Nanog Expression in Embryonic Stem Cells – An Ideal Model System to Dissect Enhancer Function

Abstract

Embryonic stem cells (ESCs) are derived from the preimplantation embryo and can differentiate into virtually any other cell type (termed pluripotency), which is governed by lineage specific transcriptions factors (TFs) binding to cis regulatory elements (CREs) to mediate changes in gene expression. The reliance on transcriptional regulation to maintain pluripotency makes ESCs a valuable model to study the role of distal CREs such as enhancers in modulating gene expression to affect cell fate decisions. This review will highlight recent advance on transcriptional enhancers, focusing on studies performed in ESCs. In addition, we argue that the Nanog locus, which encodes for an ESC-critical TF, is particularly informative because it contains multiple co-regulated genes and enhancers in close proximity to one another. The unique landscape at Nanog permits the study of ongoing questions including whether multiple enhancers function additively versus synergistically, determinants of gene specificity, and cell-to-cell variability in gene expression. See also the video abstract here: https://youtu.be/tnW2Z6bYF8E.

1. Introduction

Embryonic stem cells (ESCs) are defined by two criteria: 1) pluripotency, the capacity to differentiate into cell-types of all the primary germ layers and 2) unlimited self-renewal, the capacity to propagate indefinitely. Murine ESCs are derived from the inner cell mass (ICM) of pre-implantation blastocysts and can be perpetually cultured ex vivo while maintaining these two cardinal properties.^{[1, 2]} Thus they are a perfect model to better understand the molecular mechanisms that govern cell-identity and lineage commitment.^{[3, 4]} In addition, because they can be grown in large quantities and easily genetically manipulated using traditional homologous recombination^[5] or new genomic editing approaches based upon Cas9/Clustered Regularly Interspaced Short Palindromic Repeats CRISPR),^[6] these cells are a powerful platform for many different genome-wide approaches such as proteomics, transcriptomics, epigenomics, and classical genetic approaches. In particular, because ESCs can differentiate into virtually any cell type, the role of different pathways and/or genetic elements during differentiation studies can be studied.

Within this review, we will focus on why ESCs and the Nanog locus are an ideal system to study enhancers. First, we discuss ESC biology, and important questions in the field related to the regulation of gene expression. Second, we present the current “state of the art” in terms of how transcriptional enhancers are identified in addition to the larger role of genomic architecture to their function. Third, we focus on the Nanog locus as an invaluable model system to study transcriptional enhancers. Finally, we conclude with important open questions into how transcriptional enhancers work to govern core properties of ESCs.

2. ESC Biology – A Brief Primer on the Transcriptional Regulation of Pluripotency and Self-Renewal in ESCs

The core properties of pluripotency and self-renewal define ESCs and are regulated by maintaining a specific group of lineage specific transcription factors (TFs; Nanog, Oct4, and Sox2-NOS) that comprise the pluripotency regulatory network.^[7–10] The core ESC TFs are trans-acting factors and maintain pluripotency by positively regulating loci essential to cell-identity, while also repressing differentiation programs. Transcriptional regulation is mediated by NOS binding at cis-regulatory elements (CREs) and recruitment of co-activating or repressive complexes. Two classes of CREs, enhancers and promoters, are present on the same DNA molecule (e.g., chromosome) as the gene they regulate. Enhancers, in contrast to promoters, regulate transcription independent of distance and orientation.^{[7–9, 11, 12]} Nanog, Oct4, and Sox2 can be found at both promoters and enhancers within ESCs, although typically co-occupancy by all three factors at a promoter correlate well with gene activation, whereas occupancy by a single factor tends to correlate with gene repression.^{[9, 13]} The importance of these factors to pluripotency was highlighted by seminal studies from Shinya Yamanaka demonstrating that four TFs (Oct4, Sox2, Klf4, and c-Myc) can reprogram somatic cells into a pluripotent state.^[14] Thus, the primary action of core TFs is to establish and maintain both pluripotency and self-renewal by regulating gene expression (details reviewed in Refs.^{[3, 4, 15]}). Given the abilities of these TFs to bind both promoters and enhancers to regulate gene expression, it remains an important area of investigation to delineate how TF binding at different types of cis-regulatory elements regulate gene expression and pluripotency.

