Hoxa-2 restricts the chondrogenic domain and inhibits bone formation during development of the branchial area

Cranial neural crest cells play a pivotal role in the development of the craniofacial area (Noden, 1988; Le Douarin et al., 1993). These cells delaminate from the dorsal part of the developing neural tube and migrate into the branchial arches and frontonasal mass where they contribute to a wide variety of tissues, including bone and cartilage (Noden, 1988; Le Douarin et al., 1993). Cell tracing studies in mouse and chick embryos have shown that neural crest cells originating from different rostrocaudal levels migrate into specific areas of the developing face (Lumsden et al., 1991; Serbedzija et al., 1992; Osumi- Yamashita et al., 1994). For instance, the first branchial arch is populated by neural crest migrating from the caudal midbrain and the first two rhombomeres, while rhombomere 4 neural crest cells populate the second branchial arch. After arrival of these neural crest cells in the branchial arches, a series of epithelial-mesenchymal interactions guide their further development into a variety of structures specific for the various regions (Carlson, 1994). The core and symphysial portions of Meckel’s cartilage, as well as the dentary bone, derive from the distal part of the first branchial arch. The proximal part of the first arch (Table 1) gives rise to the squamous bone, the dorsal end of Meckel’s cartilage, two of the middle ear ossicles (malleus and incus) and the tympanic ring, which supports the tympanic membrane. From the second branchial arch, the third of the middle ear ossicles, the stapes, develops together with the styloid process and the lesser horn of the hyoid bone (Table 1).

Two alternative mechanisms underlie the development of craniofacial skeletal elements. Endochondral elements develop through the formation of cartilage templates, later replaced by bone tissue (endochondral ossification), whereas dermal (or intramembranous) elements result from the osteoblastic differentiation of cells located within the condensing mesenchyme which lay down osteoid tissue (intramembranous ossification). Skeletogenesis initiates with the induction of control genes that specify an osteogenic or chondrogenic fate in previously undifferentiated mesenchymal cells. These genes include Sox9 for cartilage (Bell et al., 1997; Ng et al., 1997) and Cbfa1 for dermal bone (Ducy et al., 1997; Komori et al., 1997; Otto et al., 1997). Epithelial-mesenchymal interactions are thought to play an important role in skeletal induction in the craniofacial area (Hall and Miyake, 1995), acting either by direct contact or through the release of diffusible factors. Subsequently, the morphogenesis of the areas fated to the skeletal pathways must be modulated to achieve the formation of functional structures. In recent years, several genes have been identified that play a role in the morphogenesis of the craniofacial skeleton. Some of them, such as Endothelin-1, Msx1, gsc and Mhox, may be involved in skeletal induction processes (Kurihara et al., 1994; Satokata and Maas, 1994; Rivera-Pérez et al., 1995; Yamada et al., 1995; Martin et al., 1995); others, including Dlx1, Dlx2, Hoxa-2, as well as Mhox, may play a role in the morphogenesis of the induced skeleton (Gendron-Maguire et al., 1993; Rijli et al., 1993; Martin et al., 1995; Qiu et al., 1995, 1997).

Gene inactivation experiments have shown that Hoxa-2 plays a central role in the development of the skeletal elements of the second branchial arch (Gendron-Maguire et al., 1993; Rijli et al., 1993). In Hoxa-2 null mutant embryos, the normal second branchial arch derivatives are replaced by a duplicated set of proximal first branchial arch elements (Table 1). Thus, Hoxa-2 appears to divert the morphogenesis of the skeletal elements in the second arch from a first arch ground program. Any explanation of this process must account for basic differences in the morphogenetic plans of the first and second branchial arches. First, both endochondral and intramembranous ossification occurs in the first branchial arch, while only endochondral ossification takes place in the second branchial arch. Second, all the second arch derivatives arise from a single cartilaginous condensation, Reichert’s cartilage, whereas proximal derivatives of the first branchial arch arise from several independent skeletogenic centers: endochondral condensations lead to the formation of the middle ear ossicles (malleus and incus), while the squamous bone and tympanic ring develop from their respective intramembranous condensations. Multiple centers also occur in the second branchial arch of Hoxa-2 mutant embryos in which the tympanic ring (and probably also the squamous bone) develops from a condensation independently of the malleus and incus (Mallo and Gridley, 1996). Therefore, Hoxa-2 normally reduces the condensation centers and permits only endochondral ossification to occur in the second arch.

