Evolutionary conserved sequences are required for the insulation of the vertebrate Hoxd complex in neural cells

During vertebrate development, proteins encoded by the Hox gene family are required to properly instruct cells about their morphological fates,subsequently leading to the emergence and organisation of different structures along the body axis. Mammals have 39 Hox genes, clustered at four genomic loci, which provide these organisational cues to a variety of embryonic axial structures and derivatives. Accordingly, the transcription of these genes must be precisely regulated in time and space, in order to ensure harmonious development. This complex task appears to rely partially upon the genomic organisation of the genes, as a correspondence exists between gene order along the clusters and their spatial and temporal sequences of transcriptional activation (reviewed by Krumlauf,1994). Although the molecular mechanisms that underlie this phenomenon are not yet fully understood, they may involve high order regulation, such as (for example) a transition in chromatin configuration(Deschamps et al., 1999;Kmita et al., 2000b).

Beside this level of transcriptional regulation, many cis-acting control sequences have been characterised by their ability to impose particular expression patterns to nearby located genes. Various enhancer sequences have thus been described, with distinct functional properties. For example, in several cases, gene-specific activation was shown to result from proximal enhancers selectively interacting with a given promoter. Alternatively, enhancer sharing mechanisms were reported to account for the co-expression of neighbouring genes(Sharpe et al., 1998), a situation favoured by the tight clustered organisation of these genes(Bell et al., 2001). Enhancer sharing processes, within Hox gene clusters, were not only shown to involve proximal enhancers, which can control the expression of neighbouring genes in the same tissue, but also more global, distally located enhancers, which are able to impose a particular regulation to series of contiguous genes. Examples of such a large-scale regulation was provided by the co-expression of severalHoxd genes in either the intestinal hernia or the developing digits(Zakany and Duboule, 1999;Kmita et al., 2000a;Spitz et al., 2001). In these latter cases, co-ordinated expression of several genes at the same place was demonstrated to be necessary to properly build up the concerned structure(Zakany et al., 1997a).

However, this particular regulatory strategy implies that other closely linked gene members of the cluster, the function of which may not be relevant in a given structure, are protected against such a global regulatory influence, such as to prevent their mis-expression. Indeed, ectopic transcription of Hox genes was shown be a potential source of severe morphological and/or physiological alterations(Knezevic et al., 1997;McLain et al., 1992;Morgan et al., 1992;Rijli et al., 1994;Yokouchi et al., 1995). Accordingly, boundary or insulator elements must exist to restrain the action of enhancers specifically to those relevant target genes, by isolating them from their neighbours (Sun and Elgin,1999; Udvardy,1999). We have previously showed that a DNA segment located between Hoxd12 and Hoxd13 could prevent both genes from responding to a distally located intestinal hernia enhancer(Kmita et al., 2000a). In much the same way, Hox clusters must themselves be isolated from external regulatory influences to prevent enhancer sequences that are necessary for closely located, non-Hox genes, to interfere with the precise and particular regulation of this gene family. This requirement for a context-dependent insulation is best exemplified by the presence of the Evx2 gene in the immediate 5′ neighbourhood of the Hoxd cluster(D'Esposito et al., 1991;Bastian et al., 1992).

Evx2 indeed displays specific expression features that are not shared by any Hoxd genes, not even by Hoxd13, whose promoter lies close to that of Evx2. This is best illustrated by discrete cell types of the developing central nervous system, in both spinal cord and more rostral parts of the brain, in various vertebrate species(Bastian et al., 1992;Brulfert et al., 1998;Dollé et al., 1994;Sordino et al., 1996). In the spinal cord, transcripts are localised in the ventrally located V0 interneurones, as well as in a population of dorsal interneurones(Moran-Rivard et al., 2001). In the developing brain, Evx2 expression is detected in the rhombencephalic isthmus area (the metencephalic-mesencephalic transition) and extends into the superficial layer of the entire midbrain. It is also expressed in the developing hindbrain and in part of the future cerebellum(Dollé et al.,1994).

