Molecular mechanisms of cranial neural crest cell migration and patterning in craniofacial development

The neural crest cells (NCCs) (see Glossary, Box 1) are a migratory cell population specific to vertebrates that originates from the dorsal part of the developing neural tube. Following induction, NCCs delaminate and migrate to different regions of the embryo, where they differentiate into a broad range of cell types, including peripheral and enteric neurons, glia, melanocytes and smooth muscle (Gammill and Bronner-Fraser, 2003; Le Douarin and Kalcheim, 1999; Morales et al., 2005; Steventon et al., 2005). Moreover, in the cranial region, NCCs contribute to most of the cartilage and bone of the skull, facial and pharyngeal skeletons. Rostral cranial NCCs extensively contribute to the frontonasal skeleton and the membranous bones of the skull, whereas more posterior cranial NCCs fill the pharyngeal arches (PAs) (see Glossary, Box 1), where they form the jaw, middle ear, hyoid and thyroid cartilages (Couly et al., 1993; Köntges and Lumsden, 1996; Noden, 1983) (for reviews, see Gross and Hanken, 2008; McBratney-Owen et al., 2008; Santagati and Rijli, 2003). Although the initial patterns of NCC segmentation and migration are mainly conserved between species, the wide diversity of vertebrate craniofacial morphologies indicates that cranial NCC subpopulations are able to generate distinctive, species-specific skeletal structures. A full understanding of the underlying molecular mechanisms involved is central to advancing our understanding of craniofacial biology and evolution.

Here, we bring a recent perspective on the complex interplay that occurs between the intrinsic NCC genetic program and the extrinsic environmental signals to which distinct NCC subpopulations are exposed during coordinated craniofacial morphogenesis. First, we focus on the molecular mechanisms that establish the segmental pattern of NCC organization when these cells colonize the craniofacial and pharyngeal regions. We then describe how cranial NCCs acquire and maintain their positional identity, and how environmental signals influence NCC transcription activity to achieve spatially and temporally coordinated craniofacial morphogenesis. Finally, we briefly discuss the modification of the shape and function of selected orofacial structures during vertebrate evolution in the context of recent findings that have identified the molecular pathways underlying such changes.

Cranial NCCs follow stereotypical migratory pathways that are conserved among vertebrate species. NCCs that originate from the diencephalon and anterior mesencephalon (see Glossary, Box 1) migrate into the frontonasal process (FNP) (see Glossary, Box 1), whereas at a more caudal level, NCCs from the posterior mesencephalon and hindbrain, which is transiently subdivided into neuroepithelial segments called rhombomeres (r), colonize the PAs, another series of metameric structures (Couly et al., 1996; Couly et al., 1993; Johnston, 1966; Köntges and Lumsden, 1996; Le Lievre, 1978; Le Lievre, 1974; Noden, 1983; Osumi-Yamashita et al., 1994) (Fig. 1A). The NCC subpopulations that target individual PAs migrate in stereotypical streams (Fig. 1A) (Birgbauer et al., 1995; Kulesa and Fraser, 2000; Lumsden et al., 1991; Sechrist et al., 1993; Serbedzija et al., 1992; Shigetani et al., 1995; Trainor et al., 2002b). Maintaining the spatial segregation of such streams has an important impact on craniofacial pattern. Indeed, NCCs from different rhombomeres remain spatially segregated through to late developmental stages, and contribute to skeletal elements of multi-rhombomeric origin forming cryptic intraskeletal interfaces (Köntges and Lumsden, 1996). Moreover, NCC-derived connective muscle tissues from a given axial level are anchored to skeletal domains derived from NCCs that originate from the same axial level (Köntges and Lumsden, 1996; Matsuoka et al., 2005). Thus, the early anteroposterior (AP) pattern of NCC generation and migration is fundamental to establishing an underlying segmental pattern upon which the pharyngeal region of the vertebrate head is built (Graham, 2008; Kuratani et al., 1997).

In most vertebrates, hindbrain NCC migratory streams are separated by crest-free regions lateral to r3 and r5 (Lumsden et al., 1991; Golding et al., 2002; Graham et al., 1993). Environmental signals force each stream into well-defined pathways. For example, the mesenchyme lateral to r3 and r5 inhibits NCC migration (Farlie et al., 1999; Trainor et al., 2002b), in part through the non-cell autonomous functions of the receptor tyrosine kinase Erbb4, which is expressed in r3 and r5 (Golding et al., 2000) (Table 1). Also involved in sculpting the NCC migratory streams are the repulsive interactions between the Eph tyrosine kinase receptors and their ephrin ligands, and between the transmembrane neuropilin (Nrp) receptors and their secreted semaphorin (Sema) ligands (see Table 1 for references; Fig. 1). For example, in zebrafish, NCCs targeting PA1-3 express nrp2a and nrp2b, and avoid cells expressing sema3f and sema3g located in the NCC-free zones lateral to r3 and r5 (Yu and Moens, 2005). Moreover, in addition to the mechanisms involved in sculpting the migratory streams, cranial NCCs need also to be provided with specific directional guidance to their final destination (see Box 2 and Table 1 for references). Twist, which is expressed in the pharyngeal mesenchyme, and Tbx1 (T-box 1), which is expressed in the mesodermal core of the PAs and in the endodermal pouches, are additionally required for the proper segregation of PA1 and PA2 NCC streams (Moraes et al., 2005; Soo et al., 2002; Vitelli et al., 2002).

In summary, to allow their appropriate positioning in the PA and, therefore, the proper assembly of skeletal structures, NCC subpopulations are guided by complex sets of cues to which they respond locally during their journey. Such positional cues are translated into stereotypical directional migratory behaviours, depending on the positional identity of NCC subpopulations. This raises questions about how cranial NCCs acquire their positional identity and how their response to several simultaneous extrinsic inputs may be integrated at the transcriptional level.

The positional identity of NCC subpopulations is established by combinatorial sets of homeodomain (HD) transcription factors, the expression of which is induced and maintained in NCCs through later developmental stages by signals from the surrounding environment. Recent work has begun to shed light on these intrinsic transcriptional mechanisms and their involvement in craniofacial and pharyngeal morphogenesis.

