Hoxa2 downregulates Six2 in the neural crest-derived mesenchyme

The facial skeleton is formed by neural crest cells(Le Douarin and Kalcheim,1999). Its development can be summarized in three main steps:cranial neural crest cells are produced in the developing neural tube, they migrate to the ventral region of the head (frontonasal mass and branchial arches), and, at the endpoint of migration, they form the different skeletal elements, each with a precise shape and in a precise position (reviewed by Chambers and McGonnell, 2002;Gammil and Bronner-Fraser, 2003; Helms and Schneider, 2003; Santagati and Rijli, 2003). As the neural crest is a dynamic population, a central issue in craniofacial development has been where these cells receive the patterning information to forge the different skeletal elements. Recent experimental observations indicate that the cranial neural crest is not pre-patterned before migration, but rather migrates as a naïve population. The epithelia, in particular the foregut endoderm, produce signals received by the neural crest at its final destinations, and these are interpreted to form the different skeletal elements(Couly et al., 2002).

Hox genes are key developmental regulators required to specify segmental identity in the developing embryo (Carroll,1995). Genetic analyses in flies and vertebrates have yielded extensive knowledge about the developmental processes regulated by Hox genes,but the molecular events directly controlled by these genes are still largely unknown (for reviews, see Krumlauf,1994; Zakany and Duboule,1999; Burke, 2000; Trainor and Krumlauf, 2001; Alonso, 2002).

Hox-negative and Hox-positive domains can be distinguished in the cranial neural crest (Prince and Lumsden,1994; Graphin-Bottom et al., 1995; Couly et al., 1996; Köntges and Lumsden,1996). The visceral skeleton of the face derives from the Hox-negative first branchial arch, with a very limited contribution from the more posterior Hox-positive branchial arches. Moreover, ectopic expression of Hox genes in anterior Hox-negative crest cells in chicken(Creuzet et al., 2002) results in the absence of the facial skeleton. A similar effect is observed in mouse,following Hoxa2 overexpression in the head mesenchyme(Kanzler et al., 1998).

Among Hox genes, Hoxa2, together with Hoxb2, shows the most anterior domain of expression in the cranial neural crest, corresponding to the population that migrates to the second branchial arch(Prince and Lumsden, 1994; Nonchev et al., 1996; Mallo, 1997). In mouse,disruption of the Hoxa2 gene mainly affects second branchial arch development. Second arch skeletal elements (stapes, styloid process, lesser horn of the hyoid bone) are transformed into first arch-specific skeletal elements (incus, malleus and tympanic ring), arranged in a mirror image disposition to their first arch counterparts(Gendron-Maguire et al., 1993; Rijli et al., 1993; Barrow and Capecchi, 1999). This arrangement suggests a common source of information located between the first and second branchial arch (Rijli et al., 1993; Mallo and Brandlin,1997), with a different interpretation of this common signal in neural crest cells expressing Hoxa2. More specifically, Hoxa2 is thought to interfere negatively with the response of the neural crest cells to skeletogenic cues(Couly et al., 2002; Bobola et al., 2003);accordingly, chondrogenesis in the second arch takes place exclusively in areas that are free of Hoxa2 expression(Kanzler et al., 1998). Thus, Hoxa2 patterns the second arch skeleton by limiting its formation. Consistent with a role in controlling the size of the condensations,overexpression of Hoxa2 in chick and in frog confers a late,postmigratory, patterning role to Hoxa2 during development of the branchial arch-derived skeleton(Grammatopoulos et al., 2000; Pasqualetti et al., 2000).

Because of the negative effect on the formation of the facial skeleton, the absence of Hox gene expression in the anterior part of the embryo has been proposed as a crucial factor to allow the evolution of the head and lower jaw in gnathostomes (Creuzet, 2002; Manzanares and Nieto, 2003). As with most developmental processes regulated by Hox genes, the molecular cascade initiated by these genes to culminate in the inhibition of craniofacial skeletogenesis is unknown.

In a subtraction approach, designed to clarify the molecular cascades initiated by Hoxa2 to control skeletogenesis in the second branchial arch(Bobola et al., 2003), we have found Six2 as a gene regulated by Hoxa2.

The Six family of homeobox transcription factors, characterized by a Six domain and a homeodomain, counts six members in mammals(Kawakami et al., 2000). Members of this family share transcriptional properties and the ability to interact physically and functionally with Eya proteins, both in Drosophila and in vertebrates(Bonini et al., 1997; Pignoni et al., 1997; Heanue et al., 1999; Ohto et al., 1999; Ikeda et al., 2002; Ozaki et al., 2002; Li et al., 2003; Ruf et al., 2004). Six genes differ largely in their expression pattern during embryogenesis, and gene inactivation experiments have revealed that these genes control a variety of developmental processes (Klesert et al.,2000; Carl et al.,2002; Li et al.,2002; Laclef et al.,2003a; Laclef et al.,2003b; Lagutin et al.,2003; Xu et al.,2003; Zheng et al.,2003; Ozaki et al.,2004).