So far it is not known how Hoxa-2 directs the correct morphogenesis in the second branchial arch. Considering the involvement of other Hox genes in establishing segmental identity in a variety of systems (Krumlauf, 1994) and the fact that Hoxa-2 is expressed in the neural crest migrating to the second but not to the first branchial arch (Krumlauf, 1993; Prince and Lumsden, 1994; Mallo, 1997), it has been proposed that Hoxa-2 specifies, in the second arch neural crest, a developmental plan responsible for the formation of second arch-specific structures (Gendron-Maguire et al., 1993; Rijli et al., 1993). A prediction from this hypothesis is that the expression of Hoxa-2 in the first arch should result in the development of second arch structures replacing those of the first arch (Rijli et al., 1993). However, previous results from our laboratory indicate that the retinoic acid-induced ectopic expression of Hoxa-2 in the first branchial arch results in variable deletions of first arch proximal derivatives but no sign of change in segmental identity was evident (Mallo and Br ändlin, 1997). This suggests that Hoxa-2 acts as more than a positive selector gene in second arch morphogenesis.

We investigated the mechanisms by which Hoxa-2 might influence the morphogenesis of second arch elements. We find that Hoxa-2 is widely expressed in the second arch mesenchyme, but is excluded from the chondrogenic condensations. Furthermore, expression of Sox9 in Hoxa-2^−/− embryos is shifted ventrally into the normal Hoxa-2 expression domain where it is normally not expressed. Using a transgenic approach, we show that this change in Sox9 expression is indeed responsible for the phenotype of the cartilaginous elements in the second arch of the Hoxa-2 mutants. In addition, Hoxa-2 inhibits dermal bone formation in the second arch, and this inhibition is mediated by the prevention of Cbfa1 expression in that area.

Hoxa-2::SOX9 construct: a 0.8 kb BglII fragment of the Hoxa-2 genomic region containing the enhancer driving Hoxa-2 expression in the second branchial arch, in which the Krox-20 binding sites were inactivated (Nonchev et al., 1996), was cloned upstream of the minimal promoter from the adenovirus 2 late region (Conaway and Conaway, 1988). The human SOX9 cDNA (Wagner et al., 1994) and the polyadenylation signal of SV40 were then introduced downstream of the minimal promoter.

Msx2::Hoxa-2 construct: the murine Hoxa-2 cDNA was cloned downstream of a 5.1 kb genomic fragment containing the Msx2 enhancer/promoter (Liu et al., 1994). The SV40 polyadenylation signal was inserted downstream of the Hoxa-2 cDNA.

The constructs were released from vector sequences, gel-purified using the Qiaquick gel extraction kit (Qiagen) and injected at a concentration of 2 ng/μl according to standard protocols (Hogan et al., 1994).

For the identification and analysis of the transgenes, DNA was obtained from the viscerae of mouse fetuses according to the method of Laird et al. (1991). DNA (5 μg) was digested with BamHI and the fragments were resolved in 0.8% agarose gels, blotted to ZetaProbe GT membranes (BioRad) and hybridized with the appropriate ³²P- labeled probes according to standard methods (Sambrook et al., 1989). Copy number was estimated by comparing the intensities of the transgenic and endogenous bands.

The genotyping of embryos and fetuses from Hoxa-2^{+/ −}intercrosses was performed by PCR as described by Gendron-Maguire et al. (1993).

Whole-mount in situ hybridization was performed according to the method of Wilkinson (1992), radioactive in situ hybridization as described by Smith and Gridley (1992), and non-radioactive in situ hybridization on paraffin sections as by Jostarndt et al. (1994) using digoxigenin-labelled probes. The following probes were used: Cbfa1: a 0.6 kb HindIII/PstI fragment of the cDNA (Ogawa et al., 1993), obtained by RT-PCR; mouse Sox9: the same as described by Wright et al. (1995), obtained by RT-PCR; human SOX9: a cDNA fragment containing the whole coding region (Wagner et al., 1994); neo: a 0.6 kb PstI/XbaI fragment from pPGKneobpA (Soriano et al., 1991); Hoxa-2: the same as described by Mallo (1997); Mhox: a 0.6 kb fragment containing the entire coding region (Cserjesi et al., 1992), obtained by RT-PCR.

