Research Article

A Switch Between Topological Domains Underlies HoxD Genes Collinearity in Mouse Limbs

See allHide authors and affiliations

Science  07 Jun 2013:
Vol. 340, Issue 6137, 1234167
DOI: 10.1126/science.1234167

Structured Abstract


During vertebrate limb development, gene members of the HoxD cluster are transcribed in two subsequent waves, following a collinear strategy. Early on, genes located in the center of the cluster are transcriptionally activated and pattern a "central" part of the limbs, such as the forearm and part of the arm. Subsequently, a partially overlapping group of genes, located at one extremity of the cluster, are activated in a distal territory, which will expand and organize the distal pieces of our limbs: the hands. Although enhancer sequences controlling the latter phase have been characterized and mapped in a gene desert centromeric to the gene cluster, the location of the early enhancers, as well as the mechanism underlying the transition from the early to the late phases of transcription, remained elusive.

Embedded Image

Regulation of Hoxd gene collinear expression in developing limb buds. The HoxD cluster is flanked by two topological domains containing distinct regulatory capacities. The telomeric domain regulates early transcription in the arm and forearm. The centromeric domain subsequently patterns the hands. During limb development, a switch occurs between landscapes as specific Hoxd genes swing from one regulatory domain to the other through a conformational change. This modular organization creates a domain in between, which will form the wrist.


To localize enhancer sequences, we screened a series of conserved sequences using a transgenic lentivector-based approach in mice. We also analyzed various histone modifications by chromatin immunoprecipitation, as well as the interaction profiles by multiplex circular chromosome conformation capture sequencing (4C-seq) on microdissected wild-type and mutant limb buds. The regulatory switch was monitored using 4C-seq, and mutant configurations were produced using embryonic stem cell targeting, targeted meiotic recombination, and sequential targeted recombination strategies. In situ hybridizations and reverse transcription quantitative real-time fluorescence polymerase chain reaction analyses were used as readouts for gene transcription.


We show that the early phase of transcription requires enhancers located in the telomeric gene desert. Therefore, the early and late phases of Hoxd gene transcription in limb buds are controlled by two opposite deserts flanking the cluster on either side and corresponding to two adjacent topological domains. The transition between early and late regulation involves a functional and conformational switch between these domains, as reflected by a subset of genes mapping centrally into the cluster, which initially interact with the telomeric domain and subsequently shift to establish new contacts with the opposite side. This polarization of the cluster between the two domains ensures a proper collinear distribution of HOX products in both proximal and distal limb structures.


The intriguing collinear correspondence between Hoxd gene topology and the patterning of proximal versus distal limb structures relies on the sequential implementation of two regulatory landscapes flanking the gene cluster. Genes located around the boundary between these two topological domains will swing from one to the other, along with the switch in regulation. The existence of independent regulation allows for cellular offset to occur between the two expression domains, where a reduced HOX protein dose is present, and which will develop into the wrist. Therefore, the mechanism patterning vertebrate proximal and distal limb pieces also contains the intrinsic capacity to build the necessary articulation in between—an adaptive value presumably explaining the selection of this complex regulatory system in tetrapods.

Collinearity Cracked in Tetrapod Limbs

During limb development, the time a nd place of Hox transcription are fixed by respective gene position within the gene cluster. Andrey et al. (p. 1234167; see the Perspective by Rodrigues and Tabin) found that this enigmatic property results from the opposite and successive actions of two large regulatory landscapes located on either side of the mouse Hox locus. In the early phase, one of these topological domains regulates transcription in the proximal limb until a switch occurs toward the other topological domain, which takes over the regulation in the distally developing digits. As a side effect of this antagonistic regulatory strategy, cells in-between have lessened Hox transcription, which generates the wrist.


Hox genes are major determinants of the animal body plan, where they organize structures along both the trunk and appendicular axes. During mouse limb development, Hoxd genes are transcribed in two waves: early on, when the arm and forearm are specified, and later, when digits form. The transition between early and late regulations involves a functional switch between two opposite topological domains. This switch is reflected by a subset of Hoxd genes mapping centrally into the cluster, which initially interact with the telomeric domain and subsequently swing toward the centromeric domain, where they establish new contacts. This transition between independent regulatory landscapes illustrates both the modularity of the limbs and the distinct evolutionary histories of its various pieces. It also allows the formation of an intermediate area of low HOX proteins content, which develops into the wrist, the transition between our arms and our hands. This regulatory strategy accounts for collinear Hox gene regulation in land vertebrate appendages.

During animal embryonic development, various axial structures must be patterned along the major body axis, such as the spine, the central nervous system, or the intestinal tract. This critical function is achieved by genes such as those in the Hox gene family, a set of transcription factors conserved throughout animal evolution. The pioneering work of Lewis using Drosophila genetics (1) revealed that these genes are colocalized in the genome and that the relative order within this cluster corresponds to the relative position of the structures they instruct along the anterior-to-posterior body axis. This intriguing correspondence is referred to as "collinearity," and its underlying molecular mechanisms have remained elusive [see, e.g., (2)].

Although Drosophila contains a single copy of the Hox gene cluster in two pieces, there are four copies of an ancestral Hox gene complex (the HoxA to HoxD clusters) in terrestrial vertebrates. As for Drosophila, the relative position of each vertebrate gene within its cluster determines the axial distribution of its future functional territory (3), reflecting the evolutionary conservation of this patterning mechanism (4, 5). The availability of multiple Hox clusters at the root of vertebrates made it possible for novel functions to emerge, associated with any one of these gene complexes, which likely participated in the appearance of several vertebrate-specific features. For example, the HoxA gene cluster is particularly important for the proper function of neural crest cells (6).