While the enhancer sequences driving Evx2 expression in the CNS have not yet been precisely identified, experiments involving targeted genomic rearrangements around the Evx2 locus have revealed some of their properties. First, targeted deletions have shown that these enhancer sequences are located at a remote position, upstream the Hoxd complex(Kondo and Duboule, 1999). Second, we showed that a Hoxd9/lacZ transgene was able to respond to the Evx2 CNS-specific enhancer sequences, whenever it was relocated upstream the Hoxd complex, 3′ to Evx2(Kondo and Duboule, 1999). However, the same transgene was unable to respond similarly when placed within the complex, even when positioned immediately next to the Evx2promoter (van der Hoeven et al.,1996). These results demonstrated that the Evx2 CNS enhancers had a weak specificity for Evx2 itself, i.e. they were able to interact with other promoters. In addition, Hoxd promoters could respond to such regulatory controls provided they would be relocated in the proper genomic environment, i.e. in 3′ of the Evx2transcription unit. These observations raised the question of which mechanism could prevent Hoxd genes to respond to these CNS enhancers, in the wild-type context. In other words, why a promoter able to respond to a given regulatory sequence, when placed outside the cluster, was unable to do so from within the Hoxd complex, even when localised right next to theEvx2 promoter.

In this set of experiments, we looked for potential sequences, located between Evx2 and the Hoxd cluster, that would be able to isolate this latter cluster from the surrounding regulatory influences. We show that an evolutionary conserved DNA stretch participates in the insulation of the cluster, as revealed by novel genomic rearrangements in this locus. However, even though this sequence was sufficient to ensure proper insulation of the cluster, additional sequences, located nearby, were also found to be involved in this process. The requirement for a combined deletion incis of these sequences in order to bypass the insulation of the cluster, raised the possibility that some functional redundancy exists between these regulatory sequences.

Targeted deletion of RXII was engineered by homologous recombination in ES cells. A 1.2 kb AvrII DNA fragment containing RXII was deleted from the 9.5 kb NotI fragment that covers the entire Evx2 toHoxd13 intergenic region. A PGK-neomycin selection cassette,flanked by loxP sites, was inserted at the NsiI site, as described previously (Hérault et al., 1996; van der Hoeven et al., 1996). The resulting targeting vector was electroporated into D3 ES cells. Clones in which homologous recombination had occurred were selected, amplified and injected into mouse embryos. After germline transmission, the Hoxd^RXII-neo line of mice were obtained and further crossed with partners carrying the CMV-Cretransgene in order to produce the Hoxd^RXII line of mutant mice that lacked the PGK-neomycin selection cassette.

Besides the Hoxd^RXII line, all mutant lines analysed in this work were produced via trans-allelic meiotic recombination (TAMERE) (Hérault et al., 1998b; Kmita et al.,2002). Each allele was obtained in the progeny of`trans-loxer' animals, i.e. males hemizygous for theSycp1-Cre transgene and trans-heterozygous for differentHoxd alleles carrying a loxP site at given positions within the Hoxd cluster (indicated inFig. 4)(Kmita et al., 2002). In particular, Hoxd^del(13) animals were obtained by combining a Hoxd allele carrying a loxP site betweenEvx2 and Hoxd13 (the EvD^GE3allele) (Hérault et al.,1996), with an allele carrying a loxP site betweenHoxd13 and Hoxd12 (Hoxd^RXI)(Hérault et al.,1998a). Hoxd^del(13-12) andHoxd^del(13-11) animals were obtained in a similar way, although in these latter cases, the EvD^GE3allele was combined either with Hoxd^RX, in which a loxP site had been inserted between Hoxd12 andHoxd11 (Beckers et Duboule,1998), or with Hoxd^RIX, containing aloxP site between Hoxd11 and Hoxd10(Gérard et al., 1996).Hoxd^RXII-del(13) mice were obtained in the progeny of trans-loxer, which were trans-heterozygous forHoxd^RXII andHoxd^RXI. Finally,HoxdRXII^(del(13-12) were produced trough trans-loxer animals trans-heterozygous for bothHoxd^RXII and Hoxd^RXalleles. All these novel lines of mice were selected by Southern blot analysis using tail DNA. The frequency of TAMERE was in the range of 5-10%, as reported previously (Hérault et al.,1998b).

