Eph signalling functions downstream of Val to regulate cell sorting and boundary formation in the caudal hindbrain

The developing vertebrate hindbrain is subdivided into a series of segments, visible transiently as morphological bulges, termed rhombomeres (Vaage, 1969). Rhombomeres form the foundation for segmental patterning of neurones and of cranial neural crest migration into the pharyngeal arches (Kimmel et al., 1985; Metcalfe et al., 1986; Lumsden and Keynes, 1989; Lumsden et al., 1991). Although all rhombomeres produce a similar array of cell types, the molecular identity of each individual rhombomere specifies distinct pattern elements.

Grafting experiments in chick show that boundaries form when rhombomeres from adjacent positions are juxtaposed, but not when rhombomeres from the same axial level are put together (Guthrie and Lumsden, 1991), suggesting that the apposition of cellular territories with distinct identities is required for formation of a boundary. Although the molecules responsible for rhombomere boundary formation are not fully understood, the Eph family of receptor tyrosine kinases and their obligate membrane-bound ligands, the ephrins, are candidates for a role. Using a dominant negative approach it has been shown that disruption of Eph signalling results in embryos showing a loss of the normal segmental restriction of gene expression within the developing hindbrain (Xu et al., 1995). More-recent experiments have shown that mosaic activation of Eph molecules leads to sorting of cells at rhombomere boundaries (Xu et al., 1999). Eph signalling has also been implicated in segment boundary formation in the paraxial mesoderm (Durbin et al., 1998).

The Eph family of receptors and their ligands, the ephrins, are divided into two classes (Flanagan and Vanderhaeghen, 1998; Holder and Klein, 1999). Class A ephrins are tethered to the plasma membrane by a glycosyl phosphatidylinositol (GPI) linkage and preferentially bind EphA receptors; class B ephrins (which preferentially bind EphB receptors) have integral transmembrane and intracellular domains and transduce signals back into the ligand-expressing cell via their intracellular domain following receptor binding (Bruckner et al., 1997; Eph Nomenclature Committee, 1997; Davis et al., 1994; Holland et al., 1996). Although binding is promiscuous within a class, different combinations of ligand and receptor interact with different affinities (reviewed in Flanagan and Vanderhaeghen, 1998). EphA4 is the only receptor shown to bind ephrins of both classes (Gale et al., 1996).

Functional studies show that in a variety of developmental contexts, cell repulsion is a major consequence of signalling between Eph receptors and ephrins. For example, localised expression of class A ephrins in the posterior optic tectum and of class B ephrins in the posterior half of each somite is required for repulsion of the axons from EphA receptor-expressing retinal ganglion cells and EphB receptor-expressing trunk motoneurones, respectively (Brennan et al., 1997; Bruckner and Klein, 1998; Monschau et al., 1997; Wang and Anderson, 1997).

Many members of the Eph family of signalling molecules are expressed in rhombomere-restricted patterns (reviewed in Flanagan and Vanderhaeghen, 1998) and are potential downstream targets of segmentally expressed transcription factors. For example, the transcription of ephA4 in r3 and r5 is under the direct control of the segmentally expressed zincfinger transcription factor, Krx20 (Theil et al., 1998).

The bzip transcription factor Val, a homologue of mouse Kreisler, is expressed in a stripe in the developing hindbrain corresponding to r5/6 (Moens et al., 1998). Zebrafish embryos homozygous for a null mutation for val have no visible rhombomere boundaries caudal to the r3/4 interface (Moens et al., 1996). The val mutant phenotype is first detected soon after the end of gastrulation, by reduced expression of krx20 in r5 (r3 expression is unaffected). Analysis of the positions of reticulospinal neurones and mapping of marker gene expression indicates that the hindbrains of val⁻ embryos are shorter than their wild-type and heterozygous siblings by the length of one rhombomere. The mutants have no region of r5 or r6 identity, instead they possess a region one rhombomere in length and of mixed identity, termed rX, that lies between (but does not form morphological boundaries with) r4 and r7 (Moens et al., 1996). Since Eph signalling is implicated in boundary formation, we have determined whether the rhombomeric expression of Eph molecules is controlled by Val, thereby addressing whether Eph receptors and ephrins are downstream of Val in a genetic hierarchy required for rhombomere boundary formation.

