It is well established that the transcriptional activity of any given promoter is influenced by regulatory input coming from its chromosomal environment. In many instances remote control elements, such as distal enhancers, exert their regulatory effects across large genomic distances (up to hundreds of kb). This mechanism is thought to involve physical contact between these regulatory elements and their target promoters, which is permitted by the folding pattern of the chromatin fiber. Examples illustrating the diversity by which enhancers mechanistically stimulate promoter activity have been reviewed elsewhere , but many key aspects of this process remain enigmatic. How does an enhancer find its target amongst the many promoters that surround it? How can several distinct regulatory elements, sometimes scattered over hundreds of kb, all participate in the control of a given promoter? What is the extent of the chromosomal neighborhood involved in the control of a given promoter, and its possible dynamics during development? Below we discuss these issues in the context of the discovery of chromatin folding into TADs.
TADs: A structural basis for regulatory landscapes?
TADs are, by definition, chromosomal regions within which DNA sequences most often establish long-range physical contact. This means that any given promoter will be contacted most frequently by sequences belonging to its TAD. It is therefore expected that enhancers, and other regulatory elements that act through physical contact with target promoters, should mainly impact the promoters of the same TAD. For these reasons it is tempting to speculate that the TAD of a given promoter embodies its regulatory domain, meaning the chromosomal region containing its cis-regulating elements. In support of this notion, most if not all of the previously reported cases of very long-range (>500 kb) functional enhancer-promoter communications are found to occur within TADs [44, 45]. These include paradigmatic loci such as Sonic Hedgehog and the Lmbr1 intron (1 Mb away) .
TADs as guides in the traffic of cis-regulatory information
Applications of high-throughput 3C, such as Hi-C and 5C, have revealed that promoters can be contacted by a large set of sequences within their TAD. What can seem surprising is that promoters are not specifically contacted by enhancers, in general – as defined genetically or according to their epigenomic signature [12-15, 47] – no matter the activity state of these promoters or enhancers (active, poised, or inactive). Likewise, enhancers do not appear only to contact promoters. This means that a given promoter is typically not tethered to a given enhancer in a one-to-one relationship , but each of these elements is generally able to associate with (or at least transiently “visit”) large chromosomal domains [8, 12, 13, 47]. This is a very important observation that has far-reaching consequences for our understanding of enhancer-promoter communication.
Using state-of-the-art genetics and genomics approaches Duboule and colleagues  have recently revealed that a constellation of enhancers spread out over hundreds of kb are active in the same tissue and interact with the HoxD genes, leading the authors to propose the existence of a “regulatory archipelago” effect. This illustrates that a given promoter can be influenced by several distal regulatory elements within the same TAD (Fig. 3A). In another scenario these various enhancers can be active in distinct cell types. This is the case for the well-studied Sonic Hedgehog (Shh) locus, where various enhancers within the Shh TAD can be active in different tissues, depending on the trans-acting factors that bind and activate them .
Figure 3. The link between TADs and domain-wide transcriptional regulation. Folding into TADs fosters long-range transcriptional regulation by allowing distal sequences to frequently contact each other. Folding into TADs allows A: multiple regulatory sequences to target the same promoter and conversely, B: multiple promoters to be targeted by a given regulatory element. C: Spatial partitioning segregates neighboring regulatory domains, allowing juxtaposed clusters to assume opposite transcriptional fates upon response to a stimulus. D: Activation of a regulatory element within a TAD can have minor yet measurable effects on secondary promoter targets, possibly explaining ripple effects of transcriptional activation. E: Elements controlling chromatin architecture or transcriptional activity can be distinct. When separate, architectural elements will control access of the transcription-controlling element to its target promoter, thereby playing an indirect but nonetheless integral role in the regulation of transcriptional activity.
Download figure to PowerPoint
The correlate of the regulatory archipelago effect is that a given enhancer will frequently visit several promoters within the same TAD (Fig. 3B). Sharing of regulatory elements within TADs could for example explain the broad yet sharply delimited radius of action of global control regions, such as the ones ensuring the co-expression of the various members of the two mouse Iroquois clusters or the proximal HoxD/Evx2/Lnp domain (for review see ). We anticipate that the topological segmentation of chromosomes, by demarcating the chromosomal range that is accessible to an enhancer, both guides and constrains enhancer-promoter communication within TADs, thereby driving the preferential allocation of enhancers to the promoters that belong to the same TAD. Such an idea is supported by the situation at the mouse HoxD cluster, where proximal regulatory elements influence the proximal HoxD members. These lie together in the same TAD, but not with the distal HoxD members, which lie in a different TAD together with their own distal regulatory elements [12, 48].
