Spatial Organization of Epigenomes

Abstract

The role of genome architecture in transcription regulation has become the focus of an increasing number of studies over the past decade. Chromatin organization can have a significant impact on gene expression by promoting or restricting the physical proximity between regulatory DNA elements. Given that any change in chromatin state has the potential to alter DNA folding and the proximity between control elements, the spatial organization of chromatin is inherently linked to its molecular composition. In this review, we explore how modulators of chromatin state and organization might keep gene expression in check. We discuss recent findings and present some of the less well-studied aspects of spatial genome organization such as chromatin dynamics and regulation by non-coding RNAs.

Introduction

Defining Spatial Chromatin Organization

Over the past decade, the emerging regulatory role of spatial chromatin organization has received much attention, particularly with respect to gene expression [1–4]. The folding of chromosomes and their position in the nuclear space have been studied extensively using a variety of light microscopy techniques [5–7], and more recently in combination with the chromosome conformation capture (3C) technologies [4, 8]. These techniques remain complementary since visual methods like fluorescence in situ hybridization (FISH) can examine chromatin architecture at the single cell level but are limited to only a few regions, whereas several 3C technologies can assess spatial organization of entire genomes although across cell populations. However, the perspective gap between the visual and molecular approaches appears to be narrowing with advancements in microscopy [9, 10] and probe synthesis that allow visualization of many regions simultaneously [11], and with the realization of a first genome-wide 3C-type analysis conducted at the level of single cells [12, 13]. Even when their limits in scale and scope finally meet, these approaches should always be paired together as some of the data produced by FISH and 3C methods differ [14, 15]. Detailed descriptions of these techniques and their limitations – including how the vast amounts of data they produce can be handled – are described in other reviews [8, 16–19].

Although exactly how and why chromosomes fold in the nucleus remains largely unknown, several key concepts are now well accepted. A number of recent reviews have focused on the spatial organization of mammalian genomes [4, 13, 20] and thus it will only be discussed in general terms here. For instance, rather than having their chromatin randomly arranged in the nuclear space, chromosomes are known to occupy distinct “chromosome territories” (CTs) reflecting preferential interactions between genomic regions located on the same DNA molecule (cis) as opposed to different chromosomes (trans). In addition, CTs themselves are not arbitrarily positioned in the nucleus. Instead, they occupy preferred radial positions from the nuclear center to the periphery with gene-rich chromosomes clustering at the center of the nucleus, while gene-poor CTs tend to reside near the lamina coating the inside of the nuclear envelope [20].

CT analysis with “Hi-C”, a variant of the original 3C technique that captures chromatin organization genome-wide, reveals the existence of two types of “compartments” designated “A” and “B” [21]. Compartments represent multi-megabase genomic regions within chromosomes that preferentially interact with each other. “A” compartments are more transcriptionally active and have a higher GC content than “B” compartments. They are also enriched in open chromatin, genes, and active/poised chromatin marks. “B” compartments are less transcriptionally active, display a stronger propensity for self-association and are enriched in the repressive H3K9me3 histone mark. Whether compartments exert specific functions or simply reflect the underlying chromatin activity remains unknown. Nonetheless, it is interesting to observe that “B” compartments along chromosomes highly correlate with late replication timing and the position of lamina-associated domains (LADs). These correlations suggest that “B” compartments might preferentially localize at the nuclear periphery in vivo [22].

At a higher resolution within the compartments are found the “topologically associating domains” (TADs), megabase scale chromatin regions that interact more frequently with themselves than with any other region in the genome [23, 24]. TADs or TAD-like chromatin domains are seen in different organisms including human, mouse, fly, and yeast [23–26]. Interestingly, many TADs appear conserved across cell types and even between species at syntenic regions. In contrast, the many types of substructures found within TADs (sub-TADs) are much more cell-type specific. These reflect different types of long-range interactions between regulatory DNA elements including promoters, enhancers and insulators.

