

Is an enhancer able to function outside its normal TAD context? An ectopic β-globin LCR reestablished contact with the β-globin gene and activated transcription albeit in only a sub-set of cells (Noordermeer et al., 2011). Moreover, new work suggests that inter-TAD interactions, although less frequent than those within TADs, may be significant for enhancer function. ES cell loci enriched for super enhancers, typically encompassed by TADs (Whyte et al., 2013), co-localize with regions of similar character for transcription (Denholtz et al., 2013). Furthermore, in the HoxD cluster, situated near a TAD border, certain genes are observed to switch enhancer contacts from one TAD to another as gene expression changes across the locus during limb development, indicating a certain fluidity of the border (Andrey et al., 2013).

The picture coming into focus is of genome folding into conserved topological domains. Within TADs, enhancers cluster together with relevant gene targets. EPUs, super enhancers and sub-TADs may all be manifestations of the same enhancer-dependent clustering phenomenon. Although descriptive, this insight is mechanistically important because it means enhancers and promoters need only scan a limited nuclear neighborhood to successfully and efficiently encounter each other (Gibcus and Dekker, 2013). Interruption of a contact site within a cluster of co-regulated genes affected transcription of other interacting genes, supporting the causality of the contacts for the transcription (Fanucchi et al., 2013). This result is consistent with a model in which enhancer-gene clustering within TADs and association of TADs of similar transcriptional character nucleate transcription factories and argues against the possibility that transcription precipitates clustering. If true, this would emphasize the critical role of enhancers and looping interactions in the spatial organization of transcription within the nucleus (Denholtz et al., 2013).

Temporal and spatial regulation of enhancer activity

High-throughput studies across development show that enhancers are progressively modified to activate specific transcriptional programs, particularly by acquisition of the H3K27ac mark (Rada-Iglesias et al., 2011; Creyghton et al., 2010). For example, ‘poised’ developmental enhancers in ES cells are characterized by p300 and Brg1 occupancy, H3K4me1 and low nucleosome density together with Polycomb Repressive Complex 2 (PRC2) and the associated H3K27me3 mark (Figure 2A). At later developmental stages, these enhancers lose PRC2 and H3K27me3 and acquire H3K27ac and the ability to activate gene expression (Rada-Iglesias et al., 2011; Ferrari et al., 2014). New work now reveals an additional ‘inactive’ state of ES cell enhancers with high nucleosome density but occupancy by the Pol II elongation factor ELL3 (Figure 2A) (Lin et al., 2013). These inactive enhancers must harbor ELL3 or they will be unable to activate target genes upon differentiation. Thus, ELL3 may mark enhancers for later activation of genes by Pol II promoter recruitment, which, interestingly, requires cohesin, consistent with looping.

Figure 2. Hallmarks of enhancers at different developmental stages.

(A) Left: Inactive ES cell enhancers have compact nucleosomes but are occupied by ELL3. Center: Poised enhancers are marked by H3K4me1, H3K27me3, PRC2, ELL3, Brg1 and p300. Right: At active enhancers, H3K27ac replaces H3K27me3 upon loss of PRC2. (B) Left: Latent macrophage enhancers lack transcription factor occupancy and enhancer-like histone modifications and contain compact nucleosomes. Center: Inactive macrophage enhancers differ from latent enhancers as they have acquired H3K4me1 and macrophage transcription factors such as PU.1. Right: Active enhancers are further marked by H3K27ac and p300 occupancy. (C) Center: In ES cells, neuroectoderm enhancers have unmethylated CpGs (white circles) and are occupied by Nanog, Sox2, and Oct4. Right: In the ectoderm, Nanog, Sox2, and Oct4 occupancy is diminished, the region remains unmethylated, and transcription of neural genes ensues. Left: Upon differentiation of ES cells to endoderm or mesoderm, Nanog, Sox2, and Oct4 occupancy is lost and the region becomes more compact and highly methylated (black circles), inactivating the enhancers.

As ES cells differentiate, enhancers of pluripotency genes are inactivated to silence their target genes. Several mechanisms have been proposed. One study invoked the activity of the LSD1 histone H3K4/9 demethylase to remove the H3K4me1 enhancer mark (Whyte et al., 2012). In ES cells lacking LSD1, enhancers of pluripotency genes fail to undergo histone demethylation associated with differentiation and their target genes are not repressed. Another study demonstrated that PRC2 deposits the H3K27me2 mark at enhancers in ES cells thereby precluding the active H3K27ac mark and inappropriate enhancer activation (Ferrari et al., 2014). Differentiating cells also require the function of previously inactive enhancers. During the transition from naïve to primed stem cells, the transcription factor Otx2 pioneers new enhancer sites allowing Oct4 to relocalize and activate a new set of target genes that allow cells to exit the naïve state (Buecker et al., 2014). Together, these data provide examples of both negative and positive enhancer regulation to allow orderly developmental progression.