Six2 function has not yet been characterized. Its expression is restricted to the head mesenchyme, foregut, stomach, kidney and genital tubercle (Oliver et al.,1995). Inactivation of Six1, its closest homolog, affects muscle, kidney, branchial arch derivatives and inner ear development(Laclef et al., 2003a; Laclef et al., 2003b; Xu et al., 2003; Zheng et al., 2003; Ozaki et al., 2004). The presence of incomplete and/or smaller cartilages and bones characterizes the craniofacial phenotype of Six1 null mice(Laclef et al., 2003b). Indeed, Six1 inactivation provided the first direct evidence implicating Six genes in the development of the facial skeleton; the strong craniofacial phenotype of the Eya1 mutant and the reported genetic interaction of Eya and Six genes give additional, indirect hints(Xu et al., 1999).

Here we show that Hoxa2 negatively regulates Six2expression during second arch development. Ectopic expression of Six2in the second branchial arch causes distinctive phenotypic features seen in the Hoxa2 mutant. Furthermore, gain-of-function experiments show that Hoxa2 is sufficient to downregulate Six2 expression in the neural-crest-derived mesenchyme. We also show that a 0.9 kb fragment of the Six2 promoter is the target of Hoxa2 regulation and that Hoxa2 physically interacts with the proximal region of this promoter.

Together, our results show that Six2 is genetically downstream of Hoxa2 in the second branchial arch and suggest that regulation of Six2 may be one of the mechanisms utilized by Hoxa2 to pattern the second arch skeleton.

Finally, we propose that the Six2 gene could be a target of Hox proteins in different developmental processes in addition to patterning the second arch.

In situ hybridization was performed as described(Kanzler et al., 1998), using Six2 (Oliver et al.,1995), Hoxa2 (Mallo,1997) and Sox9(Kanzler et al., 1998) probes. Embryonic day (E) 10.5 Msx2-Hoxa2 transgenic and control embryos were cut in half, and each half hybridized with a Hoxa2 or Six2probe to allow direct comparison. Skeletal preparations were previously described (Mallo and Brändlin,1997).

Second arches of E10.5 embryos from Hoxa2^+/-intercrosses and a2-Six2 transgenics were dissected, frozen and pooled according to genotype. RT-PCR was performed using Superscript Preamplification System (Invitrogen) and Taq polymerase (PeqLab) with the following primers: Six2F 5′-CAGCCGCCACCATGTCCATGCTG-3′; Six2R 5′-GAACTTGCGCCGCACGCGGTAC-3′; Six1F 5′-AAGAACCGGAG-GCAAAGAGACC-3′; Six1R 5′-CCAATATCTCCCCACTTAGGAA-CC-3′; Six4F 5′-AACCAGTATGGCATTGTCCAGATCC-3′; Six4R 5′-ACTGCAGAACCAAGCGCTGTTCTC-3′; Six5F 5′-GAGTGACT-GCGCTGCAACTTCCCTCG-3′; Six5R 5′-AGGGCTCCTCCACG-GGTACCGAC-3′; GadphF 5′-ACCACAGTCCATGCCATCAC-3′; GadphR 5′-TCCACCACCCTGTTGCTGTA-3′; Hoxa2F 5′-GCCT-GAGTATCCCTGGATG-3′; Hoxa2R 5′-ACCCTTCCCTCTCCAG-AAG-3′. First-strand cDNA was subjected to 24 amplification cycles. The specificity of each PCR product was confirmed by sequencing.

Hoxa2 mutant mice were described(Gendron-Maguire et al.,1993). Transgenic embryos were generated by pronuclear injection of the following transgenes: a2-Six2, containing Six2 cDNA,amplified with primers Six2F 5′-CAGCCGCCACCATGTCCATGCTG-3′ Six2R 5′-CTCTAGGAGCCCAGGTCCACAAGG-3′ cloned downstream the Hoxa2 enhancer (Kanzler et al.,1998); Msx2-Hoxa2(Kanzler et al., 1998); 900Six2-lacZ containing BamHI (-893) - Sph1 (+18) Six2 promoter fragment obtained by screening the RPCI mouse PAC library 21 (Osoegawa et al., 2000) (provided by UK HGMP Resource Centre),cloned into pCMVbeta (Clontech). 900Six2-lacZ was injected to generate both transgenic embryos and transgenic lines. A line showing high expression in the first branchial arch was crossed to Hoxa2^+/- mice to obtain 900Six2-lacZ; Hoxa2^+/-, which were mated to Hoxa2^+/-to generate 900Six2-lacZ;Hoxa2^-/- embryos.