Skeletal preparations were performed as described by Mallo and Brändlin (1997) on 18.5 days post coitum (d.p.c.) embryos.

As a step to understanding the role of Hoxa-2 in the morphogenesis of the skeletal elements derived from the second branchial arch we analyzed its expression pattern, particularly in relation to the structures affected in Hoxa-2^−/− mice. Although previous analyses indicated that Hoxa-2 is expressed in the second branchial arch mesenchyme (Krumlauf, 1993; Prince and Lumsden, 1994; Mallo, 1997), those studies were performed at early developmental stages, and no detailed analysis of Hoxa-2 expression during morphogenesis of the second branchial arch structures was conducted. Analysis at later developmental stages (10.5-14.5 d.p.c.) revealed that Hoxa- 2 is widely expressed throughout the second branchial arch mesenchyme (Fig. 1 and data not shown). However, the condensations forming the second arch skeletal elements, although at the earliest stages they might contain some Hoxa-2-expressing cells, were mostly negative for Hoxa-2 expression (Fig. 1 and data not shown). For instance, the developing stapes can be initially identified in the proximal region of the second arch as a condensation surrounding the stapedial artery (Anson and Cauldwell, 1942); this condensation was negative for Hoxa-2 expression (Fig. 1A,C,G,J). Conversely, Sox9, an early marker of chondrogenesis (Wright et al., 1995), was expressed in this condensation (Fig. 1E,H,K), confirming its cartilage-forming nature. Later, when the stapes interacts with the otic capsule to become inserted in the oval window, Hoxa-2 is still expressed around it and extends its expression through the prospective wall of the otic capsule in the region of the oval window (Fig. 1I,J). The expression of Sox9 in this area was again complementary to that of Hoxa-2: Sox9 was detected in the rest of the otic capsule, but not in the oval window (Fig. 1K). The apparent complementarity of Hoxa- 2 and Sox9 was also clear for other second arch structures. The developing styloid process contained Sox9 transcripts, in agreement with its endochondral nature (Fig. 1N), but was negative for Hoxa-2 expression (Fig. 1M).

The absence of detectable Hoxa-2 expression in the skeletal condensations makes it unlikely that this gene directs the morphogenesis of the second arch skeleton in a cell- autonomous fashion. Instead, the complementarity of Hoxa-2 and Sox9 expression domains (Fig. 1) and the finding that Hoxa-2 is expressed in the oval window (Fig. 1J), a structure absent in Hoxa-2^−/− animals (Gendron-Maguire et al., 1993; Rijli et al., 1993), suggest that Hoxa-2 may inhibit skeletal formation in this area.

Based on this hypothesis, we speculated that the Hoxa-2 mutant phenotype might derive from the development of skeletal elements within the Hoxa-2 expression domain. To test this prediction, we took advantage of the finding that the expression of neo from the Hoxa-2 null allele (Gendron-Maguire et al., 1993) resembled that of the native Hoxa-2 gene (Fig. 1D; also reported by Rijli et al., 1993), indicating a dominant effect of the Hoxa-2 second arch enhancer (Nonchev et al., 1996) on the PGK promoter. In Hoxa-2 mutant embryos, neo expression was observed in the condensations of the second branchial arch that generate the duplicated set of first arch elements (Fig. 2). By contrast, the corresponding elements developing from the first arch, which are physically connected to those of the second arch (Gendron-Maguire et al., 1993; Rijli et al., 1993), were negative for neo expression (Fig. 2), emphasizing the specificity of the neo expression pattern. These results support the notion that the set of second arch elements in the Hoxa-2 mutants arise from mesenchymal cells of the Hoxa-2 expression domain.

