Towards decoding the mechanisms that shape sub-megabase-scale genome organization

Abstract

Understanding genome organization at the kilo-base to mega-base scale is critical, as it encompasses genes and regulatory elements. Improvements in the resolution of experimental techniques have revealed novel structural motifs at this scale, including micro-compartments, nucleosome clutches, microdomains, and packing domains. Here we review recent progress on developing theories to explain these observations. Key advances include elucidating the role of nucleosome positioning and epigenetic modifications, the role and mechanisms of compartmentalization in local structure, and the interplay between loop extrusion and phase separation. This work has revealed probable mechanisms by which the observed structures emerge, but it remains unclear how these factors act together in the cell. To this end, recent studies have used chromatin conformation capture data in concert with diverse genomics datasets to create native-like models of chromatin at nucleosome-resolution and below. While several roadblocks remain, this strategy promises to decode how molecular forces sum to shape chromatin structure and ultimately regulate transcription.

Introduction

Our understanding of genome organization was revolutionized by the development of Hi-C and related methods [1, 2]. Hi-C measures the frequencies of physical contacts between all regions of the genome simultaneously. Complementing this approach, light microscopy of DNA and associated factors revealed chromatin’s spatial arrangement [3, 4]. Observations from these methods facilitated the identification of two major biophysical driving forces of genome organization.

First, chromatin is partitioned into phase-separated compartments [5]. Active/euchromatin regions partition into the A compartment and inactive/heterochromatin regions partition into the B compartment. Compartments present as a “checkerboard” pattern of preferential self-interactions in Hi-C Figure 1a and distinct dense and dilute regions when imaged Figure 1d. Compartments can be reproduced using co-polymer models that favor intra-compartment contacts [6, 7, 8, 9]. Recent work has elucidated the role of tethering to nuclear landmarks, e.g. the nuclear lamina and speckles, in establishing compartment-scale structure [4, 10, 11, 12].

Figure 1: Recently observed chromatin structural motifs.

(a) In contact maps, compartments appear as a checkerboard pattern. Data from WT mouse embryonic stem cells (mESCs) is shown in the upper triangle (GSE130275) [28] and data from mESCs with cohesin depletion is shown in the lower triangle (GSE178982) [29]. Compartments are largely unaffected by cohesin depletion. (b) Topologically associating domains appear as small triangles along the diagonal and largely disappear upon cohesin depletion. (c) High-resolution micro-C experiments [27] have observed micro-compartments, the small focal interactions, and microdomains, the insulated domains seen along the diagonal. Microdomains often, but not always, have micro-compartments at their boundaries. (d) Imaging of DAPI-stained DNA reveals compartments [25]. Heterochromatin appears as large, dense regions and euchromatin appears as more dilute, speckled regions. (e) Globule domains with a radius of a few hundred nanometers have been observed with electron microscopy [25] and imaging studies [24]. (f) Irregular, compact groups of nucleosomes, referred to as clusters or clutches, have been observed in cryo-ET [20] and super-resolution imaging [19]. (g) A cartoon depicting the potential hierarchical arrangement of clutches, microdomains, and packing domains. (h) The approximate size ranges of the recently observed structural motifs, compared to the established TAD and compartment motifs. For all contact maps, data was downloaded as pre-balanced cooler files, normalized such that the contact frequency between neighbors is 0.5 on average, and missing values were interpolated. Panels a, b, and c are at resolutions of 200 kb, 10 kb, and 500 bp, respectively

Second, topologically associating domains (TADs) are approximately 100 kb–1 MB regions that preferentially self-interact, insulated from neighboring regions [13, 14]. In Hi-C contact maps, TADs present as squares along the diagonal Figure 1b. A subset of TADs is bound by CTCF and cohesin, and their formation is dependent on cohesin-mediated loop extrusion, supported by their loss upon cohesin depletion [15]. Computational models of loop extrusion, incorporating cohesin loading, its procession along the chromatin fiber, and subsequent blockage by CTCF, accurately replicate many TAD structures observed in Hi-C data [16, 17]. An alternative hypothesis, termed the diffusion capture model, proposes that the cohesin ring directly captures large loops as they transiently interact through diffusion as opposed to the loop being extruded through the cohesin ring [18].

