Direct crossregulation between retinoic acid receptor β and Hox genes during hindbrain segmentation

During embryogenesis, the vertebrate hindbrain is transiently segmented into seven lineage-restricted units termed rhombomeres(Vaage, 1969; Fraser et al., 1990; Birgbauer and Fraser, 1994). These underlie the subsequent metameric arrangement of sensory and motor nerves within the adult brainstem. Early in hindbrain development,neuroepithelial cells are assigned a specific anterior-to-posterior (AP)identity, depending on their rhombomere of origin. This AP positional identity influences final cell fate and is regulated, at least in part, by the segmental expression of the Hox gene family of homeodomain transcription factors (reviewed by McGinnis and Krumlauf, 1992; Lumsden and Krumlauf, 1996; Capecchi,1997). To pattern the normal repertoire of cell fates along the AP axis of the hindbrain, the anterior expression borders of at least eleven Hox genes from paralogue groups 1-4 must be accurately aligned at specific interfaces between rhombomeres.

Many different genetic inputs are known to regulate the multigene complexes in which Hox genes reside. These are integrated in a stage-dependent manner by multiple enhancer elements, some of which act on only one Hox gene, while others are shared between several members of a complex (reviewed by Trainor et al., 2000; Kmita and Duboule, 2003). Within the presegmented hindbrain, initiation of Hox gene expression requires a key intercellular signal derived from vitamin A, called retinoic acid (RA)(reviewed by Marshall et al.,1996; Eichele,1997; Gavalas,2002). This small lipophilic molecule is synthesized by retinal dehydrogenase 2 (Raldh2/Aldh1a2) within the developing paraxial mesoderm and induces Hox expression in the adjacent neural tube(Niederreither et al., 1997; Berggren et al., 1999; Swindell et al., 1999). The timing of Raldh2 expression correlates well with the stages at which grafted chick somites are known to possess neural Hox-inducing activity(Grapin-Botton et al., 1995; Itasaki et al., 1996). Moreover, loss of Raldh2 function produces AP patterning abnormalities and disrupted Hox expression patterns similar to those resulting from retinoid deficiency (Maden et al.,1996; Gale et al.,1999; Niederreither et al.,2000; White et al.,2000; Begemann et al.,2001). RA concentration is not only spatiotemporally regulated at the synthetic level but also via its local inactivation by cytochrome P450 monooxygenases (McCaffery and Drager,2000; Abu-Abed et al.,2001; Niederreither et al.,2002). The importance of regulating RA availability during hindbrain development is underscored by several studies showing that the addition of exogenous RA produces dramatic rostral shifts in Hox expression boundaries (Morriss-Kay et al.,1991; Conlon and Rossant,1992; Dekker et al.,1992; Marshall et al.,1992). These and more recent studies(Gould et al., 1998; Dupé and Lumsden, 2001; Nolte et al., 2003) are consistent with the hypothesis that Hox genes are differentially sensitive to the level and/or timing of RA signalling and that this contributes to specifying their diverse expression borders within the hindbrain.

Specific enhancers containing retinoic acid response elements (RAREs)mediate RA induction of Hox gene transcription within the hindbrain. The in vivo characterization of these short DNA sequences from two genes of Hox paralogue group 1 (Hoxa1 and Hoxb1) and two from group 4(Hoxb4 and Hoxd4) has revealed that they function by binding retinoic acid receptors (RARs) and retinoid X receptors (RXRs), probably as RAR/RXR heterodimers (Marshall et al.,1994; Frasch et al.,1995; Morrison et al.,1996; Dupé et al.,1997; Kastner et al.,1997; Gould et al.,1998; Mascrez et al.,1998; Studer et al.,1998; Zhang et al.,2000). Both classes of receptors interact directly with RA: the three RARs (α, β and γ) are capable of binding all-trans and 9-cis forms, whereas the α, β and γ RXRs bind 9-cis RA only(reviewed by Chambon, 1996). Owing to extensive functional redundancy, the crucial role of RARs in hindbrain segmentation and also in Hox regulation only becomes apparent when two are removed simultaneously or when the activities of all three RARs are attenuated using a chemical antagonist(Dupé et al., 1999; Dupé and Lumsden, 2001; Wendling et al., 2001). Thus,although RARβ is unique amongst RARs in having an anterior border of expression at the r6/r7 junction, its inactivation has no apparent effect on caudal hindbrain segmentation or Hox transcription(Ruberte et al., 1991; Ghyselinck et al., 1997; Dupé et al., 1999; Folberg et al., 1999). However, double mutants for Rarb and Rara do display a severely enlarged r5, loss of the r6/r7 boundary and absence of hindbrain Hoxd4 expression (Dupé et al., 1999). On the basis of these and other observations, it has been proposed that the normal role of Rarb is to mediate a caudal increase in RA signalling, which is necessary for Hoxd4 activation(Ghyselinck et al., 1997; Wendling et al., 2001).

