3D Shortcuts to Gene Regulation

Summary of recent advances

Recent technologies have allowed high resolution genome-wide binding profiles of numerous transcription factor and other proteins. A widespread observation has emerged from studies in diverse mammalian systems: most binding events are located at great distances from gene promoters. It is becoming apparent that the traditional one-dimensional view of gene regulation via the proximal cis regulatory elements is over-simplified. True proximity and functional relevance can be revealed by studying the three-dimensional structure of the genome packaged inside the nucleus. Thus the spatial architecture of the genome has attracted a lot of interest and has intensified its significance in modern cell biology. Here we discuss current methods, concepts, and controversies in this rapidly evolving field.

Introduction

The 3.4 billion base pairs of the human genome are packaged in hierarchical structures in a cell nucleus of about 10 micron diameter. The different levels of chromatin compaction may impact numerous nuclear events such as transcription and replication. Moreover, regulation of these processes may be influenced by the three dimensional organization of the chromatin in the nuclear space. The spatial configuration of chromosomes results in inter- and intra-chromosomal interactions that bring distant genomic elements to close proximity. The nucleus is also highly compartmentalized with defined nuclear bodies like the nucleoli, nuclear speckles, PML bodies, and Cajal bodies. Here functional domains can be formed by enriched foci or compartmentalization of nuclear proteins. Despite these non-random spatial features, it is still not clear what the principles of nuclear organization are and how they are related to function.

In the past, nuclear architecture could be studied mostly by DNA or RNA fluorescence in situ hybridization (FISH) that determines the location of a genomic locus relatively to another locus or a nuclear landmark such as the nuclear periphery. Recently developed molecular approaches, chromosome conformation capture (3C) and other closely related assays, utilize crosslinking, enzymatic digestion, ligation, and amplification procedures to probe the network of spatial interactions. 3C assesses the relative contact frequency between specific pairs of genomic elements of interest. High throughput applications [1–3**] allow increased coverage and quantitative information that are necessary to understand the long-range contacts between genomic elements and the whole genome in the native state. For example, the 3C on Chip (4C) technique [4] measures the interaction frequency of a genomic point of interest (bait) with the whole genome. Now it has become possible to map the entire interaction network (all-against-all) using a novel assay, Hi-C, albeit with a limited resolution [3**].

These recent genome-wide studies demonstrated that there exist at least two tiers of environments for any given locus. A genomic locus is in contact with adjacent elements, up to a few million bases along the linear DNA, at a frequency that is distinctively higher than that for more distant genomic loci on the same chromosome or on other chromosomes [3**–5] (Figure 1). The different levels of environment have varying biophysical properties, and may have differential impact on function. In this review we will discuss the nuclear architecture at these two orders of spatial organization. The first part will focus on the long range contacts between genes and distant cis regulatory elements, within a few Mb range. The second will address the organization of the nuclear space on the order of chromosome folding structures that result in intra- and inter-chromosomal contacts. We will also discuss recent findings regarding the features of long-range contacts, nuclear sub-environments, and their function in regulating nuclear processes in mammals. Particular attention will be given to current controversies in the field and how they may relate to technical variations in imaging methods. Finally, we will discuss how new molecular approaches could be used in parallel with previous methods to address important questions in this rapidly evolving arena.

Figure 1.

Microarray profile of long range interactions by 4C. (A) Long-range contacts of the bait across the same chromosome (top profile) and with loci on another chromosome (bottom profile). The peaks represent long-range contacts and their magnitudes indicate the interaction frequency. The dominant peak cluster centered at the bait reflects the high interaction frequency of the locus with the nearby sequences within a few million bp distance. Most of the reported distant regulatory elements are found within this region (panel B). (B) Finer-scale spatial interactions within the bait region. Regulatory sites (ellipses) are in long-range contacts (arrows) with other regulatory sites and genes (gray rectangle), generating a complex network of direct and indirect contacts. (C) The chromosomes are folded in the interphase nucleus into chromosome territories. Chromosomal loci are non-randomly engaged in long-range contacts with other distant loci from the same chromosomes and with loci from other chromosomes creating spatial micro-environments.

