Signalling between the hindbrain and paraxial tissues dictates neural crest migration pathways

The cranial neural crest is a pluripotent mesenchymal population that plays a critical role in construction of the vertebrate head. Arising at the junction between the neural plate and surface ectoderm (Selleck and Bronner-Fraser, 1995) cranial neural crest forms connective tissue, nerves, ganglia, cartilage and skull bones (Le Douarin, 1983; Noden, 1982). Many craniofacial malformations, therefore, are largely attributable to defects in the patterning, proliferation, migration or differentiation of this cell population (Sulik et al., 1988). The cranial neural crest is derived primarily from the hindbrain, which is transiently subdivided into seven rhombomeres (r), and these form lineage-restricted compartments (Lumsden and Krumlauf, 1996; Trainor and Krumlauf, 2000a; Vaage, 1969). This segmental organisation is conserved and critical for patterning the pathways of neural crest cell migration. In chick and other vertebrate embryos, hindbrain-derived neural crest cells generally migrate as three distinct streams adjacent to r2, r4 and r6, contributing to the first, second and third branchial arches (ba), respectively (Kulesa et al., 2000; Kulesa and Fraser, 2000; Lumsden et al., 1991; Osumi-Yamashita et al., 1994; Sadaghiani and Theibaud, 1987; Schilling and Kimmel, 1994; Sechrist et al., 1993; Serbedzija et al., 1992; Trainor and Tam, 1995). In contrast, there are clear neural crest-free zones adjacent to r3 and r5, generating a two-segment periodicity in the patterns of neural crest migration. Since neural crest arising at distinct axial levels helps to establish the unique anterior-posterior (A-P) identity of structures in its adjacent branchial arch (Hunt et al., 1991; Noden, 1983; Trainor and Krumlauf, 2000a), it is important to understand the mechanisms that segregate these migrating populations. In this respect these crest-free zones could be important for branchial arch patterning by preventing the intermixing of neural crest populations with different positional identities.

The function and the mechanism by which the r3 and r5 crest-free zones are established are contentious. On one hand, analyses in avian embryos reported elevated levels of cell death in the premigratory neural crest populations of r3 and r5 (Graham et al., 1993), mediated by Bmp4 signalling from the even rhombomeres. This induces Msx2 in r3 and r5, leading to the elimination of neural crest before migration (Ellies et al., 2000; Graham et al., 1994; Graham et al., 1993). Hence, inter-rhombomeric signalling would be the key factor in modulating the generation of crest cells from odd rhombomeres. In contrast, lineage and time-lapse analyses in avians clearly show that r3 and r5 do generate neural crest cells, but they move anteriorly and posteriorly, joining the even rhombomere streams (Kulesa, 1998; Kulesa et al., 2000; Kulesa and Fraser, 2000; Sechrist et al., 1993). Furthermore, in frog and fish embryos, r5 produces equivalent amounts of crest cells that migrate laterally, compared to the even rhombomeres (Smith et al., 1997). The variation between species in the amount and migratory pathways of neural crest cells generated by r5, implies that the crest-free zones are not an intrinsic property of odd-numbered rhombomeres themselves. This suggests that environmental influences adjacent to r3 and r5 may be important in controlling the pathways of hindbrain neural crest migration (Farlie et al., 1999), and distinct mechanisms may be utilised by different species to pattern these events. Therefore it is important to understand the mechanisms responsible for segregating the branchial arch streams in mouse embryos and determine if they are functional in other vertebrates.

In this study, we used cell grafting and lineage-tracing techniques in cultured mouse embryos to investigate the interactions between the hindbrain and the environment with respect to patterning the pathways of mouse cranial neural crest cells. We demonstrate that the patterns of cell death in the hindbrain do not correlate with the generation or migration of neural crest. Furthermore, odd-numbered rhombomeres have the same capacity to generate neural crest cells as the even segments. Reciprocal transpositions between even and odd rhombomeres show that specific regions are inhibitory to neural crest cell migration. Our findings demonstrate that the pathways of hindbrain neural crest migration and crest-free zones adjacent to r3 in the mouse are generated by combinatorial signalling events between the hindbrain and the adjacent environment, and this mechanism appears to be conserved in the chick.

