Molecular basis and biological function of variability in spatial genome organization

Abstract

BACKGROUND:

The eukaryotic genome is hierarchically organized in the cell nucleus into DNA loops, chromatin domains, compartments, and, ultimately, chromosomes. Many of the most prominent organizational features of genomes are highly reproducible on the population level and are evolutionarily conserved, suggesting functional relevance. Numerous organizational features, such as the location of a gene or promoter-enhancer loops, have been implicated in the regulation of genome processes including transcription, replication, and repair. At the same time, single-cell analysis has revealed extensive stochasticity of gene expression, with individual genes undergoing cycles of bursts in activity and periods of inactivity. Because gene activity is closely linked to features of genome organization, the extent of variability of genome organization at the single-cell level has arisen as a question of interest. We review new findings that document extensive cell- and allele-specific variability of genome organization and discuss potential mechanisms of structural variability and its implications for genome function.

ADVANCES:

Recent advances in genome mapping using single-cell biochemical and imaging methods have enabled systematic probing of higher-order genome organization at the level of individual cells. Results from these approaches validate observations from traditional population-based methods but also reveal that genome organization is considerably more variable than anticipated. In particular, specific chromatin-chromatin interactions appear to be present in only a relatively small fraction of cells in a population, and the genome organization at individual alleles in the same nucleus differs considerably. Furthermore, the internal shapes of structural features such as chromatin domains are fluid, and the genomic positions of the boundaries between chromatin domains vary from allele to allele. The highly variable nature of genome architecture points to a high degree of intrinsic noise in genome organization, in line with the observed stochasticity in gene expression.

OUTLOOK:

The observed single-cell heterogeneity in genome organization challenges the traditional view of gene regulation, which assumes that the epigenetic landscape controlling gene activity is largely stable. Rather, it now appears that both the structure and activity of genes are dynamic and stochastic. These probabilistic features of genome organization have implications for how long-lasting cellular states are generated and how individual cells transition between states during physiological or pathological processes, in that they suggest that fluidity between cell states may be an underlying property of a population of cells. The extent and relevance of genome organization and function can be addressed experimentally using the wide arsenal of tools now available, including deep sequencing and single-molecule imaging of chromatin. A key question is whether variability in structural features of chromatin causes the stochastic activity of individual genes, or vice versa. Furthermore, although it is likely that the dynamic motion of chromatin and the randomness introduced during cell division contribute to variability in genome architecture, their importance and their roles in generating genome heterogeneity have not been characterized. An intriguing—and potentially far-reaching—possibility is that the variability of genome organization is cell type-specific and actively regulated, which would add a novel layer of control to gene expression and genome function. Rather than implying that genome organization is not functionally relevant, the observed variation in genome organization suggests new mechanisms by which chromatin topology may affect cell function.

Graphical Abstract

Genome organizational plasticity and its potential functional role. Major features of genome organization, including individual chromatin-chromatin contacts, loops, domains, and chromosome territories, are highly plastic, as is gene expression. Recent evidence suggests that this structural heterogeneity is stochastic and intrinsic to the DNA polymer and may be a functional and regulated feature of a cell.

The complex three-dimensional organization of genomes in the cell nucleus arises from a wide range of architectural features including DNA loops, chromatin domains, and higher-order compartments. Although these features are universally present in most cell types and tissues, recent single-cell biochemistry and imaging approaches have demonstrated stochasticity in transcription and high variability of chromatin architecture in individual cells. We review the occurrence, mechanistic basis, and functional implications of stochasticity in genome organization. We summarize recent observations on cell- and allele-specific variability of genome architecture, discuss the nature of extrinsic and intrinsic sources of variability in genome organization, and highlight potential implications of structural heterogeneity for genome function.