3. ESC Biology – The Role of Nanog in Pluripotency

Nanog is a homeodomain TF critical for establishment of the inner cell mass and repression of primitive endoderm (PE) TFs. Nanog null embryos are not viable and do not form an epiblast, consistent with an absolute requirement for Nanog in the earliest steps of embryogenesis.^[16] Nanog was initially identified by two groups during screens to identify factors which bypassed the need for leukemia inhibitory factor (LIF) to maintain pluripotency.^{[16, 17]} LIF is a cytokine and functions through the Jak/Stat3 and PI3 K/AKT pathways upregulating Klf4 and Tbx3 which positively upregulate Sox2 and Nanog, respectively.^[18] Nanog forms homo and heterodimers with both co-activators (Oct4, Sox2) and co-repressors (NuRD, Polycomb) via its tryptophan-rich (WR) domain, which is essential for murine ESC self-renewal and pluripotency.^{[19, 20]} Loss of a single core TF such as Nanog can collapse the pluripotency network resulting in a loss of pluripotency and subsequent differentiation. Thus, maintaining a critical threshold of each Nanog and other core TFs, which cross-regulate one another’s expression via numerous feedback loops is essential to maintain cell-identity. A fascinating property of ESCs which is shared with preimplantation embryo’s is the cell-to-cell heterogeneous expression of several TFs, especially Nanog,^[21–24] which may be partially explained by autoregulation.^{[20, 25, 26]} Nanog positively regulates itself through interactions with Oct4/Sox2 and prevents over-expression (independent of Oct4/Sox2) through interactions with the Kruppel-like zinc finger TF Zfp81 which recruits the NuRD repressor complex. Fluctuations in various NOS regulated pluripotency factors (Nanog, Dppa3, Rex1, Esrrb, Klf4) have been well documented.^{[18, 27–29]} Oscillations in Nanog levels are biologically significant and have been linked to cell fate decisions such as self-renewal (Nanog high) and differentiation (Nanog low; more details below^{[30, 31]}).

Heterogeneous expression of ESC TFs in serum/LIF results in a metastable pluripotent state whereby ESCs maintain pluripotency and self-renewal, however, some cells have a propensity to undergo spontaneous lineage commitment. Culturing murine ESCs in serum and feeder free conditions in the presence of LIF and two inhibitors (2i) of differentiation pathways (MEK and Gsk3) mimics the pre-implantation blastocyst (also referred to as the naïve or ground state ICM) and results in homogenously high levels of many pluripotency factors including Nanog and Dppa3 which are among the most variably expressed genes during transition to the ground state.^{[12, 21, 32, 33]} The transcriptome of Nanog high cells in serum/LIF are unique from homogenously Nanog high cells cultured in 2i/LIF likely due to differences in NOS occupancy. Nanog has ≈15 000 differential binding sites in 2i/LIF compared to serum/LIF. Oct4 and Sox2 have significantly fewer unique binding sites between conditions (500 and 1300, respectively^[12]). Differential TF occupancy correlates with putative enhancer activity and precedes changes in bivalent histone marks and DNA hypomethylation. Indeed, naïve ground state ESCs are epigenetically distinct from the metastable serum/LIF ESCs. Temporal changes in NOS occupancy during development and in serum/LIF conditions may permit enhancer switching resulting in heterogeneous Nanog expression.^[34] Importantly, which CREs (i.e., promoters and/or enhancers) are critical to mediating this heterogeneity remains an open question.

In summary, Nanog expression is critical to the establishment and maintenance of pluripotency. While the TFs involved in pluripotency are well-described, the cis-regulatory elements they utilize is not clearly delineated. In fact, whether promoters or enhancers are the primary element by which gene expression is regulated by NOS remains a critical question. Finally, given the highly heterogeneous expression of key pluripotency TFs such as Nanog, it remains unclear the precise molecular mechanisms which govern cell-to-cell variability, and which CREs are required. Given the importance of heterogeneity to the earliest stages of mammalian development^{[26, 30, 35]} it remains important to uncover the underlying mechanism which govern it.

4. ESC Biology-Pinpointing Factors That Fine-Tune Nanog Transcription and Cell-to-Cell Heterogeneity

Several studies have demonstrated that Nanog undergoes monoallelic transcription in early embryogenesis and switches to biallelic expression in the naïve ground state ICM.^{[36, 37]} Allelic switching is believed to play a role in maintaining homogenously high levels of Nanog in the 2i/LIF and more variable expression in 2i/LIF. RNA-FISH experiments in serum/LIF conditions (bimodal Nanog levels) revealed that a minority of cells (14%) undergo biallelic Nanog expression and 45% undergo monoallelic expression.^[36] Moreover, high variability in the number Nanog mRNA molecules per cell (ranging from 0 to 500) fit with a model of discontinuous expression (transcription bursting) at Nanog.^[37] Ochiai et al.^[38] confirmed pulsatile transcription at Nanog using a Nanog-MS2 murine ESC line and observed that transcription frequency was higher in 2i/LIF conditions (3.17 min) than in serum/LIF conditions (1.64 min). Similar results were shown by a different group as well.^[39] Transcription signal intensity (burst size) between the two conditions was similar suggesting that burst fraction (frequency of Nanog transcription) underlies differences in the number of transcripts. This is similar to findings in other cell lines suggesting burst frequency is the primary determinant of RNA quantity.^{[40, 41]} Despite compelling evidence of Nanog heterogeneity, it remains unclear if rapid allelic switching during early embryogenesis or in serum/LIF culture conditions underlies bimodal levels of Nanog expression. Moreover, no one has tested the in vivo function of enhancers in controlling heterogeneous and stochastic Nanog expression (i.e., bursts of transcription secondary to dynamic enhancer-promoter contacts). Collectively, these studies suggest that Nanog heterogeneity arises at the transcriptional level, and dynamic enhancer-promoter interactions, which vary at the single cell level^[42] may control heterogeneous Nanog expression during development, perhaps through enhancers.^[43]

