The Shh Topological Domain Facilitates the Action of Remote Enhancers by Reducing the Effects of Genomic Distances

1A substantial fraction of gene regulatory elements lie at considerable distance from the nearest promoters. While the contribution of these elements to gene expression is generally difficult to estimate, enhancers located hundreds of kilobases from their target genes but essential to their expression are increasingly identified (reviewed in, ). Accordingly, mutations or genetic variants in distant enhancers are a significant cause of genetic diseases and contribute to intra-species and inter-species phenotypic variability. Although our understanding of regulatory elements has improved tremendously in recent years, it remains unclear how enhancers find a specific target located several hundred kilobases away. There is strong evidence that such interactions require physical proximity. Yet, how this proximity is established and regulated and how it influences target gene expression is still poorly understood.

2Concomitant with the growing appreciation of distant regulatory sequences, improved chromosome conformation capture techniques have provided insights into the three-dimensional organization of the genome and cis-interaction networks between genes and surrounding elements. These approaches have revealed not only loops between distant elements but also that mammalian genomes are partitioned into sub-megabase-sized domains referred to as topologically associating domains or TADs. Several indirect lines of evidence suggest that these self-interacting regions may represent the core units of genome regulatory architecture; a large proportion of TAD boundaries are shared between cell types and largely preserved during evolution. Coordinately regulated tissue-specific enhancer-promoter pairs and associated long-range looping interactions are usually comprised within TADs. The regulatory domains defined by enhancers' range of action coincide also largely with TADs. Although internal interactions within TADs can be cell-type specific and activity dependent, these different findings support the role of TADs as basic structural and functional units.

3Correlations between regulatory and structural subdivisions of the genome suggest that TADs may constrain the range of action of enhancers, with TAD boundaries acting as functional “insulators”. TAD boundaries are indeed enriched for elements shown to have insulator activity (such as CTCF binding sites and transcriptional start sites), and insertions of a sensor gene on opposing sides of TAD boundaries show distinct expression patterns. Recent experiments deleting or altering these boundaries showed expansion of chromosomal contacts across the former boundaries, leading to ectopic activation of neighboring genes. Similarly, the consequences of multiple human pathological chromosomal rearrangements can be explained by modification of TAD boundary positions and subsequent enhancer adoption by non-target genes. Together, these experiments clearly established that TAD boundaries are essential for generating isolated domains of regulatory activities. However, other features and potential roles of TADs remain poorly studied.

4The Shh locus constitutes an ideal system to study long-range enhancer-promoter regulation. Shh expression is regulated by a series of tissue-specific enhancers distributed across a region spanning over 900 kb, which also comprises other unrelated genes and which corresponds to a tissue-invariant and evolutionary conserved TAD. In particular, the specific expression of Shh in the zone of polarizing activity (ZPA), which establishes antero-posterior patterning of the developing limbs, is fully determined by the activity of a single cis-acting enhancer. This element, the ZRS, lies 850 kilobases away from the Shh promoter, in an intron of an unrelated gene, Lmbr1. In the present work, we took advantage of this prototypic enhancer-promoter pair to study the relationship between distant enhancer-promoter interactions, 3D conformation, and gene expression. We generated a series of mouse strains carrying tagged and structurally rearranged alleles of this locus. We analyzed them in vivo, when the mechanisms associated with its regulation are functional and biologically relevant, and in situ, in the genomic context where they evolved and normally operate. Our results showed that enhancer-promoter loops occurred within the framework of much more promiscuous contacts, where enhancers scan the entire topological domain they are part of. Remarkably, altering enhancer-promoter distances in the context of the Shh TAD did not appear to affect Shh expression. In contrast, disruption of the TAD prevented physical and regulatory interactions between Shh and its limb enhancer, unless the genomic distance between the two was significantly reduced. Our observations provide evidence that TADs ensure high contact frequency between distant elements by counteracting the effect of genomic distances. TADs do not simply restrict enhancer activity to a specific region to prevent ectopic interactions. They also provide the spatial proximity that is essential for efficient action of remote enhancers on genes located within the same TAD. This regulatory role of TADs can be particularly important in large genomes and may have enabled expansion of the genomic space available for regulatory innovation during evolution.