Long-range interactions by looping

A fundamental feature of mammalian genome organization is that regulatory elements such as insulators, enhancers, silencers, and locus control regions (LCRs) are predominantly located in great distances, up to hundreds of kilobases, away from the regulated genes. These elements, as for the well studied beta globin LCR, are in some cases critical for the correct spatial and temporal expression of target genes [6]. Mechanisms involved in long-range transcription regulation remain elusive, with little conclusive evidence to support either of the leading models: looping and spreading [7]. In the spreading model, a signal is initiated at an enhancer element and is transmitted along the chromatin fiber to the target gene promoter. In the looping model, the enhancer contacts the promoter through their associated proteins, with the intervening DNA looping out.

A breakthrough for probing in vivo interaction between distal cis regulatory elements came when the chromosome conformation capture (3C) approach was applied to mammalian systems [8]. This and later studies showed that the beta-globin enhancer and other regulatory elements reside in close proximity to the active beta-globin gene. The intrinsic ability of the beta-globin LCR to upregulate the expression of nearby genes was demonstrated by knocking the LCR into an unrelated housekeeping gene cluster. The LCR was found to induce the transcription of nearby genes. Interestingly, the LCR looped to the upregulated genes, suggesting that long-range contact between regulatory element and a gene can determine the gene transcriptional activity [9**]. In the past years, many studies reported long-range interactions between distal genomic elements, starting to address the functional relevance of these interactions to gene activity. Despite the recent focus on long-range chromatin interactions, the spreading mechanism may prevail in some cases, such as in the activation of the human epsilon-globin gene by the LCR [10].

Proteins that mediate looping

If looping of DNA elements is important, what maintains the loops together? Given the inherent flexibility of the chromatin fiber, some chance encounters are expected for a locus. For example, 3C studies show that a region of interest contacts its neighboring genomic sequences, separated up to hundreds of kb, regardless of any functional relevance. Therefore, a specific long-range contact is characterized by an interaction frequency much greater than this random expectation [11]. Such a loop is most likely mediated by proteins associated with the contact points. The identity of these proteins and their regulatory roles would shed light on potential functions of the loops that they maintain. Indeed many proteins were recently found to occupy the base of the loops, and fall into different functional categories such as transcription factors (TFs) [12–16], nuclear receptors (NRs) [17–19], insulators, Polycomb [20;21], chromatin remodelers, and architectural proteins (additional papers reviewed in [22]).

For the extensively studied beta-globin locus, EKLF, FOG-1, GATA-1, GATA-2 [23;24*;25], the insulator protein CTCF [26] and the architectural protein SATB1 [27] have been shown to be involved in the loop structure. CTCF, thought to be important for loop structures, is bound across the mammalian genome at cohesin binding sites, suggesting that cohesin also has a role in the long-range structures [28–31]. It has been suggested that the TF binding at a promoter and a distant enhancer reflects the specificity observed in such interactions [32].

In addition to transcription regulators, it has been proposed that the transcriptional machinery may mediate the loop structure between distant regulatory elements and the transcribed gene. Nascent transcripts and active RNA polymerase II (Pol II) were found to aggregate in discrete foci within the nucleus termed transcription factories. In this model, transcriptionally active Pol II influences genome organization [33] as genes would migrate to the immobilized factories to be transcribed. However, the chromatin contacts between the LCR and the beta-globin gene in erythroid cells is maintained when transcription initiation and/or elongation is inhibited and Pol II binding to these elements is decreased [5;34]. Therefore, unlike TFs, Pol II may not have a direct role in maintaining the chromatin loops. The above results do not rule out the possibility that Pol II may participate in establishing the loops.