Single cell reverse transcriptase quantitative PCR (RT-qPCR) allows further insights into cell to cell variation and transcription dynamics. Several groups have shown the transcriptomes of murine ESCs cultured in 2i/LIF and serum/LIF cluster into subpopulations in vitro and in vivo.^{[21, 44, 45]} Guo et al.^[21] demonstrated many of the most variably expressed genes in serum/LIF contain bivalent chromatin marks (H3K27me3+/H3K4me3+) at their promoters. Bivalent genes were previously described to be silent or repressed genes, however, single cell RT-qPCR revealed that many bivalent genes are actively transcribed, suggesting that transcription may be highly dynamic at these loci.^[21] This represents a limitation of ChIP-Seq data which shows a population average of the histone mark at a given gene. Further, Guo et al.^[21] observed that heterogeneity does not appear to be stochastic; rather ESCs follows a defined differentiation pathway toward primitive endoderm (PE)-like cells with decreasing levels of pluripotency TFs and up-regulation of PE specific TFs. Importantly, in vitro culture conditions are developmentally relevant, as single cell transcriptomes are similar to different stages of the developing mouse blastocyst.^{[21, 44]} Thus, in vitro transcription studies provide an accurate assessment of in vivo cell fate decisions. Single cell transcriptome studies in enhancer edited ESCs or embryos have not been performed yet, and thus the contribution of distal CREs to cell fate cell fate decisions in the early embryo remain to be elucidated.

5. Transcriptional Enhancers – Critical Regulators of Gene Expression

Enhancers were previously identified over 30 years ago as short DNase hypersensitive (HS) genomic regions that modulated expression in transgenic mice or plasmid-based reporter assays.^{[46, 47]} Since the advent of next generation sequencing (NGS) based approaches combined with epigenetics, putative enhancers can be identified genome wide in any tissue by their epigenetic signature. Histone H3 lysine 4 monomethylation (H3K4me1) is a hallmark for all enhancers, whereas the presence of histone H3 lysine 27 acetylation (H3K27Ac) and DNA hypomethylation further defines an active enhancer.^[48–51] This approach revealed that the mammalian genome contains more than one million enhancers as compared to ~25 000 genes.^{[52, 53]} This complex repertoire of enhancers is critical to the intricate spatial and temporal expression of genes during development. Recently, several studies unveiled a subset of enhancers termed super-enhancers or stretch-enhancers, that co-localize with genes critical to cell-identity and disease and contain a high density of master TFs (such as NOS), the transcriptional co-activator complex Mediator, and high levels of H3K27Ac (Figure 1^[54–57]). A cardinal feature of super-enhancers is that they are RNA polymerase II (RNAPII) bound and bidirectionally transcribed producing rare, unspliced long non-coding RNAs termed eRNAs (more details below^{[55, 58]}). eRNAs have biological function and positively regulate neighboring genes in cis.^{[43, 59–63]} These large CRE clusters (often greater than 10 kb) represent less than 5% of all enhancers in a cell, but a wealth of literature supports them as key drivers of cell fate decisions. Accordingly, identification of these enhancers and their associated genes may facilitate mapping of the regulatory circuitry of different mammalian cell-types using basic models that describe transcriptional control. Thus, the advent of genome-wide ChIP-seq approaches have greatly advanced our ability to identify CREs in a range of cell-types. Nonetheless, determining whether these CREs are required in vivo to regulate gene expression remains an important area of research.

Figure 1.

Schematic of a cluster of enhancers (e.g., super-enhancer) looping into and activating expression at a neighboring gene. Both the enhancer and promoter are bound by transcription factors including Nanog, Oct4, Sox2, and co-activators such as Mediators. Enhancers preferentially recruit Brd4, p300 (catalyzes H3K27Ac) and Serine 5 phosphorylated RNA polymerase II (RNAPII). Promoters contain the elongating Serine 2 phosphorylated RNAPII and are marked by high levels of H3K4me3. Note that not all enhancers within the cluster recruit RNAPII and are bidirectionally transcribed and produce enhancer RNAs (eRNAs). eRNAs may: 1) stabilize chromatin looping through interactions with cohesion complex or Mediator; 2) interact and activate p300; or 3) act as a sponge for negative elongation factor (NELF).