Although much progress has been made in understanding how hindbrain patterns of Hox expression are generated, less is known about the regulatory mechanisms that govern the neural expression of their key upstream regulators,the RARs. We focus on Rarb, and use a combination of genetic,transgenic and biochemical analyses in the mouse and chick to dissect the mechanism generating its segmental expression border. These studies reveal that Rarb uses a two-step transcriptional regulatory mechanism:induction by a mesodermal RA signal and maintenance by Hoxb4 and Hoxd4. They also show that Rarb is a direct transcriptional target of Hoxb4, thus providing evidence that transcriptional regulation between the Hox and RAR gene families is bidirectional. We describe how these results imply the presence of a self-organizing feedback circuit that corrects for small initial differences in the anterior expression borders of multiple Hox and RAR genes, bringing them into register at a rhombomere boundary.

At least three mouse embryos of the appropriate genotype were analysed for each data point. Rarb^-/- embryos were generated and identified from heterozygous siblings by Southern blotting as described(Ghyselinck et al., 1997; Niederreither et al., 1999)and Raldh2^-/- embryos were identified by a characteristic`squashed somite' phenotype. The Rarb1lacZ and Rarb2lacZlines described here correspond to Rarb1/lacZ2 and Rarb2/lacZ respectively(Mendelsohn et al., 1994). Hoxb4^-/-; Hoxd4^-/- double mutant embryos were generated as previously described(Gould et al., 1997). Yolk sacs of Hoxb4^-/-; Hoxd4^-/- or Hoxb4^-/-; Hoxd4^-/-; Rarb1lacZ/+ or Hoxb4^-/-;Hoxd4^-/-; Rarb2lacZ/+ embryos were genotyped by PCR using the following primer combinations.

Hoxb4: CGGCTGGAAGCCGCTCTCTCGC (+ allele), CTGCATCCATGACACAGGCAAACC(- allele) and GAGCCCTATGTAAATCCTGGTGTTG (common).

Hoxd4: CCTACACCAGACAGCAAGTCCTAG (+ allele),CCCGTGATATTGCTGAAGAGCTTGG (- allele) and CTCGGGCAGGAAGGTAACCTAGTC (common)

lacZ: GCGACTTCCAGTTCAACATC (LZ3) and GATGAGTTTGGACAAACCAC(ZT4).

All-trans RA (25 mg/ml stock in dimethylsulphoxide) was diluted 1:10 in sesame oil and 200 μl per pregnant dam administered by gavage at E9.25 (∼25 mg/kg bodyweight) as previously described(Gould et al., 1998).

Staining with X-gal or a monoclonal antibody directed against murine Hoxb4 was as described (Gould et al.,1997). We have deposited anti-Hoxb4 at the Developmental Studies Hybridoma Bank(http://www.uiowa.edu/~dshbwww/). Rabbit anti-Krox20 antibody (Cambridge Bioscience) was used at a dilution of 1:100. For analysis by confocal microscopy, specimens were stained with Alexa Fluor 488 secondary antibodies (Molecular Probes), flatmounted in Vectashield with or without propidium iodide (Vector Laboratories), and images prepared using projections of several sections. In situ hybridization was performed as described (Wilkinson, 1992)except that, for Rarb, hybridization was at 70°C. Antisense riboprobes incorporating either digoxigenin or fluorescein were made using Rarb (Zelent et al.,1991), Krox20(Wilkinson et al., 1989a) or Hoxb4 (Graham et al.,1988) templates.