A variety of molecular and mutational studies have suggested the involvement of Sox9 in the initial steps of cartilage development (Foster et al., 1994; Wagner et al., 1994; Wright et al., 1995; Bell et al., 1997; Ng et al., 1997). It is therefore possible that Hoxa-2 exerts its negative effects on cartilage formation by modulating Sox9 expression. In the second branchial arch, Sox9 expression becomes detectable at 10.5 d.p.c. in the condensation of Reichert’s cartilage (Wright et al., 1995; and Fig. 3A). Its expression domain is restricted to a central area in the branchial arch and shows in its dorsal part the typical perforated appearance of the stapedial condensation, which lies in close proximity to the otic vesicle (Fig. 3A,C,D). Analysis of Sox9 expression in the second arch of Hoxa-2 mutant embryos revealed a different pattern, already evident at 10.5 d.p.c. (Fig. 3E). Careful examination of sections and whole-mount preparations indicated that the Sox9 expression domain is located more ventrally in the arch in Hoxa-2^−/− embryos (Fig. 3; and data not shown) within the Hoxa-2 expression region (compare Fig. 3G,H and Fig. 1A,C). The altered Sox9 expression in Hoxa-2^−/− embryos was particularly evident in the dorsal part of the Sox9 expression domain, where it was no longer close to the otic vesicle but more ventrally located, adjacent to the first pharyngeal pouch (Fig. 3G,H), and its shape was different from that of the normal stapedial condensation (Fig. 3E). These results indicate that the area where Sox9 is induced in the second branchial arch is modulated by Hoxa-2.

If the role of Hoxa-2 in the morphogenesis of the cartilaginous components of the second arch is to determine the chondrogenic domain by restricting the area of Sox9 induction, it is possible that forced expression of Sox9 in the rostral Hoxa- 2 domain might generate a phenotype resembling that of the Hoxa-2^−/− embryos. To test this possibility, we expressed SOX9 under the control of the Hoxa-2 second arch enhancer (Nonchev et al., 1996) in transgenic animals. We used a mutated version of this enhancer in which the Krox-20 binding sites are inactivated (Nonchev et al., 1996) to avoid complications from ectopic SOX9 expression in the hindbrain.

The effects of the transgene on the craniofacial region varied with the integration copy number. Fetuses containing low copy numbers (up to 10) of the transgene were indistinguishable from non-transgenic controls (not shown). However, phenotypic alterations were observed in fetuses containing higher copy numbers. Those alterations were restricted to the region derived from the second arch, with the rest of the craniofacial area showing a completely normal appearance (Fig. 4). The normal second branchial arch elements were absent and were substituted by a cartilaginous mass located between the otic capsule and the first arch skeletal elements on the rostral side, extending caudally into the neck region (Fig. 4E,F). The shape of these cartilaginous formations varied from one fetus to another without any obvious pattern. Interestingly, the phenotype of these transgenic fetuses also included extra elements associated with the first arch cartilages with patterns reminiscent of the ear region of Hoxa-2 mutants (Fig. 4D,F). A cartilage resembling an incus in mirror image orientation with respect to its normal counterpart was present, associated with a distal extension of the malleus and with an extended cartilage located in a position similar to that of the duplicated Meckel’s cartilage of the Hoxa-2 mutant embryos (Fig. 4D,F). Therefore, it seems that expression of Sox9 within the Hoxa-2 domain is able to partially reproduce the Hoxa-2⁻⁻ ear phenotype. Analysis of SOX9 expression in Hoxa-2::SOX9 transgenic embryos indicated that its expression in the second arch of these embryos is not restricted to the rostral Hoxa-2 domain, as Sox9 is in the Hoxa-2^−/− embryos, but encompasses the whole Hoxa-2 expression domain in this arch (Fig. 5; compare with Fig. 3E). Therefore, although it cannot be formally ruled out that SOX9 misexpression induced the Hoxa- 2::SOX9 transgenic phenotype via interference with general patterning mechanisms, it is very likely that the cartilages in the second arch of the Hoxa-2::SOX9 fetuses resulted from a dominant chondrogenic activity of SOX9 in different areas of this arch. In the most rostral areas, similar to the case in Hoxa- 2 null mutants, it would induce the observed Hoxa-2^−/−-like elements, while in more caudal and proximal regions of the second arch it would generate the extra mass of cartilage.

Together, the results presented so far indicate that the Hoxa- 2-dependent alteration of Sox9 expression in the second branchial arch plays an important role in the genesis of the Hoxa-2 mutant phenotype. It can be concluded that Hoxa-2 influences the morphogenesis of the second arch cartilages specifying the Sox9 functional domain by an inhibitory mechanism.