Such neofunctionalization events often coopted several genes from the same cluster, using their collinear properties. Consequently, vertebrates display collinear Hox gene expression in a variety of developmental contexts, generally associated with axial structures (7). However, the mechanisms leading to these collinear distributions of HOX protein products may be different from case to case, because only the final combinatorial display of proteins was selected for, rather than the underlying process (8). This is illustrated by the comparison between the trunk and the appendicular axes, where distinct collinear mechanisms are implemented. During axial extension of the trunk, Hox genes are activated in a strict, progressive time sequence associated with a stepwise transition from a repressed chromatin configuration to a transcriptionally competent context (9, 10). However, the situation is distinct in vertebrate appendages.

During limb development, gene members of the HoxD cluster are transcribed following a clear collinear strategy (8, 1113). Yet, unlike in the trunk, collinearity occurs through two subsequent waves of transcription, which split the pool of limb mesenchymal cells into two distinct domains along the limb proximodistal axis. Early on, genes located in the center of the cluster are transcriptionally activated and pattern the proximal part of the limbs leading to the prospective arm and the forearm (13). Subsequently, a partially overlapping group of genes, located at the centromeric extremity of the cluster, are turned on in the most distal aspect of the early domain. This subapical domain will then expand and organize the most distal pieces of our limbs: the hands [Fig. 1A and (14)]. In this latter case, a set of enhancer sequences lying in a gene desert located immediately centromeric to the gene cluster controls transcription in digits through long-range interactions (Fig. 1B) (12).

Fig. 1 Early phase of Hoxd gene transcription in forelimb buds.

(A) Expression of Hoxd10 during bud outgrowth (bottom) with a schematic illustration of the early (green) and late (purple) phases of expression (top). Early on, Hoxd10 expression involves most cells of the forelimb bud. Subsequently, a small group of distally located cells starts to implement the late regulation (purple at E10.5). The two domains eventually separate from one another (E11.5). The skeleton on the right shows the fate of the two domains. (B) Schematic of the HoxD locus, with its centromeric regulatory landscape (purple ovals) controlling expression in digits and hypothetic arm/forearm enhancers located telomeric to the cluster (29). Genes are indicated as gray rectangles. Centromeric (5′) and telomeric (3′) sides are indicated. (C) Absolute quantification of steady-state levels of Hoxd transcripts in entire forelimb buds at E9.5 and E10.5, as well as in proximal forelimbs (without digits) at E12.5 (schemes on the right). Hoxd9 is the most robustly expressed gene at E9.5, whereas Hoxd11 mRNA content is strongly increased at E10.5. Error bars, mean ± SD (N = 4 pairs of limbs). (D) Tiling array hybridization of E9.5 forelimb bud mRNAs showing the robust transcription of the central part of the gene cluster. The y axis represents a log2 scale of cDNA/gDNA (genomic DNA) signal intensity. (E) Centrally located Hoxd genes (Hoxd8 to Hoxd11) are decorated with the active mark H3K4me3 at all stages. The y axis represents a log2 scale of chromatin immunoprecipitation (ChIP)–enriched/input gDNA signal intensity. (F) Profiles of H3K27Ac marks, which label both active enhancers and promoters. Together with the Hoxd8 to Hoxd10 region, the 3′-located promoters (e.g., Hoxd3 and Hoxd4) are acetylated at the earliest stage, likely due to their transcription in lateral plate mesoderm. They become progressively deacetylated over time, as Hoxd11 becomes acetylated. The y axis represents the ChIP-enriched minus input gDNA normalized read count. (G) The H3K27me3 repressive mark decorates early on the inactive, centromeric part of the gene cluster. These marks are subsequently gained in the Hoxd1 to Hoxd4 region, when these latter genes are no longer transcribed. The y axis represents a log2 scale of ChIP-enriched/input gDNA signal intensity.

In the present work, we addressed the mechanism for controlling both the early phase of transcription and the transition between the early and late regulations. We show that the proximodistal collinear transcription of Hoxd genes during limb development relies on a regulatory switch between two opposite regulatory landscapes, overlapping over a common subset of Hoxd loci. These landscapes correspond to topological domains (15, 16), and we show that each Hoxd gene displays a specific tropism for either domain, depending on its relative genomic position in the cluster.


Early Phase of Hoxd Gene Transcription

To understand the transcriptional control of Hoxd genes during the early phase of expression, we quantified the level of steady-state Hoxd gene expression at three consecutive stages of forelimb bud development. We used embryonic day 9.5 (E9.5) early budding limbs, which is soon after the onset of Hoxd gene transcription, as well as entire E10.5 limb buds. We also analyzed micro-dissected proximal parts of E12.5 forelimbs (Fig. 1C). This latter domain contains those cells directly derived from early bud progenitor cells, which will remain at a proximal location during subsequent limb development. These cells are the remnants of the early expression phase (13) and will pattern the arm and forearm (Fig. 1A, green domains). Reverse transcription quantitative real-time fluorescence polymerase chain reaction (RT-qPCR) analyses revealed that Hoxd9 was expressed the strongest in early E9.5 limb buds. At later stages, however, its mRNA level decreased, whereas Hoxd11 transcription was strongly up-regulated (Fig. 1C). A comparison between the transcript profiles and the distributions of specific chromatin modifications confirmed that Hoxd8 to Hoxd11 are the main targets of this early phase of transcriptional regulation. These genes were indeed enriched in both H3K4me3 and H3K27ac chromatin marks, known to label actively transcribed genes (Fig. 1, D to F) (13, 1720). In addition, the progression in time of H2K27 acetylation toward the centromeric end of the gene cluster, combined with the delayed transcription of Hoxd11, reflected the step-wise activation of these genes occurring during this early phase (13).