In addition to its role in boundary formation, Val function is required cell autonomously for cells to contribute to r5/6 of the developing hindbrain (Moens et al., 1996). When uncommitted val⁻ cells are transplanted into the presumptive hindbrain of a wild-type host, the mutant cells are progressively expelled from r5/6. The converse experiment demonstrates that wild-type cells are not properly incorporated into rX of val⁻ hosts and instead aggregate into clumps (Moens et al., 1996). These characteristic cell mixing behaviours seen in val⁻↔wild-type genetic mosaic embryos reflect the normal cell movements at the rhombomere boundaries, although the movements take place over longer distances in the mosaics. Cell behaviour in the genetic mosaics is also reminiscent of the cell-sorting phenomena observed in chick rhombomere grafting experiments (Guthrie and Lumsden, 1991; Guthrie et al., 1993). Since interactions between Eph receptors and ephrins result in repulsive cell responses and cell sorting (Mellitzer et al., 1999), in this study we examined the possibility that Eph signalling underlies the cell sorting phenomena observed in val⁻↔wild-type mosaics.

We show that zebrafish ephB4a and ephrin-B2a are expressed in complementary rhombomere-restricted domains in the developing hindbrain, and that the receptor and ligand they encode can bind together in situ. We show that Val function is required for activation of ephB4a expression in r5/6, and for repression of ephrin-B2a expression in this same region. The absence of alternating territories of receptor-expressing and ligand-expressing cells correlates with an absence of boundaries in val mutants. Furthermore, juxtaposition of ephB4a-expressing and ephrin-B2a-expressing cells (in wild-type→val⁻ mosaics) results in ectopic boundary formation at the interface. Finally, by overexpressing ephB4a or by disrupting bi-directional EphB signalling in val⁻→wild-type mosaics, we have rescued the inability of val⁻ cells to contribute to r5/6 of wild-type embryos. These results indicate that Val controls expression of Eph molecules, and that repulsive interactions between EphB4a and Ephrin-B2a are important for cell sorting and boundary formation in the caudal hindbrain.

Breeding fish were maintained at 28.5°C on a 14-hour light, 10-hour dark cycle. Wild-type and val^b337embryos were collected by natural spawning and staged according to Kimmel et al. (Kimmel et al., 1995).

Whole-mount in situ hybridisations using digoxigenin-or fluorescein-labelled antisense RNA probes were performed essentially as described (Xu et al., 1994; Hauptmann and Gerster, 1994). Hybridisation and detection of alkaline phosphatase fusion protein affinity probes was performed as described (Cheng and Flanagan, 1994). Immunostaining using a 1/500 dilution of a polyclonal anti-GFP antibody (Clontech) was performed as for other antibodies (Xu et al., 1994), except for the inclusion of an amplification step using the ABC kit (Vector Labs). Filamentous actin was stained by overnight incubation of fixed embryos with a 1/40 dilution of Alexa Red-phalloidin (Molecular Probes).

The isolation and characterisation of a partial cDNA clone for ephB4a has already been described (rtk5; Cooke et al., 1997). This partial clone was used as a probe in a high-stringency screen of a 3-to 15-hour random-primed zebrafish cDNA library to isolate partial cDNAs encompassing the 5′ coding region of ephB4a, using standard methods (Sambrook et al., 1989). The complete open reading frame of ephB4a was thus contained within two overlapping clones. To synthesise a construct reconstituting the entire open reading frame of ephB4a, a PCR-based technique was used (Horton et al., 1990). ephB4a was subcloned into the pCS2 vector for in vitro synthesis of RNA. A construct encoding soluble Ephrin-B2a was made by inserting a stop codon in-frame after residue 218. Capped RNA was synthesised and injected as previously described (Durbin et al., 1998). The GenBank Accession Numbers are AJ005026 for ephB4a, and AJ004863 for ephrin-B2a.

Mosaic analysis was performed as described (Moens et al., 1996; Moens and Fritz, 1999). Donors and hosts were allowed to develop, and the genotype of mutants and siblings was ascertained either by visual inspection between 18 somites (18s) stage and prim-5 stage, by in situ hybridisation with krx20, or by PCR and PvuII digestion (val^b337, the allele used throughout, is characterised by a PvuII site polymorphism; Moens et al., 1998).