Functional significance of loops

Numerous studies have investigated the functional correlates of the long range loops, especially with respect to gene expression. Polycomb group (PcG) proteins are epigenetic chromatin modifiers involved in gene repression over many cell generations. The human GATA-4 locus is organized in a multi-loop structure in undifferentiated embryonic carcinoma cells. The contact regions were found to be enriched for PcG proteins and DNA hypermethylation when the locus is repressed. After receiving differentiation signals, activation of GATA-4 transcription was accompanied by DNA hypomethylation and loss of PcG proteins. Knocking down the EZH2 Polycomb protein or reduction of DNA methylation relaxed the contacts and increased expression, demonstrating a high correlation between the PcG-mediated repressive loop structure and expression [35]. Similarly, EZH2 also mediates a 35 kb repressive loop between the tumor suppressor genes INK4b and INK4a [36]. In light of the decrease of interaction frequency with the GATA-4 locus, it will be interesting to look for newly formed contacts that are correlated with gene activity. A switch of the contact partners between the active and the inactive state was demonstrated in erythrocytes, where the Kit promoter transitions from interacting with the upstream enhancer to contacting a downstream element as its transcription is silenced during differentiation [24*]. Another study showed that GATA-1 and Ikaros repress the gamma-globin gene during erythrogenesis when erythrocytes switch globin expression from the gamma to the beta gene, concurrent with conformational changes in the globin locus [13].

To better understand how the long-range loops relate to gene activity, several groups have studied the effects of knocking down the relevant TF. In many cases there was a change in the locus topology together with inhibition of transcription. Unfortunately these findings do not address whether the loop is the cause or the consequence of the transcriptional process, because the approach does not uncouple the structural effects from those of the TF action on the promoter region. Nonetheless, evidence exists that the interaction with a distant regulatory element enhances the level of transcription. During erythroid differentiation, the beta globin gene is already expressed and harbors active chromatin marks, prior to long-range contact with the distant LCR and elevated transcription [37]. In addition, the deletion of the globin LCR in mice does not completely shut down beta globin and active chromatin marks still remain [38].

Loops and dynamics

An outstanding question has generated a great deal of interest and controversy: How dynamic are the chromatin loops? NRs are particularly suitable for this problem because gene regulation can be rapidly initiated and controlled by the ligand. In addition, long-range interaction between NR binding sites and target genes is probably important, because the majority of their genomic binding sites are located at great distances from genes, and some regulated genes do not have any binding sites nearby [39–41].

A recent high throughput method, ChIA-PET, has established extensive hormone-dependent chromatin loops between multiple estrogen receptor (ER) alpha binding sites and gene promoters [42*]. However, it will be informative to look beyond the interactions between NR binding sites. In a recent study, when interaction partners of a Glucocorticoid Receptor (GR)-induced promoter were examined, the promoter was found to loop out to the promoter of another GR-regulated gene. Surprisingly, the authors did not find a GR binding site at the other side of the loop by ChIP. Transient and very short-lived protein binding events may not be preserved by the widely used formaldehyde crosslinking [43]. In addition, the long range interaction was set prior to GR activation [17]. It is possible that different NRs act via different dynamic mechanisms, but clearly more studies are required to address this important issue of long range interaction dynamics.

The controversy has intensified by a number of reports on various dynamic aspects of NR actions. Gene activation can be cyclic for some NRs such as ERα and vitamin D receptor (VDR) [44–46]. The ligand can be released in vivo in an hourly cycle, giving rise to transcriptional pulses [47]. As for long range interactions, transient loops between ER binding sites have been reported [19;48] as well as VDR binding sites interacting in a cyclic manner which correlates with gene transcription, chromatin modification, and Pol II activity [49**]. These data do not necessarily imply that the two ends of the loop are brought together by an active mechanism. Rather, dynamic formation of a loop may arise from stabilization of random collisions that exist in the cell population which results in a higher contact frequency.

Further studies for loops

Recent studies have revealed a more complex relationship between regulatory elements than previously thought [50;51]. Genomic regulatory sites are often identified as DNaseI hypersensitive sites (DHSs), regardless of the proteins occupying them. Indeed, multiple DHSs form contacts with the globin LCR in a developmental stage-dependent manner [8;52]. In light of the observed topological complexity, it will be worthwhile to identify all DHSs [53] and rigorously examine their contacts, rather than focusing on a few specific protein binding sites. DHSs could serve as useful reference points for 3C or the high throughput Chromosome Conformation Capture Carbon Copy (5C) technology [1;2], and will generate valuable insight on the nature of various loop structures.