Mouse Hoxa2 cDNA, containing a HA tag inserted in frame before the stop codon, mouse Meis1 cDNA and human Pbx1a cDNA(Di Rocco et al., 1997) were cloned in pcDNA3 (Invitrogen) and transcribed/translated using T7-coupled TNT rabbit reticulocytes (Promega).

BstEII/SspI probe, probe 1 and probe 2 were labeled withα ³²P-dCTP. The binding reaction was performed as described(Scheidereit, 1987).

The sequence of the oligonucleotides used as probes and competitors is shown in Fig. 6. For the supershift experiments, 40 ng of anti-HA antibodies (rat monoclonal 3F10,Roche) were added to the reaction.

The Hoxa2 transcription factor is essential for the development of the second branchial arch, but its mechanism of action is largely unknown. To identify target genes of Hoxa2, we analyzed the expression profiles of wild-type and Hoxa2 mutant second branchial arches(Bobola et al., 2003). Among the differentially expressed clones found in our screen, the mRNA corresponding to Six2 showed a threefold increase in the absence of Hoxa2.

Hoxa2 is expressed in neural crest cells migrating from rhombomere 4 to populate the second arch, later its main expression domain(Prince and Lumsden, 1994; Nonchev et al., 1996; Mallo, 1997). Six2expression in the branchial area is first detected at E9.5, predominantly in the first arch (Oliver et al.,1995). At this stage, no obvious differences in Six2expression pattern could be observed between wild-type and Hoxa2mutant second arches (not shown). At E10.5, Six2 expression was restricted to the posterior area of the second branchial arch. In addition, a second, more proximal domain of Six2 expression appeared in the mutant second arch (Fig. 1A,D,arrow). The spatial and temporal coordinates of Six2 expression in the mutant second arch, together with the signal intensity, were equivalent to those of the Six2 expression domain in the first arch mesenchyme. In Hoxa2 mutant embryos, the first and second arch Six2-expressing areas were symmetrical with respect to the first branchial cleft (Fig. 1A,D).

At E11.5, Six2 was mainly expressed in the first arch with a characteristic pattern; a small restricted area in the second arch was positive for Six2 (Fig. 1B, arrowhead). At this stage, the Six2 expression domain in the mutant second arch still appeared as a duplication of the first arch expression domain (Fig. 1E). In addition, a strong ectopic domain of Six2 expression was detected in the mutant mesenchyme proximal to the second branchial arch and delimited externally by the forming otic capsule(Fig. 1E, white arrow). One day later, the first and second branchial arch had completely fused. The first cleft was still evident and the pattern of Six2 expression around the cleft was essentially unchanged. However, an increase in Six2expression in the mutant could be observed in the area surrounding the otic capsule (Fig. 1C,F, white arrow).

In the absence of Hoxa2, a duplicated incus and malleus form in the second branchial arch. To see if Six2 upregulation in the second arch was spatially associated with the ectopic cartilaginous condensations, we examined mutant and wild-type embryos for the expression of the chondrogenic marker Sox9 (Ng et al.,1997; Zhao et al.,1997). In E12.5 wild-type embryos, Sox9 mRNA was mainly expressed in the area derived from the first branchial arch, consistent with the restricted chondrogenesis that takes place in the second branchial arch(Fig. 1G). In the Hoxa2 mutant, an equivalent Sox9 expression in the first and second arch anticipated the formation of incus and malleus in the first arch and their duplicated counterparts in the second arch(Fig. 1H). Six2 signal could be detected in close proximity to Sox9(Fig. 1C,D). In a more detailed inspection of the mutual localization of Six2 and Sox9mRNAs, performed by in-situ hybridization on adjacent sections of branchial arches of E11.5 Hoxa2 mutant embryos, a defined mesenchymal area,including the first and second branchial arch, was positive for both Six2 and Sox9 (Fig. 1I,L, arrows). We conclude that the upregulation of Six2takes place in the area fated to give rise to the skeletal duplication in the Hoxa2 mutant.