5We had previously shown that insertions of a regulatory sensor at the Shh locus show a Shh-like expression pattern, and reveal a large regulatory domain that overlaps with the TAD at this locus. To get further information on how Shh enhancers act within this domain, we generated additional insertions of the regulatory sensor and analyzed its expression in mouse embryos at stages E10–12 (Figures 1A and S1, Table S1). The comparison of the patterns observed with 59 different insertions across the Shh genomic region provides a fine-scale view of its regulatory architecture, extending our first observations and those performed with a different promoter. Noteworthy, like at other loci, the insertion of this naive sensor did not alter the expression of Shh or surrounding genes (Figures S1C and S1D), indicating that the activity of the sensor does not trap enhancers away from Shh. Instead, the sensor reveals the pre-existing potential of surrounding enhancers to act on a given genomic position.

6We found that in the region beginning 33 kb downstream of Shh and extending to the ZRS, most insertions showed expression patterns that closely matched Shh expression in the limb (Figure 1A) as well as in other tissues (Figure S1). Outside of this Shh regulatory domain, insertions showed no expression or a divergent one. The Shh expression patterns detected by insertions in the Shh regulatory domain included domains for which enhancers have been mapped (Figure S1), as well as domains for which no enhancers have been identified to date (e.g., choroid plexus; Figure S1G). This widespread responsiveness indicated that most Shh enhancers can act long range and not only in their vicinity or close to the Shh promoter. We found that expression of the sensor at a given position was highly reproducible, both when comparing littermate embryos and in independent replicate experiments (Table S1). Yet, lacZ stainings of neighboring insertions can sometimes differ extensively, even when only a few kilobases apart (Figures 1A and S1). A small number of insertions within the regulatory domain, such as insertion 5.2, showed no expression in any tissue. But more typically, variation was quantitative and differed depending on the tissue. For example, at position 5.1, we observed robust expression in the notochord and floor plate, but only weak staining in the limb and in the genital bud; at position C1, we observed the reverse relative intensities (Figures S1B, S1E and S1F). The reporter insertion at position −33 showed high responsiveness to the ZRS but not to other enhancers (Figure S1). We also observed this quantitative variability at stages other than E11.5 (Figures S1G and S1H). Our data show that, for the same promoter, the responsiveness to enhancer(s) can vary extensively within an otherwise largely permissive regulatory domain.

7To understand what factors modulate responsiveness to regulatory inputs, we focused on the limb where Shh expression is determined by a single enhancer, in contrast to many other tissues where it is associated with several enhancers with overlapping activities. Critically, similarly to other tissues, the sensor showed significant variability in ZPA expression at different positions within the Shh TAD (Table S2). This indicated that variability in responsiveness is not limited to complex situations involving multiple enhancers. Responsiveness to the ZRS showed no correlation with linear distance to the ZRS nor did it appear to be influenced by the orientation of the sensor (Fisher exact test p = 0.387) or local chromatin features (proximity to repeat elements such as LINEs or SINEs; accessibility measured by DNaseI hypersensitivity or chromatin acetylation/methylation) (Table S2, data not shown) signifying that it is determined by other factors.

8As noted before, the Shh regulatory domain shows strong overlap with an underlying TAD, conserved in different cell lines (Figure 1B). Since 3D conformations can vary between cell types, we performed chromosome conformation capture sequencing experiments (4C-seq) on the posterior compartment of E11.5 microdissected limb buds (Figures 1B and S2). We chose viewpoints within the TAD (Shh promoter, ZRS, and Rnf32) and outside (Rbm33, Nom1). In the posterior limb, the three viewpoints located in the Shh TAD showed prominent contacts along the entire TAD, while contact frequency with regions outside the TAD decreased quickly (Figure 1B). Reciprocally, the viewpoints located immediately outside the TAD, on either side, showed limited interactions with sequences in the Shh TAD (Figure 1B). Both the ZRS and Shh had contact frequencies that remained constantly high throughout the TAD, with limited effect of genomic distance (Figures 1C and 1D). Together, these data show that the Shh-ZRS region forms a self-interacting chromatin domain in the posterior limb bud, which corresponds well to the TAD described in other cell types.