Recent advancements in the resolution of experimental measurements have revealed new phenomena at the sub-megabase scale for which we are only beginning to develop theories to explain. Super-resolution microscopy [19] and electron microscopy [20, 21, 22] have revealed dense clusters of nucleosomes, sometimes referred to as “clutches” Figure 1f. These clutches typically span tens of nanometers and are composed of a few to hundreds nucleosomes. On a larger scale, these methods have revealed “packing domains” with diameters of a few hundred nanometers [23, 24, 25] Figure 1e. It is believed that these packing domains correspond roughly to the TADs observed in population Hi-C. However, packing domains appear even in the absence of cohesin [23, 24, 25, 26]. Finally, refinements to micro-C have provided contact maps at nucleosome resolution, revealing “micro-compartments” and small contact domains. These small contact domains, which we will refer to here as “microdomains”, differ from TADs in that they are generally smaller and insensitive to depletion of cohesin [27] Figure 1c.

Understanding the biophysical mechanisms that create these sub-megabase-scale structural motifs is critical, as this is the typical scale of genes and associated regulatory elements and thereby holds the key for understanding the connection between chromatin structure and transcriptional regulation. Here we review recent progress towards this goal.

Attractions between nucleosomes condense chromatin

Three of the recently observed structural motifs—clutches, microdomains, and packing domains—are essentially collapsed chromatin regions of various sizes. The exact relationship between these structures is not yet clear, but they appear to be arranged in a hierarchy: nucleosomes group into clutches, which group into microdomains, which group into packing domains. Whether all of these structures are present for all regions or if there are further levels of organization is not yet known. However, a variety of evidence suggests that heterochromatin and euchromatin alike are mostly composed of collapsed domains [30]. Investigating the relationship between microdomains and the other two structures is especially challenging as they have been observed in population-averaged micro-C experiments as opposed to single cell imaging experiments; clutches and packing domains that form stochastically and are not well phased might not appear in population measurements.

The condensation of chromatin regions is likely driven primarily by intrinsic interactions between nucleosomes [31, 32]. Various experimental studies have attempted to quantify the strength of these interactions, reporting values ranging from 2 to 14 kT [33, 34]. This discrepancy likely stems from the indirect nature of most measurements and uncertainties in their interpretation. Additionally, differences in salt concentrations across experiments may directly influence interaction strength. To address this inconsistency, we employed a residue-level coarse-grained model with explicit ions to estimate the strength of nucleosome interactions [35]. The inclusion of explicit ions enhances the accuracy of simulations by mitigating the need for severe approximations in calculating electrostatic forces. Under physiological conditions, we calculated an interaction energy of approximately 9 kT.

An interaction energy exceeding thermal fluctuations supports the physiological relevance of nucleosome interactions in native chromatin organization. Such strong interactions align with in vitro findings showing that unmodified nucleosome arrays condense into droplets [36] and that reconstituted native nucleosome arrays adopt compacted structures [37].

In silico simulations have elucidated the role of nucleosome interactions in the phase-separation behavior of chromatin [38, 39, 40]. Golembeski and Lequieu [38] demonstrated that a nucleosome interaction energy of merely 1.7 kT suffices to drive phase separation and that further increases in interaction energy result in a transition from liquid-like to solid-like states. Similarly, Farr et al. [39] reported that phase separation occurs with moderate nucleosome interaction energies, and additional factors, such as the DNA’s propensity to “breathe”, influence the liquidlike characteristics of individual chromatin fibers and the phase-separation behavior of chromatin melts.

