In eukaryotes transcriptional regulation often involves multiple long-range elements and is influenced by the genomic environment¹. A prime example of complex cis-regulation concerns the control of Xist expression by the X-inactivation center (Xic). The Xic has been defined by deletions and translocations as a region spanning several megabases^2,3. However the extent of Xic sequences required for proper Xist regulation and X-chromosome inactivation (XCI) initiation remain unknown. Several elements in the vicinity of Xist are known to affect its activity, including the repressive antisense transcript Tsix and its regulators Xite, DXPas34 and Tsx^4,5. However, additional control elements must exist, as single-copy transgenes encompassing Xist and up to 460kb of flanking sequences are unable to recapitulate proper Xist regulation⁶. To characterise the cis-regulatory landscape of the Xic in an unbiased approach, we performed Chromosome Conformation Capture Carbon Copy (5C)⁷ across a 4.5Mb region containing Xist.

We designed 5C-Forward and 5C-Reverse oligonucleotides following an alternating scheme⁷, thereby simultaneously interrogating nearly 250,000 possible chromosomal contacts in parallel, with a mean resolution of 10–20 kb (Fig. 1a; see Methods). Analysis of undifferentiated mouse ES cells (mESCs) revealed that long-range (>50kb) contacts preferentially occur within a series of discrete genomic blocks, each covering 0.2–1Mb (Fig. 1b). These blocks differ from the higher-order organisation recently observed by Hi-C⁸, corresponding to much larger domains of open or closed chromatin, that come together in the nucleus to form A and B-types of compartments⁸. Instead, our 5C analysis reveals self-associating chromosomal domains occurring at the sub-megabase scale. The size and location of these domains is identical in male and female mESCs (Supplementary Fig. 1) and in different mESC lines (Supplementary Fig. 2, Supplementary data 1).

Open in a separate window

Figure 1

Chromosome partitioning into topologically assocating domains (TADs)

a, Distribution of 5C-Forward and 5C-Reverse HindIII restriction fragments across the 4.5 Mb analysed showing positions of RefSeq Genes and known XCI regulatory loci. b, 5C datasets from XY undifferentiated mESCs (E14), displaying median counts in 30kb windows every 6kb. Chromosomal contacts are organised into discrete genomic blocks (TADs A-F). A region containing segmental duplications excluded from the 5C analysis is masked (white). c, Positions of DNA FISH probes. d, Interphase nuclear distances are smaller for probes in the same 5C domain. e, Structured illumination microscopy reveals that colocalisation of neighbouring sequences is greater when they belong to the same 5C domain. Boxplots display the distribution of Pearson correlation coefficient between red and green channels, with whiskers and boxes encompassing all and 50% of values respectively; central bars denote the median correlation coefficient. Statistical significance was assessed using Wilcoxon’s rank-sum test.

We explored this organisation using 3D DNA FISH in male mESCs. Nuclear distances were significantly shorter between probes lying in the same 5C domain than in different domains (Fig 1c and d) and a strong correlation was found between 3D distances and 5C counts (Supplementary Fig. 3a and b). Furthermore, using pools of tiled BAC probes spanning up to 1Mb with structured illumination microscopy, we found that large DNA segments belonging to the same 5C domain colocalise to a greater extent than DNA segments located in adjacent domains (Fig. 1e), and this throughout the cell cycle (Supplementary Fig. 3c and d). Based on 5C and FISH data, we conclude that chromatin folding at the sub-megabase scale is not random, and partitions this chromosomal region into a succession of topologically associating domains (TADs).

We next investigated what might drive chromatin folding in TADs. We first noticed a striking alignment between TADs and the large blocks of H3K27me3 and H3K9me2⁹ that are known to exist throughout the mammalian genomes^10,11,12,13 (e.g. TAD #E, Fig. 2 and Supplementary Fig. 4). We therefore examined 5C profiles of G9a^{−/ −} mESCs, which lack H3K9me2, notably at the Xic¹⁴, and Eed^−/− mESCs which lack H3K27me3¹⁵. No obvious change in overall chromatin conformation was observed, and TADs were not affected either in size or position in these mutants (Fig. 2, Supplementary Fig. 4b). Thus TAD formation is not due to domain-wide H3K27me3 or H3K9me2 enrichment. Instead, such segmental chromatin blocks might actually be delimited by the spatial partitioning of chromosomes into TADs.