9Even among the generally robust interactions detected along the TAD, the ZRS showed a particularly stronger interaction with the Shh promoter in the posterior limb bud (Figure 2A), in agreement with previous 3C and FISH data. Interestingly, based on 4C-seq, the compartmentalization of interactions and the fine-scale interactions of the ZRS did not appear very different between E11.5 posterior limbs (where Shh and the ZRS are active) and E11.5 anterior and medial forelimb samples (where Shh and the ZRS are inactive) (Figures 1C, 1D, and S2); the ZRS showed stronger contact with Shh in all limb compartments (Figure 2A), although the interaction peaks appeared more diffuse in the inactive situations than in the ZPA. Hence, similarly to other loci, conformation and enhancer-target gene contacts appear to be in part constitutive and independent of transcriptional and regulatory activity.

10To understand if responsiveness of the regulatory sensor to the ZRS was related to the native folding of the locus, we compared the interaction profile of the ZRS (from wild-type limb, without sensor insertions) to the expression of the sensor in the ZPA at the different insertion sites (Figures 2B and Table S2). We found that, within the Shh TAD, positions with ZPA expression had overall stronger contacts with the ZRS than weakly or non-expressed ones (Figure 2C). This correlation indicates a relationship between the distribution of enhancer activity and the native structural folding of the locus. It should be noted, however, that some positions contacted by the ZRS with similar efficiency (as measured by 4C, at a resolution of 5–10 kb) displayed different activation potential, indicating that average contact frequency is not the sole determinant for regulatory activation.

11To further identify the mechanisms that govern distant interactions, we decided to systematically change different genomic parameters of the locus. First, we modulated the genomic distance separating the ZRS from Shh, while respecting the TAD boundaries. To this end, we engineered mice carrying either intra-TAD deletions or duplications, using Cre-loxP in vivo recombineering (Figures 3A and 3B). We then assessed limb morphology and Shh expression in animals carrying these rearrangements over a Shh null allele (Shhdel).

12In agreement with ZRS deletion alleles, deletions that included the ZRS completely abolished limb expression of Shh (Figures 3C and S3A) and led to fore- and hindlimb monodactyly (Figures 3D and S3). We also observed loss of limb expression of the inserted sensor gene retained in DEL(C1-Z), showing that the remaining region comprised no limb enhancer. In contrast, compound embryos carrying either the DEL(5–8) deletion, which reduced the distance between Shh and the ZRS by 260 kb, or duplications that increased the distance to 1.1 Mb, DUP(5–8), DUP(C1-Z), showed normal limb morphology (Figures 3E and S3G). We did not detect major changes in Shh expression in E10.5 forelimbs as assessed by in situ hybridization (Figure 3F) and RT-qPCR (Figures 3G–3I, S3D, and S3F). We observed a slight reduction of Shh expression with the DUP(C1-Z) allele (Shh-ZRS distance = 1.08 Mb), but since the other duplication of similar size, DUP(5–8) (Shh-ZRS distance = 1.11 Mb), showed normal levels of Shh, this effect cannot be due solely to the increased distance. The difference between the two duplications may stem from the extra copy of Rnf32 in DUP(C1-Z), which could act as a competitor for ZRS activity. However, previous reports have shown that Rnf32 is not regulated by the ZRS, and we did not detect upregulation of Rnf32 (Figure 3I) beyond the 1.5-fold increase that corresponds to the increase in Rnf32 copy number from 2 in Shhdel/+ to 3 in Shhdel/DUP(1C-Z).

13Altogether, Shh expression appeared largely resilient to changes in enhancer-promoter distances when TAD boundaries were left unchanged and when no element normally external to the Shh TAD was introduced.