Recent work suggests that chromatin dissolved in a solution with only salts forms solid-like, irregular structures that do not fuse, and that liquid-like properties emerge only in the presence of macromolecular crowders (e.g. small proteins) [41, 42]. The molecular mechanisms governing the material properties of chromatin condensates—and the transition from liquid to solid—remain incompletely understood [43], presenting a promising direction for future work.

Nucleosome free regions and external forces can create insulated domains

Given that nucleosome interactions are strong enough to drive condensation of chromatin, the recent observations of compact chromatin structures in vivo are unsurprising. Perhaps, the bigger mystery is why some domains remain insulated and do not mix together into a single domain. In these next two sections, we review mechanisms that have been shown to create insulated boundaries between domains. In this section, we review the mechanisms identified to be most important at the scale of clutches and microdomains, and, in the next section, we review mechanisms identified to be critical at the scale of packing domains. However, we expect that all of these mechanisms play some role at all scales.

An important and extensive line of work from Schlick and coworkers have characterized the impact of nucleosome positioning and epigenetic factors on chromatin structure, reviewed in detail by Portillo-Ledesma et al. [44]. In their studies, they utilize a model that represents nucleosomes as a set of surface charges and includes explicitly represented coarse-grain chromatin tails and DNA. In this way, their model captures the effects of attractive nucleosomenucleosome interactions and the geometric constraints imposed by the linking DNA. Portillo-Ledesma et al. [45] find that interjecting exceptionally long linkers, often called “nucleosome free regions” (NFRs), creates insulated boundaries and the formation of nucleosome clutches Figure 2a. Gómez-García et al. [46] use experimental data to model the Pouf5f1 locus in mouse embryonic stem cells and neural progenitor cells. They use MNase data to determine nucleosome positions and ChIP-seq data to place linker histones and histone acetylations. In this native-like system, they observe nucleosome clutches, largely consistent with imaging studies [19].

Figure 2: Mechanisms of domain insulation.

(a) Long DNA linkers (green) between nucleosomes (purple) can insulate adjacent regions, due to the long persistence length of bare DNA. Nucleosome-level simulations have shown that nucleosome free regions can create clutches [45] and microdomains [47, 48]. (b) Tension can cause chromatin to break into clutches, even with uniform nucleosome positioning [49]. (c) Loop extrusion by cohesin (red) directed by CTCF (orange) can result in topologically associating domains (TADs). Here we depict the fully extruded state with cohesin blocked by two bound CTCF molecules. However, we note that live-cell imaging suggests that cohesin is more commonly partially extruded [50]. Moreover, we purposefully draw the extruded chromatin domain in a collapsed state consistent with simulation work suggesting that models integrating loop extrusion with condensation best explains the conformational ensemble seen in chromatin tracing experiments [51]. (d) Microphase separation occurs when different regions of chromatin have distinct epigenetic modifications and more favorable interactions form between regions with the same type. (e) In region capture micro-C data, microcompartments often appear at the boundaries of microdomains [27]. However, to the best of our knowledge, no theoretical work has explained this phenomena. Bridging interactions between transcription factors perhaps has a similar effect on chromatin structure as it also results in strong attractions between short genomic regions and has been shown to create microdomains in nucleosome-resolution simulations [52]

A combined experimental and simulation study characterized the conditions required to reproduce yeast chromatin domains in an in vitro system [47]. They found binding of a transcription factor and a chromatin remodeler to pack nucleosomes around it are sufficient. Together, these factors create a NFR surrounded by regularly positioned nucleosome arrays. This pattern produced insulated chromatin domains both in experiments and nucleosome-level coarse-grain simulations Figure 2a. These domains are on the scale of 1-5 kb: the size of the smallest microdomains observed in mammals [27]. A similar conclusion was reached by a study that showed that nucleosome positions obtained through MNase were sufficient to produce yeast chromatin domains [48].