RARβ intron sizes were determined by BLAST-like Alignment Tool (BLAT)(Kent, 2002) comparisons of known RARβ mRNAs (Zelent et al.,1991; Mendelsohn et al.,1994) with the UCSC Genome Browser Databases(Karolchik et al., 2003) (see http://genome.ucsc.edu)using the following genome builds: human May 2004 (hg17), chimp Nov 2003 (pan Tro1), mouse May 2004 (mm5), rat Jun 2003 (rn3) and dog Jul 2004 (canFam1). The distance between the HP site of the distal enhancer and the ATG of the first coding exon (E2) of the RARβ1 mRNA isoform is at least 241 kb in chimp and 215 kb in rat, although gaps in these genome builds indicate that these distances may be larger. There are no gaps in human (240 kb interval)and only a small number of gaps in dog (∼232 kb interval) and mouse(∼284 kb interval). The position of the P1 promoter is not yet clear in rat, chimp and dog, as 5′ untranslated exon(s) of RARβ1 have yet to be identified. For human, however, Toulouse et al.(Toulouse et al., 1996) have reported the 5′ end of the RARβ1 isoform (Genbank Accession Number U49855), which our BLAT comparisons show to be encoded by six non-coding exons, three of which lie upstream of the HP site. Multispecies HP site comparisons used MULTIZ (Blanchette et al.,2004).

GST-Hoxb4 and Hoxb4 antibody production and EMSA were performed as previously described (Gould et al.,1997), except that Cy5 oligonucleotide labelling was used and gels were imaged on a Typhoon 8700 (Amersham Biosciences). Oligonucleotide sequences were as described (Gould et al.,1997) or as follows (Hox/Pbx sites in bold, mutated residues underlined): HS1+HS2 WT (Hoxb4),GAGAATTATACAGAAAACCATTAATCACTT; HP WT (Rarb),TTTGAGGAGCAGGGTGATAAATAATGGGGCTTTTCCA; and HP MUT (Rarb),TTTGAGGAGCAGGGTGGGCCCGCCGGGGGCTTTTCCA.

The distal enhancer (a 2.3 kb NheI fragment) of Rarb was isolated by PCR from mouse genomic DNA using primers AGAATGTGTGTGCTGACTCTGC and AAGCAGTCTTACCAGGAGGG, cloned into pGEM-T (Promega) and then transferred as a SacII-NotI fragment into the BGZ40 vector(Maconochie et al., 1997). The HP site was mutated using the Quick Change kit (Stratagene) with the following 61 mer: CCTTGTGAAGTCCCCTTTGAGGAGCAGGGTGGGGCTTTTCCAATTGTTATTTGCCAAAAGG. For the 3x HP construct, annealed oligonucleotides GGCCGCAAGCTTGAGCAGGGTGATAAATAATGGGGCTTGAGCAGGGTGATAAATAATGGGGCTTGAGCAGGGTGATAAATAATGGGGCTTCCGC and GGAAGCCCCATTATTTATCACCCTGCTCAAGCCCCATTATTTATCACCCTGCTCAAGCCCCATTATTTATCACCCTGCTCAAGCTTGC were cloned into NotI-SacII cut pBSKS and then transferred into BGZ40. All constructs were confirmed by DNA sequencing.

Hamburger-Hamilton stages 9-11 chick embryos were electroporated as described (Itasaki et al.,1999). BGZ40 derivatives (1 μg/1 μl) were co-electroporated with Fast Green and CMV-GFP. Only embryos showing strong GFP expression throughout the neural tube, were analyzed further.

The murine hindbrain becomes segmented into morphologically visible rhombomeres between embryonic day 8.5 (E8.5) and E9.5. By E9.5, the anterior borders of expression of Hoxb4, Hoxd4 and Rarb are all congruent at the r6/r7 junction (Fig. 1A,B) (Wilkinson et al.,1989b; Hunt et al.,1991; Ruberte et al.,1991).

We compared the regulation of Rarb and Hoxb4 by mapping their borders of expression within the presegmented hindbrain at E8.5. At this stage, Rarb and Hoxb4 have anterior limits of expression that are much less sharp than at E9.5. Using the otic sulcus as a morphological landmark for the presumptive r5/r6 boundary, we were surprised to find that the E8.5 expression border of Rarb lies significantly more anterior than that of Hoxb4(Fig. 1C,D). The rostral borders of both genes were then mapped more accurately by performing double labelling using Krox20 expression as a landmark for presumptive r5(Wilkinson et al., 1989a). In this way, neural expression was observed up to the level of the future r5/r6 boundary for Rarb but only as far as the approximate location of presumptive r6/r7 in the case of Hoxb4(Fig. 1E,F). These results reveal that the neural expression borders of Hoxb4 and Rarbare different at early stages but become congruent at around the time when segmentation into r6 and r7 first becomes morphologically visible.