In addition to the respecification of the cartilaginous skeleton, the Hoxa-2 mutant phenotype includes the duplication of first arch bony elements (Gendron-Maguire et al., 1993; Rijli et al., 1993). In Hoxa-2::SOX9 transgenic fetuses, these dermal elements were not affected (Fig. 4). Thus, the duplicated tympanic ring and squamous bone of the Hoxa-2^−/− embryos probably do not result from a change in Sox9 expression. We previously showed that the duplication of the tympanic ring in the Hoxa-2 mutants results from the induction of an additional ossification center in the region of the second branchial arch (Mallo and Gridley, 1996). Because Hoxa-2 appears to inhibit cartilage formation, we asked whether it has similar effects on the formation of intramembranous bone. We first expressed Hoxa-2 under the control of the Msx2 gene promoter. This gene is expressed in the dermal bone precursors of the craniofacial area (MacKenzie et al., 1992) and a promoter fragment has been identified that drives expression in most of the Msx2 expression domains (Liu et al., 1994). Skeletal preparations from Msx2::Hoxa-2 transgenic embryos revealed a variable deficit of bone formation in the region of the skull (Fig. 6). This varied from small deficits in the interparietal bone and tympanic ring to extensive deletions of several bones in the craniofacial area. In an extreme case, most of the membranous bones in the craniofacial area were absent (Fig. 6C). Analysis of Hoxa-2 expression in transgenic embryos at earlier stages of development showed that this gene was indeed ectopically expressed in the areas where the affected bones develop (Fig. 7), indicating a direct role of Hoxa-2 in the transgenic phenotype. Furthermore, the extent to which bone formation was affected correlated with the transgene copy number, suggesting a direct relationship between the severity of the phenotype and Hoxa-2 expression levels. These results indicate that Hoxa-2 inhibits the formation of intramembranous bone in a dominant fashion. It should be noted that in addition to deficits in bone formation, chondrogenesis was also compromised in some of the Msx2::Hoxa-2 transgenic fetuses, in particular those presenting strong dermal bone phenotypes (Fig. 6C). Given the negative effects of Hoxa-2 on chondrogenesis, it is likely that the failure of some craniofacial cartilages to develop results from Hoxa-2 expression in their primordia. Some support for this idea is provided by the cartilage phenotype in the fetus showed on Fig 6C: cartilage derived from the first branchial arch, where the Msx2 enhancer is very active (Liu et al., 1994), are strongly affected, while the nasal capsule, which originates from a region of low Msx2 enhancer activity, is essentially conserved.

To determine whether the inhibitory effects of Hoxa-2 on dermal bone formation have physiological relevance, we analyzed neo expression during the development of the duplicated tympanic ring in the second arch of Hoxa-2^−/− embryos. In fact, neo is expressed in the condensation that initiates this structure (Fig. 2D), indicating that the additional tympanic ring of Hoxa-2^−/− embryos develops within the Hoxa- 2 expression domain. Thus, this gene might indeed inhibit the formation of this bone in wild-type animals.

We next wanted to examine the mechanisms by which Hoxa-2 inhibits the formation of dermal bone. We focused first on gsc and Mhox because they are required for the formation of the tympanic ring (Rivera-Pérez et al., 1995; Yamada et al., 1995; Martin et al., 1995) and thus, Hoxa-2 could inhibit the formation of a duplicated tympanic ring by modulating any of these genes. However, we showed previously that Hoxa-2 does not affect the expression of gsc (Mallo and Gridley, 1996), and the expression of Mhox, which includes domains in the first and second branchial arches, was also not affected by Hoxa-2 (Fig. 8B,C).