As previously observed during trunk development (10), low transcription signals were scored over Hoxd13 in early E9.5 limb buds, before Hoxd12 was switched on. Although this may reflect the activity of a spurious enhancer, it might also come from a few cells already implementing the second phase of transcription, because this latter phase primarily targets Hoxd13 (14). Also, although both transcripts and H3K27ac marks were detected over the Hoxd4 to Hoxd3 region at the earliest stage, their amounts progressively decreased subsequently (Fig. 1, C, D, and F). In contrast, the Polycomb-associated mark H3K27me3 (2123) was depleted from the telomeric half of the cluster in early limb buds, whereas it displayed a maximal coverage toward the centromeric extremity, peaking over Hoxd13 (Fig. 1G). Lower amounts of H3K27me3 signals were also detected over the promoter regions of both Hoxd11 and Hoxd10, which presumably reflects the presence of nonexpressing cells located in the anterior-most aspect of forelimb buds and included in our dissections. At later developmental stages, the Hoxd4 to Hoxd3 region became also decorated with this mark, in agreement with the loss of both transcriptional activity and H3K27 acetylation (Figs. 1C, F, and G) (13).

A Telomeric Regulatory Landscape

We used circular chromosome conformation capture (4C) to identify contacts established either by active or inactive genes during this early phase (9, 24). Hoxd13, which was mostly repressed, interacted preferentially with the centromeric gene desert. The extent of this interaction domain (Fig. 2A, upper track, gray-shaded area) precisely matched a topological domain—i.e., a domain of preferential enhancer-promoter interactions (25) reported to cover this gene desert and described in (16). Conversely, the actively transcribed Hoxd9 gene contacted a 1.5-Mb region, telomeric to the cluster and also containing a large gene desert (Fig. 2A, middle track). Again, this desert together with a large part of the gene cluster matched a reported topological domain (Fig. 2A, middle track, gray-shaded area) directly adjacent to the former (16). In early forelimb buds, the telomeric desert was specifically labeled with H3K27ac marks, which are generally associated with active enhancers (Fig. 2A, lower track) (18, 19). The amount of this histone H3 modification, however, severely decreased over developmental time (fig. S1), likely reflecting the transient activity of potential enhancers located within this gene desert and active mostly during the early stages of limb bud development.

Fig. 2 Interaction profiles and forelimb bud enhancers.

The 4C and H3K27Ac profiles were generated using early E9.5 limb buds. (A) The inactive Hoxd13 gene preferentially contacts the centromeric landscape (on the left), corresponding to a topological domain according to (16) (upper track, gray-shaded area), whereas the active Hoxd9 gene contacts the telomeric landscape with high affinity (right), also corresponding to a topological domain (middle track, gray–shaded area). The bottom track depicts the profile of H3K27 acetylation in early limb buds. The y axis represents the ChIP-enriched minus input gDNA normalized read count. (B) Two major enhancers (CNS 39 and 65) were localized within the telomeric gene desert, as shown by reporter LacZ patterns in transgenic embryos. The number of embryos with stained limbs over total transgenics is indicated. (C) The 4C interaction profiles (running mean, window size 11) with the CNS 39 (top) and 65 (bottom) enhancer sequences taken as viewpoints. The extents of both interaction domains (approximately covering the telomeric gene desert) are comparable when either region 39 or region 65 is used as bait. In both cases, the interaction domains match the same topological domain (gray-shaded area) as depicted in (A). In addition, both regions interact specifically with the transcribed portion of the HoxD cluster, with the most conspicuous peaks of interaction mapping over the Hoxd9 to Hoxd11 region. The frequency of interactions with both the Hoxd13 and Hoxd12 loci dramatically dropped down (black arrowhead) (see fig. S1).

We selected 72 conserved noncoding sequences (CNS) located within this telomeric landscape (table S1) and assayed them for enhancer activity using lentivector-mediated transgenesis (26, 27). Two regions elicited a robust expression in early limb buds (Fig. 2B, CNS 39 and 65). CNS 39 triggered lacZ transcription in very early limb buds, as expected for a proximal regulation. However, expression localized more distally at later developmental stages. CNS 65 induced transcription in a broader domain, also at an early stage, which remained restricted to the proximal limb, subsequently (Fig. 2B). Both regions are located within the telomeric gene desert, 385 and 670 Kb away from the HoxD cluster, respectively. The contacts established by these sequences with the target Hoxd genes were verified by using both CNS 39 and 65 as viewpoints in 4C experiments (Fig. 2C). This reverse 4C approach confirmed that both sequences interact specifically with the transcribed Hoxd genes. In these two instances, the overall extent of the interaction landscapes, as defined by using the enhancers as baits, precisely matched the reported telomeric topological domain (Fig. 2C, gray-shaded area). These interactions between the two enhancers and the HoxD cluster were abruptly lost over the Hoxd12 to Hoxd13 region (Fig. 2C, arrows, and fig. S1). As expected from such a conformational insulation from both enhancers, these latter two genes are not transcribed (or are poorly transcribed) at this stage.