In Drosophila, the head region is patterned by two HD transcription factors: empty spiracle (ems) and orthodenticle (Otd). The mouse orthologues of ems and Otd are the Emx1/Emx2 and Otx1/Otx2 genes, respectively (Finkelstein and Perrimon, 1991). Otx2 is expressed in forebrain NCCs that colonize the frontonasal region and in midbrain NCCs that colonize the distal, mandibular, region of PA1 (Kimura et al., 1997; Kuratani et al., 1997). Otx2 is required in cranial NCCs during a specific time window [between E8.5 and E10.5 in the mouse (Fossat et al., 2006)] and its regulation is conserved between mouse and zebrafish (Kimura et al., 1997). While Otx2 homozygous mouse mutants lack head structures, Otx2 heterozygotes have developmental defects in frontonasal and distal mandibular elements, but no structural anomalies in the rhombencephalic NCC-derived structures or the bones of the skull vault (Matsuo et al., 1995). Inactivation of histone deacetylase (HDAC) 8 in NCCs induces severe skull vault defects, in part due to the upregulation of Otx2 (Haberland et al., 2009). This finding indicates that normal skull vault development requires epigenetic-mediated repression of Otx2. Despite the variability in the defects induced by Otx2 haplo-insufficiency, no homeotic transformation has been described, suggesting that, as in Drosophila, Otx2 does not act as a selector gene (see Glossary, Box 1) (Matsuo et al., 1995).

The hindbrain-derived NCCs that colonize the mandibular region of PA1 give rise to the Meckel's cartilage and to the middle ear ossicles incus and malleus (see Glossary, Box 1), whereas NCCs migrating into PA2 generate the middle ear ossicle stapes and part of the hyoid bone. NCC AP positional identity is thought to be acquired at the pre-migratory stage (Couly et al., 1998; Hunt et al., 1991b; Hunt et al., 1998; Noden, 1983; Prince and Lumsden, 1994; Saldivar et al., 1996; Trainor et al., 2002a). However, this identity is not irreversible but displays some degree of plasticity: intrinsic NCC molecular programs can be switched to new programs when NCCs are exposed to ectopic environmental cues (Couly et al., 1998; Hunt et al., 1998; Ishikawa and Ito, 2009; Prince and Lumsden, 1994; Schilling et al., 2001; Trainor and Krumlauf, 2000a; Trainor et al., 2002a).

NCC AP positional identity in pre-migratory progenitors is established by the same molecular mechanisms that control segmentation and patterning of the rhombomeres from which they delaminate; namely, by the nested and combinatorial expression of the HD transcription factors of the Hox (homeobox) gene family (Hunt et al., 1991a; Hunt et al., 1991b; Lumsden and Krumlauf, 1996). The maintenance of Hox gene expression in migrating NCCs and developing rhombomeres is regulated by independent Hox gene enhancers, sometimes resulting in expression differences between NCC subpopulations and their rhombomere of origin (Hunt et al., 1991a; Hunt et al., 1991b; Maconochie et al., 1999; Trainor and Krumlauf, 2000b; Tumpel et al., 2008; Tumpel et al., 2002). For example, the Hoxa2 expression domain has its anterior limit at the boundary between r1 and r2, whereas r2-derived NCCs migrating into PA1 are devoid of Hox gene expression (Krumlauf, 1993; Prince and Lumsden, 1994). The NCCs that contribute to the second and more posterior arches instead express specific combinations of Hox genes (Fig. 2A).

The involvement of Hox genes in establishing the AP positional identity of NCCs was first demonstrated by the targeted inactivation of mouse Hoxa2 (Gendron-Maguire et al., 1993; Rijli et al., 1993), which resulted in the homeotic transformation of PA2 into PA1-like skeletal elements (Fig. 3B). Rhombomere 4-derived PA2 elements are transformed into proximal PA1-like structures (Gendron-Maguire et al., 1993; Rijli et al., 1993), such as the proximal part of Meckel's cartilage, incus and malleus (Fig. 3B), which are normally mainly contributed by r1-r2-derived NCCs (Köntges and Lumsden, 1996). This supports the idea that distal mandibular and maxillary PA1 structures require AP specification from a Hox-independent patterning system, partially mediated by Otx2 (see above) (Kuratani et al., 1997). Hoxa2 downregulation in Xenopus (Baltzinger et al., 2005) and zebrafish (Hunter and Prince, 2002) also results in second-to-first arch homeotic changes, underscoring a conserved role for Hoxa2 in second (hyoid) arch patterning. Some differences exist, however, among vertebrates concerning the involvement of Hoxb2, the only Hoxa2 paralogue, in PA2 NCC patterning. Indeed, in contrast to mouse and Xenopus, Hoxb2 expression in zebrafish PA2 is maintained where it genetically interacts with Hoxa2 (Baltzinger et al., 2005; Barrow and Capecchi, 1996; Davenne et al., 1999; Hunter and Prince, 2002). Irrespective of such differences, these experiments show that paralogue group 2 (PG2) Hox genes are required to select a PA2-specific mode of development. Moreover, the outcome of PG2 Hox gene inactivations indicates that NCCs contributing to the proximal region of PA1 and PA2 share an underlying Hox-free ground (default) patterning molecular program (Rijli et al., 1993). This has been further supported by complementary gain-of-function experiments, showing that the ectopic expression of Hoxa2 in the Hox-free PA1 of chick and frog, or of Hoxa2/Hoxb2 in the Hox-free PA1 in zebrafish, is sufficient to induce PA2-like derivatives in place of PA1-like derivatives (Grammatopoulos et al., 2000; Hunter and Prince, 2002; Pasqualetti et al., 2000).