Biotin-labelled donor cells were detected in fixed host embryos using the ABC kit (Vector Labs) and a fluorescein-tyramide substrate (NEN; Moens and Fritz, 1999). Donor embryos injected with RNA encoding GFP (either alone or co-injected with RNA for Eph constructs) were screened for brightly fluorescent cells at epiboly stages. Owing to mosaic distribution of RNA in injected embryos, not all donor cells contain GFP protein. To ensure that essentially all the donor cells were GFP positive, the transplant procedure was monitored frequently with fluorescence microscopy. Localisation of donor cells in host embryos was detected in live embryos by GFP fluorescence or in fixed embryos by immunostaining for GFP or enzymatic staining for β-galactosidase. For statistical analysis (see Table 1), only host embryos in which donor cells were spread throughout the hindbrain and into the spinal cord were counted. The limits of r5 and r6 were ascertained by position with respect to the otic vesicle or by in situ hybridisation with probes for ephrin-B2a or krx20.

Zebrafish ephB4a (previously rtk5; Cooke et al., 1997) and ephrin-B2a (previously ephrin-B2; Durbin et al., 1998) are expressed in complementary stripes throughout the presumptive hindbrain. ephB4a expression is strong in presumptive r2 from bud stage, with a sharp anterior boundary, but is graded caudally into presumptive r3 and there is low level expression in presumptive r5/6 at the late two-somite stage. From the three-somite stage (Fig. 1A), this caudal domain corresponds exactly to the r5/6 expression domain of val (seen at the seven-somite stage in Fig. 1B). Expression of val first appears at bud stage, slightly earlier than the first expression of ephB4a in presumptive r5/6 (Moens et al., 1998). Expression of ephB4a in the rostral hindbrain becomes restricted to r2, whereas caudally, strong expression of ephB4a is maintained in r5/6 between the eight and 18-somite stages (Fig. 1C) after which expression begins to decrease.

ephrin-B2a is expressed in a stripe in the presumptive hindbrain from late gastrulation stages (95% epiboly). Analysis of subsequent stages using probes for ephrin-B2a and for krx20 (Oxtoby and Jowett, 1993) indicates that this stripe is in r4 (Fig. 1D,E). Additional expression in r1 is first detected at the one-somite stage, and by the two-somite stage, there are stripes of ephrin-B2a expression in presumptive r1, r4 and r7 (seen at the three-somite stage in Fig. 1D). Levels of expression in r1 and r7 increase during early somite stages (Fig. 1E, F) but are always weaker than the r4 domain. ephrin-B2a expression domains in the caudal hindbrain abut the r5/6 domain of val expression both rostrally and caudally (Fig. 1F).

Confocal time-lapse image analysis of living zebrafish embryos has demonstrated that the earliest morphological indications of rhombomere boundaries appear at the five-somite stage (for the r3/4 and r4/5 boundaries), shortly afterwards for the r6/7 boundary, and by ten somites for the r5/6 boundary (Moens et al., 1998). Therefore, the expression of ephB4a and ephrin-B2a in defined stripes in the developing hindbrain prefigures the formation of morphological boundaries.

Complementary expression of Eph receptors and ephrins on either side of a presumptive boundary is required for somite boundary formation (Durbin et al., 1998). Consistent with a role in rhombomere boundary formation, the expression domains of ephB4a and ephrin-B2a in the developing hindbrain are complementary to each other. Double in situ hybridisations with probes for ephB4a and for ephrin-B2a show that their domains of expression abut at the r1/2, r3/4, r4/5 and r6/7 boundaries during the stages at which the boundaries are becoming visible (Fig. 2A).

To test if EphB4a and Ephrin-B2a can bind together in vivo, we ectopically expressed ephB4a by injecting DNA into two-cell stage embryos, then hybridised fixed embryos at gastrula stages to a fusion protein consisting of the Ephrin-B2a extracellular domain coupled to an alkaline phosphatase reporter (Ephrin-B2a-AP; see Cheng and Flanagan, 1994; Brennan et al., 1997). Alkaline phosphatase activity was detected in injected embryos (Fig. 2B), but not in uninjected embryos (Fig. 2C), indicating that the extracellular domain of Ephrin-B2a recognises and binds to the ectopically expressed receptor. Thus, EphB4a and Ephrin-B2a can recognise and bind each other in situ, and their territories of expression coincide with complementary rhombomeres in the developing hindbrain, suggesting that these molecules interact and signal to each other as a receptor-ligand pair in vivo at rhombomere boundaries.