Enhancers are typically active in a tissue-specific fashion. How does this property relate to TAD dynamics during development and cell differentiation? Maybe surprisingly, TAD positioning along chromosomes remains largely invariant during cell differentiation [12, 13]. However, subtle but reproducible variations in the internal conformation of TADs occur, and these changes depend on the cell type. These key observations suggest that stable segmentation into TADs defines core landscapes, within which tissue-specific interactions can arise during the time-course of development, most likely depending on which trans-acting factors bind to the genomic elements involved in these connections. The mouse HoxD locus illustrates how various sets of enhancers, active in different tissues at different stages of development, all use the same core TAD architecture to convey their regulatory input in a cell-type specific fashion . In other cases, TAD folding may allow a common regulatory element to be used by several distinct promoters, depending on the developmental stage. The chromatin conformation inside the TAD or tissue-specific trans-acting factors expressed at a particular time, may ensure that only one promoter is engaged at a given time with the enhancers. Such developmental switching of promoter usage by a regulatory element is, for example, illustrated by the well-studied β-globin cluster, where the locus control region (LCR) and its targets lie in the same TAD (see  for review).
Based on this it appears that sharing of regulatory elements within TADs may in some cases lead to transcriptional coordination of genes located throughout Mb-wide chromosomal domains. TADs also facilitate the physical isolation of groups of genes and regulatory elements from their neighbors that assume different transcriptional dynamics upon response to a given stimulus (e.g. during cell differentiation, hormone response, etc. (Fig. 3C)). Such a phenomenon has been reported for the TADs that split the mouse X-inactivation center into several genomic domains which are oppositely regulated during early mouse ES cell differentiation .
Given the relative pervasiveness of chromosomal contacts within TADs, how can we explain that a given enhancer is able to find its specific promoter target? Actually, we would argue that there is little evidence showing that, in their native genomic context, distal enhancers are generally able to distinguish promoters in order to exert their control over a specific target. Rather, several lines of evidence suggest that within TADs, enhancers can exert broad regulatory effects that are often loosely targeted. A consequence of this is that rapid activation of control elements – for example during exposure to a given stimulus – can lead to extensive pervasive regulatory spills. This situation is exemplified by the ripple effects on transcriptional activation that are observed upon cell stimulation with growth-factors , where rapid transcriptional induction at one locus leads to activation of several neighboring promoters. We anticipate that promoters lying in the same TAD will tend to be affected by such ripple effects in situations when dynamic changes of transcriptional patterns occur, such as during stress or hormone response (Fig. 3D). This potential for co-regulation may be exploited in certain cases to ensure coordinated transcription patterns of clustered genes.
Intra-TAD architecture and promoter-enhancer regulatory contacts
What is the structural basis for the communication between regulatory elements and promoters within TADs? A common textbook representation of enhancers is that they come into contact with promoters thanks to extensive looping out the intervening sequences. As recently discussed in depth by Fudenberg and Mirny , the detection of frequent contacts in a population of cells (i.e. peaks of 3C signal) actually does not necessarily imply the existence of such a looped-out configuration. This sort of signal can also be explained by the existence of a vast ensemble of chromatin configurations amongst the cell population, where the intervening DNA actually adopts compact but highly variable conformations without necessarily having to be extensively looped out . Such topological arrangements would actually agree better with FISH and polymer modeling data than looped-out configurations. In this situation intervening sequences are still excluded from the enhancer-promoter contact, but the overall conformation remains compact and, importantly, is very different from cell to cell. This has important consequences on how we envision transcriptional regulation. For example, it predicts extensive variability in local chromosome conformation, raising the question of how this relates to transcriptional variability and noise – possibly generating extensive “spatial effect variegation” (for review see ) – either between cells or across time.
The apparent specificity that accompanies the preferential engagement of an enhancer with a given promoter may therefore rely on their inherent ability to maintain this contact, even temporarily, rather than to establish it. The sensitivity of 3C-based assays, which sample millions of cells, may reveal a tendency in the cell population but detection of an association does not necessarily reflect the existence of a stable chromosomal structure. This also implies that the structural mechanisms that control chromatin conformation, and the probability of enhancer-promoter encounters, act upstream of promoter-enhancer regulatory communication. It also means that genomic elements (and the factors bound to them) controlling chromatin architecture can be distinct from the factors involved in modulating the actual process of transcription (Fig. 3E).
Given these considerations, it might therefore be expected that disrupting folding in TAD, either by genomic alterations or changes in the epigenomic makeup, can lead to redirection of regulatory influences. Indeed, deleting a TAD boundary at the Xist/Tsix locus has been reported to impair spatial insulation, leading to regulatory bleed-through from one domain to the other and transcriptional mis-expression . Intriguingly, it has been reported that a small fraction of TAD boundaries are cell-type specific [12, 14]. We speculate that in these cases the gain or loss of TAD-boundary activity may be a strategy to expose promoters to new large regulatory landscapes during development, something that could be tested by deleting such facultative boundaries or replacing them by stable boundaries (see “Note added in proof”).
These considerations also predict that altering the internal organization of TADs that contain enhancers that do not have an intrinsic specificity for their promoter target should alter the orchestration of long-range regulation. A compelling example of this is that shuffling genomic organization at the Fgf8 locus is sufficient to redirect underlying enhancers to different promoter targets . Another example is that tandem duplication within the TAD containing the proximal regulatory region of the mouse HoxD locus prevents distal regulatory elements from accessing and properly regulating the HoxD cluster, without creating a new TAD boundary .