14We next examined if these intra-TAD rearrangements, which showed no major impact on Shh expression, could nonetheless alter the distribution of ZRS responsiveness. Genes outside the Shh TAD (Lmbr1, Rbm33 or Nom1) showed no significant expression changes in any of these genomic configurations (Figure S3), showing that confinement of enhancer activity is maintained. To look at the responsiveness within the TAD, we compared the expression of the regulatory sensor in the native context and in the context of the genomic rearrangements (Figure 3B). Prior to rearrangement, position 5.2 is refractory to activation by Shh enhancers, constituting one of the rare “dead spots” present in the domain, while 8.2 responds to multiple Shh enhancers (Figure 3J). Surprisingly, in the context of DUP(5–8), the sensor showed robust expression in the ZPA (Figure 3K), even though its position is identical to 5.2 with respect to the ZRS (same distance, same intervening sequences). Moreover, additional Shh expression domains (not observed at positions 5.2 and 8.2, but detected with insertions elsewhere in the locus), were also un-masked in the context of DUP(5–8) and DEL(5–8) (Figures 3K and 3L, pink and green arrowheads). Some of the new expression domains in DUP(5–8) may be associated with duplicated enhancers. But as the ZRS, the only limb enhancer active in the region, is located far outside the duplicated region, the gained expression in DUP(5–8) and in DEL(5–8) requires another explanation. We considered first that expression at position 5.2 could be locally repressed. If this is done by a centromeric repressor element, the reporter in DEL(5–8) should also be repressed. If this putative repressor was telomeric to position 5.2, then the reporter in DUP(5–8) should be repressed. Since both DEL(5–8) and DUP(5–8) show activity, the hypothesis of a local repressor at 5.2 is unlikely, as it would imply the existence of a cryptic de-repressor next to 8.2 that can counteract the repressor at 5.2. Even if we cannot fully rule out the existence of such a series of local elements, we propose that the rearrangements modulate the relative 3D folding of the TAD, and therefore change which regions are functionally exposed to the influence of the enhancers dispersed throughout this domain. This model is not only more parsimonious, but also fits well with the wide distribution of sensor cold spots, which correlates with 3D conformation.

15The resilience of Shh expression to changes in enhancer-promoter distance can be interpreted as evidence for classical looping models, where Shh is directed and tethered to the ZRS. Such looping interactions could be driven by a combination of elements present at the enhancer or the promoter. To test these models, we engineered balanced inversions that should split the Shh-ZRS TAD, while keeping Shh within the range of action of the ZRS defined by the previous experiments (Figures 4A, 4B, and S4). INV(-500-C1) exchanged sequences between the Shh TAD and the centromeric En2-Rbm33 TAD, while INV(6-C2) interspersed a region telomeric to the Shh TAD between the two halves of the original Shh TAD. In both cases, Shh-ZRS distances remained below 850 kb.

16Animals carrying these inversions over a deletion of Shh (Figure 4C) or an inactivating substitution of the ZRS (Figures S4B and S4C) showed monodactyly on both fore- and hindlimbs. Expression of Shh was lost in the limb of E10 embryos homozygous for either inversion (Figure 4D). Importantly, Shh expression was detected in other tissues (Figure 4D), showing that the gene was not globally repressed. Furthermore, in both configurations, the associated regulatory sensor remained at the same position relative to the ZRS before and after inversion and maintained expression in the ZPA (Figures 4E, S4D, and S4F), indicating that the endogenous activity of the ZRS is unaltered. Shh loss of expression is also unlikely to result from the disruption of a specific accessory element, since the two inversions used different breakpoints. As further controls, we produced two additional rearrangements, INV(-330-C2) and DEL(-500-33), this time changing the sequences flanking the Shh TAD without modifying the TAD itself (Figure 4B). In both configurations, Shh expression and function appeared unaffected (Figures S4G and S4H), which led us to conclude that disruption of the neighboring domains had minimal effect on Shh regulation.

17Overall, our experiments argue against the presence of a strong specific recognition system that will suffice to bring together Shh and the ZRS, as shown by the lack of Shh limb expression in INV(6-C2) and INV(-500-C1), despite shorter genomic distances than normal. These inversions, which reshuffled sections of different TADs, were the only ones from our series of rearrangements that affected Shh-ZRS communication, further strengthening the importance of TADs as regulatory units.