Embryos were obtained from timed-pregnant matings of CBA×C57/BL6 or from transgenic line (EphA4-GFP; Hoxb1-lacZ) (Pöpperl et al., 1995) crosses with F₁ females. Host embryos used in culture experiments, 8.5-9.5 days post coitum (d.p.c.), were dissected from the uterus with an intact visceral yolk sac, amnion and ectoplacental cone (Trainor et al., 1994). Previous work has shown that the first migrating population of neural crest cells in the mouse leaves the caudal midbrain-rostral hindbrain neuroepithelium at the 5-6 somite stage (Chan and Tam, 1988). Therefore to catch the earliest waves of migrating neural crest cells in rhombomere grafting experiments, only donor and host embryos having 5 or less pairs of somites were used. After grafting, host embryos were cultured in vitro for 24-48 hours as previously described (Sturm and Tam, 1993; Trainor and Krumlauf, 2000b).

Consistent neuromeric landmarks, described by Trainor and Tam (Trainor and Tam, 1995), were used to identify the axial level of tissue to be grafted and the correct site of transplantation. Finely polished alloy and glass needles were used to separate the neuroectoderm from adjacent tissues. Tissue fragments that could not be cleanly separated from adjacent tissues were incubated in 0.5% trypsin, 0.25% pancreatin, 0.2% glucose and 0.1% polyvinylpyrolidone in phosphate-buffered saline (PBS) for 5 minutes at 37°C or in Dispase for 5 minutes at 37°C to ensure a pure population. Isolated tissue fragments were then washed in DMEM (Dulbecco’s modified Eagles medium), before being labelled with DiI or DiO by soaking in a 1:1 mix of DiI:DR50 or DiO:DR50 (Manzanares et al., 1999), for 2 minutes. Tissue fragments were then washed in DMEM and dissected in DR50 (Sturm and Tam, 1993) using glass needles into smaller fragments consisting of approximately 15-30 cells.

Small groups of approximately 15-30 cells from r3 or r5 of 8.5 d.p.c. embryos were orthotopically transplanted back to the same sites and heterotopically transplanted into r2 or r4 of isochronic embryos.

Small groups of approximately 15-30 cells from r4 of 8.5 d.p.c. embryos were orthotopically transplanted back to the same site and heterotopically transplanted into r2 and r3 of isochronic embryos.

Small fragments of mesoderm or ectoderm were isolated from adjacent to r3 and transplanted next to r4 in isochronic embryos.

Grafted embryos were analysed for DiI and green fluorescent protein (GFP) labelling by fluorescence microscopy and with a Leica TCS NT confocal microscope (567 nm excitation) before being assayed for cell death.

8.5-9.5 d.p.c. embryos were assessed for rhombomeric cell death via Nile Blue, Acridine Orange or TUNEL staining. For Nile Blue and Acridine Orange staining, embryos were cultured in DR50 (containing a 1:400 dilution of 1.5% Nile Blue in water or 5 μg/ml of Acridine Orange) for 30-40 minutes. Apoptotic cells stain intensely blue when incubated in Nile Blue and are epifluorescent (rhodamine) after incubation with Acridine Orange. TUNEL staining labels apoptotic cells fluorescently (fluorescein) and was performed according to the manufacturer’s instructions (Boehringer Mannheim).

Bmp4 and Msx2 digoxigenin-labelled riboprobes were synthesised (Boehringer Mannheim) and whole-mount in situ hybridisation was performed as previously described (Wilkinson, 1992).