Chromosome territories

In addition to the non-random and transcriptionally relevant chromatin folding in the few million base range outlined in previous sections, a locus will non-randomly associate with more distant loci on the same chromosome or on other chromosomes at lower frequency (Figure 1). A significant component of the non-random organization of the nucleus involves the heterogeneous spatial distribution of chromosomes [54;55]. Imaging-based studies of nuclear architecture have determined that individual chromosomes are not scattered over the whole nuclear space but are restrained in discrete regions termed chromosome territories (CT) [56]. CTs do not have physical barriers, and the chromatin fibers of each CT intermingle with those of neighboring chromosomes [57] thereby increasing the complexity of potential spatial neighbors for a given locus. In spherical nuclei, CTs have preferred radial positions, in that gene-rich chromosomes tend to be internally located and gene-poor chromosomes are closer to the nuclear periphery [58]. Aside from this general feature, loci organization within a CT [59] and the organization of CTs relative to each other are cell type-specific [60], suggestive of functional relevance. A recent cutting-edge work has brought new insights on the biophysical principles of CT folding. The authors developed a new technology, Hi-C, which is a high-throughput version of 3C that allows the capture of large scale chromatin interactions of the entire genome. Remarkably, the chromatin conformation of the human genome in the megabase scale was consistent with a model based on fractal globular polymers. The fractal globules enable dense packing, and are knot-free, allowing the flexibility to easily fold and unfold chromosomes during nuclear processes such as gene expression and cell cycle [3**].

Inter-chromosomal associations: Functional specificity?

Several studies have presented evidence for the intriguing phenomenon of non-random association between functionally related genes that are located very far or on different chromosomes. Regarding the extensively characterized globin genes in erythroid cells, imaging studies have shown that beta-globin co-localizes with other distant erythroid-specific genes [61–63]. However, frequent co-localization of active erythroid-specific genes does not establish the existence of functionally specialized nuclear environments. Some groups propose that the interactions occur at sites enriched with active Pol II termed transcription factories [33], while others find the associations to co-localize with nuclear splicing bodies [63].

Overcoming the limited scope in interaction partners that are probed by imaging methods, two recent studies characterized the spatial neighborhood of the beta-globin locus by genome-wide 4C approaches, with somewhat conflicting results. Simonis et al. found that the beta-globin contact loci, although interacting with many active genes, were not enriched for erythroid-specific genes [4]. On the other hand, Schoenfelder et al. probed Pol II associated beta- and alpha-globin environments and reported that the contact loci are enriched for other active and erythroid-specific genes [64*]. The authors also observed that erythroid-specific TF Klf1 is enriched in a subset of transcription factories, and mediates preferential associations of Klf1-regulated genes, suggesting that regulatory factors create specialized nuclear environments. Although they focused solely on actively transcribed globins, this does not explain the discrepant results from the two studies because the globin genes are highly transcribed and engaged with factories in about 90% of these cells [61]. Significant differences might have come from the very different approaches employed for statistical data analysis (W. de Laat, personal communication), as the former study found predominantly intra-chromosomal contacts of beta-globin while the latter reported mostly inter-chromosomal interactions.

Interestingly, inter-chromosomal co-localization of inactive genes has also been reported. The imprinting control region (ICR) of the maternally expressed H19 gene was found to engage in allele-specific inter-chromosomal interactions with other imprinted genes [65;66]. A recent 4C study showed that imprinted loci across the whole genome frequently associate each other and that the CTCF binding site within the ICR has a global effect on the spatial proximity between imprinted loci and their epigenetic state [67*]. However Sandhu et al. did not account for the gene density or the transcriptional status of imprinting-independent genes in the contacts. Therefore, it is difficult to determine whether these hubs regulate imprinted genes or are a consequence of more general features of nuclear architecture like gene density and activity.

Some studies reported remarkable findings on the functional and dynamic specificity of large-scale inter-chromosomal interactions in response to external stimuli. Co-localization of the IFN-beta gene with three distant NF-kappaB bound loci was shown to be dynamic and concomitant with transcriptional activation by a viral infection [68]. In another study, ER induced genes were found to co-localize frequently after estradiol treatment [69]. Despite these interesting observations, it remains critically important to investigate whether such dynamic phenomena are a general mechanism of gene regulation using comprehensive approaches.