18To assess the consequences of the TAD-reshuffling inversions on the topology of the locus, we repeated the 4C analysis in E11.5 limbs for the INV(6-C2) allele (Figures 5 and S5). To account for the loss of Shh expression following inversion, we compared INV(6-C2) forelimb 4C profiles with the ones obtained from the anterior compartment of E11.5 WT limbs. We found that the reciprocal interaction peaks between Shh and the ZRS found on the WT allele were lost in INV(6-C2) (Figures 5A and 5B). In addition, Shh and ZRS contacts became mostly local, focused in ∼100 kb around each viewpoint, replacing the broadly distributed contacts throughout the Shh-ZRS TAD characteristic of the wild-type chromosome (Figures 5E and 5F). This was particularly striking for the region between Shh and the inversion breakpoint, since the linear organization of this segment is not directly changed by the inversion (Figure 5B). In INV(6-C2), Shh showed some interactions with the Mnx1, Nom1, and Lmbr1 promoters, but only marginally above what was observed in WT (Figures 5B and 5D), particularly if the reduced distance is considered. We did not observe broad reciprocal contacts between the Lmbr1-Nom1 viewpoint and the Shh region, which could have indicated the reformation of a new TAD, as described at other loci. Instead, in this case, the global reduction of contact frequency suggests that the overall conformation of the locus is disrupted, changing from a dense network of interactions along the TAD, either into a diffuse structure or into small, independent domains. Interestingly, the flanking viewpoints Nom1 and Rbm33 still displayed an asymmetric distribution of their contacts, avoiding the Shh-ZRS interval, indicating that some “insulating” aspects were retained (Figures 5C and 5D).

19Our data from the INV(6-C2) and INV(-550-C1) inversions showed that gross disruption of the Shh domain abolished the ability of the ZRS to contact and activate Shh. This loss of contact could be explained by the presence of an insulator at the telomeric end of the TAD, since it would be relocated between Shh and the ZRS in the inversion. To test this possibility, we generated additional inversions, which utilized the same C2 telomeric breakpoint and different centromeric breakpoints (4.2 and 2.1), located closer to Shh (Figure 6A). The resulting rearranged alleles repositioned the same ZRS-Lmbr1-C2 intervening sequence between Shh and the ZRS as in INV(6-C2) but displaced increasing portions of the Shh regulatory domain (Figures 6B–6D). At first glance, as more enhancers were moved away, we observed a progressive increase in the severity of the phenotypes (Figures 6E–6H). This was particularly obvious for the cranio-facial and axial skeletons, which showed a stepwise increase in the extent of malformations. INV(2-C2) mice essentially copied the phenotype of complete Shh null lines or of mice where the entire regulatory region is removed.

20Importantly, these inversions also moved the ZRS progressively closer to Shh. While INV(6-C2) almost fully recapitulated the ZRS null limb phenotype (Figure 6J), we observed a gradual recovery of the limb structures, especially in the hindlimb. INV(4-C2) embryos still showed severely affected monodactylous limbs (Figure 6K), but INV(2-C2) embryos showed partially restored hindlimb morphology: feet usually comprised three digits, with an anterior big toe with two phalanges and two toes with three phalanges, while the tibia-fibula elements were distinct and only partially fused (Figure 6L). These limb phenotypes imply a gradual restoration of antero-posterior polarity and growth of zeugopod and autopod structures, consistent with a partial rescue of Shh activity. While we were unable to detect Shh expression in the limb of E10.5 embryos, prior work on other ZRS mutants has shown that reducing Shh expression to 10% of wild-type level results in a somewhat less severe hindlimb phenotype than the INV(2-C2) embryos. Therefore, the INV(2-C2) phenotype is consistent with expression that is either extremely low or that occurs only during a very limited time period.

21Compound mutants over an inactive ZRS allele (Z2D; Figure S6) also showed the same progressive restoration of limb morphology, indicating allelism to ZRS activity. In brief, this allelic series reveals that reducing Shh-ZRS distance can restore functional interactions between these elements, and that the presence of the Lmbr1-C2 region is not sufficient to block these interactions.