Open in a separate window

Figure 2

Determinants of topologically associating domains

a, Blocks of contiguous enrichment in H3K27me3 or H3K9me2¹¹ align with the position of TADs (Chip-chip from⁹) but TADs are largely unaffected in the absence of H3K9me2 in male G9a^{−/ −} cells or H3K27me3 or in male Eed^{−/ −} cells. b, Deletion of a boundary at Xist/Tsix disrupts folding pattern of the two neighboring TADs.

We then addressed whether folding in TADS is driven by discrete boundary elements at their borders. 5C was performed in a mESC line carrying a 58kb deletion (ΔXTX¹⁶) encompassing the boundary between the Xist and Tsix TADs (#D and #E; Fig. 2b). We observed ectopic contacts between sequences in TADs #D and #E and an altered organisation of TAD #E. Boundary elements can thus mediate the spatial segregation of neighboring chromosomal segments. Within the TAD #D-#E boundary, a CTCF site was recently implicated in insulating Tsix from remote regulatory influences¹⁷. However, alignment of CTCF and Cohesin binding sites in mESCs¹⁸ with our 5C data revealed that although these factors are present at most TAD boundaries (Supplementary Fig. 4), they are also frequently present within TADs, excluding them as the sole determinants of TAD positioning. Furthermore, the fact that the two neighboring domains do not merge completely in ΔXTX cells (Fig. 2b) implies that additional elements, within TADs, can be usurped when a main boundary is removed. The factors underlying an element’s capacity to act as a canonical or shadow boundary remain to be investigated.

Next we asked whether TAD organisation changes during differentiation or XCI. Both male neuronal progenitors (NPCs) and male primary mouse embryonic fibroblasts (MEFs) show similar organisation to mESCs, with no obvious change in TAD positioning. However, consistent differences in the internal contacts within TADs were observed (Fig.3a, Supplementary Fig. 2). Noticeably, some TADs were found to become lamina-associated domains (LADs)¹⁹ at certain developmental stages (Fig. 3b). Thus chromosome segmentation into TADs reveals a modular framework where changes in chromatin structure or nuclear positioning can occur in a domain-wide fashion during development.

Open in a separate window

Figure 3

Dynamics of topologically associating domains during cell differentiation

a. Comparison of 5C data from male mESCs (E14), NPCs (E14) and primary MEFs reveals general conservation of TAD positions during differentiation, but differences in their internal organisation (arrows highlight examples of tissue-specific patterns). b Lamina associated domains (LADs, from ref.¹⁹) align with TADs. Chromosomal positions of tissue-specific LADs reflect gain of lamina association by TADs, as well as internal reorganisation of lamina associated TADs during differentiation.

We then assessed TAD organisation on the inactive X (Xi), by first combining Xist RNA FISH, to identify the Xi, and super-resolution DNA FISH using BAC probe pools (as in Fig. 1e) on female MEFs. We found that colocalisation indices on the Xi were still higher for sequences belonging to the same TAD than to neighbouring TADs. The difference was however significantly lower for the Xi than for the Xa (Supplementary Fig. 5). Similarly, deconvolution of the respective contributions of the Xa and Xi from 5C data female MEFs (see methods, Supplementary Fig. 5) revealed that specific long-range contacts within TADs are lost on the Xi, but global TAD organisation remains, albeit it in a much attenuated form. This, together with a recent report focused on longer-range interactions²⁰, suggests that the Xi has a more random chromosomal organisation than its active homolog.

We next investigated how TAD organisation relates to gene expression dynamics during early differentiation. A transcriptome analysis, consisting of microarray measurements at 17 time points over the first 84 hours of female mESC differentiation was performed (Fig. 6a). During this time window, most genes in the 5C region were either up- or down-regulated. Statistical analysis demonstrated that expression profiles of genes with promoters located within the same TAD are correlated (Fig. 4b). This correlation is significantly higher than for genes in different domains (correlation coefficient of 0.03, p<10⁻⁹) or for genes across the X chromosome in randomly selected, TAD-size regions (correlation coefficient of 0.09, p<10 ⁻⁷ ). The observed correlations within TADs seems not to depend on distance between genes, and is thus distinct from previously described correlations between neighbouring genes²¹ which decay on a length scale of ~100kb (Supplementary Fig. 6). Our findings suggest that physical clustering within TADs may be used to coordinate gene expression patterns during development. Furthermore, deletion of a boundary (ΔXTX) was accompanied by long-range transcriptional mis-regulation (Supplementary Fig.7), supporting a role for chromosome partitioning into TADs in long-range transcriptional control.