Genetic Analysis of the Telomeric Gene Desert

We assessed the function of this telomeric landscape, including both CNS 39 and 65, by engineering several deletions in vivo, as well as an inversion containing the entire gene desert (Fig. 3 and tables S2 and S3). The targeted deletion in vivo of CNS 65 alone led to a 30 to 40% decrease in the steady-state levels of Hoxd8 to Hoxd11 mRNAs in proximal forelimb buds at E12.5 [Fig. 3, white arrow, and fig. S2, del(65)]. We also used targeted meiotic recombination (TAMERE) and sequential targeted recombination (STRING) in vivo recombinant strategies (28, 29) to engineer a series of alleles carrying larger modifications (Fig. 3 and table S4). When a 150-kb region directly adjacent to the HoxD cluster and including the Mtx2 gene was removed [Fig. 3, del(attP-SB2)], no detectable loss of either Hoxd10 or any other centrally located Hoxd genes was scored in proximal limb buds, by using both in situ hybridization and RT-qPCR as readouts after dissection of the proximal domain (Fig. 3, white arrow, and fig. S2). In contrast, an almost complete abrogation of Hoxd10 transcript accumulation in the proximal domain was achieved when the 525-kb del(SB2-65) deletion was used, which removed most of the gene desert (Fig. 3, white arrow). In this genetic configuration, Hoxd transcripts produced by the centrally located genes were depleted by 90 to 95% (fig. S2).

Fig. 3 Allelic series over the telomeric gene desert.

This allelic series included several deletions as well as an inversion (bottom line). The CNS 39 and 65 regions are indicated with green ovals and crossed out whenever included in the deleted material in any of the mutant configuration. Animals analyzed were all heterozygotes over a balancer allele [del(8-13)] (13), both to ensure valid and coherent comparisons and to assess the behavior of only those Hoxd genes located in cis of any rearrangement. The analysis of Hoxd10 gene expression by in situ hybridization in E11.5 embryos was used as a readout (right column). qPCR quantification of steady-state levels of Hoxd10 mRNAs in the proximal forelimb is indicated on the bottom left corner of each panel, expressed as a fraction of the wild-type control equal to 1. In each panel, the proximal domain, resulting from the early phase of expression, is indicated with a white arrow, whereas the late phase of expression leading to the digit domain is depicted with a black arrow. From top to bottom: A 150-kb deletion del(attP-SB2) had no effect on Hoxd10 transcription (P > 0.05, N = 3 pairs of proximal limbs). The 2-kb deletion specifically removing CNS 65 significantly (P < 0.01, N = 3 pairs of proximal limbs) decreased the amount of Hoxd10 transcript by 33%, whereas all deletions, including the SB2-65 interval, removed more than 90% of Hoxd10 transcripts [del(SB2-65), del(attP-65), and del(attP-SB3); P < 0.01, N = 3 pairs of proximal limbs]. However, none of these deletions totally abrogate proximal transcription (see fig. S2). In contrast, the 28-Mb inversion inv(attP-CD44) (bottom) removed all traces of Hoxd10 transcripts from the proximal forelimb, due to the dissociation between the target genes and the telomeric landscape.

We verified and confirmed that most of the early forelimb enhancers are located within this 525-kb DNA interval by analyzing two overlapping deletions spanning 1 Mb of telomeric DNA [Fig. 3, del(attP-65) and del(attP-SB3)]. The remaining transcriptional activity completely disappeared (99% decrease) when the inv(attP-CD44) allele was used. In this large inversion (Fig. 3, bottom, red arrows), the gene desert was indeed repositioned 28 Mb away from the HoxD cluster, and hence all potential interactions were abrogated. This also pointed to the existence of an additional remote enhancer(s), yet to be mapped and not included in any of the deletions used above (Fig. 3). Using this allelic series, we concluded that the early phase of Hoxd gene transcription in forelimb buds is controlled by enhancer sequences mostly located within the telomeric gene desert. Consequently, the two global regulations acting over the HoxD cluster during limb development are located in distinct topological domains, corresponding to two gene deserts present on either sides of the gene cluster.

A Transition Between Opposite Regulations

In mutant limb buds where proximal expression was almost abolished due to a telomeric deletion, we noticed a severe loss of the late expression domain in presumptive digits, for Hoxd9, Hoxd10, and Hoxd11 (Fig. 3 Fig. 4, A to C, black arrows, and fig. S2). Yet this was not the case for Hoxd13, which confirmed the presence and activity of the requested digit enhancers (12). This transcriptional down-regulation in the distal domain may have been caused by a growth defect in the distal forelimb, as induced by the absence of early Hoxd gene expression. We ruled out this possibility by looking at the expression of Hoxa13, which was used as a marker for the distal limb territory (fig. S2) (30, 31) and confirmed that these mutant embryos had normally shaped hand plates.

Thus, sustained transcription in the proximal domain is necessary for efficient subsequent transcriptional activation of Hoxd9 to Hoxd11 in the digit domain. This implies that both distal and proximal Hoxd-expressing cells derive from the same progenitors rather than distinct progenitor populations segregated at the onset of limb development (31). Indeed, the observed effect could occur only if distal cells were derived from cells that had expressed these genes previously, during the early phase. Moreover, digit progenitor cells must go through the early expression phase to be fully licensed to activate centrally located Hoxd genes during the late phase, when these latter become controlled by the opposite regulatory landscape (12).