We then addressed the in vivo mechanisms regulating Rarb and Hoxb4. For several Hox genes, it is known that hindbrain expression is initiated prior to rhombomere formation by an inductive process involving RA synthesis by the paraxial mesoderm (see Introduction). In the case of Hoxd4, RA induction in the hindbrain is abolished in embryos lacking the activity of the RA-synthetic enzyme Raldh2(Niederreither et al., 2000). We first focused on Hoxb4 and examined the role of Raldh2 during its induction phase at E8.5 (Gould et al.,1998). At this early stage, the reduction in the amount of tissue in the postotic territory of the hindbrain that occurs progressively from E8.25-E9.5 in Raldh2^-/- embryos is minimal(Niederreither et al., 2000). We find that both the mRNA and protein products of Hoxb4 are only very weakly expressed in the presegmented hindbrain of Raldh2^-/- embryos (Fig. 2A-D). Thus, like Hoxd4, the early hindbrain expression domain of Hoxb4 requires RA and, as the major site of Raldh2expression at E8.5 is the paraxial mesoderm, this requirement probably signifies a mesodermal-to-neural induction. At spinal levels, however, a different mechanism of Hoxb4 regulation must operate as neural expression remains unaffected at the mRNA level, paralleling previous Hoxd4 findings (Niederreither et al., 2000). As we observe that Hoxb4 expression in Raldh2 mutants is dramatically reduced at the protein level in spinal regions (compare Fig. 2A,B with Fig. 2C,D) it is likely that,as yet uncharacterized, post-transcriptional Hox-regulatory mechanisms are involved (Brend et al., 2003; Dasen et al., 2003).

We next sought to identify the mechanism initiating the expression of Rarb in the presegmented hindbrain. As with Hoxb4, we examined Rarb expression in E8.5 embryos lacking Raldh2 activity. Strikingly, we find that Rarb mRNA is undetectable within the presumptive hindbrain and spinal cord of these mutant embryos(Fig. 2E,F). This demonstrates clearly that RA, which is produced by the mesoderm, is required to induce the early neural expression of Rarb. Hence, the repertoire of RA-inducible genes within the hindbrain extends beyond the Hox family to include Rarb. Our combined results demonstrate that, despite initial differences in their rostral expression borders, both Rarb and Hoxb4 are induced in the hindbrain at the mRNA level by the same mesodermal source of RA.

We next focused attention on later stages of development, to identify the mechanism regulating the expression border of Rarb in the segmented hindbrain. During the E8.5-E9.5 time window, the anterior expression border of Rarb regresses by approximately one rhombomere in length, matching the sharp segmental border of Hoxb4 expression at r6/r7. Given this late border congruency, we tested the possibilities that either Hoxb4regulates Rarb or that Rarb regulates Hoxb4. Functional overlap among RAR genes results in the overall pattern of Hoxb4 mRNA appearing normal in Rarb single mutants(Folberg et al., 1999). In addition, confocal analysis with single-cell resolution in Rarbsingle mutants provided no evidence for a non-redundant requirement for this particular RAR in specifying the sharpness or positioning of the r6/r7 border of Hoxb4 protein expression at E10 (data not shown). However, a previous study showed that blocking the activities of all types of RAR abolished hindbrain Hoxb4 induction (Gould et al.,1998). Together with the analysis of mutants lacking multiple RARs(see Introduction), we conclude that, although the set of RARs participating in Hoxb4 induction does include Rarb, other types of RAR can substitute for its role.

To test for the reciprocal regulatory relationship between Rarband Hoxb4, we generated double homozygous mice lacking the functions of Hoxb4 and Hoxd4. The activities of both of these paralogues were removed, as previous studies indicated that they play redundant roles in activating a Hox-responsive enhancer in r7(Gould et al., 1997). In Hoxb4^-/-; Hoxd4^-/- embryos, caudal hindbrain tissue aberrantly expresses several region-specific molecular markers (data not shown). We find that the early phase of neural Rarbexpression is normal in double homozygous embryos at E8.5 and E9.0(Fig. 3A and data not shown). Moreover, no alteration in the late phase of Rarb expression was observed at E10.5, when only two out of four alleles were mutated(Hoxb4^+/+; Hoxd4^-/- embryos; Fig. 3B). However, in double homozygous embryos, older than E9.0, the position of the Rarb border is significantly disrupted. Thus, by E9.5, the Rarb expression border in Hoxb4^-/-; Hoxd4^-/- embryos lies approximately one rhombomere in length caudal to the normal r6/r7 location(Fig. 3C,D). In addition, at this stage, the border is less sharply defined than in wild-type embryos. One day later, at E10.5, the caudal shift and diffuseness of the border in Hoxb4^-/-; Hoxd4^-/- embryos have become more dramatic, leaving most of the hindbrain void of detectable Rarbexpression (Fig. 3E,F). These results demonstrate that Rarb expression in the hindbrain is initially independent of group four Hox genes but subsequently becomes positively regulated by overlapping inputs from Hoxb4 and Hoxd4. Late Hox-dependent regulation comes into play at the time of morphological segmentation, serving to sharpen and fix the regressing expression border of Rarb at the r6/r7 boundary.

