The Dynamic Architecture of Hox Gene Clusters

See allHide authors and affiliations

Science  14 Oct 2011:
Vol. 334, Issue 6053, pp. 222-225
DOI: 10.1126/science.1207194


The spatial and temporal control of Hox gene transcription is essential for patterning the vertebrate body axis. Although this process involves changes in histone posttranslational modifications, the existence of particular three-dimensional (3D) architectures remained to be assessed in vivo. Using high-resolution chromatin conformation capture methodology, we examined the spatial configuration of Hox clusters in embryonic mouse tissues where different Hox genes are active. When the cluster is transcriptionally inactive, Hox genes associate into a single 3D structure delimited from flanking regions. Once transcription starts, Hox clusters switch to a bimodal 3D organization where newly activated genes progressively cluster into a transcriptionally active compartment. This transition in spatial configurations coincides with the dynamics of chromatin marks, which label the progression of the gene clusters from a negative to a positive transcription status. This spatial compartmentalization may be key to process the colinear activation of these compact gene clusters.

During mammalian development, Hox genes are activated sequentially relative to their positions along the four genomic clusters (HoxA to HoxD). As a result, this process leads to a corresponding distribution of transcripts along the rostral-to-caudal body axis. This process of colinear activation is essential for the organization of the body plan (1, 2). Accompanying this process, a dynamic transition occurs in the chromatin microenvironment, from a repressive (histone H3K27me3) to a transcription-permissive (histone H3K4me3) state (3). Changes in higher-order chromatin organization have been reported to accompany the transcription of developmentally relevant genes (4), and the 3D organization of the HoxA cluster is changed upon gene activation in mammalian cultured cells (57). Furthermore, the HoxB and HoxD clusters adopt a decondensed conformation along with gene activation (8, 9) accompanied by modifications of the Polycomb repressive complex 1, as shown in cultured cells (10). We analyzed the architectures of these genomic loci in embryonic tissues at different stages of the colinear transcriptional activation and describe a bimodal state, where active and inactive genes are found in distinct three-dimensional (3D) domains and genes progressively lose their interactions with the repressive domain to associate with a transcriptionally active structure.

We used gene expression microarrays to compare Hox gene activity in three tissue samples obtained from embryonic day 10.5 (E10.5) mouse embryos: “anterior” dorsal trunk cells (from upper forelimb to upper hindlimb levels), “posterior” dorsal trunk cells (from upper forelimb level to tailbud), and forebrain cells. The latter cells do not express any Hox genes and were used as negative control (Fig. 1A, fig. S1, and table S1). We determined which genes were either transcribed or silent in these samples and positioned the dissection limit between the two trunk samples approximately at the level of the Hoxd10 expression boundary (Fig. 1A, arrowheads; fig. S1 and table S2).

Fig. 1

The inactive HoxD cluster forms a discrete 3D compartment. (A) Schematized E10.5 mouse embryo highlighting tissue samples used in this work: forebrain (green), anterior trunk (red), and posterior trunk (blue). Approximate positions of expression boundaries for either Hoxd9 (d9), Hoxd10 (d10), or Hoxd11 (d11) are indicated with arrowheads. (B) Running-mean 4C-seq interaction patterns of Hoxd9 in a 4-Mb-large genomic region surrounding the HoxD cluster in forebrain tissue. Significant interactions are depicted below (see also fig. S6). The position of the HoxD cluster is shown in red, flanked by two gene deserts (below). (C) Quantitative local 4C-seq interactions reveal the local 3D domain of the inactive HoxD cluster (forebrain). Below, H3K27me3 signal is aligned. Dashed lines emphasize the discrete borders of the local 3D domain and the coincidence with the H3K27me3 domain. Three viewpoints are used (Hoxd13, Hoxd9, and Hoxd4, from top to bottom) and are indicated with arrowheads. Excluded regions around these viewpoints are depicted with vertical light gray boxes. The locations of Hoxd genes (red) and of other transcripts (black) are shown below.

Using these tissue samples, we examined the 3D architecture by high-resolution chromatin conformation capture [Multiplex 4C-seq (11, 12); fig. S2 and table S3] with multiple genes as anchor points (“viewpoints”), taken in all four Hox clusters (figs. S3 to S5 and table S4). Although most of the intrachromosomal associations were restricted to the Hox clusters themselves (Fig. 1B and fig. S6), a statistical algorithm reliably identified nondynamic long-range interaction landscapes surrounding each locus and extending slightly past the flanking gene deserts (13), which may reflect a generic organization of Hox clusters and their surroundings in these cells (Fig. 1B, fig. S6, and table S5).