Recent studies have identified changes that occur genome-wide within the regulatory landscape as cells differentiate and gene expression patterns change. Epigenetic profiling of the H3K27ac active enhancer mark in different developing mouse tissues identified approximately 90,000 enhancers that exhibited precise tissue and stage specific windows of predicted activity (Nord et al., 2013). A similar conclusion was reached when DNase I hypersensitive sites were used as the enhancer proxy (Stergachis et al., 2013). Enhancer usage varied widely among cell lineages in an additional study and was linked to recruitment of lineage-determining transcription factors (Kieffer-Kwon et al., 2013). Intriguingly, enhancer usage by commonly expressed genes such as Myc and Pim1 also varies among cell types suggesting that such genes make use of different tissue specific enhancers and transcription factors as cells progress through development.

In differentiated cells, enhancers can respond to external stimuli to activate appropriate genes. For example, TGFβ signaling in pro-B cells or myotubes results in SMAD2/3 recruitment to silent enhancers occupied by PU. 1 or Myod, respectively, and target genes are activated (Mullen et al., 2011). A similar mechanism is employed in macrophages. Upon treatment with lipopolysaccharide endotoxin (LPS), ‘inactive’ enhancers marked by H3K4me1 and tissue specific transcription factors, including PU.1, are rapidly occupied by p300 and activate inflammatory response target genes (Ghisletti et al., 2010) (Figure 2B). Interestingly, a set of ‘latent’ enhancers in macrophages are not sensitive to nucleases and lack both H3K4me1 and PU.1 (Figure 2B) (Ostuni et al., 2013). Upon exposure to LPS, latent enhancers acquire attributes of active enhancers and activate target genes. Although transcription factors can mark inactive enhancers in unstimulated differentiated cells that will be required for later activation of target genes, it is not clear how latent enhancers can be recognized in order to be ‘unveiled’.

Together, these studies reveal a dynamic enhancer landscape at distinct developmental stages linked to cell specific gene expression but how modifications ‘poise’ or ‘activate’ enhancers mechanistically and functionally is much less clear. It is tempting to propose that activation of genes by enhancers during development depends on establishment of loops to target promoters. However, in the HoxD locus, select enhancer-gene contacts are detectable before gene activation (Andrey et al., 2013). Moreover, both active and poised enhancers in IMR90 cells are equally likely to be engaged in looping before gene activation by signaling to TNF-α-responsive enhancers (Jin et al., 2013). Therefore, it will be important that future work focus on the precise temporal relationship of enhancer recognition, poising and activation with enhancer looping and transcriptional activation and to determine the activities required to sequentially specify these outcomes.

DNA methylation as a modulator of enhancer activity

Consistent with the idea that DNA methylation interferes with transcription factor binding (Wiench et al., 2011), genome-wide studies revealed that enhancers are negatively impacted by DNA methylation. A study of the murine methylome in ES and neural progenitor cells established a significant association of DNA hypomethylation at regions distal to promoter CpG islands with transcription factor binding, enhancer histone marks and the activity of enhancers (Stadler et al., 2011). Subsequent studies in human cells affirmed that DNA hypomethylation at enhancers is associated with an open chromatin state, p300 occupancy, and target gene expression (Thurman et al., 2012; Varley et al., 2013).

New studies looked broadly at the dynamics of methylation at regions outside of genes that are likely to be enhancers. While only a small fraction of autosomal CpGs are differentially methylated across numerous human cell and tissue types during development, 26% co-localize with enhancer-like regions (Ziller et al., 2013). Many insights can be gleaned from a large study showing lineage specific gene silencing through asymmetric loss of enhancers by methylation (Gifford et al., 2013). For example, in human ES cells, enhancers destined to regulate genes required for neural specification are poised, unmethylated, marked by H3K4me1 and are occupied by pluripotency factors (Figure 2C). Upon differentiation to neuroectoderm, these enhancers become activated, lose pluripotency factors and, presumably, acquire neural lineage transcription factors. In contrast, when ES cells differentiate into mesoderm or endoderm, pluripotency factors are lost and the enhancers become highly methylated. A consistent theme appears to be loss of DNA methylation and transition to H3K4me1 and either H3K27me3 or H3K27ac to poise or activate enhancers, respectively (Gifford et al., 2013; Sheaffer et al., 2014).

