Nuclear reorganisation and chromatin decondensation are conserved, but distinct, mechanisms linked to Hox gene activation

Chromatin condensation and organisation in the nucleus are considered important facets of the regulation of gene activation and gene silencing. Nuclear reorganisation of genes relative to `transcription factories'(Osborne et al., 2004) and chromosome territories (CTs) has been correlated with active transcription. In the latter case, fluorescence in situ hybridisation (FISH) detects a relocalisation of some active genes to positions beyond the visible limits of CTs (Volpi et al., 2000; Williams et al., 2002; Mahy et al., 2002). An`opening', or decondensation, of higher-order chromatin structure is also thought to be linked to gene activation(Mohd-Sarip and Verrijzer,2004). Indeed, it has been suggested that the `looping out' of gene loci from CTs is a visual manifestation of a large-scale chromatin decondensation (Volpi et al.,2000; Gilbert et al.,2004).

To understand how changes in complex chromatin structure relate to the expression of genes, and to work towards an understanding of the mechanisms that initiate and regulate them, requires an inducible system of co-regulated gene expression. One system that affords this is Hox gene loci. Transgene studies and genetically induced rearrangements of Hox loci have led to a suggestion that there might be a progressive opening of chromatin structure through Hox loci from 3′ to 5′ that controls sequential gene activation colinear with the order of the genes on the chromosome(van der Hoeven et al., 1996; Kmita and Duboule, 2003). Support for this model has come from our previous analysis of the murine Hoxb locus both ex vivo, during the differentiation of embryonic stem(ES) cells, and in vivo in cells of the primitive streak and adjacent mesoderm during early embryogenesis and also along the rostrocaudal (head to tail) axis of the embryo later in development. In both cases, there is visible chromatin decondensation and a looping out of Hoxb genes from their CTs as the expression of Hoxb is induced(Chambeyron and Bickmore, 2004; Chambeyron et al., 2005).

If these large-scale chromatin remodelling events are fundamental to the regulation of Hox loci, and possibly also their colinear expression, then they should be conserved between paralogous gene clusters. The four mammalian autosomal Hox gene loci are the result of an ancestral duplication that occurred at the origin of vertebrate evolution(Ferrier and Minguillon, 2003). Each cluster still exhibits temporal and spatial colinearity of gene expression along the rostrocaudal axis of the trunk. However, Hoxd,but not Hoxb, genes have been co-opted more recently in evolution into the regulation of developing limbs and external genitalia and this required the appearance of novel regulatory elements around Hoxd. An enhancer controlling expression of Hoxd10-d13 and the adjacent Lnp gene in digits and in the distalmost part of the limb buds has been identified ∼160 kb 5′ of Hoxd(Spitz et al., 2003; Spitz et al., 2005). Similarly, control regions 3′ of Hoxd have been postulated to control colinear Hoxd expression in the nascent limb bud(Fig. 1)(Zakany et al., 2004; Deschamps and van Nes, 2005)and gut (Spitz et al., 2005). Limb malformations in humans with chromosomal rearrangements 5′ and 3′ of HOXD also suggest that regulatory elements in the large-scale genomic context around the locus are important for correct gene expression or for protection from position effects(Spitz et al., 2002; Dlugaszewska et al., 2006).

No such regulatory elements flanking Hoxb have been described. Therefore, it was unclear whether chromatin decondensation and looping out from the CT, similar to that at Hoxb, would occur at Hoxd,nor whether the nuclear reorganisation accompanying Hoxd expression would be the same along the rostrocaudal embryonic axis and in the limb bud.

Here, we have analysed the nuclear organisation of a ∼1 Mb region around the Hoxd cluster in both differentiating ES cells and the embryonic day (E) 9.5 embryo. As at Hoxb, we see chromatin decondensation and a looping out of the Hoxd region from its CT,suggesting that these are part of ancestral chromatin-based mechanisms for regulating Hox loci. However, contrary to expectations, we have dissociated these two facets of nuclear reorganisation from each other. During ES cell differentiation we show that movement of the Hoxd region towards the outside of the CT can occur before its apparent chromatin decondensation. In the embryo, both looping out and decondensation occur in the tailbud region,but in the limb bud there is decondensation of the Hoxd region in the absence of discernable looping out. This may reflect different modes of Hoxd regulation that occur in different developmental contexts.

