Zebrafish hox genes: expression in the hindbrain region of wild-type and mutants of the segmentation gene, valentino

Studies on a range of different vertebrates have revealed that the rostrocaudal axis of the developing hindbrain is transiently subdivided into a series of reiterated segments termed rhom-bomeres (reviewed by Guthrie, 1996; Lumsden and Krumlauf, 1996) which play a pivotal role in organizing the structure and function of the vertebrate head. Cellular studies in the chick have revealed that rhombomeres are lineage restricted com-partments (Fraser et al., 1990; Birgbauer and Fraser, 1994) which correspond to the segmental organization of reticular neurons, branchiomotor nerves and sensory ganglia (Lumsden, 1990). Rhombomeric organization also correlates with the formation and migration of the cranial neural crest (Lumsden et al., 1991; Schilling and Kimmel, 1994; Trainor et al., 1994). Comparative studies have suggested that these general proper-ties of hindbrain organization are conserved within the ver-tebrates (Gilland and Baker, 1993).

There appears to be a fundamental two-segment periodicity\ along the rostrocaudal axis of the developing hindbrain which influences both the formation and the patterning of the rhom-bomeres. Studies in chick have shown that rhombomere boundary formation requires alternating rhombomeric states; when odd and even-numbered rhombomere cells are juxta-posed, a third state is achieved leading to boundary formation (Guthrie and Lumsden, 1991; Guthrie et al., 1993). This two-segment periodicity is reflected by neuronal organization (Lumsden and Keynes, 1989), and by the expression patterns of developmental control genes. These genes include receptors, such as eph-like receptor tyrosine kinases, their ligands, and transcription factors such as the zinc-finger gene Krox-20 and the Hox genes (reviewed by Lumsden and Krumlauf, 1996). In mouse and chick, the anterior limits of Hox gene expression generally define two segment regions out of register with those defined by the branchiomotor neuron pools (Wilkinson et al., 1989b; Hunt et al., 1991a,b). Seven rhombomeres have been described for the zebrafish, each sharing a common internal structure (Hanneman et al., 1988; Trevarrow et al., 1990). The earliest neurons to differentiate, the primary reticulospinal neurons that are born near the end of gastrulation, form a ladder-like array that corresponds to locations at the centres of the indi-vidual rhombomeres (Mendelson, 1986a,b; Metcalfe et al., 1986; Hanneman et al., 1988). The cranial motoneurons derive from pairs of rhombomeres, for example, the trigeminal (Vth nerve) derives from r2 and r3, the facial (VIIth nerve) from r6 and r7, and the abducens (VIth nerve) from r5 and r6 (Chan-drasekhar et al., 1997; Trevarrow et al., 1990; Gilland and Baker, 1993). A similar pairwise derivation of motoneurons has been described for the mouse, although, interestingly, the specific rhombomeres in which individual motoneurons differ-entiate vary between species (Gilland and Baker, 1993).

The Hox genes are implicated in conferring segmental identity, both by homology to the Drosophila homeotic genes and from functional approaches in vertebrates. For example, mice with targeted disruptions of the Hoxb-1 gene show changes in r4 identity (Goddard et al., 1996; Studer et al., 1996). Disruptions of the Hoxa-2 gene, which is normally expressed in r4-derived neural crest, lead to abnormal pattern-ing of the crest-derived second branchial arch cartilage elements (Gendron-Maguire et al., 1993; Rijli et al., 1993). When Hoxa-1 is overexpressed, either in transgenic mice (Zhang et al., 1994) or by ectopic expression in zebrafish (Alexandre et al., 1996), the phenotype is reminiscent of classic homeosis: r2 takes on aspects of r4 identity. However, the phe-notypes of Hoxa-1 disruptions hint at the possibility that some vertebrate Hox genes may not only play a role in conferring segmental identity but also in segmentation per se. The Hoxa-1 gene has an early anterior expression limit at the boundary between rhombomeres 3 and 4 (r3/r4 boundary) but expression regresses rapidly out of the hindbrain. Two separate mutations for this gene have been generated (Carpenter et al., 1993; Mark et al., 1993; reviewed by Wright, 1993) both lead to severe dis-ruption of the hindbrain between r4 and r7, with r5 being reduced or deleted. This potential dual role of the Hox genes, in setting up segmentation, as well as in conferring segmental identity, may explain the apparent paucity of candidate seg-mentation genes. Only two other candidate segmentation genes have been described, the zinc-finger transcription factor Krox-20, which is expressed in presumptive r3 and r5 (Wilkinson et al., 1989a), and the kreisler (kr) gene which encodes a bZIP transcription factor (Cordes and Barsh, 1994). Targeted dis-ruptions of Krox-20 lead to loss of r3 and r5 (Schneider-Maunoury et al., 1993; Swiatek and Gridley, 1993), and Krox-20 directly regulates transcription of Hoxa-2 and Hoxb-2 (Sham et al., 1993; Nonchev et al., 1996). In kr mutant mice, the neural tube appears unsegmented caudal of the r3/r4 boundary (Frohman et al., 1993; McKay et al., 1994).

Normal segmentation of the hindbrain is similarly disrupted in zebrafish mutant for valentino (val) (Moens et al., 1996); val⁻ embryos also lack visible hindbrain segmentation caudal to the r3/r4 boundary. It has recently been demonstrated that val is the zebrafish homologue of mouse kr (Moens et al., 1998). The val gene product is expressed throughout r5 and r6 and mosaic analysis has suggested that val is required cell-autonomously for normal subdivision of r5 and r6 from a hypo-thetical common precursor region (Moens et al., 1996). In the absence of functional val gene product, this common precursor region is maintained as rX, a region of one rhombomere’s length that lies between, yet fails to form boundaries with, r4 and r7. Analysis of neurons and neuronal cell types within the mutant hindbrain has shown that rX has a unique identity whilst incorporating some aspects of the identities of both r5 and r6. For example, the primary reticulospinal neurons MiD2 and MiD3, characteristic of r5 and r6 respectively, are both present in rX in the normal anteroposterior order, although the spacing between these neurons is reduced. However, the later-differentiating r5- and r6-specific cells of the abducens (VI) cranial nerve are absent in val⁻ embryos. Furthermore, mosaic analysis has shown that wild-type cells are unable to contribute in a normal fashion to rX, revealing the distinct character of this region (Moens et al., 1996).