22Although the ability of enhancers to act in a distance-independent manner is part of their original definition, this property was established on plasmid assays (i.e., at distances up to 10 kb). In their native genomic environment, enhancers have been shown to select their target gene through mechanisms influenced by proximity, even though promoter preference, occupancy by specific transcription factors, and/or tethering elements may modulate these effects. Our present study of the Shh locus provides new insights into the organizing principles of long-distance enhancer-promoter interactions.

23Confirming previous reports, our data demonstrate that enhancers act not only on their immediate neighborhood or on their target gene(s) but more generally across large domains. Our high-resolution characterization of the Shh regulatory domain highlights that the potential to respond to a given enhancer shows peaks and troughs throughout an otherwise largely permissive interval. This potential can be different, depending on the promoter; for example, insertions immediately adjacent to Rnf32, which does not respond to the ZRS, showed expression in the ZPA; inversely, some insertions next to Shh were inactive (which could be also due to competition). But we also uncovered substantial variation in expression between insertions of the same reporter, even when separated by only a few kilobases. This variation indicates that other factors than promoter sequence modulate responsiveness. We found a good correlation between the physical proximity to the ZRS, as measured by 4C, and the propensity to respond to its enhancer activity. This suggests that the Shh region folds in a pattern that acts as a mold for enhancer action (Figure 7). This framework is flexible, as it comprises only a few regions that are completely unresponsive (Figure 7A). The re-activation of unresponsive positions after internal rearrangements (DUP/DEL(5–8)) indicates that these positions are not necessarily locally repressed but simply excluded from contacting enhancers. Interestingly, some responsive positions showed contact frequencies that were as low as unresponsive regions, revealing either the influence of other factors or the limits of 4C to measure some interaction parameters (e.g., duration of contacts in the context of an ensemble of dynamic conformations).

24Previous studies have suggested that sequences close to or within the ZRS may target it to Shh, through the formation of a large loop. Our ability to detect the action of the ZRS throughout the TAD with a reporter gene argues against the need for a specific promoter to respond to the ZRS. The inability of the ZRS to contact Shh in most inversions further shows that the ZRS cannot find Shh if it is not located in the same TAD, demonstrating the absence of a TAD-independent system targeting the ZRS to Shh.

25Recent studies have substantiated models proposing that enhancers act within a space delineated by insulators. The existence of insulators is widely supported by experimental evidence that identified short regions that can block enhancer-gene interactions, and our data do not challenge the general existence and role of insulators. Yet, previous studies have also suggested that, even in the absence of specific insulators, certain loci show restricted enhancer-promoter interactions, questioning the universal necessity for insulators. Supporting this alternative view, many TAD boundaries appear not to be strict boundaries but correspond to a gradual effect, in terms of contact frequencies or blocking enhancer activities. At the Shh locus, the centromeric boundary between Rbm33 and Shh appeared much more marked than the telomeric one between the ZRS and Lmbr1, both from a structural (based on 4C and Hi-C data) and a regulatory (changes in enhancer responsiveness) viewpoint, suggesting the telomeric boundary may be less robust (or organized). Modeling of insulator action has indicated their effect is largely distance insensitive. The restoration of a functional ZRS-Shh interaction in the INV(C2) allele, when the distance separating Shh and the ZRS is reduced, therefore argues against the presence of a strict, well-defined insulator element. Interestingly, whereas Lmbr1 is not responsive to the ZRS in mice, its ortholog in the more compact chicken genome shows distinct expression in the ZPA. Based on our observations, we suggest that large genomic distances can act as a buffer for regulatory interactions, without the need to invoke the presence of specific insulators.

