Large-scale Chromatin Organization: The Good, the Surprising, and the Still Perplexing

Abstract

Traditionally large-scale chromatin structure has been studied by microscopic approaches, providing direct spatial information but limited sequence context. In contrast, newer 3C (Chromosome Capture Conformation) methods provide rich sequence context but uncertain spatial context. Recent demonstration of large, topologically linked DNA domains, hundreds to thousands of kb in size, may now link 3C data to actual chromosome physical structures, as visualized directly by microscopic methods. Yet, new data suggesting that 3C may measure cytological rather than molecular proximity prompts a renewed focus on understanding the origin of 3C interactions and dissecting the biological significance of long-range genomic interactions.

Introduction

Traditionally, large-scale chromatin organization has been studied by microscopy, providing spatial information but limited DNA sequence context. In contrast, genomic methods now allow a new, orthogonal approach to investigating large-scale chromatin organization, providing rich sequence context but uncertain spatial context. Both approaches have their own limitations, assumptions, and technical concerns, and both typically are applied by different scientific communities possessing different expertise and posing somewhat different questions. While these new genomic methods have dramatically elevated interest in large-scale chromatin organization, there remains a notable gap that must be bridged to relate genomic results to actual physical models for large-scale chromatin organization.

Here I attempt to provide a conceptual framework for possibly bridging this gap while reviewing, from a microscopist's perspective, key advances from the previous two years. In particular, I focus on the recent demonstration of large, topologically linked DNA domains, hundreds to thousands of kb in size, which may represent constitutive units of chromosome folding.

Folding of chromatin fibers into large-scale chromatin domains

Local chromatin structure, corresponding to the positioning and composition of nucleosomes and the higher-order folding of nucleosome arrays into 30 nm chromatin fibers, has been reviewed recently elsewhere [1-3]. Emerging developments include the concept of a family of "30 nm" structures, replacing a single, canonical 30 nm fiber structure, as well the proposal that the 30 nm chromatin fiber might exist predominantly in regions of low chromatin density, whereas in areas of higher chromatin density intermingling of nucleosomes from adjacent chromatin fibers forms a "polymer melt" [4,5]. More difficult to resolve is to what extent mixtures of 10 and 30 nm chromatin fibers may co-exist within interphase chromosome regions and be distinguished experimentally from regions of polymer melt.

Easier to address are the types of differential chromatin compaction that might exist through the genome. The old "heterochromatin" / "euchromatin" division, derived largely from nonspecific, heavy metal staining of nuclei using transmission electron microscopy (TEM), has been misleading. Multiple DNA specific TEM stains have revealed that the so called "euchromatin" compartment, lightly stained by heavy metals, contains little DNA (Fig. 1A-B); instead most of the genome in the typical mammalian nucleus is packaged into large-scale chromatin structures with higher diameters than 30 nm [6-12] (Fig. 1A-G).

Fig. 1. Large-scale chromatin domains in search of genomic context- genomic interactions in search of structural context.