The results thus far are consistent with a two-phase model for the regulation of Rarb within the developing hindbrain. Expression is first induced within the presegmented hindbrain by an RA signal from the mesoderm, generating a diffuse border at the approximate level of presumptive r5/r6. This early phase of expression at E8.5 is independent of Hoxb4and Hoxd4. Approximately 1 day later, around the time of rhombomere boundary formation, the border becomes sharpened and keyed onto the r6/r7 junction by overlapping regulatory inputs from Hoxb4 and Hoxd4. Importantly, these results demonstrate clearly that RARs can act downstream of Hox genes. In combination with a large body of published evidence already placing them upstream, our results reveal the existence of bidirectional regulation between the Hox and RAR gene families.

To dissect the Hox-RAR feedback mechanism and, in particular, the way in which RA and Hox inputs are temporally coordinated by the Rarb gene,we used two transgenic regulatory DNA constructs from this locus(Mendelsohn et al., 1991; Mendelsohn et al., 1994): (1) Rarb2lacZ, expressing β-galactosidase under the control of a 3.8 kb genomic fragment (termed the proximal enhancer) that includes the proximal(P2) promoter; and (2) Rarb1lacZ, containing a 2.3 kb NheI genomic fragment (termed the distal enhancer) that encompasses the distal (P1)promoter. At E9.5, Rarb1lacZ and Rarb2lacZ drive neural expression with anterior borders at r6/r7 and r7/r8, respectively(Mendelsohn et al., 1994). However, in the context of the endogenous Rarb gene, transcripts from both the P1 and the P2 promoters are known to be expressed with a r6/r7 boundary at E10.5 (Mollard et al.,2000). This indicates that, when in their normal chromosomal environment, the proximal and distal enhancers may not be selective for one promoter. More specifically, it suggests that the distal enhancer may be capable of regulating both P1 and P2.

We analysed early proximal enhancer activity, using E8-E8.5 embryos transgenic for Rarb2lacZ. We find that Rarb2lacZ is robustly expressed within the neural plate of two-somite stage embryos(Fig. 4A). By the seven-somite stage, reporter activity is observed extending up to the level of the otic sulcus, corresponding to the future r5/r6 boundary(Fig. 4B). This pattern recapitulates the anterior limit of Rarb mRNA at E8.5 that we describe here (compare Fig. 1Dwith Fig. 4B). One day later,at E9.5, proximal enhancer activity has regressed dramatically and is now missing from most of the caudal hindbrain but remains present in the spinal cord (Fig. 4C)(Mendelsohn et al., 1994). This late loss of expression throughout the caudal hindbrain is more extensive than the regression observed for the endogenous gene, indicating that the proximal enhancer accounts for the early but not the late phases of Rarb expression. To investigate the factors regulating the proximal enhancer, we initially examined Rarb2lacZ expression in Hoxb4^-/-; Hoxd4^-/- embryos. However,removing the functions of both Hox paralogues does not affect the neural Rarb2lacZ expression pattern at E9.5 or E10.5(Fig. 4C,D; data not shown).