Though we are just starting to map chromatin organization across entire genomes [3, 21, 23–25, 27, 28], it is clear that gene regulation is affected by chromatin architecture [5, 29–35]. The hierarchical model of functional chromatin organization emerging from these studies makes intuitive sense considering all that is involved in transcription and that control elements must first find the gene they regulate in three-dimensional (3D) space at specific times and under various conditions. Thus, TADs are thought to spatially compartmentalize the linear genome into regulatory units of chromatin where, for example, genes can be regulated independently from their neighbours. This hypothesis is supported by the observation that gene expression patterns within TADs tend to correlate during differentiation [24], and that the activity of a regulatory sensor inserted at several hundred different sites consistently spans large regions correlating with the size and position of TADs [36].

The idea that TADs partly represent spatial regulatory units of transcription is also supported by the observation of enhancer-promoter contacts mostly within TADs and that displacement of such boundaries may lead to misexpression of genes [24, 37]. Furthermore, removal of TAD boundaries can lead to spreading of activating histone marks and transcription of genes in the neighbouring TAD [38]. Although these findings support the compartmentalization of genes and regulatory elements into TADs, they do not exclude the possibility of regulatory contacts across TADs. Indeed, the existence of transcription factories suggests some level of interactions between chromatin from different TADs. Transcription factories are specialized nuclear sites where high transcriptional levels are detected. They are thought to represent another level of genome compartmentalization driven by co-regulated genes that permit rapid activation of associated genes regardless of their genomic distance [39–41]. Whether or not transcription factories affect the organization of TADs or their position in the nuclear space is unknown.

Chromatin Organization in Single Cells

Microscopy-based studies clearly demonstrate the inherent dynamic nature of chromatin organization in vivo, even in the nucleus of genotypically and phenotypically identical cell types [42–44]. Given that 3C-based methods yield spatial chromosome maps averaged from millions of cells, new methods are needed to answer questions for which averages no longer suffice. One study pioneered this venture by modifying the Hi-C protocol for contacts within individual CD4⁺ T_H1 cells [12]. As with the conventional Hi-C method, the existence of CTs and the presence of TADs are observed with single-cell Hi-C. At the resolution considered, individual chromosomes appear to maintain TAD organization from cell to cell, but contact patterns vary significantly at larger scales and between domains. This single-cell analysis also recapitulates the presence of active domains at CT boundaries. On the other hand, contacts between chromosomes appear at highly distinct regions instead of more uniformly distributed as with the original Hi-C approach, although whether this difference stems from technical biases such as PCR amplification is unclear.

While single-cell analysis methods will be indispensable to fully understand chromosome architecture and its relationship with genomic activities like transcription, developing new sufficiently robust approaches may take some time. Rather than conducting experiments in individual cells followed by genome-wide statistical analyses, alternative approaches such as combinatorial cellular indexing might be immediately useful. Combinatorial cellular indexing is a technique involving sequential barcodings and dilutions whereby nuclei can be effectively “compartmentalized” by the barcodes they receive [45]. The power of this technique lies in the fact that single-cell resolution data can be obtained from within cell populations without having to process cells individually. Applying this method to 3C-type datasets might eventually become feasible enough to circumvent the need for novel ultra-specialized methodologies, as is seen for chromatin states in singles cells [45].

How Dynamic is Chromatin Organization?

Most current techniques thus far can only provide insight into the complexity of chromosome organization at given times without directly informing on the dynamic component. To work around this, one team has compared genome-wide chromatin interactions across multiple cell lineages [46]. By contrasting chromatin organization in embryonic stem (ES) cells and ES-derived lineages with their respective gene expression and chromatin state, they would uncover cell type-specific changes in spatial organization. Assessing changes in active and inactive states during differentiation may provide a possible mechanism for lineage specification [46]. They show that while around 36% of the genome can switch states during differentiation, these changes tend to be restricted to within TADs [46]. The intra-domain interactions of TADs appear to determine whether a domain is active or inactive, with frequency related to gene expression [46]. H3K4me1 density appears to be an important predictor of these intra-domain interactions [46] – an interesting result given that H3K4me1 is often found at poised and/or active enhancers [47, 48], and considering the fact that enhancers exists at many looping sites [49]. Since enhancer density correlates with interaction frequency and H3K4me1 associates with enhancers, H3K4me1 is thought to help determine interactions during lineage specification through enhancer looping [46].