In a different epigenetic analysis of human ES cells and their differentiated progeny, it was observed that DNA methylation is broadly present at enhancers at all stages and negatively associated with their activity (Xie et al., 2013). In some progeny, but not in others, early enhancers acquired methylation when they became inactive. A further study from this group introduced the term vestigial enhancers to describe early enhancers that retain a hypomethylated landscape in adult tissues, possibly as a memory mark. These enhancers exist in closed chromatin and are heavily marked by repressive H3K27me3 which, it is proposed, may be inhibitory to methylation of local DNA residues (Hon et al., 2013). These studies show that DNA hypomethylation is a common hallmark of active enhancers but highlight that hypomethylation is insufficient for activity of enhancers. Contrarily, recent work suggests the opposite, at least at some loci. Loss of Lsh, a modulator of DNA methylation, results in decreased enhancer methylation, increased H3K4me1 occupancy, and increased target gene expression (Yu et al., 2014). More experimentation is needed to resolve this issue.

How is DNA methylation at enhancers regulated? At promoters, Tet1 has been implicated in regulating DNA demethylation through conversion of 5-methylcytosine to 5-hydroxymethylcytosine (5hmC) (Williams et al., 2012). Enrichment of 5hmC and Tet1 occupancy is also associated with enhancers, suggesting that this mechanism might be conserved (Stroud et al., 2011; Pulakanti et al., 2013). Is enhancer DNA methylation the cause of inactivation or a step following other silencing mechanisms? Xie et al interpreted their data to support the latter possibility, consistent with earlier reports (Thurman et al., 2012; Xie et al., 2013; Stadler et al., 2011). Further studies of the dynamics of enhancer DNA methylation-demethylation and function during development may resolve this fundamental question.

Regulated production of transcripts from enhancers and from their target genes

It had long been known that transcription occurs at enhancer elements, but it was widely assumed to be ancillary to transcription activation of target genes. Enhancer transcripts of various types are now known to occur on a genome scale and can be related to the transcription of select neighboring regulated genes (De Santa et al., 2010). Transcription of an eRNA upstream of the Arc gene requires the intact promoter and gene hinting that eRNA transcription and enhancer-promoter communication are linked (Kim et al., 2010).

Several different kinds of stimuli can affect eRNA transcription. Androgen receptor, p53, MyoD and MyoG positively regulate eRNAs by direct binding to enhancers (Wang et al., 2011; Melo et al., 2013; Mousavi et al., 2013). In contrast, REV-ERB signaling can negatively regulate target eRNAs as evidenced by reduced eRNA transcription following Rev-Erbα/Rev-Erbβ over-expression (Lam et al., 2013). Enhancer DNA methylation may also play a role in eRNA transcription (Pulakanti et al., 2013). Demethylation of mature B cell enhancers accompanies their transcription into eRNAs and their activation (Schlesinger et al., 2013). In each of these examples, regulation of eRNA transcription was associated with the appropriate positive or negative effect on mRNA transcription of target genes supporting a mechanistic link.

Mechanism by which eRNA may function to regulate target gene expression

High-throughput studies established a significant association between eRNAs, enhancer-promoter interactions and target gene expression (Sanyal et al., 2012). Does the eRNA transcript per se have a role? Although the mechanism of nuclear targeting remains unclear, siRNAs and locked nucleic acids (LNAs) have been used to successfully deplete nuclear transcripts (Berteaux et al., 2008; Sarma et al., 2010). Using these approaches, when eRNAs transcribed from estrogen receptor-alpha (ERα)-sensitive enhancers were reduced in breast cancer cells, target gene expression was reduced (Li et al., 2013b). The eRNAs were associated with cohesin, and both looping to the genes and cohesin occupancy were reduced after eRNA inhibition, suggesting that eRNAs and cohesin stabilize the contacts (Figure 3A). eRNAs can also interact with the Mediator complex (Lai et al., 2013). Knockdown of MED1 and MED12 compromised eRNA activation of target genes and decreased loop formation between the enhancers and genes. Consistent with this, an eRNA transcribed from an androgen response element is required for Med1 occupancy, chromatin looping and target gene expression (Hsieh et al., 2014). These studies suggest a direct role of eRNAs in formation or stabilization of loops between enhancers and target genes.