Previous studies have shown that the proximal enhancer region(Rarb2lacZ construct) contains a RARE required for its RA responsiveness in cell culture (de Thé et al., 1990; Hoffmann et al., 1990; Sucov et al., 1990). Furthermore, in an embryonic context, the Rarb2lacZ transgene responds rapidly to a pulse of exogenous RA at E9.25, giving a robust response in the hindbrain within 4 hours (Fig. 4E) (Mendelsohn et al.,1994). Strong ectopic expression is observed up to the r3/4 boundary with levels in r5 much lower than in r4 and r6. To assess whether the proximal enhancer is sufficient to maintain a long-lasting in vivo response to a transient RA signal, we examined Rarb2lacZ activity 24 hours after a single RA dose. In contrast to the 4 hour time point, we find that no ectopic lacZ expression is detectable after 24 hours(Fig. 4F). At this time, neural Rarb2lacZ activity is confined to spinal cord levels, resembling its pattern in untreated E9.5-E10.5 embryos (compare Fig. 4D with 4F). The rapid and transient nature of the RA response indicates that neural activity of the proximal enhancer provides a readout for RA availability and, together with the previous cell culture studies, suggests that this represents a direct response to RARs. The Rarb2lacZ analysis also shows that while the proximal enhancer responds to RA and accounts for the early Raldh2-dependent phase of Rarb expression, it does not contain sequences capable of stably maintaining an RA response and it is not regulated by Hoxb4 or Hoxd4.

We analysed the activity pattern of the second transgenic construct, Rarb1lacZ, containing the distal enhancer. In contrast to Rarb2lacZ, no reporter expression is seen at E8.5(Fig. 5A). Approximately 24 hours later, activity within the neural tube is initiated with a well-defined limit at r6/r7 that is stably maintained until E10.5 and even later stages(Fig. 5B and data not shown). Therefore, the distal enhancer recapitulates the late but not the early phase of expression of the endogenous Rarb gene. To assess whether the distal enhancer is regulated in a similar way to its proximal counterpart, Rarb1lacZ embryos were treated with a single dose of exogenous RA. As no expression was detected anterior to r6/r7 after 4 or 24 hours, we conclude that the distal enhancer differs from the proximal enhancer in that it does not mediate a direct ectopic response to RA(Fig. 5C,D). As the r6/r7 expression border of distal enhancer activity at E9.5-E10.5 is shared with Hoxb4 and Hoxd4, we next examined Rarb1lacZexpression in embryos carrying various combinations of knockout alleles of these two group 4 Hox paralogues. Although loss of one of the four wild-type alleles has no reproducible effect on expression at E10.5, removing two or three copies does lead to significantly reduced neural tube expression(Fig. 5E-G). Removing all four alleles, in Hoxb4^-/-; Hoxd4^-/-embryos, leads to the complete abolition of all distal enhancer activity with 100% penetrance (Fig. 5H). Therefore the r6/r7 border and all other aspects of distal enhancer activity display an absolute requirement for Hoxb4 and Hoxd4function.

To investigate whether the regulation of Rarb by Hox4genes might be direct, we searched the 2.3 kb distal enhancer sequence for potential Hox response elements. A close match to a bipartite consensus binding site for Hox and Pbx proteins (HP) was identified 83 bp upstream from the characterized murine RARβ1/3 transcription start site(Fig. 6A)(Mendelsohn et al., 1994). This 10 bp element and its surrounding sequences are highly conserved in syntenic regions upstream of Rarb in other mammalian genomes such as human, chimp, rat and dog (Fig. 6B).

To determine whether Hoxb4 was capable of direct binding to the HP site, a double-stranded oligonucleotide containing the motif was used in electrophoretic mobility shift assays (EMSA, Fig. 6C). EMSA revealed that the HP site displays concentration-dependent binding of Hoxb4, similar to that seen with a previously characterized Hoxb4-binding site (HS1+HS2) from the Hoxb4 gene itself (Fig. 6C) (Gould et al.,1997). The specificity of the Hoxb4 protein interaction with HP was confirmed by supershifting the protein-DNA complex with a Hoxb4 antibody and by using various oligonucleotide competition assays(Fig. 6D). Whereas oligonucleotides corresponding to either the HP or the HS1+HS2 sites compete efficiently with one another for Hoxb4 binding, mutated variants (HS1+HS2 MUT and HP MUT: see Materials and methods) do not. Thus, Hoxb4 binds to the conserved HP site within the distal enhancer of Rarb in a direct and sequence-specific manner.