26According to simple polymer models, contact frequency should decline sharply with increasing distances. Yet, Hi-C data have revealed that, below 700 kb (approximately corresponding to the size of TADs), interactions occur more frequently than predicted, suggesting that loops and long-lived crosslinks may facilitate interactions at shorter scales. A recent study modeling the Igh locus emphasized the importance of spatial confinement to establish interactions and proposed that this is the main determinant for enhancer-promoter communication. Our data also demonstrate that the interactions weaving the Shh TAD are necessary for efficient long-distance enhancer-promoter interactions; in the context of the Shh TAD, genomic distance has a minimal effect on enhancer-promoter interactions, whereas distance becomes a critical factor when this TAD is disrupted. TADs increase interaction frequency between elements and reduce the otherwise limiting effect of genomic distances. TADs can therefore actively extend the functional reach of enhancers to distantly located target genes. It will be important to see to what extent genes are dependent on this functionality of TADs or if other, independent mechanisms have also evolved to ensure proper long-range regulation.

27The principles that lead to TAD formation are still debated, although mounting evidence suggests that loops between CTCF sites, possibly mediated by cohesin complexes, are involved. The presence and relative orientation of CTCF sites at both ends of the Shh TAD (Figure S7) partially fit with recent CTCF-based models. Yet, our observations also show noticeable deviations from what could be predicted from such models.

28Firstly, some of our sensors integrated beyond the CTCF site separating Rbm33 and Shh showed expression in the ZPA, implying that the ZRS is not blocked by this CTCF site or limited to a strictly defined CTCF loop. With respect to the predicted CTCF loops, the ZRS would be just outside the CTCF loop containing Shh, and the WT and INV configurations would be similar, whereas their functional outcome is strikingly distinct. In contrast, one would expected a more important effect in DUP(C1-Z), as the ZRS is now moved away from the potential CTCF-mediated loop containing Shh. Our functional data therefore underline that binding and orientation of CTCF are not sufficient to predict regulatory outcomes.

29Beyond the underlying mechanism(s), the decisive factor governing enhancer-promoter functional interactions is the frequency of physical interactions between these elements. In this respect, the relative degree of insulation (which essentially is how TADs are identified) is far less important than the 3D volume of a TAD and its internal dynamics. Addition or removal of sequences normally present in the Shh domain have a small but noticeable impact on enhancer action, whereas interspersing external sequences into the Shh TAD, like with INV(6-C2), leads to a loss of compaction associated with the TAD and reduced long-range interactions. This shows that interaction frequency within a TAD may depend on its internal sequence or chromatin organization and not only on loops determined at its extremities.

30Controlling regulatory interactions is an essential function of genomes, and current models have put a lot of emphasis on insulation. In animals with more compact genomes, insulators may be critical to avoid unwanted interactions between close neighbors, which could explain why Drosophila evolved multiple types of insulators. But in animals with large genomes and large intergenic distances, genomic distance per se can often suffice to limit functional interactions. In these conditions, promoting long-range interactions becomes crucial to ensure robustness of a system that would otherwise depend on rare, stochastic collisions. Absence of such a mechanism would lead to phenotypic variability as illustrated by INV(2-C2) animals (Figure 7). In this view, the formation of compact genomic domains like TADs and the diverse mechanisms that ensure both robust and specific long-range regulatory interactions may have been essential to expand the genomic toolbox of evolution.

31The founder ShhSB-C1, ShhSB-C2, and Z2D mice were generated by homologous recombination in E14 embryonic stem cells (ESCs). We inserted a Sleeping Beauty transposon that carries a LacZ reporter gene and a loxP site at chr5:29,413,901 for ShhSB-C1 and at position chr5:29,854,582 for ShhSB-C2. The ShhSB-C2 insert also contained a second loxP site outside the transposon. For Z2D, the ZRS enhancer (chr5:29,641,240-29,642,424) was substituted with a Dach1 limb enhancer (chr14:97,485,490–97,486,674). Remobilization of the SB transposon and mapping of new insertions was performed as described. Targeted rearrangements were produced by in vivo recombineering. Shhdel mice carry a deletion of the second and third exon of Shh, produced by Cre-mediated recombination of the Shh-nLZ transgene. Shh-nLZ mice were kindly provided by Andreas Kottmann (Columbia University, New York) and are referred to as Shh::LacZ mice in this paper. All lines were maintained by breeding with C57BL/6J mice. Genomic positions are given for using the mm9/NCBI37 assembly. Mouse experiments were conducted in accordance with the guidelines in place at the European Molecular Biology Laboratory, as defined and overseen by its Institutional Animal Care and Use Committee, in accordance with the European Directive 2010/63/EU.