(A-C) Three staining methods demonstrate large-scale chromatin domains in thin-section TEM of freeze-substitution (A) or conventional (B-C) glutaraldehyde fixed samples: (A) DNA post-embedding immunostaining (dark, positive staining) of TEM thin section reveals most nuclear DNA is present in large-scale, frequently fiber-like structures. Scale bar = 1 μm. Reprinted from Fig. 4a, reference [7], Copyright 1994 J Histochem Cytochem. (B) DNA post-embedding staining by osmium ammine of rat liver nucleus shows no "euchromatin" texture. Large regions of nucleus are DNA free. Staining sensitivity is shown by visualization of decondensed chromatin within nucleolus (arrow). Reprinted from Fig. 1, reference [9], Copyright 1989 J Histochem Cytochem. (C) Uranyl and lead staining of detergent permeabilized Hela nucleus in buffer preserving large-scale chromatin structure. Staining is white in negative image. Extraction of nucleoplasmic protein results in absence of "euchromatin" texture after fixation and reveals large-scale chromatin structures. (D-G) Thin section, serial sections of CHO nuclei reveal (D-E) fiber-like, large-scale structures (small arrows) consisting of ~Mbp size segments punctuated by less condensed chromatin (large arrows). (F-G) Abrupt transitions from thick (large arrows) (~100-130 nm) to thinner (~60-80 nm) diameter size are seen, with local discontinuities by decondensed chromatin (thin arrows). Reprinted from Fig. 6 (D-E) and Fig. 7 (F-G), reference [16] , Copyright JCB 1994. Scale bars = 1 μm (A,B) and 500 nm (C), 120 nm (F-G). (H) Hi-C scatterplot showing long-distance DNA interactions in Drosophila embryonic nuclei over region of chromosome 3R. Off-diagonal, box-like, yellow regions define topological domains within which 3C interactions are favored over neighboring regions. X-Y axes show regions of 3R (bp) with associated chromatin average states. Reprinted from Fig. 3F, reference [37], Copyright 2012, Cell. (I) Cartoon illustrating that topological domains detected by 3C methods correspond to local regions of compact, but irregular chromatin folding separated by extended chromatin. (J) Boundary between topological domains (grey bar) is defined by asymmetry in DNA interactions up versus downstream of boundary region, with DNA within a topological domain interacting preferentially. (I-J) reprinted from Fig. 1b, reference [38], Copyright 2012 Nature.

In this context, old autoradiography studies showing concentrated sites of transcription at "heterochromatin /euchromatin boundaries" can now be reinterpreted as suggesting transcription at the edges of large-scale chromatin domains [13]. This interpretation is supported by demonstration of transcription on a condensed template in BAC transgene arrays [14] and by recent super-resolution light microscopy showing Br-UTP incorporation, RNA pol II immunostaining, and enrichment of some active chromatin marks at the periphery of large-scale chromatin domains [15].

Thus the old dichotomy of "heterochromatin" versus "euchromatin" now needs replacement with differentiation between multiple types of chromatin. From a structural perspective, TEM shows a range or continuum of large-scale chromatin compaction. In less compact chromosomal regions, large-scale chromatin domains form discontinuous fibers which can be traced in late G1 and S phase nuclei for up to several microns in selected regions; these fibers frequently fold back and supercoil on themselves to form plectonemic chromosome loops [16]. A range of large-scale chromatin domain diameters is seen within the same nucleus and these diameters can vary through the cell cycle [11,16,17]. Pericentric heterochromatin versus the facultative heterochromatin of the inactive X chromosome are easily distinguished, with the condensed Barr body of the inactive X showing large-scale chromatin domains comparable in diameter to more condensed regions elsewhere in the nucleus [18].

Sharp transitions between different compaction states can be visualized at the TEM level (Fig. 1D-G). A simple interpretation is that discontinuous fiber segments Mbp in size are connected by sharp transitions to less folded states; these fiber segments may correspond to structurally stable chromatin domains of similar, ~Mbp size consisting of clusters of DNA replication origins which initiate DNA replication synchronously [19-21]. While it has been assumed that these apparent fibers form from hierarchical folding of chromatin, topological looping of DNA may co-exist within these fiber-like structures [22].

The major problem with these structural results has been the difficulty testing whether these distinct structural states are reproducible for specific DNA sequences and mapping structure to sequence. Visualization of engineered chromosome regions by light and electron microscopy has correlated large-scale chromatin compaction with transcriptional activity, but this has relied on transgene arrays that may not completely recapitulate normal chromosome structure [14,22-24]. Fluorescent in situ hybridization (FISH) has demonstrated variable large-scale compaction correlated with gene transcriptional activity over specific gene loci [25-27] but this approach is low resolution, given both limitations of light microscopy and possible structural perturbations induced by the hybridization procedure [28-30].