Eukaryotic genomes must be condensed by several orders of magnitude to fit into the nucleus of a cell. Compaction of genomes is achieved by coiling the DNA around histone proteins to form a chromatin fiber, which is subsequently folded into complex higher-order structures such as loops, domains, and compartments (1) (Fig. 1). Chromatin loops of various sizes affect gene expression by modulating interactions between gene promoters and distal regulatory elements and preferentially occur within chromatin domains, referred to as topologically associating domains (TADs). These chromatin domains are typically between 200 kb and 1 Mb in size and are defined by boundary elements occupied by architectural chromatin proteins (2). On a larger scale, euchromatic and heterochromatic regions tend to self-segregate and form transcription-permissive and repressive compartments, roughly corresponding to euchromatin and heterochromatin, respectively (1). More globally, chromosomes and specific genes show preferred localization in three-dimensional space, at times correlating with their expression status (3) (Fig. 1). Many of these architectural features of genomes are evolutionarily conserved (4), suggesting a fundamental role of spatial organization at various scales in genome function. Indeed, genome reorganization events have been linked to functional changes—for example, in differentiation-related genes including those that encode Hox proteins (5–7), globin proteins (8), and several T cell-specific transcription factors (9). Abnormal organization may also be a sign of disease as certain cancers also show disease-specific repositioning (10) or aberrant loop formation at genes, including driver genes such as TP53 (11), and viral infection drastically remodels nuclear architecture (12–14).

Fig. 1. Variability at all levels of genome organization.

The genome is organized at multiple levels, and all of these levels show variability. DNA wraps around histone proteins at variable positions to form a chromatin fiber of variable width. Chromatin forms domains made up of individual loops, each of which occurs at only low frequency. Domains in turn assort within compartments, but their position within that compartment is variable. Similarly, chromosomes have preferred positions but can be found anywhere in the nucleus.

Despite the ubiquitous presence of all major architectural features of genomes, there is considerable variability and heterogeneity in genome organization at the single-cell level. Although the positions of chromosomes and genes are nonrandom in the cell population, their patterns of localization are variable among individual cells (Fig. 1) (15). Furthermore, individual chromatin interactions identified in bulk biochemical assays, such as Hi-C and ChIA-PET, typically occur in a relatively small fraction of cells in the population at any given time (16, 17) (Fig. 1). The fundamental variability of genome organization is mirrored by stochasticity in the transcription process itself (18). Recent in-depth studies of heterogeneity in genome organization, facilitated by new quantitative imaging and single-cell sequencing, suggest new models of genome function based on variability and stochasticity (16, 17, 19, 20).

In assessing the role of heterogeneity in genome organization and function, three major types of variability must be considered: extrinsic, allele-specific, and intrinsic variability (Fig. 2). Extrinsic variability arises from processes acting on the cell level, including stochastic, cell type-specific, and disease state-specific differences in levels of cellular machinery, as well as changes associated with the cell cycle (21–23) (Fig. 2A). Allele-specific variability occurs when the two alleles in the same cell are organized, and consequently function, in a mutually exclusive manner, especially in cases where one allele is silenced, such as imprinted genes, olfactory receptor loci, or X-chromosome inactivation (24–26) (Fig. 2B). Finally, intrinsic variability is caused by stochastic intracellular events such as dynamic motion and binding of protein complexes (Fig. 2C). Below, we discuss how different types of variability affect genome organization and function.

Fig. 2. Classes of variability in spatial genome organization.

(A) Extrinsic variability emerges when two cell types within a population have different preferred genome conformations. This variability results in a bimodal distribution of chromatin-chromatin distances. The two alleles in a cell will behave the same. (B) Allele-specific variability is created when chromatin loops form at one allele in a cell but not the other allele. This type of variability results in a bimodal distribution, but alleles will behave differently. (C) Intrinsic variability occurs when each allele represents an individual looping event. In this case, cells will occur in the population with zero, one, or two alleles forming loops. The distance distribution will be unimodal, and the alleles will show no correlation in distances.

Variability in gene expression and genome organization

Stochastic gene expression

It is now recognized that gene expression at the single-gene level occurs in a stochastic fashion and that stable phenotypes on a population level are derived from variable single-cell gene expression patterns (27, 28) (Fig. 3A). Stochasticity in gene activity is evident even in clonal populations of cells, which show variable activity of exogenous gene reporters (29, 30) and generate variable numbers of mature mRNA molecules at any given time (18). Active transcription generally occurs in short bursts at irregular intervals (31) (Fig. 3A). This pattern of punctuated gene transcription is common in all organisms, is particularly widespread in mammals (18, 27, 32), and has been observed even at genes with very high levels of expression and very tight developmental regulation (33, 34). Transcriptional stochasticity has been implicated in cell culture asynchronicity (35), viral latency (36), and differentiation (37).