Plasmid-based assays which link a reporter (such as luciferase) to CREs have been invaluable at identifying DNA sequences which may potentially play a role as an enhancer. However, one limitation of these assays is because they are plasmids they are unable to fully recapitulate the chromatin environment within which a gene normally resides. Because of this issue, many labs endeavored to develop new approaches to measure the proximity of DNA elements through molecular based approaches. A shared feature of all enhancers is that they must physically interact with a promoter to regulate gene expression.^[64] This is accomplished by bringing the enhancer and promoter into close physical proximity by “looping out” the intervening DNA segment. Chromatin looping is mediated by architectural proteins including CTCF and the cohesin complex.^{[65, 66]} Chromosome conformation capture (3C) related assays have been used to detect local and genome wide CRE interactions (Table 1). Spatial chromosome organization datasets in conjunction with the binding profile of the nuclear architecture protein CTCF, led to the discovery that CREs exist within topology associated domains (TADs^{[67, 68]}). TADs are large chromatin regions (≈1 MB) that demarcate high interaction of nearby genomic elements but fewer interactions with chromatin segments in neighboring domains. At a finer level, ESC super-enhancers and their associated genes were shown to exist within “insulated neighborhoods” which permit high interaction frequency of enhancers and lineage critical genes.^[69] It is important to note that until recently these technologies were limited by their reliance on bulk analysis of fixed cells. More recent approaches indicate that they can be “scaled-down” to the single cell level,^{[42, 70]} a technical advance that will yield exciting new understanding of how CREs regulate cell-type specific gene expression.

Table 1.

Higher order chromosome architecture studies in murine embryonic stem cells.

Technique	Species	Comments	Reference
3C	Mouse	DNase HS site interaction with Nanog and 5 kb CRE	[82]
3C	Mouse	Enhancer interactions at Sox2	[76]
3C	Mouse	Super-enhancer interactions at Nanog	[43]
4C	Mouse	Genome-wide interaction with Nanog in ESCs, pre-IPSCs, and IPSCs	[90]
4C	Mouse	Genome-wide interaction with Nanog in ESCs	[91]
4C	Mouse	Genome-wide interactions in ESCs (including Oct4, Dppa3, Dppa2)	[76]
Hi-C	Mouse	Genome-wide interaction with Nanog	[89]
Hi-C	Mouse	X chromosome interactions	[68, 112]
Hi-C	Mouse	Promoter or gene enriched Hi-C	[75, 112]
ChIA-PET	Mouse	Smc1a pull-down	[69]
Imaging + NGS	Mouse	Genome architecture mapping	[113]

A listing of papers which dissect higher order chromatin structure in ESCs. We would like to note the listing is not comprehensive, but instead intended to provide a series of key references.

6. Transcriptional Enhancers – Three-Dimensional Genome Architecture of ESCs

Nearest neighbor analysis is commonly used to correlate enhancers with the genes they regulate. However, many enhancers regulate genes over extremely long distances (>100 kb) by looping-out intervening chromatin containing repressed or autonomously regulated genes. Thus, specific chromatin looping events coordinate dynamic spatial and temporal expression of genes during development.^{[71, 72]} Several proteins with roles in chromatin looping have been identified including the insulator protein CTCF, the cohesin complex, and Mediator.^{[65, 73]} Mediator and cohesin physically and functionally connect enhancers and promoters of active genes in ESCs. Depletion of Mediator (Med12), cohesin (Smc1a), Nipbl/Wapl (roles in loading/unloading cohesin) results in decreased NOS expression and up-regulation of lineage factors.^{[65, 74]} Despite clear biological roles in maintaining pluripotency, how these proteins mediate cell-type specific looping events such as insulated neighborhoods are not well understood.

While interaction of an enhancer:gene dyad through 3C-type approaches can imply the enhancer regulates the gene’s expression, it has becoming increasingly clear that many of these interactions between regions may be secondary to the formation of higher order chromatin domains, which organize the genome in three dimensions. Spatial organization of the genome is tightly linked to gene expression and therefore predicted to be variable between different tissue types. High throughput chromosome capture (Hi-C) studies in murine ESCs revealed that megabase sized TADs are largely preserved between species and cell-types (Table 1).^{[67, 68]} Thus, TADs alone are not sufficient to understand tissue specific gene expression. To better identify long-range CRE interactions in murine ESCs at greater resolution, Schoenfelder et al.^[75] used enriched Hi-C libraries for >22 000 annotated promoters. This group observed that ≈40% of promoters interact with several (2–10) enhancers, whereas ≈12% of promoters interact with more than ten enhancers. Similar to promoters, enhancers showed multiple interacting partners with a majority (≈69%) contacting one to five promoters, and a minority (2%) of “highly connected” enhancers interacting with more than five promoters. A positive correlation between gene expression and the number of interacting CREs suggests additive effects of putative enhancers on transcription output. Moreover, this group showed that active and inactive promoters were largely spatially segregated. This parallels findings by Denholtz et al.^[76] demonstrating that transcriptionally active regions of the genome (NOS occupied) segregate from repressed Polycomb bound regions. Finally, promoters within the same expression category (high or low expression)were shown to preferentially interact. Colocalization of highly expressed genes and active CREs may enhance the likelihood of chromatin interactions or availability of transcription activators, thereby enhancing transcription. Collectively, this implies that while TADs are relatively cell-type invariant, intra-TAD interactions are highly lineage specific and critical to driving cell-type specific gene expression.