Nuclear ‘hubs’ that can be defined by distinct factors and genomic elements, like the nucleoli, are an appealing concept for coordinated regulation of specific genes. However, a clearer picture will emerge from probing the whole environment of the loci of interest. Another point to consider, as we expand our understanding, involves the nature of the cell types under study. For example, it is unclear how much of the extensive knowledge about erythroid cells, a highly specialized cell type dedicated to globin production, translates into other cell types of physiological importance. For a cell type that needs to be versatile in response to various stimuli, the nuclear organization may be more complicated.

Nuclear bodies and landmarks for genome organization

Nuclear bodies, such as nucleoli, PML, and Cajal bodies, are dynamic environments enriched for specific protein factors and genomic loci. Complete mapping of genomic regions in proximity to these bodies could be useful in building three dimensional maps of nuclear architecture. A disadvantage in this approach is that the number of nuclear bodies is highly variable, depending on the cell type, transcriptional activity, the cell cycle stage, and also varies in the cell population. Moreover, the association of the genome loci with the nuclear bodies may vary.

The nuclear envelope is also a well-defined nuclear landmark with associated genomic regions. Although nuclear lamina associated loci are generally thought to be gene-poor and repressed [70] several different approaches, including gene tethering to the periphery, have revealed a more complex relationship [71–74]. These studies underscore the fact that the nuclear periphery is not an exclusively silent environment.

The nuclear envelope has also served in recent years as a convenient reference for mapping the nuclear architecture in a more quantitative fashion [75]. In addition to mapping individual chromosomes along the center-periphery axis, attempts have been made to correlate radial positioning to gene expression. However, the positioning in the radial axis is poorly associated with gene activity [76]. Another study examined the radial position of 11 cancer-related genes during early mammary tumor formation and development [77]. Although each gene had a preferred radial position, alterations of positioning patterns during differentiation and tumorigenesis were unrelated to transcription.

Taken together, it seems that there is no clear correlation between gene activity and the distance from the nuclear periphery. This is in agreement with the fact that sites of active transcription are evenly distributed throughout the nucleus with no preference towards the nuclear center [78]. In addition, there may be other repressive environments in the nucleus. Therefore, although the radial position of a gene relative to the periphery serves as a convenient reference, it is not clearly linked to the functional status [75].

‘C’ assays and DNA FISH: Which is validating what?

Since the early studies that introduced 3C and 4C methodologies, DNA FISH has been used to ‘validate’ a selected sample of interacting pairs to see if they are indeed in close physical proximity in the nucleus. Here we discuss some inherent differences between the two assays and how they affect the apparent degree of correlation.

3C-based assays obtain information from DNA-DNA interactions that are summed over millions of cells. At any given moment, certain contacts may be abundant and exist in the majority of cells, while other interactions may be transient or present in a small subpopulation of cells. Although direct translation of the 4C/5C signals to the contact frequency scale is very difficult, it is generally thought that high and low signals roughly correspond to frequent and rare interactions, respectively. The dramatic shift of the signal strength within versus outside of the bait neighborhood in 4C data exemplifies the wide spectrum of contact frequency.

DNA FISH produces information at the level of single cells with the spatial resolution dictated by the optical limit and the probe size. In a widely used protocol, each locus is represented by BAC probes that typically span a few hundred kilobases and therefore may cover a larger region than the relevant contact locus. Also, the probe-annealing step requires denaturing the nuclear DNA in a high temperature which may perturb the native positions of genomic loci [79]. Although the contact frequency is directly estimated by the fraction of cells containing co-localizing loci, the assay is more prone to sampling error because the number of nuclei that are examined ranges in the hundreds at best.