OS25 ES cells allow for the selection of undifferentiated(Oct4-expressing) cells using Hygromycin (hyg), and differentiated cells using Ganciclovir (selection against Oct4 expressing cells)(Billon et al., 2002). They were maintained undifferentiated by growth on 0.1% gelatin-coated dishes in Glasgow's minimal Eagle's medium supplemented with 10% fetal calf serum,nonessential amino acids, 1 mmol/l sodium pyruvate, 0.3 mg/ml L-glutamine, 0.1 mmol/l 2-mercaptoethanol, 1000 U/ml human recombinant leukaemia inhibitory factor (LIF) and 100 μg/ml hyg(Chambeyron and Bickmore,2004). Differentiation was induced by plating the cells at low density without LIF or hyg for 1 day. Retinoic acid (RA)5×10^-6 M was then added. On day 3 the cells were replated in RA-containing medium complemented with 2.5 μmol/l Ganciclovir. Samples for RNA extraction and nuclei preparation were collected every 2 days of differentiation. Similar results were obtained on samples from two independent differentiation experiments. Data presented in this paper all come from the same differentiation.

Analysis of E9.5 embryos was as previously described(Chambeyron et al., 2005). Embryos from crosses between Dct-LacZ homozygous transgenic CD1XCD1 mice were collected at E9.5, fixed in 4% formaldehyde overnight at 4°C,dehydrated through a graded ethanol series, cleared in xylene and embedded in paraffin blocks. Adjacent serial sections were cut at 4 μm and used for DNA FISH and Haematoxylin-Eosin staining. For FISH, sections laid on Superfrost slides were heated to 60°C for 20 minutes and washed four times in xylene for 10 minutes each before rehydration through an ethanol series. They were then microwaved for 20 minutes in 0.1 mol/l citrate pH 6 buffer, washed in water and rinsed once in 2× SSC before use. FISH was performed as described below, except for the denaturation step, which was for 3 minutes at 75°C in 70% formamide/2× SSC pH 7.5 followed by 3 minutes in ice-cold 70% ethanol.

Total RNA was extracted using Tri-reagent (SIGMA) and assessed by electrophoresis. Random-primed reverse transcriptions were performed at 42°C for 1 hour on 5 μg total RNA by using 1 Unit Superscript II Reverse transcriptase (RT) (Invitrogen) and 4 μl random hexamers (50 mmol/l; Amersham Pharmacia) in a 50 μl reaction. A negative control reaction, lacking enzyme, was performed in parallel for each RNA sample. One microlitre of these reactions were amplified by PCR using 1 Unit Taq polymerase (Amplitaq, Roche), 2 mmol/l magnesium chloride and 1 μmol/l forward and reverse primers in a 20 μl reaction. All primer pairs used span exons and therefore only detected the spliced transcripts. PCR conditions and primer sequences are described in Table 1.

Nuclei for 2D and 3D FISH were prepared as previously described(Chambeyron and Bickmore,2004). FITC-labelled paint for MMU2 was purchased from CAMBIO. BAC and fosmid clones covering the Hoxd region were chosen from ENSEMBL v37, Feb 2006(http://www.ensembl.org/Mus_musculus/index.html). BAC clones were purchased from BACPAC resources center and fosmid clones were kindly provided by the Sanger Institute (Cambridge, UK) (for coordinates and names, see Table 2). DNA from clones was prepared using standard alkaline lysis and labelled by nick-translation with digoxigenin-11-dUTP or biotin-16-dUTP (Roche). Approximately 200 ng FITC-paint, 100 ng biotin-labelled BAC or fosmid probe and 100 ng digoxigenin BAC or fosmid probe were used per slide, together with 15 μg mouse Cot1 DNA (GIBCO BRL) and 5 μg salmon sperm DNA. Digoxigenin-labelled probes were detected using Rhodamine anti-digoxigenin and Texas Red anti-sheep IgG (Vector Laboratories); biotin-labelled probes were detected using Cy5 streptavidin and biotinylated anti-avidin (Vector Laboratories); and the FITC signal from the MMU2 chromosome paint was amplified using F1 rabbit anti-FITC and F2 FITC anti-rabbit (CAMBIO). Hybridisation, washes and detection were as described previously(Chambeyron and Bickmore,2004). Slides were counterstained with 0.5 μg/ml DAPI.