An important drawback to comparing cell lineages to infer how chromatin organization contributes to gene expression changes is the difficulty in assessing the actual contribution of architecture to transcription. This relationship can be more clearly examined when systematically tracked over short differentiation time-courses as we previously demonstrated [50]. Alternatively, we might benefit from models that can at least help us predict chromosome structure and topology. In the past, we have seen the success of energy models in determining protein structure [51–53]. However, unlike most folded proteins, which tend to prefer their equilibrium state [54, 55], DNA is large enough to prevent itself from reaching its low energy state [56]. Further complications include the numerous proteins and DNA elements, which determine contacts as well as variations between cell types and cell cycle phases [56].

To get around this, one study proposes an entropy approach with interesting results [56]. They show that the relaxed interphase chromatin landscape could be simulated, complete with contact probabilities. They also recapitulate topology and indicate that TADs minimize knots of chromatin and might allow discrete unfolding to occur. Although further modeling of the different stages of the cell cycle will be possible as contact maps become available, a limitation is that this approach only presents ideal conformations. Nevertheless, a model that mimics how dynamic structure/conformation relates to gene function will help us better understand gene regulation [56].

How Modulators of Chromatin State and Organization Might Keep Gene Expression in Check

Overall genome architecture is guided by chromatin interactions with nuclear landmarks like the lamina as well as by intra- and inter-chromosomal contacts mediated by chromatin-binding proteins (Figure 1). As such, chromosome organization is inherently linked to chromatin state, and any change to its composition has the potential to alter folding and the proximity between regulatory DNA elements. Chromatin composition can be altered through various pathways, including the more stable epigenetic types like DNA and histone modifications. Epigenetic regulation is central to all aspects of human life including development, health, and disease. These mechanisms modify gene expression and can have a significant impact on genome architecture. By mutually affecting each other, chromatin state and architecture may take part in a self-enforcing feedback process to propagate cell-fate memory [57].

Figure 1. Genome architecture is guided by contacts with nuclear landmarks, and by intra- and inter-chromosomal interactions mediated by chromatin-binding proteins.

(A) Schematic representation of a mammalian nucleus illustrating how chromosome territories (CTs) can relate to each other and with respect to the nuclear lamina (brown dashed line). Darker areas within CTs represent heterochromatic regions near the lamina. The nucleolus is highlighted in white. The lines within CTs represent contacts within (intra) and between (inter) chromosomes. Arrows point to inter-chromosomal contacts. (B) Lamina-associating domain (LAD) showing heterochromatin clustered near the nuclear lamins. (C) Close-up indicating silencing marks (H3K9me3) with CTCF at the borders of LADs. (D) Topologically associating domain (TAD) showing regions defined by CTCF boundary elements at the megabase level. (E) Close-up demonstrating sub-TADs within TADs can be differentially regulated. Chromatin can either be repressed (H3K27me3) or actively transcribed (H3K36me3). H3K4 methylation is seen at active promoters (H3K4me3) and enhancers (H3K4me1).

The importance of spatial genome organization in human disease is clearly demonstrated in cancers where single nucleotide polymorphisms (SNPs) are found to create novel enhancers acting long-range to activate distal genes through intra-chromosomal contacts and chromatin looping [58–60]. Improper binding to nuclear landmarks is also linked to human disease. For example, studies on laminopathies like Emery-Dreifuss muscular dystrophy demonstrate that this disease can originate from mutations in a lamin protein, which specifically causes abnormal retention and silencing of muscle-specific genes at the nuclear envelope [61]. A particularly interesting group of proteins likely to contribute to these changes are the polycomb group (PcG) proteins. PcG proteins play an essential role during differentiation by repressing key developmental genes until their appropriate time of expression. Importantly, they are modulators of stem cell differentiation that can be deregulated in cancers.

