Epigenetic Temporal Control of Mouse Hox Genes in Vivo

See allHide authors and affiliations

Science  05 Jun 2009:
Vol. 324, Issue 5932, pp. 1320-1323
DOI: 10.1126/science.1171468


During vertebrate development, the temporal control of Hox gene transcriptional activation follows the genomic order of the genes within the Hox clusters. Although it is recognized that this “Hox clock” serves to coordinate body patterning, the underlying mechanism remains elusive. We have shown that successive Hox gene activation in the mouse embryo is closely associated with a directional transition in chromatin status, as judged by the dynamic progression of transcription-competent modifications: Increases in activation marks correspond to decreases in repressive marks. Furthermore, using a mouse in which a Hox cluster was split into two pieces, we document the necessity to maintain a clustered organization to properly implement this process. These results suggest that chromatin modifications are important parameters in the temporal regulation of this gene family.

Hox genes, which are generally arranged in clusters at genomic loci, are essential for patterning the anterior to posterior animal body axis (13). In vertebrates, these genes are activated in a time sequence that follows their physical order within the cluster, a process referred to as temporal collinearity (4). This property is observed in animals developing their trunk via a rostral to caudal time sequence, yet the underlying molecular mechanism is elusive (5, 6). A progressive transition in chromatin state was hypothesized (7, 8), whereby an initially repressed configuration becomes open for transcription. The subsequent observation of chromatin decondensation at these loci when transcription is induced supported this hypothesis (9).

Hox genes are repressed by Polycomb group (PcG) proteins (10). Mutation of PcG genes induces ectopic Hox expression and results in posterior homeotic transformations (11, 12). PcG proteins form large complexes with histone-modifying activities; for example, Polycomb Repressive Complex 2 (PRC2) trimethylates histone H3 at lysine 27 (H3K27me3) (1316), an essential modification for long-term repression of target genes. In contrast, Trithorax group (TrxG) proteins antagonize PcG proteins and activate target gene expression (10). TRX complexes trimethylate histone H3 at lysine 4 (H3K4me3), a mark generally associated with active transcription (17). Genome-wide studies of both H3K27me3 and H3K4me3 modifications in embryonic stem cells (ESC) and other cultured cells have revealed specific profiles during the maintenance phase of Hox gene expression in vitro (1821). We looked at the in vivo dynamics of chromatin marks during the sequential activation of Hoxd genes in developing murine tail buds.

We dissected out mouse tail buds during late somitogenesis when the last Hox genes become transcribed (22) and performed expression profiling at E8.5 (embryonic day 8.5), E9.0, and E9.5 (Fig. 1A) using tiling arrays covering 2 Mb of DNA containing the HoxD cluster. This highly syntenic region (23) also contains four ubiquitously expressed genes, Atp5g3, Lnp, Mtx2, and Hnrpa3, and two gene deserts (fig. S1). Transcription of Hoxd1 to Hoxd9 was active at all three time points (Fig. 1B), reflecting the onset of Hox gene transcription during early gastrulation. However, transcriptional progression was observed for more posterior genes, with Hoxd10 and Hoxd11 transcribed at E9.0 (Fig. 1B), whereas by E9.5 transcriptional activity had spread over Hoxd12, Hoxd13, and the nearby neighbor gene Evx2 (Fig. 1B). Low transcript levels were detected for Hoxd13 before activation of Hoxd10 (Fig. 1B, arrow).

Fig. 1

Collinear activation of Hoxd genes during axial development. (A) E8.5, E9.0, and E9.5 embryos, with dissected samples indicated in red. Scale bar, 200 μm. (B) Transcript profiles on tiling arrays using reverse-transcribed total RNA (green). Bound RNA Pol II (orange) and the AcH3 pattern (magenta) are also displayed for the HoxD cluster. The y axis indicates the log2 ratio of cDNA/genomic DNA or ChIP-enriched/input signal intensity.