For 2D FISH, slides were examined using a Zeiss Axioplan II fluorescence microscope with Plan-neofluar objectives, a 100 W Hg source (Carl Zeiss,Welwyn Garden City, UK) and a Chroma #84000 quadruple band pass filter set(Chroma Technology Corp., Rockingham, VT) with the excitation filters installed in a motorised filter wheel (Ludl Electronic Products, Hawthorne,NY). Greyscale images were captured with a Hamamatsu Orca AG CCD camera[Hamamatsu Photonics (UK) Ltd, Welwyn Garden City, UK]. Image capture and analysis were performed using in-house scripts written for IPLab Spectrum(Scanalytics Corp., Fairfax, VA). The distance between two probes, and probe position relative to the CT edge, were calculated as described previously(Chambeyron and Bickmore,2004).

For 3D FISH, slides were examined using a Zeiss Axioskop fluorescence microscope with Plan-neofluar or Plan apochromat objectives, a 50 W Hg source(Carl Zeiss, Welwyn Garden City, UK) and Chroma #83000 triple band pass filter set (Chroma Technology Corp., Rockingham, VT) with the excitation filters installed in a motorised filter wheel (Ludl Electronic Products, Hawthorne,NY). A piezoelectrically driven objective mount (PIFOCi model P-721, Physik Instrumente GmbH & Co., Karlsruhe) and a Princeton Instruments Micromax CCD camera with Kodak 1400e sensor (Universal Imaging, Maldon, UK), were used to control movement in the z dimension and collect image stacks with a 0.2 μm step. Hardware control, image capture and analysis were performed using in-house scripts written for IPLab Spectrum (Scanalytics Corp., Fairfax,VA). Images were deconvolved using a calculated PSF with the constrained iterative volume algorithm of Microtome (Scanalytics Corp., Fairfax, VA). Three-dimensional distance measurements were as described previously(Chambeyron et al., 2005). A minimum of 50 nuclei was analysed for each tissue.

The statistical relevance was assessed using the non-parametric Kolmogorov-Smirnov test to examine the null hypothesis that two sets of data show the same distribution. Data sets consisted of at least 50 nuclei (100 territories) for each differentiation time point or embryonic tissue, and for each combination of probes. A P-value <0.05 was considered statistically significant.

We have previously shown that nuclear reorganisation of Hoxboccurs in expressing tissues during embryonic development(Chambeyron et al., 2005). In the E9.5 embryo, differential looping out of Hoxb from its CT occurs along the rostrocaudal axis of the neural tube. Hoxb1, at the 3′ end of Hoxb, is extensively looped outside its CT in nuclei of rhombomere 4 (r4), where it is strongly expressed, but not more caudally in r5 or more posterior parts of the spinal cord, where it is not expressed. The tailbud is the most caudal part of the spinal cord, resulting from the regression of the primitive streak earlier in development. Both 3′ and 5′ Hoxb genes are widely expressed in the tailbud and,concomitantly, both 3′ and 5′ ends of the Hoxb locus locate just outside the CT in this embryonic region(Chambeyron et al., 2005). Hoxd genes are also widely expressed in the tailbud region(Tarchini and Duboule, 2006; Zakany et al., 2001). Therefore, we used 3D DNA FISH on E9.5 mouse embryo sections to determine whether the Hoxd region is looped out from its CT in the tailbud. Rhombomeres 1, 2 and 4, where Hoxd genes are not expressed(McKay et al., 1994; Zakany et al., 2001; Zhang et al., 1994), were used as control tissues. Nuclear distances between the probe signals and the edge of mouse chromosome 2 (MMU2) CT, where Hoxd is located, were determined using a previously described script(Mahy et al., 2002; Chambeyron and Bickmore, 2004; Chambeyron et al., 2005). In this analysis a probe signal is considered to be outside the CT when its distance to the CT edge is negative. Distances >0 correspond to probe signals inside the CT.