PcG proteins are part of two distinct complexes called PRC1 and PRC2. These complexes repress transcription and have distinct enzymatic activities. While PRC1 catalyzes the mono-ubiquitination of H2A on lysine 119 (H2AK119Ub1), PRC2 catalyzes the di- and tri-methylation of H3 on lysine 27 (H3K27me2/3) [62–64]. Genetic and biochemical studies show that PRCs control transcription in pluripotent cells by catalytic and non-catalytic mechanisms [65, 66]. They are shown to induce chromatin compaction [65–67], but can also form long-range contacts within and between chromosomes in vivo, supporting their potential role in regulating gene expression through genome architecture. For example, deletion of a PRC2 subunit that leads to H3K27me2/3 depletion correlates with chromatin decompaction at HOX loci as detected by FISH [66]. H2A ubiquitinylation is also lost under these conditions since PRC1 recruitment requires PRC2 [68]. Interestingly, chromatin decompaction also occurs in cells where the PRC1 Ring1B ubiquitinase subunit is deleted, even though H3K27me2/3 levels are maintained [66]. Since complementation of Ring1B−/− cells with either the wt protein or a catalytically dead mutant (I53A) restore compaction at HOX loci by FISH [66], Ring1B can likely mediate long-range contacts in a non-catalytic manner.

As indicated by the example of muscle-specific gene silencing described above, directing chromatin regions to the nuclear lamina can have a significant impact on gene expression. Genomic regions naturally targeted to the nuclear lamina are termed LADs and the genes contained within them are mostly repressed [69–71]. LADs are enriched in the H3K9me2/3 silencing marks, while their borders also contain H3K27me3 and CTCF binding sites [35, 71–73]. CTCF (CCCTC-binding factor) is viewed as a key factor in genome organization [1]. It can act as an insulator/barrier at heterochromatin boundaries and divide chromatin into silent and active domains. It is enriched at TAD boundaries and can also mediate long-range chromatin interactions such as those observed during enhancer-promoter looping.

Genomic regions associated with the nuclear lamina can be regulated as shown for developmentally regulated variable LADs (vLADs). vLAD borders appear to be critical for their dynamic association with the lamina, and are enriched for YY1 binding sites [74]. That YY1 plays a role in this process agrees with previous studies demonstrating the recruitment of PRC2 by YY1 [75–82]. Accordingly, chromatin repositioning to the nuclear lamina is dependent on YY1 and H3K27me3 enrichment at LADs, in addition to H3K9me2/3 [74, 83–85].

Recently, we have reported a link between CTCF-CTCF interactions, PcG silencing, and 3D chromatin organization [50]. In human and mouse cells, the HOXA cluster is divided between two TADs with 3′ genes residing into one TAD and those at the 5′ end extending into the other. Like most TAD boundaries, the HOXA middle region is enriched in sites bound by CTCF [50]. We show that silencing by PcG is associated with the reconfiguration of CTCF-CTCF contacts at the HOXA TAD boundary, which induces a kink upon differentiation. Our results support a model whereby regulated looping between CTCF-bound sites at TAD boundaries might control the folding path of chromatin, the proximity between TADs, the formation of contacts between them, and thereby the topology of genomes.

Non-coding RNAs as Modulators of Chromatin State and Organization

Another mechanism by which PcG proteins regulate genome architecture is through long non-coding RNAs (lncRNAs) [86]. Over the last decade, advances in high-throughput sequencing technology have fuelled interest in mapping the transcriptome of eukaryotic cells. These studies have revealed thousands of non-coding RNA throughout mouse and human genomes [87, 88], which are now classified based primarily on their length and origin of transcription. Long non-coding RNAs represent the broadest class of these molecules described to be at least 200 nts in length with little to no protein-coding potential [89]. Additional categories such as long intergenic non-coding RNAs (lincRNAs) and enhancer RNAs (eRNAs) are more descriptive nomenclatures based on the site of origin.

Despite being largely transcribed by RNA polymerase II and having post-transcriptional processing similar to mRNAs, lncRNAs appear poorly conserved between species [90]. This observation led to early skepticism as to whether or not lncRNAs possess any relevant function in the cell. Numerous studies of individual lncRNAs now show that they play important roles in regulating gene expression. These transcripts adopt a variety of mechanisms by which they can affect their target genes. Most notably, they associate with histone modifying complexes and direct changes to the chromatin state of their targets. For example, the lncRNA transcripts Air and kcnq1ot1 play important roles in gene imprinting by recruiting PRC2 and G9a complexes to their target regions on the paternal allele for silencing by methylating H3K27 and H3K9, respectively [91, 92]. These lncRNAs act in cis, contrary to the lncRNA HOTAIR which regulates gene expression across different chromosomes (trans) [93]. Transcribed from the HOXC locus, HOTAIR recruits both PRC2 and the histone H3K4me2 demethylase LSD1 to supress gene expression at the HOXD locus [93–95]. By serving as a scaffold for different histone modifying complexes, HOTAIR shows that lncRNAs can direct a variety of changes to the chromatin state of their target genes.