The propensity of NFRs to create insulated domains is likely governed by DNA mechanics. Double-stranded DNA exhibits a persistence length of 50 nm, equivalent to approximately 170 base pairs in a linear conformation. Consequently, nucleosomal DNA ( 147 base pairs), when not bound to histone proteins, preferentially adopts extended conformations. This inherent tendency of DNA to favor linear over looped structures counteracts nucleosome interactions, thereby preventing the collapse of the two chromatin regions bridged by the free DNA.

In addition to intrinsic chromatin properties, external factors such as tension may also oppose nucleosome interactions, fragmenting chromatin into smaller domains or clutches [49] (Figure 2b). Tension could arise in vivo through force generating processes like transcription and chromatin remodeling or from tethering to nuclear landmarks, like the nuclear lamina or speckles. Clutch formation is energetically favorable under tension because it allows a subset of the highly favorable nucleosome stacking interactions to be retained.

Phase separation and cohesin-mediated loop extrusion act together to shape packing domains

In addition to their positioning, a key characteristic of native nucleosomes is their epigenetic modifications. Variations in these modifications across chromatin regions directly influence nucleosome interactions [53]. Such modifications transform chromatin into block copolymers, which exhibit more complex phase behaviors compared to simple bulk phase separation. Notably, block copolymers are known to undergo microphase separation, leading to the demixing of adjacent chromatin regions with distinct histone modifications into spatially segregated domains.

The importance of phase separation has been recognized in many studies, emerging as an important mechanism for the formation of chromatin domains. Conte et al. [51] develop models of chromatin structure based on phase-separation, loop extrusion, or both. They find that all models can reproduce contact maps from Hi-C, but phase separation based models better capture the variability observed in single-molecule chromatin tracing experiments. Along similar lines, Jeong et al. [54] investigate why some TADs don’t vanish after cohesin depletion. They conclude that these “persistent” TADs tend to be in different epigenetic states than their flanking regions, facilitating their maintenance upon cohesin depletion via phase-separation. Together, these works support a revised model in which chromatin intrinsically tends to phase separate into TAD-sized packing domains and cohesin-mediated looping largely acts to guide the placement of the domains [55]. In the case of persistent TADs, epigenetic modifications are also sufficient to guide their placement through microphase separation Figure 2d.

The role of compartments in kilobase-scale chromatin structure

The traditional view is that the separation of chromatin into A and B compartments occurs at a relatively large scale with large contiguous regions being entirely partitioned into either compartment. Two recent studies have challenged this view. First, very high-resolution Hi-C has shown that, when computed using smaller genomic bins (500 bp), compartments alternate more frequently than previously believed [56]. In particular, often only a small region near gene promoters and enhancers will partition into the A compartment. Second, region capture micro-C (RCMC) revealed short regions, often at promoters and enhancers, that interact strongly and non-specifically together [27] Figure 2d. Goel et al. [27] call these small scale compartments “micro-compartments”.

There is not yet consensus on the mechanism of micro-compartment formation. One hypothesis is that it could be driven by bridging interactions between transcription factors (TFs). Portillo-Ledesma et al. [52] study the role of TF-mediated bridging interactions using a nucleosome-level model with TFs represented as additional favorable interactions between binding sites. In model systems, they find that bridging interactions can create loops similar in appearance to micro-compartments and that certain arrangements of TF bridging interactions result in microdomains, consistent with the RCMC data Figure 2e. A notable finding is that properties of the chromatin fiber can significantly modulate the effect of TF binding on chromatin structure. For example, in simulations of a native-like Eed gene system, varying the amount of linker histone changes whether the promoter is occluded due to bridging interactions, which could explain the activation of Eed during differentiation.

An alternative mechanism for non-cohesin mediated looping is attractive interactions between Pol II or the attached nascent RNA. Salari et al. [57] propose that some transcription mediated process results in attractive interactions between active regions, leading to micro-compartments. They create a polymer model that includes transcribing Pol II and show that this model recapitulates the increased contacts for gene bodies and enriched contacts between transcribed genes observed in micro-C. Their work presents a compelling mechanistic hypothesis for microcompartment formation. However, experimental evidence shows that transcription inhibition does not significantly affect the strength of most micro-compartments [27], suggesting that transcription is at least not the sole driver of micro-compartment formation.