Aside from these general features, it is noteworthy that DNA FISH assays are implemented in various ways. These include the three methods that we describe here. In the first method, one simply collects images from several focal planes through the nuclei and scores two loci as co-localizing if their FISH signals overlap in the same z plane [80]. In the second method, cryoFISH is performed and a thin physical section through the nuclei is examined to count those that have co-localizing FISH signals [4;57]. In these two methods, one obtains an estimate of the contact frequency for the pair of interest by counting the cells that have ‘co-localizing’ FISH signals. Typically the percent of nuclei showing co-localization is reported for each pair of loci. Co-localization of a pair of loci is determined based on an overlap between the two FISH signals, which is influenced by the particular image acquisition and thresholding of the intensities for the fluorescence channels. In an alternative method (Hakim, Sung, Hager, unpublished data), all the focal planes from DNA FISH images are utilized to calculate the three dimensional position of each signal. Effects from imaging and thresholding can be minimized if the position of a FISH signal is determined by its brightest pixel (the center) rather than its edges. Then the center-to-center distance in 3D is calculated in an automated fashion for a large number of cells. Unlike the previous methods, one obtains the information on the distribution of 3D distances, and can also report the contact frequency by introducing a distance cut-off for co-localization. Here it has to be taken into account that even overlapping loci may well have a significant (center-to-center) distance between them. Interestingly, it has been observed that the distance distribution itself has some correlation with interaction measurements from 4C or 5C [Job Dekker, personal communication; Hakim, Sung, Hager, unpublished]. It seems possible that spatially proximal loci represent those that contact each other in a dynamic fashion.

Because the apparent ‘validation’ rate is directly affected by the limited spatial resolution, small sample size, and variable quantification methods for DNA FISH, it can be difficult to establish a high concordance with 3C-based assays. Employment of objective automated quantification method will not only increase the number of cells examined but also ensure a standardized protocol for treating and reporting DNA FISH data accurately.

Challenges and new opportunities

Imaging approaches like FISH and immunofluorescence, that have been instrumental in making interesting observations, are nevertheless descriptive and limited in the number of loci that can be studied simultaneously. They do not provide comprehensive information about the spatial environment for a given locus. Similarly, although nuclear bodies are known to associate with particular loci in certain conditions, imaging results do not convey the full spectrum of genomic contacts for these nuclear bodies.

Current high-throughput technologies allow higher-resolution quantitative studies of DNA-DNA interactions across the genome. Many variations of the 3C approach and 3C combined with protein interactions are now available [1;42*;64*]. The 4C (3C on chip, circular 3C) approaches identify all the genomic regions in contact with a locus of interest, while the 5C assesses all the interactions spanning a large locus of interest. Although rather high-resolution, 5C analysis is limited to a confined genomic region and does not cover the entire genome. A major advance in genomic technologies for studying nuclear architecture has recently been achieved by the Hi-C method [3**]. It is based on selective purification of spatially proximal ligation products followed by massively parallel sequencing, which reveals chromatin interactions across the whole genome. The assay resolution is currently about 1Mb, and may potentially approach that of 4C, in the range of 10 kb, as the Hi-C library is sequenced much deeper. The ChIA-PET approach is a powerful combination of ChIP and 3C, developed for long range interactions mediated by a protein of interest [42*].

The different versions of the high-throughput methods provide data on different aspects of nuclear organization, and are therefore complementary. For example, ChIA-PET provides information for interactions between genomic elements that are in contact with a specific protein, but does not seem to capture inter-chromosomal interactions efficiently. 4C and Hi-C detect long-range interactions regardless of whether they bind to a specific protein.

Conclusion

Recent investigations of the functional roles of nuclear organization have led to many interesting discoveries as well as new questions. While imaging methods like DNA FISH have been extremely useful, great advances have been made for high resolution molecular assays that dramatically improve quantification of genome-wide interaction patterns. Combining the genome-wide chromatin configuration methodologies with existing genomic tools will be instrumental for obtaining comprehensive insights on principles of genome organization. In parallel, it will be necessary to standardize and increase the throughput of imaging approaches for a more accurate picture at the single cell level. In addition to FISH and other fixed cell assays, live cell imaging approaches will be essential in addressing dynamic aspects of chromatin interactions. Lastly, correct interpretation of data remains an important issue in this fast-evolving, technology-driven field. Sound statistical analyses and computational modeling can facilitate almost all aspects of the relevant techniques, not only by minimizing sources of discrepancies but also by uncovering novel insights. Interdisciplinary collaborations including researchers with mathematical and biophysical expertise will significantly enhance many studies in the future. Armed with these exciting new approaches, the field has high hopes for witnessing fruitful contributions toward a clear relationship between nuclear architecture and function.