We quantified intracluster 3D organization at highest resolution [figs. S5 and S7 (11)], using the HoxD cluster in forebrain cells, where all Hox genes are inactive. Seven different viewpoints revealed comparable domains of 3D association, spanning from Evx2 to a few kilobases downstream of Hoxd1 (Fig. 1C and fig. S7). Likewise, three viewpoints within each of the other Hox clusters uncovered association domains covering the entire clusters plus a few kilobases on either side (fig. S8). Therefore, silent Hox clusters form 3D compartments with discrete separation from flanking DNA regions. Little specific organization was scored within these domains, suggesting mostly random contacts. Furthermore, these association domains precisely matched the distribution of H3K27me3 marks decorating these loci (Fig. 1C and figs. S7 to S9), both in the positions of the borders and in the organization within each cluster, supporting a functional interplay between these two parameters in vivo (10).

We examined these architectures in “anterior” and “posterior” embryonic tissue samples, where different HoxD genes are transcribed at this stage of development (Fig. 2 and fig. S10). In contrast to the single interaction domain observed in brain cells (Fig. 2, A to C, green), both anterior (in red) and posterior (in blue) trunk cells generated bimodal profiles of association, dividing the gene cluster into two distinct 3D compartments. However, the boundaries between these two compartments were located at different positions in anterior versus posterior trunk samples. In anterior trunk, transcribed genes like Hoxd4 no longer contacted the silent (centromeric) part of the cluster, thus forming an “active domain” (Fig. 2A). Accordingly, Hoxd13, the most centromeric gene, no longer contacted the telomeric part of the cluster (Fig. 2C, “inactive domain”). The transition of genes from an inactive to an active 3D domain was best exemplified by Hoxd9, expressed strongly in posterior trunk cells but only weakly in the anterior sample (Fig. 2B and fig. S1). Contacts of Hoxd9 were stronger with the centromeric part of the cluster in anterior cells (gene mostly off; in the “inactive domain”), but they clearly shifted toward the telomeric part of the cluster in posterior trunk cells (gene on, “active domain”; Fig. 2B, see ratio). The same was observed for Hoxd11, which strongly contacted the negative domain in anterior trunk cells (off state), whereas most contacts were with the positive domain in posterior trunk cells (on state; fig. S10). This bimodal organization also applied to the other three Hox clusters, with slight variations in the location of the internal boundary, depending on the progression of gene activation within each cluster (fig. S11).

Fig. 2

Dynamic architectures of the HoxD cluster at different stages of colinear activation. The frequencies of associations are shown with Hoxd4 (A), Hoxd9 (B), or Hoxd13 (C) as viewpoints in forebrain (profiles in green), anterior trunk (profiles in red), or posterior trunk (profiles in blue) tissues. The ratios of interactions are indicated between the profiles for the same viewpoint in different embryonic tissues. The colinear expression status (blue, active; red, inactive) of Hoxd genes (bottom line) is schematized below each profile. For the anterior trunk sample (in red), the corresponding H3K27me3 and H3K4me3 signals are indicated just below (profiles in black). Residence of viewpoints in the active [(A), anterior trunk Hoxd4 and (B), posterior trunk Hoxd9] and inactive [(B), anterior trunk Hoxd9 and (C), anterior trunk Hoxd13] domains is indicated with arrowheads.

In the trunk samples, not only did we observe a coincidence between the inactive 3D domain and the extent of H3K27me3 modifications, as in the brain sample, but the active compartments also matched the presence of H3K4me3 chromatin domains (Fig. 2, A to C, and figs. S9 to S11). The distribution of these chromatin marks correlated with the 3D organization at these Hox clusters. In this context, the HoxB cluster was particularly interesting because an 80-kb, repeat-rich intergenic region separates Hoxb13 from the rest of the cluster (14). Using viewpoints in the cluster and within this intergenic region, we observed a weak association only (if any) between this region and the rest of the HoxB cluster (fig. S12), suggesting that it loops out from this bimodal architecture. The same interruption was seen in the distribution of H3K27me3 marks (fig. S12), illustrating again the precise correspondence between chromatin marks and 3D architecture and showing that these spatial domains do not necessarily involve an uninterrupted linear chromatin fiber. This latter conclusion was further illustrated by strongly increased and targeted associations between the inactive domain of the HoxD cluster (Hoxd13) and the Dlx1 locus, which are separated by a distance of 3 Mb and both heavily decorated with H3K27me3 marks (fig. S13). In contrast, such interactions were not scored with the active part of the HoxD cluster.