In control tissues (r1, 2 and 4), signals from a BAC probe covering the Hoxd cluster (`Hoxd' RP23-15M17 in Fig. 1 and Table 2) were mainly inside the MMU2 CT, but in the tailbud 40% of them were located outside the MMU2 CT(Fig. 2A). This looping out was seen in >80% of tailbud nuclei, but was usually restricted to only one of the two alleles. Therefore, in the tailbud, movement of the Hoxdlocus to the edge and to the outside of the CT coincided with its expression(P=0.001). This nuclear reorganisation was not restricted to Hoxd itself. Signals from a BAC clone just 5′ of Hoxdand overlapping the Lnp gene (`Lnp' RP24-267L11) were also relocalised towards the outside of the CT in 70% of tailbud nuclei(P=0.003 compared with control), even though there is no evidence that Lnp expression is induced there(Spitz et al., 2003). However,signals from a BAC located more 5′ ('5′flank' RP23-288B11 in Fig. 1) remained inside the CT in both control and tailbud (P=0.66)(Fig. 2A). Even in the nonexpressing control tissues, Hoxd was located significantly(P<0.001) closer to the CT edge than the two more 5′regions, suggesting that the CT may be organised such that Hoxd is poised before its activation (Fig. 2B).

Throughout the emerging limb bud of the E9.5 embryo, 3′ Hoxdgenes, up to and including Hoxd9, are also expressed(Tarchini and Duboule, 2006; Tarchini et al., 2006). However, we found no significant looping out of the region from the CT in the forelimb bud, compared with control tissues (P>0.09 for all three BACs) (Fig. 2A-C). Therefore activation of Hoxd in limb bud occurs in the absence of a relocalisation of the locus outside the CT, whereas in the tailbud Hoxd expression occurs in the context of a looping out of a large region from the CT. There appears to be a boundary to the extent of this looping out located between the BACs Lnp and 5′ flank.

At Hoxb, movement away from the CT was also accompanied by a visible increase in the nuclear distance between the 3′ and the 5′ends of the cluster, which is thought to represent a decondensation, or unfolding, of higher-order chromatin structure(Chambeyron and Bickmore, 2004; Sachs et al., 1995; Yokota et al., 1995). To establish whether this also occurs along the Hoxd region, we measured the interphase distances (d) between the 3D DNA FISH signals for the Hoxd, and Lnp BACs in nuclei from E9.5 embryos. There was significant(P<10^-3) movement to larger d values in both the tailbud and limb bud compared with the Hoxd nonexpressing control tissues. Interphase decondensation was also detected between the 5′flank and Lnp BACs (P=0.01 and <10^-3 in tailbud and limb bud, respectively) (Fig. 2D,E).

Another manifestation of chromatin decondensation lies in the shape of the BAC hybridisation signals. Whereas the signals detected with the 5′flank BAC were well-defined pinpoints, the Lnp and Hoxd BACs often gave stretched-out tracks of signals. This was especially prominent for the Hoxd probe in the limb bud (Fig. 2F,G).

Therefore, long-range decondensation of the entire region extending from Lnp to the Hoxd cluster correlated with Hoxdexpression in E9.5 embryos. In the tailbud, 20% of Hoxd loci appeared decondensed, which is similar to the percentage of loci localised outside the MMU2 CT. However, there was also decondensation of Hoxd in the limb bud, where there was no relocalisation of this region with respect to the CT. This suggests that `looping out' and decondensation are not just different manifestations of the same event of nuclear reorganisation.

To dissect the events of nuclear reorganisation in more detail and in a more tractable system, we used an ES cell differentiation system. Gene expression and nuclear reorganisation could be induced at Hoxb by triggering the differentiation of murine ES cells with retinoic acid (RA)(Chambeyron and Bickmore,2004). To determine whether similar activation occurs at Hoxd we used RT-PCR to analyse the expression of Hoxd genes(Fig. 3) in undifferentiated OS25 ES cells, and during 18 days after the withdrawal of LIF and the addition of RA. As for Hoxb, there was no detectable expression of Hoxd genes in undifferentiated cells. The extinction of Oct4expression upon differentiation was accompanied by the rapid induction (by day 2) of Hoxd1 expression, but not of the more 5′ genes Hoxd3 through to Hoxd12(Fig. 3). Expression of Hoxd3 and Hoxd4 were detected by day 6, Hoxd8 by day 8, Hoxd9 and Hoxd10 by day 10 and Hoxd12expression was not detected until day 18. Hoxd1 expression declined at later stages of differentiation, but not as rapidly as seen for Hoxb1 (Chambeyron and Bickmore,2004).