Figure 3. Mechanism of eRNA function.

(A) To activate target genes, eRNAs can interact with Mediator and cohesin to form long range interactions between the enhancer and target genes. Loss of eRNAs or Mediator results in decreased target gene expression. (B) eRNAs can mediate target gene expression through influencing chromatin remodeling at target promoters. It is unclear if this mechanism occurs independently of chromatin looping.

In contrast, other studies questioned the role of eRNAs in looping. One study focusing on eRNAs transcribed at p53 dependent enhancers found that the transcribed enhancers physically interacted with multiple target genes whose expression was reduced after eRNA depletion (Melo et al., 2013). However, the loops were not diminished after p53 or eRNA reduction. Similarly, chemical inhibition of transcription elongation of eRNAs activated by ERα binding did not reduce looping (Hah et al., 2013). In these studies, enhancer-gene loops are maintained after reduction of eRNAs required for transcription of target genes but they do not rule out a requirement for eRNAs to initially form loops or, alternatively, for another function related to target gene activation. In fact, work in which chromatin looping was not directly examined implicates eRNAs in chromatin remodeling at target genes (Mousavi et al., 2013). Reduction of eRNAs regulating Myod1 or Myog expression results in decreased promoter Pol II occupancy and DNase I hypersensitivity, while eRNA over-expression suggests the transcript could function from an ectopic location (Figure 3B).

Together these data suggest that eRNA transcripts themselves may be functional and directly involved in target gene activation, at least in some cases. However, caution is required in drawing conclusions and new approaches are needed to strengthen or refute current findings regarding eRNA functions. Moreover, underlying mechanisms of eRNA function on target gene transcription are still unclear. It is of considerable interest that eRNAs may function in chromatin looping via interaction with cohesin and/or the Mediator complex (Li et al., 2013b; Lai et al., 2013; Hsieh et al., 2014) and this issue warrants further study.

Conclusions

Genome-wide studies have changed the perspective in which we view enhancers and opened new possibilities. One important goal is to identify active enhancers in a high-throughput fashion. STARR-seq in Drosophila, provides a promising approach (Arnold et al., 2013). However, of primary importance is the ability to identify natural targets of enhancers as nearby genes are frequently skipped in favor of more distant ones (Li et al., 2012; Sanyal et al., 2012; Zhang et al., 2013). Computational improvements in Hi-C and greater sequencing depth in ChIA-PET bring us closer to this goal by identifying enhancer-gene pairs with greater resolution (Jin et al., 2013; Zhang et al., 2013; Kieffer-Kwon et al., 2013). Ultimately, however, candidate enhancer must be functionally tested. Genome editing with TALE nucleases (TALENs) has been deployed to delete or inactivate enhancers (Kieffer-Kwon et al., 2013; Mendenhall et al., 2013). TALEN or CRISPR/Cas9 (Jinek et al., 2013; Miller et al., 2011) methodologies appear destined to become the sine qua non to define enhancer activity in an endogenous environment. However, these approaches are still low-throughput and considering that one gene may have more than one enhancer and vice versa they may be painstaking.

Do enhancers control gene expression by looping? Single cell studies capable of visualizing proximity of genes and enhancers at high resolution will be critical to reveal the repertoire and dynamics of enhancer-gene looping and its relationship to transcription. Hi-C can now be performed on single cells revealing considerable variation in the interactome among cells of the same type (Nagano et al., 2013). Approaches like this may allow us to understand the relationship between biochemically determined enhancer-gene loops and proximity determined by DNA-FISH, which are not always concordant (Belmont, 2014). How do the patterns of enhancer-gene proximity relate to the firing of nascent transcripts? What is the relationship of physical clustering of genes in transcription factories observed by microscopy and topological features of chromatin, including enhancer looping and clustering in TADs, gleaned from populations of cells by 3C-related methodologies? These fundamental question should become approachable, and in live cells (Ghamari et al., 2013).

Finally, it is becoming widely appreciated that sequence variation or other disruption of distant enhancers, is a critical aspect of mis-regulation of gene expression in disease (Maurano et al., 2012). Both sequence and methylation polymorphisms in enhancers are likely to be important but recent work indicates a better correlation with altered target gene expression of the latter compared to the former (Aran and Hellman, 2013). A more in-depth understanding of enhancer function is needed to point the way to disease mechanisms and potential interventions. Improvements to existing technologies are emerging at a fast and furious pace. We can expect our understanding of enhancers to broaden and deepen rapidly.