The coincidence of val and ephB4a expression, and their complementarity to ephrin-B2a expression, suggests that Val may regulate the expression of these genes in the caudal hindbrain. To test this hypothesis, we analysed the expression of ephB4a and ephrin-B2a in val⁻ embryos. In situ hybridisation analysis of val⁻ embryos indicates that Val is indeed required to spatially regulate ephB4a and ephrin-B2a expression. The r5/6 stripe of ephB4a expression present in wild-type embryos from the late two-somite stage to the prim-5 stage (see Fig. 1A-C, Fig. 3A, and data not shown) is virtually absent in val⁻ embryos throughout the same range of stages (Fig. 3B,C). A low level of ephB4a expression is sometimes seen in the caudal hindbrain of val⁻ embryos at early somite stages (Fig. 3B), but this is lost by the 10-somite stage (Fig. 3C). Expression of ephB4a in the midbrain, in r2/3 and in the 3rd arch cranial neural crest appears unaffected (Fig. 3B,C).

In contrast to the downregulation of ephB4a, ephrin-B2a is upregulated in the caudal hindbrain of val⁻embryos. Instead of defined stripes of expression in r1, r4 and r7 (see Fig. 1D-F, Fig. 3D), in val⁻ embryos, ephrin-B2a is expressed in r1, and in an enlarged caudal domain encompassing r4, rX and r7 (Fig. 3E,F). Initially (at the three-to four-somite stage), expression in r4 is stronger than in rX/r7 (Fig. 3E), however, by eight to ten somites, there is uniform high-level expression of ephrin-B2a throughout the caudal hindbrain of val⁻ embryos (Fig. 3F). Therefore Val promotes expression of ephB4a and represses expression of ephrin-B2a in the territory between r4 and r7.

Val function is also required for r5 expression of ephA4, a second Eph receptor able to bind Ephrin-B2a (Durbin et al., 1998). ephA4 is expressed in r3 and r5 in the wild-type embryo (Fig. 3G), but its r5 expression domain is severely reduced in val⁻ embryos (Fig. 3H). It has previously been shown that ephA4 expression in r5 is under the direct transcriptional control of krx20 (Theil et al., 1998). Since krx20 expression is also downregulated in val⁻ embryos (Moens et al., 1998), the effect of Val on ephA4 expression is likely to be indirect.

The loss of alternating territories of receptor-expressing and ligand-expressing cells in the caudal hindbrain of val⁻ embryos correlates with the loss of some of the rhombomere boundaries associated with this mutant phenotype, and suggests that Eph signalling functions downstream of val in boundary formation in the caudal hindbrain.

Genetic mosaic experiments show that val⁻ cells cannot contribute to r5/6 of wild-type hindbrains and that wild-type cells do not contribute normally to rX of val⁻ embryos (Moens et al., 1996). We propose that spatial disruption of the normal signalling interfaces between ephB4a-and ephrin-B2a-expression domains is responsible for the loss of morphological boundaries in val⁻ embryos and for the characteristic cell-sorting behaviours observed in val⁻↔wild-type genetic mosaics.

Wild-type donor cells that form clumps in rX of val mutant host embryos autonomously express krx20 (when located at the rostral end of rX) and val, genes appropriate for cells in the equivalent region of a wild-type embryo (Moens et al., 1996; Moens et al., 1998). In situ hybridisations performed on wild-type→val⁻ genetic mosaic embryos show that ephB4a is also autonomously expressed by the clumps of wild-type donor cells present in rX (Fig. 4A,B). Thus, when wild-type cells are transplanted into rX of val⁻ hosts, a new Eph receptor/ephrin interface is established between the transplanted ephB4a-positive cells and the surrounding host cells which ectopically express ephrin-B2a (see Fig. 3E, F). This ectopic interaction between EphB4a and Ephrin-B2a may explain the repulsion of wild-type donor cells from the mutant host rX environment, resulting in their aggregation into clumps.

To assess whether a boundary forms at ectopic EphB4a/ Ephrin-B2a interfaces established in wild-type→val⁻ mosaic embryos, we looked at actin localisation. Wild-type embryos stained with fluorescent phalloidin (which visualises filamentous actin) show an accumulation of actin at the rhombomere boundaries between approximately the 12-somite and 18-somite stages (Fig. 4C). This actin accumulation is transient and is not synchronous for all boundaries. Phalloidin staining of mosaic embryos demonstrates an accumulation of actin at the interface between the wild-type cell clumps and surrounding host cells (Fig. 4D-F). This observation indicates that a boundary is established at the interface between ephB4a-positive cells (clumped wild-type donor cells) and ephrin-B2a-positive cells (the surrounding host rX cells).