In this study we have investigated the processes of hindbrain segmentation and the acquisition of segmental identity by analyzing the expression of zebrafish hox genes in the devel-oping hindbrain. To date, detailed expression analysis of only one such gene, hoxa1, has been reported (Alexandre et al., 1996). We have carried out expression analyses for a further 7 hox genes; our results allow a comparative approach by corre-lation with available data from the mouse, human, chick, and Xenopus systems and we find that the zebrafish hox genes share many expression properties with their tetrapod homologues, although some details of temporal and spatial pattern do differ. In addition, we have investigated hox gene expression in mutants of the hindbrain segmentation gene val, and have related our results to those obtained with mutants of its murine homologue kr, allowing a comparison of the patterning mech-anisms at work in the zebrafish and mouse hindbrains. We have used our results to address the apparent conundrum that some aspects of segmental identity are maintained in the mutants (i.e. the reticulospinal neurons are normally patterned) despite lack of overt segmentation in the caudal hindbrain. Our detailed analysis of gene expression at a variety of developmental stages in mutant and wild-type embryos has helped to shed new light on the detailed temporal changes in cell identity occurring during rhombomere formation and patterning. Our results suggest that the val gene product functions both in the seg-mentation process, by subdividing two rhombomeres from their common precursor, and in aspects of the acquisition of segmental identity, by regulating hox gene expression.

Zebrafish (Danio rerio) embryos were obtained from natural spawnings and staged as described by Kimmel et al. (1995). val^b337/val^b337 embryos were produced by crossing val^b337/val⁺ fish together, yielding wild-type and mutant embryos in a 3:1 ratio. Com-parisons of the expression of hox genes in wild-type, heterozygous val^+/− mutants and homozygous val^−/− mutants showed that the het-erozygotes were indistinguishable from wild types.

RACE-PCR was carried out as previously described (Frohman, 1993). cDNA was reverse transcribed, using the Gibco-BRL Superscript kit according to manufacturer’s instructions, from 24-hour zebrafish embryo RNA prepared as described by Chomczymski and Sacchi (1987). PCR reaction conditions were : 1 cycle at 94°C 2 minutes; 35 cycles at 94°C 1 minute, 45-50°C 2 minutes, 72°C 2.5 minutes; 1 cycle at 72°C 10 minutes. PCR products were cloned into the Promega pGEM-T cloning vector, or the EcoRV site of pBluescript SK(−) (Stratagene) according to manufacturer’s instructions.

Primers were designed based on published sequences of homeobox cDNAs (Njølstad et al., 1988a; Rundstadler and Kocher, 1991; Misof et al., 1996).

Primer sequences (names previously given to these genes are indicated in parentheses): hoxb1 (Misof et al., Z-3): 5′ GCAAGTATCTGACGCGAGCAC and 5′ CACGGCGTGTG-GAGAATTGCTG hoxa2 (Misof et al., Z-75): 5′ AACAAG/ATAAT/CCTG/TTGC/TCGGCG/C and 5′ GGC-CAAGGCGTGTGGAAATC hoxb2 (Misof et al., Z-151): 5′ AACAAG/ATAAT/CCTG/TTGC/TCGGCG/C and 5′ GGCGCAGGCGCGTTGAAATT hoxb3 (Misof et al., Z-56): 5′ TTCAACCGA/CTACCTGTGC/TCG and 5′ GGCCGAG-GCGTGTGGAAATG hoxd3 (Misof et al., Z-92): 5′ TTCAACCGA/CTACCTGTGC/TCG and 5′ GCCCCAGAA-GAGTGGAGATG hoxx4 (Rundstadler and Kocher, zf26): 5′ AGAGGTCTCGCACCGCCTAC and 5′ CCAGCAGGCTCTTGAGCTTG hoxb4 (Misof et al., Z-17; Njølstad et al., zf13) 5′ TTACAACCGCTATCTGACCCG and 5′ CAGAA-GAAGGGTGGAAATCGC Clones were screened by sequencing of double stranded templates (Sequenase, US Biochemicals Inc.) from forward and reverse primers, the most 5′ 400 bp (at minimum), of coding sequence and 3′ untrans-lated region, were then sequenced in both directions using internal oligonucleotide primers. Sequence analyses and comparisons were performed using the Wisconsin genetics GCG software package. Phy-logenetic tree analysis was performed with the Megalign module of the Laser gene programme (DNASTAR, Inc.). All the sequences described are available in the EMBL database under accession numbers Y13944-13950. The sizes (in base pairs) of the cDNAs obtained for each gene are indicated as follows. hoxb1, 1100; hoxa2, 1000; hoxb2, 900; hoxb3, 600; hoxd3, 800; hoxx4, 900; hoxb4, 1600.

In situ hybridizations were performed essentially as previously described (Thisse et al., 1993) with the following modifications. Entire 3′RACE-PCR derived cDNAs were used as templates to syn-thesize antisense riboprobes with T7, T3 or SP6 RNA polymerase (Promega) as appropriate; probes were not hydrolyzed. Proteinase K treatment (10 µg/ml in PBT) time was reduced to approximately 30 seconds per somite; the proteinase K reaction was stopped by refixa-tion in 4% paraformaldehyde in PBS.

Two colour in situ hybridizations were performed essentially as described by Hauptmann and Gerster (1994); Jowett and Lettice (1994). Briefly, the antisense riboprobe to one gene was labeled with digoxigenin-UTP (Boehringer Mannheim) and the second probe labeled with fluorescein-UTP (Boehringer Mannheim), the probes were then probe corresponds to 490bp of 3′ untranslated region (Moens et al., submitted).

We have obtained cDNAs of 7 zebrafish hox genes, from paralogue groups 1 through 4, which are expressed within the developing hindbrain. The 3′ RACE-PCR technique (Frohman, 1993) was used to amplify cDNAs based on published sequences of homeobox regions (Materials and Methods). Nested primers were targeted to the 5′ half of the homeobox domain and RT-PCR was used to amplify the region between the specific primers and the 3′ end of the message. We have used this approach to obtain relatively long cDNAs (600-1600 bp; see Materials and Methods) facilitating further characteri-zation of the genes and the synthesis of specific riboprobes. We have identified individual genes based upon sequence compar-ison, and in the majority of cases by locating the genes to specific hox clusters (see below). Furthermore, similarities between the expression patterns of our zebrafish genes and the equivalent murine and chick genes are consistent with our assignments. The sequences of the homeoboxes of the 7 isolated genes are shown in Fig. 1.