Whole-mount in situ hybridisation (WISH) were carried out on 11.5- and 12.5-day-old foetuses, using a standard procedure and previously described probes (Hérault et al.,1996; Kondo et al.,1998).

The Evx2 gene, a mammalian gene orthologous to the Drosophila even skipped gene (eve), is localised about 8 kb upstream of the most posterior gene member of the Hoxd complex; Hoxd13(Fig. 1)(Bastian et al., 1992). Because its transcriptional orientation is opposite to that of all Hoxdgenes, its promoter lies close to the Hoxd13 promoter(Fig. 1; arrows). Even though this homeobox-containing gene does not in the strictest sense belong to the Hox gene family, it shares some important regulatory features with thoseHoxd genes located at the `posterior' end of the cluster, such asHoxd13. During limb development, the timing of Evx2expression follows that of Hoxd genes and it is eventually co-expressed with 5′-located Hoxd genes in developing digits(Fig. 1A). This shared regulatory feature is dependent upon the action of a remote enhancer sequence located in 5′ of the Hoxd complex(Spitz et al., 2001;van der Hoeven et al., 1996). Unlike Hoxd genes, however, Evx2 was shown to be transcribed in subset of cells within the central nervous system(Fig. 1A)(Dollé et al., 1994),in response to regulatory sequences that are also located upstream the cluster, as revealed by engineered targeted deletions(Kondo and Duboule, 1999).

These differences in regulation between Evx2 and Hoxd13could hardly be accounted for by the specificity of enhancer/promoter interactions, because a Hoxd9/lacZ transgene was able to respond to these neural enhancers when placed 3′ to Evx2(Fig. 1B; RelII). This transgene, however, behaved as a proper Hox gene when placed betweenEvx2 and Hoxd13, a position at which it failed to show expression in rostral parts of the brain and in spinal cord(Fig. 1B). These results indicated that the capacity of a Hox promoter to respond to Evx2 CNS enhancers was abrogated when this promoter was positioned within the cluster,suggesting that a potential insulating element was present between the Re10 insertion site and Evx2 (Fig. 1; red bar). Because in birds, fish and mammals Evx2 lies at the same relative position with respect to Hoxd13(Sordino et al., 1996), we anticipated that a DNA sequences that would prevent the Evx2 neural enhancers from affecting Hox gene expression may have been conserved between these different genomes.

Comparison between Evx2 to Hoxd13 intergenic DNA sequences, obtained from either the murine, the chick or the zebra fish loci,revealed only two stretches of high sequence similarity localised betweenEvx2 and the Re10 position (Fig. 2A,B; red bar). In the mouse genome, these two motives are located within a 1.2 kb large fragment, starting about 1 kb upstream from the first exon of Evx2. This region of significant sequence conservation was referred to as region XII (RXII), following previously characterised conserved regions within the Hoxd cluster(Renucci et al., 1992;Beckers and Duboule, 1998;Gérard et al., 1996;Hérault et al., 1998a). While sequence conservation was high between murine and avian DNAs for both motives (67% identity over 206 nucleotides), it was less conspicuous when compared with the fish DNA, as only short stretches of sequence identity were scored for both motives. In this latter case, however, the core of the second motif was clearly identified in the zebra fish locus and found at the same relative position (Fig. 2A,B). This unambiguously demonstrated the existence, in the zebra fish locus, of at least one of these two blocks of homologies.