Our data indicate that in both wild-type and genetic mosaic embryos, cell repulsion and boundary formation occur when ephB4a-expressing (receptor-positive) cells are apposed to ephrin-B2a-expressing (ligand-positive) cells.

When val⁻ cells are transplanted into wild-type embryos, we predict that Eph signalling will be activated inappropriately, since ephrin-B2a will be expressed on the val⁻ donor cells in r5/6, and ephB4a will be expressed in r5/6 of the wild-type host hindbrain. This activation may underlie the repulsion of val⁻mutant cells from wild-type r5/6. To test whether the loss of ephB4a expression in val⁻ donor cells contributes to their repulsion from r5/6, we reconstituted ephB4a expression in val⁻ donor embryos and analysed donor cell distribution in wild-type host hindbrains.

For these experiments, we injected mRNA encoding ephB4a into val⁻ donor embryos and green fluorescent protein (GFP), translated from co-injected RNA, was used as a tracer to identify cells ectopically expressing ephB4a. Co-localisation of protein products from these co-injected mRNAs indicates that the majority of GFP-expressing cells also express ephB4a (data not shown). To test the longevity of ectopic EphB4a protein, ephB4a mRNA-injected donor embryos were hybridised to the ephrin-B2a-AP affinity probe. Ectopic EphB4a protein could be detected up to and beyond those stages at which rhombomere boundaries are forming (data not shown). To control for nonspecific effects of mRNA overexpression on donor cell behaviour, we injected val⁻ donors with GFP mRNA alone. In the vast majority of cases, GFP-overexpressing val⁻ donor cells did not contribute to r5/6 of wild-type hosts (Fig. 5A, Table 1). The transplanted cells appeared healthy and behaved in an identical fashion to val⁻ donor cells injected with lineage tracers (Moens et al., 1996). val⁻ donor cells do frequently contribute to the r5/6 boundary region (Fig. 5A and Moens et al., 1996), indicating that cells in the boundary region are phenotypically distinct from cells in the body of the rhombomere.

When we overexpressed ephB4a and GFP in val⁻ donor embryos and transplanted cells from such individuals into wild-type hosts, we found that donor cells were often present in r5 and/or r6 (Fig. 5B, Table 1). These donor cells contributed to r6 about twice as frequently as to r5, and often exhibited abnormal morphology and unilateral distribution (Fig. 5B; see Discussion). In 50% of the wild-type hosts of mutant donors in which there was a good spread of donor cells from hindbrain to spinal cord, the ephB4a/GFP-overexpressing val⁻ donor cells were present in r5 and/or r6, the region from which GFP-overexpressing or lineage-labelled val⁻ donor cells are excluded. Therefore, overexpression of ephB4a can at least partially rescue the inability of val⁻ donor cells to contribute to r5/6 of a wild-type host, indicating that EphB4a is a downstream effector of val, whose function is required for proper cell sorting in the caudal hindbrain.

Since overexpression of ephB4a by val⁻ donor cells resulted in an incomplete rescue of the val⁻→wild-type mosaic phenotype (see Discussion), we decided to use a complementary approach to test more rigorously the role of ectopic EphB signalling in the repulsion of val⁻ donor cells from wild-type r5/6. To do this, we used a soluble Ephrin-B2a construct (Durbin et al., 1998), which disrupts bi-directional signalling through all EphB molecules. Soluble B-class ephrins have been shown to block bi-directional signalling in other systems by preventing receptor clustering and competitively inhibiting endogenous ligand binding (Krull et al., 1997; Durbin et al., 1998). We injected mRNA encoding soluble Ephrin-B2a into val⁻ donor embryos, transplanted cells into wild-type hosts and analysed donor cell distribution in wild-type host hindbrains. In these experiments, GFP translated from co-injected RNA was used as a tracer to follow donor cells in living embryos. RNA for β-galactosidase was also injected in some experiments to enable localisation of donor cells in fixed embryos.