We have assigned zebrafish hoxb1, hoxb2, hoxb3 and hoxb4 to the hoxb cluster based upon their presence on a single YAC clone (A. Fritz, unpublished; YAC library, L. Zon., personal communication) which also contains the previously identified genes hoxb5 and hoxb6 (previously designated ZF21 and ZF22 respectively; Njølstad et al., 1988b,c, 1990). Presence of the genes on the YAC was determined by PCR with primers from within the 3′ untranslated sequence of each gene (data not shown). Consistent with these results the zebrafish hoxb2 and hoxb4 genes are absent in a mutant with a large deletion encompassing the hoxb cluster (Andreas Fritz pers. comm; Fritz et al., 1996). The homeobox sequence of zebrafish hoxb1 has only a single amino acid difference to chick Hoxb-1, the zebrafish hoxb2 and hoxb3 homeobox sequences both show 100% identity to their mouse and human homologues. The hybridized to the embryo simultaneously. The digoxigenin-labeled probe was visual-ized with anti-digoxigenin alkaline phos-phatase (Boehringer Mannheim) reacted with NBT and BCIP to produce a blue colour; glycine treatment (0.1 M glycine-HCl pH 2.2, 0.1% Tween-20, 10 minutes at room temperature) was then used to remove anti-digoxigenin alkaline phosphatase conjugate before application of anti-FITC alkaline phosphatase conjugate (Boehringer Mannheim) and detection with Fast Red (Sigma). In situ hybridizations were stopped by washing in PBS, 20 mM EDTA. Double stained embryos were mounted in 80% glycerol in PBS, single stained embryos were dehydrated through a methanol series, cleared with benzyl benzoate and mounted in Permount. The zebrafish krox-20 riboprobe was as previously described (Oxtoby and Jowett, 1993), the valentino zebrafish hoxb4 gene is 100% identical at the DNA level to the previously cloned zf-13 gene (Njølstad et al., 1988a), this gene was previously incorrectly identified as hoxd4 based on sequence alignments (Misof et al., 1996). The homeobox of the zebrafish hoxb4 gene has just a single amino acid change compared to its mouse and human homologues.

The homeobox of the gene we assign as zebrafish hoxd3 has 100% amino acid identity with mouse and human Hoxd-3, and is linked to hoxd4 on hybrid chromosomes (Marc Ekker, personal communication; Ekker et al., 1996). The homeobox of the gene we identify as zebrafish hoxa2 shows 100% amino acid identity with mouse Hoxa-2 but also with Hoxb-2. We assign this gene to the a cluster because its identity with mouse Hoxa-2 is 67% over the entire 210 amino acids sequenced, compared to only 41% with human Hoxb-2 (no Hoxc-2 or Hoxd-2 genes have been reported from other species). In addition, this gene shows diagnostic features of the Hoxa-2 expression pattern (see below) and we already have a candidate hoxb2 gene. Finally, the last gene we analyzed, which we term zebrafish hoxx4, is 100% identical to the previously cloned zf-26 gene (Rundstadler and Kocher, 1991); this gene was previ-ously identified as hoxb4 based on sequence alignments (Misof et al., 1996; Rundstadler and Kocher, 1991), however we already have a clear example of a hoxb4 gene. The homeobox of hoxx4 shows 95% amino acid identity with both mouse and human Hoxa-4 and phylogenetic tree analysis suggests hoxx4 to be most closely related to the Hoxa-4 genes of other species, however we are unable to unambiguously assign this gene to the a cluster.

We analyzed the expression patterns of the 7 zebrafish hox genes by whole-mount in situ hybridization in normal zebrafish embryos at stages between 6 hours (50% epiboly) and 30 hours of development (Fig. 2). In many cases double in situ hybrid-izations are shown using krox-20 (the zebrafish homologue of murine Krox-20; Oxtoby and Jowett, 1993) as a second molecular marker to indicate the locations of rhombomeres 3 and 5. The expression patterns of the zebrafish hox genes show many properties in common with their murine homologues, consistent with our gene assignments.

The zebrafish hoxb1 gene shares the characteristic expression domain in rhombomere 4 of the developing hindbrain exhibited by the Hoxb-1 genes of mouse, chick, human, Xenopus and another teleost fish, the carp (Frohman et al., 1990; Sundin and Eichele, 1990; Vieille-Grosjean et al., 1997; Godsave et al., 1994, Stevens et al., 1996; Fig. 2A). The onset of zebrafish hoxb1 expression is during gastrulation, between 80% and 100% of epiboly with an anterior limit at approximately the level that will give rise to r4; this is similar to the situation in the carp (Stevens et al., 1996), but unlike the situation described for mouse and chick in which initial expression is confined to the posterior of the embryo with the expression domain then spreading forward until the anterior expression limit is reached (Frohman et al., 1990; Sundin and Eichele, 1990). Up to the 1 somite (1s) stage, zebrafish hoxb1 expression is continuous from pre-r4 posterior throughout the CNS (Fig. 2Ai). At the 3s stage, expression is down-regulated in the region immediately caudal to r4, this is clearly indicated by the reduced level of hoxb1 staining in r5 (Fig. 2Aii); r5 is easily identified by krox-20 expression (red signal). The regres-sion of hoxb1 expression (continued down-regulation moving posterior from r4), a phenomenon which has also been reported for other species, continues during the next few hours of devel-opment (Fig. 2Aiii, iv); concomitant with this process expression levels in r4 become significantly up-regulated. At the 10s stage, a low level of hoxb1 expression is visible from the level of r7 and posterior (Fig. 2Aiii), but by the 15s stage only the r4 expression domain is clearly visible in the CNS (data not shown; Fig. 2Aiv). Transverse sections through r4, at the 5s and 10s stages, reveal transient expression of hoxb1 in neural crest cells migrating out of r4 (data not shown).

The hoxb1 expression pattern shares some features with that of the paralagous gene hoxa1 (Alexandre et al., 1996). The onset of expression of hoxa1 is somewhat earlier, at 50% epiboly or germ ring stage. However, by tailbud stage the two genes share rather similar expression domains with distinct anterior borders approximating to the location of presumptive r4. As in other species hoxa1 expression rapidly regresses caudally, without leaving behind an r4-specific expression domain as seen for hoxb1. The differences in the expression patterns between the two genes are consistent with their assign-ment to the hox a and b clusters respectively.

The zebrafish hoxa2 gene has an anterior expression limit in the CNS similar to those of the Hoxa-2 genes of chick, human and mouse, namely at the r1/r2 boundary (Prince and Lumsden, 1994; Vieille-Grosjean et al., 1997; Frasch et al., 1995). The onset of detectable expression is at approximately the 2s stage (Fig. 2Bi), and commences as a single hindbrain ‘stripe’ in the region destined to give rise to r2 and r3. Similar to hoxb1, this early expression pattern differs from that observed in the chick, where expression commences in the most posterior part of the embryo and is gradually activated from posterior to anterior until the final r1/r2 boundary limit is reached (Prince and Lumsden, 1994). By the 10s stage, zebrafish hoxa2 expression has expanded posteriorly into r4 and r5, although this posterior expression domain is at significantly reduced levels compared to the r2+r3 domain (Fig. 2Biii). Note that by this stage the expression limits have become sharp, possibly reflecting the time of morphological rhombomere boundary formation. Com-mencing at approximately the 12s stage (data not shown), and clearly visible by the 20s stage (Fig. 2Biv), expression can be seen in neural crest cells migrating out into the second and third branchial arches (i.e. the hyoid and first gill arches). Expression is maintained in r2 through r5 at the 20s stage, but there are differences in the expression levels in individual rhombomeres (Fig. 2Biv); in particular the expression level in r4 is reduced relative to the other rhombomeres.