In addition to Six2, the genes Six1, Six4 and Six5 are also expressed in the branchial area(Oliver et al., 1995; Klesert et al., 2000; Ozaki et al., 2001). To distinguish if the upregulation of Six2 observed in the Hoxa2 mutant second arch is specifically restricted to Six2or if it is a common feature shared by the other Six genes, we performed semiquantitative RT-PCR on total RNA extracted from E10.5 Hoxa2mutant and wild-type arches, using Six1-, Six2-, Six4- and Six5-specific primers. As expected, we observed a significant increase in Six2 signal in the mutant versus wild-type second arch, but Six1, Six4 and Six5 levels (as well as Gadph control) remained unaffected(Fig. 2). These data indicate that Hoxa2 negatively controls Six2 expression in the second branchial arch specifically and does not affect any of the other Six genes expressed in this area.

In the absence of Hoxa2, Six2 is expressed in the proximal domain of the second arch, symmetrically to the Six2 expression domain in the first arch. Duplicated first arch skeletal elements arise in the very same area in the Hoxa2 mutant (Fig. 3A,B) (Gendron-Maguire et al.,1993; Rijli et al.,1993; Barrow and Capecchi,1999).

To test if these two events could be associated, namely if Six2upregulation might be one of the causative events in the generation of the Hoxa2 mutant phenotype, we overexpressed Six2 cDNA in the second branchial arch of wild-type embryos using a Hoxa2 enhancer(Nonchev et al., 1996). In-situ hybridization on E10.5 transgenic embryos (a2-Six2) showed that Six2 mRNA was additionally detected in the second and more posterior branchial arches. Strong ectopic Six2 expression was also observed in the somitic mesoderm (Fig. 3C,D).

We then analyzed the effects of Six2 overexpression on the skeletal phenotype of E18.5 a2-Six2 embryos. The second arch skeleton is composed of three cartilages (stapes, styloid process and lesser horn of the hyoid bone), which are transformed into different skeletal elements in the Hoxa2 mutant. All these elements displayed morphological changes in transgenic embryos overexpressing Six2(Fig. 3E,F,H,I).

The stapes, which is lost in the Hoxa2 mutant, was either reduced or absent in a2-Six2 transgenics (2/6). An ectopic cartilage, fused to the proximal part of the styloid process, extended into the oval window and in front of the incus. This cartilage, which in shape and position resembled a stapedial arch, may be an intermediate between loss of the stapes and formation of a duplicated incus, as observed in the Hoxa2 mutant(4/6) (Fig. 3E,F). The styloid process was thicker (Fig. 3E,F)or abnormally elongated and fused to the lesser horn of the hyoid bone,forming a long Meckel-like cartilage (2/6)(Fig. 3H,I). The lesser horn,absent in the Hoxa2 mutant, was misshapen, enlarged and fused to the greater horn (6/6); in most cases, it appeared as a bifurcation of the greater horn. By contrast to the noticeable effects in cartilages, bone formation was relatively unchanged and in only one case tympanic ring growth was mildly affected (Fig. 3H,I).

Molecular analysis of a2-Six2 transgenic embryos showed no noticeable effect of overexpression of Six2 on the levels of Hoxa2 in the second branchial arch(Fig. 3J), thus ruling out Hoxa2 downregulation as a possible cause of the transgenic phenotype.

Compared with the Hoxa2 phenotype, we observed two main defects:one was as featured in the Hoxa2 mutant (absent stapes, incus duplication), while the other consisted of enlargements of the elements patterned by Hoxa2 (lesser horn, styloid process abnormally elongated). In all cases, overexpression of Six2 in the second branchial arch resulted in the formation of ectopic cartilage, one of the phenotypic characteristics of the Hoxa2 mutant(Kanzler et al., 1998).

Six2 is widely expressed in the head mesenchyme(Oliver et al., 1995). By contrast, Hoxa2 expression is mainly confined to the second branchial arch (Prince and Lumsden,1994; Nonchev et al.,1996; Mallo,1997). To gain insight into how Hoxa2 regulates Six2, we asked whether Hoxa2 repressor activity is restricted to the second arch, or, alternatively, if Hoxa2 is sufficient to downregulate Six2 expression in the craniofacial mesenchyme. For this, we took a transgenic approach and ectopically expressed Hoxa2under the Msx2 promoter, able to direct gene expression to the head mesenchyme (Liu et al., 1994; Kanzler et al., 1998).

To detect expression of Six2 and Hoxa2 in the same embryo, E10.5 control and Msx2-Hoxa2 transgenic littermate embryo halves were hybridized with either Six2 or Hoxa2 probe. Six2 mRNA showed an abundant distribution in the maxillary and nasal mesenchyme of wild-type E10.5 embryos, areas that are negative for Hoxa2 expression (Fig. 4A). Transgenic E10.5 littermates (Msx2-Hoxa2; n=4)displayed a markedly reduced expression of Six2 in the maxillary mesenchyme and to a lesser extent in the periocular mesenchyme, the areas of Hoxa2 ectopic expression (Fig. 4B). We conclude that Hoxa2 is sufficient to repress Six2 in the head mesenchyme in vivo; this observation is particularly interesting because this very same transgenic expression results in reduction or absence of the facial skeleton (Kanzler et al., 1998).