We analyzed H3K27me3 coverage over the HoxD cluster in del(attP-SB3) mutant early forelimb buds, where more than 90% of Hoxd9, Hoxd10, and Hoxd11 transcripts were lost (fig. S2). A robust gain of H3K27me3 was scored over these three genes in the deleted configuration (Fig. 4D), coinciding with their subsequent transcriptional down-regulation in forming digits. This result suggests that in presumptive wild-type digit cells, centromeric (digit) enhancers readily access the middle part of the HoxD cluster due to an open chromatin configuration inherited from early limb bud cells, where this region is actively transcribed. In limbs carrying the telomeric deletions, however, this early transcription is abolished due to the absence of the appropriate enhancers. Consequently, this abnormally inactive portion of the cluster is now fully decorated with H3K27me3 chromatin marks. The unexpected presence of this Polycomb-associated repression likely induced the observed delay in the establishment of the second phase of expression; when mobilized, the digit enhancers had to compete with this "negative" chromatin domain rather than interacting with a more open chromatin configuration, as in wild-type limbs. In contrast, transcription of Hoxd13 was not modified because this gene is mostly repressed early on. H3K27me3 coverage of Hoxd13 in mutant proximal cells was similar to what was seen in wild type, and hence its accessibility remained unchanged in the mutant.

Fig. 4 Transcriptional connectivity of the early and late domains.

The abrogation of the early phase of transcription affects the efficiency of the subsequent transcriptional activation in digits. Δ is for the del(8-13) allele, used as a balancer in our crosses. (A) Scheme of the del(attP-SB3) deletion, removing most of the telomeric landscape. (B) This deletion strongly depletes both Hoxd11 and Hoxd10 transcripts from the proximal domain in E11.5 buds (white arrows). In both cases, both the intensity and extent of the distal domain was reduced too (black arrows), yet not for Hoxd13, which is normally not expressed in the proximal domain (left panels, black arrow). The panel showing Hoxd10 expression is taken from Fig. 3. (C) The RT-qPCR quantifications of Hoxd transcripts present in the presumptive digit domain of such mutant digits show a significant decrease, from Hoxd9 to Hoxd11, whereas neither Hoxd12 nor Hoxd13 are significantly affected. Error bars, mean ± SD (N = 3 pairs of distal limbs). *P < 0.05, **P < 0.01. (D) H3K27me3 profiles covering the centromeric part of the HoxD cluster in both E10.5 control (top) and mutant forelimb buds carrying a deletion of the telomeric landscape (middle). The ratio between both conditions is shown below (positive and negative values in green and orange, respectively). H3K27me3 is significantly increased over the Hoxd9 to Hoxd11 region in mutant limb buds, yet not over Hoxd13, where it remains unchanged. In the absence of Hoxd gene transcription, cells in the proximal domain display H3K27 trimethylation throughout this part of the cluster.

A Regulatory Switch Between Topological Domains

From this data set, we conclude that Hoxd9, Hoxd10, and Hoxd11 switch from telomeric regulation in early limb cells to centromeric regulation in digit cells. To confirm this hypothesis, we looked at potential modifications in three-dimensional (3D) conformations that would occur concomitantly with the transition in regulations. We generated 4C profiles using several Hoxd promoters as viewpoints, in both the early and late transcription domains. In all samples analyzed, Hoxd1 was found to interact strongly with the telomeric neighborhood, whereas Hoxd13 was preferentially engaged toward the centromeric side. These opposite tropisms were observed regardless of the transcriptional status—i.e., whether the gene was active or inactive (Fig. 5, A and B). In contrast, both Hoxd9 and Hoxd11 displayed more dynamic interaction patterns. Whereas they established preferential contacts with the telomeric domain in early E9.5 forelimb cells, they both increased their interactions with the centromeric domain in E12.5 digit cells (Fig. 5, C and D), indicating that the regulatory switch is accompanied by a transition between two conformational states. This transition did not occur in late proximal forelimb cells, showing that they merely maintain the early phase of transcription (Fig. 5, C and D). Notably, the conformational switch appeared more extensive for Hoxd11 than for Hoxd9 (Fig. 5, C and D). This may reflect the relative location of Hoxd9 within the gene cluster, which is closer to the telomeric extremity and, as such, may have an increased preference to interact with the telomeric regulatory landscape.

Fig. 5 A conformational switch underlies collinearity in limbs.

A 4C analysis of the transition between the early and late phases of Hoxd gene expression in forelimb buds, as seen from different viewpoints within the gene cluster. On the left side of each track, a schematic representation of the expression pattern is shown for the corresponding gene. Green depicts the early and proximal domain, whereas the late and distal domain is shown in purple. The absence of any color reflects the absence of expression. At the top of both (A) and (C), the presence and extent of topological domains are shown, with data taken and adapted from (16). These domains, defined by HiC, reflect regions of high enhancer-promoter interactions. The red lines label their borders. The 4C interaction profiles shown below are made from a running mean. (A) Hoxd1 is silent during the early (tracks 1 to 3) and late (track 4) expression phases. This gene establishes constitutive contacts with the telomeric domain (gray-shaded area and 86 to 91% of contacts), regardless of which regulation is implemented, whereas 9% of the contacts are scored with the centromeric domain. (B) Likewise, Hoxd13 strongly interacts with the centromeric landscape (gray-shaded area), regardless of whether this landscape is active or inactive, but with some differences, such as the additional contact with island 3 in digit cells, where it is transcriptionally active (arrow). (C) In contrast, Hoxd11 can display preferential contacts with either gene desert. In the early buds and in the subsequent proximal domain, it interacts mostly with the telomeric desert (gray-shaded areas; 77 and 80%, respectively). In the distal domain, however, Hoxd11 now switches its contacts toward the centromeric desert (62%, gray-shaded area), including a specific interaction with island 3 (arrow). (D) Likewise, Hoxd9 mostly contacts the telomeric domain during the early phase of expression (gray-shaded area; 81 percent), yet it substantially increases its contacts with the centromeric landscape in digits, including with island 3 (arrow). In this case, however, the final balance between centromeric and telomeric interactions is almost equalized.