7. Transcriptional Enhancers-Super-Enhancers in Gene Regulation and Genomic Architecture

Whyte et al.^[54] identified 231 super-enhancers in murine ESCs, however, this study relied on nearest neighbor gene analysis and did not examine physical CRE interactions. Subsequent work by Schoenfelder et al.^[75] demonstrated that 142 of the 210 neighboring genes physically interact with super-enhancers. In addition, ESC super-enhancers interact with 361 other genes, and thus may control expression of many more genes than previously thought. While super-enhancers do not contact more promoters than typical enhancers, highly expressed genes are overrepresented among their targets.^[75] This provides further evidence that clusters of enhancers may co-localize and reinforce robust expression of genes critical to cell-identity. Interestingly, almost all genes that interact with super-enhancers in ESCs (98.2%) also physically interact with other typical enhancers, thus super-enhancers act in the context of larger 3D regulatory networks. In sum, NOS and cohesin occupied CREs are highly centralized and associate at higher than expected frequencies suggesting spatial chromosome organization contributes to coordinated tissue specific gene expression.

There is a great deal of debate in the field as to whether super-enhancers have greater roles in activating gene expression compared to typical enhancers. A recent study by Moorthy et al.,^[77] compared in vivo function of super-enhancers and H3K27Ac high typical enhancers. They demonstrated that typical enhancers have equivalent capacity to drive robust gene expression and that similar to Nanog, super-enhancer function at Tet1 and Sall1 was highly variable, controlling transcription output from 12% to as much as 92%. This group confirmed that enhancers can functionally regulate numerous genes as suggested by a previous Hi-C study.^[78] Moreover, they demonstrated that constituent enhancers within a super-enhancer have partially redundant control in gene regulation. Analogous enhancer editing studies at the alpha globin locus showed enhancers act additively to drive robust expression in vivo.^[79] In contrast, constituent enhancers at Wap (mammary tissue) and Mitoferrin1 (erythroid cell line) appear to function in a hierarchy.^{[80, 81]} Only a handful of these studies have been performed with no consistent paradigm (additive vs. synergistic function) and thus at present multiple enhancer control appears to be locus dependent. This illustrates the importance of continued work utilizing a combination of 3C-type approaches in conjunction with genomic editing to elucidate how enhancers and genes functionally interact.

To gain a better understanding of sub-TAD gene loops in controlling pluripotency, Dowen et al.^[69] used Chromatin Interaction Analysis by Paired-End Tag Sequencing (ChIA-PET) with Smc1a as an anchor (Table 1). This study confirmed that super-enhancers and genes critical to cell-identity co-localize within sub-TAD structures formed by cohesin and CTCF termed insulated neighborhoods. A majority of super-enhancers (>80%) are restricted to CTCF insulated looping structures (averaging 106 kb) in contrast to only 48% of typical enhancers. CRISPR-mediated deletion of CTCF sites at insulated neighborhood boundaries showed perturbation of neighboring gene expression on either side of the boundary. Specifically, deletion of a CTCF site downstream of the −45 Nanog super-enhancer resulted in a ≈40% decrease in Nanog expression. This is similar to the decrease in expression following deletion of the −45 enhancer,^[43] signifying that CTCF permits the distal enhancer to interact with Nanog. Collectively, these findings argue that intra-TAD CTCF sites play important roles in gene expression and do not exclusively block or insulate enhancers from interacting with promoters. Rather they play a role in organizing the genes into structural units that mediate proper gene expression. This suggests that the intra-TAD landscape is necessary to permit proper gene expression. The next logical step is to query which of these DNA elements are critical to regulating gene expressions and/or 3D chromatin architecture.

8. The Nanog Locus – Different Pluripotency Associated Genes Used to Connect Enhancers to Gene Expression

Many studies have utilized ESCs as a model system to study how chromatin domains interact to regulate gene expression. One of the earliest was a 3C based study in ESCs by Levasseur et al.^[82] focused on the intra-TAD landscape surrounding the pluripotency associated transcription factor encoded by Nanog. This locus was selected for study because a large number of nearby genes are expressed in pluripotent cells and demonstrated that the Nanog promoter as well as the −5 enhancer interact with many NOS occupied DNase HS elements including the −45 enhancer, Gdf3, Dppa3 (Table 1 and Figure 2). Oct4 ablation resulted in decreased Nanog expression secondary to loss of chromosome architecture at this locus, particularly at distal enhancers. This was the first study in ESCs demonstrating a role of a core pluripotency TF in organizing putative regulatory elements into a compact chromatin structure. Similar studies have found that Klf4^[83] plays a similar role in regulating two distinct enhancers which play distinct roles in regulating Oct4 expression in early development.^[84] These studies highlight how informative studies utilizing 3C-based approaches can be to study not just which enhancers are involved in regulating a single locus, but also how TFs differentially shape 3D genomic architecture in different lineages.