Hoxd is flanked by structurally and functionally unrelated genes. Expression of Mtx2, located 3′ of Hoxd1(Fig. 1), is induced by day 2(Fig. 3), suggesting that this gene might also be subject to temporal colinearity. However, at the 5′end of Hoxd, the early detection of Hoxd13 expression (by day 2) suggests a break in the temporal colinearity at this end of the cluster in this system. A large conserved noncoding region 5′ of Hoxd(Fig. 1B), termed a global control region (GCR), contains digit enhancers that act on Hoxd13,Lnp and Evx2 (Tarchini and Duboule, 2006). Neural enhancers in the GCR also act on Evx2 and Lnp (Spitz et al., 2003). As Evx2 is also activated early in the timecourse of differentiation, this suggests that the GCR may have some activity in ES cells. Lnp expression in ES cells is constitutive(Fig. 3)(Spitz et al., 2003). This analysis shows that the colinear activation of the Hoxd cluster is mostly recapitulated upon RA-induced ES cell differentiation.

We used 2D interphase FISH to determine whether the events of nuclear reorganisation at Hoxd, seen in the embryo, also occur in differentiating ES cells. As in control embryonic tissues, hybridisation signals from the Hoxd and Lnp BAC probes were almost all located inside the CT in undifferentiated cells (Fig. 4A,B), even though Lnp was expressed in them(Fig. 3). These regions relocalise outside the MMU2 CT upon differentiation (P<0.005), but whereas the highest proportion (32%) of Hoxd alleles located >0.3 μm outside the CT occurred by day 4 of differentiation, the maximum frequency of extra-territory Lnp alleles was not seen until day 8(Fig. 4A). The nuclear behaviour of the region 3′ of Hoxd and Mtx2, detected by the RP23-7D13 (3′ flank) BAC clone(Fig. 1), was similar to that of Lnp, with a maximal frequency of looping out from the CT at day 8 of differentiation. After 14 days, signals from all three probes were relocated back inside the CT (Fig. 4A,B). As in the embryo, there was no movement detected with the 5′ flank BAC at any time point during differentiation(Fig. 4B). Therefore, there is a remarkably similar domain of chromatin around Hoxd that is located outside its CT in the tailbud of the embryo, and that is induced to transiently loop out from the CT during ES cell differentiation.

There is also a significant (P<10^-3) increase in the interphase distances separating the Hoxd cluster (RP23-15M17) from Lnp (RP24-267L11), and Lnp from the 5′ flank (RP23-288B11) during differentiation (Fig. 4C). At day 14, and concomitant with the relocalisation of the region back inside the CT, there is recondensation of the whole region(Fig. 4D). As in the embryo,extended tracks of hybridisation signals were often detected in differentiating ES cells with the Hoxd and the Lnp BAC probes (data not shown).

The relative kinetics of the intra-CT movements detected with probes across the Hoxd region (Fig. 4A) suggests that these events initiate within the Hoxdcluster itself, rather than in flanking regions. As the temporal activation of Hox genes begins at the 3′ ends of the clusters, we analysed nuclear behaviour within the Hoxd locus itself using three fosmid clones (121N10, 860J8 and 469P2; Fig. 1 and Table 2). In undifferentiated ES cells, almost all (>88%) 3′ Hoxd signals(121N10 probe) were within the CT (Fig. 5A,B). By day 4 of differentiation, there was a significant(P=0.01) shift of signals away from the CT, such that 25% of them were then located outside. This is the same time point at which `looping out'of Hoxb1 signals is seen from its CT(Chambeyron and Bickmore,2004). By day 14, 3′ Hoxd signals were back within the CT, similar to the situation before differentiation (P=0.95)(Fig. 5A,B). Similar results were obtained with 860J8, encompassing the central Hoxd genes (data not shown). At the 5′ end of Hoxd (469P2), there was also movement towards and outside the CT edge, but this did not reach maximal significance (P=0.004) until day 8(Fig. 5B). This suggests that nuclear reorganisation initiates at the 3′ end of Hoxd and this is also supported by a comparison of the mean probe positions during the timecourse of differentiation (Fig. 5C).