The transition from the telomeric to centromeric regulation results in the strong activation of Hoxd13 in digit cells, a gene heavily covered by H3K27me3 in early forelimb bud cells, from where digit cells will subsequently derive. Therefore, we compared the enrichments of H3K27me3 between the early and late transcription phases and noted a large de novo H3K27 trimethylation of the now inactive telomeric part of the cluster. This trimethylation occurred in parallel with the demethylation of the same histone H3 residue in the most centromeric—now active—part containing Hoxd12 and Hoxd13, which illustrates the antagonistic behaviors of the two topological domains (Fig. 6A).

Fig. 6 Functional independence of the regulatory landscapes.

(A) Dynamics of H3K27me3 chromatin modifications covering the HoxD gene cluster during forelimb development. In early limb buds (upper track) where the telomeric regulation is implemented, genes located at the centromeric extremity of the cluster (e.g., Hoxd13) are covered by H3K27me3 (upper track; see also Fig. 1). During the late phase of expression in the presumptive digit domain (lower track), the centromeric part of the gene cluster has now been cleared from these repression-associated marks and has become fully active. In contrast, the telomeric part of the cluster is now inactive and decorated with this posttranslational modification. The comparison between both tracks reflects the perfect complementarity of the profiles and illustrates the switch in regulations. The y axis represents a log2 scale of ChIP-enriched/input gDNA signal intensity. (B) The H3K27me3 profiles over the telomeric desert show little difference between the early forelimb bud (first track) and the proximal domain of a late forelimb (second track). In contrast, this desert becomes heavily decorated with H3K27me3 in digit cells (third track, red-shaded area), where the telomeric regulation no longer operates. In the del(Nsi-Atf2) deletion, which removes the centromeric gene desert (bottom track, scheme), the late phase of transcription, controlled by the centromeric regulation, is abrogated (scheme on the left). In this mutant condition, however, the trimethylation of H3K27 still occurs over the telomeric gene desert (red-shaded area), indicating that the termination of the telomeric regulation does not depend upon the start of the centromeric regulation. The y axis represents a log2 scale of ChIP-enriched/input gDNA signal intensity.

Functional Independency of Topological Domains

We also scored a reenforced distribution of H3K27me3 marks over the entire telomeric gene desert in digit cells (Fig. 6B). This was unexpected because a robust presence of these repressive marks was not observed previously in other contexts where Hoxd genes are silent, such as in embryonic stem cells or in the brain (10, 32). Within this gene desert, we selected those regions with the highest content of H3K27me3 modifications and compared them with another set of eleven DNA regions mapping within the desert (table S5) and showing an enrichment in the acetylation of H3K27 in early E9.5 limb buds, at a time when the telomeric regulation is fully operational. Eight out of these eleven regions were scored identically, indicating that DNA sequences heavily decorated by H3K27 acetylation in early bud cells were progressively deacetylated and trimethylated at the same H3K27 residue in digit cells. We interpret this observation as reflecting the termination of the early phase of expression (Fig. 6B) (33).

The extensive telomeric H3K27 trimethylation occurred in parallel with the acetylation of H3K27 over the opposite, centromeric gene desert (12), raising the possibility that this massive telomeric H3K27 trimethylation may be necessary to push the regulatory balance toward the active centromeric domain and secure the switch, in particular for the Hoxd9, Hoxd10, and Hoxd11 swing genes. Alternatively, the implementation of the centromeric regulation in digits may in itself induce the switch, leading to the subsequent abrogation of the telomeric regulation and further H3K27 trimethylation of the gene desert. To discriminate between these alternatives, we looked at the H3K27me3 coverage in digits of mutant mice lacking the centromeric domain [del(Nsi-Atf2)] (12). In these mutant limbs, where the late regulation was no longer implemented, the telomeric desert remained largely trimethylated at H3K27, indicating that the termination of the telomeric regulation occurred independently from the activation of the centromeric regulation (Fig. 6B).

A Fine-Tuned Balance of Interactions

These results show that the HoxD cluster is the target of two global and independent regulations, implemented from both flanking gene deserts. A key factor determining the activity of any gene of the cluster is its tropism toward one or the other domain, a tropism depending upon its relative position within the cluster. In early forelimb cells, the interactions with the telomeric desert are maximal for Hoxd9, whereas they are minimal for Hoxd13, correlating with their transcriptional status (fig. S3). In this tissue, a conformational insulation of Hoxd13 from the early telomeric enhancers prevents any transcriptional leakage of this gene at this stage, a situation known to be deleterious for the limb morphology (34, 35). Conversely, the pervasive contacts established by Hoxd13 with the centromeric desert underlie its high level of transcription in digit cells, whereas Hoxd11 and Hoxd9 are transcribed at progressively lower levels (fig. S3) (14), likely due to resilient contacts with the telomeric region.