When soluble Ephrin-B2a and GFP were overexpressed in val⁻ donor embryos and cells from such individuals were transplanted into wild-type hosts, donor cells were often present in r5 and/or r6 and displayed a near-normal morphology (Fig. 5C,D; Table 1). In 62.5% of the wild-type hosts of mutant donors in which there was a good spread of donor cells from hindbrain to spinal cord, the soluble Ephrin-B2a-positive/GFP-positive val⁻ donor cells were present in r5 and/or r6, the region from which val⁻ donor cells are normally excluded. Therefore, blocking EphB signalling in val⁻→wild-type mosaics can rescue the inability of val⁻ donor cells to contribute to r5/6 of a wild-type host. This suggests that signalling between Ephrin-B2a and its cognate receptors (e.g. EphB4a, EphA4) plays a role in cell sorting in the val⁻→wild-type genetic mosaic embryos, and, more generally, may be required for establishing and maintaining segmental integrity during normal development.

Our observations of gene expression and cell behaviour in val⁻ embryos and in genetic mosaics indicate that the spatial control of ephB4a and ephrin-B2a expression by Val is crucial for cell sorting and boundary formation in the caudal hindbrain (see Fig. 6). We present several lines of evidence to support this. Firstly, EphB4a and Ephrin-B2a can bind together in vivo and are expressed in complementary domains with boundaries forming at the interface between these domains (Fig. 6A). Second, Val is required to induce complementary expression of ephB4a and ephrin-B2a, such that in val⁻ embryos, there is a loss of Eph/ephrin interfaces and a corresponding loss of boundaries (Fig. 6B). Third, an ectopic EphB/ephrin-B interface correlates with ectopic boundary formation (i.e. when wild-type cells populate rX of a val⁻ embryo; Fig. 6C). Fourth, the val⁻→wild-type mosaic phenotype (expulsion of donor cells from r5/6, Fig. 6D) is partially rescued by overexpressing ephB4a in val⁻ donors (Fig. 6E). Finally, blocking all bi-directional EphB signalling results in a more complete rescue, allowing val⁻ cells to contribute to wild-type r5/6 in most cases (Fig. 6F and Table 1). These results suggest that the repulsion of val⁻ cells from wild-type r5/6 is mediated by signalling between EphB molecules.

The complementary expression of ephB4a and ephrin-B2a arises after the r5/6 expression domain of val is established, but before the first appearance of morphological boundaries. Our results show that the receptor EphB4a requires Val for its expression in r5/6, whilst Val inhibits expression of the ligand Ephrin-B2a in these rhombomeres, either directly or indirectly. In some val⁻ embryos a very low level of ephB4a expression is detectable at early stages (e.g. four-somite stage), but is not detected later (e.g. ten-somite stage). This transient, low-level expression of ephB4a seen in rX of some val⁻ embryos may be a result of transcriptional activation by early-expressed krx20 whose expression in the caudal hindbrain is initiated normally in val⁻ embryos before being lost (Moens et al., 1996). Even though some ephB4a expression may be retained in some val⁻ embryos, the high levels of ephrin-B2a in the same cells may suppress any activity of ephB4a co-expressed transiently at a low level (see below, and Bohme et al., 1996).

In the absence of Val, the consequent loss of Eph receptor/ephrin expression domain interfaces correlates with a failure of the r4/5 and r6/7 boundaries to form. The r5/6 boundary also fails to form in val⁻embryos, probably because r5 and r6 identities are not specified. However, we have not addressed formation of the r5/6 boundary in this study, since it is not normally established by interactions between EphB4a and Ephrin-B2a. Further evidence supporting a role for EphB4a↔Ephrin-B2a signalling in boundary formation comes from the wild-type→val⁻ genetic mosaics, in which an ectopic boundary forms at the interface between the ephB4a-positive wild-type donor cells and the ephrin-B2a-positive host rX cells. This result suggests that introduction of an Eph/ephrin interface is sufficient to establish boundary features, even in the normally unsegmented caudal hindbrain of a val⁻ embryo.

Alterations in Eph signalling can also explain the behaviour of cells in val⁻→wild-type mosaics. When val⁻ donor cells are transplanted into a wild-type host hindbrain, those donor cells that find themselves in r5/6 will not be able to synthesise a functional Val transcription factor. The absence of Val in these cells means that they will also not activate ephB4a expression and will not repress ephrin-B2a expression. The ephrin-B2a-expressing donor cells will be surrounded by ‘hostile’ ephB4a-positive host cells of r5/6 and may therefore be repelled into the adjacent ‘permissive’ ephrin-B2a-positive host rhombomeres 4 and 7. Donor cells tend to pile-up at the edges of r5 and r6 (e.g. see Fig. 5A), suggesting that the cells move no further than to escape from rhombomeres 5 and 6. val⁻ donor cells are also often present at the wild-type r5/6 boundary, suggesting a difference in repulsive/adhesive properties between the body of the rhombomere and the boundary region. Indeed the unique extracellular matrix composition of rhombomere boundaries (as seen in the chick; Heyman et al., 1995) may lead to a restriction to the movement of repelled val⁻ donor cells in boundary regions.