Chromatin decondensation also occurs within Hoxd. The midpoints of the 3′ (121N10) and 5′ (469P2) Hoxd probes were 90 kb apart and their hybridisation signals were barely separable in undifferentiated cells (Fig. 5A). At days 4-8 of differentiation, there was a significant(P<10^-3) increase in the interphase separation between them (Fig. 5D). There was some recondensation by day 14, but their separation was still greater than that before differentiation (P=0.006), suggesting that the structure of the Hoxd cluster remains in an altered state. To confirm that decondensation over such short genomic distances was not an artefact of the 2D DNA FISH, we confirmed that there was also a significant increase in the interphase separations between 121N10 and 860J8 fosmids during differentiation, using 3D FISH (Fig. 5E). A similar relative degree of decondensation was detected by 2D and 3D FISH, even though the former exaggerated the absolute interphase distances (Fig. 5F).

Superficially, the similar kinetics of looping out and chromatin decondensation of the Hoxd region during ES cell differentiation are consistent with the suggestion that looping out from CTs is a visual manifestation of chromatin decondensation that has been propagated over a long range (Volpi et al., 2000; Gilbert et al., 2004). However,our observation in the embryo proper, of extensive decondensation in the absence of looping out, led us to question this. To determine whether interphase decondensation and looping out are indeed synonymous events, we used four-colour DNA FISH to simultaneously measure the nuclear positions of different parts of Hoxd both relative to each other and relative to the CT during ES cell differentiation. When we looked between the 5′flanking region, which remained fixed inside the CT, and either the Lnp or Hoxd regions, which moved away to locations outside the CT during differentiation, the increase in inter-probe distances was indeed correlated with looping of Hoxd outside the MMU2 CT (R²=0.44 at day 4) (Fig. 6A-C).

However, a very different pattern emerged when we examined probes within the region that loops out of the CT (from Lnp to the 3′ flank, including Hoxd itself). Movement of Lnp and the 3′ flank occurred together: it was rare to find one probe outside a CT while the other remained inside(Fig. 6D,E). However, this movement preceded interphase separation, as both probes found outside the CT at day 4 were still spatially close together (<0.5 μm)(Fig. 6F). At day 8, when there was both looping out and apparent decondensation, there was no correlation between the two events (in linear regression analysis R²<0.001)(Fig. 6E). Lnp and 3′flank signals located far outside the bounds of the visible CT could appear to still be quite closely juxtaposed, and widely separated signals could be seen within the CT (Fig. 6D). The same lack of correlation between movement and interphase separation was also observed between Lnp and the Hoxd cluster itself, and even between the 3′ and 5′ ends of Hoxd (data not shown).

We have shown previously that the murine Hoxb locus loops out from its CT (mouse chromosome 11) and decondenses when the locus is activated during ES cell differentiation (Chambeyron and Bickmore, 2004), or during embryonic development(Chambeyron et al., 2005). Here we show that similar events of nuclear reorganisation also occur at Hoxd in the tailbud region of the E9.5 embryo(Fig. 2) and during ES cell differentiation (Fig. 4). Nuclear reorganisation of Hox loci therefore reflects a basic mechanism underlying the specific activation of these gene clusters in differentiation,and in development along the rostrocaudal embryonic axis. Thus it seems likely that this mechanism predates the duplication and divergence of vertebrate Hox clusters more than 500 million years ago(Pollard and Holland, 2000; Ferrier and Minguillon,2003).

Although activation of Hoxd is accompanied by its nuclear reorganisation, a simple escape from the CT cannot be sufficient to explain the temporal activation of gene expression during ES cell differentiation. Whereas the kinetics of Hoxd1 activation is contemporaneous with its movement away from the CT by day 4, expression of Hoxd3 (on the same fosmid clone as Hoxd1) is not detected by RT-PCR until day 6(Fig. 2). Conversely, maximal relocalisation of the 5′ Hoxd fosmid is seen by day 8 of differentiation, but Hoxd12 expression is not detected by RT-PCR until day 18, at which time the Hoxd region has been retracted back into the CT. This suggests that movement away from the CT is not sufficient to activate the expression of all Hoxd genes and that additional chromatin modifications must occur to allow for gene expression per se. Also,it seems unlikely that it is necessary for 5′ Hox genes(Hoxd12) to be outside the CT when they are expressed, although we cannot exclude the possibility that Hoxd12 expression is originating from that small proportion (10%) of alleles that remain outside the CT at the end of differentiation (Fig. 5B). Combined RNA and DNA FISH might resolve this, but we have been unable to detect Hox expression by RNA FISH. This may be due to the fact both that Hox genes have only one small intron, and that microarray analysis shows that levels of Hoxd gene expression are not very high (data not shown).