As a transcription factor, Hoxa2 could block the accumulation of Six2 mRNA in the second arch by directly repressing its transcription. Alternatively, the absence of Six2 in the second branchial arch could be achieved indirectly, as the result of more upstream events that are in turn regulated by Hoxa2.

As a first step in understanding the role of Hoxa2 in regulating Six2, we examined mouse and human genomic Six2 sequences. The sequence from -900 bp to the putative transcriptional start site(identified as the 5′ end of the Six2 first exon, GenBank NM_011380) displayed a high conservation between the two species.

To assay whether this promoter region harbors regulatory elements controlling Six2 transcription, we cloned it upstream of a lacZ reporter gene and injected the resulting construct(900Six2-lacZ) into fertilized mouse oocytes. Transgenic embryos,collected at E11.5, showed a β-galactosidase (β-gal) staining consistent with the Six2 expression pattern at various embryo locations, including the branchial area(Fig. 5A) (E.K. and N.B.,unpublished). Here, the most proximal 900 bp of the Six2 promoter directed lacZ expression in the proximal part of the first branchial arch (arrowhead) and in mesenchyme proximal to the first branchial arch(arrow), a pattern that faithfully recapitulates endogenous Six2expression. The activity of the transgene in the second branchial arch appeared stronger compared with the restricted Six2 endogenous expression in this area (Fig. 1B). Insertion of the transgene in multiple copies may have reduced the efficiency of Hoxa2 repressor activity; alternatively, other repressor-responsive elements controlling Six2 transcription independently from Hoxa2 may not be contained in our transgene.

The identified Six2 promoter region recapitulated the expression of the endogenous gene in the first arch; Hoxa2 blocked Six2expression in an equivalent domain in the second branchial arch. To test if this promoter region retains Hoxa2-dependent regulation, we introduced the 900Six2-lacZ transgene into the Hoxa2 mutant background. In the absence of Hoxa2, a higher expression of the transgene was observed in the second arch, where β-gal staining became equivalent to that observed in the first arch; the lacZ-expressing areas were arranged symmetrically with respect to the first cleft (Fig. 5B, arrowheads). As for endogenous Six2 signal, lacZ expression was also upregulated in the mesenchyme proximal to the branchial arches (black arrows). All together, lacZ expression reproduced the expression pattern of Six2 in Hoxa2 mutant embryos. In addition, higher activity of the transgene was observed in the caudal part of the second branchial arch, in an area in which Six2 is not differentially regulated (white arrow).

These results show that the proximal region of the Six2 promoter is sufficient to direct Six2 expression in the branchial area. More importantly, they demonstrate that Hoxa2 controls the Six2 gene at the transcriptional level and that this control, direct or indirect, is confined to a proximal 900 bp of genomic sequence.

Our data suggest the possibility that Hoxa2 may directly repress Six2 transcription, interacting with the proximal 900 bp of the Six2 promoter. As noted above, the first kilobase of Six2genomic sequences upstream of the transcription start site (TSS; +1) is highly similar between mouse and human. Sequence conservation is extremely high between position -249 and -11, reaching 95% similarity between the two species. To test promoter, we performed an electrophoretic mobility shift assay (EMSA) using the conserved Six2 region located immediately upstream of the TATA box (BstEII-SspI fragment; Fig. 6A) as a probe. Incubation of the probe with in-vitro translated HA-tagged Hoxa2 resulted in the formation of two retarded complexes. These bands represent the interaction of Hoxa2-HA with the probe, as they were supershifted by the addition of the anti-HA antibody. By contrast, incubation of the probe in the presence of unprogrammed reticulocytes did not result in any retarded complex, nor did addition of the antibody have any effect(Fig. 6B). When the probe was fragmented and each half-fragment (probe 1 and probe 2) incubated with Hoxa2-HA, we still observed formation of the characteristic doublet(Fig. 6C,D). A close inspection of the BstEII-SspI genomic area revealed the presence of two conserved 5′-GAATAAT-3′ motifs, one in each of the two fragmented probes. According to in vitro binding experiments, the Hox consensus sequence contains a TAAT core (Graba et al.,1997); to test whether Hoxa2 recognizes the GAATAAT sequence, we performed competition experiments using wild-type and mutant oligonucleotides. The complex formed in the presence of probe 1 (BstEII-SmaI, Fig. 6A) and Hoxa2-HA was competed at two different concentrations of a cold oligonucleotide reproducing the GAATAAT sequence and flanking nucleotides of probe 1. A similar effect was observed upon adding a molar excess of oligonucleotide wt2, which reproduces the GAATAAT motif and flanking nucleotides contained in probe 2(SmaI-SspI, Fig. 6A). By contrast, the addition of the same molar excess of m1 or m2 oligonucleotides, containing three or four nucleotide substitutions in the GAATAAT, left the complex unaffected (Fig. 6C). Incubation of Hoxa2-HA in the presence of probe 2 resulted in the formation of the same retarded complexes, and these were efficiently competed by wt1 or wt2 oligonucleotides, but not by mutant oligonucleotides(Fig. 6D). The ability of Hoxa2 to recognize the two sites contained in the Six2 promoter was further confirmed by using the wild-type and mutant oligonucleotides as probes(Fig. 6E). Hoxa2-HA formed slower migrating complexes when incubated with oligonucleotides wt1 and wt2,while no higher complex formation was observed when the mutant oligonucleotides were used as probes in the same assay.