We challenged this polarizing effect of the flanking gene deserts by using the 3-Mb inv(Nsi-Itgα6) inversion, which completely disconnects Hoxd13 from the centromeric gene desert (Fig. 7). We compared the interactions established by Hoxd13 in a control proximal limb bud at E12.5 with those scored in the absence of the neighboring gene desert and observed both a slight loss of contacts centromeric to the gene cluster (Fig. 7) (from 70 to 56%) and a concomitant increase in the interactions with the telomeric gene desert on the opposite side (from 30 to 44%). This result suggests that the disconnection between the centromeric gene desert and the HoxD cluster decreased the strength and/or number of contacts between Hoxd13 and the new flanking DNA landscape, leading to a slightly increased tropism of this gene for the telomeric domain. This break in the conformational equilibrium occurred in parallel with an ectopic transcription of Hoxd13 in the proximal limb (34), sufficient to induce a measurable shortening of forearm bones due to the suppressive effect of HOXD13 over other Hox gene products in the proximal limb bud (34, 36). These data illustrate that it is necessary to keep Hoxd13 bound to the centromeric gene desert, even in the absence of transcription, to prevent its partial relocation toward the opposite regulatory domain and concurrent deleterious effects.

Fig. 7 Disrupting the equilibrium between telomeric and centromeric interactions.

Comparison between the profiles of 4C interactions (running mean) established by Hoxd13 in the proximal limb domain of either a control specimen (top track) or a mutant one carrying the inv(Nsi-Itgα6) inversion that relocates the centromeric gene desert several megabases away from the gene cluster (middle track). In this mutant limb, the centromeric regulation can no longer operate over the HoxD gene cluster due to the increase distance, and hence Hoxd13 expression is lost in digits. Because of this change in centromeric neighborhood, the interactions established by Hoxd13 on this side are reduced (56%; green bars on the left). All contacts with the original, now displaced, centromeric domain (in orange) are lost. In contrast, contacts are gained with the new centromeric neighborhood, which however do not fully compensate for the loss. Concomitantly, contacts are increased with the telomeric desert (44%; green bars on the right). The ratio between both conditions is shown below and illustrates the significant switch from a preference for the centromeric domain to rather equally distributed contacts between both sides (P = 0.005; log-likelihood ratio test for independence), suggesting that Hoxd13 is normally prevented from responding to the telomeric regulation through its sequestering contacts with the centromeric gene desert. Animals carrying this inversion express Hoxd13 ectopically in the proximal domain, leading to a weak mesomelic phenotype [scheme on the left and (34)].


Our results indicate that the correspondence between the positions of Hoxd genes within their genomic cluster and the proximal-to-distal sequence in the various pieces of appendicular skeleton that these genes determine is established by the antagonistic partition between two topological domains (Fig. 8A). In early limb buds, all cells are engaged into a telomeric regulation, which involves genes located centrally into the cluster. Hoxd13 and Hoxd12, to some extent, are preserved from this regulation due to their high tropism for the centromeric topological domain, which prevents them from establishing substantial telomeric contacts (Fig. 8B). Soon after the start of limb bud growth, in a subset of subapical and posterior cells, the telomeric regulation terminates, whereas the centromeric regulatory landscape becomes activated. Hoxd13 constitutively interacts with this topological domain and is up-regulated first, followed by a set of central genes swinging from one domain to the other, with decreasing strength up to Hoxd9. This process accounts for the collinear expression pattern originally described in limbs (8), with a proximal-only pattern for genes closer to the telomeric part of the cluster, a distal-only pattern for Hoxd13, and a combined pattern for genes located at a central position, which will adopt both configurations, but at different times and in different cells.

Fig. 8 A regulatory model for Hoxd gene collinearity in limbs.

(A) Two nonoverlapping regulatory landscapes control distinct, yet partially overlapping, sets of Hoxd genes during limb bud development. (B) During early bud outgrowth (E9.5), the telomeric domain (T-DOM) and its set of early limb enhancers (green ovals) induce the first wave of gene expression in distal and proximal progenitors (green domain) with the robust activation of Hoxd9, followed by Hoxd10 and Hoxd11, which are delayed due to their progressively increased proximity to the inactive centromeric domain (C-DOM). Hoxd13, which is constitutively part of the C-DOM, is closed, as illustrated by the presence of H3K27me3 (in red). The activity of these telomeric enhancers progressively decreases as the presumptive forearm grows. At E10.5, in a subset of subapical and posterior cells, the telomeric landscape becomes trimethylated at H3K27 (middle right scheme, red hexagons). At the same time, external signals activate the C-DOM (purple ovals). Hoxd13 starts to be transcribed and concurrently demethylated at H3K27, whereas Hoxd9, Hoxd10, and Hoxd11 change their contacts and relocate toward the active C-DOM. In presumptive digits (E11.5, left), the T-DOM and the inactive part of the cluster become trimethylated at H3K27 (red bar and hexagones). Cells where the telomeric regulation is terminated but where the switch was not implemented, due to an increased distance to the apical signaling, do not express any Hoxd genes and produce a zone that will subsequently turn into the wrist (black arrow). (C and D) Schematic representation of the switch, as illustrated by the transition of three swing genes (Hoxd9, Hoxd10, and Hoxd11) from T-DOM to C-DOM, along with C-DOM becoming active and T-DOM inactive. Hoxd13 is constitutively associated with C-DOM, whereas Hoxd1, Hoxd3, and Hoxd4 are constitutively associated with T-DOM. The position of the boundary between these two topological domains [taken and adapted from (16)] matches the group of genes that can go either direction, depending on which landscape is activated (enhancers are colored dark and light, whether active and inactive, respectively). The set of HOX proteins produced in response to the T-DOM regulation leads to the patterning of the arm and forearm, whereas the presence of HOXD13 protein, once the regulation has switched to the C-DOM, will terminate the system and produce digits (the corresponding morphologies are schematized within each of the topological domains).