Zebrafish hoxb2 also shows a similar anterior expression limit to that of its murine and human homologues, at the r2/r3 boundary (Wilkinson et al., 1989b; Hunt et al., 1991a; Vieille-Grosjean et al., 1997; Fig. 2C). The onset of zebrafish hoxb2 expression is at approximately the 1s stage, and similar to hoxa2, expression commences at the anterior expression limit, in this case in pre-r3 (determined by double labeling with krox-20; data not shown). There is also expression in pre-r5 (Fig. 2Ci, ii); this colocalization of hoxb2 and krox-20 is suggestive of hoxb2 activation by the krox-20 gene product, as has been described for the mouse (Sham et al., 1993). By the 10s stage, expression is primarily localized to r3, r4 and r5 with highest expression levels towards the anterior (Fig. 2Ciii); in some specimens a region of low level expression was also noted from r7 and posterior. This pattern continues through the 20s stage (Fig. 2Civ), with spinal cord expression levels gradually decreasing, and is maintained to at least 30h of development. From about the 12s stage, expression is seen in neural crest cells migrating from r4 into the second arch. These cells are clearly visible at the 20s stage (Fig. 2Civ). There is also transient lower-level expression, between approximately the 12s and 18s stages, in neural crest cells migrating into the third branchial arch.

The zebrafish hoxb3 gene shares an expression limit at the r4/r5 boundary with the group 3 paralogue genes of mouse, human and chick (Wilkinson et al., 1989b; Hunt et al., 1991a; Vieille-Grosjean et al., 1997; Itasaki et al., 1996; Fig. 2D). The onset of hoxb3 expression is at the 1-2s stage, commencing posterior to the r5 krox-20 domain in both the CNS and the developing somites (data not shown). By the 3-4s stage, the expression domain expands forward to reach an anterior limit at the r4/r5 boundary (Fig. 2Di). At the 5s stage the most anterior expression domain in r5+r6 is at significantly higher levels than more posterior (Fig. 2Dii). This expression pattern is maintained at the 10s stage (Fig. 2Diii); neural crest cells just beginning to migrate out of the posterior limit of r6, towards the 3rd branchial arch, also express hoxb3. From the 20s stage, a low level expression is observed more anteriorly, throughout r4 (see Fig. 5C). Expression of the paralogue group 3 genes in mouse and chick has not been reported in r4 (Wilkinson et al., 1989b; Hunt et al., 1991a; Itasaki et al., 1996). At the 30 hour stage, expression within r4 is still visible although it is now confined to the ventral (basal) part of the rhombomere (Fig. 2Div).

The expression of the zebrafish hoxd3 gene also begins at the 1-2s stage, in this case expression commences in the most posterior part of the embryo (Fig. 2Ei), more in line with the expression patterns reported in tetrapod species. The expression domain expands rapidly toward the anterior to reach the r5/r6 boundary by the 3-4s stage (Fig. 2Eii). This anterior expression limit, which is one rhombomere more posterior than the limit reported for the mouse or human (Wilkinson et al., 1989b; Hunt et al., 1991a; Vieille-Grosjean et al., 1997), is very prominent and is maintained up to at least the 20s stage (Figs 2Eiii, 5I). However, by the 30 hour stage there is also limited expression within r5; this expression is confined to a small region in the basal plate of the rhombomere (Fig. 2Eiv). Low levels of hoxd3 expression were seen in neural crest cells migrating towards the 3rd branchial arch in some overstained specimens at the 20s stage (data not shown).

The murine Hoxa-4, Hoxb-4 and Hoxd-4 genes are reported to have anterior expression limits at the r6/r7 boundary, although there are differences in the expression levels of these genes within their anterior domains (Wilkinson et al., 1989b; Hunt et al., 1991a). The two zebrafish group 4 paralogues which we have analyzed share this anterior expression limit. The zebrafish hoxx4 gene has an expression onset at the 1s stage, the anterior limit of expression appears to be approximately within r7 by comparison to krox-20 (Fig. 2Fi). This gene continues to have an anterior expression limit lying within r7, but not reaching the r6/r7 boundary, up to at least the 30 hour stage (Fig. 2Fii-iv). The zebrafish hoxb4 gene has a similar time of onset and expression pattern at early stages (Fig. 2Gi, ii). By the 10s stage the anterior limit of hoxb4 expression is a little ahead of that of hoxx4 (compare Fig. 2Fiii and Giii). At approximately the 15s stage, hoxb4 reaches an anterior limit at the r6/r7 boundary (data not shown; Fig. 3G). This limit is maintained until at least 30 hours of development (Fig. 2Giv).

The spatial limits of hox gene expression are very prominent and robust; between the 5 and 20 somite stages of development they provide useful molecular markers for all the rhombomere boundaries from the r1/r2 boundary through to the r6/r7 boundary. We have made use of these markers to further analyze the val mutant, in which presumptive r5 and r6 do not subdivide but instead produce a region the length of a single rhombomere, rX. The rX region has some features of both r5 and r6, including characteristic reticulospinal neurons; never-theless, the identity of rX is unlike that of r5 and r6 based on mosaic analyses and marker gene expression (Moens et al., 1996). We have used hox gene expression to investigate both rhombomere identity and the nature of the interfaces between rX and the adjacent rhombomeres.

We have compared the expression domain of the val gene with those of the hox genes in normal embryos, and analyzed the expression patterns of the hox genes in val⁻ embryos at three developmental stages: the 4-6 somite (5s) stage, the 9-11 somite (10s) stage and the 18-22 somite (20s) stage. At the two early stages it is not possible to identify mutants morphologi-cally and therefore double in situ hybridizations were performed using krox-20 as a second marker; the r5 domain of krox-20 is radically reduced in val⁻ embryos facilitating iden-tification of mutant embryos.

Comparison with the expression pattern of krox-20 has shown that val expression is localized to r5 and r6 between its onset of expression, at the end of gastrulation, and the 20s stage; after this stage val expression decreases starting at the anterior and by 24 hours, the r5 domain has disappeared (Moens et al., submitted). At the 5s, 10s and 20s stages the r5+r6 expression domain of val precisely corresponds to the high level expression domain of hoxb3 (compare Fig. 3A and C with B and D). val is also expressed in the Mauthner neurons (Moens et al., 1998; Fig. 3B) and in a subset of the r6-derived neural crest (Fig. 3B). The neural crest expression domain is very restricted in comparison to that of hoxb3 (compare Figs 3B and D). The posterior limit of val expression at the r6/7 boundary corresponds to the mature anterior limit of hoxb4 expression (Fig. 3F).