In contrast to the val⁻→wild-type mosaics, wild-type donor cells in rX of a val⁻ host are not repelled to adjacent segments, indeed there are no adjacent ‘permissive’ (ephB4a-positive) segments, perhaps explaining why the wild-type donor cells aggregate into clumps in rX. Alternatively, the different cell behaviours of donor cells in the val⁻→wild-type mosaics (repulsion) and wild-type→val⁻ mosaics (clumping) may be a result of differences in the downstream signalling pathways activated by B-class ephrins and Eph receptors, respectively.

When val⁻ donors were injected with mRNA for ephB4a, 50% of hosts analysed showed donor cells in r5 and/or r6, suggesting that cell-autonomous expression of ephB4a increases the probability that cells will contribute to r5/6. However, the ephB4a-positive val⁻ donor cells in wild-type r5/6 sometimes form clumps and are often unilateral (see Fig. 5B), in contrast to the appearance of such cells elsewhere in the host hindbrain, where they have a characteristic elongated morphology and form bilaterally symmetric clones. The unilateral appearance of the rescued cells suggests that some degree of repulsion may be taking place between the rescued cells and their wild-type neighbours. One possible explanation for this is that by the stages at which the host embryos were analysed (18 somites to prim-5), ectopic ephB4a expression is decreasing, so val⁻ donor cells may belatedly be repelled from host r5/6.

It should be noted that the ephB4a-overexpressing val⁻ donor cells do not necessarily reflect the wild-type situation, since we have not ruled out the possibility that these cells also express ephrin-B2a. Evidence from in vitro co-transfection assays suggests that activity of B-class Eph receptors is reduced when ephrin B ligands are co-expressed in the same cell (Bohme et al., 1996). This raises the possibility that signalling by overexpressed EphB4a may be compromised in the val⁻ donor cells by co-expression of Ephrin-B2a. This co-expression may explain why there is only partial, and abnormal, contribution of these cells to r5 and r6. It may also help explain another curious observation that the ephB4a-expressing val⁻ cells appear to contribute normally to r4 and r7, both of which express ephrin-B2a. Again, we suggest that co-expression of ligand and receptor may alter the Eph signalling-mediated events within these cells.

It is interesting to note that ephB4a-overexpressing val⁻ donor cells incorporate into wild-type host r6 approximately twice as often as into r5. The ephA4 receptor, which also binds ephrin-B2a (Durbin et al., 1998) is expressed with a similar timecourse to ephB4a in r5, but is not expressed in r6. Similar to ephB4a, expression of ephA4 is downregulated in rX of val⁻ embryos. Perhaps EphB4a and EphA4 interact within the cells of r5 such that signalling through both receptors is required for incorporation of cells into r5. Overexpression of both receptors together may increase the probability that val⁻ donor cells incorporate into r5. The possibility of a role for other Eph family members is supported by the observation that blocking all bi-directional ephrin-B-mediated signalling produces a more efficient rescue of the val⁻→wild-type mosaic phenotype than simply upregulating expression of ephB4a.

Blocking all bi-directional EphB signalling results in a more complete rescue of the val⁻→wild-type mosaic phenotype than reconstituting ephB4a expression, confirming the importance of the EphB signalling pathway in mediating expulsion of val⁻ cells from wild-type r5/6. Using soluble Ephrin-B2a to prevent interactions between Ephrin-B2a and receptors such as EphB4a and EphA4 (Krull et al., 1997), the repulsion of val⁻ donor cells from wild-type r5/6 can be overcome, allowing the cells to contribute to this region. The improved efficiency of this rescue, compared with that achieved with ephB4a overexpression, suggests a possible involvement for additional family members such as ephA4. It will be interesting to see which cell types these rescued donor cells are able to give rise to, as this may provide a clue as to the importance of segmentation in hindbrain neuronal patterning.