We mapped the sites occupied by RNA polymerase II using chromatin immunoprecipitation combined with hybridization on tiling array (ChIP-chip) (Fig. 1B and fig. S1). The Pol II profile corresponded to transcribed regions; whereas virtually no Pol II was scored centromeric to Hoxd10 at E8.5, signals were detected for both Hoxd10 and Hoxd11 at E9.0. At E9.5, the whole centromeric part of the cluster was fully occupied by Pol II (Fig. 1B), indicating that it was recruited in a collinear manner too. In agreement with transcript profiling, a weak Pol II binding was scored at the Hoxd13 locus at E8.5. Similarly, high levels of H3K9/K14 acetylation (AcH3) were found in E8.5 tail buds, covering from Hoxd1 to Hoxd9 (Fig. 1B). However, AcH3 marks were also scored over the silenced Hoxd10 and Hoxd11 (Fig. 1B, arrowheads). Along with the transcriptional activation of these two genes, the levels of AcH3 were expanded to cover the entire gene cluster by day 9.5, matching both the presence of Pol II and the robust transcription of the Hoxd10 to Hoxd13 interval. As for Hox genes, AcH3 modification appeared at the Evx2 locus before Pol II binding was scored, consistent with a role for this modification in transcriptional initiation (17).

We next investigated the status of both H3K27 and H3K4 trimethylation. H3K27me3 levels were assessed in E8.5 and E9.5 tail buds and in ESC. In ESC, consistent with previous studies (19, 20), H3K27me3 marks, associated with transcriptional repression, covered the entire gene cluster (Fig. 2). Accordingly, transcription of Hoxd genes was not detected in these cells (Fig. 2, RNA). During collinear activation in tail buds, a complete loss of H3K27me3 mark was progressively observed over the telomeric part of the cluster, initially from Hoxd1 to Hoxd4 at E8.5 and subsequently extending until Hoxd11 at E9.5 (Fig. 2). Because H3K27me3 disappeared upon gene activation, we conclude that tail bud cells that do not express any Hox gene do not implement this repression. H3K27me3 marks slightly overlapped with transcriptionally active regions, for example, over the Hoxd11-Hoxd12 loci at E9.5 (Fig. 2 and fig. S2), likely illustrating some temporal heterogeneity in the activation of Hoxd genes within neighboring cells. Also, samples may have included mixtures of cells expressing and cells not expressing a particular Hox gene because of the anterior-posterior extent of the dissected domains. Finally, the H3K27me3 signals were higher over the silenced part of the Hox gene cluster in E8.5 tail bud cells than in ESC, suggesting that a tighter repression is implemented during axial development.

Fig. 2

Chromatin marks during temporal collinearity. Transcriptional activities (green) are shown in ESC and in E8.5 and E9.5 embryos. In ESC, a 120-kb domain is decorated by H3K27me3 (blue), yet at rather low density. In the embryo, H3K27me3 marks progressively retract from the telomeric to the centromeric extremity of the cluster. In addition, levels of H3K27me3 modifications found over silenced genes are higher than in ESC. Low enrichment (log2 ≤ 1) for H3K4me3 (red) marks all CpG islands within HoxD in ESC, as opposed to the strong levels (log2 = 4) detected over Hoxd1 to Hoxd9 at E8.5. An increase (log2 ≤ 3) in H3K4me3 marks was also detected at the silenced Hoxd10, Hoxd11, and Evx2 loci at this early stage, in the absence of detectable transcripts.

We checked whether this progressive demethylation of H3K27 was paralleled by an increased trimethylation of H3K4. Consistent with previous results (24), we scored low levels of H3K4me3 over HoxD in ESC, with residual signals on CpG islands (Fig. 2). In E8.5 tail buds, however, H3K4me3 marks drastically increased and covered the telomeric part of the cluster up to both Hoxd10 and Hoxd11, which are still silenced at this time point (Fig. 2). Therefore, as for AcH3, H3K4me3 marks were detected before Pol II binding and prelabeled future sites of transcription. At E9.5, elevated levels of H3K4 trimethylation over the Hoxd12 and Hoxd11 loci corresponded to their robust transcriptional activation (Fig. 2). Altogether, temporal collinearity in tail buds corresponds to chromatin dynamics, progressing along the cluster and involving the removal of H3K27me3 marks, the methylation of H3K4, and the acetylation of H3. Transcriptional activation along the gene cluster occurs within a region of transition between H3K27me3 and H3K4me3 marks, a window that shifts with time toward the centromeric extremity of the cluster (fig. S2).