Open in a separate window

Figure 4

Transcriptional co-regulation within topologically associating domains

a, Female mESCs were differentiated towards epiblast stem cell lineage for 84 hours. Transcript levels were measured every 4–6 hours at 17 different time points by mircoarray analysis. b, Pearson’s correlation 15 coefficients over all time points were calculated for (1) gene pairs lying in the same TAD, (2) pairs in different TADs and (3) for pairs in randomly defined domains on the X-chromosome that contain a similar number of genes and are of comparable size. Box plots display the distribution of Pearson’s correlation coefficients, with whiskers and boxes encompassing all and 50% of values respectively, and central bars denoting the median correlation coefficient. * represents significant difference with p<10⁻⁷ using Wilcoxon’s ranksum test. c, Pearson’s correlation coefficients for gene pairs in TADs #D, #E and #F with red denoting positive and blue negative correlation. Boxes indicate the TAD boundaries.

A more detailed analysis of each domain (Supplementary Fig. 6) revealed that co-expression is particularly pronounced in TADs #D #E and #F (Fig. 4b and c). Although correlations are strongest within TADs, there is some correlation between TADs showing the same trend, such as TADs #D and #F, which are both down-regulated during differentiation. Only TAD #E, which contains Xist and all of its known positive regulators (Jpx, Ftx, Xpr, Rnf12, supplementary Fig. 4d)²² is anti-correlated with most other genes in the 4.5 Mb region, being up-regulated during differentiation (Supplementary Fig. 4c). The fact that these coordinately up-regulated loci are located in the same TAD suggests that they are integrated into a similar cis-regulatory network, potentially sharing common cis-regulatory elements. We therefore predict that TAD #E (~550kb) represents the minimum 5′ regulatory region required for accurate Xist expression, explaining why even the largest transgenes tested so far (covering 150kb 5′ to Xist, Fig. 5) cannot recapitulate normal Xist expression⁶.

Open in a separate window

Figure 5

5C maps reveal new regulatory regions in the Xic

(a) 5C maps at the restriction fragment resolutions (male E14 mESCs) show that Tsix and Xist promoters belong to two distinct neighbouring TADs, respectively spanning >200kb and >550kb (male E14 mESCs). b, Corresponding statistically significant looping events (5C peaks) for restriction fragments within Xite, Tsix promoter or Xist promoter within their respective TAD. The Tg80 YAC transgene lacks genomic elements found to interact physically with Xite/Tsix that are present in Tg53. c, RNA FISH analysis of Tsix Tsix expression is detected in the inner cell masses of heterozygous transgenic male E4.0 embryos by RNA FISH from single-copy paternally inherited Tg53 but not Tg80 transgenes. Transgenic (star) and endogenous Tsix alleles (arrowhead) were discriminated by subsequent DNA FISH as in supplementary Fig. 5. n=20 ICM cells (2 embryos each). d, Linx transcripts (green, wi1–1985N4 probe) are expressed in both E4.0 ICM cells and mESCs, together with Tsix (red, DXPas34 probe), and unspliced transcripts accumulate locally in a characteristic cloud-like shape. e, RNA FISH in differentiating female mESCs revealing synchronous downregulation of Linx and Tsix with concomitant up-regulation of Xist (detected with a strand-specific probe). Barrs are the standard deviation around the mean of three experiments. f, Triple-color RNA FISH allowing simultaneous detection of Linx, Tsix and Xist RNAs. Scoring of Xist-negative alleles demonstrates that prior to Xist up-regulation Tsix expression is more frequent from Linx-expressing alleles than from Linx non-expressing alleles, at all time-points tested. Presented is the mean of three experiments. Statistical differences were assessed using Fisher’s exact test. Cells were differentiated in monolayers by LIF withdrawal.