The Good

Enter 3C (chromosome conformation capture) methods, possibly to the rescue. As reviewed recently [31-36], multiple variations on 3C use a proximity- ligation assay, typically together with PCR, to probe long-distance interactions between loci on the same or different chromosomes: the original 3C examines pairwise interactions, 4C examines interactions of one particular "bait" region with all other genomic loci, 5C examines pairwise interactions over a defined genomic interval, and Hi-C examines all pairwise interactions over the entire genome but at lower resolution than 5C. Reassuringly, these 3C methods have revealed general features of large-scale chromatin organization, discovered through decades of microscopy on specific genomic regions by FISH, but now demonstrated in higher throughput and genomic context. This includes folding of DNA into spatially segregated chromosome territories, proximity of arms from the same chromosome, clustering between centromeres and between telomeres, and clustering between transcriptionally active genome regions.

Several independent groups now have demonstrated distinct topological domains in the hundreds of kb to Mbp size range using 5C and Hi-C methods in Drosophila or mammalian systems [37-41]. 3C interactions occur at significantly higher frequency with other sequences lying in these same "topological associated domains" (TADs) [39] relative to sequences lying outside these domains (Fig. 1).

To date the highest resolution Hi-C studies have been in Drosophila, given its ~20 fold smaller genome size. Two studies, one using mixed embryonic nuclei from different developmental stages and lineages [32] and another using Kc tissue culture cells [40], each describe the delineation of roughly 1000 domains, with median and mean values of 60 or 62 kb and 107 or 100 kb, respectively. Half the domain boundaries are the same in both studies with the overall domain pattern even more similar. Thus a large fraction of these domains are common to a range of cell types. This similarity in domain structure is striking, given the differences in starting material and the different computational methods used to extract domain boundaries from still noisy data sets.

These topological domains largely demarcate regions enriched for either repressive or active epigenetic chromatin classifications, as defined by cluster analysis of protein binding and histone modification genomic patterns [42,43]. Larger domains are biased heavily towards repressive epigenetic classes and smaller domains heavily towards transcriptionally active epigenetic classes. Domain borders are significantly enriched in marks of transcriptionally active chromatin, including H3K4me3, DNase I hypersensitivity sites, transcriptional start sites (TSSs) and overall gene density, rather than numbers of active genes. One study concluded that domain borders were determined by multiple insulator sites, particularly those combinations bound by multiple insulator proteins including CP190, that may or may not be associated with transcriptionally active sites [37]. The other study concluded that domain boundaries are particularly enriched in insulators containing sites for three or four of the insulator protein types (+/− Su(Hw)) combined with bound RNA pol II and TSSs [40].

Several Hi-C and 5C studies in mammalian cells strikingly have described quite similar topological domain structure [38,39,41] with median size of ~900 kb. A large fraction of these are conserved between mouse and human and invariant during development (Fig. 2A-B). As in Drosophila, domain boundaries are enriched in insulator (CTCF) binding sites, active promoters, gene bodies, transcription start sites, engaged RNA pol II, housekeeping genes, tRNA genes, and active histone marks (Fig. 2D). Again, domain boundaries appear to demarcate chromatin domains varying in replication timing, lamin B binding, or distinctive histone modifications [38,39]. However, these chromatin domain types can change during development and can be perturbed by mutation or knockdown without changing the TAD domain organization [39].

Fig. 2. An incomplete list of important biological features of 3C defined topological domains.

(A) Conservation of topological domains- comparison of mouse versus human domain organization over syntenic region. Reprinted from Fig. 3g-h, reference [38], Copyright 2012 Nature. (B) Preservation of topological domain organization during differentiation- comparison of domain organization in undifferentiated ES cells versus neuronal progenitor cells. Topological domains are fixed, while lamin associated domains (LADs) change organization. However, topological boundaries appear to mark boundaries of LADs and other epigenetic domains. (C) In topological domains identified in one 5C study, genes within topological domains show higher correlation of gene expression than genes lying in different domains. Reprinted from Fig. 3 (B) and Fig. 4 (C) from reference [39], Copyright 2012 Nature. (D) Topological boundaries are enriched in marks for active chromatin and transcription. Reprinted from Fig. 4a, reference [38], Copyright 2012 Nature.