This collinear chromatin dynamic suggests a mechanism whereby modifications would spread from the telomeric extremity of the cluster to the opposite end. We assessed this possibility by using mice where the HoxD cluster is split into two pieces, separated by a 3-Mb inversion (25) (Fig. 3A). In this configuration, the Hoxd11 to Hoxd13 region becomes isolated from the rest of the cluster. This allows a test of whether early establishment and dynamic progression of both H3K27me3 and H3K4me3 marks require an integral gene cluster.

From Hoxd1 to Hoxd9, the mutant (inv) transcript profile was as in wild type (Fig. 3B), demonstrating that cis regulations required to initiate transcription of these genes either lie within this segment of the cluster or are telomeric (26). However, differences were observed close to the break point. First, Hoxd10 was transcribed at E8.5, whereas this gene is normally silent at this stage. Second, ectopic antisense transcripts were detected in the small posterior half-cluster (fig. S3A), likely triggered by the new genomic neighborhood. As in wild type, premature Hoxd13 transcription was scored in inv mutants, indicating that this transcriptional activity does not require telomeric-located regulatory sequences controlling other genes of the cluster. This late-occurring leakage in temporal collinearity may reflect the spurious activity of enhancers located nearby and dedicated to strongly activate this gene in subsequent morphological contexts (27). In E9.5 inv tail buds, neither Hoxd12 nor Hoxd11 showed transcriptional increase (fig. S3B). Their expression was slightly elevated between E8.5 and E9.5 but remained very low when compared to the wild-type situation. On the other side of the breakpoint, Hoxd10 transcription peaked as in wild type, whereas transcripts located upstream of Hoxd10 and originating from the Hoxd11 locus were reduced in amount due to the break (fig. S3B).

Fig. 3

Transcriptional activation in a split HoxD cluster. (A) Scheme of the Integrin-alpha6 (Itga6)-HoxD inversion (in gray). Dashed lines and arrowheads indicate both breakpoints (BP1 and BP2), either between exons 24 and 26 of Itga6 or between Hoxd11 and Hoxd10. The inversion positions the centromeric Hoxd13 to Hoxd11 DNA segment 3 Mb away from the rest of the cluster. (B) Transcript profiles (green) in wild-type (wt) and mutant (inv) tail buds at E8.5. Animals with a split cluster activate Hoxd10 prematurely (E8.5, *1). Only the RNA representing the Hox-coding DNA strand is shown (see also fig. S3A). The position of both the break point (BP2) and the 3-Mb interval is indicated. (C) Enlargement of the Hoxd13 to Hoxd8 region shown in (B) with both H3K4me3 (red) and H3K27me3 (blue) profiles in E8.5 tail buds. Elevated levels of H3K4me3 marks are scored over Hoxd10, Hoxd11, and Hoxd12 (black arrows and arrowheads). H3K27me3 marks are reduced at and around the Hoxd10 locus (red arrowheads) and, to a lesser extent, over Hoxd11 and Hoxd12 (red arrows).

ChIP-chip analyses of H3K4me3 distribution in E8.5 homozygous inv tail buds showed enrichment over the Hoxd12 to Hoxd10 region, on both sides of the break point (Fig. 3C), whereas the profile from Hoxd9 to Hoxd1 was comparable to wild type. The robust gain in H3K4me3 marks over Hoxd12 was not scored in older wild-type tail buds (fig. S2) and did not match any transcriptional activity, neither for Hoxd12 nor for Hoxd11 (Fig. 3B). In this case, both Hoxd11 and Hoxd12 were ready to be transcribed (28), yet they remained silent because they were moved away from the required enhancer sequence located telomeric to the breakpoint. In contrast, increased H3K4 trimethylation on the other side of the breakpoint (Fig. 3C) matched the premature activation of Hoxd10.