Fig. 3. Stochasticity in gene expression and association.

(A) Stochastic gene expression results from random activity at gene promoters and causes a skewed distribution of transcription events, nascent mRNAs, and mature transcripts, in which at any given time a few cells produce most of the mRNA in the population. This may occur at genes with overall low expression levels (i.e., blue) or high expression levels (i.e., orange). (B) Stochastic chromatin interactions result from randomness inherent in the position and movement of the DNA fiber. They result in a population of cells in which interactions between any two regions are possible and interactions between each individual pair are rare. (C) Stochastic domain boundaries form as a result of the movement of DNA and the architectural proteins that associate with it. They result in a population of cells in which boundaries between domains can occur at any genomic position, with a preference for enriched boundary sites.

Both extrinsic and intrinsic noise act on gene expression (29). Extrinsic noise comes from cell-to-cell variability in gene expression machinery, including differences in cell cycle stage (38), differential cell type-specific expression of transcription factors (39), and random stochasticity acting at the cell level, such as the number of RNA polymerase molecules in the cell (40). In contrast, intrinsic noise derives from the underlying randomness of the gene expression process itself, including the dynamic binding of transcription factors and chromatin motion (28). In addition, different promoters have different sensitivities toward intrinsic noise, depending on their activity (41), their genomic location (42), and the biochemical mechanism of their activation (43). Given the tight interplay between gene expression and chromatin structure, the stochastic nature of transcription points to structural heterogeneity of the chromatin fiber and of the genome as a whole.

Heterogeneity in genome organization

The stochasticity observed in gene expression is indeed paralleled by high variability in genome organization. Direct evidence for extensive heterogeneity in genome organization comes from single-cell methods (Fig. 3B) (16, 17, 20). Results from these approaches indicate that large-scale genome structures, such as chromatin compartments, are consistently present in individual cells (44), but smaller organizational units, such as chromatin domains, are variable between cells (45, 46) (Fig. 3C). This model is supported by super-resolution imaging, which reveals that although domains can be observed in individual cells, their boundaries are highly variable (20) (Fig. 3C). In addition, single-cell Hi-C studies have shown that the relative position of individual domains within compartments varies considerably (47), and modeling suggests dynamic movement within individual chromatin domains (48).

Looping events are also highly variable between cells and are generally rare in a cell population (16, 17) and even the most enriched chromatin-chromatin interactions occur in only a fraction of cells, typically on the order of 10 to 30% (16, 17). Most loops are likely to be dynamic structures, forming and reforming many times during a single interphase (49). Furthermore, the behavior of the two alleles in the same nucleus is not correlated, which suggests that the variability present is largely intrinsic and not dependent on cell-level variables such as cell cycle stage (17). These observations are consistent with single-cell mapping data showing that individual interactions underlying the formation of chromatin domains and loops are highly variable (47).

In all, these observations demonstrate a high degree of heterogeneity in genome organization; they suggest that in individual cells in a population, genomes can assume many distinct, albeit related, spatial conformations mediated by rare, short-lived chromatin-chromatin interactions rather than by persistent and pervasive associations. This variability does not imply that structural features of chromatin organization are not relevant for function, but rather it suggests that structural heterogeneity may be another layer modulating the stochasticity in gene expression.

Types of organizational variability

Allele-specific genome organization

Because genome organization is inherently flexible, the two alleles in a cell may be differentially organized. Allele-specific stable differences in chromatin architecture are often associated with cellular processes that require persistent differences in allele function, such as dosage compensation, imprinted genes, and monoallelically expressed genes (Fig. 2B). Genes such as the permanently inactive immunoglobulin loci, as well as the entire inactive X chromosome, are sequestered at the nuclear periphery, whereas their active counterparts are found nearer the center of the nucleus (50, 51). Allele-specific condensation, looping, and clustering occur within imprinted domains (52) and the active X chromosome contains numerous loops and smaller domains, whereas the inactive X is condensed into two “mega-domains” (2, 53). In olfactory neurons, thousands of olfactory receptor (OR) loci that are not expressed are sequestered into a single heterochromatin focus (54), and their enhancers cluster into a single super-enhancer (25, 55), similar to other monoallelically expressed genes such as IFNβ (56).