Hoxa2 performs its main patterning function in post-migratory NCCs, as assessed by its temporal inactivation in the mouse, and is thus directly involved in the morphogenesis of PA2 structures (Fig. 3B) (Santagati et al., 2005). Conversely, temporal induction of Xenopus XHoxa2 ectopic expression at post-migratory stages in PA1 is sufficient to induce homeotic transformation of mandibular elements towards hyoid morphology (Pasqualetti et al., 2000). These complementary experiments demonstrate that: (1) the skeletal pattern of mandibular and hyoid crest is not irreversibly committed before migration but NCC positional identity must be maintained through post-migratory stages in order to provide information about shape, size and orientation of PA2 skeletal elements; and (2) Hoxa2 is a selector of hyoid fate, representing an integral component of the PA2 NCC morphogenetic program (Santagati et al., 2005). At the molecular level, Hoxa2 participates in PA2 morphogenesis by modulating the competence of post-migratory NCCs to respond to local skeletogenic signals such as fibroblast growth factor (Fgf) (Bobola et al., 2003). This results in direct or indirect negative regulation of relevant transcription factors normally expressed in PA1 [including Pitx1 (paired-like homeodomain transcription factor 1), Lhx6 (LIM homeobox protein 6), Six2 (sine oculis-related homeobox 2 homologue), Alx4 (aristaless-like homeobox 4), Bapx1 (Nkx3-2 – Mouse Genome Informatics) and Barx1 (BraH-like homeobox 1)] (Fig. 3A) (Bobola et al., 2003; Kanzler et al., 1998; Kutejova et al., 2005; Kutejova et al., 2008; Minoux et al., 2009; Santagati et al., 2005).

The inactivation of Hox PG3 genes, particularly of Hoxa3, induces malformations of PA3 and PA4 NCC-derived skeletal elements (Chisaka and Capecchi, 1991; Manley and Capecchi, 1995). More severe phenotypes are observed in compound mutants, indicating synergistic interactions between members of the same paralogue group (Condie and Capecchi, 1994; Manley and Capecchi, 1997). However, no homeotic transformations of NCC-derived elements are seen in either single or compound Hox PG3 or PG4 mutants (Boulet and Capecchi, 1996; Chisaka and Capecchi, 1991; Condie and Capecchi, 1994; Horan et al., 1995; Manley and Capecchi, 1995; Manley and Capecchi, 1997; Ramirez-Solis et al., 1993). By contrast, selective deletion of the Hoxa cluster in NCCs results in the partial homeosis of PA3 and PA4 derivatives into morphologies that are characteristic of PA1-derived structures, in addition to the PA2 homeosis induced by the lack of Hoxa2 (Minoux et al., 2009). Thus, pre- and post-otic NCCs may share the same Hox-free PA1-specific ground patterning program (Fig. 3A). By modifying this common ground patterning molecular program in an arch-specific manner, Hox genes provide each PA with unique rostrocaudal identities (Fig. 2A, Fig. 3A) resulting in the formation of arch-specific skeletal elements that precisely connect with arch-specific subsets of muscles (Köntges and Lumsden, 1996; Noden, 1983).

Furthermore, the results of Minoux et al. (Minoux et al., 2009) rule out a simple `posterior prevalence' model (Duboule and Morata, 1994) in which Hox PG3 or PG4 expression would provide positional identity to the NCCs of PA3 or PA4, respectively. Indeed, the NCC-specific Hoxa cluster deletion has uncovered a role for Hoxa2 in the patterning of PA3 and PA4 that only becomes apparent in the absence of Hoxa3, supporting a prevalent role for Hoxa3 over Hoxa2 in patterning PA3 and PA4 (Minoux et al., 2009). The concomitant removal of Hoxa2 and Hoxa3 induces homeosis of PA3- and PA4-derived structures, unveiling synergistic genetic interactions between Hoxa2 and Hoxa3 (Minoux et al., 2009). Differences in Hoxa2 and Hoxa3 relative expression levels and/or target specificity might underlie the distinct effects of their inactivation. Moreover, expression of Hox genes in PA epithelia might further contribute to establish PA-specific patterning. The additional removal of the Hoxd cluster in a NCC-specific Hoxa-deleted background does not increase the extent of the homeotic phenotype (Minoux et al., 2009). Thus, the Hoxa cluster may have a primary role in the biology of skeletogenic NCCs (Duboule, 2007; Minoux et al., 2009), whereas Hoxb and Hoxd genes could provide a `quantitative backup' (Rijli and Chambon, 1997) that might become functionally relevant only in the absence of Hoxa genes.

The maintenance of Hox gene expression in NCCs is under epigenetic control. In the absence of the zebrafish histone acetyltransferase Moz (monocytic leukemia zinc finger) or its partner Brpf1 (bromodomain and phd finger containing 1), both of which are involved in chromatin remodelling, the expression of Hox PG1 to PG4 genes in NCCs is initiated but not maintained, resulting in anterior homeotic transformations of PAs similar to those observed in the Hoxa cluster deletion mutants (Laue et al., 2008; Miller et al., 2004). The nuclear matrix protein special AT-rich binding protein 2 (SATB2) represses Hoxa2 in the FNP and developing calvarial bones (Dobreva et al., 2006), whereas Hox gene repression in PA1 NCCs may be influenced by Fgf8-mediated signalling from the mid-hindbrain boundary (Irving and Mason, 2000; Trainor et al., 2002a). Indeed, Hox gene expression is incompatible with jaw and craniofacial development (Couly et al., 1998; Creuzet et al., 2002). The elucidation of the molecular mechanism(s) involved in Hox gene cluster repression in the first arch will be of particular interest in the future and will be crucial for understanding jaw development and evolution.

Altogether, the above findings underscore the importance of establishing the rostrocaudal positional identity of NCCs at pre-migratory stages and maintaining it through late post-migratory stages for the metameric assembly of the pharyngeal region of the vertebrate head. Indeed, a skeletogenic NCC-specific molecular ground pattern, shared by all PAs, is modified in an arch-specific manner to select appropriate segment-specific skeletal morphologies. This is achieved by the integration of local signals (see below) with intrinsic combinatorial programs of Hox gene function that, in turn, regulate directly or indirectly downstream effector genes through the distinct phases of NCC development. However, in addition to Hox-dependent AP positional addresses that distinguish the segmental identity of each arch from that of its neighbours, NCCs need also to be provided with dorsoventral positional information to establish intra-arch identity.