In order to understand the mechanisms regulating the pathways of neural crest migration in mouse embryos it was necessary to precisely map the neural crest contributions made by each rhombomere to the branchial arches. We therefore performed focal injections of DiI into the developing hindbrain of 8.25 d.p.c. mouse embryos to mark small groups of cells in specific pre-rhombomeric territories (r1-r6), preceding the migration of cranial neural crest cells. Care was taken to label only very small groups of cells to ensure that any labelled neural crest cells were derived from only one rhombomere. These embryos were cultured in vitro for 24-36 hours and assayed for the relative contribution and patterns of migration of neural crest cells from each marked group of cells (Fig. 1A-F). Labelling in r1 and r2 produce neural crest cells, which migrate laterally and extensively populate the entire proximo-distal extent of the ba1 (Fig. 1A,B). By comparison, r3 produces relatively few neural crest cells, but some migrate anteriorly to join the r2 stream and contribute to the most proximal region of the ba1 (Fig. 1C, arrowhead). Hence, the first branchial arch (ba1) is populated by neural crest cells derived from the midbrain and r1-3. The remainder of the r3-derived neural crest cells migrate posteriorly and make a minor contribution to the proximal region of ba2 (Fig. 1C, arrow). A lineage tag in future r4 reveals that neural crest cells derived from this segment primarily populate the second arch and they colonise the entire proximo-distal extent of ba2 (Fig. 1D). r5-derived neural crest migrates anteriorly and posteriorly to join the even-numbered neural crest streams from r4 and r6 (Fig. 1E) in a manner similar to cells derived from r3. However, r5 generates more neural crest cells than r3 and contributes less neural crest to ba2 than r4, but significantly more than r3. As a consequence of these patterns, ba2 is composed of neural crest derived from three rhombomere segments, r3-r5 (Fig. 1C-E,G). Lineage tracing r5 and r6 reveal that the third arch (ba3) is populated primarily by r6-derived neural crest cells (Fig. 1F) and also, to a lesser extent, by the posteriorly migrating neural crest derived from r5 (Fig. 1E, arrow, and data not shown). Together, these results show that mouse hindbrain-derived neural crest cells migrate in three segregated streams adjacent to the even-numbered rhombomeres into the branchial arches (Fig. 1G), and that this pattern is very similar to that observed in the chick embryo. Each rhombomere in the mouse produces neural crest cells, with the even-numbered segments the most, followed by r5 and then r3. There are also clear neural crest-free zones adjacent to r3 and r5.

Since the patterns of neural crest migration from r3 and r5 were similar between the mouse and chick, it was important to examine whether there was also a conservation in the rhombomeric patterns of apoptosis (Graham et al., 1994; Graham et al., 1993) that could account for the diminished production of neural crest cells relative to the other rhombomeres.

To score for general patterns of apoptosis, we used the vital dyes Nile Blue Sulphate and Acridine Orange, which stain the highly chromatin-rich apoptotic bodies. This enabled the detection of apoptotic cells in embryos cultured in vitro, so that a picture of cell death can be accumulated over a number of hours during development rather than at a single time point. Using mouse embryos from 8.5-9.5 d.p.c., vital staining demonstrated that, during the period of neural crest migration, cell death occurs in a temporally and spatially dynamic manner (Fig. 2A-D). Although we observed consistent and reproducible cell death in the hindbrain, there was no specific pattern that could be attributed to either odd or even rhombomeres. Even within a single embryo, the unfused neural folds often display completely different patterns of cell death, highlighting the dynamic nature of events (Fig. 2A, arrowheads). The elevated levels of cell death detected between 8.5-9.5 d.p.c. were associated with the normal caudal-to-rostral progression of neural tube closure and subsequent formation of the roof plate in the hindbrain (Fig. 2A-D). Similar results were obtained independently by TUNEL staining of fixed embryos (data not shown).