The hoxb1 gene has a characteristic expression domain in r4 of the normal embryo. At the 5 somite stage, the r4 expression domain is clearly delimited but the borders of expression are not yet totally sharp (Fig. 4A), presumably reflecting the incomplete formation of morphological boundaries at this early stage. There are no distinguishable differences in hoxb1 expression between wild-type and val⁻ embryos at the 5s stage (compare Fig. 4A and B). By the 10s stage the r4 expression domain has developed sharp borders in normal embryos (see Fig. 2Aiii). In 10s val⁻ embryos the anterior expression limit of hoxb1 at the r3/r4 boundary has become sharp, but the posterior border of r4 expression is less clearly defined and patchy expression can be seen in the most rostral part of rX (Fig. 4C). By the 20s stage there is extensive expression of hoxb1 within rX, although this expression is only present in scattered cells (Fig. 4D). No hoxb1-expressing cells are present within the small, localized domain of krox-20 expression that remains within rostral/dorsal rX.

The hoxb4 gene represents a convenient marker for the r6/7 boundary, although this limit is not sharp until the 20s stage. Results of in situ hybridizations using hoxb1 and hoxb4 probes simultaneously are shown in Fig. 4E-J. At the 5s stage, pre-sumptive r5+r6 in both normal and mutant fish does not express hoxb1 (Fig. 4A,B), but hoxb1 is expressed at a significant level within r7 and posteriorly and thus it is not possible to differen-tiate between the posterior hoxb1 domain and the overlying hoxb4 domain. Nevertheless, at the 5s stage, no significant dif-ferences were noted in the size of the presumptive r5+r6 non-expressing domain between wild-type and mutant embryos (Fig. 4A and B). By the 10s stage, hoxb1 expression in r7 and posterior is drastically reduced (see Fig. 2Aiii). In 10s wild-type embryos there is a clearly demarcated non-expressing region, between the hoxb1 and hoxb4 expression domains, correspond-ing to r5+r6 (Fig. 4E). In 10s val⁻ embryos the size of this region is reduced in comparison to normal embryos (compare Fig. 4E and F), probably due to lack of separation of the presumptive r5+6 domain into definitive r5 and r6 territories (Moens et al., 1996). By the 20s stage there is no clearly demarcated non-expressing region between the expression domains of hoxb1 and hoxb4 in the val⁻ embryos (compare Fig. 4G, I with H, J; dorsal and lateral views). The expression within rX is at lower levels than in r4 or r7, again probably reflecting a ‘pepper and salt’ type distribution of expressing and non-expressing cells. There is a small region at the border of r4 and rX where hox expression is at a very low level; this region is at the correct location to correlate with the small rX krox-20 expression domain (compare Fig. 4D and H). The krox-20 expression domain in val⁻ embryos is rather variable in precise size and level, but is confined to the dorsal and anterior parts of rX; consistent with this observation the low level hox expression domain is also confined to this region in a lateral view of a val⁻ embryo (Fig. 4J). In situ hybrid-ization experiments using the hoxb4 probe alone, confirmed that expression of this gene spreads gradually into the rX region (data not shown), similar to our observations with hoxb1.

In normal embryos hoxb3 has an anterior expression limit at the r4/r5 boundary. There is expression throughout the caudal hindbrain and spinal cord but expression is at an elevated level within r5 and r6 (Fig. 2Diii), the two rhombomeres in which the val gene product is normally expressed (Fig. 3). In val⁻ embryos this elevated expression is lost at both the 5s stage (data not shown) and the 10s stage (compare Fig. 5A and B), although low level expression persists in rX. In addition, hoxb3 does not show a clear anterior limit of expression in val⁻ embryos, instead expression fades out at approximately the position of the r4/rX interface (Fig. 5B). Analysis of multiple embryos, and correla-tion with the position of the otic vesicle, suggests that hoxb3 expression may extend into the posterior part of r4 in val⁻ embryos, although low expression levels make this difficult to assess. At the 20s stage the high level r5+r6 expression domain is maintained in wild-type embryos, in addition to appearance of a weak expression domain in r4 (Fig. 5C,E). The r4 expression domain also appears in val⁻ embryos (Fig. 5D), but a level of expression equivalent to that in normal r5+r6 is not attained by rX (Fig. 5D,F). In approximately half the embryos examined there is a slight elevation of the hoxb3 expression level within rX (Fig. 5D), in the remaining half no difference was noted between the expression levels in rX and the more posterior neural tube (Fig. 5F). This lack of high level hoxb3 expression in those rhom-bomeres in which val is expressed suggests a role for the val gene product in up-regulating hoxb3 transcription. Neural crest expression of hoxb3 is maintained in val⁻ embryos, although inspection of multiple specimens suggests that there are fewer hoxb3-expressing cells in the mutant (compare Fig. 5C and D).

The hoxd3 gene is expressed with a clear anterior limit at the r5/r6 boundary in normal embryos (Fig. 5G – 10s stage; I – 20s stage). In val⁻ embryos there continues to be hoxd3 expression in rX (Fig. 5H,J). The anterior limit of hoxd3 expression is somewhat diffuse in val⁻ embryos at the 10s stage, similar to the situation for hoxb3, however, hoxd3 expression does not extend throughout the rX krox-20 domain and thus cannot extend into r4 (compare Figs 5B and H). By the 20s stage hoxd3 expression is limited to approximately the posterior half of rX (Fig. 5J), suggesting some cryptic AP pat-terning information remains along the length of rX.