A distal-less homeobox (Dlx) code provides cranial NCCs with patterning information and intra-arch polarity along the dorsoventral (DV) (i.e. proximodistal in the mouse) axis (Fig. 2B). The vertebrate Dlx genes are HD transcription factor homologues of Drosophila Distal-less. Mammals have six Dlx genes (Dlx1-Dlx6) (Stock et al., 1996), which exhibit nested expression patterns. In each PA, Dlx1/2, Dlx5/6 and Dlx3/4 transcripts overlap distally but display offset proximal expression limits, with Dlx1/2 extending more proximally than Dlx5/6, which in turn extends more proximally than Dlx3/4 (Fig. 2B). In PA1, Dlx1 and Dlx2 are expressed in both the maxillary (upper jaw) and mandibular (lower jaw) processes. Dlx5 and Dlx6 are expressed only in the mandibular process; their expression extends close to the position of the future hinge region between the upper and lower jaws, which in mammals includes both the primary (incudo-malleal) and secondary (dentary-squamosal) jaw articulations (Depew et al., 2005). Dlx3 and Dlx4 expression domains are further restricted to the distal-most end of the mandibular process (Depew et al., 1999; Depew et al., 2002; Jeong et al., 2008; Qiu et al., 1997). In posterior PAs, the nested DV expression domains of Dlx genes intersect with the AP Hox code in NCCs (Santagati and Rijli, 2003) (Fig. 2).

The role of Dlx genes has been mainly investigated by loss-of-function mutations in the mouse. [A detailed account of the craniofacial alterations of such mutants is outside the scope of this review (Depew et al., 2005).] Here, we briefly discuss the functional importance of the Dlx code in PA patterning. As Dlx5/6 control Dlx3/4 expression (Depew et al., 2002; Jeong et al., 2008) (Fig. 4A), the partitioning of PA1 is mainly achieved with two Dlx combinations: Dlx1/2 for the maxillary and Dlx1/2/5/6 for the mandibular process. Single or compound Dlx1/Dlx2 mutants selectively affect the development of upper jaw elements and upper components of the hinge region (Depew et al., 2005; Qiu et al., 1997; Qiu et al., 1995). Conversely, inactivation of Dlx5 results in proximal lower jaw and hinge region abnormalities (Acampora et al., 1999; Depew et al., 1999). The simultaneous inactivation of Dlx5 and Dlx6, although having no effect on Dlx1/2 expression, results in a homeotic transformation of lower jaw into a mirror image of upper jaw derivatives, thus generating a mouse with two upper jaws facing each other with their associated soft tissue, including vibrissae (Beverdam et al., 2002; Depew et al., 2002) (Fig. 4B). Such an outcome provides the best evidence that a Dlx combinatorial code in NCCs establishes intra-arch polarity (Figs 2, 4). Conceptually, it may be similar to the above-mentioned Hox code for inter-arch patterning. Indeed, intra-arch polarity is also achieved by modifying an underlying Dlx1/2-positive ground (default) patterning program, corresponding to the PA1 maxillary process molecular program, upon which Dlx5/6 select a mandibular identity (Fig. 4). However, it is unclear at the moment whether this Dlx ground patterning program also extends to more posterior PAs, as for the Hox genes (Minoux et al., 2009). Even though PA2 and PA3 skeletal elements are affected by the double Dlx5/Dlx6 inactivation, such malformations cannot be assigned to clear homeotic transformations (Depew et al., 2002; Depew et al., 2005). Similar to the Hox code, synergistic interactions between distinct Dlx paralogue groups, namely Dlx1/2 and Dlx5/6, are observed, indicating both qualitative and quantitative features of the code (Depew et al., 2005). For example, when Dlx1 or Dlx2 are inactivated on Dlx5- or Dlx6-deficient backgrounds, the mandibular process derivatives are reduced in size and/or transformed into structures that resemble maxillary elements, with disruption of the hinge region partially reproducing the Dlx5/Dlx6 mutant phenotype (Depew et al., 2005; Jeong et al., 2008).

In PA1, Dlx5/6 induce and/or maintain the expression of several genes involved in the development of the mandibular process, including Dlx3/4, Hand1/2 (heart and neural crest derivatives expressed transcripts), Alx3/4, Pitx1, Gbx2 (gastrulation brain homeobox 2), Bmp7 (bone morphogenetic protein 7) and Evf2 [Dlx6os1 – Mouse Genome Informatics; Dlx6 opposite strand transcript 1; itself involved in Dlx6 regulation (Feng et al., 2006)], while repressing other genes, the expression of which is normally restricted to the maxillary process and is under the control of Dlx1/2, including Pou3f3 (POU domain, class 3, transcription factor 3), Foxl2 (forkhead box L2) and Irx5 (Iroquois-related homeobox 5) (Beverdam et al., 2002; Depew et al., 2002; Jeong et al., 2008). Interestingly, Dlx1/2 and Dlx5/6 also act partially redundantly in the mandibular process, where they upregulate mandibular and repress maxillary process-specific genes, such as Pou3f1. Dlx1/2 have therefore opposing roles in specifying the mandibular versus maxillary processes of PA1 (Jeong et al., 2008), perhaps through differential interactions with distinct co-factors. Further studies will be required to resolve this issue.

Thus, the molecular information provided at the intersection of AP and DV positional values is converted into NCC patterning and differentiation programs that are appropriate for each spatial coordinate, prompting the question `how are such intrinsic programs influenced by the local cues to which NCCs are exposed?'.

In recent years, considerable effort has been made to identify the environmental signals that are involved in craniofacial development. Here, we discuss recent insights into the signalling systems that establish and maintain the spatiotemporal identity of NCCs, and how they impinge on the NCC transcriptional program to form structures of the appropriate shape, size and orientation.