To test whether the conserved HP site is functionally relevant in vivo, its activity was monitored by chick in ovo electroporation. We first showed that the 2.3 kb murine distal enhancer can direct lacZ reporter expression within the developing chick hindbrain up to the r6/r7 boundary (17/18 embryos, Fig. 6E), the same pattern as observed in transgenic mouse embryos. When present in three copies, a 26 bp oligonucleotide spanning the murine HP site (3xHP) is sufficient to direct a weak r6/r7 pattern (8/12 embryos, Fig. 6F), indicating that rhombomere-specificity information resides within the HP site and/or its immediate flanking sequences. Moreover, within the context of the 2.3 kb distal enhancer, the HP site is required for activity as mutating it to a sequence unable to bind Hoxb4 (HP MUT) results in the complete abolition of lacZ expression (0/36 embryos, Fig. 6G). Taken together, the sequence analysis, EMSA and chick electroporation experiments provide strong evidence that Hoxb4 regulates the distal enhancer of Rarb in a direct manner via the HP site.

Many studies have focused on how one signalling molecule can trigger several different gene expression patterns, as in the morphogen paradigm. Less attention has been directed to the other important problem of how one signal can specify the precisely congruent expression of multiple genes. We have explored this issue in the context of how the expression domains of two RA-induced genes, Rarb and Hoxb4, become aligned at a segment boundary in the developing hindbrain. We focused on the regulation of Rarb and identified a two-step transcriptional mechanism. We now discuss this mechanism and how, in combination with the previously identified Hoxb4 regulatory mechanism, it provides new insights into the way in which sharp segmental borders of gene expression are generated.

Raldh2-dependent RA signalling from the paraxial mesoderm induces a diffuse and unstable border of Rarb mRNA within the presegmented hindbrain. In the absence of Raldh2, no Rarb expression is detected within the neuroepithelium, indicating that initial induction by RA is probably transduced by RARα and/or RARγ and not by Rarbitself. This also implies that early hindbrain expression of at least one of these RARs is independent of induction by RA. The presegmental phase of Rarb expression is regulated by the proximal RARE-containing enhancer, which responds rapidly and transiently to RA with similar kinetics to a RARE that is a known direct target of RARs (cf. Gould et al., 1998). Together with evidence that the proximal RARE directly binds RARs and mediates an RA-response in cultured cells (de Thé et al., 1990; Hoffmann et al., 1990; Sucov et al., 1990), our findings strongly suggest that the induction of neural Rarbexpression by RARs is direct. Hence, following initial neural induction via RARα and/or RARγ, RARβ would then be available to augment its own expression via direct binding to the RARE of the proximal enhancer.

Rarb is a well-known regulator of Hox genes, including those of paralogue group 4 but, surprisingly, our study indicates that reciprocal regulation also occurs. Together, our mouse knockout and transgenic analyses reveal that Rarb lies downstream of Hoxb4 and Hoxd4. Moreover, the combined EMSA and chick electroporation approaches indicate that regulation is mediated in a direct manner via Hoxb4 binding to the HP site of the Rarb distal enhancer. Given the similar homeodomain sequences and DNA-binding specificities of Hoxb4 and Hoxd4, it is likely that the observed genetic requirement for Hoxd4 is also mediated in the same direct manner.

We have shown that the mechanisms regulating both the early and late phases of Rarb expression within the hindbrain operate at the transcriptional level. At E8.5, the proximal retinoid-dependent enhancer recapitulates the initial diffuse and transient Rarb expression up to the presumptive r5/r6 border. Likewise, at E9.5-E10.5, the distal Hox-dependent enhancer maintains the stable segmental border at r6/r7 via a direct input from Hoxb4 and probably Hoxd4. Thus, the summation of the activities of the proximal and distal enhancers accounts for the establishment and maintenance of hindbrain expression of the endogenous Rarb gene.

The mechanism regulating Rarb in the presegmented hindbrain is similar to that of Hoxb4. Both genes are transcriptionally induced by a Raldh2-dependent RA source and both possess RARE-containing enhancers [for Hoxb4, this is termed the early neural enhancer (ENE)(Gould et al., 1998)] that direct neural expression with borders that recede after E8.5. These caudal shifts presumably reflect regression of the inducing ability of the paraxial mesoderm with increasing embryonic age(Itasaki et al., 1996; Gould et al., 1998). Although there are strong parallels between the RARE enhancers of Rarb and Hoxb4, there are also some differences. For example, proximal enhancer activity begins at around the two-somite stage, whereas ENE activity begins at the nine-somite stage. In addition, at E8.5, the anterior border of the Rarb proximal enhancer is at presumptive r5/6 but that of the Hoxb4 ENE is at presumptive r6/r7. Although the DNA element responsible for these expression differences is undefined, it may be relevant that the DR5 class of RARE, present in both enhancers, differs at 3/12 nucleotide positions.