We then asked whether Hoxa-2, which had a clear effect on the expression of one of the earliest regulators of cartilage formation, Sox9 (Graves, 1997), might also affect the expression of Cbfa1, a key regulator of membranous bone development (Ducy et al., 1997; Komori et al., 1997; Otto et al., 1997). At 11.5 d.p.c. Cbfa1 is expressed in several osteogenic domains of the wild-type embryo, including the developing mandible and maxilla (Fig. 8D); in the ear region, expression was detected rostral to the pit where the external acoustic meatus originates (Fig. 8D,E). Analysis of sections at 12.5 d.p.c. (Fig. 8F) confirmed that the Cbfa1-expressing region corresponds to the site where the tympanic ring is initiated (Mallo and Gridley, 1996; Mallo, 1998). Cbfa1 expression in the second arch of wild type embryos is very low, being detectable only upon over development of the whole-mount in situs (Fig. 8D,E; and not shown). In the Hoxa-2 mutant, an additional expression domain with an intensity similar to that in the first arch, was observed caudal to the prospective external acoustic meatus (Fig. 8G,H) in an area where Hoxa-2 is normally expressed (Fig. 8A). Analysis of sections at 12.5 d.p.c. (Fig. 8I) showed that this extra Cbfa1 domain corresponds to the region where the supernumerary tympanic ring develops in the Hoxa-2^−/− animals (Mallo and Gridley, 1996; Mallo, 1998). These results suggest that prevention of Cbfa1 expression mediates the inhibition of dermal bone formation by Hoxa-2.

Previous studies have shown that the Hoxa-2 gene is essential for proper development of the skeletal structures derived from the second branchial arch (Gendron-Maguire et al., 1993; Rijli et al., 1993). When inactivated, the second branchial arch cartilages fail to develop and are replaced by a duplicate set of proximal first arch elements. Therefore, it has been suggested that, like other Hox genes in a variety of developmental settings (Krumlauf, 1994), Hoxa-2 provides the second arch with regional identity (Gendron-Maguire et al., 1993; Rijli et al., 1993; Hall and Miyake, 1995). Our present results suggest that Hoxa-2 does not provide the second arch skeletal elements with an intrinsic and specific developmental plan since, except for a few cells at the earliest stages, this gene is not expressed in the condensations that initiate and form those elements. Theoretically, it is still possible that Hoxa-2 expression in those cells is sufficient to give a specific plan to the second arch condensations. However, our results rather suggest that Hoxa- 2 influences the morphogenesis of the second branchial arch elements by preventing dermal bone and cartilage formation in particular areas.

Cartilage is clearly induced in the second arch both in the presence and absence of Hoxa-2. However, our data indicate that the site of cartilage formation and the resulting set of structures differ depending on whether Hoxa-2 is expressed. In wild-type embryos, chondrogenesis occurs in a group of Hoxa- 2-negative cells in the core of the second branchial arch; this chondrogenic center generates the stapes and other second arch-derived cartilages. In Hoxa-2^−/− embryos, chondrogenesis is activated in a different subset of second arch mesenchymal cells, located more ventrally within the typical Hoxa-2 expression domain, resulting in a duplicated malleus and incus. Several findings support such an interpretation. neo (which is expressed from the mutant allele similar to Hoxa-2) is expressed in the chondrogenic condensations located in the second arch of Hoxa-2^−/− embryos. In addition, in Hoxa-2 null mutants the Sox9 expression domain in the second arch, in particular its dorsal part, is shifted into the Hoxa-2 ‘territory’. Finally, when SOX9 was expressed under the control of the Hoxa-2 promoter in transgenic animals, the ear region of the resulting fetuses contained cartilaginous elements resembling those of Hoxa-2 null mutants, indicating that upregulation of SOX9 in the Hoxa-2 domain can indeed mimic part of the Hoxa-2 mutant phenotype. Therefore, Hoxa-2 influences the formation of cartilaginous elements in the second branchial arch by restricting the chondrogenic areas.

In addition to cartilages (malleus and incus), the duplicated set of first arch skeletal elements in the second arch of Hoxa- 2^−/−embryos includes dermal bones (tympanic ring and squamous bone) (Gendron-Maguire et al., 1993; Rijli et al., 1993). Since no membranous ossification occurs in the second branchial arch of wild-type embryos, Hoxa-2 must somehow interfere with the formation of dermal bone occurring in the second arch of Hoxa-2 null mutants. This process is expected to involve mechanisms different from those affecting cartilage formation, because the tympanic rings and squamous bones were not affected in Hoxa-2::SOX9 transgenic fetuses containing Hoxa-2^−/−-like cartilage elements in the second arch region. This finding also makes it unlikely that Hoxa-2 normally respecifies bone condensations into the cartilage pathway. Instead, our experiments suggest that Hoxa-2 inhibits dermal bone formation in this area. The ability of Hoxa-2 to inhibit bone formation is evident from our transgenic experiments in which this gene was ectopically expressed in several osteogenic areas. Furthermore, the supernumerary tympanic ring arises in the second arch of Hoxa-2^−/− embryos in the neo (i.e. Hoxa-2) expression domain, indicating that this mechanism operates normally in the second branchial arch. Moreover, Cbfa1 is upregulated in the second branchial arch of Hoxa-2^−/− embryos. Thus, the absence of membranous bone in the second arch of normal embryos is the result of inhibition in areas where it would be made in the absence of Hoxa-2.