Enhancers are not the only genomic elements that use chromatin architecture to convey regulatory information from afar. For example, some non-coding RNA loci also appear to rely on a similar process to modulate transcription. Human HOTTIP produces an RNA that partners with chromatin-modifying enzymes and regulates transcription of the portion of the HOXA locus that lies in the same TAD . Importantly, interfering with HOTTIP expression does not affect the topology of the locus, again suggesting that organization in TADs is controlled by mechanisms that appear to act upstream of transcription. One interesting case of a non-coding RNA concerns the Xist transcript, that can modify chromosome architecture by randomizing chromosomal contacts on the inactive X chromosome [13, 58] – see Box 1.
Altogether these findings suggest that rather than being highly specific of a target promoter, enhancers should be considered as elements that can come into contact with a wide set of sequences, including but not limited to their target promoters, thanks to the underlying chromosomal architecture. Their radius of action would therefore be determined by the mechanisms that control three-dimensional chromatin organization. It is expected that long-range regulatory contacts within TADs will depend on the kinetics of chromatin diffusion. Determining the actual frequency and duration of these contacts (as discussed by ), as well as the physical parameters that govern them , and how these integrate with the actual act of transcription and its enhancement, should shed important light onto the mechanisms of long-range transcriptional regulation.
Future work will be needed to elucidate the mechanistic rules that underlie how architectural elements control the probability of enhancer-promoter contacts (Fig. 3E). Understanding how architectural elements have participated in the evolution of phenotypic diversity by their broad impact on the flow of cis-regulatory information, and to what extent they are implicated in human disease are other topics of interest. This will be the subject of our last section.
A link between chromosome folding and genome evolution
It is well known that the presence of cis-regulatory elements lying a long way from their targets can lead to evolutionary constraints that would select against the breaking of their synteny . This results in maintenance of linkage for the whole regulatory block, including bystander loci that are not necessarily under the control of the underlying regulatory elements, simply because simple recombination cannot disentangle them from the rest of the domain . Given the observation that, at least in some cases, TADs provide a structural basis for such cis-regulatory landscapes , we anticipate that loci within the same TAD will have a tendency to be syntenic across species. Preliminary inspection of available data on conserved synteny blocks  supports this idea, and this could now be explored further quantitatively (Fig. 4). This phenomenon would also explain the overall conservation of the position of TAD boundaries between mouse and human . We hypothesize that such counter-selection of synteny breaks within TADs underlie a modular evolutionary pattern of the genome, where groups or TADs – or even single TADs – on a given chromosome in a given species would correspond to a similar arrangement on another chromosome in another species (Fig. 4). Cases of evident synteny loss within a TAD are also of interest as they may be linked to the appearance of species-specific expression patterns, whereby a single recombination event would have broken or reshuffled the underlying cis-regulatory landscape. Such a rearrangement would lead to dramatic and instantaneous rewiring of the cis-regulatory network. We therefore speculate that such “cis-ruption”  of TADs, when advantageous, may represent a mechanism of saltational phenotypic divergence (see [64, 65] for insightful reviews on the theme of cis-regulatory evolution).
Figure 4. TAD-driven Mb-wide synteny. Synteny breaks within TADs would be expected to be counter-selected in general because they would disrupt underlying cis-regulatory connections. Such a phenomenon would lead to synteny of large chromosomal regions corresponding to groups of TADs, or even single TADs, with macrosynteny breaks occurring close to their boundaries. Expansion or retraction of these macrosyntenic regions can be observed, so that syntenic TADs or groups of TADs do not necessarily have the same genomic size in different species. The example shown here is illustrative.
Download figure to PowerPoint
Such considerations also have implications for the genomic (or epigenomic) features that may define given sequences to act as TAD boundaries. For example, the observation that promoters of house-keeping genes are often found close to TAD boundaries has been proposed to reflect a possible mechanistic explanation for boundary formation and topological insulation. An alternative – and non-exclusive – hypothesis is that housekeeping genes would tend to become positioned close to TAD boundaries during evolution because their presence within TADs would interfere with their constitutive expression and might also perturb the proper functioning of long-range regulatory landscapes. The same reasoning holds for SINE elements or even CTCF binding sites, as well as for the reported tendency of P-element transgenes to insert close to TAD boundaries in Drosophila . One can envision several ways for testing these possibilities, such as removing a boundary-associated feature and measuring whether this disrupts topological insulation (as has been shown in one case at the Xist/Tsix locus ); or else analyzing the effects of placing an ectopic feature (SINE, housekeeping gene, transgene, …) within a TAD on proper long-range regulation.
In the context of such considerations it seems highly likely that segmental folding of chromosomes has an impact on genome evolution, mainly because of the evolutionary pressure exerted by the cis-regulatory landscapes it helps to create. Mapping of TADs in diverse species will open the possibility to further explore this exciting avenue.