The regulatory parallels between Rarb and Hoxb4 also extend to the later phase of segmental expression. Like Rarb, Hoxb4uses a two-step regulatory strategy of establishment and maintenance within the hindbrain, involving two enhancer elements that are mechanistically and physically separable (Gould et al.,1998). For Hoxb4, the late hindbrain element is termed the late neural enhancer (LNE) (Gould et al., 1997). Both the Rarb distal enhancer and the Hoxb4 LNE drive expression with a sharp r6/r7 border and respond to stabilizing inputs from group 4 Hox genes. In both cases, these late Hox inputs serve to halt the caudal regression of diffuse borders that were established by RARE-containing enhancers. However, when the functions of Hoxb4 and Hoxd4 are completely removed, Hoxb4 LNE activity is lost only from r7 (Gould et al., 1997), whereas Rarb distal enhancer activity is abolished within the entire neural tube. This suggests that, although group 4-6 Hox paralogues activate the Hoxb4 LNE(Gould et al., 1997), only some or all of the group 4 Hox genes may be capable of activating the Rarb distal enhancer.

We have argued that the early induction and late maintenance mechanisms regulating hindbrain Rarb transcription are strikingly similar to those of Hoxb4. As both genes are transcriptionally regulated by RARs and Hox proteins, numerous types of feedback loop are possible. Importantly,the Hox-to-Rarb direction of regulation that we have uncovered in this study identifies a new molecular link completing a circuit between the two different gene families. Combining our results with those of several previous studies (see Introduction), it can be seen that bidirectional Hox-RAR regulation forms the core of a complex genetic circuit involving direct and indirect interactions between Rarb, Hoxb4 and Hoxd4(Fig. 7). It is likely that the feedback circuit we describe also includes a fourth gene, as Hoxb3shares the LNE with Hoxb4 and thus upregulates its expression posterior to r6/r7 at late stages (Gould et al., 1997). One important requirement for the indirect type of feedback loop to occur is that the RAR→Hox and Hox→RAR interactions must overlap in developmental time rather than being strictly sequential. More specifically, RAR→Hox regulation must persist long enough after the initial stage of RA-dependent Hox induction at E8.5 to coincide with the onset of the reciprocal Hox→RAR regulation, first detectable at around E9.5. Consistent with this temporal overlap, the isolated ENE of Hoxb4 does remain active within the neural tube at E9.5(Gould et al., 1998).

The complex transcriptional feedback circuit identified here maintains gene expression and has the capacity to align the expression borders of multiple Hox and RAR genes at a single segment boundary. Early positional information conveyed by

RA induction is not sufficient to specify the late segmental expression border of Rarb or Hoxb4. Furthermore, although both genes interpret the same Raldh2-dependent signal, small differences in the responsiveness of their RARE-containing enhancers produce early expression borders that are out of register. Despite this, and some variations in the Hox-responsive late enhancers of Hoxb4 and Rarb, when all of the components are appropriately combined within the in vivo genetic circuit,they function to generate congruent expression at the r6/r7 boundary. In addition to these self-organizing properties, the Hox-RAR circuit is also genetically robust such that overlapping inputs from both gene families ensure that loss of any one gene has a minimal effect on the expression of the others. For example, we found that the r6/r7 expression border of Hoxb4 is unaffected by loss of all Rarb activity and,conversely, that the Rarb mRNA pattern remains normal when all Hoxd4 activity is removed. Finally, the identification of the above border-matching circuit was initiated by the observation that group 4 Hox genes share a segmental expression pattern with Rarb. As Rara is reported to be expressed up to the r3/r4 boundary(Ruberte et al., 1991),coincident with the expression limits of Hoxa1 and Hoxb1, a related Hox-RAR feedback circuit may also operate at this segment border.

We thank Cathy Mendelsohn, Eumorphia Remboutsika and Robin Lovell-Badge for providing transgenic and knockout lines; and Sarah Hodges, Peter Dawson and Kathleen Mathers for technical support and advice. We also thank Siew-Lan Ang,James Briscoe, Francois Guillemot, Nobue Itasaki, Juan-Pedro Martinez-Barbera,Jean-Paul Vincent, David Wilkinson and members of the Gould laboratory for advice and critical reading of the manuscript. This work was supported by the UK Medical Research Council, the Stowers Institute for Medical Research, CNRS,INSERM and the Collège de France. S.W.T. was supported by Boehringer Ingelheim Fonds.