The paralogue group 2 genes are expressed in cranial neural crest migrating anterior to the otic vesicle into the second branchial arch, and to a lesser extent posterior of the otic vesicle into the third branchial arch (Figs 2Biv, Civ). Second arch neural crest derives from the level of r4 and r5, third arch neural crest from r5, r6 and r7 levels (Schilling and Kimmel, 1994). In normal embryos hoxb2 expression in neural crest migrating into the 3rd arch is transient and by the 20s stage is no longer visible (Fig. 6A). By contrast, in 20s stage val⁻ embryos there continues to be hoxb2 expression in post-otic migrating neural crest (Fig. 6B). Similarly, hoxa2 continues to be expressed in wild-type 3rd arch neural crest at the 20s stage (Fig. 6C), but expression is at higher levels in val⁻ postotic neural crest cells (Fig. 6D). The otic vesicle of the val⁻ embryos is reduced in anteroposterior length, but maintains its approx-imate normal location. The postotic migrating neural crest in val⁻ embryos appears to derive from approximately the AP level of rX and may represent a misdirected neural crest pop-ulation that would normally migrate anterior of the otic vesicle, as has been shown for the mouse (Nonchev et al., 1996). This hypothesis leads to the prediction that the small domain of krox-20 expression in rX should be coincident with expression of hoxa2 at levels found in wild-type r5; unfortunately due to the relatively low level of hoxa2 expression in this region it has not been possible to show this conclusively. By the 20s stage, hoxa2 expression reaches high levels in normal r5, yet remains at low levels in rX (compare Fig. 6C and D), thus rX is again showing a rather similar expression profile to normal r4 at this later stage. The relative expression levels of hoxb2 in r4 and r5 change with developmental stage. At the 5s stage the expression level in r5 is significantly higher than that in r4 (Fig. 2Cii), but at the 10s and 20s stages this situation has reversed (Fig. 2Biii, Fig. 6A). In val⁻ embryos at the 5s stage, high level hoxb2 expression in rX is confined to the small domain of krox-20 expression (data not shown), again consistent with the idea that krox-20 may directly up-regulate hoxb2 expression as has been described for the mouse (Sham et al., 1993). At the 10s stage, hoxb2 expression in rX of val⁻ embryos once again seems confined to the domain of krox-20 expression (data not shown), by this stage r4 expression has been up-regulated, yet the posterior part of rX continues to show low level expression of hoxb2, and thus does not share r4 identity. By the 20s stage, hoxb2 expression levels in rX of val⁻ embryos have increased to approximately r4 levels, and are significantly higher than expression levels in normal r5 (compare Fig. 6A and B). Thus, once again by the 20s stage rX seems to have acquired some aspects of the identity of r4.

We have analyzed the expression of 7 different hox genes within the developing hindbrain of wild-type and val⁻ or alternatively a mis-specified crest population that is migrating in the normal direction but has taken on an identity more appropriate to r4-derived neural crest. Either interpreta-tion is consistent with the observation that 3rd arch cartilage elements in val⁻ embryos take on some characteristics of normal 2nd arch elements (Moens et al., 1998).

We also analyzed the expression of the paralogue group 2 genes in the hindbrain at 5s, 10s and 20s, in order to further evaluate the changing identity of rX. Specifically, we wished to assess to what extent rX identity is correctly specified early in development, and to what extent rX takes on aspects of r4 identity, as revealed by the gradually attained mis-expression of the r4 marker hoxb1 within rX (described above). Expression of hoxa2 in r4 and r5 is apparent at the 10s stage (Fig. 2Biii), but the expression levels are approximately equivalent between these two rhombomeres and thus hoxa2 does not usefully dif-ferentiate between r4 and r5 identity at this stage. In 10s stage val⁻ embryos, hoxa2 expression is at a slightly reduced level in rX in comparison to r4 or r5 expression levels in normal embryos (data not shown), perhaps reflecting lack of krox-20 expression in the mutant; krox-20 may be involved in up-regulating hoxa2 zebrafish. We find many similarities between the expression domains of the hox genes of zebrafish and other species, however, some distinct differences in timing, anterior limits and levels of expression do exist; these changes may be reflected in the observed interspecies modifications in neu-roanatomy. We find that many aspects of hox gene expression are relatively normal in val⁻ embryos at early stages before rhombomere boundary formation. This early phase of expression may correspond to the time at which the reticu-lospinal neurons acquire segmental identity, and hence this observation is consistent with the retention of normal pattern-ing of these early born neurons in val mutants. By contrast, later in development we observe inappropriate hox gene expression within rX, the region in val⁻ embryos that lies between, but fails to form boundaries with, r4 and r7. This late inappropriate expression may reflect lack of normal inter-rhombomeric interactions or inappropriate cell mixing along the anteroposterior (AP) extent of the val⁻ hindbrain.

The expression patterns of mouse and zebrafish Hox genes are summarized in Fig. 7. Despite broad similarities in expression patterns between these species, we have noted several discrep-ancies. For example, the onsets of zebrafish hox gene expression are generally relatively late, at the beginning of somitogenesis, compared to gastrulation stages in other species. In order to make useful cross-species comparisons regarding the timing of Hox gene function, the most useful parameter to consider is the time at which the anterior limit of expression is reached; differ-ences between the posterior expression domains may not play any functional role as in general Hox genes seem to function at or close to their anterior limits of expression (reviewed by McGinnis and Krumlauf, 1992). For the majority of the zebrafish genes we analyzed, expression appeared to commence with the final anterior limit already set. This is quite different to the general situation reported in tetrapod vertebrates where expression commences in the posterior of the embryo and gradually spreads forward until the final anterior limit is reached (Deschamps and Wijgerde, 1993). We would therefore suggest that the apparent late onset of zebrafish hox gene expression does not correlate with late function; for example, Hoxb-1 expression reaches its anterior limit in the murine neurectoderm at the onset of neurulation (Frohman et al., 1990), an approximately equiv-alent stage to that at which zebrafish hoxb1 is first expressed with its final anterior limit already set. It is possible that the rapid development of the zebrafish embryo precludes detection of an equally rapid anterior spread of hox expression; for example, for hoxd3, a gene which does show an anterior spread of expression, the anterior expression limit shifts from the posterior most part of the embryo to the hindbrain within one hour (Fig. 2D). Alter-natively, the general lack of preliminary posterior expression of zebrafish hox genes may reflect differences in the mechanisms used to activate Hox expression in zebrafish and mice.

We find several differences in relative expression levels in individual rhombomeres between fish and mouse (Fig. 7A). Murine Hoxa-2 and Hoxb-2 have high expression levels in r3 and r5, correlating with Krox-20 expression domains; indeed Krox-20 has been shown to directly activate Hoxa-2 and Hoxb-2 expression (Sham et al., 1993; Nonchev et al., 1996). In the zebrafish, hoxb2 expression is first localized to r3 and r5, shortly after the onset of krox-20 expression (Oxtoby and Jowett, 1993), suggesting similar mechanisms are at work. However, by the 10 somite (10s) stage, r5 expression of hoxb2 is at relatively low levels, and hoxa2 expression does not attain high levels in r5 until the 20s stage. For both of these genes, expression in the CNS posterior to r5 is at very low levels, yet in other species these genes are expressed caudally along the length of the hindbrain and spinal cord.