The absence of localised cell death in r3 and r5 prompted us to examine the patterns of Bmp4 and Msx2 gene expression during mouse hindbrain development and neural crest migration (Fig. 3). In contrast to the chick, mouse Bmp4 is not expressed in r3 or r5 during this period. At 8.5 d.p.c., mouse Bmp4 is completely absent from the dorsal neural tube but is expressed in more ventral tissues, such as the heart (Fig. 3A). By 9.5 d.p.c., Bmp4 is expressed in Rathke’s pouch (arrow) and along the anterior epithelium of the first branchial arch, but is still absent from the hindbrain (Fig. 3B,C). Similarly, Msx2 expression, which is activated by BMP signalling in apoptosis, also differs markedly from the patterns of its chick homologue, which are confined to r3 and r5 during hindbrain development (Graham et al., 1994; Graham et al., 1993). In the mouse Msx2 is expressed uniformly in the dorsal edges of the neural tube along the entire A-P axis (Fig. 3D,E). By 9.5 d.p.c., mouse Msx2 expression is reduced in the hindbrain but is elevated in the roof plate and the distal halves of the branchial arches (Fig. 3E,F). Therefore, Bmp4/Msx2 mediated apoptosis of odd-rhombomere neural crest cells co-ordinated by signalling between even- and odd-numbered segments is unlikely to be responsible for patterning the pathways of neural crest migration in the mouse.

The absence of rhombomere-specific cell death raised the possibility that the reduced capacity of r3 to generate neural crest is an intrinsic property of this rhombomere. To address this issue we isolated and transposed cells from r3 to different rhombomeres to test their ability to generate neural crest cells in novel environments. In homotopic control grafts, r3 cells (labelled with DiI) transplanted back to r3 generated very few neural crest cells (Fig. 4A). In agreement with the lineage studies described above, these cells migrated anteriorly or posteriorly to populate the proximalmost region of the first and second branchial arches (Fig. 4A, and data not shown). Surprisingly, r3 cells transplanted heterotopically to r2 or r4 gave rise to large numbers of neural crest cells, which populated the entire proximodistal extent of the first and second branchial arches, respectively (Fig. 4B,C). Similarly r5 also has the capacity to generate large amounts of neural crest when transplanted into r2 and r4 (Fig. 4D, and data not shown). Hence r3 and r5 have the intrinsic capability to generate neural crest cells in a manner indistinguishable from even-numbered rhombomeres. However, this ability appears to be restricted during normal development.

One explanation for the ability of r3 to generate elevated levels of neural crest in r2 or r4 is that these cells have changed their identity. To address this issue we utilised an EphA4-GFP transgenic line expressing the reporter specifically in r3 and r5 under the control of Krox20 as a marker for segmental identity (Theil et al., 1998). We isolated genetically marked r3 cells, labelled them with DiI and then transplanted cells homotopically to the same site or heterotopically to other rhombomeres in host embryos (Fig. 4E-H). This enabled us to test for the ability of r3 to generate neural crest cells in novel environments and to monitor the regional identity of the grafted tissue. As expected from the results above, r3 tissue grafted into r4, generates increased amounts of neural crest cells (Fig. 4G,H). In the hindbrain the transposed r3 cells retain their original identity, confirmed by GFP expression in the ectopic r4 environment (Fig. 4E,F). In contrast, nearly all the graft-derived crest cells downregulated the reporter, consistent with the idea of plasticity in dispersed migrating cranial neural crest populations, as described in previous studies (Kulesa et al., 2000; Kulesa and Fraser, 2000; Schilling, 2001; Trainor and Krumlauf, 2000a; Trainor and Krumlauf, 2000b). Therefore the ability of r3 to generate increased amounts of neural crest in the r2 or r4 environment cannot be attributed to a change in identity of the transplanted rhombomere cells.