Therefore many TAD boundaries may act as constitutive boundaries to spreading of chromatin states. Across several neighboring TADs identified by 5C, expression of genes with promoters within a single TAD correlated better than expression of genes in different TADs [39] (Fig. 2C). Focusing on putative enhancer / promoter interactions within single TADs identified a lncRNA likely involved in X chromosome inactivation regulation [39]. More generally, mapping TAD structure may facilitate identification of functional long-range interactions important in gene regulation. Dissecting determinants of TAD organization and long-range interactions within TADs is beginning, leading to suggestions that constitutive, Mbp size TADs are demarcated by CTCF/cohesin while Mediator and cohesin bridge enhancer/promoter interactions over shorter distances [41].

The Surprising

These new Hi-C and 5C results clearly direct a renewed focus on large-scale chromatin architecture. Yet, conceptually there remains a large contrast between the richness of 3C interaction data versus our understanding of how to relate these interaction frequencies to actual large-scale chromatin structures and dynamics. Normalization schemes now correct for technical biases- for instance, varying ligation rates as a function of restriction fragment length and nucleotide composition [44]. A more general correction for "visibility" to Hi-C interactions across the genome corrects for unknown biases that vary over the 200kb-1Mbp bin sizes used now for Hi-C [45].

Less tractable are questions related to variations in fixation efficiency across the genome and, even more basic, whether detected 3C interactions reflect molecular (~1-10 nm) versus cytological (100-1000 nm) proximity. For instance, DNase I HS sites show elevated 3C interaction frequencies [44,46]. But they also show lower cross-linked protein after formaldehyde fixation. Indeed, the FAIRE genomic method for identifying HS sites is based on HS sites purifying as "naked DNA" in a standard phenol / chloroform DNA extraction [47,48]. Likewise, genomic interactions detected by 3C methods are nearly always conceptualized as the result of ligation of two DNA fragments self-associating in vivo through a direct molecular interaction. Yet formaldehyde concentrations much lower than used in 3C protocols result in extensive nucleosome cross-linking within and between chromatin fibers [49], raising the possibility of large networks of cross-linked chromatin fragments.

Indeed, Razin and colleagues now have shown that 3C signals are predominately, if not exclusively, generated from the insoluble fraction of the "lysed nuclei" fraction (Fig. 3A) [50]. This insoluble fraction is comprised of moderately swollen nuclei with basic architectural features- nucleoli, SC35 speckles, heterochromatic domains, and chromosome territories- largely preserved at the light microscope level (Fig. 3C) but severely perturbed as seen by TEM (Fig. 3D). Using 3C interactions within the mouse beta-globin locus as their assay, nearly all long-range 3C interactions were demonstrated as arising from the insoluble fraction, with disruption of nuclear structure by sonication decreasing these interactions proportionally. These results raise the intriguing possibility that 3C interactions might measure cytological rather than molecular proximity (Fig. 4).

Fig. 3. 3C Interactions predominately derive from insoluble fraction corresponding to swollen nuclei.

(A) Comparison of 3C interactions between beta-globin promoter region and flanking regulatory sequences in beta-globin locus from soluble (dotted lines) versus insoluble (solid lines) fractions. Data is normalized for DNA amount in each fraction. Red lines correspond to fetal liver where gene is active, while blue lines correspond to brain where gene is inactive. 3C peaks from known long-distance interactions in beta-globin locus derive from insoluble fraction containing swollen nuclei. (B) Nuclear structure immediately after fixation (top) versus just prior to 3C ligation step (previously assumed to act on soluble, small, DNA-protein molecular complexes). Nuclei enlarge with SDS and other treatments prior to ligation step, chromosome territories swell, but nuclear bodies such as nucleoli, SC35 domains, and patches of epigenetically marked chromatin domains are still recognizable. DNA (DAPI) staining blue, immunofluorescence or FISH signals red. (C,D) TEM images of nuclei after fixation (C) or after fixation but just prior to ligation step of 3C procedure (D)- 3C procedures result in obvious disruption of condensed chromatin masses in fetal liver nuclei. Scale bars = 5000 nm (B), 1000 nm in (C-D). Reprinted from Fig. 2C (A), Fig. 3 (B), and Fig. 4 (C-D) from reference [50], Copyright 2013 Nucleic Acids Research.