The inherent randomness and variability in spatial genome organization play an important role in selecting alleles for silencing or activation in systems where this choice is random. As an example, XIST, the long noncoding RNA that causes X-chromosome inactivation, is located at the boundary between two TADs, and its expression fluctuates as it associates either with the upstream or downstream TAD until expression from one allele solidifies the organization and drives permanent silencing (26). Similarly, initially stochastic and variable interactions between individual OR promoters and a regulatory super-enhancer cluster allow for the selection of a single OR gene to be expressed, while minimizing biases due to genomic position (55).

In addition to these well-established cases, genome organization and its stochastic nature may be a universal mechanism in establishing differences in the cellular properties of alleles (57). Indeed, allelic imbalance in expression, observed at nearly one-quarter of all testable genes, has been correlated with allelic differences in chromatin marks and promoter-enhancer contacts in multiple lineages (22), suggesting more widespread use of this mechanism than currently appreciated.

Spatial variability due to intrinsic noise

A major driver of variability in gene expression and genome function is the intrinsic noise present in individual cells. In allele-specific organization, the two alleles within a cell assume persistent differential organization and each cell in a population shows the same pattern (Fig. 2B). By contrast, intrinsic variability refers to the situation where two alleles in a cell may be similarly organized but stochastic biophysical features, such as transient fluctuations in transcription factor binding or random diffusion of genomic loci, cause intrinsic variation in the activity and organization of the two alleles. Most of the variability in chromatin interactions appears to be intrinsic, as indicated by the fact that there is little correlation between interactions of the two alleles in the same cell (17). In addition, allele-specific mapping data demonstrate that the two alleles in a cell generally contact different sets of partners (58, 59).

One obvious source of intrinsic variability in spatial genome organization is the dynamic nature of chromatin. Chromatin undergoes persistent subdiffusive movement, caused by thermal motion as well as the continuous action of adenosine triphosphate–dependent chromatin remodeling complexes, and at the same time is constrained by association with nuclear subcompartments (60). In addition, individual loci also undergo active, directional, long-range movements in response to transcriptional induction (61, 62). However, these dynamic events are not sufficient to fully account for all intrinsic variability, because the range of distances sampled by most locus pairs is considerably larger than the roughly 1-μm radius of constraint typically observed for DNA movement in live cells, and long-range motion events are rare (17, 60). Thus, although dynamics plays a role in generating variability, especially over short distances, at larger scales other chromatin features must determine the variability between alleles.

One possible contributing factor to intrinsic variability is the active remodeling of chromatin, which may alter the density and shape of large regions of the genome and thus change higher-order chromatin configurations (63). Stochastic association with structural features of the nucleus, including the nuclear lamina and intranuclear bodies, likely also contributes to structural differences between alleles (64). In addition, the interaction of chromatin proteins with their substrate is dynamic (65), even for the architectural proteins that form “stable” loops, such as CTCF and cohesin (49). Given the high levels of intrinsic variability in spatial genome organization, it is likely that most features that contribute to genome organization also contribute to its heterogeneity.

Functional consequences of structural variability

Structural heterogeneity in genome organization has been linked to function in several settings. Most prominent is the regulation of transcription via the dynamic formation of promoter-enhancer loops. In transgenic systems in Drosophila, colocalization between promoters and their cognate enhancers is necessary for proper expression in cis and in trans (66, 67). Although the initial contact between the promoter and an enhancer occurs largely stochastically as a result of dynamic motion in the absence of a loop (66, 67), active transcription serves to stabilize the interaction (66), and spatially distant loci compete for enhancer binding and gene activation (67). The role of promoter-enhancer looping in endogenous transcription is likely to be more complex, as there is no strong, systematic link between changes in genome organization and changes in gene expression (68). At Drosophila Hox genes, for example, several enhancers are more likely to be near their cognate promoters when transcription is taking place, but the correlation is overall weak (69). This observation suggests a mechanism in which active enhancers are those capable of sampling the volume of a domain but do not necessarily form stable loops with their target promoters (69). Similarly, at the mouse Sox2 locus, enhancer-dependent activation occurs even without spatial proximity between enhancer and promoter (70). In some cases, active loci move less than silenced ones (71), whereas for other genes, actively transcribing alleles show increased mobility (72, 73). Additionally, active transcription induced by viral activation disrupts loops (14). On the basis of these observations, it appears that the movement of enhancer elements, as well as their position relative to promoters, is functionally relevant.