These experiments suggest that the environment adjacent to r3 plays an important role in restricting the delamination, migration and/or proliferation of neural crest cells in ba2. Hence, an important issue is whether even-numbered rhombomeres retain their ability to generate substantial amounts of neural crest cells in ectopic r3 environments. To test this, we used a Hoxb1-lacZ transgenic line expressing the reporter specifically in r4, to isolate, DiI-label and graft small groups of r4 cells into host embryos (Fig. 5). In control transplantations, r4 cells grafted back into r4 gave rise to numerous neural crest cells, which extensively populated ba2 (Fig. 5A). Similarly, when r4 cells were transposed into r2 they retained the ability to generate numerous neural crest cells, which migrate into ba1 (Fig. 5B). In contrast, when r4 cells were transplanted into r3, their capacity to produce migrating neural crest cells was repressed (Fig. 5C,D). We confirmed that this repression is not due to a change in their segmental identity, as indicated by continued lacZ expression in the grafted hindbrain tissue (data not shown). To exclude the possibility that the absence of migrating neural crest cells was due to cell death of the grafted tissue, we performed TUNEL staining (Fig. 5E-G). TUNEL and staining for β-gal show that the grafted cells are retained in r3, express the appropriate marker and are hence not eliminated by cell death. These results imply that the local environment adjacent to r3 prohibits the formation and migration of neural crest cells equally well from odd- and even-numbered rhombomeres.

The inability of r3-derived neural crest cells to migrate laterally, and the existence of a neural crest-free territory adjacent to r3, suggest the existence of environmental zones with combinatorial signals that prohibit neural crest formation and migration. To explore the source of the inhibitory signal(s), we examined whether paraxial tissues, such as the cranial mesoderm or surface ectoderm, were responsible for patterning the migration pathways of neural crest cells. To test if these tissues are sufficient to block the lateral migration of neural crest cells, we isolated mesoderm or surface ectoderm adjacent to r3 and transplanted small amounts of each tissue next to r4 (Fig. 6). The grafted DiO-labelled mesoderm and ectoderm tissue colonised the host, proliferated and contributed to the second branchial arch, as expected for tissues placed in this location (Fig. 6B,D). In all cases, irrespective of the proximo-distal location of the transplanted tissue, mesoderm and ectoderm did not inhibit migration of r4-derived neural crest cells (DiI labelled) into ba2 (Fig. 6A,C,D). These results imply that neither the cranial mesoderm nor the cranial ectoderm alone is sufficient for dictating the pathways of neural crest migration in the mouse. It may be possible that the cranial mesoderm and surface ectoderm act together to pattern the neural crest pathways. However, this seems unlikely given that in the chick, 180° rotations of the cranial ectoderm and mesoderm adjacent to r3-r4 do not alter the pathways of neural crest migration (Sechrist et al., 1994). This implies that mesoderm and surface ectoderm do not appear to secrete a broadly diffusible signal that dominantly inhibits neural crest migration.

One caveat to these experiments is that it is difficult to control or maintain the relative position of the grafted mesoderm and surface ectoderm. If these tissues need to be in direct contact with the neural crest cells as they are beginning to delaminate or migrate from the neural epithelium, we might not have been able to assay for short-range signals or interactions. Hence we cannot formally exclude the possibility that surface ectoderm or mesoderm alone are critical in these events, as the proper balance of inhibitory signals might not have been established due to improper positioning or time of exposure. Nonetheless, together these analyses argue that signalling between the hindbrain and these tissues in the developing arch environment is required to mediate the inhibition of neural crest formation and migration. This is consistent with the fact that neural crest is generated at the junction between the surface ectoderm, mesoderm and neural plate and a series of interactions may be involved in setting up and maintaining the inhibitory influences adjacent to r3.

Our rhombomere transpositions in mouse showed that the relatively low production of r3-derived neural crest cells is not intrinsic to the rhombomere itself. Rather, combinatorial interactions between r3 and the surrounding arch environment as it develops, restrict the generation and migration of cranial neural crest cells. In contrast, in chick embryos it has been postulated that inter-rhombomeric signalling between even- and odd-numbered rhombomeres is responsible for limiting the production of neural crest (Ellies et al., 2000; Graham et al., 1994; Graham et al., 1993). In this model the ability of r3 to generate neural crest cells is repressed by signals received from its interaction with the adjacent r2 and r4 territories.