From these data sets, we propose a model whereby Hox genes move stepwise from an inactive compartment marked by H3K27me3 to another, transcriptionally active domain labeled with H3K4me3 marks (Fig. 3). We challenged this view by using two deletions in vivo where Hox gene activities are differentially perturbed (15). Deletion of the Hoxd8 to Hoxd10 DNA fragment [Del(8–10)] does not severely change the expression of Hoxd11 (fig. S14). However, the additional deletion of the intergenic region i [Del(i8-10)] results in a strong activation of Hoxd11 in anterior tissues (fig. S14). These overlapping deletions thus have distinct transcriptional outcomes, with Hoxd11 ectopically activated in the anterior trunk sample of Del(i8-10) embryos only. We first assessed whether such deletions had changed the overall cluster architecture in brain cells (Fig. 4A and figs. S15 and S16) and observed that the inactive domains maintained the same borders on both sides, indicating that the mechanism underlying this 3D compartmentalization is likely intrinsic to the gene cluster. We then studied the interaction profiles using Hoxd11 and Hoxd4 as viewpoints in anterior trunk samples (Fig. 4, B and C, and fig. S17). In the Del(i8-10) mutant, where Hoxd11 is ectopically activated (15), the association between Hoxd11 and the “positive” compartment was strongly increased. This was scored either by using Hoxd11 as a viewpoint (Fig. 4B, shaded purple/blue ratio), or Hoxd4 (Fig. 4C, shaded purple/blue ratio). However, contacts remained as in wild-type embryos when the shorter Del(810) deletion was analyzed with the same viewpoints (Fig. 4, B and C, shaded blue/yellow ratios). Ectopic activation, rather than a deletion per se, was thus paralleled by enhanced association between Hoxd11 and the “active” anterior domain.

Fig. 3

Model of the 3D organization of Hox gene clusters, at various stages of colinear gene activation. Transcriptionally inactive genes are depicted in red and active genes in blue. Gene activation is paralleled by a transition from one 3D domain, matching the presence of H3K27me3, to another domain of active transcription (marked with H3K4me3). Although the same dynamics are observed for the HoxA, HoxC, and HoxD clusters (left), the HoxB cluster (right) shows a slight variation with a large piece of intergenic DNA looping out from these two domains.

Fig. 4

Ectopic activation of Hoxd11 in anterior tissue increases its association frequency with other active Hoxd genes. (A) Three-dimensional organization of the inactive HoxD cluster in wild-type forebrain (blue profiles, middle) or in forebrain tissues carrying two distinct internal deletions into the HoxD cluster [Del(i8-10), profiles in purple on the top, and Del(8-10), profiles in yellow, bottom]. The deletions are indicated between small arrowheads and, for each sample, both Hoxd11 and Hoxd4 are used as viewpoints (large arrowheads). Regardless of cluster size, the 3D inactive domains remain demarcated by the same outside borders. (B) Three-dimensional organization of mutant HoxD clusters in anterior trunk with Hoxd11 as a viewpoint (arrowhead). Hoxd11 is expressed ectopically in the Del(i8-10) mutant anterior tissue, but not in the Del(8-10) mutant (see figs. S14 to S17). Accordingly, increased association frequencies are observed between Hoxd11 and the active part of the cluster in the Del(i8-10) mutant (purple box and ratio), as compared to both wild-type (in blue) and the Del(810) mutant embryos (yellow box and ratio). (C) Same experiment as in (B), but with Hoxd4 as a viewpoint (arrowhead). Again, interactions are increased between Hoxd4 and the posterior part of the HoxD cluster (purple box and ratio) containing Hoxd11, which is ectopically expressed in the mutant Del(i8-10) anterior tissue. In contrast, the Del(8-10) mutant tissue, where Hoxd11 is not expressed anteriorly, does not show such increased interactions (yellow box and ratio).

This work suggests that the colinear activation of Hox genes involves a stepwise transition of each gene from a negative to a positive compartment, which display different biochemical properties and thus results in a physical separation of their regulatory modalities. Although it remains to be fully demonstrated whether such a process underlies colinear activation or is a consequence of it, it is noteworthy that the former possibility would provide a mechanistic solution to three crucial problems encountered during the activation of this gene family: (i) to ensure a proper colinear sequence in gene activation, such that axial morphologies are respected [see e.g., (16)]; (ii) to prevent the most posterior genes from being activated too early, which leads to deleterious phenotypes (17); and (iii) to fix and memorize transcriptional states at various body levels. These critical constraints are well addressed by our cis-acting model, whereas other potential mechanisms, such as relying upon trans-acting interactions, may not allow the same level of precision and reliability.

Supporting Online Material

Materials and Methods

Figs. S1 to S17

Tables S1 to S6

References (1824)

References and Notes

  1. Material and methods are available as supporting material on Science Online.
  2. Acknowledgments: We thank B. Mascrez for assistance with mouse handling and genotyping, P. Descombes and members of the National Research Centre genomics platform for high-throughput sequencing, and members of the Duboule laboratories for discussion. Computations were performed at the Vital-IT Center for high-performance computing ( at the Swiss Institute of Bioinformatics. This work was supported by funds from the Ecole Polytechnique Fédérale (Lausanne), the University of Geneva, the Swiss National Research Fund, the National Research Centre “Frontiers in Genetics,” and the European Research Council grant (to D.D.). Data are all based on ENSEMBL Mouse assembly NCBIM37. 4C-seq patterns can be obtained from or Microarray and ChIP-seq data have been submitted to the Gene Expression Omnibus (GEO) repository ( under accession no. GSE31570.
View Abstract

Navigate This Article