Spatial genome organization and its heterogeneity also affect genome functions other than transcription. Cancer-relevant translocations occur more frequently among chromosomes that are in spatial proximity, which suggests that variability in spatial organization strongly determines the repertoire of translocations in a given cell type or tissue (74–76). Additionally, haploinsufficiency, mutation, and down-regulation of the architectural protein CTCF, all of which lead to destabilization of chromatin loops and causemorevariability in genome organization, have tumor-promoting effects including deregulation of cancer-relevant gene expression programs, disruption of cell polarity, and a decrease in patient survival (77, 78).

Finally, it is plausible that the variability in genome organization may facilitate cell fate decisions during differentiation, development, and disease. Analogous to transcriptional noise, stochasticity in genome architecture generates functional heterogeneity in the population, with individual cells expressing related but varying sets of genes. Cells with particular patterns of transcriptome-wide noise are more prone to transitioning when stimulated to differentiate (37). Consistent with the observation of extensive heterogeneity in tumors, it can be speculated that the emergence of cancer-initiating cells may represent the transition of a subpopulation of healthy cells whose gene expression program is particularly sensitive to cancer-causing mutations, possibly due to their epigenetic predisposition (79).

Generation of a stable output from variable inputs

The extensive structural heterogeneity in genome organization on the one hand, and on the other hand the requirement for establishment and maintenance of stable cell states defined by characteristic levels of proteins and mRNA and common morphological features are not mutually exclusive. Stable functional states and cell populations can be generated by two mechanisms: time- or population averaging of gene activity (Fig. 4A) or the formation of functionally equivalent but morphologically diverse cellular structures (Fig. 4B).

Fig. 4. Averaging and phase separation create stability out of organizational variability.

(A) Because of the stability of the mRNA and protein products, highly variable and transient promoter-enhancer loops can nonetheless result in consistent numbers of mRNA molecules being transcribed, as well as consistent transcription profiles, within a cell or tissue. (B) Phase separation results in the segregation of molecules or DNA loci and can result in a stable nuclear body forming from a complex combination of variable associations.

Stability generated by population averaging

Because the intervals of transcriptional bursts are typically shorter than the half-lives of RNA and protein, fluctuations in the amount of mRNA produced during stochastic gene expression are averaged out over the lifetime of the cell to generate an overall stable cellular state (80) (Fig. 4A). The same principle applies to population averaging. An example of this behavior is the early embryonic patterning in Drosophila, where, before cell membranes arise to separate the embryo into different cells, mRNAs are free to diffuse throughout the embryo, resulting in low spatial variability despite high transcriptional noise (34). The stable and reproducible transcription profiles at the level of cell populations and tissues are thus consensus averages and represent the result of highly variable transcription of individual genes averaged over time and among all cells in the population (Fig. 4A).

Variability in chromatin interactions in individual cells is similarly averaged out in the population. Regulatory loops may be variable between cells at any given time, but as a result of their dynamic nature, each promoter may encounter its regulatory element and initiate transcription a similar number of times per cell cycle (66). This results in overall uniform gene expression (Fig. 4A). Similarly, although the distance between two potential chromatin interaction partners varies in individual cells, mean distance and interaction frequency are stable on the level of the population (Fig. 4A). Thus, although every cell shows dynamic movement of every DNA locus, and although at any given moment the population shows a wide variety of conformations, averaging on the basis of either of these variables results in a stable output.

Stability generated by higher-order physical structures

A further mechanism to generate stability from variability is the formation of higher-order physical structures in the nucleus, such as peripheral heterochromatin compartments, nucleoli, and other nuclear bodies (21). Although these nuclear features are variable in appearance in individual cells in a population, they are functionally equivalent among cells and thus serve to equalize the effect of structural variability in the population. One mechanism that has emerged in the biogenesis and maintenance of such stable nuclear structures and genome features from dynamic interactions is phase separation (81) (Fig. 4B).