To distinguish between these mechanisms we performed transpositions from mouse to chick. Mouse r3 tissue grafted into r4 in the chick, becomes incorporated and proliferates, generating neural crest cells, which populate ba2 (Fig. 7A). This is further evidence that r3 has the intrinsic capacity to generate neural crest cells. Conversely the ability of r4 to generate neural crest is lost when mouse r4 tissue is transplanted into r3 in the chick (Fig. 7B). This is not due to cell death or changes in identity as the grafted cells continue to express the lacZ reporter (Fig. 7C). This illustrates that the environment adjacent to r3 in the chick also restricts the ability of r4 to generate migratory neural crest and that mouse rhombomeric cells have conserved the ability to respond to these signalling interactions. Since in both of these types of transpositions, r3 and r4 cells are juxtaposed and capable of interacting with each other, our experiments show that interactions between the arch environment and rhombomeres are a critical determinant in controlling the production of neural crest. Together with the neural crest-free zones adjacent to r3 in other vertebrates, these results suggest that there is a common mechanism involving interactions between r3 and arch tissues used to govern the patterning of neural crest derived from r3.

In this study we demonstrate that environmental signals adjacent to r3 in the mouse, established by combinatorial tissue interactions, inhibit the lateral migration of neural crest cells. This is consistent with time-lapse analyses in the chick showing that crest cells arising in r3 extend filopodia into the adjacent territory but retract these projections and fail to migrate laterally (Kulesa, 1998). Recently, evidence has been obtained for a neurite growth inhibitor in the r3-adjacent mesenchyme that patterns central projections of cranial sensory axons, and this axon-growth inhibitory activity may be part of the mechanism that restricts neural crest migration from r3 into the adjacent mesenchyme (Golding et al., 1999). As a consequence of this lateral inhibition, r3-derived neural crest cells migrate anteriorly and posteriorly to join the neural crest streams emigrating from r2 and r4, and this is a common feature of all vertebrates (Fig. 8).

Our lineage analyses in the mouse show that r3 has a reduced capacity to generate neural crest cells, and reciprocal transpositions experiments between r3 and r4 revealed that this is not an intrinsic property of the rhombomere. Since the interfaces between r3 and r4 and their respective axial identities in these transposition experiments were maintained, this implies that inter-rhombomeric signalling is not responsible for the reduced levels of neural crest cells generated from r3. Furthermore, lineage analyses in the chick have shown that the reduced level of neural crest cell delamination from r3 occurs throughout the entire segment and not just at the borders with even rhombomeres (Kulesa, 1998; Sechrist et al., 1993). This implies that environmental signals are not only limiting the lateral migration of neural crest but also, directly or indirectly, influencing the ability of r3 to generate neural crest. It may be that the inhibition of lateral neural crest migration feeds back on r3 itself to regulate the amounts of neural crest cells that can be made. Alternatively, signals in the adjacent environment could act directly on r3 to modulate the induction of neural crest cells.