In order to assess the function of these two conserved sequences, we deleted them from their native genomic context by homologous recombination in ES cells. We constructed a targeting vector containing the Evx2 toHoxd13 intergenic region, but in which the 1.2 kb fragment had been deleted (Fig. 2C). After electroporation in ES cells, clones carrying a targeted deletion of RXII were selected and further injected into mouse blastocysts. After germline transmission, the Hoxd^RXII-neo line of mice was established. In order to prevent regulatory interferences caused by the presence of the PGK-neomycine selection cassette,Hoxd^RXII-neo animals were crossed with transgenic mice producing the Cre recombinase (CMV-Cre mice)(Dupe et al., 1997) to delete the selection cassette. Therefore, the final genomic configuration of theseHoxd^RXII mice was a single deletion of the 1.2 kb fragment containing RXII (Fig. 2C), along with the presence of a loxP site. We subsequently obtained Hoxd^RXII homozygous mice,which were fully viable and fertile.

The expression of several Hoxd genes was examined at various developmental stages, in animals homozygous for the deletion of RXII, but no detectable difference was scored when compared with their wild type or heterozygous littermates. In particular, ectopic expression of Hoxdgenes showing an Evx2 related CNS pattern was not observed. This result suggested that the deleted 1.2Kb DNA fragment was not able, on its own,to function as a boundary-like or insulator element, to isolate theHoxd cluster from the upstream located Evx2 CNS enhancers. Alternatively, the apparent lack of effect of this deletion may illustrate some redundancy in this regulatory process.

Within the 5′ part of the Hoxd cluster, several regions of high interspecies conservation were previously identified(Fig. 3) (RVIII to RXI). Each individual region was assayed for potential regulatory function through targeted deletion/mutation (Gérard et al., 1996; Zakany et al.,1997b; Beckers and Doboule, 1998;Hérault et al., 1998a). Although slight variations in Hoxd gene expression were occasionally observed following these targeted modifications, none of them indicated a potential role for these regions, by themselves, to restrict the accessibility of Hoxd promoters to the Evx2 cis-regulatory sequences. In order to look for their possible cooperation in the implementation of an insulting process, we used the targeted meiotic recombination (TAMERE)strategy (Hérault et al.,1998b) to generate novel genomic configurations in vivo through Cre-mediated meiotic recombination between loxP sites carried intrans by homologous chromosomes. In this way, we produced a set of progressive deletions of the 5′ end of the Hoxd complex,involving one, two or three gene loci, as well as RXI, RX and RIX/RVIII,respectively (see Kmita et al.,2002).

First, we generated mice containing the SYCP-Cre transgene(Vidal et al., 1998), along with a Hoxd complex carrying, on one chromosome, a loxP site positioned in the middle of the Evx2 to Hoxd13 intergenic region. The other chromosome had a loxP site recombined either upstream Hoxd12, between Hoxd12 and Hoxd11, or upstream Hoxd10. During meiotic prophase, in some male germ cells,recombination occurred between these loxP sites in trans,leading to unequal chromosomal exchanges, thereby producing sperms carrying a deletion of the DNA fragment located in between. In this way, mice were produced which carried different deletions; a 12 kb large DNA fragment covering the Hoxd13 locus (Hoxd^del(13)in (Fig. 3); a 18 kb large fragment covering both Hoxd13 and Hoxd12 loci(Hoxd^del(13-12)), and a 23 kb large fragment encompassing all three Hoxd13, Hoxd12 and Hoxd11 loci(Hoxd^del(13-11) inFig. 3). The same 5′break point was used to engineer all three deletions, such that increasingly large deletions concomitantly removed either one (RXI), two (RXI and RX) or four (RXI, RX, RIX and RVIII) conserved sequences, respectively(Fig. 3)(Kmita et al., 2002). Homozygous embryos were collected for each configuration and the expression patterns of the remaining 5′ Hoxd genes were examined by whole-mount in situ hybridisation. Again, Evx2-like expression in the CNS was not detected in any of these configurations (data not shown). This suggested that sequences responsible, either alone or in combination, for the insulation of the Hoxd cluster were not exclusively located within these 23 kb large DNA fragment containing the Hoxd13 toHoxd11 loci, if at all present in this fragment.