Fig. 4. Do long-range cis and trans 3C interactions imply molecular versus cytological proximity?

(A-C) Conventional models interpret 3C interactions as demonstrating molecular proximity. Top: After fixation (A), chromosomes (dark blue) within nuclei (light green) are digested and solubilized by restriction enzyme digestion (B), followed by dilution of soluble fraction prior to ligation to favor intramolecular ligation events (C). Bottom: DNA strands from distant chromosome sites (blue or red) are brought into proximity through specific binding of trans factors (green and purple) binding to different cis elements; this proximity is fixed by formaldehyde cross-linking of trans and cis factors (A). Restriction digest cuts restriction sites (black lines) (B) which then ligate DNA fragments together from different genomic regions (C). (D-F) Alternative model suggests 3C interactions correspond instead to proximity over cytological rather than distances. Top: Interphase chromosome structures (blue) after fixation (D) are disrupted progressively during 3C procedure. Restriction digest (E) results in fragments that remain cross-linked to insoluble, cross-linked nuclear structures and ligation (F) occurs within these partially disrupted chromosome residual structures. Bottom: Chromosome regions (red and blue) adjacent to nuclear bodies (brown) are densely cross-linked by formaldehyde fixation (D). SDS treatment and heat prior to restriction digest denatures histone and nonhistone proteins (purple) that act as linkers creating a protein-DNA network anchoring restriction fragments after restriction digest (E). Ligation events (black lines) connect fragments over large distances within this disordered network (F). In this model, long-range 3C interactions might derive for instance from co-association of two gene loci to the same nuclear body, even if they are separated in the live cell by cytological (100-1000 nm) distances.

The Still Perplexing

Many examples now exist in which 3C interactions bridging LCR, enhancer, and/or promoter sequences depend on the recruitment of specific transcription factors or co-factors to these cis-regulatory sequences. Enhancer-promoter looping is the dominant model for enhancer function. The demonstration of transcription factor dependent 3C interactions connecting enhancer and promoter elements therefore appears to confirm this looping model while demonstrating the molecular origin of 3C interactions.

An excellent example is the 3C interaction between DNase I HS site 2 of the beta-globin LCR and the beta-globin promoter [51]. This interaction is dependent on the Lbd1 protein and the hematopoietic-restricted factors KLF1, GATA1, and the GATA1 cofactor FOG1. Ldb1 is present at both the LCR and promoter and GATA1 is required for Lbd1 recruitment at the promoter. In one model, Ldb1 homo-dimerization or oligomer formation anchors a loop between the LCR and the promoter, generating a 3C detectable interaction [52]. In an elegant experiment, the GATA1 dependence of this 3C interaction was bypassed by targeted tethering of Lbd1 directly to the promoter via a zinc finger-Lbd1 fusion protein or the same zinc finger protein fused to the self-interacting domain of Lbd1 [52]. Transcriptional activation via these Lbd1 fusion proteins was within 20% of that seen after GATA1 activation.

And yet there also exist numerous published examples where validation of 3C interactions of very long-distance cis and trans interactions by FISH showed adjacent (~200-1000 nm) rather than overlapping signals. These adjacent but non-overlapping signals suggest cytological versus molecular interactions, connected by extensive cross-linking consistent with the origin of 3C interactions from the insoluble rather than soluble template. Such cytological interactions would also explain how genes on different chromosomes that co-localize at nuclear bodies, such as transcription factories, are identified by 3C interactions [53].