Looping out of chromatin away from the CT, initiated within Hoxd,then spreads out in both directions to encompass unrelated flanking genes(Figs 2, 4). The activation of Mtx2 expression early in differentiation 3′ of Hoxd(Fig. 3) might be a response to the spread of nuclear reorganisation. Similarly, the timecourse of Evx2 expression immediately 5′ of Hoxd appears to parallel the spread of chromatin looping and decondensation(Fig. 4). Interestingly the proximity of Evx genes to another Hox locus (Hoxa) suggests that Evx2 was part of an ancestral Hox cluster(Pollard and Holland, 2000; Ferrier and Minguillon, 2003),so it may have evolved to respond to global chromosomal cues regulating Hox regulation. However, beyond Evx2, Lnp is a recent evolutionary recruit to this Hox locus. Its expression does not respond to Hoxdnuclear reorganisation during ES cell differentiation(Fig. 3), or in the tailbud of the E9.5 embryo (Spitz et al.,2003), even though it is relocalised towards the outside of the CT in these situations (Figs 2, 4).

Our previous analyses showing both interphase probe separation and looping out at Hoxb, in ES cell differentiation(Chambeyron and Bickmore,2004), and in the embryo(Chambeyron et al., 2005),appeared to be consistent with the idea that looping out from CTs and chromatin decondensation are two sides of the same coin(Volpi et al., 2000; Gilbert et al., 2004). However,our analysis here of Hoxd has thrown up discordances between interlocus interphase distances and movement of the loci to the outside of CTs. In two different Hoxd-expressing regions of the E9.5 embryo, one- the tailbud - shows both looping out and decondensation at Hoxd,whereas the other - the limb bud at E9.5 - shows extensive chromatin decondensation but no evidence of any looping out of Hoxd(Fig. 2). We therefore assayed simultaneously, at each allele during ES cell differentiation, the localisation with respect to the CT, and the interphase probe separation. Whereas the movement of the Hoxd region to sites outside the CT correlates with its interphase separation from a 5′ probe (288B11) that remains inside the CT, a more complex situation was seen within the Hoxd region itself - from Lnp to the 3′ flank(Fig. 6). The first (day 4)nuclear events seen within this region are movement to the outside of the CT,without chromatin decondensation as assayed by interphase separation. When a movement to larger inter-probe distances is seen at day 8, this occurs regardless of position relative to the CT. One possibility is that long-range unfolding of chromatin structure allows the locus a greater degree of mobility so that it can move dynamically in and out of the CT. Thus, although both events are triggered upon ES differentiation, looping out does not depend on a prior decondensation of the region, and chromatin decondensation may not lead directly to a looping out from the CT.

Most published in situ hybridisations on embryos at around E9.5 suggest that levels of Hoxd expression in the tailbud and limb bud are similar (e.g. Tarchini and Duboule,2006), but there is evidence for stronger expression of Hoxd4 in the tailbud compared with the limb bud(Akasaka et al., 2001)(http://genex.hgu.mrc.ac.uk/das/jsp/submission.jsp?id=EMAGE:48). Therefore at this stage, we cannot exclude that the differential nuclear organisation of Hoxd in tailbud and limb bud is due to different levels of Hoxd transcription. Also, different regulatory pathways might be responsible for Hoxd activation along the rostrocaudal and the secondary axis of the embryo. The temporal activation of Hoxd1-d9 in the early limb bud is thought to be under the control of an early limb bud activating element (ELCR) located 3′ of the cluster (Tarchini and Duboule,2006), but its role in the regulation of Hoxd along the rostrocaudal axis has not been fully explored. Understanding how different modes of regulation feed in to altered chromatin structures at Hox loci will be key to further elucidating how the expression of these developmental players is regulated so exquisitely in time and space.

C.M. was an EMBO long-term fellow and is currently a recipient of a Marie Curie Intra European Fellowship; W.A.B. is a Centennial fellow of the James S. McDonnell foundation. We thank Séverine Chambeyron (IGH, Montpellier)for assistance in initiating this project and for reading the manuscript. We are also grateful for the insightful comments of Clémence Kress and Marie Kmita. The work was supported by the UK Medical Research Council and in part by the EU FP6 Network of Excellence Epigenome (LSHG-CT-2004-503433).