This set of data demonstrated that none of the engineered deletions that removed unique evolutionary conserved sequences had an effect on the insulation of the Hoxd cluster. It also showed that larger deletions,i.e. those that removed more than one such sequence from the cluster, were equally ineffective in altering this particular mechanism. One remaining possibility that could account for the insulation effect was the presence of an element located between the Re10 site and Evx2(Fig. 2; red bar), but outside the 1.2 kb large fragment that contains region XII, as deletion of this fragment had no effect. An alternative explanation is that the combined effect of region XII and other regions included in the series of deletions described above are responsible. We did not favour the first possibility, assuming that such a tight mechanism, present in many vertebrate species, may likely rely upon some sequence specificity. Therefore, we challenged the second possibility by producing multiple deletions in cis.

We used Hoxd^RXII as a parental allele in targeted meiotic recombination, to engineer novel genetic configurations in which the RXII deletion was combined in cis with larger deletions(Fig. 3). This was made possible by the strategy that was used to delete region XII, which involved the positioning of a selection cassette flanked by loxP sites, within the Re10 insertion site, i.e. in the middle of the Evx2 toHoxd13 intergenic region (Fig. 2C). Consequently, mice carrying the deletion of RXII had aloxP site at this position (Fig. 2C; Hoxd^RXII), as a left over of the Cre-mediated deletion of the PGKneo selection cassette. We first produced males carrying either the Hoxd^RXII andHoxd^RXI alleles (Fig. 4B), or the Hoxd^RXII andHoxd^RX alleles (Fig. 4C), along with the Cre. In the progeny of thesetrans-loxer males, we isolated bothHoxd^RXII-del(13) andHoxd^{RXII-del(13-12)} animals, respectively(Fig. 4). Although a strain ofHoxd^RXII-del(13) homozygous mice could be established,Hoxd^{RXII-del(13-12)} animals died at birth. Homozygous embryos of both genotypes could nevertheless be collected to look at the expression of the remaining 5′ Hoxd genes. We first analysed the expression of Hoxd12, Hoxd11 and Hoxd10 in theHoxd^RXII-del(13) strain, i.e. mice that lack both region XII and the Hoxd13 locus. In these animals, ectopic activation ofHoxd genes was not detected within the rostral brain or in the spinal cord (not shown), as one would have anticipated from an alteration of the insulating process.

We next looked at the deletion of both RXII and the 18 kb large fragment containing the Hoxd13 and Hoxd12 loci(Fig. 5). In marked contrast to the previous configuration, a robust ectopic expression of bothHoxd11 and Hoxd10 in the anterior CNS was detected in embryos carrying these two deletions in cis. Ectopic expression ofHoxd11 and Hoxd10 was scored in anterior neural tube, in a subset of cells located dorsally (arrow), as well as in the developing hindbrain, an expression pattern clearly reminiscent of that seen forEvx2 (Fig. 5).Hoxd11 and Hoxd10 transcripts were also detected in the isthmus and in specific domains within the mesencephalon, where Evx2is also normally detected. Although the complete Evx2 neural pattern was not entirely recapitulated by either Hoxd11 or Hoxd10,the observed gain of expression encompassed several domains that were previously defined as specific for Evx2. Ectopic expression was also observed in heterozygous embryos with a weaker staining intensity, as expected if only one copy of each gene has been activated. From these results, we concluded that the insulation of the Hoxd complex from theEvx2 regulatory influence, in a large subset of CNS cells, was achieved as a result of the presence of two DNA fragments, one of them being RXII, the other(s) lying around the Hoxd12 locus.