The major disparity we note between zebrafish and mouse Hox gene expression domains is for the paralogue group 3 genes (Fig. 7A). Up to at least the 20s stage, zebrafish hoxd3 has a clear anterior expression limit at the r5/r6 boundary, at later stages of development there is a small hoxd3 expression domain in r5, but this is confined to a small population of ventro-lateral cells (Fig. 2Eiv). Conversely, mouse Hoxd-3 has an anterior limit one rhombomere more anteriorly, at the r4/r5 boundary, although the r5 expression is at a lower level than more posterior expression (Fig. 7A). The hoxb3 gene of the zebrafish does have the same anterior limit of expression as its murine homologue, at the r4/r5 boundary (Fig. 7A), at least until late stages when a low-level r4 domain of expression appears. However, in the mouse there is an elevated expression domain of Hoxb-3 confined to r5 (Sham et al., 1993), but we find elevated hoxb3 expression in the zebrafish in both r5 and r6. Observed differ-ences in neuronal architecture between murine and zebrafish hindbrains may reflect the differences we observe in Hox gene expression. For example, in the zebrafish, the motor nuclei of the abducens (cranial nerve VI) lie in r5 and r6 (Gilland and Baker, 1993; Moens et al., 1996), whereas in mouse the abducens nucleus is confined to r5; it is interesting that in each instance the abducens nucleus colocalizes with the high level expression domain of hoxb3. In addition, the cell bodies of the facial nuclei (cranial nerve VII) lie within r6 and r7 of the zebrafish (Chandrasekhar et al., 1997), but within r4 and r5 of the mouse. Such differences in rhombomere identity presum-ably need to be patterned at the level of gene expression, and thus differences in Hox gene expression between mouse and fish at this rhombomeric level may not be unexpected. Indeed, the existence of such interspecies differences in Hox gene expression was accurately predicted from comparative studies of neuroanatomy (Gilland and Baker, 1993).

Expression of hoxb1 and hoxb4 changes dramatically during development of val⁻ embryos. At early stages, shortly after the normal onset of val expression (10 hours; Moens et al., 1998), the r4 hoxb1 expression domain is indistinguishable in wild-type and val⁻ embryos, as are the r7 hoxb4 and hoxx4 expression domains. This suggests no requirement for the val gene product in setting up r4 and r7 identity. However, between the 5s and 20s stages, expression of both hoxb1 and hoxb4 becomes progres-sively more apparent in rX, although remaining at lower levels than in the normal r4 and r7 territories, which could reflect a dis-tribution of expressing and non-expressing cells (summarized in Fig. 7). This observation can be explained in one of two ways, either cells within the rX territory have an incorrect positional specification that allows them to activate inappropriate markers, or hox gene expressing cells are crossing the r4/rX and r7/rX interfaces between the 5s and 20s stages.

The timing of val expression, and its location in r5 and r6 (Moens et al., 1998; Fig. 3), might suggest a role for the val gene product in down-regulating hoxb1 expression within these two rhombomeres. However, our observations do not support this idea as we see clear absence of hoxb1 expression in presump-tive r5+r6 in 5s stage val⁻ embryos (Fig. 4B). Thus, early expression of hox genes seems to be specified normally in the absence of val gene product; the possibility remains, however, that the val gene product is involved in ensuring that hoxb1 does not become re-activated in the r5+r6 region later in development. Hence, if inappropriate hox gene expression in rX is a manifes-tation of incorrect specification, this must be a relatively late effect, perhaps resulting from lack of normal inter-rhombomeric interactions in the absence of definitive r5 and r6. The alterna-tive idea, that cells may migrate into rX from adjacent rhom-bomeres, is supported both by the apparent gradual spread of ectopic gene expression through rX suggesting cell movement, and by the results of experiments which suggest that the r4/rX and r7/rX interfaces are not fully formed boundaries. In val⁻ embryos Moens et al. (1996) report an absence of morphologi-cal boundaries caudal to the r3/r4 boundary, and loss of expression of the boundary marker mariposa. Guthrie and col-leagues (Guthrie and Lumsden, 1991; Guthrie et al., 1993) found that when two ‘like’ rhombomeres are juxtaposed, cells can migrate across the interface; boundary formation requires juxta-position of alternating odd- and even-numbered rhombomere states. The results of Moens et al. (1996) are consistent with this idea and further suggest that both of these two states are absolutely required for boundary formation to occur, i.e.odd next to even leads to a boundary, but neither odd (r7) nor even (r4) next to unspecified territory (rX) can cause boundary formation, presumably due to lack of recognition between the adhesion systems of presumptive and definitive rhombomeres.

It is interesting to note that hoxb1 expression does not spread into the region of rX where krox-20 expression is maintained (Fig. 4D,H,J), suggesting that this region maintains r5-like properties incompatible with hoxb1 expression. This observa-tion suggests that if cell mixing is occurring across the r4/rX interface then cells must down-regulate hoxb1 as they pass through the krox20 expression region, or, perhaps more likely, that they must migrate ventral of this dorsally restricted domain. Currently, there are arguments to support or refute either of the two hypotheses we propose to explain the gradual changes in molecular identity of rX (i.e. incorrect positional specification of rX versus cell mixing across the r4/rX and r7/rX interfaces); resolution of this question will ultimately require cell labelling and time-lapse analyses.

The major differences in hox gene expression between normal and val⁻ embryos are summarized in Fig. 7. Loss of high level hoxb3 gene expression in rX suggests that the val gene product may be involved in up-regulating transcription of this gene. The relative timings of gene expression are consistent with this idea; in normal embryos high level hoxb3 expression is present in r5 and r6 from the 3-4s stage, shortly after the onset of val expression (Fig. 3; Moens et al., 1998). A similar interpretation was proposed by McKay et al. (1994) based upon analysis of Hoxa-3 expression in the mouse kr mutant. Furthermore, the kr gene product (the murine homologue of val) has recently been shown to interact with enhancer elements for Hoxb-3 which are required for transcriptional activity (Manzanares et al., 1997). Lack of high level hoxb3 expression correlates with the lack of an abducens nucleus in val⁻ fish. As described above, the high level hoxb3 expression domain colocalizes with the abducens nucleus in both zebrafish and mouse; these observations may suggest a possible role for the hoxb3 gene product in abducens specification. The recently demonstrated role of murine Hoxb-1 in specification of the facial nerve provides a precedent for this hypothesis (Goddard et al., 1996; Studer et al., 1996).

The zebrafish hoxd3 gene is normally expressed in r6 but not in r5. We observe expression of hoxd3 in rX, perhaps implying that some specific characteristic of zebrafish r5 normally prevents hoxd3 expression. At the 20s stage, hoxd3 expression appears to be confined to the posterior half of rX suggesting that some elements of normal AP patterning are retained along the length of rX; only the posterior half of rX exhibits the r6-like property of maintaining hoxd3 expression. This result is consistent with other observations that suggest some AP identity is retained by rX. For example, in mosaic experiments, if wild-type cells are placed in rX they are only able to express krox-20 if they lie in the anterior portion, close to r4. Similarly, the reticulospinal neurons retain their normal identity and AP order in val mutants.