The Hoxa-2 mutant phenotype includes the deletion of the oval window, located in the otic capsule (Gendron-Maguire et al., 1993; Rijli et al., 1993). The absence of this structure is not simply the consequence of the failure of the stapes to develop, because oval windows can be found in the absence of this ossicle (Mallo, 1997). Our results provide a rationale for the formation of this structure. Because Hoxa-2 is expressed in the area where the oval window is formed, it might render the cells in that area unable to respond to the signals that induce the cartilage surrounding the inner ear. The stapedial condensation would then interact with this area and become inserted in the window.

Our experiments indicate that Hoxa-2 acts at very early stages of skeletogenesis since the expression of Sox9 and Cbfa1, which are so far the earliest markers for the chondrogenic and osteogenic pathways (Graves, 1997; Ducy et al., 1997), is affected by Hoxa-2 from the first moment these genes can be detected. However, we cannot be certain about the molecular mechanisms of the Hoxa-2-mediated inhibition of skeletal formation. One possibility is that Hoxa-2 represses directly Sox9 and Cbfa1 transcription. Alternatively, or in addition, Hoxa-2 could down regulate other genes that mediate the induction of Sox9 and Cbfa1 in the second branchial arch. Biochemical studies are necessary to distinguish between these possibilities.

Mutational analyses have shown that Hoxa-2 plays an important role in patterning the second branchial arch (Gendron-Maguire et al., 1993; Rijli et al., 1993), but exactly how this gene exerts its role has remained unclear. Grafting experiments in birds (Noden, 1983) have suggested that the cranial neural crest is largely prespecified before its migration. In this context it has been proposed that Hoxa-2 provides the neural crest cells with a differentiation program to form the second arch elements (Gendron-Maguire et al., 1993; Rijli et al., 1993). According to this view, wild-type and Hoxa-2^−/− embryos develop different skeletal structures in their second branchial arches because of the loss of this specific morphogenetic program in the mutants. However, our results do not support this view and suggest instead that Hoxa-2 acts later by preventing chondrogenesis and dermal bone formation in specific areas.

We previously hypothesized that Hoxa-2 modulates the response of the mesenchyme to skeletogenic signals provided locally by the branchial arches (Mallo and Br ändlin, 1997). In the absence of Hoxa-2, the first and second branchial arches would be endowed with similar developmental potential in terms of their ability to respond to local patterning cues. In this ‘basic state’, equivalent areas of the first and second branchial arches would respond to one or several common skeleton- inducing signals located probably between both branchial arches. Dermal bone and cartilage developing from those induced areas would form elements with first arch characteristics and opposite polarity (Gendron-Maguire et al., 1993; Rijli et al., 1993). The presence of Hoxa-2 modifies the competence of the second arch mesenchyme to respond to inducing signals. Cells responding to those signals in the absence of Hoxa-2 are rendered unresponsive by an inhibitory action of this gene. As a result, dermal bone formation is completely inhibited and the cartilage-forming area is spatially shifted into a Hoxa-2-deficient domain. As a consequence, typical second arch structures form.

This hypothesis is consistent with our data and with the Hoxa-2 null mutant phenotype. The area of the second branchial arch in which chondrogenesis is induced is different in wild-type and Hoxa-2 mutant embryos. In the mutant situation, cartilage is induced in a region within the Hoxa-2 expression area. When this gene is present, cartilage is formed more dorsally in the second arch in a Hoxa-2-deficient region. Thus, Hoxa-2 appears to restrict the domain of cartilage formation inhibiting the mesenchymal response to chondrogenic signals. Similarly, our results suggest that dermal bone is not formed in the second arch because Hoxa-2 blocks the mesenchymal response to the corresponding signals.