The fact that the deletion of both Hoxd13 and Hoxd12 loci did not induce expression of Hoxd11 in the Evx2 CNS domains,indicated that RXII, which is located in the immediate neighbourhood of theEvx2 start site, was able by itself to mediate such an insulation. Interestingly, in Drosophila, a GAGA-dependent enhancer blocking activity was identified within the promoter region of the orthologous geneeven-skipped (eve), and this activity was shown to prevent 5′ located genes to respond to 3′ located enhancers(Ohtsuki and Levine, 1998). Thus, in both organisms, an enhancer blocking activity was found associated with the eve/Evx2 locus. Whether or not this observation has a phylogenetic meaning, rather than being a mere coincidence, remains to be established. In any case, the underlying molecular mechanisms are likely to be distinct, as RXII does not seem to contain any GAGA-binding site.

The morphological effect of expressing Hoxd genes in the developing anterior CNS, and hence the biological relevance of this insulation, was difficult to assess as Hoxd^{RXII;del(13-12)}homozygous specimens died at birth. However, this lethality may not be directly associated to the abrogation of insulation, as neonatal death was also observed for Hoxd^del(13-12) homozygous animals, i.e. animals that carried a wild-type RXII and, consequently, did not expressHoxd genes in anterior CNS. In this latter configuration, the deletion of both Hoxd13 and Hoxd12 induced the mis-expression of other Hoxd genes in a variety of embryonic structures, which may have caused lethality (data not shown). Consequently, it is as yet unclear whether such an insulator activity is required to prevent one particular gene to be expressed in developing CNS, or alternatively, if all posterior Hoxd genes would be equally detrimental when expressed there. To precisely assess the biological relevance of this insulation mechanism, specific gain of expression of 5′ Hoxd genes, using conventional transgenic approaches, will be necessary.

The presence, at one extremity of a Hox gene cluster, of sequences with insulating potential suggests a general requirement for isolating these chromosomal loci from their surrounding genomic contexts. Interestingly,various gene complexes seem to implement different mechanisms to protect themselves from regulatory interferences(Bell et al., 2001). For example, the β-globin gene complex, which shows some analogies with Hox clusters in its functional organisation, is flanked by sequences carrying properties of insulators (Bell et al.,1999; Saitoh et al.,2000). These latter sequences were proposed to prevent crosstalk between β-globin regulation, on the one hand, and unrelated regulatory influences emanating from closely located genes, such as those encoding odorant receptors, on the other (Bulger et al., 1999; Prioleau et al.,1999). This insulating potential was tightly associated with the 5′ HS4 and the 3′ HS DNAse I hypersensitive sites(Bell et al., 1999;Saitoh et al., 2000). These sites were identified in all cell types and tissues examined, suggesting that insulation of this gene complex is a rather generic mechanism with little cell specificity. By contrast, the insulating activity described in this paper,which prevents Hox genes from responding to upstream located CNS enhancers,was ineffective in a different cellular context. Indeed, the same series of genes was able to respond to another remote enhancer sequence, also located upstream the cluster, which controls Hoxd gene expression in developing digits (Spitz et al.,2001). This indicates that insulation of the Hoxd cluster is tissue-specific; it is effective in CNS cells, but not in limb mesenchymal cells (Kmita et al.,2002).

In the Drosophila Bithorax complex (BX-C), the gene orthologous to mammalian 5′ Hoxd genes (AbdB) is controlled, in defined parasegments, by a series of regulatory elements(Boulet et al., 1991;Celniker et al., 1990;Sanchez-Herrero, 1991). Such sequences (Iab genes) are often flanked by frontabdominal elements (Fab genes), which display insulating or boundary properties. Fab sequences are essential for proper parasegmental identity as they prevent crosstalk between distinct Iab (Barges et al.,2000; Mihaly et al.,1997; Zhou et al.,1999). Instead, in the vertebrate Hoxd complex, the insulating activity may rather reflect a general, complex-wide protection against anterior CNS regulation, rather than a way to implement properly a regulatory circuitry in space and time, as is the case forDrosophila. Therefore, it is unlikely that the mechanisms involved in these two processes serve identical purposes. It is nonetheless possible that RXII, as do Fab8 and the promoter targeting sequences (PTS)identified adjacent to it (Zhou and Levine, 1999), contains both insulating and `enhancer positioning'activities. Indeed, the bipartite RXII element was also shown to be involved in the mechanism that triggers preferential interaction between the digit enhancer and the most 5′ Hoxd gene (Kmita et al., 2002a). Therefore, the digit enhancer may have a `positioning activity', which might help to bypass the RXII blocking activity in limbs, in a way related to the PTS element which was shown to allow distal enhancers to overcome theFab8 insulation activity (Zhou and Levine, 1999). This capacity of the digit enhancer to overcome the effect of RXII may not be shared by neural enhancers which, as a consequence, would not be capable of bypassing RXII in CNS cells.