Figure 2.

Schematic of the extended Nanog locus on murine chromosome 6. Gdf3, Dppa3, and Nanog are co-regulated by NOS. The Slc2a3 promoter is not occupied by NOS. The Nanog locus contains three ESC specific super-enhancers. Genes are indicated by black rectangles and super-enhancers are indicated by red rectangles.

While these studies highlight the power of 3C-type approaches, one limitation is they do not directly address whether these CREs are required for proper gene expression. Plasmid based reporter assays, as mentioned above, can often detect whether a CRE possesses enhancer potential, but are unable to definitively establish enhancer activity in vivo. Multiple groups have turned to genomic editing to address this issue. One key example is at the Sox2 locus, where two groups^{[85, 86]} utilized a combination of ChIP-seq and 3C to identify a distal enhancer >100 kb downstream of the locus in addition to an intronic enhancer, both of which interacted with the Sox2 promoter specifically in ESCs and MEFs. Interestingly, deletion of the intronic enhancer by genomic editing had minimal impact on Sox2 expression whereas deletion of the distal enhancer caused a near complete loss of Sox2 expression. The key point of both studies is that the combination of reporter assays and 3C was able to identify DNA elements with the potential to regulate Sox2 expression, but determining which enhancer(s) are required for gene expression necessitated a genetic approach. A second example is the Pou5f1/Oct4 locus, in which the proximal enhancer (1 kb upstream) is more active in “primed” (post-implantation) ESCs and the distal enhancer (2 kb upstream) preferentially controls expression in “naïve” ground-state ESCs.^{[84, 87, 88]} The ease of genomic editing in ESCs will dramatically improve our understanding of which enhancers play a role in regulating which genes.

9. The Nanog Locus – Key Reasons Why Nanog Represents an Ideal Locus to Study How Enhancers and Genomic Architecture Regulate Gene Expression

One critical caveat of the Sox2 locus is that it resides within a gene desert, with >100 kb on either side of it. Thus, while an excellent model system to study enhancer:promoter interactions and regulation, the locus does not permit detailed studies on how different enhancers and genes interact in concert to regulate gene expression. In addition, because essentially a single gene lies within this TAD,^[89] it makes it harder to study the more complex but common situation of how higher order genomic organization impacts the expression of multiple genes. In this respect, the Nanog locus displays distinct advantages. First, there are multiple pluripotency associated genes within close proximity (≈150 kb, Figure 2) including both Dppa3 and Gdf3, in addition to ubiquitously expressed genes (Slc2a3) and other genes which are expressed in early embryogenesis but in adult animals are tissue restricted (Apobec1, Foxj2). Also, the locus contains three different super-enhancers^{[43, 54]} and a number of different traditional enhancers. Thus, theNanog locus present an important opportunity to study not just how enhancers interact with a given gene to regulate expression, but also how multiple enhancers interact with each other and multiple genes simultaneously to regulate tissue specific and ubiquitous gene expression.

Two subsequent studies examined the interactions between Nanog and other CREs using circular chromosome conformation capture (4C) and identified a Nanog centered pluripotency interaction network (Table 1). Apostolou et al.^[90] noted that a large fraction of Nanog interacting sites were bound by Mediator and cohesin and that depletion of these looping factors resulted in a loss of ESC specific interactions (including upstream cis elements in Levasseur et al.^[82]) and acquisition of a differentiation set of CRE interactions that preceded transcriptional changes. In parallel, de Wit et al.^[91] showed very little role of Oct4 and Nanog in controlling overall genome wide topology (consistent with findings that TADs are tissue independent), but critical functions in driving a pluripotency specific network (similar to findings by Ref.^[90]) including an inter-chromosomal interaction with Esrrb. In addition, de Wit et al.^[91] noted that in contrast to Oct4, only complete Nanog knockout resulted in loss of architecture. This suggests a greater tolerance for Nanog loss than Oct4 in organizing pluripotency specific interactions. Further, this group demonstrated that artificial targeting of Nanog can induce inter-chromosomal contacts. In sum, spatial clustering of pluripotency loci into nuclear transcription hubs facilitates coordinated regulation of critical ESC genes (e.g., Nanog and Esrrb), however, these 3D architecture studies only correlate with gene expression and do not directly assay their contribution to transcriptional output. This raises a broader and intriguing hypothesis – specifically that intra-TAD organization may be critical to permitting dynamic gene expression changes on a cell-to-cell basis or in response to extracellular signals.