A role for lncRNA in mediating trans-chromosomal interactions is seen with the lncRNA FIRRE. RNA antisense purification (RAP) and FISH experiments show FIRRE localizing near its site of transcription and bridging both cis and trans chromosomal interactions to that region [96]. Interestingly, the maintenance of these inter-chromosomal interactions requires both the lncRNA and its associated protein component hnRNPU. This is particularly interesting as hnRNPU is essential for localizing XIST to an X chromosome for silencing [97].

While numerous lncRNAs interact with histone modifying complexes to regulate gene expression, much less is known about their role in regulating 3D chromatin architecture. A recent study shows that the lncRNA XIST, responsible for X-chromosome inactivation in female mammalian cells, also plays a role in the chromatin organization of the inactive X chromosome (Xi) [98]. XIST is a 17-kb long lncRNA transcribed from the X chromosome, which recruits PRC2 for suppression [99]. Further investigations identified several additional proteins that interact with XIST such as the ATRX RNA helicase, YY1 and hnRNPU [97, 100, 101]. Each of these components plays an important role in maintaining XIST interaction with PRC2 or tethering the complex to the X chromosome itself. Recently, analysis of the XIST interactome revealed over 80 proteins associating with the lncRNA, including CTCF and members of the cohesin complex (SMC1A, SMC2 and RAD21), which are known to regulate 3D chromatin architecture [98]. Xi does not have any of the TADs present in the active X-chromosome (Xa) and is organized instead into megadomains. However, upon removal of XIST, the chromatin architecture of Xi reverts to that of Xa with cohesin demarking the boundaries of reformed TADs [98]. This suggests that XIST plays a direct role in shaping chromatin architecture by actively preventing the formation of TADs on Xi.

These findings open a new field in epigenomics for lncRNA-mediated chromatin organization. As these transcripts are often long, likely possessing highly-folded structures, they may serve as docking sites for a wide variety of proteins involved in nuclear architecture. This may permit the directed localization of these components and therefore could affect both intra- and inter-chromosomal interactions.

Conclusions and Perspectives

Microscopy-based studies and those using 3C-type technologies have altogether led to the current model of genome architecture wherein chromosomes are organized in hierarchical length scales [4]. From high to low resolution, genomic DNA wrapping around histone octamers forming nucleosomes represents the first level of genome organization in this model. By engaging in long-range cis contacts, chromatin fibers fold into substructures (sub-TADs), with topologies varying in a tissue-specific manner [27]. These conformations exist within larger megabase-sized chromatin domains, the TADs, which in contrast appear largely conserved between cell types and across species. TADs contained within compartments “A” (open/active) or “B” (closed/silent) interact together within their respective chromosome territories. The nuclear position of CTs is not random but rather defined at least in part by transcriptional activity and interactions with nuclear landmarks like the nucleoli and lamina through heterochromatic regions.

Although compelling, this model does not answer many pressing questions about genome organization, including why and how this folding occurs, the relationship between each folding state, and how changes at one level may impact the conformation and position of chromosomes within the nucleus. Also, despite the existence of many robust and complementary technologies, the field would nonetheless benefit highly from new approaches that capture chromatin organization along with transcriptional and epigenomic changes in real-time and preferably at the level of single cells – a tall order not likely to be reached anytime soon. Regardless of whether it will come through the development of new methodologies to probe chromatin organization or through computational/mathematical modeling of population-based datasets, understanding the actual relationship between chromatin architecture and state will be essential to appreciate the natural plasticity of chromatin and identify conformation patterns that lead to improper gene expression and disease. In this respect, chromatin organization is often altered in human diseases such as cancer [102], and we have recently provided evidence of its value as a discovery tool for candidate cancer biomarkers [103]. It will be interesting to see whether chromatin conformations actually represent ideal biomarkers and to define the value of such signatures in the clinic [104].