Endothelin 1 (Edn1/ET1) is secreted by the surface ectoderm [where it performs its main role (Nair et al., 2007)], and by the mesoderm and endoderm of PAs. It acts by binding to its G protein-coupled endothelin A receptor (Ednra), which is expressed in NCCs (Clouthier et al., 1998; Dettlaff-Swiercz et al., 2005; Ivey et al., 2003; Kurihara et al., 1994; Miller et al., 2000; Nair et al., 2007; Nataf et al., 1998; Offermanns et al., 1998; Sato et al., 2008a) (Fig. 4A). In mouse, the targeted inactivation of Edn1, endothelin receptor A (Ednra) or of endothelin converting enzyme 1 (Ece1, which cleaves Edn1 to form an active peptide) induces severe malformations of ventral PA elements, whereas dorsal structures are less affected (Clouthier et al., 1998; Kurihara et al., 1994; Yanagisawa et al., 1998). This Edn1-ventralizing activity is particularly evident in PA1 where most of the mandibular process-derived structures are homeotically transformed into a second set of maxillary-like elements, thus mimicking the compound Dlx5/Dlx6 mutation (Ozeki et al., 2004; Ruest et al., 2004) (Fig. 4B). Indeed, Dlx5, Dlx6 and some of their molecular targets, such as Hand1/2 and Dlx3, are downstream effectors of the Edn1/Ednra pathway. Thus, Edn1/Ednra establish a mandibular process identity partly by initiating a Dlx5/Dlx6-dependent transcriptional program (Charite et al., 2001; Clouthier et al., 2000; Ozeki et al., 2004; Thomas et al., 1998; Yanagisawa et al., 2003) (Fig. 4B). Conversely, constitutive activation of Ednra throughout the pharyngeal mesenchyme induces the replacement of the maxillary components by a second set of mandibular elements (Sato et al., 2008b) (Fig. 4B). A similar homeotic transformation is also induced by the ectopic expression of Hand2, thus indicating that, within PA1, the Edn1/Ednra→Dlx5/Dlx6→Hand2 signalling pathway works as a genetic switch that is responsible for the choice of NCCs to select a mandibular, instead of a maxillary, process molecular program (Sato et al., 2008b) (see above). Nonetheless, the mandibular process molecular program also requires an Ednra-independent pathway (Ruest et al., 2004; Sato et al., 2008a).

Some evidence in zebrafish indicates that within the arch epithelium, and the mesodermal core, edn1 is regulated by Tbx1 (Piotrowski et al., 2003). To specify mandibular process identity, Ednra/Edn1 interaction is only necessary during a narrow temporal window between E8.25 and E9.0 of mouse embryogenesis (Fig. 4A). If Ednra-mediated signalling is inactivated after this period, mandibular structures develop normally and Dlx5/Dlx6 expression is no longer downregulated, indicating that Ednra signalling is mainly required for the initiation of a mandibular process patterning program in NCCs (Fukuhara et al., 2004; Ruest and Clouthier, 2009; Ruest et al., 2005). The molecular mechanisms involved in the maintenance of Dlx5/Dlx6 expression, as well as the potential later roles of these genes in craniofacial development, remain to be elucidated.

The Ednra/Edn1 pathway is conserved in jawed vertebrates (Kempf et al., 1998; Miller et al., 2000; Walker et al., 2006; Walker et al., 2007). In zebrafish, the edn1 defective mutant sucker displays severely reduced ventral PA cartilages that are fused with dorsal cartilages of the same arch (Kimmel et al., 1998; Miller et al., 2000; Piotrowski et al., 1996). Moreover, edn1 may act as a morphogen to pattern specific skeletal elements along the entire DV axis (Kimmel et al., 2003; Miller et al., 2003). In PA2, high Edn1 levels induce the formation of ventral dermal bones (the branchiostegal rays), whereas low levels of Edn1 specify the dorsal bones (the opercles) (Kimmel et al., 2003). In PA1, edn1 is involved in the correct positioning of the primary articulation between upper and lower jaws by regulating the expression of bapx1, while also inducing hand2 ventrally (through dlx function, see above), which in turn represses bapx1 (Miller et al., 2003) (Fig. 4A). As a result, the expression domain of bapx1, which regulates growth differentiation factor 5 (gdf5) and chordin expression, is restricted to a subset of NCCs at an intermediate DV position in PA1, defining where the jaw joint will form (Miller et al., 2003). Edn1 may additionally repress dorsal fate in ventral skeletal precursors by inhibiting Jagged-Notch signalling, which ensures dorsal identity of mandibular and hyoid skeletal components (Zuniga et al., 2010). In contrast to mouse, zebrafish possess two Ednra genes: ednra1 and ednra2. The knockdown of ednra1 alone disrupts the jaw joint, whereas the knockdown of both eliminates the ventral arch skeletal elements, thus reproducing the edn1 mutant phenotype (Nair et al., 2007). These data indicate that higher levels of Edn1 are required to pattern joint cells compared with ventral cartilages (Nair et al., 2007).

Notably, in lamprey (a jawless vertebrate), homologues of Edn1 and Ednra are expressed as in jawed vertebrates, indicating the involvement of this signalling pathway in craniofacial patterning early in vertebrate evolution (Kuraku et al., 2010). However, the expressions of Dlx or bapx1 genes are not spatially restricted in lamprey PAs (Kuraku et al., 2010), suggesting that the elaborate genetic program leading to the Dlx code is likely to have been acquired uniquely in jawed vertebrates (see below).

FGF and BMP molecules have multiple roles during craniofacial and pharyngeal skeletal morphogenesis. For example, FGFs are key NCC survival factors (Abu-Issa et al., 2002; Frank et al., 2002; Macatee et al., 2003; Szabo-Rogers et al., 2008; Trumpp et al., 1999; Walshe and Mason, 2003). In the mouse, Fgf8 conditional inactivation in PA1 ectoderm induces massive NCC apoptosis, resulting in the absence of most PA1 skeletal elements (Trumpp et al., 1999). Conversely, Fgf8-soaked beads placed in presumptive PA1 ectoderm causes r3-derived NCC proliferation, overcoming the failure of facial development induced by removing NCCs from di-, mes- and rhombencephalon down to r2 (Creuzet et al., 2004). In zebrafish, endodermal fgf3 expression also exerts a trophic effect on NCCs. Early fgf3 downregulation, or preventing its maintenance through later stages, causes apoptotic elimination of PA3 and PA4 NCCs (David et al., 2002; Nissen et al., 2003). fgf8 and fgf3 expressed in the mesoderm and hindbrain can also act as chemoattractants to promote the lateral migration of endodermal cells, which is required for the segmentation of the pharyngeal endoderm into pouches and, in turn, for the correct patterning of NCC-derived skeletal elements (Crump et al., 2004; Graham, 2008). Moreover, FGFs are involved in directing NCCs to adopt the ectomesenchymal fate (Blentic et al., 2008).