The DNA interval decorated by H3K27me3 marks in inv mutants was virtually identical to wild type (Fig. 3C), indicating that an integral cluster is not necessary to define the initial extent of the repressive domain; H3K27me3 marks were positioned over posterior genes even though these genes were disconnected from the rest of the cluster, thus ruling out the existence of a spreading mechanism sensu stricto for the implementation of this repression. In addition, the overall density of these marks on both sides of the break point was considerably below the wild-type situation (Fig. 3C). In the posterior half-cluster, H3K27me3 marks were distributed almost as in wild type over Evx2 and Hoxd13, whereas a decrease was scored over the Hoxd12 to Hoxd11 intergenic region and 3′ to Hoxd11 (Fig. 3C). In the anterior half-cluster, a similar reduction was detected at the Hoxd10 locus, consistent with its premature activation and, to a lesser extent, over Hoxd9 (Fig. 3C). This weakening in H3K27me3 signal over Hoxd10 was not observed at the wild-type locus, even in older tail buds (fig. S2). The general decrease in H3K27 trimethylation around the break point suggests that a dense coverage of the HoxD cluster by this histone modification requires an intact clustered configuration. Whereas isolated parts of the gene cluster can be trimethylated at H3K27 independently of one another, these various parts may cooperate and synergize to mediate a dense pattern of methylation, potentially through local cis interactions.

These results shed light on the general regulatory strategy implemented by Hox gene loci during the earliest steps of mouse trunk development. Unlike in Drosophila, mammalian Hox gene loci appear refractory to transcription before transcription initiates, as indicated by high levels of H3K27me3 marks covering the HoxD locus early on. This likely reflects the necessity to prevent the premature activation of posterior genes at a time when anterior structures are being determined, which would be deleterious to the embryo. During gastrulation, this repression is counteracted by an activity progressing from the telomeric to the centromeric extremity of the cluster, illustrated by both an elevation of H3K4me3 level and the demethylation of H3K27me3. The region of transition between these two states of chromatin corresponds to the dynamic window wherein Hoxd genes become transcriptionally active. Alternatively, Hox genes could be activated from a persisting pool of nonexpressing stem cells. In this view, the chromatin modifications observed in our samples reflect the average of successive waves of transcriptional activation rather than a dynamic process occurring in the same cells. We do not favor this possibility because such a pool of Hox-negative cells would constitute a large fraction of the tissue sample, yet it has never been observed in gastrulating tail buds. Also, the nucleosomes of these stem cells would lack the repressive marks over the HoxD cluster, unlike in ESC. Finally, Hox genes are activated in cells already expressing more anterior combinations thereof.

We have shown that gene clustering is not necessary for the initial definition of the H3K27me3 landscape. However, clustering is required for a full repression to be consolidated and/or maintained over the cluster, which suggests a synergistic effect due to Hox genes’ density. Likewise, whereas an integral cluster appears dispensable for selecting the sites of H3K4 trimethylation, gene clustering helps the coordination of this general transition in chromatin status because split clusters displayed premature H3K4me3 marks on either side of the breakpoint. Although the gain of H3K4me3 and the concurrent weakening of H3K27me3 at the mutant Hoxd10 locus coincided with its early ectopic transcription, similar imbalances at the inverted Hoxd11 and Hoxd12 loci did not elicit the same transcriptional response. From this, we conclude that H3K4me3 chromatin modification is necessary but not sufficient for proper Hox gene transcriptional control and that remote enhancer sequences must have contributed to the maintenance of clustered organization during animal evolution.

Supporting Online Material

Materials and Methods

Figs. S1 to S3


References and Notes

  1. We thank T. Montavon, A. Puglisi, and D. Schübeler for advice; F. Chabaud for cell culture; and P. Descombes and members of the Genomics Platform for their help with tiling arrays. N.S. was supported by a European Molecular Biology Organization long-term fellowship. This work was funded by the University of Geneva, the Federal Institute of Technology in Lausanne, the Swiss National Research Fund, the National Research Center Frontiers in Genetics, and the European Union program Crescendo. Data and analysis are available for download from ArrayExpress (accession E-TABM-677).
View Abstract

Navigate This Article