We show that only a combined deletion of both region XII and an 18 kb piece of the cluster would lead to ectopic transcription of both Hoxd11 andHoxd10 in CNS. This observation suggests that the DNA fragment that is able, along with RXII, to insulate the Hoxd complex lies around the Hoxd12 transcription unit. Two DNA fragments were shown to display significant interspecies sequence conservation within this interval;regions XI and X (Beckers and Duboule,1998; Hérault et al.,1998a). A role for RXI in insulation is unlikely as: (1) it has no counterpart in the fish genome(Hérault et al.,1998a); and (2) its deletion together with RXII inHoxd^RXII;del(13) animals had no apparent effect. Therefore, region X appears as the best candidate element to mediate this activity at the Hoxd12 locus. However, its inactivation in vivo,through targeted deletion, had no detectable effect upon 5′Hoxd gene regulation, similar to the case of RXII. This unexpected observation was tentatively explained by the existence of redundant regulatory processes (Beckers and Duboule,1998).

Regulatory redundancy is a difficult concept to accommodate with our current views of gene regulation. However, if we assume that both regions have insulating potentials, we may understand redundancy as a property associated with one particular cellular context. For example, in order to be functional in a given cell type, RXII may require factors partially specific for this cell type, to properly insulate the cluster. Likewise, in another cell type,RX may recruit a different set of factors to insulate the cluster from the influence of a different enhancer. In the case where both sets of factors would be present in CNS cells, both insulation processes would operate, hence only multiple deletions in cis would reveal this mechanism. Accordingly, the evolution and stability of either one of these two regions might have been driven separately, in different contexts, to become redundant in CNS cells. In this scheme, the question nevertheless remains as to why single deletions have no visible effect, at least in the original context wherein a given element is specifically required? Such tissues or organs might simply have been overlooked; they may, for example, involve vertebrate specific functions (rather recent evolutionary features), the alteration of which may have as yet escaped our attention.

The Hoxd cluster has been thoroughly investigated, in vivo, for the functional relevance of evolutionary conserved DNA sequences. In its most posterior part, i.e. between Hoxd10 and Evx2, five stretches of non-coding sequences were found significantly conserved amongst vertebrates. Using targeted approaches in ES cells, all five sequences were either deleted, mutagenised and/or exchanged for an orthologous sequence(Gérard et al., 1996;Beckers and Duboule, 1998;Hérault et al., 1998a)(this work). Interestingly, although in some cases, slight variations in the expression of the neighbouring genes were scored, none of these drastic genetic modifications led to major regulatory alterations, a counter intuitive observation that is at odd with current speculations regarding sequence conservation outside coding sequences. The results presented in this paper may shed some lights on this puzzling issue, as they suggest that such sequences might relate to high order regulatory processes, rather than to gene-specificcis-acting controls.

We thank Nadine Fraudeau and Marianne Friedli for technical assistance, as well as Takashi Kondo for sharing data, and colleagues from the Duboule laboratory for discussions and comments on the manuscript. This work was supported by funds from the Canton de Genève, the NCCR `Frontiers in Genetics', the Swiss National Research Fund, the Claraz, Latsis, Cloetta and Louis-Jeantet foundations.