While high-throughput sequencing technologies and genome editing have catapulted our knowledge of transcriptional regulation in the last decade, many questions remain as to how numerous enhancers function to regulate a single gene or a cluster of co-regulated genes in their native genomic context. For example, do multiple enhancers cooperate at a single locus to drive robust transcriptional output? Or does a single enhancer operate on a gene at any one point in time, but enhancers may “switch” dynamically? Despite a plethora of transcriptional regulation studies at Nanog, the regulatory events at neighboring CREs (looping and/or recruitment of activating and repressive complexes) that mediate heterogeneous Nanog expression during development and in serum/LIF remains unknown, and even if different enhancers play a role in heterogeneous expression of Nanog. Future single cell studies of transcription kinetics following enhancer editing may uncouple how transcribed enhancers fine-tune Nanog levels in pluripotency.

To begin to address these types of question, our own group utilized CRISPR-mediated deletion of each Nanog super-enhancer and revealed differential control of neighboring genes with only the proximal −5 enhancer necessary for robust Nanog expression and pluripotency.^[43] Interestingly, ablation of the −45 enhancer (which lies 35 kb downstream of Dppa3) resulted in decreased expression of both Dppa3 and Nanog. Deletion of the +60 enhancer, however, did not alter expression of genes at the extended Nanog locus. Finally, our group showed that −45 enhancer deletion did not disrupt spatial organization at this locus.^[43] In sum, despite reporter assays demonstrating equivalent activating properties of all three enhancers for Nanog and 3C indicating all three enhancers are in close proximity to the Nanog promoter, in vivo genomic editing revealed a functional enhancer hierarchy, with the −5 kb enhancer most critical to Nanog expression, the −45 kb enhancer playing a modest role, and the +60 enhancer having no influence on expression. Functional studies of Nanog super-enhancers during early embryogenesis and primordial germ cell specification have yet to be performed. This may reveal temporal differences in enhancer strength secondary to re-wiring of the endogenous sub-TAD chromosomal architecture during development. It is important to note that enhancer proximity to the gene on the linear chromosome does not imply enhancer strength. Thus, the current rules to assess enhancer activity in vivo may be too locus-specific to be generalizable.

Taken together, genomic editing is required to parse out which CRE interactions (including super-enhancers) are functional. Plasmid based reporter assays or newer genome-wide approaches such as STARR-seq^[92] can be very useful in identifying DNA elements with the potential to be tissue-specific enhancers, but are insufficient to identify whether these elements function as enhancers in vivo.^[93] Finally, it is of critical importance that enhancer function is tested in different contexts.

10. The Nanog Locus – Enhancer-Transcribed RNAs and Their Function in Regulating Nanog Expression

The majority of DNA hypomethylated and H3K27Ac high enhancers (e.g., super-enhancers) are bound by the “paused” Serine5 phosphorylated RNAPII and bidirectionally transcribed to produce nuclear-restricted eRNAs (Figure 3).^{[58, 94–98]} This contrasts with long intergenic non-coding RNAs (lincRNAs) which are unidirectionally transcribed, spliced, polyadenylated, and can function in trans (Figure 3). LincRNAs, may be transcribed on the antisense strand of an enhancer but are marked by high levels of H3K4me3 at the promoter and H3K36me3.^{[96, 99]} While many RNAPII bound CREs are strong enhancers in vivo, the functional contribution of eRNAs in transcriptional regulation is widely debated. Pioneering studies on eRNA function revealed diverse roles in transcriptional regulation including stabilizing chromatin architecture through interactions with looping factors such as the cohesin complex and Mediator (Figure 1).^{[60, 100–103]} eRNAs have been shown to interact with the histone acetyltransferase (HAT) RNA binding domain of CBP/p300, which in turn stimulates its HAT activity and corresponding changes in gene expression.^[63] In addition, eRNAs in neural progenitors were shown to act as a “sponge” for negative elongation factor (NELF-Figure 1), thereby permitting Serine2 phosphorylation of RNAPII (elongating RNAPII) by positive transcription elongation factor (pTEFb).^[104]

Figure 3.

eRNAs and lincRNAs are epigenetically and functionally distinct classes of long non-coding RNAs. eRNAs are bidirectionally transcribed at enhancers by the paused form of RNAPII (Ser5P), not spliced, and function in cis at focal regions of the nucleus. lincRNAs are unidirectionally transcribed by the elongating form of RNAPII (Ser2P), spliced, polyadenylated, and function in trans in the nucleus and cytoplasm. lincRNAs share similar histone marks as protein coding loci, namely H3K4me3 at their promoters and H3K36me3 in their gene bodies.