The transposition of cranial mesoderm and surface ectoderm suggest that the signals influencing the generation and migration of neural crest are established by interactions between the hindbrain and these tissues. This suggests several possibilities. Positive, emigration-provoking signals, potentially in concert with extracellular matrix differences, may be present in some mesodermal populations either intrinsically or through interactions with the endoderm. Furthermore, neural crest survival, emigration and migration into the arches may result from the sequential interactions initiated in the hindbrain during neural plate stages. Hence, in a manner similar to what we have found in studying cranial neural crest plasticity in A-P patterning, the proper program of events governing the migration of crest may need first to be established in the hindbrain, to allow migratory crest cells to interpret and respond to environmental signals set up through a series of tissue interactions. This contrasts with the trunk, where the segmental migratory pattern of neural crest through the rostral sclerotome is directly imposed by the rostrocaudal character of the mesoderm (Bronner-Fraser and Stern, 1991; Keynes and Stern, 1984). The exclusion of neural crest adjacent to r3, and the rostral and caudal patterns of migration of r3-derived neural crest observed in the mouse, create a neural crest-free zone. This is a common feature in vertebrates (Farlie et al., 1999) and suggests that the signalling interactions we have observed in the mouse reflect a conserved mechanism for patterning the distribution of neural crest cells from r3. This raises several interesting issues with respect to neural crest and craniofacial patterning.

An important question is why is it necessary to establish exclusion zones and restrict r3 neural crest migration? One major reason may be a need to prevent mixing between non-Hox and Hox expressing neural crest cells, which populate ba1 and ba2, respectively (see Fig. 8). A variety of gain- and loss-of-function studies have shown that Hoxa2 is primarily responsible for specifying second branchial arch fate and can also inhibit the lower jaw skeleton (Couly et al., 1998; Gendron-Maguire et al., 1993; Grammatopoulos et al., 2000; Kanzler et al., 1998; Pasqualetti et al., 2000; Rijli et al., 1998; Rijli et al., 1993). Therefore Hoxa2-expressing populations must be excluded from the first branchial arch neural crest. This is achieved in part by restricting the generation of neural crest from r3 and also by changes in identity of the small number of anteriorly migrating cells, due to plasticity and cell community effects (Schilling, 2001; Trainor and Krumlauf, 2000b). Furthermore, the tight contacts and communication between migrating neural crest cells derived from a similar location may help to feed back the influences of inhibitory signals in the arch environment (Kulesa et al., 2000; Kulesa and Fraser, 2000).

While this mechanism excludes the lateral migration of r3-derived neural crest, other mechanisms must operate to prevent in-filling and mixing of neural crest from adjacent territories to maintain an exclusion zone. For example, in ErbB4 mutants a population of r4-derived neural crest cells acquires the ability to migrate through the dorsal mesenchyme adjacent to r3 (Golding et al., 2000). This phenotype arises due to changes in the mesenchyme and is not autonomous to the neural crest (Golding et al., 2000). Since ErbB4 is expressed only in r3 and r5, this phenotype reflects defects in signalling between r3 and its adjacent environment (Golding et al., 2000). There is also evidence that Eph/ephrin signalling is another mechanism that contributes to restrictions in the mixing of branchial arch neural crest populations (Fig. 8). In Xenopus embryos, the neural crest delaminates as a contiguous A-P band and separation is achieved only during the later phases of migration through the differential expression of the Eph receptor and ephrin ligand families of genes (Sadaghiani and Theibaud, 1987; Smith et al., 1997). Hence these mechanisms keep neural crest streams segregated and prevent infilling from adjacent territories, working in concert with restrictions in the lateral migration of r3 neural crest to establish crest-free zones.

There are important differences in the mechanisms governing neural crest migration from r3 and r5. Firstly, in mouse, fish and frog embryos r5 generates considerably more neural crest than r3. Secondly, the migration patterns of r5 neural crest are not as well conserved in vertebrates as they are for r3. In frog and fish embryos r5 generates equivalent amounts of neural crest compared to the even rhombomeres, which also migrate laterally (Smith et al., 1997). In contrast, in mouse (Fig. 1) and chick embryos (Kulesa, 1998; Kulesa et al., 2000; Kulesa and Fraser, 2000; Sechrist et al., 1993), crest from r5 migrates rostrally and caudally to join adjacent even-rhombomere streams (Fig. 8). Whilst the general migration pattern of r5-derived cells is similar to r3, it does not appear to arise by the same mechanism. The otic vesicle is positioned immediately adjacent to r5, which provides a physical barrier rather than an exclusion zone that inhibits lateral migration from r5 (Fig. 8). This is consistent with the fact that in fish and frog embryos, the otic vesicle is positioned more laterally, thus allowing the unimpeded lateral migration of neural crest cells from r5 that is seen in these species (Sadaghiani and Theibaud, 1987; Schilling and Kimmel, 1994; Smith et al., 1997; Snape et al., 1991). Furthermore, in mouse and chick transplantations, moving r5 from the proximity of the otic vesicle results in the lateral migration of r5-derived neural crest cells (Fig. 4D) (Saldivar et al., 1996). Thus there are differences in the ability of the environments adjacent to the odd-numbered rhombomeres to influence neural crest migration. Hence the interactions between r3 and the environment, which generate signals that establish a neural crest-free zone, appear to be unique to that rhombomere.