The switch in functionality between these two constitutive topological domains may be considered as an allosteric-like transition (37), but applied to chromatin micro-architectures. Tissue-specific factors activating either of these two regulatory landscapes may act similarly to allosteric ligands, selecting between a limited number of preorganized 3D conformations and triggering the transformation of a topological domain from a negative to a positive state. In this view, the recruitment of a transcription factor by enhancer sequences would not cause a complete reorganization of the 3D structure, for example by looping. Instead, a preformed structure may help factors to recognize their target sequences and induce minor conformational changes accompanying the formation of an active state. The fact that the transition of a topological domain from a negative to a positive configuration did not involve dramatic variations in the interaction profiles supports this view. However, although the comparison between these various profiles in active and inactive tissues indeed showed minor changes in the distribution of contacts, some of the observed differences involved DNA sequences of critical importance for the regulatory specificity. For instance, when Hoxd13 becomes transcriptionally active, new contacts are established with one of the regulatory islands necessary for its full activity in digits (Fig. 5B, arrow, and fig. S4) (12, 16).

Thus, limb patterning by Hoxd genes involves a transition in the functional states of two flanking topological domains (Fig. 8, C and D). The existence of these two independent regulatory domains, one necessary to pattern the hand, the other required for patterning the forearm, further illustrates the modular nature of our appendages. It also provides a potential explanatory framework as to how these distinct modules may have evolved at different times, in response to various environmental conditions. In this context, however, the question of the articulation between these morphological modules is critical, because the evolutionary emergence of digits without an appropriate wrist may not have represented a strong adaptive value. It is noteworthy that the subpopulation of early limb bud cells where the telomeric regulation is maintained does not participate to the further distal extension of the limb bud and will thus remain in the proximal aspect of the developing limb (Fig. 8B, green domain at E11.5). By contrast, the distal growth leading to digit development (31) will be achieved by cells implementing the centromeric regulation. Because this distal growth depends on factors released by both the apical ectodermal ridge (AER) and the zone of polarizing activity (31, 38), cells located near these signaling centers will switch toward the centromeric regulation.

Because the termination of the early phase of Hoxd gene expression does not require the activation of the late phase, the possibility exists for a physical separation to occur between the cells implementing the telomeric regulation, which will stay behind (Fig. 8B, green area), and cells implementing the centromeric regulation, which will be under the influence of the AER and hence localized at the most distal aspect of the limb bud (Fig. 8B, purple area). This segregation between the two types of Hoxd expressing cells leads to the appearance of an intermediate cellular territory where the global dose of HOXD proteins is minimal because cells escape both regulatory controls (Fig. 8B, arrow). This zone of low HOX protein content generates the wrist (39), which is made out of small roundish bones used to articulate the two sets of long bones found in both the forearm and the hand. Previous analyses of dominant negative mutations in both mouse and human (36, 40), which severely reduce the functional contributions of HOX proteins, have revealed a transformation of metacarpal bones into a carpal-like morphology with the loss of cortical ossification and the formation of lateral joints (40, 41). Therefore, the morphological transition between the two sets of long bones in our arms coincides with a transition between two regulatory modalities, at the boundary of topological domains. This intriguing correspondence may explain why this apparently complex system of two distinct regulations was selected and fully implemented in tetrapod species: It contained the intrinsic capacity for a built-in articulation to evolve concomitantly, which may have given a novel tetrapod appendage its full adaptive value.

Supplementary Materials

Materials and Methods

Figs. S1 to S4

Tables S1 to S5

References (4252)

References and Notes

  1. Acknowledgments: We thank I. Barde, S. Verp, and A. Quazzola, from the Ecole Polytechnique Fédérale de Lausanne transgenic platform, for their help; M. Docquier and P. Descombes, from the National Center in Genetics genomic platform; U. Gunthert and L. Thévenet for the CD44 and inv(attP-CD44) mice, respectively; B. Ren for permission to use and adapt data; and J.-P. Changeux for discussions. Computations were performed at the Vital-IT ( computing center of the Swiss Institute of Bioinformatics using tools described in We also thank M. Friedli, N. Soshnikova, E. Joye, P. Schorderet, N. Lonfat, and other members of the Duboule laboratories for discussions and sharing reagents. This work was supported by the Ecole Polytechnique Fédérale de Lausanne, the University of Geneva, the National Center for Competence in Research Frontiers in Genetics, the Boninchi Foundation, and European Research Council grant (to D.D.). The GEO accession number for the data is GSE454570. Authors' contributions: G.A., T.M., and D.D. designed the experiments; experiments were mostly performed by G.A., with some performed by T.M. F.G., B.M., and F.S. provided mutant mice. D.N. provided material and helped with 4C experiments. M.L. helped with data analysis, and D.T. collaborated on the transgenic screen. G.A. and D.D. wrote the manuscript with input from other coauthors.
View Abstract

Stay Connected to Science

Navigate This Article