This interpretation also provides an explanation for the mirror image disposition of the duplicated ear elements in the Hoxa-2^−/− embryos. If these elements are the result of skeletal induction from a center located between the first two arches, signaling similarly to both arches, the induced elements are likely to develop with opposite polarity. The existence of such a signaling center, already suggested by Rijli et al. (1993), is supported by the expression of gsc in the first and second branchial arches (Gaunt et al., 1993), although it is only required in the first arch (Rivera-Pérez et al., 1995; Yamada et al., 1995).

An important step in further understanding the development of the branchial area will be the identification of the inducing factors. Considering that the cartilage-forming area in the Hoxa-2 mutants is located closer to the signaling center than that in normal embryos, it is possible that, at least for the cartilage, diffusible molecules are involved in the induction which would be prevented from further diffusion upon activation of responsive cells. Shh and members of the Wnt, BMP and FGF gene families are prime candidates since they have already been identified as signaling molecules in a variety of developmental settings (Echelard et al., 1993; Krauss et al., 1993; Riddle et al., 1993; Laufer et al., 1994; Yang and Niswander, 1995; Vainio et al., 1993; Neubüser et al., 1997). Furthermore, several of these genes are expressed in the branchial area (Parr et al., 1993; Heikinheimo et al., 1994; Wall and Hogan, 1995; Helms et al., 1997; Neubüser et al., 1997), where they have been shown to participate in morphogenetic processes (Vainio et al., 1993; Helms et al., 1997; Neubüser et al., 1997). Endothelin-1 is another candidate to play a role in the induction process: the inactivation of this gene, which encodes a secreted peptide (Yanagisawa et al., 1988) with a restricted expression between the branchial arches (Kurihara et al., 1994), results in craniofacial defects characterized by extensive deletions of the skeletal derivatives of the branchial arches, without obvious effects on the rest of the craniofacial area (Kurihara et al., 1994).

A wide variety of homeobox-containing genes have been implicated in the development of skeletal elements in different body areas. It is possible that some of them function through mechanisms similar to those described here for Hoxa-2. For instance, null mutations in MHox and Dlx2 produced alterations in the proximal region of the branchial arches (Martin et al., 1995; Qiu et al., 1995, 1997). In both cases, the phenotype includes the altered development of the incus, which remains attached to an extra cartilage in the lateral wall of the skull. It is thus possible that in particular areas, these genes also modulate the responsiveness of the mesenchyme to chondrogenic signals in achieving the proper morphogenesis of skeletal elements like the incus. In this respect, it should be noted that in the absence of Dlx1 and Dlx2, Sox9 was found to be upregulated in the molar-forming area of the maxilla, a region where Dlx1 and Dlx2 are normally expressed (Thomas et al., 1997). For other genes, however, the null phenotypes cannot be easily interpreted in terms of inhibitory mechanisms on chondrogenesis or osteogenesis. For instance, genetic data suggest that more positive mechanisms mediate the function of Msx1 or gsc (Satokata and Maas, 1994; Rivera-Pérez et al., 1995; Yamada et al., 1995), and differential proliferation rates provide a simpler explanation for the role of Hox genes in the morphogenesis of the axial skeleton (Condie and Capecchi, 1994). Therefore, homeobox-containing genes in general, and Hox genes in particular, might act in different skeletogenic areas according to different mechanisms.

Finally, modulation of skeletogenic competence might represent a general mechanism for the evolution of the craniofacial area. It has been suggested that the Hoxa-2 mutant phenotype includes atavistic elements (Rijli et al., 1993), and inactivation of MHox and Dlx2, which as discussed above might also down regulate skeletogenesis, produced phenotypes that have been interpreted as the region of the middle ear adopting a configuration reminiscent of the reptilian jaw joint (Martin et al., 1995; Qiu et al., 1995, 1997), the homolog of the middle ear in non-mammalian jawed vertebrates (Novacek, 1993). Therefore, these genes (and others) might be involved in the evolution of this region by modulating the areas competent to respond to skeletal inducing signals. The availability of mutant strains for those genes will facilitate the design of experiments to test these possibilities.

We are grateful to Gerd Scherer and Peter Südbeck for providing us the human SOX9 cDNA and to Tom Gridley for the genomic clone containing the Hoxa-2 enhancer. We also thank Tom Gridley, Bernhard Herrmann, Andreas Kispert, Heiner Schrewe and Davor Solter for critical review of the manuscript.