These data show that Hoxa2 interacts with the proximal region of the Six2 promoter and that this interaction is sequence-specific. Both the identified sites are bound with similar affinity and contain a GAATAAT sequence. Moreover, disruption of their ATAAT core abolishes Hoxa2 binding.

The interaction of Hox proteins with their target promoters often requires co-factors (Mann and Affolter,1998), and we have indeed detected Pbx and Meisbinding to the Six2 promoter in close proximity to the Hoxa2 binding sites. Pbx1a alone bound the BstEII/SspI fragment with very low affinity, while no binding was detectable for Meis1 alone with the probe. When the two proteins were co-translated, an intense retarded band was seen. In contrast to a previous study of Hoxb1 auto-regulatory element(Ferretti et al., 2000), we could not detect the formation of a slower molecular complex by simultaneous incubation of the probe with Hoxa2, Pbx1a and Meis1(Fig. 6F).

Along with second arch-derived structures, transgenic mice overexpressing Six2 displayed abnormal growth and morphology of the thyroid and cricoid cartilages and of the hyoid bone, derived from the third and more posterior pharyngeal arches.

The hyoid bone was malformed, curved and fused to the thyroid cartilage,bilaterally or unilaterally (Fig. 7B,C, arrow). Fusion of the greater horn to the lesser horn and to the thyroid cartilage in a structure independent of the main body of the hyoid was often observed (Fig. 7B,C,arrowhead). The thyroid and cricoid cartilages were abnormally thickened, and larger areas of fusion were observed (Fig. 7B,C). Occasionally, the rings of the trachea were abnormally fused to each other and disorganized (Fig. 7C).

Hoxa3, Hoxb3 and Hoxd3, alone or in combination, control the formation of mesenchymal derivatives of the neural crest(Chisaka and Capecchi, 1991; Condie and Capecchi, 1993; Condie and Capecchi, 1994;Manley and Capecchi, 1997). Intriguingly, skeletal defects observed in mice carrying mutations in the Hox paralogous group 3, such as the fusion of the greater horn to the thyroid cartilage, fusion of the hyoid bone to the thyroid cartilage, and malformation and fusion of thyroid and cricoid cartilages(Condie and Capecchi, 1994)are reminiscent of the phenotype observed in a2-Six2 transgenic embryos. This suggests a possible genetic interaction of Six2 with the Hox genes of paralogous group 3. Indeed, paralogous group 3 mutants and a2-Six2 transgenic embryos also display common defects in the growth and differentiation of skeletal elements derived from somitic mesoderm(Condie and Capecchi, 1994;Manley and Capecchi, 1997). Fusion of the atlas to the exoccipital bone, a characteristic of the Hoxd3 mutant phenotype(Condie and Capecchi, 1994;Manley and Capecchi, 1997), was observed in transgenic embryos overexpressing Six2 (Fig. 7D,E). Similarly to a Hoxa3; Hoxb3 double mutant (Manley and Capecchi,1997), the supraoccipital bone displayed only very partial ossification(Fig. 7F,G). These defects are consistent with the high expression of the transgene detected in the anterior somitic mesoderm (Fig. 3D).

Here we have shown that Hoxa2 regulates the expression of Six2 in the second branchial arch. Two main observations suggest that Hoxa2 repression of Six2 has a functional effect: (1) gain of function of Six2 in the second arch results in ectopic cartilage formation with characteristics similar to the Hoxa2 phenotype; (2)Hoxa2 controls the activity of a Six2 promoter fragment in vivo and binds within this promoter in vitro.