FGF signalling is also involved in specification of NCC spatial identity and in establishing the AP and DV polarity of PAs. The mouse mandibular process is subdivided into oral (anterior) and aboral (posterior) domains. This AP polarity is established by Fgf8, which diffuses from the oral epithelium and controls, in a concentration-dependent manner, the differential expression of HD transcription factors in the underlying mesenchyme (Grigoriou et al., 1998; Tucker et al., 1999), such as Lhx6 and Lhx7, which are expressed in the oral domain in response to high Fgf8 levels, and goosecoid (Gsc), which remains restricted to the aboral mesenchyme (Grigoriou et al., 1998; Tucker et al., 1999) (Fig. 5A). Gsc expression, which is required for patterning aboral skeletal elements (Rivera-Perez et al., 1995; Yamada et al., 1995), is also controlled by the Edn1 pathway (Clouthier et al., 1998; Clouthier et al., 2000), underscoring that distinct pathways may interact and converge on the same targets to specify NCC spatial identity. In this respect, in chick PA1, Fgf8- and Bmp4-mediated signalling also contribute to the correct positioning of the Bapx1-positive jaw joint (Wilson and Tucker, 2004), probably by interacting with the Edn pathway (Fig. 5A). In addition, Edn1 expression is downregulated in the Fgf8 conditional knockout mouse (Trumpp et al., 1999).

At early developmental stages, Fgf8 and Bmp4 pattern the presumptive maxillo-mandibular region and define it from the pre-mandibular domain (Shigetani et al., 2000). Before mesencephalic NCCs arrive at their destination, ectodermal Fgf8 prefigures the prospective oral cavity (Haworth et al., 2004; Shigetani et al., 2000). This Fgf8 expression domain, which is induced by sonic hedgehog (Shh) signalling from the endoderm (Haworth et al., 2004; Haworth et al., 2007), is delimited by Bmp4 expressed on both sides of the adjacent Fgf8-expressing ectoderm (Shigetani et al., 2000). Ectodermally derived Fgf8 and Bmp4 control, in turn, the regionalization of incoming NCCs through the activation of specific patterning genes: Fgf8 induces the expression of Dlx1 and Barx1, while Bmp4 induces Msx1 (muscle segment homeobox like 1) expression in the underlying mesenchyme (Fig. 5A). These epithelio-mesenchymal interactions are crucial for specifying the identity of the pre-mandibular and maxillo-mandibular regions (Shigetani et al., 2000). In addition to its role in establishing AP and DV PA identity, Fgf8 is required for left-right symmetry of the craniofacial skeleton (Albertson and Yelick, 2005; Albertson and Yelick, 2007).

Blocking Shh signalling in chick, mouse and zebrafish induces severe head skeleton abnormalities, including holoprosencephaly and cyclopia, which result from defects in NCC survival, proliferation and patterning (Ahlgren and Bronner-Fraser, 1999; Chen et al., 2001; Chiang et al., 1996; Cordero et al., 2004; Hu and Helms, 1999; Jeong et al., 2004; Washington Smoak et al., 2005). Several sources of Shh, such as the foregut endoderm, the neuroepithelium and the facial ectoderm, are crucial in these processes. The foregut endoderm (see Glossary, Box 1) is a major source of the patterning signals that provide NCCs with information about the size, shape and orientation of the skeletal elements that are generated in the first and more posterior PAs (Couly et al., 2002; Graham, 2008; Ruhin et al., 2003). Recent studies have identified Shh as one such signal (Benouaiche et al., 2008; Brito et al., 2006; Brito et al., 2008). Its absence in the ventral foregut endoderm prevents the development of Meckel's cartilage and associated PA1 structures because of massive NCC apoptosis (Brito et al., 2006). By contrast, providing an extra source of Shh to the presumptive PA1 induces the formation of two supernumerary Meckel's cartilages that develop in a mirror-image to the normal one. The formation of these structures is preceded by Bmp4, Fgf8 and Shh ectopic expression in the caudal ectoderm of PA1 (Brito et al., 2008). The Shh patterning effect on PA1 is not seen in PA2 (Brito et al., 2008), suggesting that other signals act independently, or additionally, to Shh in more posterior PAs. In zebrafish, fgf3 downregulation in the pharyngeal pouch endoderm leads to the formation of a hyoid cartilage with an inverted AP polarity (David et al., 2002), suggesting that fgf3 expressed in the dorsal half of the pouch polarizes PA2 skeletal structures.

In addition to its role in foregut endoderm, Shh signalling arising from the ventral brain primordium is required in zebrafish for the proper specification of the roof of the stomodeum (see Glossary, Box 1), which in turn promotes NCC condensation to form the anterior neurocranium and upper jaw cartilages (Eberhart et al., 2006). This early function of Shh is followed by a later role in NCCs to promote their differentiation into cartilages (Eberhart et al., 2006; Jeong et al., 2004; Wada et al., 2005). Moreover, Shh plays a role in establishing skeletal polarity along the mediolateral axis of the embryo (Eberhart et al., 2006; Wada et al., 2005).

Shh signalling from the facial ectoderm is also involved in the specification of NCC spatial identity. In the ectoderm overlying the FNP, a signalling centre called the frontonasal ectodermal zone (FEZ), defined by the juxtaposition of Fgf8 and Shh expression domains, regulates the growth and DV polarity of the upper beak in birds (Hu et al., 2003) (Fig. 5). When grafted ectopically, the FEZ can reprogram the developmental fate of the underlying NCCs, inducing upper beak duplications with a DV polarity that reflects the orientation of the grafted tissue. However, mandibular NCCs form a supernumerary lower (and not upper) beak when in contact with an ectopic FEZ, and Hox-positive NCCs are unresponsive to FEZ cues (Hu et al., 2003). Thus, the epithelial-mediated patterning instruction is not absolute, but is interpreted by the NCCs according to their relative AP positional identity. The same observation holds also true for signals arising from the endoderm (Couly et al., 2002).