Phase separation refers to the demixing of proteins leading to the formation of spatially distinct membraneless protein aggregates (81). The process of phase separation appears to promote the formation of heterochromatin domains and of super-enhancer clusters (59, 82–86). In both cases, phase separation acts on large sets of loci rather than individual locus pairs, thus conferring flexibility and stability at the same time. Individual loci may rearrange within the structure without disrupting it, and the dissociation of individual loci from the structure will not cause the dissolution of the structure as long as a critical mass is maintained (Fig. 4B). The counteracting effects of stochasticity and stability are illustrated in the organization of the OR genes, whose ability to phase-separate into heterochromatin and to form a single super-enhancer hub ensures monoallelic expression (87, 88). However, the underlying stochasticity of positioning of individual OR loci ensures that the choice is random and unique to each cell (55). The intricacies of the relationships among chromatin movement, phase separation, and genome function are currently unclear, but recent work has shown that phase separation of proteins is able to segregate loci by excluding DNA from dense protein droplets and can also bring them together by tethering them to the same droplet (89); this work suggests that phase-separating proteins maybe capable of generating stable structures from dynamic interactions.

Functional benefits of structural variability

At first glance, the heterogeneous nature of spatial genome organization may appear to complicate genome function. However, variability in structure has important functional benefits. In particular, fluctuations in chromatin configurations ensure that occasional extreme responses at individual genes are short-lived and reset transcription to a baseline level (28). In addition, the dynamic nature and structural variability of genome regions enhances responsiveness to external stimuli. For example, a system in which enhancer-promoter contacts fluctuate rapidly is more responsive to stimulation than one with static enhancer-promoter configurations that would take longer to assemble or disassemble. Such rapid responses are particularly relevant in fast-activating transcriptional networks such as hormone stimulation (90). Furthermore, the structural plasticity of chromatin facilitates the formation of phase-separated structures, which cannot form in static, dense chromatin regions (89). Taken together, the dynamic nature and variability of genome organization, rather than being an impediment to controlled gene expression, may thus facilitate the establishment and maintenance of stable cellular phenotypes.

Perspectives

Recent observations of genome organization by biochemical mapping and imaging approaches have revealed that the genome is highly plastic in its spatial arrangement. The fact that any chromatin-chromatin interaction in the genome is relatively rare in a cell population is in line with fundamental observations about the nature of gene regulation, including stochastic gene expression, the inherently dynamic nature of the chromatin fiber and the proteins that bind to it, and the plasticity of gene expression programs in differentiation and disease. It thus appears that multitudinous variable or low-frequency chromatin interactions, rather than pervasive stable sequence-specific loops, shape the organization of chromatin and chromosomes in a cell type-specific manner and determine gene expression (16).

The recent studies on heterogeneity in genome organization have opened several avenues for future research. It will be essential to map the degree of variability in genome organization in various cell types, particularly in situations where cells undergo changes in their states, such as during differentiation, in development, and in disease. The possibility that the variability of genome topology is regulated or regulable presents an intriguing hypothesis to test in these models. In addition, it will be essential to delineate whether chromatin interactions drive gene activity or vice versa. It will also be critical to simultaneously label DNA and RNA to determine which conformations of the DNA fiber are conducive to active transcription and regulation. The investigation of how variability in spatial genome organization contributes to function will be greatly aided by single-cell biochemical and imaging techniques that allow the probing of many loci per cell and in a large number of individual cells (16, 17, 20). Finally, it is noteworthy that much, albeit not all, of the evidence for heterogeneity in genome organization comes from the analysis of cultured cells, and it will be important to determine the corresponding degree of variability in intact tissues and organisms.

Heterogeneity in structure and function is a hallmark of most biological processes and one that most biologists have been acutely aware of. However, experimental limitations have impeded the interrogation of cellular heterogeneity. The advent of single-cell sequencing and biochemistry methods and the use of quantitative imaging methods are enabling us to address this important aspect of genome function. Although variability in genome organization has in the past complicated attempts to understand the functional relevance of genome organization, the recent progress summarized here has made structural heterogeneity an experimentally tractable parameter and suggests that variability in genome organization is a driving force in genome function, rather than just a confounding factor in its analysis.