Our expression analyses of hoxa2 and hoxb2 do not suggest regulation of the paralogue group 2 genes by the val gene product. In normal 20s stage embryos, hoxa2 expression is upregulated in r5. In val⁻ embryos there is no equivalent elevated expression in rX, perhaps reflecting lack of krox-20 expression; Krox-20 has been shown to directly upregulate Hoxa-2 transcription in mouse r5 (Nonchev et al., 1996). Expression of hoxb2 in both r3 and r5 of normal embryos shortly follows the onset of krox-20 expression, and rX expression in val⁻ embryos is limited to the krox-20-expressing region, again consistent with a role for Krox-20 in activating transcription of this gene as shown in the mouse (Sham et al., 1993). However, hoxb2 has a relatively low level of expression in r5 of normal 20s stage zebrafish. In val⁻ embryos the rX expression of hoxb2 is at a higher level than normal r5 expression, more similar to the expression level in r4. It seems likely that the gradual attainment of r4-like aspects of expression for the paralogue group 2 genes in rX, may reflect the same processes as do the changes in hoxb1 expression discussed above; namely aberrant cell mixing or incorrect pos-itional specification.

In wild-type embryos, the neural crest cells which migrate posterior to the otic vesicle toward the third branchial arch, express hoxb3 and, at low levels, hoxd3 and val. In val⁻ embryos this population of cells expresses hoxa2 and hoxb2; molecular markers characteristic of neural crest cells that migrate anterior to the otic vesicle at equivalent stages in normal zebrafish embryos. A similar caudally migrating crest cell population was revealed by CRABP I staining in the mouse kr mutant (McKay et al., 1994). We suggest that these neural crest cells transfer hox information appropriate to the second arch into the third arch, thus leading to a cartilage transformation. In val⁻ fish and kr⁻ mice, the 3rd arch cartilage shows shape changes sugges-tive of an anterior transformation towards 2nd arch characteris-tics (Moens et al., 1998; Frohman et al., 1993). The aberrant hox expression status of neural crest migrating into the third arch may reflect inappropriate cell migration, resulting from a shortening of the hindbrain in val⁻ and kr⁻ embryos, thus allowing primarily r4-derived 2nd arch neural crest to inappro-priately mix with more caudally derived 3rd arch crest. Alter-natively, aberrant hox expression in neural crest might result from inappropriate specification of the neural crest primordium. According to its hox expression status, rX has some aspects of r4 identity and thus rX-derived neural crest cells migrating into the 3rd arch may be incorrectly specified with 2nd arch identity. However, it should be noted that these r4-like aspects of rX hox expression are all seen well after the time at which neural crest cells migrate away from the neural keel.

Our detailed expression analyses at several developmental stages have allowed us to determine differences between primary and secondary defects in val mutant zebrafish. For example, we propose that loss of high level hoxb3 expression in val⁻ mutants which occurs from the earliest stages, likely reflecting a direct transcriptional regulation of hoxb3 by the val gene product, rep-resents a primary defect. However, we propose that changes in expression of other hox genes are due to secondary defects. For example, our analyses of several different developmental stages has revealed that hoxb1 expression posterior to r4 in val⁻ embryos is indistinguishable from that of wild-type embryos at early stages. However, at later stages there is an ectopic posterior extension of hoxb1 expression into rX of val⁻ embryos. Similarly, hoxb2 expression in rX of val⁻ embryos becomes increasingly similar to the expression in r4 with developmental time. These results suggest that rX only gradually attains aspects of r4-like identity during its development. Our interpretation that changes in hoxb1 expression are a secondary defect extends the observations in kr⁻ mice where a posterior extension of Hoxb-1 expression (Frohman et al., 1993; McKay et al., 1994) led to the idea that the territory posterior to normal r4, analogous to rX, has aspects of r4 identity.

Our results indicate that some aspects of hox gene expression in the val⁻ hindbrain are specified appropriately, but later in devel-opment there are secondary changes in expression, perhaps as a result of inappropriate cell mixing or lack of normal inter-rhom-bomeric interactions. Frohman et al. (1993) noted that expression of Hoxb-1 overlapped with that of Hoxb-4 and Hoxb-3 in kr⁻ mice; they interpreted this result to mean that the region posterior to r4 has a mixed identity, with properties of more than one rhom-bomere. Our results with val⁻ zebrafish suggest that any mixed identity of rX is a secondary event; at early times no region expresses markers appropriate to more than one rhombomere level. As many aspects of hox gene expression seem relatively normal in val⁻ embryos at early stages, this might suggest that hox-dependent AP patterning events that occur early would be correspondingly insensitive to val genotype. This idea is supported by the disposition of the early born reticulospinal neurons in val⁻ embryos: the wild-type complement of neurons is present and they retain the normal AP order. Similarly, as hox expression does not begin to differ markedly between wild-type and val⁻ embryos until after the 5s stage, this may also suggest that the proposed function of the val gene product in subdividing a presumptive r5+r6 territory occurs after this time. In mosaic experiments in which wild-type cells are placed in val⁻ embryos, the wild-type cells lying in r4 and r7 have not been seen to enter rX; however, our suggestion that rX gradually takes on a more r4-like identity, predicts that if such manipulated embryos were allowed to develop further, then rX might gradually become colonized by wild-type cells from adjacent r4 and r7.

There are many similarities between the val and kr pheno-types, although a few differences are apparent. For example, elevated levels of apoptotic cell death have been observed in the r4 region of kr embryos (McKay et al., 1994); no such cell death has been found in val⁻ embryos analyzed at the 20s stage (Moens et al., 1996). Another difference lies in the details of the expression profiles of the val and kr gene products; after rhombomere boundary formation val continues to be expressed throughout r5 and r6 (Moens et al., 1998), whereas at equiva-lent stages kr expression is confined to r5 plus the anterior part of r6 (Cordes and Barsh, 1994). These differences between mouse and zebrafish do not seem extensive enough to suggest a fundamentally different role for the val and kr gene products. We propose that in both systems the val/kr gene product is necessary for allowing subdivision of presumptive r5 and r6 into definitive rhombomeres. Additionally, by regulating hox gene expression, the val gene product presumably has an important role in imparting final segmental identity; thus the late differences in expression profiles of val and kr may have profound effects on mature neuroanatomy.

We are extremely grateful to Andreas Fritz, Len Zon and Marc Ekker for assisting us and sharing information regarding mapping of the hox genes. We would also like to thank Anand Chandrasekhar for sharing unpublished observations and Laure Bally-Cuif for comments on the manuscript. V. E. P. and C. B. M. are Human Frontiers Science Program long-term fellows. This work was supported by a Basil O’Connor Starter Scholar Research Award from the March of Dimes to R. K. H. who is a Rita Allen Foundation Scholar, by a donation from the Rathmann Family Foundation to the Molecular Biology Department at Princeton University, and by NIH grants RO1 HD34499 to R. K. H and NS17963 to C. B. K.