At early stages, Shh emanating from the forebrain acts on the NCCs, which in turn induce Shh expression in the FEZ (Marcucio et al., 2005) (Fig. 5A). Part of this signalling network could include BMPs (Foppiano et al., 2007) (Fig. 5A). Once the FEZ is established, reciprocal inhibitory interactions contribute to the maintenance of the transient Fgf8/Shh molecular boundary, which marks the dorsal tip of the upper beak (Abzhanov et al., 2007) (Fig. 5A). In avian embryos, the FEZ spans the entire FNP, whereas in mouse, two FEZ are present – in the left and right median nasal processes (Hu and Marcucio, 2009b). If the Shh pathway is ectopically activated in the forebrain of chick embryos, two instead of one FEZ are induced in the facial ectoderm, mimicking the mouse organization (Hu and Marcucio, 2009a). Thus, modifications in the molecular organization of the FEZ might in turn generate distinct patterns of growth in the upper face and contribute to morphological diversity among species (Hu and Marcucio, 2009a; Hu and Marcucio, 2009b).

In mouse, the double inactivation of the retinoic acid (RA)-synthesizing enzymes Raldh2 and Raldh3 (Aldh1a2 and Aldh1a3 – Mouse Genome Informatics), which are expressed in the forebrain and facial ectoderm leads to a partial lack of skeletal structures derived from the FNP, reproducing the compound retinoic acid receptor (RAR) α/RARγ mutant phenotype (Halilagic et al., 2007; Lohnes et al., 1994). Whereas RARα and RARγ act cell-autonomously in NCCs, RARβ is dispensable (Dupe and Pellerin, 2009). A local source of RA in the chick rostral head coordinates FNP morphogenesis through Fgf8 and Shh induction in the forebrain and facial ectoderm (Schneider et al., 2001) (Fig. 5A). RA is also required for FNP spatial patterning. Increasing RA levels in the presumptive maxillary field, while inhibiting BMP signalling, transforms the side of the beak (derived from the maxillary process) into a second set of midline structures (normally derived from the FNP) (Lee et al., 2001). This suggests that an endogenous source of RA is necessary to pattern the FNP and its derivatives: the pre-nasal and pre-maxillary skeletal elements. Accordingly, the nasal pit is an important RA source (Dupe et al., 2003) that, together with Fgf8 and Bmp4, coordinates local cell survival around the nasal placode, thus contributing to the shaping of the face (Song et al., 2004; Szabo-Rogers et al., 2008) (Fig. 5B).

The morphogenesis of the pharyngeal region also requires RA signalling (Dupe et al., 2003; Lohnes et al., 1994; Mark et al., 2004; Mendelsohn et al., 1994; Niederreither et al., 1999; Vermot et al., 2003; Wendling et al., 2000). The double inactivations of RARα and RARγ, or RARα and RARβ, induce PA hypoplasia and NCC-derived cartilage malformations (Dupe et al., 1999; Lohnes et al., 1994; Mendelsohn et al., 1994; Wendling et al., 2000). However, the endoderm could be the primary target of RA action in mediating morphogenesis of PA2 and more posterior arches (Matt et al., 2003), as RARs are mostly dispensable in cranial NCCs (Dupe and Pellerin, 2009). Not only RA deficiency, but also exogenous RA excess during pregnancy or the inactivation of the cytochrome P450 (Cyp) RA-degrading enzymes induce teratogenesis in the craniofacial region (Abu-Abed et al., 2001; Maclean et al., 2009; Mulder et al., 2000; Reijntjes et al., 2007; Uehara et al., 2007). Hence, a tight balance between the production and degradation of RA is necessary for correct NCC-derived craniofacial skeleton morphogenesis.

Recent findings support the view that heterochronic (in time), heterotopic (in place) or quantitative changes in the expression of key signalling molecules in the face epithelia and/or NCCs might underlie the evolution and variation of facial morphology. Modifications in the level and/or spatiotemporal distribution of the HD transcription factors that modulate the intrinsic NCC ground patterning molecular programs could also lead to the morphological evolution of skeletal elements. Such mechanisms could modify the positional identity of cranial NCCs, thus potentially resulting in species-specific skeletal elements of different size and/or shape.

Variation in beak size and shape among birds is one example of craniofacial morphological evolution that is well studied. The exchange of NCCs between the presumptive beak regions of quails and ducks generates quail-like beaks in duck hosts, and vice versa (Schneider and Helms, 2003). NCCs from donors respond to local signalling from the host environment, while partially maintaining their original molecular programs, and regulate expression in adjacent host tissues to specify the morphology of the beak in a species-specific manner (Schneider and Helms, 2003; Tucker and Lumsden, 2004). The spatial organization and activity of proliferation zones in the FNP or between the FNP, maxillary and nasal prominences may also result in species-specific beak shape (Wu et al., 2004; Wu et al., 2006). Bmp4 mediates the activity of such proliferation zones, and its expression level in the FNP is crucial for final beak morphology (Abzhanov et al., 2004; Wu et al., 2004). Indeed, Darwin's finch species with deeper and broader beaks express Bmp4 in their beak prominences at higher levels and at earlier stages than do species with narrow and shallow beaks (Abzhanov et al., 2004). In chick embryos, Bmp4 overexpression in the FNP mesenchyme induces a significant enlargement of the beak width and depth, while Noggin-mediated inhibition induces its reduction (Abzhanov et al., 2004). Recently, calmodulin (CaM), a molecule involved in mediating Ca²⁺ signalling, has been suggested to control beak length in Darwin's finch. In chick, the upregulation of the CaM-dependent pathway in the FNP leads to an elongated upper beak, reproducing the beak morphology of Darwin's cactus finches (Abzhanov et al., 2006). Thus, two different pathways, Bmp4- and CaM-dependent signalling, act to shape the beak along different axes: CaM controls the length, whereas Bmp4 controls the width and depth (Fig. 5B). This provides evidence that variations in Bmp4 and/or CaM expression levels may have contributed to species-specific beak morphologies (Abzhanov et al., 2004; Abzhanov et al., 2006).

Temporal variations of signalling molecule activity could also underlie morphological evolutionary changes. Fgf8 and Shh, which are expressed in the FEZ of the facial epithelia, promote cartilage outgrowth, in part, by inducing Bmp4 expression in the underlying NCCs (Abzhanov and Tabin, 2004; Hu and Marcucio, 2009b) (Fig. 5B). Hence, modifications in the timing of their expression could induce shape diversity. Accordingly, it has been proposed that the persistence of Fgf8 expression in the facial ectoderm of duck, but not of chick, contribute to the distinct morphology of the duck beak, inducing it to produce more cartilage (Wu et al., 2006). In quail-duck chimaeric embryos, bone and cartilage formation are controlled by the NCCs that also control the stage-specific size and shape of skeletal structures, in part by temporally regulating their own expression of Bmp4 (Eames and Schneider, 2008; Merrill et al., 2008).