In patterning the embryo, Hox gene activity has been proposed to be transduced by a battery of genes, termed the realizator genes, that directly influence cell processes such as cell adhesion, apoptosis or rate of cell division (Garcia-Bellido,1975). Like Hoxa2, Six2 is a transcription factor. The next crucial question will therefore be to identify the cellular processes controlled by Six2 in the developing branchial arches.

The Hoxa2 mutation affects skeletal development of the second branchial arch (Gendron-Maguire et al.,1993; Rijli et al.,1993; Barrow and Capecchi,1999). Second arch skeletal elements are lost and replaced by first arch duplicated elements arranged in a mirror image disposition with respect to their first arch counterparts. How does Hoxa2 pattern second arch skeleton? Previous reports have shown that Hoxa2negatively interferes with the development of the facial skeleton, and in the second arch Hoxa2 surrounds, but is excluded from, the endochondral ossification centers required to form the hyoid cartilage(Kanzler et al., 1998; Creuzet et al., 2002). The most likely mechanism of action is that Hoxa2 restricts skeletogenesis in the second arch, preventing the formation of first arch duplications, while, at the same time, shaping second arch-specific elements. Six2 could be one of the genes regulated by Hoxa2 to restrict skeletogenesis in the second arch. In support of this, Six2 overexpression in the second branchial arch interferes with the normal patterning of the second arch-derived skeleton; endochondral ossification is increased overall,producing a skeletal phenotype reminiscent of the Hoxa2 phenotype. Discrepancies in the shape of the skeletal elements between transgenic and mutant embryos indicate that other factors, acting in parallel or in concert with Six2, are required to generate the full Hoxa2 phenotype.

However, even in the scenario with Six2 as the only factor responsible for the Hoxa2 phenotype, the few intrinsic differences that characterize the mutant and transgenic second arches could alone account for discrepancies in the final shape of the skeletal elements. First, being driven by a heterologous promoter, the expression pattern of Six2 in the second arch of transgenic embryos is different from Six2 endogenous expression in the second arch of the mutant. Second, in contrast to the mutant, a functional Hoxa2 protein is present in the second arch of transgenic embryos.

Final evidence that repression of Six2 is one of the mechanisms employed by Hoxa2 in second arch patterning will require analysis of a Hoxa2; Six2 double mutant. Our prediction is that Six2 inactivation should, at least partially, rescue the Hoxa2 phenotype. The lack of a Six2 mutant hampers the accomplishment of this experiment, but the inactivation of Six1, the closest homolog to Six2, supports our prediction. Six1mutant mice display an evident craniofacial phenotype, thereby identifying Six genes as important regulators of neural-crest-derived craniofacial skeleton(Laclef et al., 2003b; Ozaki et al., 2004). We found,both by RT-PCR (see Fig. 2) and by in-situ hybridization (B.E. and N.B., unpublished), that Six1 is not regulated by Hoxa2. However, Six1 is required for development of part of the skeleton that is affected by the Hoxa2mutation (Laclef et al.,2003b; Ozaki et al.,2004), suggesting, together with the high similarity in the encoded proteins and the expression pattern of the Six1 and Six2 genes (Oliver et al.,1995), a functional contribution of Six2 to the Hoxa2 phenotype.

In the Six1 mutant, second-arch-derived cartilages fail to form(Laclef et al., 2003b; Ozaki et al., 2004). If Six2 is also required for second arch skeletal growth, a likely scenario for second arch skeletal patterning is that Hoxa2 `tunes'the size of the skeletal elements to be produced in the second arch by regulating the domain of Six2 expression in this area.

Evidence accumulated in recent years indicates that Hox genes inhibit development of the facial skeleton in the areas where they are expressed. Gain of function of Hoxa2, both in pre-migratory neural crest (Couly,2002) and in the facial mesenchyme (Kanzler, 1998), prevents the formation of the facial skeleton. In the second branchial arch, its normal domain of expression, Hoxa2 negatively regulates skeletal development(Kanzler et al., 1998). If the broad effects of Hoxa2 on skeletal development are attributable to a general mechanism, we should expect the same molecular mediators acting downstream of Hoxa2 in the facial mesenchyme and in the second branchial arch. An important requisite for such mediators would be a broad expression in the area fated to form head cartilages and bones. We would also expect that gain of function of Hoxa2 would affect the spatial distribution of such a mediator. Indeed, Six2 displays a widespread expression in the craniofacial mesenchyme (Oliver et al.,1995) and its expression is downregulated following ectopic expression of Hoxa2. However, while Hoxa2 indistinctly inhibits bone and cartilage formation, overexpression of Six2 in the second branchial arch produces ectopic cartilages but does not affect intramembranous bone growth (the tympanic ring is fairly normal). Does Six2 specifically promote cartilage formation? The craniofacial defects of the Six1 mutant are not restricted to cartilage, but affect bones as well (Laclef et al.,2003b; Ozaki et al.,2004). In addition, the domain of Six2 upregulation around the otic vesicle is spatially associated with the squamous bone duplication observed in the Hoxa2 mutant. Finally, the analysis of Six mutants indicate that Six genes positively regulate cell proliferation,which explains how these genes control processes as diverse as muscle formation, retina development and skeletal development. On this basis, a likely prediction would be that Six2 promotes both cartilage and bone development. The lack of effect on bone development in our gain-of-function experiment could be explained by the absence, in the wild-type second arch, of a factor acting in concert with Six2. Alternatively, as the process of intramembranous ossification begins later than chondrogenesis, Six2overexpression might occur too early to affect bone formation. We favor the second hypothesis, because a2-Six2 transgenics show a high level of Six2 mRNA at E10.5; 1 day later, there are barely detectable differences in Six2 mRNA levels between transgenics and wild-type embryos (data not shown).