The Xist and Tsix promoters lie in two neighbouring TADs with transcription crossing the intervening boundary (Fig. 2b), consistent with previous 3C experiments²². While the Xist promoter and its positive regulators are located in TAD#E, the promoter of its antisense repressor, Tsix, lies in TAD#D, which extends up to Ppnx/Nap1L2, >200kb away (Fig. 2b). Thus, in addition to the Xite enhancer, more distant elements within TAD #D may participate in Tsix regulation. To test this we used two different single-copy transgenic mouse lines, Tg53 and Tg80²³. Both transgenes contain Xist, Tsix and Xite (Fig. 5a). Tg53 encompasses the whole of TAD #D, while Tg80 is truncated just 5′ to Xite (Fig. 5a; Supplementary Fig. 8). In the inner cell mass (ICM) of male E4.0 mouse embryos, Tsix transcripts could be readily detected from Tg53, as well as from the endogenous X (Fig. 5b). However, no Tsix expression could be detected from Tg80, which lacks the distal portion of TAD #D (Fig. 5b). Thus, sequences within TAD#D must contain essential elements for the correct developmental regulation of Tsix.

Within TAD#D, several significant chromosomal looping events involving the Tsix promoter and its enhancer Xite were detected (Fig. 2a and ​and5a,5a, Supplementary Fig. 9). Alignment of 5C maps with enhancer signatures (Supplementary Fig. 8), suggested the existence of multiple regulatory elements within this region. We also identified a transcript initiating ~50kb upstream of the Ppnx promoter (Fig. 5a), from a region associated with pluripotency factor binding and corresponding to a predicted promoter for a large (80kb) intervening non-coding RNA (LincRNA²⁴, Supplementary Fig.9) which we termed Linx (Large INtervening transcript in the Xic). Linx RNA shares several features with ncRNAs, such as accumulation around its transcription site²⁵ (Fig. 5c), nuclear enrichment and abundance of the unspliced form²⁶ (Supplementary Fig.11 and 12). Linx and Tsix transcripts are co-expressed in the ICM of blastocysts from E3.5–4.0 onwards, as well as in male and female mESCs (Fig. 5c). Linx RNA is not detected earlier on in embryogenesis, nor in extra-embryonic lineages, implying an epiblast-specific function (Supplementary Fig. 8). Triple RNA FISH for Linx, Tsix and Xist in differentiating female mESCs (Supplementary Fig. 13) revealed that prior to Xist up-regulation, the probability of Tsix expression from alleles co-expressing Linx is significantly higher than from alleles that do not express Linx (Fig. 5d). Furthermore, Linx expression is frequently monoallelic, even prior to Xist up-regulation (Supplementary Figure 13), revealing a transcriptional asymmetry of the two Xic alleles prior to XCI. Taken together, our experiments based on 5C, transgenesis and RNA FISH, point towards a role for Linx in the long-range transcriptional regulation of Tsix – either through its chromosomal association with Xite and/or via the RNA it produces. This analysis of the Xist/Tsix region illustrates how the spatial compartmentalisation of chromosomal neighborhoods partitions the Xic into two large regulatory domains, with opposite transcriptional fates (Supplementary Fig. 12).

In conclusion, our study uncovers that sub-megabase folding of mammalian chromosomes results in the self-association of large chromosomal neighborhoods The stability of these TADs throughout differentiation, X-inactivation and in cell lines with impaired histone-modifying machineries suggests that this level of chromosome organisation may provide a basic framework onto which other domain-wide features, such as lamina association and blocks of histone modification, can be dynamically overlaid. Our data also point to a role for TADs in shaping regulatory landscapes, by defining the extent of sequences that belong to the same regulatory neighborhood. We anticipate that TADs may underlie previously proposed regulatory domains based on functional and synteny conservation studies^27,28. We believe that the principles we have revealed here will not be restricted to the Xic, as spatial partitioning of chromosomal neighborhoods occurs throughout the genome of mouse and human (see Dixon et al. in this issue of Nature), as well as Drosophila²⁹ and E. coli³⁰,. Future work will clarify the mechanisms driving this level of chromosomal organisation, and to what extent it generally contributes to transcriptional regulation. In summary, our study provides new insight into the cis-regulatory architecture of chromosomes that orchestrates transcriptional dynamics during development, and paves the way to dissecting the constellation of control elements of Xist and its regulators within the Xic.