32LacZ staining, whole-mount in situ hybridization, and skeletal preparation were performed according to standard protocols. Full details are in the Supplemental Experimental Procedures.

33Total RNA was isolated from microdissected tissue embryos using a PureLink RNA Mini Kit (Invitrogen) with on-column DNase I treatment; 200 ng to 1 μg of isolated RNA was reverse transcribed with a ProtoScript M-MuLV First Strand cDNA Synthesis Kit (New England Biolabs) using oligo-dT as primer. qPCR was performed on an ABI7500 system with SYBR Green (Applied Biosystems), and analyzed using the ΔΔCT method. For data normalization, TBP, GusB, or Hif1 was used as the reference gene, and each condition was normalized to stage-matched littermate controls. Primers are listed in Table S3.

344C libraries were generated from microdissected embryonic limb tissues following published protocols (see Supplemental Experimental Procedures for details). 4C libraries were generated by PCR amplification with primers containing barcodes and Solexa adapters (see Table S3). Viewpoints were analyzed in duplicate and approximately 40 libraries were pooled per sequencing lane. All samples were subjected to 50 bp single-read sequencing using Illumina HiSeq 2000.

35For the analysis of 4C libraries, FASTQ files were de-multiplexed based on barcode and viewpoint primer sequences, allowing no mismatch (first eight bases were used). Primer sequences were trimmed keeping the first restriction site, and de-multiplexed libraries were aligned to the mm9 reference genome using Bowtie version 1.0.0. Aligned reads were then mapped to an in silico NlaIII-digested mm9 genome in order to filter out non-informative reads. Only reads mapping to fragment ends in the correct orientation were kept and assigned to the corresponding fragment end. Fragment read counts correspond to the sum of the counts obtained for each of their extremities. We assessed the quality of the libraries by determining the percentage of mapped reads and the percentage of reads mapping in cis (intra-chromosomal reads relative to the viewpoint) for cross-linking and digestion efficiency. All samples showed similar library quality based on these parameters (see Table S4). 4C-seq reads were filtered as described in and down-sampled to match the number of the library with the lowest read count. Interaction values with the viewpoint were calculated using two measures: normalized raw read counts smoothened across 11 fragments and a hit percent rate, including a minimal threshold. For the latter approach, we transformed the 4C signal to a binary value (0 or 1) for each fragment, depending on whether the normalized read count was below or above a certain threshold (e.g., over 1, 10, or 100 counts). Fragments that fulfill the threshold criteria are termed hits. We calculated the hit percentage in a given window (e.g., 25, 51, or 101 fragments) as an estimate of the contact frequency and reliability of a given fragment. We compared the effect of different thresholds and window size on the reproducibility of the signals obtained with biological replicates (see log2 ratios plots in Figure S2). For the experiments displayed here, parameters with a 51-fragment binning size (∼10–20 kb length) with a read count threshold of 10 showed robust and reproducible contact patterns across the region of interest and were therefore used.

36For the analysis of 4C data from samples carrying a genomic inversion, we inverted the reference genome in silico between the breakpoint coordinates and removed the fragments containing the breakpoints. To estimate the asymmetry of the interaction profiles, we calculated cumulative count distributions on each side of the viewpoint by using the counts of the sub-sampled libraries. In this analysis, we disregarded the fragments located at a distance less than 10 kb from the viewpoint to reduce the strong influence of the most proximal fragments. Data have been deposited on ArrayExpress (E-MTAB-4980 ).

37F.S. conceived the project and designed it with O.S. O.S., L.P, T.A., and S.R. performed the experiments. F.K. and W.H. performed computational analysis of 4C data. F.S., O.S., L.P., S.R., and T.A. analyzed the data. O.S. and F.S wrote the manuscript with input and comments from the other authors.