Despite the vast literature on Hox genes, the nature of the genes regulated by Hox proteins in vertebrates is still largely unknown. Solving this riddle is fundamental, if we want to explain how Hox genes control development in the vertebrate embryo.

We have shown that Hoxa2 controls the Six2 gene at the transcriptional level, as indicated by the identification of a Six2promoter fragment regulated by Hoxa2. As a transcription factor, Hoxa2 could directly regulate the Six2 promoter, and the observed binding of Hoxa2 to this promoter fragment in vitro strongly suggests that Six2is a direct target of Hoxa2.

Within the Six2 promoter, Hoxa2 recognizes two GAATAAT motifs near the transcription start site. The consensus Hoxa2 binding motif has not been previously described, but the sequence recognized by Hoxa2 on the Six2 promoter meets the requirements for Hox proteins binding to DNA(Graba et al., 1997). The interaction of Hox proteins with their target promoters often requires co-factors such as Pbx (Mann and Affolter,1998), and we have indeed detected Pbx and Meis binding to the Six2 promoter, in close proximity to the Hoxa2 binding sites. Another hint,albeit indirect, that Six2 might be a direct target of Hoxa2 is that the absence of Hoxa2 in the second arch specifically affects Six2expression, leaving the levels of Six1, Six4 and Six5unaffected.

Additional analyses will be required to definitely prove that Six2is a direct target of Hoxa2. However, the experimental analysis of Six2 regulation by Hoxa2 is complicated by the fact that Hoxa2 behaves as a repressor. Acting as a repressor, Hoxa2 might inhibit the basal transcription machinery, counteract the activity of a positively acting transcription factor or, alternatively, interact with proteins that remodel chromatin (Gaston and Jayaraman,2003). The elucidation of most of these mechanisms will be greatly facilitated by the identification of the proteins acting as activators of Six2. Currently, efforts in this direction are proceeding in our laboratory.

Regulation of Six2 by Hox genes was described in the Hoxa11;Hoxc11; Hoxd11 mutant, characterized by loss of metanephric kidney induction (Wellik et al.,2002). As in the Eya1 mutant(Xu et al., 1999), which has a remarkably similar phenotype, in the Hox11 triple mutants Six2 expression disappears(Wellik et al., 2002). This effect is opposite to the one observed in the second arch, where Hoxa2 represses Six2 expression.

Overexpression of Six2 under the Hoxa2 enhancer resulted in profound effects on the development of the skeleton patterned by Hox genes of paralogous group 3; some of the defects observed in Hox paralogous 3 single and compound mutants were mimicked by Six2 gain of function(consistent with ectopic expression of the transgene in the branchial arches and somitic mesoderm).

All together, these observations raise the possibility that Six2expression might be under the control of other Hox genes in the development of the vertebrate embryo. In Drosophila, common targets for different Hox proteins have been described, as well as the ability of Hox proteins to behave both as activators and repressors on their target genes (reviewed in Graba et al., 1997). To learn whether this holds true in vertebrates and if Six2 is a common target of different Hox proteins will require a profile of Six2 expression patterns in different Hox mutants, the entire spectrum of which is currently available.

The authors thank Alexander Aulehla for many motivating discussions; Davor Solter for his generous and constant support; Heiner Schrewe for his generous help during this project; Randy Cassada and Takashi Hiiragi for critical reading of the manuscript; and Elsa Huber, Laurent Morawiec and Sabine Ohnemus for excellent technical assistance. We are grateful to Peter Gruss for the Six2 probe and to Fulvio Mavilio for the Pbx1a cDNA. M.M. was supported by FCT/FEDER grant POCTI/MGI/43466/2001.