Thus there is strong evidence that 3C interactions connecting regulatory elements in trans or over very long-distance in cis arise from cytological rather than molecular interactions. Higher frequency 3C interactions over ~100 kb distances are most simply explained by molecular looping interactions, but this has not yet been clearly established by the high-resolution imaging approaches needed to distinguish molecular interactions from longer-range interactions generated by extensive cross-linking. Interestingly, in the beta-globin example, Lbd1 is also required for the migration of the beta-globin locus from the nuclear periphery to the interior [54]. One can imagine therefore an alternative model in which Lbd1 at both the LCR and promoter recruit the beta-globin locus to a transcription factory or nuclear speckle.

Regardless of the origin of shorter-range, enhancer-promoter interactions, the demonstration that many Mbp-scale 3C interactions arise from cytological rather than molecular interactions, and from the insoluble rather than soluble fraction, supports the idea that TADs may correspond to distinct large-scale chromatin folding motifs. Structural perturbations induced by the 3C protocol might even increase 3C interactions across the TAD, facilitating their detection.

Certainly the rough similarity between the ~1000 number of Drosophila TADs and the 3286 Drosophila polytene band/interband units estimated by electron microscopy is tantalizing [55]. Like TADs, polytene banding patterns are also highly conserved between different tissues [56]. The ~1000 TAD number is almost certainly an underestimate given the still limited Hi-C resolution. Indeed, internal structure and sub-TAD domains exist within larger TAD domains, as revealed more clearly by higher resolution 5C analysis [38,39,41].

Yet this analogy of TADs with polytene banding patterns raises several provocative questions. FISH of polytene chromosomes have generally always shown distinct stripes of target sequences within individual bands and interbands with no evidence for long-range looping interactions; also at best very few highly reproducible, long-range interactions between different chromosome regions were seen in polytene nuclei [57]. What then is the biological meaning of the very long-range (inter-TAD) and trans interactions detected by 3C, whose ligation frequencies are orders of magnitude lower than shorter-range 3C interactions? Most Drosophila larval tissues are polytene and carry out the vast majority of biological functions required of diploid nuclei. Might this argue for limited functional significance in diploid cells for these long-range interactions for most genes?

For example, while 3C long-range interactions connect the alpha-globin locus with distant chromosome sites, comparison of 22 species revealed an ~120kb kb syntenic region sufficient to drive normal developmental expression patterns [58]. Perhaps very long-range interactions rather have evolved to play highly specialized roles for a small number of more complex examples of gene regulation. An extreme example of the latter might be the mono-allelic expression selected from the thousands of olfactory receptor genes in mammals in which clustering of inactive alleles into heterochromatic foci in the nuclear interior may be key to their silencing [59]. A more typical developmental example might be the transient Mbp enhancer-promoter looping at the Shh locus seen in tissues where Shh expression is lower and more transient than in tissues regulated by short-range enhancers [60].

Alternatively, might these inter-TAD and trans interactions in diploid nuclei fine-tune gene regulation for a larger number of genes, either through small changes in gene expression or more robust regulation of expression levels? Here an interesting example might be the fine-tuning of expression of the haploinsufficient genes PTHLH (in cis) and Sox9 (in trans) by a cis-regulatory region 24 Mbp downstream of the PTHLH gene coding for a lncRNA, with loss of this fine-tuning leading to brachydactyly [61].

Most BAC transgenes express within several fold of endogenous genes in mammalian systems [62]. Those that don't would identify gene loci in which long-range interactions are essential. But exploring the basis for the several fold reduction in expression relative to endogenous genes would be one means to explore the fine-tuning role played by very long-range interactions.

Moving forward

Through their identification of topological domains, Hi-C, 5C, and other 3C approaches open new avenues, where none existed previously, for dissecting the molecular basis for long-range genome organization and gene regulation. A major goal for the future will be understanding how this topological organization revealed by 3C methods relates to actual large-scale chromatin ultrastructure directly visualized by high-resolution microscopy. Meanwhile a combination of sequence and gene knockdown manipulations should lead to rapid progress in unraveling the determinants of this level of genome organization and their functional impact on gene regulation.