Our group used antisense oligonucleotides to deplete eRNAs produced at the −45 Nanog enhancer.^[43] Interestingly, in contrast to enhancer deletion (which resulted in reduced expression of both Nanog and Dppa3), −45 eRNA depletion specifically reduced Dppa3 expression secondary to decreased interaction of the −45 enhancer and Dppa3 promoter. Thus, eRNAs are required for proper function of the −45 enhancer. An earlier study by Sigova et al.^[105] investigated roles of the RNA binding TF Yin-Yang 1 (YY1). Addition of a transcription elongation inhibitor resulted in decreased RNA and YY1 levels at enhancers and promoters (particularly at super-enhancers) suggesting eRNAs can stabilize TF occupancy. Inhibition of the exosome complex (which degrades RNAPII released eRNA transcripts) resulted in decreased binding of YY1, secondary to dispersal of eRNAs away from the CRE. Finally, CRISPR-mediated tethering of RNAs at enhancers resulted in increased YY1 binding. Thus, transcription at CREs results in positive feedback loops that mediate small but significant changes transcriptional output. TF trapping at transcribed enhancers (which represent less than 5% of all enhancers in ESCs) may help explain why TFs bind to only a fraction of their consensus DNA motifs. Further, subtle changes in levels of cis acting eRNAs may contribute to fine-tuning of TF expression during development. This is biologically important as eRNAs may have functional roles in not just enhancer activity, but also specificity for an enhancer with a subset of nearby genes. Finally, given oncogenic activity of super-enhancers, eRNAs are sequence and tissue specific targets which could be harnessed as therapeutic targets.^[106]

11. Conclusions and Outlook

The molecular mechanisms underlying regulatory control of enhancers at Nanog provides numerous opportunities to address issues ranging from how enhancer(s) operate and how they assist in controlling cell fate decision. Future studies at the single cell level in genomically edited cells will provide greater insight into the mechanism which underlie Nanog heterogeneity, but also which CREs are required. Key questions which the Nanog locus provides a fertile ground to study: 1) Do multiple enhancers function additively or synergistically to regulate transcription of a single gene? 2) Can the interactions of multiple enhancers with a single gene explain changes in cell-to-cell variability in the expression of a single gene (Figure 4). 3) Does a single enhancer coordinately co-activate or compete for neighboring genes? 4) Do insulated neighborhoods mediate precise enhancer-promoter interactions within a cluster of numerous co-regulated genes? 5) Do eRNAs have roles in fine-tuning transcriptional regulation by increasing frequency of enhancer contacts? Collectively, these studies will help determine the molecular basis of CREs in controlling heterogeneous gene expression. Moreover, a better understanding of transcriptional regulation of ESC TFs will increase our understanding of dysregulation that results in pathological states of less tractable tissues. It is well established that large chromosomal rearrangements (translocations and inversions) contribute to oncogenesis. Several studies have shown that most disease associated non-coding variation occurs near enhancers, particularly at the CTCF DNA-binding motif.^[107–111] Thus, increasing our knowledge of sub-TAD enhancer-gene specificity and how numerous enhancers function to fine-tune expression is essential to better comprehend disease pathogenesis. Finally, understanding how higher order chromosome structure, TF feedback loops, and enhancer elements regulate Nanog likely parallels critical TF regulatory networks in other stem cell models.

Figure 4.

A possible model of enhancer looping at a heterogeneously transcribed embryonic stem cell transcription factor in serum/LIF and 2i/LIF conditions. A) In serum/LIF, there is dynamic expression of an ESC TF such as Nanog over time (left). These fluctuations are secondary to an intermittent, stochastic interaction between a single enhancer (red) with the promoter which promotes a low level of transcriptional bursting (right). B) In 2i/LIF condition, the ESC TF is instead expressed more consistently over time (left), due to the activation of a novel, 2i/LIF specific enhancer (right, purple). The presence of two enhancers maintains a higher level of transcriptional bursting, which results in more consistent mRNA levels. In this proposed model as illustrated, burst size (number of RNA molecules produced per enhancer-promoter contact) is similar between different enhancers. All enhancers are within an insulated chromosome neighborhood (do not cross CTCF insulated boundaries).

Abbreviations

3C

chromosome conformation capture

4C

circular chromosome conformation capture

ChIA-PET

chromatin interaction analysis by paired-end tag sequencing

ChIP-Seq

chromatin immunoprecipitation sequencing

CRE

cis-regulatory element

CRISPR

clustered regularly interspersed palindromic repeat

eRNA

enhancer RNA

ESCs

embryonic stem cells

Hi-C

Hi-throughput chromosome conformation capture

HS

hypersensitive

ICM

inner cell mass

iPSC

induced pluripotent stem cell

lincRNA

long intergenic non-coding RNA

lncRNA

long non-coding RNA

NOS

Nanog, Oct4, Sox2

RNAPII

RNA polymerase II

TAD

topology associated domain

TF

transcription factor

TSS

transcription start site.