In the jawless lamprey, the oral apparatus consists of a lower lip and a velum, derived from the mandibular region, and an upper lip, derived from the pre-mandibular region. By contrast, in gnathostomes (jawed vertebrates), the lower jaws, the primary jaw joint and part of the upper jaws are derived from the mandibular region (Shigetani et al., 2002; Takio et al., 2004) (but see Cerny et al., 2004; Lee et al., 2004). Hence, phenotypically similar protrusions, i.e. the lips in lampreys and the jaws in gnathostomes, originate from non-equivalent cell populations (Shigetani et al., 2002). A heterotopic (posterior) shift of epithelial-mesenchymal interactions could underlie these changes, placing jaws as evolutionary innovations rather than as simple modifications of the ancestral oral apparatus (Shigetani et al., 2002).

One feature of the bilaterian body plan is its organisation in successive metameric units along the AP axis. Segment-specific morphology is achieved by modifying underlying ground (default) patterning molecular programs that are shared by all segments of a given series, to provide successively more posterior positional identities. In the hindbrain and pharyngeal regions of vertebrate embryos, such ground patterning programs correspond to the molecular programs of the Hox-free rostralmost elements of the series, namely rhombomere 1 and PA1, respectively (see above) (Minoux et al., 2009; Rijli et al., 1993; Waskiewicz et al., 2002) (Fig. 3A). Such a strategy may be widely conserved in animal evolution because in the short germ band beetle Tribolium, the complete inactivation of Hox genes induces the transformation of all embryonic segments into antennae (Brown et al., 2002).

The morphological diversity of NCC-derived skeletal elements among vertebrates could arise by modifying the level, timing and/or spatial expression of HD factors involved in the specification of spatial identity, or by modifying the expression pattern of genes involved in establishing the molecular ground patterning program. This is illustrated by the dramatic morphological and molecular modifications that have occurred in PA1 during the transition from jawless to jawed vertebrates, yielding the appearance of the primary jaw articulation between upper and lower jaw processes. In gnathostomes, this is likely to be accompanied by the modification of the PA1 ground pattern by a Dlx code, which generates DV polarity in the NCC-derived skeletal elements (Beverdam et al., 2002; Depew et al., 2002; Depew et al., 2005) (Fig. 4). It is noteworthy that lampreys possess multiple Dlx genes; however, their expression is not nested within PAs (Kuraku et al., 2010; Neidert et al., 2001). Focal expression of bapx1 is also a gnathostome feature (Kuraku et al., 2010). Both dlx5/6 and bapx1 expression is under the control of edn signalling (Fig. 4A), which might therefore be involved in the evolutionary changes resulting in the appearance of the jaw joint (Miller et al., 2003). A further modification in the PA1 DV patterning program may have subsequently occurred in tetrapods because, in mouse (unlike in zebrafish), Bapx1 is unable to regulate Gdf5 and Gdf6, two genes that are essential for the formation of the jaw joint (Tucker et al., 2004). Therefore, the regulatory cascade initiated by bapx1 in fish is not maintained in mouse, thus potentially contributing to species-specific morphologies.

The existence of a PA1 molecular ground patterning program shared by all segments predicts that any genetic change modifying DV patterning information in PA1 would be integrated into the basic genetic program of all PAs and, in turn, would be modified by segment-specific Hox codes along the AP axis to yield segment-specific structures. Indeed, the inactivation of Hoxa2 in post-migratory PA2 NCCs results in the ectopic induction of Bapx1 in the mutant PA2 (Santagati et al., 2005) (Fig. 6). This indicates that the DV molecular changes leading to the appearance of the jaw joint in PA1 of jawed vertebrates have also been fixed in the molecular ground pattern of PA2, where they undergo PA2-specific repression in response to the post-migratory expression of Hoxa2 in NCCs. This additionally suggests that a late function of Hoxa2 in PA2 NCCs is to interfere with Edn-mediated induction of Bapx1 and to prevent the formation of a jaw joint in PA2. Such a function appears to be conserved, as Hoxa2 overexpression in Xenopus PA1 after NCC migration is sufficient to downregulate Xbap and cause loss of articulation and mandibular-to-hyoid transformation (Pasqualetti et al., 2000). Changes in the levels and/or local spatial distribution of Hoxa2 in PA2 of different species might therefore differently modulate the underlying molecular ground patterning program and the response of downstream targets to endothelin signalling, resulting in the generation of species-specific elements along the PA2 DV axis (e.g. the mammalian sound-conducting stapes or fish hyomandibular skeleton). Thus, interactions between AP ground patterning and DV molecular programs might provide the molecular framework to link the DV morphological evolution of PA1 to the coordinated morphological changes in the other elements of the rostrocaudal series (Fig. 6). Further work is required to support such a speculative model and to address potential genetic interactions between the Hox and Dlx codes.

Many birth defects are associated with craniofacial malformations. It is becoming increasingly clear that these craniofacial abnormalities could be attributed to defects in the generation, proliferation, migration and differentiation of cranial NCCs. Thus, a full comprehension of craniofacial development will rely on a complete dissection of the molecular processes that underlie NCC differentiation and patterning. Significant progress has been made recently in revealing conserved gene regulatory modules that are crucial for patterning the cranial neural crest and signalling pathways that regulate tissue interactions during craniofacial development. The next challenge is that of achieving a comprehensive systems biology understanding of how collections of genes integrate into functional networks to generate normal or defective complex craniofacial phenotypes. It will be important to reconstitute the molecular aspects that make up the signalling mechanisms of epithelial-mesenchymal interactions and the spatiotemporal epigenetic regulation of NCC transcriptional programs of differentiation at the single and cell population levels. Integrating information on genetic, cellular and systems levels will hopefully provide insights into the mechanisms that underlie complex craniofacial phenotypes.

We thank Christophe Laumonnerie for help with the artwork and discussion. Work in F.M.R.'s laboratory is supported by the Swiss National Science Foundation and the Novartis Research Foundation.