Our results implicating Eph signalling in the expulsion of val⁻ donor cells from r5/6 of wild-type host hindbrains suggest that similar repulsive interactions in the wild-type hindbrain function to separate cells from adjacent rhombomeres. The cell movements seen in val⁻→wild-type mosaics take place over relatively larger distances than the movements that occur during normal boundary formation, nevertheless, they appear to be governed by the same molecular interactions. Supporting a role for Eph-mediated repulsion in rhombomere-dependent cell sorting, recent work using a co-culture system of zebrafish cells expressing different Eph receptor and ephrin constructs has shown that bi-directional Eph signalling results in restriction of cell intermingling and loss of intercellular communication (Mellitzer et al., 1999). Interactions between activated Eph molecules and components of the cytoskeleton are likely to mediate the repulsive response (reviewed in Bruckner and Klein, 1998).

Eph molecules are important downstream targets in the Val pathway, providing a mechanistic link between position along the anteroposterior axis and the sorting of cells into segments. However, Val function is required not only for proper boundary formation in the caudal hindbrain, but also for expansion of a precursor region of one segment in length into definitive r5/6. Since no direct link between Eph signalling and cell division has been demonstrated, we suspect that Eph molecules do not play a role in this process. Segmental identity is likely to be imparted by Hox genes (e.g. Bell et al., 1999), and so targets of Val (other than Eph molecules; such as Hox genes), may be responsible for growth of the protosegment rX into r5/6 and for acquisition of appropriate regional identity. Indeed, Kreisler/ Val has been shown to regulate expression of hoxb3 in both the mouse (Manzanares et al., 1997) and the zebrafish (Prince et al., 1998).

Grafting experiments in chick suggest that an odd/even segmental alternation of cell-surface properties normally restricts the mixing of cells from adjacent rhombomeres and show that juxtaposition of odd and even rhombomeres is required for boundary formation (Guthrie and Lumsden, 1991; Guthrie et al., 1993). Could Eph molecules provide the basis for this odd/even alternation of cell-surface properties shown experimentally in the chick? In mouse and Xenopus, there is alternating expression of an interacting Eph receptor-ligand pair, with ephrin-B2 expression in r2/4/6 (Bergemann et al., 1995; Smith et al., 1997) and ephA4 expression in r3/5 (Gilardi-Hebenstreit et al., 1992; Hirano et al., 1998; Nieto et al., 1992; Winning and Sargent, 1994; Xu et al., 1995). Thus, Ephrin-B2 and EphA4 are good candidates for mediating cell sorting between odd and even rhombomeres, at least in mouse and Xenopus.

We have shown that zebrafish Ephrin-B2a is expressed in r1/4/7, and an interacting receptor, EphB4a, is expressed in double-rhombomere domains r2/3 and r5/6. This interacting ligand-receptor pair do not, therefore, observe a simple single-rhombomere-alternation of expression domains. Zebrafish EphA4 (which also interacts with Ephrin-B2a) is, like its counterpart in mouse and Xenopus, expressed in odd-numbered rhombomeres r3 and r5 (Xu et al., 1995), partially overlapping with ephB4a expression domains. There are also additional, yet to be characterised, members of the zebrafish Eph family expressed in rhombomere-restricted patterns. For instance, a second zebrafish ephrin-B2 orthologue, ephrin-B2b, is expressed in r1 and r4 (L. D., unpublished observations). This complexity may in part be the consequence of a postulated genome duplication in the lineage leading to teleosts (Amores et al., 1998; Meyer and Malaga-Trillo, 1999). One might predict that an ephrin able to interact with EphA4 but not with EphB4a might be expressed in r2 and r6, and be required for formation of the r2/3 and r5/6 boundaries. Thus, the presence of additional Eph family members in the zebrafish may result in increased levels of complexity or redundancy in the genetic regulation of hindbrain patterning, compared with other vertebrates.

In this study, we have addressed the role of the Val bzip transcription factor in establishing segment boundaries in the hindbrain. We have shown that Val regulates rhombomere-restricted complementary expression domains of an EphB receptor-ligand pair in the developing hindbrain. This receptor-ligand pair interact repulsively at the interfaces of their expression domains, leading to cell-sorting and subsequent formation of rhombomere boundaries.

We thank Jose Campos-Ortega for supplying a cDNA library, and Tom Schilling for helpful discussions and comments on the manuscript. This study was supported by the HFSP and by grants from the BBSRC and Wellcome Trust to N. H. and S. W., NIH grants NS17963 and 22486 to C. K., and NIH grant HD37909-01 to C. M. J. C. received a Bogue Research Fellowship.