Data-driven modeling of gene-sized chromatin regions

While various studies have endeavored to model gene-sized regions of chromatin from the bottom up, inaccuracies in force fields and missing accessory factors make it hard to accurately model native chromatin structure. A different paradigm is to inform simulations with experimental data, a strategy often referred to as “data-driven” or “integrative” modeling. This strategy has long been used at the chromosome to genome scale to generate 3D conformational ensembles that recapitulate Hi-C data [58]. However, until recently, the limited resolution of the data has prevented application to gene-sized regions.

Two recent studies have used data-driven modeling to generate nucleosome-resolution models. Li and Schlick [59] model 50-100kb regions, including linker histone, histone tail acetylation, and non-uniform linker lengths derived from ChIP-seq and MNase, respectively. They impose constraints on the models by distributing the experimental contacts across a set of replicas, i.e. if a pair is in contact 4/10ths of the time, then in 4 of 10 simulations a restraint on that contact is added. Brandani et al. [60] model the 10 kb region surrounding the nanog promoter. They find that tetranucleosome folding motifs differ between genomic regions with promoter and enhancer regions showing more open conformations.

Data-driven simulations have also been applied at coarser resolution to elucidate the conformational ensemble of larger genomic regions. In addition to their nucleosome-level simulations, Brandani et al. [60] simulated 200 kb regions around the nanog and sox2 genes at 1 kb resolution. They conclude these regions are arranged as crumpled globules with enhancers and promoters often being brought together by large-scale collapse of the polymer. Conte et al. [61] fold the 2 MB LTN1 locus at 2.5 kb resolution. They too find that the resulting structures are generally compact. They highlight the high-degree of conformational heterogeneity, which they propose is due to the high degeneracy inherent to micro-phase separation. Future studies at this scale will benefit from the principled framework for parameterizing polymer models at a range of resolutions proposed by Kadam et al. [62].

Data-driven approaches have also been developed to make use of chromatin tracing data. Neguembor et al. [63] uses restraints from both imaging and Hi-C. Shi and Thirumalai [64] describe a maximum entropy approach for generating a conformational ensemble given pairwise distances from chromatin tracing.

Perspective

Great progress has been made on characterizing the molecular mechanisms that can influence sub-megabase chromatin structure. However, it remains unclear how these factors act together to shape the conformational ensemble of specific chromatin regions. Perhaps the biggest obstacle is that the nano-scale structure of chromatin has primarily been observed through imaging modalities that do not distinguish the genomic content of the feature being observed. This has made it challenging to directly connect structures with profiles of chromatin accessibility, histone modifications, and protein binding, and thereby the cast of potential driving forces.

The advent of ultra-resolution micro-C techniques, such as RCMC [27], allows us to examine the chromatin structure at single-nucleosome resolution for specific genomic regions. This positions data-driven modeling using diverse genomics datasets to produce native-like systems as a particularly exciting area for future work. Performing this work using residue-level coarse-grained models is of particular value, as it facilitates the inclusion of arbitrary proteins. Along these lines, a recent study has produced residue-level coarse grain structures including non-uniform nucleosome positions, linker histone, BRD4, Pol II, and a range of transcription factors [65]. Works such as this hold the key for understanding how higher-level structures and transcriptional regulation emerge from the sum of molecular interactions.

Highlights.

• Recent experimental observations have provided unprecedented views into sub-megabase-scale genome organization, driving theoretical advances.

• Simulations, at a variety of resolutions, have clarified the role of nucleosome positioning and other factors in shaping local chromatin structure.

• Phase-separation and cohesin-mediated loop extrusion act together to shape packing domains.

• Data-driven models synthesize diverse genomics data providing a view into native chromatin structure.