Our interspecies grafting experiments, in which mouse r3 was transposed into chick r4, confirms that r3 has the capacity to generate neural crest in a different species (Fig. 7). Conversely transplants of mouse r4 into chick r3 display a repressed ability to generate neural crest cells. This suggests that the environmental signals adjacent to r3, which restrict the initial generation and migration of neural crest cells, are conserved between the species. In contrast, other mechanisms contributing to neural crest patterning appear to have diverged. In the chick, interactions between odd and even rhombomeres generate segmental patterns of apoptosis mediated by a Bmp4-Msx2 signalling loop that contributes to the reduction of neural crest from r3 (Ellies et al., 2000; Graham et al., 1994; Graham et al., 1993). However, in the mouse hindbrain the patterns of Bmp4 and Msx2 expression are not segmentally restricted in comparison to the chick. Bmp4 is not expressed in the mouse hindbrain and Msx2 is expressed uniformly along the dorsal edge of the neural tube. While it is possible that other Bmps may be involved, neither Bmp7 or Bmp2 are segmentally expressed in odd rhombomeres. Furthermore, this is consistent with our findings in the mouse that while cell death occurs in a temporally and spatially dynamic manner, there is no evidence for cell death specifically in odd rhombomeres during the period of neural crest migration (Fig. 2).

Since cell death in odd-numbered rhombomeres does not contribute to the patterning of mouse cranial neural crest, this raises the issue of why there are differences between the species. One explanation may be related to fundamental differences in the timing and manner in which neural crest cells are generated in each species. Neural crest migration in the mouse occurs during neural fold elevation prior to neural tube closure, whereas in the chick, neural crest migration occurs well after neural tube closure (Le Douarin, 1983; Le Douarin and Kalcheim, 1999). As a consequence the conserved interactions adjacent to r3 that inhibit crest migration may cause crest cells to build up between the neural tube and surface ectoderm in the chick and not the mouse. Hence, in chick, cell death may be required to clear cells in r3 and r5 that are unable to migrate.

In conclusion, our grafting experiments, together with existing studies in the chick (Farlie et al., 1999; Saldivar et al., 1996), suggest that signalling interactions between r3 and the surrounding environment restrict the generation and migration of cranial neural crest. To date few molecules that influence the pathfinding of cranial neural crest cells have been identified, but examples include the ephrins and their Eph receptors (Helbling et al., 1998; Smith et al., 1997) and evidence is emerging to indicate that collapsin-1/semaphorin-III might also be involved (Eickholt et al., 1999). Future analyses in mouse embryos will be aimed at identifying the molecular networks that interact to generate and pattern the migration pathways of cranial neural crest cells.

We thank P. Mealyer for help with animal husbandry, and N. Itasaki, L. Ariza-McNaughton and members of the Krumlauf laboratory for valuable discussions. P. T. was supported by EMBO and HFSP postdoctoral fellowships. This work was funded by Core MRC Programme support, The Stowers Institute for Medical Research and an HFSP International Network Research Grant (RG0146/2000-B) with R. K. as Principal Investigator.